MD Nastran

MSC.Nastran 2005 r3
Design Sensitivity and Optimization
Users Guide
Main Index Main Index
Corporate
MSC.Software Corporation
2 MacArthur Place
Santa Ana, CA 92707
Telephone: (800) 345-2078
FAX: (714) 784-4056
Europe
MSC.Software GmbH
Am Moosfeld 13
81829 Munich
GERMANY
Telephone: (49) (89) 43 19 87 0
Fax: (49) (89) 43 61 71 6
Asia Pacific
MSC.Software Japan Ltd.
Shinjuku First West 8F
23-7 Nishi Shinjuku
1-Chome, Shinjuku-Ku
Tokyo 160-0023, JAPAN
Telephone: (81) (3)-6911-1200
Fax: (81) (3)-6911-1201
Worldwide Web
www.mscsoftware.com
Disclaimer
MSC.Software Corporation reserves the right to make changes in specifications and other information contained in this document without prior notice.
The concepts, methods, and examples presented in this text are for illustrative and educational purposes only, and are not intended to be exhaustive or to apply
to any particular engineering problem or design. MSC.Software Corporation assumes no liability or responsibility to any person or company for direct or indirect
damages resulting from the use of any information contained herein.
User Documentation: Copyright 2006 MSC.Software Corporation. Printed in U.S.A. All Rights Reserved.
This notice shall be marked on any reproduction of this documentation, in whole or in part. Any reproduction or distribution of this document, in whole or in part,
without the prior written consent of MSC.Software Corporation is prohibited.
This software may contain certain third-party software that is protected by copyright and licensed from MSC.Software suppliers.
MSC, MSC., MD, MSC.Dytran, MSC.Marc, MSC.Nastran, MD Nastran, MSC.Patran, the MSC.Software corporate logo, and
Simulating Reality are trademarks or registered trademarks of the MSC.Software Corporation in the United States and/or other
countries.
NASTRAN is a registered trademark of NASA. Windows is a registered trademark of Microsoft Corporation.
MSC.Nastran for Windows includes a licensed version of FEMAP from Unigraphics Solutions, Inc. Portions of FEMAP and related documentation are derived
from GHS3D software under license from INRIA, other portions are copyrighted by and are the property of Unigraphics Solutions, Inc. and Spatial Technology Inc.
All other brand names, product names or trademarks belong to their respective owners.
NA*V2005R3*Z*DSO*Z*DC-USR
Main Index Main Index
C O N T E N T S
MSC.Nastran Design Sensitivity and Optimization Users Guide

Table of Contents
Preface I About this Book, x
I List of MSC.Nastran Books, xvi
I Internet Resources, xvii
1
Getting Started I Introduction, 2
What is Design Optimization?, 2
What is Design Sensitivity?, 2
Why Use Design Sensitivity and Optimization?, 4
How Does Design Optimization Differ from Analysis?, 6
I Numerical Optimization Basics, 11
How Much Do I Need to Know About Optimizers?, 11
A Simple Illustration, 12
Numerically Searching for an Optimum, 16
I Structural Optimization, 26
Relating the Finite Element Analysis Model to the Design
Model, 26
Managing the Structural Optimization Task, 27
Overview of Approximation Concepts Used in Structural
Optimization, 28
I Summary, 31
2
Fundamentals of
Design Sensitivity
and Optimization
in MSC.Nastran
I Overview of Fundamentals, 34
Initial Design, 35
Structural Analysis, 35
Constraint Screening, 36
Sensitivity Analysis, 36
Optimizer, 36
Approximate Model, 36
The Improved Design, 37
Converged, 37
MSC.Nastran Design Sensitivity
and Optimization Users Guide
Table of Contents
Main Index Main Index
I The Design Model, 39
Design Variables, 39
Designed Properties, 39
Type-1 Properties, 44
Type-2 Properties, 46
Type-3 Properties, 47
Designed Shapes, 47
Basis Vectors in Shape Optimization, 48
Design Responses, 53
Type-1 Responses, 53
Type-2 Responses, 53
Type-3 Responses, 54
I Multidisciplinary Analysis, 58
Response Calculation, 59
Mode Tracking, 62
I Constraint Screening, 64
I Design Sensitivity Analysis, 67
General Considerations, 67
Element and Grid Responses, 71
I Optimization, 97
Connection Between the Optimizer and the Approximate
Model, 97
I The Approximate Model, 101
Formal Approximations, 101
A Simple Linear Design Space, 103
Move Limits, 104
Design Variable Limits, 104
Property Limits, 106
Transformation of the Approximate Optimization Task to a
Feasible Design, 110
Function Evaluation, 111
Gradient Evaluation, 119
I Tests for Convergence, 125
Convergence of Design Cycles: Hard and Soft Convergence, 125
I Special Topics, 133
Discrete Variable Optimization, 133
Fully Stressed Design, 135
Topology Optimization, 137
Manufacturability Constraints with Topology Optimization, 138
Minimum Member Size (Constraints), 138
Casting Constraints (Draw Direction Constraints), 138
Extrusion Constraints, 139
Main Index Main Index
Symmetry Constraints, 139
Advanced Numerical Techniques in SOL 200, 139
Matrix Domain Automated Component Modal Synthesis
(MDACMS), 143
Superelement Optimization, 144
Special Considerations When Designing One-Dimensional
Bending Elements, 147
3
Developing the
Design Model for
Sensitivity and
Optimization
I Overview of Design Modeling, 153
I Case Control for Design Optimization, 155
I Defining the Design Variables, 158
I Relating Design Variables to Properties, 162
I Relating Design Variables to Shape Changes, 166
I Designating the Design Responses, 188
DRESP1 Bulk Data Entries, 188
DRESP2 Bulk Data Entries, 188
DRESP3 Bulk Data Entries, 195
I Defining the Objective Function, 199
I Defining the Constraints, 201
4
Input Data I File Management, 205
Shape Optimization, 205
User Defined Beam Libraries, 205
External Responses, 206
I Executive Control, 207
Solution 200, 207
I Case Control Section, 208
Analysis Discipline Definition, 208
Design Task Definition, 209
Mode Tracking, 212
Shape Basis Vector Computation, 214
Generation of a New Bulk Data File, 215
I Bulk Data Entries, 216
BNDGRID, 216
DCONADD, 218
DCONSTR, 218
Main Index Main Index
DDVAL, 221
DEQATN, 222
DESVAR, 228
DLINK, 230
DOPTPRM, 231
Optimization Method (OPTCOD, METHOD), 233
DRESP1, 244
DRESP2, 267
DRESP3, 274
DSCREEN, 278
DTABLE, 279
DVBSHAP, 281
DVCREL1, 282
DVCREL2, 284
DVGRID, 287
DVMREL1, 288
DVMREL2, 290
DVPREL1, 293
DVPREL2, 296
DVSHAP, 298
MODTRAK, 299
TOPVAR Topological Design Variables, 300
I Parameters Unique to Design Sensitivity and Optimization, 304
5
Solution
Sequences
I Solution 200 Program Flow, 309
I Design Sensitivity and Optimization SubDMAPS, 314
I Design Sensitivity and Optimization Modules, 317
6
Output Features
and Interpretation
I Output-Controlling Parameters, 327
I Design Optimization Output, 329
DSOUG4 Output, 329
Modification of Move Limit Parameters, 358
Special Prints for Fully Stressed Design, 359
Special Prints for Discrete Variable Optimization, 360
Special Prints for Topology Optimization, 363
I Design Sensitivity Output, 364
Formatted Design Sensitivity, 364
Unformatted Design Sensitivity, 369
Main Index Main Index
I Design Punch Output, 374
Punch Parameters, 374
Example for DESPCH, 375
Special Punch Considerations for Topology Optimization, 376
Special Punch in the Case of Shape Optimization, 377
New Bulk Data File, 377
I Postprocessing Output, 380
Comma Separated Values File, 380
Output from PARAM POST, 381
7
Example Problems I Three-Bar Truss, 388
Analysis Model Description, 389
Design Model Description, 389
I Vibration of a Cantilevered Beam (Turner's Problem), 395
Analysis Model Description, 395
I Cantilevered Plate, 401
Analysis Model Description, 402
Design Model Description, 402
I Stiffened Plate, 412
Analysis Model Description, 413
Design Model Description, 416
I Shape Optimization of a Culvert, 419
Problem Description, 419
Modeling Considerations, 420
Creation of Auxiliary Model, and Generation of Basis
Vectors, 420
Design Optimization Input, 424
I Analytic Boundary Shapes, 430
Problem Description, 430
Boundary Shape Changes Using Auxiliary Boundary
Models, 432
Shape Changes over the Interior, 441
Modeling Summary, 442
Shape Basis Vectors, 442
Design Task, 443
Optimization Results, 443
I Dynamic Response Optimization, 447
I Twenty-Five Bar Truss, Superelement and Discrete Variable
Optimization, 461
Main Index Main Index
Analysis Model Description, 461
Design Model Description, 462
I Design Optimization with Composite Materials with Fully Stressed
Design, 470
I Acoustic Optimization, 478
I RMS Response, 490
I Transient Dynamic Optimization, 497
I External Response to Include Alternative Buckling Response, 504
I Topology Optimization, 513
I Continuing the Design Process in a Subsequent Job, 521
Continuing a Property Optimization Design Task , 521
Continuing a Shape Optimization Design Task, 525
Restarting from a Previous Analysis, 526
A
Nomenclature,
Glossary of Terms,
and References
I Nomenclature, 532
I Glossary of Terms, 535
I References, 545
B
Adding Your Own
Beam Cross-
Section Library
I Introduction, 548
I BSCON Subroutine, 550
I BSBRPD Subroutine, 552
I BSGRQ Subroutine, 566
I BSBRT Subroutine, 568
I BSBRID Subroutine, 572
I BSBRGB Subroutine, 580
I BSBRCD Subroutine, 587
I BSMSG Subroutine, 594
I Linking Your Library to MSC.Nastran, 596
I Example of Building and Linking a Beam Server, 598
Main Index Main Index
C O N T E N T S
MSC.Nastran Design Sensitivity and Optimization Users Guide

C
MSC.Patran
Support for Design
Sensitivity and
Optimization
I Tools Concepts and Definitions, 603
Design Studies and Design Optimization, 603
I The Tools Menu, 605
Menu Conventions, 605
I Tools Commands, 607
Tools>Model Variables, 607
Creating Model Variables, 607
Values as Variables, 608
Element Properties as Variables, 609
Beam Dimensions as Variables, 611
Material Properties as Variables, 613
Displaying Variables, 614
Modifying Variables, 615
Deleting Variables, 616
Tools>Design Studies, 617
Creating Design Studies, 618
Design Variables, 620
Design Cycle Select, 622
Design Objective, 623
Design Constraints, 624
Constraints for Linear Static Solutions, 624
Stress as Constraint, 627
Strain as Constraint, 630
Force as Constraint, 631
Composite Stress/ Composite Strain as Constraint, 633
Constraints for Normal Modes, 634
I Submitting a Job from MSC.Patran for SOL 200, 635
Optimization Parameters, 636
Subcases, 638
Subcase Select Optimize, 639
I Supported OUTPUT2 Result and Model Quantities, 641
Main Index Main Index
C O N T E N T S
MSC.Nastran Design Sensitivity and Optimization Users Guide

C
Numerical
Optimization
D
Numerical
Optimization
I Basic Optimization, 645
I The Modified Feasible Direction Algorithm, 652
I Sequential Linear Programming, 671
I Sequential Quadratic Programming, 674
I Summary, 676
INDEX I MSC.Nastran Design Sensitivity and Optimization Users Guide 677
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
Preface
I About this Book
I List of MSC.Nastran Books
I Internet Resources
Main Index Main Index


x
About this Book
MSC.Nastran is a general purpose finite element program which solves a wide variety
of engineering problems. It is developed, marketed and supported by the
MSC.Software Corporation.
The MSC.Nastran Design Sensitivity and Optimization Users Guide is intended to
explain the capability of MSC.Nastran to predict the effect of changes in the structural
model on structural responses (sensitivity) and to resize the structure so as to satisfy
imposed design conditions while taking a particular response to minimum (or
maximum) value (optimization). As such, it is the Users Guide for two MSC.Nastran
modules: Design Optimization and Topology Optimization. The Design Optimization
capability performs general tasks that are the focus of this volume while Topology
Optimization is a special purpose capability intended to find the optimal distribution
of material and is described in a special section of this volume. Note that sensitivity
analysis is available in standard MSC.Nastran and does not require additional
modules. The guide provides guidance in preparing the input and assessing the
output. A number of examples are provided to illustrate the use of MSC.Nastran
Design Sensitivity and Optimization.
This guide has been made possible by the work of numerous people at the
MSC.Software Corporation. Notably, the Optimization Project within the Nastran
Development organization has provided the technical developments of the capability.
Members of this group are Dr. Gee (David) Chou, Dr. Shenghua Zhang, Dr. Xiaoming
Yu, Dr. Leonard H.Woo and myself. Mr. Mike Reymond has led the development
effort associated with the DMAP (Direct Matrix Abstraction Program) used in the
implementation of this capability. On the documentation side, Ms. Lori J.Lampert
initiated the development of this publication, which was completed by Ms. Wendy
Webb. Ms. Donna Thaete provided an editorial review.
The development of the Design Sensitivity and Optimization capability has been
greatly influenced by numerous client requests and feedback. Four clients who stand
out in this respect are: Mr. Ingo Raasch of BMW, Mr. Peter Hougardy of Audi, Dr.
Wayne Nack of General Motors and Mr. Dan Barker of Lockheed-Martin Aeronautics.
Dr. Srinivas Kodiyalam provided MSC.Software with his enhanced version of the
ADS source code at no cost. It is the interest of these people and all the others that
make providing this capability such an interesting and challenging endeavor.
This is the fourth edition of this guide. The first was issued in conjunction with
Version 67 of MSC.Nastran in 1992 and the second was issued in 1994 in conjunction
with Version 68. Both of these documents were edited by Dr. Gregory Moore and his
contributions are gratefully acknowledged. The organization from these earlier
Main Index Main Index
xi Preface
guides, as well as much of the material, has been retained for subsequent editions. The
third edition was issued in conjunction with the release of MSC.Nastran 2004. This
fourth addition corrects errors and omissions in the third edition and adds material
for the 2005 release of MSC.Nastran.
Organization
The Guide is divided into seven chapters and five appendices. A brief description of
each of these is:
Chapter 1. Getting Started. This chapter provides an overview of design sensitivity
and optimization and is presented at an introductory level. This chapter assumes
limited knowledge of the field and can be skimmed or skipped altogether by
knowledgeable workers in this area.
Chapter 2. Fundamentals of Design Sensitivity and Optimization in
MSC.Nastran. This chapter can be regarded as a theoretical overview of the Design
Sensitivity and Optimization capability that identifies and describes the concepts that
are used. There is minimal reference to the user interface in this chapter. Instead, it
provides information on the algorithms used and is intended for the high-end user
and for developers.
Chapter 3. Developing the Design Model for Sensitivity and Optimization.
Chapter 3 serves as a bridge between the concepts of Chapter 2 and the input
description of Chapter 4. It is intended to provide an overview of the concept of the
Design Model as it is implemented in MSC.Nastran and provides guidelines for the
use of the capabilities.
Chapter 4. Input Data. This chapter provides explicit descriptions of each of the
Case Control Commands, Bulk Data Entries and Parameters that are related to design
sensitivity and optimization. In goes into greater depth in describing these items than
is possible win the MSC.Nast ran Quick Reference Guide.
Chapter 5. Solution Sequences. Chapter 5 provides a high-level overview of the
SOL 200 DMAP sequence. A detailed description is beyond the scope of this guide, but
it is felt that this introduction can give readers some bearings when they set about to
explore this complex sequence.
Chapter 6. Output Features and Interpretation. This chapter provides a
description of the output available from running design sensitivity and optimization
tasks. In addition to a comprehensive discussion of the output provided in the.f06
Main Index Main Index


xii
output file, there is also mention of the design sensitivity and optimization outputs
that are available in the.pch file and that can be used in post-processors, especially
MSC.Patran.
Chapter 7. Example Problems. A series of example problems are given to illustrate
the capabilities available in MSC.Nastran for Design Sensitivity and Optimization. All
of the example problems are available in the MSC.Nastran 2002 TPL (Test Problem
Library) with the naming convention DSOUGxx.dat. It is expected that all readers will
profit from reading this chapter and exercising the example problems. Studying the
effects of variations on the provided input should be particularly profitable.
Appendix A. Nomenclature and Glossary of Terms. This appendix provides a
quick reference for the nomenclature used in this guide and to some technical terms
used throughout.
Appendix B. MSC.Patran Support for Design Sensitivity and Optimization. This
appendix is extracted from MSC.Patran documentation and is given for completeness.
Appendix C. Numerical Optimization. This final appendix is taken almost directly
from a Users Guide published by Vanderplaats Research and Design entitled DOT,
Design Optimization Tools Users Manual, Version 5.0.
Developments Introduced Since Version 68 of MSC.Nastran
New versions of MSC.Nastran are released periodically. This guide has been prepared
with minimal reference to the version number and is consistent with the 2005 Release
of MSC.Nastran. As mentioned above, the previous version of this guide was released
in conjunction with Version 68. There have been several significant releases in the
meantime and it is felt to be of historical interest to show how each of these releases
added to the Design Sensitivity and Optimization capability. It is also of great
practical interest to those who are not using the 2005 release to see which of the
capabilities described in this Guide are not available to them. Note that MSC.Nastran
has changed its convention for naming releases. Previously, each release was given
the next integer in line, with minor releases identified as a point release; i.e., a number
after the decimal point. This convention changed after Version 70.7 was released so
that the version now corresponds to the year in which it first became available and the
term Version is not used. Since the Version 68 release, there have been seven
additional releases and each has had some enhancement in the Design Sensitivity and
Optimization capability. The major enhancements are identified here. Discussions of
each of these topics are included in this Guide.
Main Index Main Index
xiii Preface
Version 68.2
Support for inertial relief in a statics analysis
Consideration of differential stiffness effects in buckling sensitivity analysis
Version 69
Design of the PBARL
This was done in conjunction with the introduction of the beam library in the
same release. The beam library also allowed for the analysis of PBEAML
properties, but their design was deferred until the Version 70.7 release.
Mode tracking
Formatted design sensitivity prints
Support for multiple boundary conditions (normal modes, buckling and
flutter) Multiple static boundary conditions were first supported in Version
68.
Version 70
Adjoint sensitivity analysis
Version 70.5
Rigid element sensitivities with shape design changes
Simplified design of beams
There has been an ongoing project to simplify the user requirements and
calculations when beams are designed. This has been particularly significant
when tapered PBEAMs and PBEAMLs (Version 70.7, et seq.) are designed.
Version 70.7
Design of the PBEAML
Material and connectivity sensitivity
Total weight response
In earlier releases, the weight had been computed by multiplying the
structural volume by the density of the material. This neglected any non-
structural mass and concentrated masses.
Extend the DRESP2
Main Index Main Index


xiv
This entailed a number of new features. More significant ones includes the
ability to include material and connectivity properties and the ability to
reference a DRESP2.
Enhanced design data in the punch file
Output of a comma separated values (CSV) file with design data that can be
read into a spreadsheet.
MSC.Nastran 2001
Complex eigenvalue sensitivity and optimization
Discrete variable optimization
Random response sensitivity and optimization
Fully stressed design
MSC.Nastran 2004
External Response (DRESP3)
This is an undocumented feature of the 2001 release. Users should now be
able to access this capability with MSC.Nastran 2004.
Eigenvector sensitivity and optimization.
Subcase dependent frequency excitation sets.
Prior to this enhancement, it was necessary to use the same set of excitation
frequencies in each frequency response subcase.
DMP and ACMS technology.
Punch of a complete, unsorted Bulk Data file.
Rayleigh Quotient Approximation for eigenvalues/ eigenfrequencies.
Spanning of subcases and superelements.
PSD response sensitivity and optimization.
DESVAR Case Control command.
Removal of property limit constraints that are redundant with design
variable limits.
MSC.Nastran 2005
Topology optimization
Main Index Main Index
xv Preface
Composite strength ratio response
Beta and match function for the DRESP2 ADS optimizer
BIGDOT optimizer
Transformation of approximate optimization task for a feasible design
MSC.Nastran 2005 r3
MSC ADS as default optimizer
Manufacturing constraints for topology optimizer
Erwin H. Johnson
Main Index Main Index


xvi
0.1 List of MSC.Nastran Books
Below is a list of some of the MSC.Nastran documents. You may order any of these
documents from the MSC.Software BooksMart site at www.engineering-e.com.
Installation and Release Guides
Installation and Operations Guide
Release Guide
Reference Books
Quick Reference Guide
DMAP Programmers Guide
Reference Manual
Users Guides
Getting Started
Linear Static Analysis
Basic Dynamic Analysis
Advanced Dynamic Analysis
Design Sensitivity and Optimization
Thermal Analysis
Numerical Methods
Implicit Nonlinear (SOL 600)
Aeroelastic Analysis
Superelement
User Modifiable
Toolkit
Main Index Main Index
xvii CHAPTER
Preface
Internet Resources
MSC.Software (www.mscsoftware.com)
MSC.Software corporate site with information on the latest events, products and
services for the CAD/ CAE/ CAM marketplace.
Simulation Center (simulate.engineering-e.com)
Simulate Online. The Simulation Center provides all your simulation, FEA, and other
engineering tools over the Internet.
Engineering-e.com (www.engineering-e.com)
Engineering-e.com is the first virtual marketplace where clients can find engineering
expertise, and engineers can find the goods and services they need to do their job
CATIASOURCE (plm.mscsoftware.com)
Your SOURCE for Total Product Lifecycle Management Solutions.
Main Index Main Index


xvii
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
CHAPTER
1
Getting Started
I Introduction
I Numerical Optimization Basics
I Structural Optimization
I Summary
Main Index Main Index


2
1.1 Introduction
This chapter introduces some of the basic concepts of numerical optimization with an
emphasis on the methods used in MSC.Nastran.
Some of the questions answered in this chapter include:
What is design optimization, and how does it differ from analysis?
What is the relationship between design sensitivity and optimization?
How is an optimization problem formulated?
How does an optimizer search for an optimum?
How does an optimizer communicate with the structural analysis?
What is Design Optimization?
Optimization is a broad and active area technology and we need to first focus the
discussion on the particular application that is the subject of this Users Guide. A
recent search of an online bookseller for titles that included the word optimization
resulted in 1740 matches. The majority of the titles indicated that they dealt with topics
related to this guide in that they detailed methods, generally involving a computer
solution, that could be used to improve a design or process. It is, however, impractical
and unnecessary to try and absorb all this material in order to utilize the methods
described here. Instead, we must first restrict ourselves to the optimization of
structures that can be analyzed with MSC.Nastran. In this context, design
optimization refers to the search for a structural design that is, in some sense, optimal,
or the best, while varying structural parameters. While performing this search, the
design is guided to satisfy operating limits that are imposed on the response of the
structure and by further limits on the values the structural parameters can assume.
This, in a nutshell, represents the design optimization task in MSC.Nastran and leads
directly to concepts that are covered in this chapter of objective, design variables,
design constraints, side constraints and search.
What is Design Sensitivity?
The design optimization capability in MSC.Nastran benefits significantly from being
based on a design sensitivity analysis. Design sensitivity analysis computes the rates
of change of structural responses with respect to changes in design parameters. These
design parameters, or design variables, can be such things as shell thicknesses, beam
dimensions, hole radii, material variables and so on. In civil engineering we may be
Main Index Main Index
3 CHAPTER 1
Getting Started
interested in how changes in the deflection of a bridge span can be affected by changes
in the dimensions of the bridge sections. In automotive chassis design, we may want
to investigate changes in cabin resonant frequencies given changes in panel
thicknesses. These rates of change (what we call partial derivatives in the language
of calculus) are called design sensitivity coefficients.
These sensitivities are computed explicitly in MSC.Nastran and are extremely
valuable in their own right as they can be used to predict how a design change will
affect an important response. When they are applied with an optimizer, they greatly
improve the efficiency of the search since the algorithm now not only knows the
current state of the design but also has an idea of which way to look for an improved
design. MSC, therefore, takes considerable care that the predicted sensitivities are as
accurate as possible and applies special techniques to insure that they predict accurate
response values as the design variables change.
The basic design optimization capability in MSC.Nastran depends on having design
sensitivity information available and optimization cannot be performed without it. It
should be noted that this is not true in general and that it is possible to perform design
optimization without having design sensitivity information available. It is possible,
for example, to infer the affects of design changes by making a series of analyses for
different design variable values. There are any number of products available that
perform optimization in this way and they have their advantages in certain situations.
In particular, design optimization in MSC.Nastran is limited to those analyses and
design parameters that have been explicitly designated in the design of the capability
and that can be accessed by you in the performance of an optimization task. As you
will learn from studying this guide, the range of design tasks that MSC.Nastran can
address is very broad and expands as new versions of the code are released. There will
always remain, however, situations that cannot be addressed with MSC.Nastran and
it is necessary to study this guide to determine if your design task is one that can be
treated by the capability.
As a final note to motivate the value of sensitivity analysis, it is useful to consider what
would occur if you did not have this information available. If the design task was to
find the value of a single design variable that provided the best performance, one
could conceive that performing analyses with several values of this parameter would
provide information that could be displayed as a function of the variable and the
optimum design could be visually determined. Now consider the case with two
design variables. The number of analyses required is expanded and the plot now
becomes three dimensional, but it still should be possible to find an optimum in this
fashion. But now consider 10 variables and then one hundred and then one thousand.
Main Index Main Index


4
At some point, the most powerful computer would be overwhelmed by the number
of analyses required and in making sense of the analysis results. With sensitivity
analyses, it is quite reasonable to pose design tasks in MSC.Nastran with a set of
design variables that is far beyond that possible without the sensitivity information.
Even for a small number of design variables, the presence of design sensitivities
makes finding solutions much more efficient than what would be possible without
these sensitivities.
Why Use Design Sensitivity and Optimization?
Why use design sensitivity and optimization? Perhaps you have been requested to
investigate product improvement using optimization, or perhaps you are aware that
these tools can be used to create better engineered designs. Design sensitivity and
optimization are used when we seek to modify a design whose level of structural
complexity exceeds our ability to make appropriate design changes. What is
surprising is that an extremely simple design task may easily surpass our decision-
making abilities. Experienced designers, those with perhaps decades of experience,
are sometimes fantastically adept at poring through mounds of data and coming up
with improved designs. Most of us, however, cannot draw upon such intuition and
experience. A basic goal of design optimization is to automate the design process by
using a rational, mathematical approach to yield improved designs. Ways in which
this might be put to use include:
1. Producing more efficient designs having maximum margins of safety.
2. Performing trade-off or feasibility studies.
3. Assisting in design sensitivity studies.
4. Correlating test data and analysis results (model matching).
In addition to providing a complete description of the optimization tools in
MSC.Nastran, part of the aim of this users guide is to suggest various ways in which
design sensitivity and optimization might be used. Consider the following examples:
Example A
A complex spacecraft is in a conceptual design stage. The total weight of the spacecraft
cannot exceed 3,000 pounds. The nonstructural equipment including the payload is
2,000 pounds. Static loads are prescribed based on the maximum acceleration at
launch. Also, the guidance systems require that the fundamental elastic frequency
must be above 12 Hz. It is extremely important to reduce the structural weight since
it costs several thousand dollars to place one pound of mass in a low earth orbit. There
Main Index Main Index
5 CHAPTER 1
Getting Started
are three types of proposed designs: truss, frame, and stiffened shell configurations.
Currently all of the designs fail to satisfy at least one design requirement and are
expected to be overweight. We are to determine which configuration(s) promises the
best performance and warrants detailed design study. Also, the payload manager
needs to know how much weight could be saved if the frequency requirement were
to be relaxed from 12 Hz to 10 Hz. The spacecrafts structure contains about 150
structural parameters, which we may want to vary simultaneously.
Example B
One part of a vehicles frame structure was found to be overstressed. Unfortunately,
it is too expensive to redesign that particular frame component at this stage in the
engineering cycle. However, other structural components nearby can be modified
without severe cost increases. There are nearly 100 structural design parameters that
can be manipulated. The design goal is to reduce the magnitude of the stresses by
reducing the internal load to the overstressed member.
Example C
A frame structure, which supports a set of sensitive instruments, must withstand
severe in-service dynamic loads. Modal test results are available from comprehensive
tests performed on the prototype structure. We need to create a finite element model
for dynamic analysis that is much less detailed than the original model created for
stress analysis since the costs of dynamic analysis using a complex model would be
prohibitive. We must ensure, however, that the first ten modes obtained from our
simplified model are in close agreement with those obtained from the test results. The
goal is to determine suitable properties for the lumped quantities in our simplified
dynamic model so that the first ten eigenvalues correlate well with the prototype.
Example D
Your company supplies automotive components and one of your products is a
connecting rod that is produced in the millions for a number of vehicles. The design
of the component is well understood and has been refined for years based on detailed
design loads. It is now decided to see if shape optimization techniques can be applied
to produce a design that satisfies the design requirements at a final weight that is less
than the current weight. Even a 1-2% improvement will translate into significant cost
savings when it is applied across the large production run.
Main Index Main Index


6
How Does Design Optimization Differ from Analysis?
Although design optimization and analysis can be viewed as complementary, there
are some important conceptual differences between the two which must be clear in
order to make effective use of both.
When we perform an analysis, we create a mathematical idealization of some
physical system in order to obtain estimates of certain response quantities. The class
of responses that we are interested in defines the applicable analysis discipline to be
used, while the accuracy of these responses is dependent on the quality of the analysis
model and our general knowledge of the true system. Our choice of finite element
types, representation of boundary conditions, loads, and definition of the finite
element mesh all play critical roles in determining how well our model is able to
predict the responses of the physical structure. The goal is to obtain an accurate
prediction of the responses which can be expected from the real structure. For
example, consider the plate subjected to uniform tensile loads in Figure 1-1. The
corresponding analysis model in Figure 1-2 is a discretized finite element
representation of idealized geometry, loads, and boundary conditions.
Figure 1-1 Flat Plate with Hole
y
x
R
l
T
x
T
x
Main Index Main Index
7 CHAPTER 1
Getting Started
Figure 1-2 Analysis Versus Design Models
In contrast, a design model is an idealized statement of changes which might be made
to the structure to improve its performance or response. In order to accomplish this,
we need to define what we mean by an improved design. It may be the minimum
weight or maximum stiffness, but whatever our choice is, this constitutes a statement
of the design objective. The design may be varied such that certain bounds on
responses are not exceeded. Expressions of maximum allowable stresses or minimum
permissible frequencies are termed design constraints. And, in our description of how
the design might be changed, we use design variables to express what we mean by a
suitable variation. Limits on the range of the design variables are called side
constraints. By convention, the mathematical region over which our design variables,
objective, and constraints are defined is called the design space. Figure 1-2 also shows
the design model corresponding to a redesign of the hole radius for weight
minimization. It gives allowable structural variations (hole radius) subject to limits on
structural responses (stress).
R
Analysis Model Design Model
Finite element discretization of:
Structure (Mesh).
Loads.
Boundary Conditions
(1/ 4 Plate Representation).
Find R such that:
Weight is minimized.
Stresses do not exceed
allowables.
(R is the design variable, weight is the
design objective, and stresses are the
design constraints.)
Main Index Main Index


8
Probably the biggest difference between analysis and design is that analysis leads to
the solution (within the limits of the analysis model), while design optimization
leads to a solution. In other words, in analysis, we are usually guaranteed a unique
solution, while more than one solution may be possible in design optimization.
Mathematically, we can say that our design space may contain local minima. This is
analogous to the situation in which we may want to find the low point in a valley, yet
various low points are separated by hills that we cannot see over. Simply finding a low
point itself represents an acceptable solution, but there may be more than one such
solution. In some instances we may be able to restate the problem and, in effect, shift
the locations and contours of the hills to allow more efficient convergence. However,
the fact remains that our goal has been stated in the context of design improvement
and not in determining a global optimum or a unique solution (as in the case of
analysis).
One thing the analysis and design models share is that the results are dependent upon
the skill and judgment used in their construction. A poorly-meshed finite element
model may lead to inaccurate and misleading results. Similarly, an ill-posed design
optimization task may produce unexpected or useless results. Since the redesign
process is based on analysis results, the results of design optimization are strongly
dependent on the integrity of the analysis model. In fact, the optimizer can be
expected to exploit inaccuracies in the analysis model in a way that helps it perform
the design task.
Optimizer Limitations
A numerical optimizer seeks to find an improved design by trying to minimize or
maximize a specified objective. Throughout this process, it must adhere to the bounds
on responses and design variables given in the design model. It does not have the
intelligence to modify the objective or relax any of these limits. For example, suppose
you asked a friend to find you a nice apartment on his street. Your friend, the
optimizer, may have a somewhat different definition of nice than you do. His
income might be higher than yours, so that the optimal design he proposes may be
infeasible in terms of your bank account. Even though he is searching just on his street,
the next block may turn out to have an apartment that you consider a better value. The
optimizer is not able to go beyond your specifications to search out other possible
configurations.
The optimization problem statement requires an explicit description of the design
objective, as well as bounds which define the region in which it may search. You may
ask for a design satisfying a minimum flexibility requirement such as wing tip
Main Index Main Index
9 CHAPTER 1
Getting Started
deflection, but without a weight budget, the design which the optimizer proposes
may turn out to be unrealistic. You might get an extremely stiff, 200 ton wing out of
this process. You also might have asked for a minimum weight design but allowed for
negative physical dimensions. The optimizer may have no trouble minimizing the
weight by adding negative mass, but this design may produce a challenge on the
factory floor!
What Do I Need to Use Design Optimization Effectively?
Since design sensitivity and optimization depends on the results of analyses, a
reasonable level of skill preparing MSC.Nastran analysis models is required. Design
optimization in MSC.Nastran can combine analysis results from a number of
disciplines, including statics, normal modes, buckling, direct and modal frequency,
modal transient, acoustic, direct and modal complex eigenanalysis, static aeroelastic
response and flutter analysis. In addition, design models can also employ
superelements. Proficiency in any or all of these disciplines is useful.
The topics covered in this users guide are intended to describe design sensitivity and
optimization in MSC.Nastran. No prior knowledge of structural optimization is
required. However, it must be recognized that the discussion of structural
optimization contained in this guide is restricted to the features appearing in design
sensitivity and optimization in MSC.Nastran. This guide attempts to describe all
aspects completely so that someone new to the field of design optimization, as well as
those with more experience, has enough information available to use it wisely and
effectively. More information may, at times, be desired on a topic than is available is
this guide. In these instances, any one of the excellent references on design
optimization listed in the reference section may prove useful. Many, in fact, are
authored by distinguished researchers who have made direct contributions to design
optimization in MSC.Nastran.
Probably the most critical requirement in the effective use of design optimization is
common sense. Any sufficiently general and powerful tool possesses the capacity for
misuse. This is especially true of numerical optimizers. The old adage that engineering
is more art than science is probably more true of design optimization than of many
other disciplines. As you gain experience with numerical optimization, you will
probably discover a few pitfalls that are particular to your fields of application, and in
the process you may discover some especially useful and efficient tricks for clearly
stating your optimization tasks. In any field of application, there is no substitute for a
well posed problem. If your design goals are clear, the constraints are meaningful and
well-conditioned, and the design variables are chosen carefully to produce useful
Main Index Main Index


10
designs, then success becomes more certain. The description of methods for
accomplishing these goals forms the majority of the material of this guide. Before
moving on to a detailed discussion of MSC.Nastran design optimization in
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran in
Chapter 2 and Developing the Design Model for Sensitivity and Optimization in
Chapter 3, the next two sections introduce some of the basic concepts of numerical
optimizers and how they are used to solve problems in structural optimization.
Main Index Main Index
11 CHAPTER 1
Getting Started
1.2 Numerical Optimization Basics
This section introduces some of the basics of numerical optimization in an intuitive
manner, stressing overall concepts over technical details. Since few universities
include design optimization in their curricula, this section may be a useful
introduction to numerical optimization for those entirely new to the subject. Once the
basics of numerical optimization have been covered, Structural Optimization on
page 26 introduces the extension of these methods to the field of structural design.
How Much Do I Need to Know About Optimizers?
Some knowledge of the basic procedures involved in numerical optimizers will aid in
understanding why an optimizer does what it does, in interpreting the final results of
an optimization run, and in understanding what may have happened if the results are
unexpected. As an analogy, it is not necessary to know all of the details of sparse
matrix decomposition in order to perform a linear static finite element analysis, but if
singularities are present in your model and the decomposition procedure fails, the
results are far less mystifying, and a modeling solution to the problem may be much
more apparent if you know something of the basics of the solution procedure. Most of
the parameters that control the optimizer in MSC.Nastran can be changed to improve
performance for various classes of problems. Understanding the significance of these
choices allows you to take full advantage of the tools at your disposal.
Design optimization in structural redesign is actually an application of operations
research (a branch of applied mathematics) to problems in engineering design.
Generally, these are classes of problems in which the optimum allocation of scarce
resources is desired. Operations research is frequently used to solve scheduling
problems such as the routing of airplanes among various airport facilities. The
allocation of 100 airplanes among 68 airports may not seem to have much to do with
engineering design, but the optimization methods employed are similar and can be
extended to structural problems. An efficient engineering solution to a design
problem involves the optimum allocation of scarce resources. For example, tensile
stresses cannot be allowed to assume unlimited values and must be restricted to
within reasonable limits (the distribution of strain energy density must be made in an
optimal manner). Likewise, reducing structural mass leads to a savings of material
and possibly maintenance, fuel, or other indirect costs.
Main Index Main Index


12
A Simple Illustration
A useful way of introducing basic optimization concepts is to visualize a design task
of finding the lowest spot in the immediately surrounding terrain. Suppose we are
standing on the side of a hill and would like to find the point of lowest elevation; this
is our objective. We will quantify the location of any point in terms of coordinates
and these are our design variables. Suppose also that some fences exist which force
us to restrict our search to within the region enclosed by these fences. These fences or
constraints act as bounds in our design space, which is the region that defines all
of our possible positions on the hill. Only one out of all points on the hill can be
considered an optimum, though. For simplicity, we are neglecting the presence of
local minima.
Finding the lowest point on the hill while staying inside the fences is no real problem.
All we really need to do is have a good look about and note, from our perspective,
which point on the hill appears to be the lowest. We have scanned the hill, analyzed
thousands of possible candidates at a glance, and made an immediate decision. If we
were blindfolded, though, our decision-making process would not be as simple, and
that is exactly the task a numerical optimizer is faced with.
In a computational sense, the elevation of a single point on the hill, or the numerical
value of our objective function, must be determined by an analysis which may take
considerable effort. Evaluating hundreds or thousands of candidate designs may be
prohibitive. We need a systematic method of searching for an optimal design. There
are numerous techniques available to solve such a problem, all of which are classified
as numerical optimization algorithms.
Generally, numerical optimization methods seek to determine a direction of travel or
search direction that moves us down the hill as quickly as possible, yet allows us to
find an optimum that lies within the fences. A sequence of search directions is usually
employed during the overall search procedure.
In our hill example, we could easily find a search direction, even though blindfolded,
by taking small steps from side to side and then forward and back to test for elevation
changes. This will allow us to determine which direction will move us down the hill
the fastest and, based on this estimate, we can then proceed until we hit a fence or the
hill starts to climb up again. What we have done is to find the local value of the
gradient of our objective function and then used this information to establish a
probable direction in which to search for a minimum. Numerical optimization
algorithms that rely on gradient information are termed gradient-based. Once we
have done the best we can possibly do in this direction, we find another search
Main Index Main Index
13 CHAPTER 1
Getting Started
direction, and again proceed as before. We continue to repeat this procedure until we
cannot reduce the objective function any further; that is, we have reached a point at
which any move will take us up hill (an unconstrained option) or a point where we
can move further downhill only by climbing over a fence (a constrained optimum).
To quantify the location of a point on the hill, we might use north-south and east-west
coordinates corresponding to the elevation at a given point as our design variables. In
design optimization terms, this is a two design variable space since two coordinate
values are required to uniquely specify a point in the design space. Two design
variables are the most we can easily visualize. The task becomes understandably more
complex when considering that an optimizer may have to deal with hundreds of
design variables. In fact, recent topology optimization developments have considered
one million design variables.
There are three other conditions that may apply to our hill climbing exercise that
illustrate two additional concepts. The fences on the hill are properly considered
constraints, but we may have been told than under no circumstances can we go north
further than a specified amount, even though there is no fence there. This constraint
that is imposed directly on the design variable is called a side constraint. We might
further have the condition that we want the design to lie on some prescribed path or
curve drawn on the hillside. This is an equality constraint. Note that if there are as
many equality constraints as design variables, a unique solution exists (as long as the
equalities are linearly independent). This solution can be found using standard
algebraic methods. (A linear finite element analysis belongs to this category of
problem.) When the number and type of constraints do not enable a direct, unique
solution, the job becomes complicated and numerical optimizers are the best way to
attack the problem. Another condition is that we may be restricted to coordinates that
are a multiple of 10 (10.,20.,30.....1000.). This is the concept of discrete design
variables and in MSC.Nastran this is handled by first finding the continuous
optimum and then stepping over to the closest discrete solution that minimizes the
elevation.
This illustration is too simple in some ways and we will learn that following the
maximum gradient is not always the fastest way to find the bottom of the hill. But we
can build on this experience to express the equations which describe the basic
optimization problem statement.
Main Index Main Index


14
The Basic Optimization Problem Statement
We are now in a position to express our optimization task in a quantitative form. This
mathematical expression of the design problem is called the basic optimization
problem statement and can be found in many textbooks.
Find X to minimize (or maximize)
objective Eq. 1-1
subject to
inequality constraints Eq. 1-2
equality constraints Eq. 1-3
side constraints Eq. 1-4
where:
design variables Eq. 1-5
In this notation, items that are in upper case and bold are vectors while members of
the vectors are designated using a lower case symbol with a subscript to indicate the
member.
The objective function is the scalar quantity to be minimized. It is a function of the set
of design variables. (Although we stated the problem as a minimization task, we can
easily maximize a function by minimizing its negative.) Side constraints are placed on
the design variables to limit the region of search, for example, to plate thicknesses that
are nonnegative or tubes whose wall thicknesses are less than one-tenth of the outer
radii. The inequality constraints are expressed in a less than or equal to zero form by
convention; that is, a constraint is satisfied if its value is negative. The location of the
fence lies at . Equality constraints, if present, must be satisfied exactly
at the optimal design.
The objective and constraint functions may either be linear or nonlinear functions of
the design variables. If all of these functions are linear, we may use linear techniques
to find an optimal solution if one exists. If just one of these functions is nonlinear, then
F X ( )
g
j
X ( ) 0 j 1 n
g
, , =
h
k
X ( ) 0 k 1 n
h
, , = =
i
L
x
i
x
i
U
i 1 n , , =
X x
1
x
2
x
n
, , , { } =
j -th g
j
X ( ) 0 =
Main Index Main Index
15 CHAPTER 1
Getting Started
search algorithms which can deal with this nonlinearity must be used. MSC.Nastran
includes capabilities for solving both linear and nonlinear optimization problems,
with nonlinear problems making up the vast majority of the prescribed tasks.
As seen throughout the remainder of this users guide, the basic optimization problem
statement is used directly in MSC.Nastran and influences the nomenclature adopted
here. For example, the design objective is defined in the Case Control Section by the
DESOBJ command, while the design variables and constraints are defined in the Bulk
Data Section using the DESVAR and DCONSTR entries respectively. We will see how
these are used in Developing the Design Model for Sensitivity and Optimization
in Chapter 3.
In conjunction with problems in structural optimization, we will discuss design
spaces, objectives, constraints, and design variables. It is useful to take a look at a
simple problem that is not explicitly related to structural optimization, just to become
familiar with these concepts.
Example
Consider the following optimization problem.
Find and to minimize the objective
Eq. 1-6
subject to the constraints
Eq. 1-7
Eq. 1-8
The objective and constraint functions are dependent on two design variables and
. The objective is linear in the design variables, while the constraint is
nonlinear. Figure 1-3 shows the two variable design space, where shading is used to
denote regions in which the constraint or side constraints on the design variables are
violated. If no constraints are violated, we say the current design is feasible (although
it is probably not optimal). A design is infeasible if one or more of the constraints are
violated. For example, the point (2,2) is a feasible design, whereas the point (0.05,3)
violates not only the inequality constraint but also the lower bound constraint on .
x
1
x
2
F X ( ) x
1
= x
2
+
g
1
X ( )
1
x
1
----- =
1
x
2
----- 2 + 0
x
1
0.1 x
2
0.1
x
1
x
2
F X ( ) g
1
X ( )
x
1
Main Index Main Index


16
Figure 1-3 Two-Variable Function Space
The optimal design at (1,1) can be found by inspection. The explicit description of the
design problem has allowed a graphical solution in two dimensions. In practice, we
usually have more than two design variables and non-explicit constraints and
objective function.
Numerically Searching for an Optimum
The optimization algorithms in MSC.Nastran belong to the family of methods
generally referred to as gradient-based, since, in addition to function values, they
use function gradients to assist in the numerical search for an optimum. (One
exception to the gradient based optimizers is the Fully Stressed Design on
page 135).
The numerical search process can be summarized as follows: for a given point in the
design space, we determine the gradients of the objective function and constraints and
use this information to determine a search direction. We then proceed in this direction
for as far as we can go, whereupon we investigate to see if we are at an optimum point.
If we are not, we repeat the process until we can make no further improvement in our
objective without violating any of the constraints.
0
1
2
3
4
1 2 3 4
F=4
3
1
2
g=0
x
1
x
2
Feasible Region
Optimum
x
1
0.1 =
x
2
0.1 =
Main Index Main Index
17 CHAPTER 1
Getting Started
Essentially this is the procedure used by the optimizer in MSC.Nastran, although the
task is complicated by the structural optimization context. Here, we consider each of
the aspects of this process in more detail, with an emphasis placed on the intuitive
aspects rather than a rigorous mathematical treatment.
The first step in a numerical search procedure is determining the direction to search.
The situation may be somewhat complicated if the current design is infeasible (one or
more violated constraints) or if one or more constraints are critical. For an infeasible
design, we are outside of one of the fences, to use the hill analogy. For a critical design,
we are standing right next to a fence. In general, we at least need to know the gradient
of our objective function and perhaps some of the constraint functions as well. The
process of taking small steps in each of the design variable directions (suppose we are
not restricted by the fences for this step) corresponds exactly to the mathematical
concept of a first-forward finite difference approximation of a derivative. For a single
independent variable the first-forward difference is given by
Eq. 1-9
where the quantity represents the small step taken in the direction x. For most
practical design tasks, we are usually concerned with a vector of design variables. The
resultant vector of partial derivatives, or gradient, of the function can be written as
Eq. 1-10
where each partial derivative is a single component of the dimensional vector.
Physically, the gradient vector points uphill, or in the direction of increasing objective
function. If we want to minimize the objective function, we will actually move in a
direction opposite to that of the gradient. The steepest descent algorithm searches in
the direction defined by the negative of the objective function gradient, or
Eq. 1-11
since proceeding in this direction reduces the function value most rapidly. S is
referred to as the search vector.
df x ( )
dx
-------------
f x x + ( ) f x ( )
x
------------------------------------------
x
F X ( )
F
x
1
---------
.

F
x
n
---------
)



`



F X x
1
+ ( ) F X ( )
x
1
----------------------------------------------------
.

F X x
n
+ ( ) F X ( )
x
n
----------------------------------------------------
)



`




=
S F =
Main Index Main Index


18
For now, just note that MSC.Nastran uses the steepest descent direction only when
none of the constraints are critical or violated and then only as the starting point for
other, more efficient search algorithms. The difficulty in practice stems from the fact
that, although the direction of steepest descent is usually a very good starting
direction, subsequent search directions often fail to improve the objective function
significantly. In MSC.Nastran we use other, more efficient methods which can be
generalized for the cases of active and/ or violated constraints. We will briefly
introduce these methods later in this section. The next question to consider is: once we
have determined a search direction, how can this be used to improve our design?
In the hill example, once we found a search direction, we proceeded downhill until
we bumped into a fence or until we reached the lowest point along our current path.
Note that this requires us to take a number of steps in this given direction, which is
equivalent to a number of function evaluations in numerical optimization. For a
search direction S and a vector of design variables X, the new design at the conclusion
of our search in this direction can be written as
Eq. 1-12
This relation allows us to update a potentially huge number of design variables by
varying the single parameter . We have been able to reduce the dimensionality from
n design variables to a single search parameter . For this reason, this process is called
a one-dimensional search. When we can no longer proceed in this search direction, we
have the value of which represents the move required to reach the best design
possible for this particular direction. This value is defined as . The new objective
and constraints can now be expressed as
Eq. 1-13
Eq. 1-14
From this new point in the design space, we can again compute the gradients and
establish another search direction based on this information. Again, we will proceed
in this new direction until no further improvement can be made, repeating the
process, if necessary.
At some point we will not be able to establish a search direction which can yield an
improved design. We may be at the bottom of the hill, or we may have proceeded as
far as possible without crossing over a fence. In the numerical search algorithm, it is
necessary to have some formal definition of an optimum. Any trial design can then be
X
1
X
0
=
*
S
1
+

*
F
1
F X
0

*
S
1
+ ( ) =
g
j
1
g
j
X
0

*
S
1
+ ( ) j 1 n
g
, , = =
Main Index Main Index
19 CHAPTER 1
Getting Started
measured against this criteria to see if it is met and an optimum has been found. This
required definition is provided by the Kuhn-Tucker conditions (see Numerical
Optimization in Appendix D), which are physically quite intuitive.
Figure 1-4 shows a two design variable space with constraints and and
objective function . The constraint boundaries are those curves for which the
constraint values are identically zero. A few contours of constant objective are shown
as well; these can be thought of as contour lines drawn along constant elevations of
the hill. The optimum point in this example is the point which lies at the intersection
of the two constraints. This location is shown as .
Figure 1-4 Kuhn-Tucker Condition at a Constrained Optimum
If we compute the gradients of the objective and the two active constraints at the
optimum, we see that they all point off roughly in different directions (remember that
function gradients point in the direction of increasing function values). For this
situation-a constrained optimum-the Kuhn-Tucker conditions state that the vector
sum of the objective and all active constraints must be equal to zero given an
appropriate choice of multiplying factors. These factors are called the Lagrange
multipliers. (Constraints which are not active at the proposed optimum are not
included in the vector summation.) Indeed, Figure 1-5 shows this to be the case where
and are the values of the Lagrange multipliers which enable the zero vector sum
condition to be met. We could probably convince ourselves that this condition could
not be met for any other point in the neighboring design space.
g
1
X ( ) g
2
X ( )
F X ( )
X*
x
2
g
2
X ( )
F X* ( )
g
1
X ( )
x
1
g
1
X* ( )
g
2
X* ( )
X*
F X ( ) constant =

1

2
Main Index Main Index


20
Figure 1-5 Graphical Interpretation of Kuhn-Tucker Conditions
The Kuhn-Tucker conditions are useful even if there are no active constraints at the
optimum. In this case, only the objective function gradient is considered, and this is
identically equal to zero; i.e., any finite move in any direction will not decrease the
objective function. A zero objective function gradient indicates a stationary condition.
Not only are the Kuhn-Tucker conditions useful in determining if we have achieved
an optimal design; they are also physically intuitive. The optimizer in MSC.Nastran
tests the Kuhn-Tucker conditions in connection with the search direction
determination algorithm. See Numerical Optimization in Appendix D for
theoretical details.
A Simple Structural Example
In this section we covered the basic optimization problem statement, the concept of a
design space, gradient-based search techniques, and the meaning of an optimum in
terms of satisfying the Kuhn-Tucker conditions. We pause here to take a look at a
simple example to help fit these pieces together and to introduce some qualitative
aspects of the optimizer used in MSC.Nastran.
The cantilever beam in Figure 1-6 has a rectangular cross section. Suppose we want to
minimize the volume, and thus the weight of the beam, subject to constraints on
maximum bending stress and deflection due to the tip loading. In addition, the beams
cross section must remain below a maximum beam height-to-width ratio to guard
against the introduction of twisting modes of failure.
F X* ( )

1
g
1
X* ( )
g
1
X* ( )
F X* ( )
g
2
X* ( )

2
g
2
X* ( )
Main Index Main Index
21 CHAPTER 1
Getting Started
Figure 1-6 Cantilever Beam
The optimization problem statement could be written as follows
minimize
Eq. 1-15
subject
Eq. 1-16
Eq. 1-17
Eq. 1-18
Eq. 1-19
Eq. 1-20
Since the objective and constraints are available explicitly, we can graphically display
the two-variable design space as shown in Figure 1-7.
A
H
A
P = 2250 N
L = 500 cm
Section A-A
B
V B = H L

Mc
I
--------
6PL
BH
2
----------- 700 = =

PL
3
3EI
----------
4PL
3
BH
3
------------- = = 2.54
H
B
---- 12
1 B 20
20 H 50
Main Index Main Index


22
Figure 1-7 Cantilever Beam Design Space
In Figure 1-7, note that the optimum lies at the vertex formed by the intersections of
the beam bending stress constraint and the constraint on maximum allowable ratio of
cross-sectional height to width. This optimum occurs at an objective of approximately
1,000 . Let us examine the path which the optimizer might take as it searches for
this constrained minimum by referring to Figure 1-8.
Assume that we begin with an initial design of H = 44 cm and B = 7 cm. Since none of
the constraints are active, the direction of steepest descent is used as an initial search
direction, and the optimizer proceeds in this direction until it encounters a constraint
boundary. From Figure 1-8 we see that the structural volume cannot be reduced any
further in this search direction without violating the maximum allowable tip
deflection requirement. The optimizer is now faced with a choice. A finite move in the
direction of steepest descent would not be admissible since constraints would then be
violated, yet we know that the objective function can still be reduced.
Optimum
Height H (cm)
65
60
55
50
45
40
35
2.5 3 4 5 6 7 7.5
Width B (cm)
50000
100000
125000
150000
175000
V = 200000
H/ B = 12
H = 50
75000
2.54 =

b
700 =
cm
3
Main Index Main Index
23 CHAPTER 1
Getting Started
Figure 1-8 Sequence of Iterations: Modified Method of Feasible Directions
The optimizer in MSC.Nastran resolves the situation by choosing a search vector that
effectively follows the active constraint boundary in the direction of decreasing
objective function. (If the optimizer could not find any direction in which to move, an
optimum would be at hand since the Kuhn-Tucker conditions have implicitly been
satisfied.) The objective can be reduced further and we observe that by the time two
such iterations have been completed, the true optimum has been reached.
Note that for both of these iterations, one or more constraints have been slightly
violated in the interim. This is a characteristic of the default optimizer used in
MSC.Nastran, the modified method of feasible directions, which establishes a search
direction tangent to the critical constraint(s). If the constraint is nonlinear, a finite
move in this direction may lead to a small constraint violation. Thus, continual
corrections must be made by stepping back toward the constraint boundary along the
current search direction. These corrections are performed as part of the search process,
and are qualitatively represented in Figure 1-8 as small steps back to the constraint
boundary.
Optimum
Height H (cm)
60
55
50
45
40
35
3 4 5 6 7
Width B (cm)
H/ B = 12
H = 50
2.54 =
700 =
X
Main Index Main Index


24
If the initial design is infeasible, the optimizers first task is to return to the feasible
region. Once this has been achieved, the optimizer can then proceed to minimize the
objective, if possible. Often simply finding a design in which none of the constraints
are violated is an engineering success. If no feasible designs are found, this still
provides useful information about the original design formulation, as one or more of
our performance criteria may need to be relaxed somewhat if we hope to produce a
feasible design. By reexamining our design goals, we may learn something about the
problem that was not evident before.
A critical but related issue involves the identification of various applicable modes of
failure. In developing a set of design criteria, we must ensure that all possible failure
modes are adequately addressed by the design specifications. This is true in all aspects
of engineering design but is especially so in design optimization. To use the current
example, if we did not specify a maximum allowable beam height to width ratio, we
might run the risk of introducing twisting or other buckling modes beyond the simple
bending stress criteria we had already accounted for. Without this constraint, the
optimizer would be able to reduce the structural volume even further (see the design
space of Figure 1-7), but would be completely unaware of the introduction of other
failure modes possible with narrow beam sections. Consequently, the engineer, not
the optimizer, must accept ultimate responsibility for the integrity of the final design.
We can also preview a concept that we will deal with in great detail in Fundamentals
of Design Sensitivity and Optimization in MSC.Nastran on page 33; optimizers
work best when they are dealing with linear problems. MSC.Nastran applies concepts
that make the structural response quantities a nearly linear function of the design
variables (see The Approximate Model on page 101). The constraint contained in
Eq. 1-18 is highly nonlinear the way it is written and the representation of Figure 1-7
suggests that this can be linearized as
Eq. 1-21
In practice, this linear specification of the constraint between the two design variables
is a much easier constraint to deal with than the non-linear one.
In summary, the intent of this example is to help illustrate the concepts presented in
this section as well as to give a general idea of the approach used by the modified
method of feasible directions. The discussion was simplified by the fact that we had
an explicit functional description of the design space beforehand, as well as only two
design variables. In any real structural optimization task, each of the data points in the
design space can only be determined based on the results of a complete finite element
H 12B 0
Main Index Main Index
25 CHAPTER 1
Getting Started
analysis. This may be quite expensive. Also, since a numerical optimizer usually needs
a number of these function evaluations throughout the search process, the costs
associated with this analysis can quickly become enormous.
These factors combined with tens or even hundreds of design variables and thousands
of constraints force us to consider methods for efficiently coupling structural analysis
routines with numerical optimizers. The field of structural optimization as
implemented in MSC.Nastran is based on the introduction of approximation
concepts, which reduce the need for repeated finite element function evaluations.
Approximation concepts are actually quite intuitive and are a topic of the next section.
Main Index Main Index


26
1.3 Structural Optimization
In Introduction on page 2 and Numerical Optimization Basics on page 11, we
introduced some examples of ways in which numerical optimization might be used to
solve design problems and gave a brief overview of gradient-based numerical
optimizers. This final introductory section focuses more closely on how the
optimization methods are coupled with a structural analysis in general and with the
MSC.Nastran product in particular. This is done by first introducing the concepts of
design properties and design responses that link the features of MSC.Nastran that are
familiar to the analyst to the design variables and constraints that are familiar to
optimizer. Following paragraphs address the difficulties encountered when trying to
make the optimization process feasible for real-world design tasks. Optimization
tools yield an orderly, rational approach to solving a minimization problem subject to
a set of constraints; however, a large number of function evaluations may need to be
performed. These function evaluations may be expensive, especially in a finite
element structural analysis context. Methods used in MSC.Nastran to reduce the
number of finite element analyses required are introduced, as are methods for
minimizing the effort involved in providing the information required during the
optimization.
Relating the Finite Element Analysis Model to the Design
Model
In analysis, our goal is to construct a mathematical idealization of an actual physical
system. We must determine analysis model properties, select appropriate response
quantities, and determine a suitable finite element mesh so that the desired responses
are computed within an acceptable degree of accuracy. The properties define
materials, element thicknesses, element meshes and other quantities needed to
perform the analysis. The results of the analyses depend on the type of analysis
performed, but include such things as element stresses and modal frequencies. In
design modeling, on the other hand, we wish to define how our model can change in
pursuit of an improved design along with the criteria which we (or the optimizer) use
to judge the effects of these changes. Numerical Optimization Basics on page 11
has presented the basics of the optimization task and uses words such as objective,
constraints and design variables. MSC.Nastran links these two models by introducing
two intermediate concepts. The first is the designed property and provides a link
between the design variable used by the optimizer and the finite element property
used in the finite element analysis. The second concept is the design response. A
design response is a physically based result that can be used as the design objective or
Main Index Main Index
27 CHAPTER 1
Getting Started
it can be used, with user imposed limits, as a design constraint. The presence of this
intermediate layer between the design quantities and the analysis quantities is a key
feature of Design Sensitivity and Optimization that adds to the flexibility of the
capability. Design properties and design responses are treated in detail in
Developing the Design Model for Sensitivity and Optimization on page 151.
Managing the Structural Optimization Task
Historically, the first attempts at linking structural analysis with numerical
optimization were based on a direct coupling or black box type of approach as seen
in Figure 1-9. Whenever the optimizer needed a function evaluation, the finite
element analysis would be invoked to provide the necessary information. One must
also remember that these methods were being developed during the infancy of
computerized structural analysis when a finite element analysis that we now consider
trivial consumed a measurable amount of computer resources. Therefore, the sheer
number of analyses quickly tended to make this approach useless in all but the
smallest of problems. Recall that the numerical optimizer may not only require
response derivatives with respect to the design variables, but a number of function
evaluations must also be performed during each of the one-dimensional searches.
This situation could quite easily lead to hundreds of analyses. MSC.Nastran therefore
employs concepts that limit the number of required finite element analyses. It is
remarked that some commercial structural analysis codes that include an
optimization capability still perform their optimization tasks in the black box
manner of Figure 1-9. Although they benefit from the high performance capabilities
of present day computers, they are still limited in the size of the design task (in terms
of number of design variables and responses and the cost of finite element analysis)
they can address relative to what can be achieved using MSC.Nastran.
Figure 1-9 Early Structural Optimization Attempts
The principal complicating factor in structural optimization is that the response
quantities of interest are usually implicit functions of the design variables. For
example, a plate elements stress variation with changing thickness can only be
determined in the general case by performing a finite element analysis of the
Numerical
Optimization
Finite Element
Analysis
Main Index Main Index


28
structure. When we consider that the optimizer may be asked to deal with perhaps
hundreds of design variables and thousands of constraints, it becomes apparent that
we do not have the luxury of invoking a full finite element analysis each time the
optimizer proposes an incremental design change.
To avoid these difficulties, certain approximations can be implemented to reduce the
computational overhead. The remainder of this section will introduce each of these
methods, all of which are available in MSC.Nastran.
Overview of Approximation Concepts Used in Structural
Optimization
Approximation concepts are actually nothing more than the computational
implementation of methods generally used by experienced design engineers. In many
instances, an engineer is handed a stack of analysis data and asked to propose an
improved design. This raw data usually contains much more information than is
necessary to suggest possible design improvements. The question becomes one of
how to reduce the problem sufficiently so that only the most pertinent information is
considered in the process of generating a better design.
A first step may be to narrow the design task to that of determining the best
combination of just a few design variables. There is virtually no way for a designer to
consider fifty or one hundred variables simultaneously and expect to find a suitable
combination from the group. It is much more efficient to link these together if possible.
That is, it would be advantageous if all the design variables could be varied in a
suitably proportional manner according to changes made to a much smaller set of
independent variables. Describing a shape defining polynomial surface in terms of
just a few characteristic parameters, or allowing only linear variations of plate element
thicknesses, are both examples of types of design variable linking.
The next step might then be to identify the few constraints that are violated or nearly
violated. There may be just a few constraints which are currently guiding the design,
such as stress concentrations due to thermal loads, while others, such as the first
bending mode, may be nowhere near critical and can be temporarily disregarded. In
structural optimization, this is a constraint deletion process. Constraint deletion
allows the optimizer to consider a reduced set of constraints, simplifying the
numerical optimization phase. This also reduces the computational effort associated
with determining the required structural response derivatives (sensitivity analysis
costs are reduced as well).
Main Index Main Index
29 CHAPTER 1
Getting Started
Once the engineer has determined the constraint set that seems to be driving the
design, the next step might be to perform some sort of parametric analysis in order to
determine how these constraints vary as the design is modified. The results of just a
few analyses might be used to propose a design change based on a compromise
among the various trial designs.
A parametric study of the problem is carried one step further in structural
optimization with formal approximations, or series expansions of response quantities
in terms of the design variables. Formal approximations make use of the results of the
sensitivity analysis to construct an approximation to the true design space which,
although only locally valid, is explicit in the design variables. The resultant explicit
representation can then be used by the optimizer whenever function or gradient
evaluations are required instead of the costly, implicit finite element structural
analysis.
This coupling is illustrated in Figure 1-10. The finite element analysis forms the basis
for creation of the approximate model which is subsequently used by the optimizer.
The approximate model includes the effects of design variable linking, constraint
deletion, and formal approximations. Design variable linking is established by the
engineer, while constraint deletion and formal approximations are performed
automatically in MSC.Nastran.
Figure 1-10 Coupling Analysis and Optimization Using Approximations
Once a new design has been proposed by the optimizer (based on the information
supplied by the approximate model) the next step would most likely be to perform a
detailed analysis of the new configuration to see if it has actually managed to satisfy
the various design constraints and reduce the objective function. This reanalysis
update of the proposed designs is represented by the upper segment of the loop in
Figure 1-10. If a subsequent approximate optimization is deemed necessary, the finite
Finite Element
Analysis
Numerical
Optimization
Approximate
Design Model
Design Improvements
Main Index Main Index


30
element analysis serves as the new baseline from which to construct another
approximate subproblem. This cycle may be repeated as necessary until convergence
is achieved. These loops are referred to as design cycles.
Main Index Main Index
31 CHAPTER 1
Getting Started
1.4 Summary
By now you should have a fairly good idea of some of the various applications of
design optimization as well as some of the numerical optimization capabilities
available to assist you in your design tasks. We introduced the basic ideas of
numerical optimization and some of the methods that are used to couple it with
structural analysis codes. Central to this has been the recognition that it is impractical
to directly link numerical optimizers and structural analysis codes to create a
structural optimizer, and that some sort of approximations are required.
It has also been shown that the set of approximation concepts acts as an interface
between the analysis and the optimizer. There is really nothing extraordinary about
these approximations since in many ways they are similar to the methods experienced
designers already use. Of course, formalizing these methods in structural
optimization allows problems of considerably greater size and complexity to be
solved than with traditional design methods alone.
Main Index Main Index


32
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide/
CHAPTER
2
Fundamentals of Design Sensitivity and
Optimization in MSC.Nastran
I Overview of Fundamentals
I The Design Model
I Multidisciplinary Analysis
I Constraint Screening
I Design Sensitivity Analysis
I Optimization
I The Approximate Model
I Tests for Convergence
I Special Topics
Main Index Main Index


34
2.1 Overview of Fundamentals
Chapter 1 presented a Getting Started overview of design optimization in general and
introduced a number of concepts that are specific to MSC.Nastrans implementation
of Design Sensitivity and Optimization. This chapter goes into greater depth on the
MSC.Nastran methodology, again in terms of concepts and equations. Mention of
user interface features (that is, the case control commands and bulk data entries) is
sometimes unavoidable, but the intent is to defer a description of the interface to
Developing the Design Model for Sensitivity and Optimization in Chapter 3.
This chapter has been prepared for several classes of readers:
New users who want to obtain an overview of features of MSC.Nastrans
implementation of Design Sensitivity and Analysis.
More experienced users who will use this chapter as reference material to
clarify details of the features and to gain insight into the algorithms
employed.
Support engineers and software developers who need to be able to explain,
and possibly extend, the capability.
It is recognized that this covers a broad spectrum of readers and you probably will not
read this chapter from beginning to end. Instead, you should skim the material until
you find the topic that interests you and return to the chapter as you gain more
experience and want to obtain a deeper level understanding.
The majority of this chapter can be thought of as a description of the flow chart shown
in Figure 2-1, which is itself an expanded version of the sketch of Figure 1-10. The
figure shows a design loop with a number of blocks. The blocks, and the section in
which they are described, is as follows:
Main Index Main Index
35 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Figure 2-1 MSC.Nastran Implementation of Structural Optimization
Initial Design
The initial design is a combination of the analysis model familiar to MSC.Nastran
users and the design model, which is a unique feature of the Design Sensitivity and
Optimization capability. The design model is discussed in The Design Model on
page 39.
Structural Analysis
This is the analysis function that is, again, familiar to the MSC.Nastran user. However,
for design sensitivity and optimization, it is frequently necessary to perform multiple
types of analyses. This reflects the fact that the design responses can be created by a
number of analysis types and it is necessary for the optimizer to synthesize these
results when performing its redesign. Multidisciplinary Analysis on page 58
discusses this multidisciplinary feature.
Initial
Design
Improved
Design
Structural
Analysis
Constraint
Screening
Sensitivity
Analysis
Approximate
Model
Optimizer
Finite Element
Analysis
Many Times
Converged
Stop
(One time around this loop is referred to as
a design cycle or design iteration.)
Y
N
Main Index Main Index


36
Constraint Screening
This concept was introduced in Structural Optimization on page 26 and refers to
the process that is used to identify those constraints that are likely to drive the
redesign process. A detailed discussion on this is presented in Constraint
Screening on page 64.
Sensitivity Analysis
Accurate sensitivity analyses are the hallmark of the MSC.Nastran implementation of
design sensitivity and optimization. The individual analysis types each have their
own techniques for performing sensitivity analysis and these are developed in
Design Sensitivity Analysis on page 67.
Optimizer
MSC.Nastran uses a variety of optimization algorithms with the DOT optimization
algorithms from VR&D (see Reference 15.) serve as the workhorse set for the releases
prior to 2005 r3. Starting with the MSC.Nastran 2005 r3 release, the MSCADS suite of
optimization algorithms are the default with DOT available by user selection. The
ADS suite of optimization algorithms are available in MSC.Nastran 2005. These
algorithms were adapted from the public domain version of ADS codes described in
Reference 11. Appendix C presents to mathematical underpinnings of the DOT
algorithm while Optimization on page 97 discussed how the optimizers interact
with the approximate model when analysis and sensitivity results are required.
The BIGDOT algorithm, also provided by VR&D,is a special purpose optimization
algorithm to address design tasks with a large number (> 3-4,000) of design variables.
This algorithm is only available from MSC when the Topology Optimization module
of MSC.Nastran is acquired and has its primary application to topology optimization
tasks. If the BIGDOT algorithm is to be applied to the conventional design tasks that
are discussed in this volume, then both the Design Optimization and the Topology
Optimization modules must be available.
Approximate Model
Structural Optimization on page 26 has also introduced this concept that involves
the construction of high-quality approximations to the finite element results so that
the number of full scale finite element analyses is kept to a minimum. The
Approximate Model on page 101 documents the approximation concepts used in
MSC.Nastran.
Main Index Main Index
37 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The Improved Design
This is the point at which the finite element model is updated based on the results
from the optimizer so that a new finite element analysis can occur. This is a
straightforward operation and its discussion is included in the Converged write-up of
Tests for Convergence on page 125.
Converged
As Figure 2-1 indicates, Design Optimization is an iterative process and a key part of
the implementation is therefore determining when to stop the iterations. Tests for
Convergence on page 125 discusses the many factors that enter into making this
decision.
The chapter concludes with a section called Special Topics on page 133 that
contains fundamental descriptions of several other topics that are outside those of
Figure 2-1. These include:
Discrete Variable Optimization. An underlying assumption of the Design
Sensitivity and Optimization capability of MSC.Nastran is that the variations in the
design variables are continuous and that therefore the responses and their sensitivities
are also continuous. Practical engineering considerations frequently dictate that
values of the designed properties be chosen from a discrete set. The first special topic
discusses how discrete variables are handled in MSC.Nastran as a postprocessing task
to the continuous optimization process.
Fully Stressed Design. Fully Stressed Design is a rapid redesign method that is
quite distinct from the mathematical programming techniques that are the main topic
of this users guide. The basics of this methods are discussed and it is shown how
MSCs implementation works together with the mathematical programming solution.
Topology Optimization. Topology optimization is a special version of design
optimization that finds an optimal distribution of material, given package space, loads
and boundary conditions. Basically, it makes a design variable out of each finite
element that can vary from 0 (remove) to 1 (keep) and the algorithm strives to force
real design variables to one of these limits.
Distributed Memory Parallel. Distributed Memory Parallel (DMP) algorithms allow
for splitting up a given task to run on multiple processors. This section discusses the
DMP options that are available in design optimization.
Main Index Main Index


38
Superelements. Superelements provide a powerful analysis tool. Special
considerations and limitations when using superelements with design sensitivity and
optimization are discussed.
Main Index Main Index
39 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
2.2 The Design Model
Numerical Optimization Basics on page 11 introduced the general concepts of
design variables, constraints, objective function, and side constraints. Structural
Optimization on page 26 presented further concepts of designed properties and
design responses and indicated that these intermediate quantities are used to fill the
gap between standard finite element analysis and optimization procedures. This
section of the guide indicates how these concepts are implemented in MSC.Nastran.
Design Variables
Design Variables are quantities that are known to the optimizer and that can be
directly changed to satisfy the optimization statement. In MSC.Nastran, the design
variables are defined on the DESVAR entry (see Defining the Design Variables on
page 158 and Bulk Data Entries on page 216) and can only affect the finite element
analysis if they are used with designed properties and/ or design responses. It is the
users responsibility to define the initial values of these design variables and,
optionally, upper and lower bounds on the values of these variables. The bounds
make up the side constraints (See Eq. 1-4) of the optimization problem statement
and these bounds can never be exceeded at any time.
MSC.Nastran distinguishes between dependent and independent design variables.
Dependent design variables are linked (see the entry, DLINK on page 230) to the
independent variables using a relationship of the form
Eq. 2-1
MSC.Nastran also supports the concept of discrete design variables wherein the
value of the design variable is restricted to a user specified set of real numbers. This
topic is discussed in more detail in Discrete Variable Optimization on page 133.
Designed Properties
Designed properties are quantities that do have a direct bearing on the finite element
analysis and include such things as element thicknesses, material properties, mass
values, and grid locations. Designed grid locations, which are used in shape
optimization, are defined in a way that is distinct from the remaining properties.
Therefore, the link between grid locations and design variables is presented
x
j
D
c
0
= c
i
x
i
I
i 1 =
ndv
i

+
Main Index Main Index


40
separately in Designed Shapes on page 47 that follows immediately after this
discussion which deals with element properties, material properties and connectivity
properties.
The first order of business is to explicitly define what these properties are. Element
properties refer to quantities defined in MSC.Nastran using element property bulk
data entries (those that begin with the letter P). Table 2-1 contains a list of the bulk
data entries that have element properties that can be designed and identifies which of
the fields on the entry can be designed. In general, real numbers that appear on an
element property bulk data can be designed.
Main Index Main Index
41 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Table 2-1 Element Properties Available as Designed Properties
Property
Entry
Property
PACABS TESTAR, CUTFR, B, K, M
PACBAR MBACK, MSEPTM, FRESON, KRESON
PBAR A, I1, I2, J, NSM, C1, C2, D1, D2, E1, D2, F1, F2, K1, K2, I12
PBARL DIMi, NSM
PBEAM (A(i), I1(i), I2(i), I12(i), J(i), NSM(i), C1(i), C2(i), D1(I), D2(i), E1(i),
E2(i), F1(i), F2(i), i = A, B, 1 ... 9), K1, K2, S1, S2, (NSI(j), CW(j), M1(j),
M2(j), N1(j), N2(j), j = A, B)
PBEAML (DIMi(j), NSM(j), j = A, B, 1 ... 9)
PBUSH (Ki, Bi, GEi, i = 1, 6), SA, ST, EA, ET
PBUSH1D K, C, M, SA, SE (only linear elements are supported for
optimization)
PCOMP Z0, NSM, SB, TREF, GE, Ti, THETAi
PCONEAX T1, HI, T2, NSM, Z1, Z2, (PHIj; i = 1, 14)
PDAMP B1, B2, B3, B4
PELAS K1, GE1, S1, K2, GE2, S2
PGAP U0, F0, KA, KB, KT, MU1, MU2
PMASS M1, M2, M3, M4
PROD A, J, C, NSM
PSHEAR T, NSM, F1, F2
PSHELL T, 12I/ T**3, TS/ T, NSM, Z1, Z2 (The 12I/ T**3 term can be designed
but must be referenced by Field ID = 6 rather than by name.)
PTUBE OD, T, NSM
PVISC CE1, CR1, CE2, CR2
Main Index Main Index


42
Table 2-2 contains a list of the bulk data entries for material properties and again
identifies which fields on these entries can be designed. Table 2-3 performs the same
function for connectivity entries (bulk data entries that begin with the letter C). It is
seen that this enables the support of properties such as beam offset vectors,
concentrated mass values, and tapered shell elements.
Main Index Main Index
43 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Table 2-2 Material Properties Available as Designed Properties
Material
Entry
Material Properties
MAT1 E, G, NU, RHO, A, TREF, GE
MAT2 G11, G12, G13, G22, G23, G33, RHO, A1, A2, A3, TREF, GE
MAT3 EX, ETH, EZ, NUXTH, NUZTH, NUZX, RHO, GZX, AX, ATH, AZ,
TREF, GE
MAT8 E1, E2, NU12, G12, G1Z, G2Z, RHO, A1, A2, TREF, Xt, Xc, Yt, Yc, S,
GE
MAT9 G11, G12, G13, G14, G15, G16, G22, G23, G24, G25, G26, G33, G34,
G35, G36, G44, G45, G46, G55, G56, G66, RHO, A1, A2, A3, A4, A5,
A6, TREF, GE
MAT10 BULK, RHO, C, GE
Table 2-3 Connectivity Properties Available as Designed Properties
Connectivity
Entries
Connectivity Properties
CBAR X1, X2, X3, W1A, W2A, W3A, W1B, W2B, W3B
CBEAM X1, X2, X3, BIT, W1A, W2A, W3A, W1B, W2B, W3B
CBUSH X1, X2, X3, S, S1, S2, S3
CDAMP2,4 B
CELAS2 K, GE, S
CELAS4 K
CGAP X1, X2, X3
CMASS2,4 M
CONM1 M11, M21, M22, M31, M32, M33, M41, M42, M43, M44, M51,
M52, M53, M54, M55, M61, M62, M63, M64, M65, M66
CONM2 M, X1, X2, X3, I11, I21, I22, I31, I32, I33
CONROD A, J, C, NSM
Main Index Main Index


44
Type-1 Properties
Two methods are provided to relate the properties of Table 2-1, Table 2-2, and
Table 2-3 to the DESVAR values. The first, which is designated type-1, specifies a
linear relationship which can be expressed as
Eq. 2-2
where is the j-th property, expressed as a linear combination of n design variables
plus an optional constant value. In a typical application, the property is a function of
a single design variable, but Eq. 2-1 can be used to express a large number of
properties in terms of a much small set of design variables. This is shown in the
following example.
Example
Assume the plate thickness distribution of the simple structure shown in Figure 2-2 is
to be determined using a combination of constant, linear, and quadratic basis
functions as shown in Figure 2-3. The design variables act as multipliers of these basis
functions which are, in turn, used to specify the element plate thicknesses.
CQUAD4,8,R THETA, ZOFFS, T1, T2, T3, T4
CTRIA3,5,R THETA, ZOFFS, T1, T2, T3
CTRIAX6 TH
Table 2-3 Connectivity Properties Available as Designed Properties (continued)
Connectivity
Entries
Connectivity Properties
p
j
c
o
= c
i
i

+ x
i
p
j
Main Index Main Index
45 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Figure 2-2 Plate Thickness Distribution
Figure 2-3 Basis Functions
The plate thicknesses are to vary in the x-axis direction, but are to remain constant in
the y-direction. Evaluating the basis functions for each of these ten thickness stations
and writing these design variable-to-thickness relations in matrix form, we get
x 0 =
t
1
t
2
t
10
x L =
y
1.0
1.0
0.5
F x L ( )
F
1
F
2
F
3
x L
F
1
x
L
---
\ .
| |
1.0 =
F
2
x
L
---
\ .
| |
c
1
1
x
L
---
\ .
| |
=
F
3
x
L
---
\ .
| |
c
1
1
x
L
---
\ .
| |
2
=
Main Index Main Index


46
Eq. 2-3
The thicknesses of the ten element groups are controlled using only three design
variables. Each of the matrix columns correspond to constant, linear, and quadratic
basis functions, respectively. Relations of this sort are termed reduced basis
formulations.
In general, a reduced basis formulation can be expressed as
Eq. 2-4
The most powerful form of Eq. 2-4 occurs when the number of rows of is much
greater than the number of columns or . Cantilevered Plate on page 401
presents an example that uses this concept in the design of a cantilevered plate.
Type-2 Properties
The second method, designated type-2, utilizes a general equation utility to define the
relationship between design variables and designed properties
Eq. 2-5
where the j-th property is a function of a collection of design variables and constants
as indicated by the bold notation.
t
1
t
2
t
3
t
4
t
5
t
6
t
7
t
8
t
9
t
10
)









`










1.0 1.0 1.0
1.0 0.9 0.81
1.0 0.8 064
1.0 0.7 0.49
1.0 0.6 0.36
1.0 0.5 0.25
1.0 0.4 0.16
1.0 0.3 0.09
1.0 0.2 0.04
1.0 0.1 0.01
x
1
x
2
x
3
)


`



=
p { }
M
T [ ]
MXN
x { }
N
=
T [ ]
M N
p
j
f X C , ( ) =
Main Index Main Index
47 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
A simple example of the application of a type-2 relation is in the design of beam cross-
sectional properties. For a rectangular cross section, the logical design variables would
be the width and height of the beam but MSC.Nastran expects cross-sectional
properties such as area and moment of inertia. If represents the width and
represents the height, these properties can be written using equations
Eq. 2-6
It should be noted that the beam library feature of MSC.Nastran permits the direct
design of the PBARL and PBEAML dimensions. This is a much simpler user interface
and is to be preferred if the beam library supports the cross section type of interest.
Type-3 Properties
The beam library does introduce a third type of designed property that is somewhat
hidden from the user. The user has direct access to the beam dimension in terms of
designed properties. The actual beam properties, such as area and moment of inertia,
that are used in the development of the element stiffness matrices, are functions of
these dimensions. When a beam property is designed indirectly because it is a
function of a beam dimension, the designed beam property is referred to as a
spawned design property. The relation between the spawned property and the
design variable can be linear (stress recovery points are typically a linear function of
the beam dimensions) or nonlinear (inertia properties are typically a nonlinear
function of the beam dimensions). MSC.Nastran makes a special effort to determine
the nature of this relationship because the sensitivity calculations are much simpler
when the relationship is linear. Spawned beam properties that are spawned using
nonlinear relations are referred to subsequently as type-3 properties.
Designed Shapes
Shape sensitivity and optimization in MSC.Nastran requires the definition of design
variables and a relationship between these variables and the allowable shape
variations. The shape variations are similar to the basis vectors of the previous section
in that a single shape typically affects multiple grid locations. The amount the design
variable is changed during optimization results in a corresponding shape change.
x1 x2
A x1 x2 =
I
x1 x2
3

12
--------------------- =
Main Index Main Index


48
The allowable shape changes are defined using shape basis vectors. The optimizer
then determines how much the structure can change by modifying the design
variables.This section begins with a brief theoretical discussion of shape basis vectors,
followed by an introduction to auxiliary models, which can be used in MSC.Nastran
as an aid to shape basis vector generation. Following these discussions is an overview
of the various modeling methods available as well as an example highlighting some
of the differences among these various approaches.
Basis Vectors in Shape Optimization
Shape basis vectors are used in MSC.Nastran to characterize the allowable shape
changes. The optimizer then determines the best linear combination of these vectors,
given the design criteria established by the engineer.
The angle bracket redesign of Figure 2-4 can be used to illustrate the concepts of shape
basis vectors.
Figure 2-4 Angle Bracket
Suppose the design goal is to vary the lengths of the x- and y-axis legs of the bracket
while keeping the outer edge straight. The location of grid points 2 and 3 can easily be
described in terms of the bracket leg lengths and . For simplicity, assuming
only four grids along this edge, we can write
Eq. 2-7
y
x
1
2
3
4
G
1y
G
4x
G
2x
G
2y
G
3x
G
3y
)



`




0 1 3
2 3 0
0 2 3
1 3 0
G
1y
G
4x
)

`


=
Main Index Main Index
49 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
A more realistic model certainly would include more grid points along this edge in
addition to a number of grids in the structures interior. Adding these grids would
simply add more equations to the set in Eq. 2-7, leaving its basic form unchanged.
We can write Eq. 2-7 in terms of grid variations as
Eq. 2-8
The coefficient matrix is unchanged in Eq. 2-8 from that of Eq. 2-7.
Eq. 2-8 suggests that a convenient choice of design variables would be
Eq. 2-9
or
Eq. 2-10
Note that Eq. 2-10 expresses a vector of grid coordinate changes in terms of a vector
of design variable changes. Each column of the matrix on the right-hand side of the
equation is called a shape basis vector. Generating the shape basis vectors is one of the
design engineers primary tasks in shape optimization.
If we graph each column of the matrix in Eq. 2-10 as a set of grid displacements, we
see that column 1 represents a vertical (or y-component) variation in shape, while
column 2 represents a horizontal shape variation. (The column order is arbitrary and
G
2x
G
2y
G
3x
G
3y
)



`




0 1 3
2 3 0
0 2 3
1 3 0
G
1y
G
4x
)

`


=
x
1
G
1y
=
x
2
G
4x
=
G
1y
G
2x
G
2y
G
3x
G
3y
G
4x
)





`






1 0
0 1 3
2 3 0
0 2 3
1 3 0
0 1
x
1
x
2
)

`


=
Main Index Main Index


50
just depends on the order of the design variables.) Note that both shape basis vectors,
as well as any linear combination of them, keep the edge of the bracket as a straight
line, consistent with our design goals.
Figure 2-5 Basis Vectors for the Angle Bracket
Once a shape basis vector description is available, the optimizer can vary the shape of
the structure by changing the corresponding design variables.
We can generalize the relation given in Eq. 2-10 as
Eq. 2-11
where:
and i = design cycle number.
As the design variables change, the grid locations are computed as
Eq. 2-12
Eq. 2-11 expresses a vector of grid point changes as a function of changes in design
variables. The linear combination of design variable changes times each shape basis
vector results in the total change in shape. Typically, there are more grid points than
design variables, or . For this reason, Eq. 2-11 is often called a reduced basis
formulation.
4
1
y
2
3
x
1
y
x
2
3
4
[T], Column 2 [T], Column 1
G
1y
G
2y
G
3y
G
2x
G
3x
G
4x
G { }
mx1
T [ ]
mxn
x { }
nx1
=
G { } G { }
i 1 +
G { }
i
=
x { } x { }
i 1 +
= x { }
i

G { }
new
G { }
o
T [ ] x
new
x
o
{ } + =
m n
Main Index Main Index
51 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
It needs to be reemphasized that the designed shape is the perturbation of the original
shape defined by the GRID locations. The perturbation, as expressed in Eq. 2-11, is the
product of the shape basis vector and the change in the design variable value. This
contrasts with Designed Properties on page 39 of the previous subsection where
the designed property is determined based on the specification of Eq. 2-2 or Eq. 2-5
and the input property from the analysis model is ignored.
Input of Shape Basis Vectors
MSC.Nastran supports four methods to describe shape basis vectors:
1. Manual grid variation
2. Direct input of shapes
3. Geometric boundary shapes
4. Analytic boundary shapes
An overview of these methods is given here with details of the user interfaced
deferred until Relating Design Variables to Shape Changes on page 166.
Manual Grid Variation. This is the most general, as well as the most tedious, method
of generating shape basis vectors. A DVGRID Bulk Data entry defines the direction
and magnitude of a grid variation for a given change in a design variable. This relation
is shown in Figure 2-6 and Eq. 2-13. A single DVGRID entry is required for every
design variable-grid pair.
Figure 2-6 Grid Point Variation Defined by a DVGRID Entry
x
z
y
Z
Y
X
G { }
i
G { }
i
Main Index Main Index


52
Eq. 2-13
Direct Input of Shapes. This approach greatly simplifies the process of defining
shape basis vectors. With this method, externally-generated vectors are DBLOCATEd
and used to define shape basis vectors. These externally generated vectors are solution
vectors from an auxiliary model that shares the geometry (grid numbers and
locations) with the primary designed model but the boundary conditions and
analyses are chosen to produce deformations (or mode shapes) that are candidate
shape changes. Relating Design Variables to Shape Changes on page 166 details
the steps required to use this method while Shape Optimization of a Culvert on
page 419 provides an example application of this technique.
Geometric Boundary Shapes. This method is similar to the Manual Grid Variation
method except that the user is required to define the allowable shape variations only
on the boundary of the structure. In addition to supplying the DVGRID information
on the boundary, it is also necessary to supply BNDGRD Bulk Data entries to define
the structures boundaries. With this method, the shape basis vectors for the entire
structure are automatically generated by the code through a process of interpolation
of the boundary shape changes to the interior of the structure. An advantage of this
method is that this interpolation is updated on every design cycle, minimizing the
problems associated with mesh distortion for large shape changes.
Analytic Boundary Shapes. This approach extends the geometric boundary shapes
method to avoid having to supply DVGRID entries on the boundary. Instead, the
designer provides an auxiliary model over the boundaries of the structure that are to
be changed. This may be a collection of BAR elements along the edges of a two-
dimensional structure or a skin of plate elements over a three-dimensional part. We
refer to this model over the boundaries as an auxiliary boundary model. The designer
then constrains and statically loads the auxiliary model to produce a shape variation
over the boundary, which the code then interpolates to the interior of the structure.
The result is a shape basis vector for optimization. Unlike the Direct Input of Shapes
method, the auxiliary model is included in the same input file as the primary model
and the analysis of the auxiliary model feeds the shape design task directly. Relating
G
x
G
y
G
z
)


`



i
COEFF
N1
N2
N3
)

`


\ .
|
|
|
| |
i j
x
j
=
Main Index Main Index
53 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Design Variables to Shape Changes on page 166 details the steps required to use
this method while Analytic Boundary Shapes on page 430 provides an example
application of this technique.
Design Responses
Design responses are used in MSC.Nastran as the basis for defining the design
objective and design constraints. In a manner similar to the designed properties, they
are an intermediate layer that serves to connect the MSC.Nastran analysis with the
Basic Optimization Problem Statement. Designating the Design Responses on
page 188 goes into the specifics of the available response quantities so that this
discussion simply introduces the three types of responses.
Type-1 Responses
The first type of response is entitled type-1 or first-level responses. These responses
are available directly from an MSC.Nastran analysis. Structural weight,
displacements at grid points, element stresses, and so on, are all examples of type-1
responses. The DRESP1 Bulk Data entry is used to select this type of response and is
discussed at length in Bulk Data Entries on page 216.
Type-2 Responses
The second type is entitled type-2, or second-level responses. This class of responses
is also frequently called user-defined since they utilize the equation input feature in
MSC.Nastran. This response allows the user to develop a response of the form
Eq. 2-14
where the and vectors are design variables, constants, type-1
responses, designed properties, type-2 responses and grid locations, respectively. This
is an extremely powerful feature of the MSC.Nastran Design Sensitivity and
Optimization capability that allows users to synthesize a wide variety of design
conditions that are not available from the MSC.Nastran analysis directly. Example
applications of type-2 responses are to formulate stress failure criteria not available in
MSC.Nastran, to evaluate local buckling behavior, to develop root mean square
responses and, in shape optimization, to impose limits that prevent the geometry from
becoming physically meaningless (e.g., to make sure a hole radius does not extend
beyond the boundary of the plate it is contained in). The DRESP2 Bulk Data entry is
used to develop this type of response.
r
j
2
f X C R1, P, R2, XYZ , , ( ) =
X C R1, P R2 , , , XYZ
Main Index Main Index


54
Type-3 Responses
The final type of response is called type-3 or third level responses. This can be
regarded as a major extension of the type-2 response that creates a response using a
representation similar to that of Eq. 2-14
Eq. 2-15
The evaluation of the type-3 response is performed using an API (Application
Programming Interface) that invokes a process running externally to MSC.Nastran.
The additional string argument in the type-3 formulation permits the passing of
general information that can be used to direct the execution of the external process.
For example, it could identify an input file needed by the external evaluator or it could
be a simple flag that directs the execution of the external evaluator in some way. It is
seen that the type-3 response provides a completely general way of synthesizing
response quantities that can drive the design but that are not MSC.Nastran responses
and are too complex to be evaluated using the type-2 formulation. Examples are
limited only by the imagination and cleverness of the user, and include proprietary
buckling criteria, cost models, detailed component analysis and other analyses not
performed in MSC.Nastran. The DRESP3 Bulk Data entry is used to develop this type
of response.
Design Objective
The objective function is a scalar quantity that is either minimized or maximized by
the optimizer. The design objective is selected using the DESOBJ (DESign OBJective)
Case Control command. This command points to a design response defined on either
a DRESP1, DRESP2 or DRESP3 Bulk Data entry which must define a single scalar
response.
Design Constraints
Design constraints are invoked by using the DCONSTR entry to identify a design
response (DRESP1, DRESP2 or DRESP3) and to impose limits on the response
Eq. 2-16
r
j
3
f X C R1 P R2 XYZ string , , , , , , ( ) =
r
j
L
r
j
X ( ) r
j
U

Main Index Main Index
55 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
where is the lower bound on the j-th response and is the corresponding upper
bound. Unlike the design objective, constraints can be applied to responses that have
multiple values. For example, a stress response that invokes a particular property ID
will create a separate response value for every finite element that references that
property ID.
Normalized Constraints
The inequalities of Eq. 2-16 are converted to normalized constraints that satisfy the
convention of Eq. 1-2 in which satisfied constraints are negative
Eq. 2-17
where
Eq. 2-18
where GSCAL is an MSC.Nastran parameter with a default of 0.001.
The normalized constraints of Eq. 2-17 are especially useful since the dependence on
the magnitude of the response quantity has been removed. A constraint having a
value of +1 represents a 100% violation irrespective of the magnitude of the response
or whether it is an upper or lower bound. It will be seen that this normalization
enables the constraint screening discussed in Constraint Screening on page 64.
Since response bounds are used as normalizing factors, it is recommended that the
bounds of zero be avoided. To avoid a divide by zero error in the code, the small
nonzero number GSCAL is used. This may or may not be acceptable from a design
standpoint. For example, consider the synthetic response
Eq. 2-19
and assume the response had been formulated using a type-2 response as
r
j
L
r
j
U
g
2j 1
X ( )
r
j
L
r
j
X ( )
GNORM
-------------------------- = 0
g
2j
X ( )
r
j
X ( ) r
j
U

GNORM
--------------------------- = 0
GNORM
LALLOW for lower bounds if LALLOW GSCAL >
UALLOW for upper bounds if UALLOW GSCAL >
GSCAL otherwise

1

2
+
2
--------------------
max

Main Index Main Index

56
Eq. 2-20
At first glance, this seems acceptable since it is expressed in standard form. However,
the upper bound normalization uses GSCAL and the resultant normalized upper-
bound constraint then becomes
Eq. 2-21
To understand why this is not a good formulation, lets assume is 10000. and the
responses exceeds this by 1% to equal 10100. With the default value of
GSCAL, Eq. 2-21 gives a constraint value of 100,000.(!), implying a grossly violated
constraint. Further, a small change in the design could result in
and now the constraint is -100,000. The optimizer will have a
difficult time making sense out of these wild gyrations and is likely to become
confused.
If the constraint is expressed instead as
Eq. 2-22
then the limit (provided as an upper bound on the DCONSTR entry) is used as
the normalizing factor or
Eq. 2-23
now a 1% excedence by of results in a constraint value of 0.01 and
the optimizer will have a much better chance of doing its job.
Equality Constraints in MSC.Nastran
Equality constraints can be defined simply by specifying equivalent upper and lower
bounds. This will yield a pair of equal and opposite constraints that force the response
to be satisfied with equality at the optimum

1

2
+
2
--------------------
max
0
g
u

1

2
+
2
--------------------
max

\ .
| |
GSCAL
----------------------------------------------- = 0

max
1 2 + ( ) 2
1 2 + ( ) 2 9900 =

1

2
+
2
--------------------
max

max
g
u

1

2
+
2
--------------------
\ .
| |

max

max
----------------------------------------------- = 0
1 2 + ( ) 2
max
Main Index Main Index
57 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-24
In practice, this is not a good idea since the optimizer does not have any special
algorithms for dealing with equality constraints so that one of the two relations in
Eq. 2-24 will always be violated. It is recommended that the equality constraint be
replaced with a tolerance by expressing the bounds as and on the
DCONSTR entry, where is the allowable tolerance and is perhaps 1% of the target
value.
R
j
r
j
X ( ) R
j

g
2j 1
X ( )
R
j
r
j
X ( )
R
j
-------------------------- = 0
g
2j
X ( )
r
j
X ( ) R
j

R
j
-------------------------- = 0
R
j
R
j
+

Main Index Main Index

58
2.3 Multidisciplinary Analysis
A structural design requires the synthesis of design requirements that can arise from
any number of areas, including strength, stiffness, and stability. Historically, this often
led to additional design cycles as additional analyses revealed shortcomings in the
designs that were based on preliminary analyses with a limited set of requirements.
As an example from aircraft design, the term flutter penalty was applied to the extra
weight that was required in the design to satisfy flutter speed requirements once the
initial strength design had been performed. The MSC.Nastran implementation of
design sensitivity and optimization recognizes this by combining analyses that were
previously considered separate, with separate input files and job submittals, into a
single job. The responses from these combined analyses can then be included in the
optimization step so that all the applicable requirements are considered
simultaneously.
Table 2-4 indicates the analyses that are available for design sensitivity and
optimization and lists some of the attributes of the analyses. The column of Solution
Numbers refers to the number MSC.Nastran would use to perform the separate
analyses. The Design Optimization Solution Number is 200.
Table 2-4 Analysis Disciplines Supported in Multidisciplinary Analysis
and Design of SOL 200
Analysis
SOL
Numbers
Multiple
Subcases
Multiple Boundary
Conditions
Statics 101 Y Y
Normal Modes 103 Y Y
Buckling 105 Y Y
Direct Complex
Eigenanalysis
107 N N
Direct Frequency 108 Y Y
Modal Complex
Eigenanalysis
110 N N
Modal Frequency 111 Y Y
Modal Transient 112 N N
Main Index Main Index
59 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
It is seen that SOL 200 has value simply in the performance of multidisciplinary
analysis. In fact, clients have been known to use SOL 200 to submit combined analyses
without any design sensitivity or optimization simply because it simplifies the
maintenance of input data and provides some performance efficiency (that is,
operations are not repeated for the separate analyses). The capabilities (and
limitations) of the separate Solution Numbers are incorporated into SOL 200 for the
most part. Table 2-4 indicates that many of the analysis types support multiple
subcases and/ or multiple boundary conditions. Analyses from SOLs 103, 110, 111, 112
and 145 require a normal modes analysis. When a SOL 200 job with multiple subcases
is submitted, a determination is made of the number of unique eigenanalyses that are
required based on the boundary conditions and the eigenanalysis method selected in
the separate subcases. Only the unique analyses are performed with the results shared
with the remaining subcases.
Response Calculation
Type-1 Responses
The type-1 responses are, in most cases, response quantities that are available as part
of standard MSC.Nastran results processing such as displacements, stresses and
eigenvalues. These responses are recovered from standard MSC.Nastran tables.
Several of the responses are not standard MSC.Nastran results and their development
requires additional description.
Static
Aeroelasticity
144 Y Y
Flutter 145 Y Y
Table 2-4 Analysis Disciplines Supported in Multidisciplinary Analysis
and Design of SOL 200
Analysis
SOL
Numbers
Multiple
Subcases
Multiple Boundary
Conditions
Main Index Main Index


60
Eq. 2-25
Eq. 2-26
Weight Response This response is calculated using the grid point weight generator
that is used to produce the weight and balance information that
is printed based on PARAM,GRDPNT. The WEIGHT type-1
response is identified by the selection of the row and column of
the 6X6 rigid body mass matrix that is to be used for the
response.
Volume Response This response is calculated by summing the individual volumes
of each of the finite elements.
PSD Response PSD responses are included as available design responses and
the development of these responses uses the same theoretical
development as that given in the MSC.Nastran Advanced Dynamic
Analysis Users Guide. On page 238 of that document, the power
spectral density of the an output response quantity is
where and denote two loading conditions (two subcases)
and the superscript * indicates the complex conjugate of the
response.
As an example, with two subcases, = 1 and = 2, the
spectral density is
Note that because the term , the resulting spectral
density, is a real number. Individual terms of can
be specified as PSD responses.
RMS Response RMS responses are the RMS value, , for the response
quantity, . This is theoretically developed as an integral that
is implemented in MSC.Nastran using a numerical
approximation
S
j
( ) H
j a
( )
b

H
j b
*
( ) S
ab
=
a b
a b
S
j
( ) S
jf
S
11
H
j1
H
j1
*
= = S
12
H
j 1
H
j 2
*
+
S
21
H
j2
H
j 1
*
S
22
H
j2
H
j2
*
+ +
S
21
S
12
*
=
S
j
( ) S
j
( )
u
j
j
Main Index Main Index
61 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-27
Eq. 2-28
Type-2 Response
The type-1 responses of the previous paragraphs provide responses that are directly
available from MSC.Nastran. Type-2 (or synthetic) responses, on the other hand,
provide responses that are not directly available form MSC.Nastran but can be
formulated using a combination of arguments and user-defined equations. A number
of predetermined(?) functions are available for this response type. These eliminate the
task of preparing user-defined equations. DRESP2 Bulk Data Entries on page 188
provides details on constructing these responses.
Type-3 Response
This type of response has been provided for situations where the desired design
response is beyond what is available using the type-2 response. The type-3 (or
external) response applies the same client-server technology that has been used in
MSC.Nastran for user-supplied beam libraries and for p-element geometry. Figure 2-
7 illustrates MSC.Nastrans ability to act as a client through an application
programming interface (API) to obtain response quantities from one or more servers
that are external to MSC.Nastran. This enables, for example, the incorporation of
proprietary local buckling criteria that are too complex to be developed using a
DRESP2. DRESP3 Bulk Data Entries on page 195 provides detail on constructing
these responses.
where, from the trapezoidal integration method, the
frequency-dependent coefficients, , are multiplied by
the frequency factors
u
j
S
j f
f d
f

C
f
S
j f
f

=
S
j f
f f
i
= ( )
C
f i
1
2
--- = f
i 1 +
f
i 1
( ) 1 i N < <
C
f 1
1
2
--- f
2
f
1
( ) = i 1 =
C
f N
1
2
--- f
N
f
N 1
( ) = i N =
Main Index Main Index


62
Figure 2-7 The External Response Client-Server Capabilities
Added Benefit of Synthetic and External Responses
SOL 200 can offer a benefit in the analysis phase that is in addition to the features
available in the separate solution numbers. This benefit relates to The Design
Model on page 39 and, in particular, the type-2 and type-3 responses. The results of
evaluation of the responses can be reported in the output file and the user could
exploit this by developing responses that are in addition to what are available from
standard data recovery. As an example, suppose it is desired to assess the buckling
behavior of a column that is contained within a large finite element model. One could
develop a type-2 response that is some combination of element response, element
dimensions, materials, geometry, and restraint conditions. Normally, this type of
calculation would require some type of postprocessing operation on the part of the
user, but this could be automated within SOL 200 to provide this result as part of the
standard analysis.
Mode Tracking
The selection of a particular eigenvalue from a normal modes analysis as a design
response is done by specifying its mode number. The users intent when specifying
this number may be to design this particular mode number regardless of the physical
nature of the original mode shape. However, in many applications, most notably
Server 1
Server i
Server 2
API
In-House
Program
MSC.Nastran
i 110 =
Main Index Main Index
63 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
automotive, the intent is to control a particular physical mode; for example, the
vibration of the vehicles roof. In this case, the users intent is actually to design the
frequency of the physical mode shape and the mode number could vary due to the
redesign. (That is, the example roof mode may be the 10th mode of the initial design,
but after a redesign it becomes the 12th.)
The eigenvalue analysis of SOL 200, includes a feature to track the modes so that the
modes are ordered based on their original physical behavior rather than on their
frequency order. This is done by constructing cross-orthogonality check between the
current design cycle and the previous cycle
Eq. 2-29
where the subscript denotes the current design cycle, the previous, and the
prime over indicates mass-normalized eigenvectors.
If there had been no changes in either mode shape, order, or mass of the system, the
cross-orthogonality check would yield a diagonal matrix. If the mode shapes change
due to a mass and/ or stiffness variation, however, their correlation will be less than
1.0. If the mode numbers have changed, dominant off-diagonal terms will be present
in . Where sufficient correlation exists, the results of this cross-orthogonality check
can be used as a basis to determine which modes have switched.

i
T
M
i

i 1
t
i
=
i i 1

t
i
Main Index Main Index


64
2.4 Constraint Screening
The concept of constraint screening was introduced in a general way in Structural
Optimization on page 26. The idea is to limit the constraints that are considered in
the redesign to those that are likely to play an active role in the redesign. To motivate
the importance of this, consider an example where the user has defined four stress
responses (say , , and von Mises) at the top and bottom of each QUAD4 finite
element and say the model contains 10000 of these elements. Further, suppose 10 static
loads cases represent the design loads for the design task. This rather modest problem
statement results in 4 stresses * 2 surfaces * 10000 elements * 10 subcases * 2 constraints
(upper and lower bounds) = 1.6 million constraints. Clearly, it is not practical or
desirable to consider all of these constraint conditions when performing a design task;
hence the use of constraint screening.
The Design Model on page 39 discussed how the normalization of the design
constraints has put all the design conditions on an equal basis regardless of whether
the underlying design response is a pressure, displacement, stress or whatever. The
screening process is a two stage process that does an initial screening based on a
threshold and a second screening based on the region. The threshold screening simply
says that it is possible to ignore constraints unless they exceed a specified value.
Figure 2-8 illustrates screening based on normalized constraints by representing the
current value of each constraint function in a bar chart format. If any constraint
exceeds the truncation threshold value denoted by TRS, we retain it for the ensuing
approximate optimization, while temporarily deleting all others below TRS.
Figure 2-8 Constraint Deletion

x

y

xy
TRS
0
-1.0
G(X)
XXXXXXXX
List of Constraints
Main Index Main Index
65 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
If there are a large number of constraints below this truncation threshold (as is typical
in structural optimization), the constraint screening phase will greatly simplify the
optimization task. However, the number of constraints can still be further reduced
using regionalization.
As an example, suppose we have an automobile roof panel, modeled with a number
of quadrilateral elements. This panel is to be of constant thickness for manufacturing
considerations and is found to be overstressed in a number of locations. It would not
make much sense to retain the stress constraints for every element in the panel since
all of these stresses will vary nearly in unison as the panel thickness is varied. Thus, it
is probably safe to retain only a few of the largest valued constraints from this
region.
Constraint regionalization is shown in Figure 2-9, which is the same as Figure 2-8, but
with the constraints grouped into three regions. For now, we will assume these
regions have been established based on some design model characteristics. Of the
three regions shown, only two have constraints that are numerically greater than the
truncation threshold. These constraints will pass the first screening test. Since the
retained constraints within each region are likely to contain redundant information,
only the largest constraints from each region are retained. These constraints are
denoted by the check marks in the figure. NSTR, which in this example is 2, stands for
the maximum number of constraints to be retained per region.
Figure 2-9 Constraint Regionalization
TRS
0
-1.0
G(X)
X XXXXXXXX

Region 1
1
Region 2
0
Region 3
2
NSTR = 2
Retained Responses
Main Index Main Index


66
It should be apparent that this screening process could result in a dramatic reduction
in the 1.6 million constraints mentioned in the example above; perhaps to as few as
100 retained constraints. This is a much more tractable problem for the optimization
algorithm to deal with. There is another more subtle, but equally important, benefit
that can be realized from constraint screening. The example given above assumed that
10 load cases were used to develop the constraints. It is quite conceivable that some of
these loads are unimportant in terms of producing critical design responses. The
entire load case can then be eliminated from consideration in the sensitivity phase of
the analysis with a direct effect on the performance of the sensitivity task of Design
Sensitivity Analysis on page 67. This concept is referred to as load case deletion
and plays an important role in MSC.Nastran design sensitivity and optimization.
As a final comment on constraint screening, sometimes it is desirable to retain
constraints that would not normally survive the screening process. For example,
perhaps only a sensitivity analysis is being performed and it is desired to know the
sensitivity of all the responses, regardless of the value of the constraints. In another
case, it might happen that a particular constraint is very sensitive and while not critical
for a particular design, becomes critical, or even violated after a redesign and then gets
deleted again after the second redesign. The design could then bounce back and forth
between two designs (one with the constraint well satisfied but with a large objective
and the second with the constraint violated but with a small objective) and not
converge. For these reasons, the threshold parameter TRS and the regionalization
parameter NSTR are made accessible to the user and can be adjusted to increase the
likelihood that a particular response is retained.
Main Index Main Index
67 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
2.5 Design Sensitivity Analysis
Design sensitivity analysis, as defined in this document, computes the rates of change
of structural response quantities with respect to changes in design variables. Other
sensitivities, such as change in a constraint value due to a change in the design
variable or change in the response due to a change in a property, can be derived from
chain rule operations on the sensitivities featured here. Since these partial derivatives
provide the essential gradient information to the optimizer, they are always computed
in connection with design optimization. Sensitivity analysis is always performed
automatically in MSC.Nastran whenever design optimization is requested. There may
be times when you want MSC.Nastran to compute just the sensitivity coefficients and
not perform optimization. You can then use this sensitivity information to perform
parametric design studies or to link with your own optimizer.
This section presents the theoretical basis for design sensitivity analysis in
MSC.Nastran. This is a broad subject that is organized by first presenting some
general concepts that are applicable to all of the sensitivity analyses. This is followed
by a detailed discussion of element and grid responses. Adjoint and direct techniques
for sensitivity are detailed and then sensitivity analyses for statics, dynamics and
static aeroelasticity are described. This is followed by a discussion of the sensitivity of
a number of responses that are global quantities: weight/ volume, eigenvalues
(normal modes and buckling), complex eigenanalysis, and flutter.
General Considerations
A design sensitivity coefficient is defined as the rate of change of a particular response
quantity r with respect to a change in a design variable x or . These coefficients
are evaluated at a particular design characterized by the vector of design variables ,
giving
Eq. 2-30
where subscripts are used to indicate the i-th design variable and the j-th response.
Eq. 2-30 is just the slope of the response with respect to as is shown in Eq. 2-9.
r x
X
0

ij
r
j
x
i
-------
X
0
=
x
i
Main Index Main Index


68
Figure 2-10 Design Sensitivity Coefficient -- Graphical Interpretation
For Which Responses Are Sensitivities Computed?
Sensitivities are computed for all of the responses (types-1, 2, and 3) that are used to
define the objective function and retained constraints.
Due to the constraint screening described in Constraint Screening on page 64,
fewer response sensitivities are typically computed than there are responses defined
in the design model.
Internal Representation of the Sensitivity Coefficients in
MSC.Nastran
For reasons of efficiency as well as accuracy, MSC.Nastran uses a slightly different
internal representation of the coefficients of Eq. 2-30 depending on how the design
variables are related to the analysis model properties.
The set of independent design variables is taken as the basis for the design sensitivity
coefficients when the linear relation between the designed property and design
variable is used. This is an efficient choice since a large number of properties may be
a function of a much smaller set of design variables.

i j
r
j
x
i
( )
x
i
0
x
i
Main Index Main Index
69 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
For nonlinear relations, the basis for sensitivity analysis is taken to be the set of
analysis model properties referenced on all of these second-level property relations.
Inside the code, the sensitivities are then first computed as
Eq. 2-31
Choosing properties as the basis greatly improves the accuracy of the approximate
responses used in design optimization. Since an explicit relation between the
properties and the associated design variables is known, this information can be used
to evaluate the properties for a given change in the set of design variables. Computing
the sensitivities with respect to the design variables would have linearized this
relation, resulting in a less accurate approximation.
To summarize, the sensitivity coefficients used internally in MSC.Nastran are
i = 1, NDVI
Eq. 2-32
k = 1, NIPR
where NDVI is the number of independent design variables and NIPR is the set of
designed properties which are a nonlinear function of the design variables (including
the spawned designed properties from the beam library).
To illustrate the advantage in using a mixed design variable-property basis, consider
responses that depend on properties which are linear functions of the design
variables. The approximation for a change in the response due to a change in the
design variable is
Eq. 2-33
where the notation is used to indicate that the ith design variable is being
perturbed while the remaining variables on the vector are unchanged.

ij
r
j
p
i
--------
X
0
=
r
j
x
i
-------
r
j
p
k
---------
r
j
X
0
x
1
+ ( ) r
j
X
0
( )
r
j
x
1
---------
X
0
+ x
1

X
0
x
i
+ ( )
X
0
Main Index Main Index


70
For responses that depend on the properties that are nonlinear functions of the design
variables, the approximation can be written as
Eq. 2-34
where the properties at the perturbed design in Eq. 2-34 can be evaluated precisely
from the input equations.
Finite Difference Methods
Finite difference techniques are used in places in MSC.Nastran where it is not practical
to compute analytical sensitivities.
MSC.Nastran employs two alternative finite difference schemes in its sensitivity
calculation: forward difference and central difference. Conceptually, forward
difference techniques determine the slope of a function at point by solving
Eq. 2-35
while central differencing solves for
Eq. 2-36
The forward differencing provides an estimate of the true sensitivity with an error that
is on the order of the step size. Central differencing reduces the expected error to
be on the order of . Central differencing therefore provides a higher quality result
but at the added cost of two difference steps rather than one.
Sensitivities of type-2 and type-3 responses are always calculated using central
differencing techniques to obtain the change in the response due to the change in one
of the inputs. Chain rule techniques are employed to obtain the overall sensitivity of
one of these responses to the design variable. For example, suppose one has
constructed a type-2 response that is a function of a single displacement, stress and
design variable
Eq. 2-37
r
j
X
0
x
i
+ ( ) r
j
X
0
( ) =
r
j
p
k
---------
X
0 k

+ p
k
X
0
x
i
+ ( ) p
k
X
0
( ) ( )
f X ( ) X
0
f X
0
( )
x
i
-----------------
f X
0
x
i
+ ( ) f X
0
( )
x
i
---------------------------------------------------- =
f X
0
( )
x
i
-----------------
f X
0
x
i
+ ( ) f X
0
x
i
( )
2x
i
-------------------------------------------------------------------- =
x
x
2
r
j
r
2
u
r

s
x
t
, , ( ) =
Main Index Main Index
71 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Then the sensitivity with respect to the ith design variable is computed using
Eq. 2-38
where the and terms are calculated using central differencing
techniques and and terms are derived from the response sensitivity
analysis described in the following sections. The term is 1.0 for and 0.0
otherwise.
Element and Grid Responses
Direct and Adjoint Sensitivity Methods
MSC.Nastran supports two distinct methods for computing sensitivity coefficients:
direct sensitivity and adjoint sensitivity. The difference between the two methods can
be motivated by expressing a response as a function of the finite element responses
and the design variables
Eq. 2-39
The the sensitivity of this response with respect to the design variables can be
expressed as
Eq. 2-40
The direct method solves for and uses that in a semi-analytic formulation such
as that given in Direct Sensitivities of Static Responses on page 72 to develop the
response sensitivity. If the system equation for the finite element analysis can be
written in a general form as
Eq. 2-41
where is simply in a statics analysis and is in a frequency response
analysis, then the derivative of Eq. 2-41 gives
Eq. 2-42
dr
j
dx
i
-------
r
j
x
t
-------
x
t
x
i
------- =
r
j
u
r
--------
u
r
x
i
--------
r
j

s
---------

s
x
i
--------- + +
r
j
x
t
r
j
u
r
, r
j

s

u
r
x
i

s
x
i

x
t
x
i
i t =
r f u X , ( ) =
r
x
------
f
u
------
u
x
------ =
u x
S [ ] u { } P { } =
S K
2
M iB K + +
S [ ]
u
x
------
)
`

P
x
-------
)
`

S
x
------ u { } =
Main Index Main Index


72
The important point about Eq. 2-42 for now is that it requires the solution of
right-hand sides, where of Eq. 2-32 and is the number of
retained solution vectors. For statics solutions, this is the number of retained
subcases while for a frequency response analysis, it is the number of retained
frequencies per subcase summed over the number of subcases.
The adjoint method employs a scheme that first solves for adjoint solutions using
Eq. 2-43
and then substitutes in Eq. 2-40 to give
Eq. 2-44
Where has been used to eliminate the system matrix.
Eq. 2-43 requires the solution of nresp equations, where nresp is the number of retained
responses. There is an important class of problems where the adjoint method had a
distinct advantage over the direct method: frequency response optimization. As an
example, consider a design model that has 20 design variables and 100 frequency
excitation points that results in 150 active responses distributed over the 100
frequencies. (Note that excitation frequencies that do not result in active responses are
screened out and do not contribute to the nlc calculation.) In this case, the adjoint
method requires 150 solutions while the direct method requires 2000. This indicates
that the adjoint method can be expected to require roughly 7.5% of the resources (CPU
and disk space) needed by the direct method.
This section continues by discussing the direct sensitivity method with a subsequent
discussion of the adjoint method. The adjoint method has been implemented in
MSC.Nastran for grid responses.
Direct Sensitivities of Static Responses
For a given structure with specified geometry, material properties, and boundary
conditions, a displacement-based linear static analysis computes the displacement
responses due to the applied loads. All other static responses such as stresses and
strains are determined from the displacement solution. For an arbitrary response, r,
the functional dependency on the displacement is written as,
ndv nl c
NDV NDVI NIPR + = nl c
u { }
S [ ]
T
{ }
f
u
------
)
`

=
r
x
------ { }
T P
x
-------
S
x
------ u { }
)
`

=
S [ ]
T
S [ ] I [ ] =
Main Index Main Index
73 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-45
For the direct method in MSC.Nastran, the sensitivities of these responses with respect
to changes in the design variables are approximated using first forward finite
differences as
Eq. 2-46
where is the j-th design response. It should be noted that Eq. 2-46 uses and in
a generic way to encompass the non-linear properties of Eq. 2-32. The steps required
to evaluate Eq. 2-4 include:
1. The designed properties and/ or designed shape basis vectors are perturbed
using a finite difference step and perturbed element matrices are developed.
2. Pseudo-loads are constructed from the perturbed loads, the perturbed
element matrices and the solution vectors.
3. The perturbed displacements resulting are solved from the pseudo-loads.
4. The perturbed displacements are added to the baseline solution vectors to
produce the total solution vectors.
5. Data recovery is applied to these total vectors to obtain the responses when
the design quantities have been perturbed.
6. Eq. 2-46 is applied to obtain the required sensitivity.
Each of these steps is now described in greater detail.
Property and/or Shape Basis Vector Perturbation
For property optimization, the size of the move is given by
Eq. 2-47
or
Eq. 2-48
where DELB is user accessible parameter with a default value of 0.0001. If in Eq. 2-
47 or in Eq. 2-48 is exactly zero, then the respective move is given by .001.
r r U X , ( ) =
r
j
x
i
-------
r
j
x
i
--------
r
j
X
0
x
i
U , U + + ( ) r
j
X
0
U
0
, ( )
x
i
--------------------------------------------------------------------------------------------- =
r
j
X x
U U +
x
x
i
DELB = x
i

p
i
DELB = p
i

x
i
p
i
Main Index Main Index


74
For shape basis vectors, the selection of the step size is more involved because, as the
discussion on designed shapes in The Design Model on page 39 indicates (Eq. 2-
11), the magnitude of the change in the grid location is the product of the shape basis
vector and the change in the design variable.
For shape sensitivities, grid coordinates are perturbed a finite amount
Eq. 2-49
The subscript i in Eq. 2-49 characterizes the relation as a variation in shape for the i-th
design variable.
is the maximum strain energy norm computed using the basis vector. STPSCL
is a user accessible parameter with a default value of 1.0.
The maximum strain energy norm used in Eq. 2-49 is computed by first determining
the grid forces due to a shape basis vector displacement
Eq. 2-50
Note that the components of T associated with rotational DOFs (degrees-of-freedom)
are set to zero so that no forces are generated on these DOFs. The forces can be used
to compute a strain energy per grid location as
Eq. 2-51
Eq. 2-51 is an open product, resulting in a vector of strain energies. Locating the
maximum term in the vector and taking its square root yields
Eq. 2-52
We get one such for each design variable-shape basis vector pair. This has been
found to yield a consistent set of finite difference move parameters for shape
sensitivity.
Pseudo-Load Vectors
The term Pseudo-Load vectors is used to refer to the terms on the right hand side of
Eq. 2-42. For static analysis this equation, can be written as
Eq. 2-53
G { }
i
STPSCL
E
i
----------------------- T { }
i
=
E
i
T { }
i
K [ ] T { }
i
F
s
{ } =
F
s
{ } T { }
i
E { }
i
=
E
i
max E { }
i
( )
1 2
=
E
i
K [ ]
U { }
x
i
--------------
P { }
x
i
-------------- =
K [ ]
x
i
------------ U { }
Main Index Main Index
75 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Computationally, Eq. 2-53 and Eq. 2-41 are identical except that they have different
right-hand side load vectors, hence the term pseudo-loads. The cost of the
displacement sensitivity solution can be reduced if the stiffness matrix, already
available in decomposed form as a result of the finite element solution, is reused. This
observation is taken advantage of in MSC.Nastran.
Note that the right-hand side of Eq. 2-53 includes the partial derivatives and
. Since these are generally implicit functions of the design variables, these
derivatives are approximated using finite differences.
The selection of the finite difference method (forward vs. central) is controlled in
MSC.Nastran by PARAM CDIF. The default is NO (forward differencing) for
property sensitivity and YES (central differencing) for shape sensitivity. These
defaults reflect MSCs experience that it is generally more difficult to obtain accurate
shape sensitivities relative to the property sensitivities.
For forward differencing,
Eq. 2-54
For central differencing,
Eq. 2-55
The product of the change in the matrices times the solution vectors is formed directly
rather than first forming the difference in the matrices and performing the
multiplication by the solution vectors later. This greatly improves the performance
and disk space usage in the sensitivity calculation.
P { } x
i

K [ ] x
i

K [ ]
x
i
------------ u { }
K X
0
x
i
+ ( ) [ ] K X
0
( ) [ ]
x
i
------------------------------------------------------------------- u { }
P { }
x
i
--------------
P X
0
x
i
+ ( )
)
`

P X
0
( ) { }
x
i
----------------------------------------------------------------------
K [ ]
x
i
------------ u { }
K X
0
x
i
+ ( ) [ ] K X
0
x
i
( ) [ ]
2x
i
-------------------------------------------------------------------------------- u { }
P { }
x
i
--------------
P X
0
x
i
+ ( )
)
`

P X
0
x
i
( )
)
`

2x
i
-----------------------------------------------------------------------------------
Main Index Main Index


76
In most cases, the load vector is invariant with respect to design variable changes.
Exceptions are gravity loads that vary with the mass of the structure and thermal
loads which are derived from the element properties. MSC.Nastran does not account
for changes in loading due to changes in shape in its sensitivity calculations so that,
for example, a pressure load is assumed to be invariant in the shape sensitivity
calculation. The changed load is accounted for when a finite analysis is performed on
the redesigned shape.
Solution for the Perturbed Displacements
Once the pseudo-loads have been developed, standard MSC.Nastran solution
techniques can be used to solve for the perturbed displacements. This entails
reduction techniques to bring the loads to the degrees-of-freedom required by the
decomposed stiffness matrix and subsequent recovery to bring the vectors back to
the full set of physical degrees-of-freedom.
Addition of Perturbed and Baseline Solution Vectors
It needs to be recognized that in general
Eq. 2-56
so that it is necessary to form the for the recovery of the perturbed
responses. This entails a straightforward addition except that the vector has
columns for each design variable for each column in . Therefore it is necessary to
precede the addition by an operation which duplicates the vectors to make this
matrix conformable with the vectors.
Data Recovery
Standard MSC.Nastran data recovery techniques are applied to extract retained
design responses for each of the perturbed design variables.
Final Sensitivity Calculations
With the perturbed and baseline responses, it is now possible to apply Eq. 2-46 to
compute the actual sensitivity of the type-1 one responses with respect to the
independent design variables and the nonlinear properties.
Inertia Relief Sensitivity Analysis
A special case in static sensitivity analysis occurs when the structure being analyzed
is free to move as a rigid body. In this case inertia relief techniques are applied,
resulting in additional terms in the sensitivity calculations. A complete derivation of
u
u
r u u + ( ) r u ( ) r u ( ) +
u u + { }
u { }
u { }
u { }
u { }
Main Index Main Index
77 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
the sensitivity calculations for this case is not warranted; however, a sketch of the
method, indicating how inertia relief sensitivity analysis is implemented in
MSC.Nastran, is provided.
For static inertia relief analysis, a revised loading is developed using
Eq. 2-57
where:
These accelerations can be either specified by the user with a DMIG,UACCEL Bulk
Data entry or computed as part of the solution using
Eq. 2-58
where the mass and loading terms have been reduced to the SUPORTed degrees-of-
freedom.
The sensitivity of the load is then
Eq. 2-59
is considered invariant in this derivation, indicating that shape changes are not
supported. The term is calculated from a perturbation of Eq. 2-58 to give
Eq. 2-60
Note that second order terms, such as , are retained in this special analysis.
L = the applied loading.
D = a rigid body mode matrix that provides accelerations at all grid points due to
rigid body accelerations .
P
i
P
L
MDu

r
=
u

r
u

r
{ } M
r
[ ]
1
P
r
l
)
`

=
P
i
{ } P
L
MDu

r
MDu

r
MDu

r

)
`

=
D
u

r
u

r
{ } M
r
M
r
+ [ ]
1
P
r
L
M
r
u

)
`

=
MD u

r
Main Index Main Index


78
Shape Sensitivity with Rigid Elements
Rigid elements provide a convenient way of specifying linear relationships among
degrees-of-freedom. This has implications in shape sensitivity analysis since the
perturbations in the grid locations produce changes in the linear relationships.
Basically, the relationship between dependent (m-set) and independent (n-set)
degrees-of-freedom is expressed as
Eq. 2-61
This gives rise to a relationship between dependent and independent degrees-of-
freedom of the form
Eq. 2-62
where
Eq. 2-63
or
Eq. 2-64
For sensitivity analysis, it is then necessary to develop from perturbing Eq. 2-64
Eq. 2-65
or
Eq. 2-66
If and are defined as being finite forward steps for the and matrices,
then Eq. 2-64 can be used in Eq. 2-66 to give
Eq. 2-67
The rigid element sensitivity calculation, therefore, entails including this when
computing the pseudo loads. The details of including these terms are not shown here,
but differ by analysis type and whether direct or adjoint methods are employed.
R
m
[ ] u
m
{ } R
n
[ ] u
n
{ } 0 = +
u
m
{ } G
m
[ ] u
n
{ } =
G
m
[ ] R
m
[ ]
1
R
n
[ ] =
R
m
[ ] G
m
[ ] R
n
[ ] =
G
m
[ ]
R
m
[ ] G
m
[ ] R
m
[ ] G
m
[ ] R
n
[ ] = +
G
m
[ ] R
m
[ ]
1
R
n
[ ] R
m
[ ] G
m
[ ] + ( ) =
R
n
[ ]
+
R
m
[ ]
+
R
n
R
m
G
m
[ ] R
m
[ ]
1
R
n
[ ]
+
R
m
[ ]
+
G
m
[ ] + ( ) =
G
m
[ ]
Main Index Main Index
79 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Static Aeroelastic Sensitivity Analysis
Static Aeroelastic analysis performs a static analysis of a free-flying vehicle immersed
in an airflow. This is a separate discipline within MSC.Nastran SOL 200 and the
available responses are:
Static aeroelastic responses, including not only displacement, stresses,
forces, etc., but also trim variables. This allows the evaluation, for example,
the change in the trim angle of attack due to a structural change.
Stability derivative sensitivities. This, for example, allows the determination
of the effect that strengthening a wing spar has on the rolling moment
produced by a deflection of the aileron.
This section contains a brief description of the theory for aeroelastic sensitivity
analysis. For both theoretical and design modeling details as well as example
problems, see the MSC.Nast ran Aeroelast ic Analysis Users Guide.
The form of the equations are the same as for sensitivity analysis of static responses
Eq. 2-68
where, for static aeroelastic analysis,
K [ ] u { } P { } =
Main Index Main Index


80
Eq. 2-69
This equation is differentiated with respect to a design variable and solved for the
displacement sensitivities using a semi analytic approach as has been described in
Eq. 2-53 for static sensitivity analysis. Static Aeroelastic Sensitivity (see
Aeroelastic Design Sensitivities and Optimization in Chapter 2 of the
MSC.Nastran Aeroelastic Analysis Users Guide) provides further details of the
sensitivity and indicates the special steps that are required to accommodate the mean-
axis constraint embedded in Eq. 2-69. Also, the stability derivative responses
necessitate the generation of additional solution vectors for a unit deflection of the
associated aerodynamic variable.
Direct Sensitivities for Dynamics
Dynamic response sensitivities are computed in Solution 200 in connection with
design sensitivity analysis and optimization. The following disciplines are available:
Direct frequency
Modal frequency
Modal transient
K [ ]
K
l l
K
ll
a
+ K
l r
K
l r
a
+ M
l l
D M
lr
+ K
lx
a
D
T
M
ll
M
rl
+ D
T
M
l r
M
rr
+ 0 0
D
T
K
ll
a
K
rl
a
+ D
T
K
l r
a
K
rr
a
+ m
r
D
T
K
l x
a
K
rx
a
+
0 0 0 0
=
u [ ]
u
l
u
n
u

r
u
x
)



`




=
P { }
P
l
0
P
r
D
T
P
l
+
P
x
)



`




=
Main Index Main Index
81 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
All of these disciplines support fluid-structural coupling.
General Considerations
The general methodology employed for dynamic responses are very similar to those
just described for static analysis so that this description uses far less detail and the
reader is referred to the previous section for a more complete description.
The complete stiffness matrix for dynamic response analysis consists of a
superposition from damping as well as direct matrix input
Eq. 2-70
where:
The DMIG input of cannot be designed, nor can PARAM G. The matrix can be
affected by any of the available property or shape design parameters. The
parameter is accessible to design as one of the material properties shown in the table
Material Properties Available as Designed Properties on page 43.
The total damping matrix is the sum of the contributions from viscous elements
and direct matrix input at the grids
Eq. 2-71
For structural mass,
Eq. 2-72
The direct matrix inputs for damping and mass are assumed constant in the
sensitivity analysis as is the case with the direct matrix input for stiffness.
Direct Frequency Response Sensitivity Analysis
Sensitivities for direct frequency response are obtained by differentiating the
governing equation
= structural stiffness
= uniform structural damping coefficient PARAM,G
= structural damping coefficient, GE, on a MATi entry
= direct matrix input at grids DMIG
K [ ] K
1
[ ] = ig K
1
[ ] i g
e
K
e
[ ]

K
2
[ ] + + +
K
1
[ ]
g
g
e
K
2
[ ]
K
2
K
1
g
e
B
1
[ ]
B
2
[ ]
B [ ] B
1
[ ] = B
2
[ ] +
M [ ] M
1
[ ] = M
2
[ ] +
B
2
[ ] M
2
[ ]
Main Index Main Index


82
Eq. 2-73
to obtain
Eq. 2-74
where indicates differentiation with respect to a design variable, or . Eq. 2-74
is solved once for each forcing frequency, load condition, and design variable.
Although the force derivative term is shown in Eq. 2-74 for completeness, this
term is assumed to be zero in MSC.Nastran. Therefore, if gravity, thermal, or
geometry dependent loadings are significant, some error will be introduced into the
approximation. This assumption of zero also applies to the modal frequency
and modal transient formulations.
There is a subtle issue in the data recovery step of the dynamic sensitivity calculation
that can be mentioned here. MSC.Nastran calculates results in the analysis set and
then recovers them to the physical set. One step in this standard recovery is the
calculation of the omitted degrees-of-freedom using a relation of the form
Eq. 2-75
(See Eq. 3-23 of the MSC.Nastran Advanced Dynamic Analysis Users Guide).
This ignores terms that arise from the structures mass and the loads applied to the
omitted degrees-of-freedom that is given in Eq. 3-20 in the MSC.Nastran Advanced
Dynamic Analysis Users Guide.
Eq. 2-76
In a standard dynamic analysis, users are cautioned to not apply loads to omitted
degrees-of-freedom and not to omit degrees-of-freedom that have appreciable mass
so that the approximation involved in Eq. 2-76 can be considered negligible. The
omitted degrees-of-freedom are typically nodal rotations that have no load or mass
associated with them. However, in a sensitivity analysis, terms can easily
produce loads on these omitted degrees-of-freedom with the result that the
sensitivities of element responses can be in significant error. MSC has provided a
remedy in the form of PARAM SENSUOO. The default for this parameter is NO so

2
M [ ] i B [ ] K [ ] + + ( ) U { } P { } =

2
M [ ] i B [ ] K [ ] + + ( ) U { } P { } =
2
M [ ] i B [ ] K [ ] + + [ ] U { }
x
i

P { }
P { }
u
o
{ } G
o
[ ] u
a
{ } =
u
o
o
)
`

K
oo
[ ]
1
P
o
M
oa
[ ] u
a
{ } [ ] =
K [ ] u { }
Main Index Main Index
83 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
that the terms of Eq. 2-75 are neglected. Setting SENSUOO to YES includes the term.
With the de-emphasis on omitted degrees-of-freedom in present day finite element
analysis, this is typically not a major issue.
Modal Frequency Response Analysis
The governing equations for modal frequency response analysis are obtained from the
equations for frequency response by introducing the modal transformation
Eq. 2-77
where is a matrix of eigenvectors, perhaps augmented by residual vectors and
is a vector of modal coordinates.
If Eq. 2-77 is applied to Eq. 2-73 and the result is remultiplied by , the following
set of equations result
Eq. 2-78
MSC.Nastran also supports a modal damping term that can be added to the diagonal
elements of the modal damping matrix. This terms expressed as
Eq. 2-79
where is the undamped vibration frequency, m is the generalized mass in the ith
mode and is a user input value. (See Eq. 2-26 in the MSC.Nastran Advanced Dynamic
Analysis Users Guide). It is recommended that this form of damping be avoided in
design sensitivity and optimization. This is because MSC.Nastran neglects the
sensitivity of this term when performing a sensitivity analysis in a modal frequency
response analysis. It has been found that this can result a substantial error in the
sensitivity results, particularly when the excitation frequency is near a resonant
frequency. It is recommended that the PARAM G form of structural damping be used
in place of the modal damping when performing modal frequency response
sensitivity analysis.
An important assumption that is made is that the normal modes technique used in
performing the modal analysis is equally adequate in performing a sensitivity
analysis. That is, the basis that is used to span the analysis space that is used to justify
a modal analysis in the first place should be just as valid in a sensitivity analysis. The
benefit of making this assumption is that it avoids the need for computing the
sensitivities of the eigenvectors. Support for this assumption can be obtained from the
realization that modal analyses can be performed without any loss of accuracy relative
U { } [ ] { }
[ ] { }
[ ]
T

2
[ ]
T
M [ ] [ ] i [ ]
T
B [ ] [ ] [ ]
T
K [ ] [ ] + + ( ) { } [ ]
T
P { } =
b
i
g
i

i
m
i
=

i
g
i
Main Index Main Index


84
to a direct analysis if a complete set of modes (that is all the modes are extracted from
the physical model) is used. Therefore a sensitivity analysis that uses a complete set of
modes applied to the pseudo-loads will result in the same sensitivity results as would
be obtained using a direct analysis. Another way of stating this is that it assumes that
Eq. 2-80
Applying this to Eq. 2-74 yields
Eq. 2-81
With the derivatives of the modal coordinates computed from Eq. 2-81, the
displacement derivatives are available by direct substitution into Eq. 2-80. The data
recovery steps of the static sensitivity analysis can then be applied to obtain the
sensitivities of the grid and element responses.
The above discussion of the assumption that accurate sensitivities can be obtained
without including eigenvector sensitivities should be qualified by pointing out that
the pseudo-loads generated in a sensitivity analysis are likely to excite the structure in
ways that typical structural loads do not (for example, the pseudo-loads are likely to
be in the plane of the structure while applied loads are typically transverse to the
structure). For this reason, it is recommended that more modes be retained when
performing a sensitivity analysis than would be used in frequency response analysis.
A rule of thumb is that twice as many modes should be retained to obtain accurate
sensitivities as are required to obtain accurate responses. However, this is only a
guideline and it is strongly recommended that a convergence study be performed by
increasing the number of retained normal modes until the sensitivity values do not
change significantly.
Modal Transient Response Sensitivities
The governing differential equation for transient dynamic response is
Eq. 2-82
The modal transformation is again given by
Eq. 2-83
U { } [ ] { }

2
[ ]
T
M [ ] [ ] i [ ]
T
B [ ] [ ] [ ]
T
K [ ] [ ] + + ( ) { } =
[ ]
T
P { }
2
M [ ] i B [ ] K [ ] + + ( ) U { } ( )
M [ ] U

{ } B [ ] U

{ } K [ ] U { } + + P { } =
U { } [ ] { } =
Main Index Main Index
85 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Introducing the modal transformation into Eq. 2-82 and differentiating with respect to
a design variable yields the expression for the modal transient displacement
sensitivities
Eq. 2-84
As in modal frequency sensitivity analysis, Eq. 2-84 makes the assumption that a
sufficient number of modes are included such that the relation of Eq. 2-80 is a good
approximation.
Once the solution for the modal coordinates and their time derivatives are available,
the displacement derivatives are computed from Eq. 2-83.
Coupled Fluid-Structure Interaction Sensitivity
Coupled fluid-structure analysis is used to solve problems in which the interaction
effects between fluid and structure are significant. MSC.Nastran uses a fully-coupled
formulation for this analysis. Although its principal application is in acoustic
problems, it can also be applied to other general problems, such as fluid storage (tank
sloshing), or other situations in which an inviscid, irrotational formulation is valid.
Refer to the Coupled Fluid-Structure Interaction on page 754 of the MSC.Nastran
Reference Manual for details on the analysis formulation.
The sensitivity equations for fluid-structure interaction are identical in form to the
dynamic response sensitivity equations and can be solved using the existing dynamic
response sensitivity solution algorithms. To understand why this is so, the general
equations of motion for the fluid-structure system can be written as
Eq. 2-85
The solution vector consists of both structural displacements and the fluid
pressure .
Eq. 2-86
The vector of external loads is comprised of both structural as well as fluid forces
or
[ ]
T
M [ ] [ ]

{ } [ ]
T
B [ ] [ ]

{ } [ ]
T
K [ ] [ ] { } + +
[ ]
T
P { } M [ ] U

{ } B [ ] U

{ } K [ ] U { } + + ( ) ( ) =
M [ ] x

{ } K [ ] x { } + F { } =
X { } U { }
P { }
x { }
U { }
P { }
)
`

=
F { }
Main Index Main Index


86
Eq. 2-87
The mass matrix in Eq. 2-85 is defined as
Eq. 2-88
where:
The stiffness matrix consists of
Eq. 2-89
where:
If the loads in Eq. 2-85 are frequency-dependent, a direct formulation can be written
just as for the structural case where no fluid effects are present. Differentiating the
direct frequency equation yields an expression similar to Eq. 2-74, which can be
solved by exactly the same methods.
In very large problems, it is often desirable to apply the modal decomposition method
in order to reduce the cost of the analysis. Modal formulations in coupled fluid-
structure analysis are derived from a separate consideration of both fluid and
structural components.
= structural mass
= fluid mass (including compressibility effects)
= fluid-structure coupling matrix
= structural stiffness
= effective fluid stiffness
F { }
F
s
{ }
F
f
{ }
)

`


=
M [ ]
M
s
[ ] 0
A [ ]
T
M
f
[ ]
=
M
s
[ ]
M
f
[ ]
A [ ]
K [ ]
K
s
[ ] A [ ]
0 K
f
[ ]
=
K
s
[ ]
K
f
[ ]
Main Index Main Index
87 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The structural modes are computed for a structure in a vacuum, that is, in the absence
of any fluid effects. The modal transformation for the structural degrees-of-freedom
can then be written as
Eq. 2-90
where:
The modes for the fluid are computed under the effects of rigid wall boundary
conditions, which eliminate the structural coupling effect. The resulting modal
transformation for the fluid degrees-of-freedom is then
Eq. 2-91
where are the modes of the fluid enclosed by a rigid container.
After modal reduction for both structural and fluid degrees-of-freedom, the equations
of motion can be written as
Eq. 2-92
This modal transient formation can be differentiated with respect to a design variable
to obtain the equations for coupled fluid-structure interaction sensitivity. The form of
the equations is similar to Eq. 2-84 for modal transient analysis, so it is not repeated
here. However, the solution process is identical.
If the excitation is frequency-dependent rather than time-dependent, the resultant
sensitivity equations assume the form given by Eq. 2-81 for structural problems. Due
to the similarities of their equations, the solution process is again identical.
= are the modes for structure in a vacuum
u [ ]
s
[ ]
s
{ } =
m
s
[ ]
s
[ ]
T
M [ ]
s
[ ] =
k
s
[ ]
s
[ ]
T
K [ ]
s
[ ] =

s
[ ]
P { }
f
[ ]
f
{ } =
m
f
[ ]
f
[ ]
T
M [ ]
f
[ ] =
k
f
[ ]
f
[ ]
T
K [ ]
f
[ ] =

f
[ ]
m
s
0

f
T
A
T

s
\ .
| |
m
f

f
)

`


k
s
s
T
A
f
\ .
| |
0 k
f

f
)

`


+

s
T
F
s

f
T F
f
)

`


=
Main Index Main Index


88
Adjoint Sensitivity Analysis for Grid Responses
As detailed above, the criterion for invoking the adjoint sensitivity method is
Eq. 2-93
The actual selection of the adjoint sensitivity method involves further
factors/ restrictions:
The adjoint method is only available for statics (not including inertia relief,
thermal loads or static aeroelasticity) and frequency response analyses (both
direct and modal).
Only grid responses are supported. SPC Forces are not supported.
The method is not available with p-element models
The limitation to grid responses is motivated by two factors: 1) The development of
the vector required in Eq. 2-40 is not trivial for certain elements and 2) it is
unlikely that a design task that included element responses would pass the criterion
of Eq. 2-93.
The adjoint sensitivity analysis shares the property or shape basic vector perturbation
step detailed by Eq. 2-47 thru Eq. 2-52 with the direct method, but the two methods
diverge at that point. The adjoint method requires solution of the adjoint equation
Eq. 2-94
where the vectors are established based on the retained grid response(s) and
the rows of these vectors are all zeroes except for a 1.0 in the location corresponding
to the degree-of-freedom of the active response. The sensitivity can then be developed
directly using
Eq. 2-95
where the terms on the right-hand side are developed using the semi-analytic
methods similar to those given in Eq. 2-54 and Eq. 2-55.
NRESP NDV < NLC
f u
S [ ]
T
{ }
f
u
------
)
`

=
f u { }
r
x
------
)
`

{ }
T

2
M [ ] i B [ ] K [ ] + + [ ] u { } =
Main Index Main Index
89 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Because many of the steps required in the direct sensitivity analysis are not needed in
the adjoint method, this permits additional savings in terms of CPU time and required
disk space over and above the savings produced from the fewer number of adjoint
loads relative to the number of pseudo-loads.
As a final comment on the adjoint sensitivity method, Eq. 2-94 is similar to the
equation solved in a frequency response analysis. This similarity can be exploited by
solving for the adjoint solution vectors of as part of the analysis. This is
particularly useful in a direct frequency response analysis where a considerable
portion of the time can be expended in the decomposition of the matrix at each
frequency. See the description of PARAM SOLADJC in SOLADJC on page 306 for
details as to when and how the adjoint solution vectors are computed during the
frequency response analysis.
Weight Sensitivities
Weight sensitivities are calculated using matrix and matrix . The latter
matrix provides the displacement at all the degrees-of-freedom in the model based on
a unit displacement at the six degrees-of-freedom at the grid identified by PARAM
GRDPNT.
Eq. 2-96
This is a six by six matrix for each design variable and the appropriate term in the
matrix is selected based on the users specification of the weight response.
Volume Sensitivities
Volume sensitivities are computed by computing the volume of all the finite elements
for a perturbation in each of the design variables, subtracting out the baseline volume
and dividing by the perturbation.
Eq. 2-97
Real Eigenvalue Sensitivities
The eigenvalue equation is
Eq. 2-98
{ }
S [ ]
M [ ] DR
g
[ ]
W [ ] DR
g
[ ]
T
M [ ] DR
g
[ ] =
V
x
i
--------
V x
0
x
i
+ ( ) V x
0
( )
x
i
------------------------------------------------------ =
K [ ]
n
M [ ] ( )
n
{ } 0 =
Main Index Main Index


90
where and are the n-th eigenvalue and eigenvector, respectively. is the
structural stiffness, and is the structural mass.
The governing Eq. 2-98 can be differentiated with respect to the i-th design variable
to yield,
Eq. 2-99
When Eq. 2-99 is pre-multiplied , the term that then appears at
the front of Eq. 2-99 is the transpose of Eq. 2-98 and therefore zero. Eq. 2-99 can then
be solved for the eigenvalue derivatives
Eq. 2-100
In practice the solution to the above equation is based on a semi analytic approach.
The derivatives of the mass and stiffness matrices are approximated using finite
differences of the form in Eq. 2-54 and Eq. 2-55. Eq. 2-100 is solved for each retained
eigenvalue referenced in the design model and for each design variable. It should be
mentioned that this equation is valid only for the case of distinct eigenvalues.
Real Eigenvector Sensitivities
The eigenvector sensitivity method available in MSC.Nastran is based on Nelsons
Method (see Nelson, R. B., in Reference 6.. For the method, Eq. 2-99 can be rewritten as
Eq. 2-101
Eq. 2-101 cannot be solved directly for the eigenvector sensitivity because the
term is singular. This requires that the matrix size be reduced by one and
the reduction technique differs depending on what method the user has specified for
the normalization of the eigenvector. If the NORM=MAX option has been selected,
then the maximum term is assumed invariant so that this term can be removed from
the system, the reduced system solved, and a zero placed in the location of the
maximum term.

n

n
{ } K [ ]
M [ ]
x
i
K [ ]
n
M [ ] ( )

n
{ }
x
i
----------------
K [ ]
x
i
------------
n
M [ ]
x
i
-------------
\ .
| |

n
{ } +

n
x
i
--------- M [ ]
n
{ } =

n
{ }
T

n
{ }
T
K [ ]
n
M [ ] ( )

n
x
i
---------

n
{ }
T K [ ]
x
i
------------
n
M [ ]
x
i
-------------
\ .
| |

n
{ }

n
{ }
T
M [ ]
n
{ }
----------------------------------------------------------------------------- =
K [ ]
n
M [ ] ( )

n
{ }
x
i
----------------
K [ ]
x
i
------------
n
M [ ]
x
i
-------------

n
x
i
--------- M [ ] +
\ .
|
| |

n
{ } =
K [ ]
n
M [ ] ( )
Main Index Main Index
91 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
For NORM=MASS, a substitution is made
Eq. 2-102
Without loss of generality, one of the terms in can be set to 0.0 and this row and
column can be partitioned out of the system to produce a reduced system than can be
designated with a superscript bar (
-
)
Eq. 2-103
Eq. 2-102 contains a term that can be derived from the requirement that the
generalized mass is invariant with respect to the design change
Eq. 2-104
Substituting Eq. 2-102 into Eq. 2-104 yields
Eq. 2-105
The total sensitivity for the NORM=MASS method can now be determined by
substituting results from Eq. 2-103 and Eq. 2-105 into Eq. 2-102.
It should be noted that NORM=POINT is not supported for eigenvector sensitivity
analysis.
Buckling Load Factor Sensitivities
The derivation of buckling sensitivity is similar to that of eigenvalue sensitivities. The
equation for determining of the buckling load factor is
Eq. 2-106
Differentiating this expression and solving for the buckling load factor sensitivities
yields

n
x
i
---------
)
`

V
i
{ } C
in

n
{ } + =
V
i
k [ ]
i
M [ ] V
n
{ }
K [ ]
x
i
------------
n
M [ ]
x
i
-------------

n
x
i
--------- M [ ] +
\ .
|
| |

n
{ } =
C
i n
2
n
{ }
T
M [ ]

n
x
i
---------
n
T M
x
i
--------
n
0 = +
C
in
1
2
--- 2
n
{ }
T
M [ ] V
n
{ }
n
{ }
T M
x
i
--------
n
{ } +
\ .
| |
=

K [ ]
n
{ }
n
K
d
[ ]
n
{ } + 0 =
Main Index Main Index


92
Eq. 2-107
As before, this equation is solved semi analytically by approximating the structural
property matrix derivatives using finite differences.
The approximation of the differential stiffness is worth discussing. This stiffness is a
function not only of geometry but of displacement as well
Eq. 2-108
where the displacement is of a particular static analysis that is used in the buckling
analysis.
In the finite difference approximation, can be approximated for the perturbed
configuration as
Eq. 2-109
MSC.Nastran supports two variations for the sensitivity of the differential stiffness,
neither of which is exactly of the form given in Eq. 2-109. In the first, the differential
stiffness matrix is ignored in its entirety. In some situations, this provides an adequate
approximation to the buckling sensitivity and certainly requires less computational
resources. The second alternative is to include the sensitivity of the differential
stiffness using
Eq. 2-110
This form relies on the assumption that the differential stiffness matrix is a linear
function in the displacements so that
Eq. 2-111

n
x
i
---------

n
{ }
T K [ ]
x
i
------------
n
K
d
[ ]
x
i
--------------- +
\ .
|
| |

n
{ }

n
{ }
T
K
d
[ ]
n
{ }
------------------------------------------------------------------------------- =
K
d
[ ] K
d
X U , ( ) [ ] =
K
d
[ ]
K
d
[ ]
x
i
---------------
K
d
X
0
x
i
U U + , + ( ) [ ] K
d
X
0
U , ( ) [ ]
x
i
----------------------------------------------------------------------------------------------------------
K
d
[ ]
x
i
--------------- K
d
X
0
x
i
+ U ( ) [ ] K
d
X
0
U , ( ) [ ]
K
d
X
0
U , ( )
x
i
------------------------------- + =
K
d
x
0
U u + , ( ) [ ] K
d
x
0
U , ( ) [ ] K
d
x
0
U , ( ) [ ]
Main Index Main Index
93 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Experience has shown that this is a more accurate formulation than using differencing
techniques to obtain the effects of the perturbed displacements. The selection of which
method is applied in a buckling sensitivity calculation is based on PARAM DSNOKD.
The default value for this parameter is 0.0, which neglects the differential stiffness
sensitivity. In the case where the analysis is being carried out on a determinant
structure, the differential stiffness matrix is invariant with respect to property changes
so neglecting this term is appropriate. DSNOKD=1.0 uses the formulation of Eq. 2-110
and is recommended when accuracy is important.
Complex Eigenvalue Sensitivity Analysis
The basic equation for complex eigenvalue analysis is
Eq. 2-112
and there is an associated left-handed eigenvalue analysis
Eq. 2-113
where , and denote the complex eigenvalue and right and left complex
eigenvectors, respectively, of the nth mode. Note that the left and right eigenvectors
are distinct and share a common eigenvalue, which can be defined as
Eq. 2-114
where and are the real and imaginary parts of the complex eigenvalue, . The
structural matrices , , and include contributions both from structural
elements as well as direct input via K2GG, B2GG, M2GG, K2PP, B2PP, and M2PP.
Relations used in computing the design sensitivities can be derived by differentiating
Eq. 2-112 and Eq. 2-113 with respect to a design variable. Using to refer to a
perturbation, we obtain, from Eq. 2-112,
Eq. 2-115
Expanding, ignoring higher order terms, and utilizing Eq. 2-113 yields the following
approximation for the variation in the n-th complex eigenvalue
Eq. 2-116
M [ ]p
n
2
B [ ]p
n
K [ ] + + ( )
n
{ } 0 =
M [ ]
T
p
n
2
B [ ]
T
p
n
K [ ]
T
+ + ( )
n
{ } 0 =
p
n

n

n
p
n
p
n
= i +
p
n
K [ ] B [ ] M [ ]

M M + [ ] p p + ( )
n
2
B B + [ ] p p + ( )
n
K K + [ ] + + ( ) + ( )
n
{ } 0 =
p
n

n
{ }
T
M [ ]p
n
2
B [ ]p
n
K [ ] + + ( )
n
{ }

n
{ }
T
2p
n
M [ ] B [ ] + ( )
n
{ }
-------------------------------------------------------------------------------------------------------------
Main Index Main Index


94
where
Eq. 2-117
and and are the real and imaginary variations, respectively, of the complex
eigenvalue. Note that the approximation in Eq. 2-116 does not require complex
eigenvector sensitivities. As the case of modal dynamic analysis of Eq. 2-80 and Eq. 2-
53, the assumption is that enough modes have been retained such that the perturbed
subspace can be adequately represented by the original mode set. For this reason, a
larger mode set than would normally be required for analysis purposes is
recommended for modal complex sensitivity analysis. Because this number is highly
problem-dependent, a logical approach would be to include enough modes until
acceptable accuracy convergence is achieved.
Flutter Sensitivity Analysis
The eigenvalue problem for flutter adds aerodynamic terms to the derivation given
for complex eigenanalysis in the previous section. The flutter equation is given by
Eq. 2-118
where the terms in the above equation are defined in the Flutter Solution
Techniques in Chapter 2 of the MSC.Nastran Aeroelastic Analysis Users Guide.
Flutter sensitivity computes the rates of change of the transient decay rate coefficient
with respect to changes in the design variables. is defined in connection with the
complex eigenvalue
Eq. 2-119
The sensitivity calculation for is expressed in terms of the sensitivities of the
eigenvalue as
Eq. 2-120
The eigenvalue sensitivity computation is conceptually straightforward and
somewhat along the lines of the complex eigenvalue sensitivity given above. The
complete derivation can be found in Aeroelastic Design Sensitivities and
Optimization on page 84 of the MSC.Nastran Aeroelastic Analysis Users Guide.
p
n
= i +

M
hh
p
2
B
hh
1
4
--- cV
Q
hh
l
k
----------
\ .
|
| |
p K
hh
1
2
--- V
2
Q
hh
R

\ .
| |
+ + u
h
{ } 0 =

p
p i + ( ) p
R
= = p
I
+

x
------
1
p
I
-----
p
R
x
---------
p
I
x
-----------
\ .
| |
=
Main Index Main Index
95 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
PSD Response Sensitivities
The calculation of the sensitivities of the PSD responses of Eq. 2-25 is based on
sensitivities calculated from a dynamic sensitivity analysis
Eq. 2-121
where the terms are sensitivities at a particular degree-of-freedom and frequency.
RMS Response Sensitivities
The sensitivity of the RMS quantity versus each design variable x is calculated from
a numerical integration of the PSD sensitivities of Eq. 2-121 and the definition of RMS
response given by Eq. 2-27
Eq. 2-122
Sparse Data Recovery in Sensitivity Analysis
A final aspect of sensitivity analysis bears mentioning simply because of the dramatic
effect it can have on performance and data storage requirements. As finite element
model sizes become even larger, it has become apparent that there is a benefit to be
obtained in only computing minimum required values of physical displacements.
Data recovery refers to the steps that occur following the solution of equations like
Eq. 2-41 in physical coordinates or Eq. 2-78 in model coordinates. The concept of
sparse data recovery will be explained here in the modal context, since it is here that
the most dramatic gain can be achieved. The recovery from modal to physical
coordinates occurs with the matrix multiply given in Eq. 2-77. Sparse data recovery
entails limiting the terms in the matrix to those that are to be used subsequently.
In the context of design sensitivity, the total degrees-of-freedom required in a
sensitivity analysis are a union of sets defined by
The user output requests.
Those required to evaluate the design responses.
Those touched by the design model and needed either to compute the
pseudo-loads or in the sensitivity calculations of Eq. 2-95.
In extreme cases, a finite element model may have millions of degrees-of-freedom
while only a few hundred need to be recovered.
S
j f
x
S
ab
b

H
ja
H
jb
*
H
j a
H
jb
*
+ ( )
a

=
H
u
j
u
j
1
2
---
1
u
j
----
\ .
| |
C
f
S
j f
( )
f

=
[ ]
Main Index Main Index


96
Sparse data recovery is currently employed in three analysis disciplines in SOL 200:
1. Normal Modes
2. Direct Frequency Analysis
3. Modal Frequency Analysis
While other disciplines could be added to this list, the three listed above are
considered the most important.
Main Index Main Index
97 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
2.6 Optimization
A variety of optimization algorithms are available for use in SOL 200. The MSCADS
suite of optimization algorithms is the default set for the 2005 r3 Release of
MSC.Nastran. This is an enhanced version of the ADS suite described in Reference 11.
Prior to the 2005 r3 release, the DOT algorithm from VR&D has been the default
optimization algorithm and the methods of this algorithm are discussed in some detail
in Numerical Optimization in Appendix D. The BIGDOT algorithm (Reference
14.) is available for design tasks as part of the Topology Optmization option in
MSC.Nastran. Guidelines for the selection and use of each of these optimizers is given
as part of the discussion of DOPTPRM on page 231. This subchapter is limited to
a discussion of the interaction between the optimizer and the approximate model and
is applicable regardless of the selected optimization algorithm.
Connection Between the Optimizer and the Approximate
Model
Figure 2-11 is a flow chart of the approximate optimum design task using the specified
optimization algorithm as it is embedded within MSC.Nastran. Note that the
optimizer is called within a do-while loop and that, based on a the INFO parameter,
the approximate model discussed in The Approximate Model on page 101 is used
to provide function values or gradient values. The figure shows the possibility of an
error being encountered by the optimizer that makes further progress impossible. This
is, hopefully, a rare occurrence that is triggered by, for example, attempting an
optimization task that is too big or asking the optimizer to overcome a violated
constraint for which the gradient is zero. In this latter case, there is no possibility of
achieving a feasible design and this usually indicates a user error in specifying the
design task.
As seen in Figure 2-11, if INFO=2, a gradient evaluation is required. This is a request
for sensitivities of the objectives and all the constraints the optimizer considers active.
If INFO is 0 or 1, a function evaluation is required. This is a request for the value of the
objective and all of the constraints for the current values of the design variables.
INFO=0 indicates that the design task is complete, the do-while loop is exited and
control is returned to the MSC.Nastran system.
Main Index Main Index


98
Figure 2-11 Interaction Between the Optimizer and the Approximate Function
and Gradient Evaluations
Numerically Identifying the Active and Violated Constraints
The optimization algorithm makes a determination as to which of the retained
constraints are violated and which are active. The constraints which are neither active
or violated can be ignored in the gradient evaluation of Figure 2-11, thereby reducing
the amount of computations required, the amount of computer memory required and
the size of the mathematical programming task. Figure 2-12 illustrates the concept of
active and violated constraints for a single inequality constraint in a simple two-
Initialization
Optimizer
(MSCADS, DOT or BIGDOT
Fatal
Error
INFO=2
INFO=0
Function
Evaluation
Exit
STOP
Gradient
Evaluation
Y
Y
Y
Y
N
N
N
Main Index Main Index
99 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
design variable space and Figure 2-13 presents the same information in a plot of a
constraint value as a function of a single design variable or search direction. A
constraint is considered active if its numerical value exceeds CT. The default value for
CT is -0.03, but can be changed by the user. Once a constraint is active, its gradient is
included in the search direction computation. An active constraint may subsequently
become inactive if its value falls below CT.
Figure 2-12 Active and Violated Constraints
x
2
X
1
Feasible Region
Infeasible Region
Active Constraint
Inactive Constraint
Violated Constraint
g
j
X ( ) CT =
g
j
X ( ) 0 =
g
j
X ( ) CTMN =
g
j
X ( ) 0 >
g
j
X ( ) 0 <
Main Index Main Index


100
Figure 2-13 CT and CTMIN
The optimizer designates a constraint as violated if its value is greater than CTMIN.
The CTMIN default is 0.003 as is seen in Figure 2-13. Thus, some small constraint
violation (three-tenths of one percent, by default) is tolerated.
Feasible Violated
Active
Numerical Constraint
Boundary
0
CTMIN (0.003)
CT (-0.03)
True Constraint Boundary
x
g x ( )
g x ( )
Main Index Main Index
101 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
2.7 The Approximate Model
MSC.Nastran employs a number of techniques, collectively referred to as
Approximation Concepts, to make design optimization possible for large finite element
models. Design variable linking, reduced basis vectors, constraint screening and load
case deletion have already been described and these approximation concepts all play
a role in reducing the cost of the design task. This subsection deals with the concept of
the Approximate Model that can be thought of as enabling design optimization as it
is performed in MSC.Nastran. MSC.Nastran uses formal approximations to the finite
element analysis and the sensitivity analysis to avoid the high cost associated with
repeated finite element analyses during design optimization. As shown in Figure 2-
11, the optimizer interacts with the approximate model when it requires information.
Given a set of design variables, the optimizer needs information on the function
values (the value of the objective and the values of the retained constraints) and the
gradient values (gradient of the objective with respect to the design variables and the
gradient of the active constraints with respect to the design variables). This section
first introduces the concept of using Taylor Series expansion to approximate the finite
element analysis and follows this with a discussion of how move limits are applied to
try to restrict the changes in the design model to a region where the approximations
are valid. With this background, it is possible to describe the Function Evaluations and
Gradient Evaluations of Figure 2-11.
Formal Approximations
The approximating functions used in MSC.Nastran are based on Taylor series
expansions of objective and constraints. For any function , an infinite series about
a known value in terms of the change in an independent variable can be
written as
Eq. 2-123
In addition to the function value , this series requires that all derivatives at be
known as well. Determining these derivatives may present some difficulty, so the
series is often truncated to a given power in , yielding an approximate
representation of the original function.
For example, if we only included through the first derivative term in the series, we
would obtain a linear approximation
f X ( )
f X
0
( ) x
f X
0
x + ( ) f X
0
( ) =
df
dx
------
X
0
x
d
2
f
dx
2
---------
X
0
+ +
x
2
2!
---------
d
3
f
dx
3
---------
X
0
+
x
3
3!
--------- +
f X
0
( ) X
0
x
Main Index Main Index


102
Eq. 2-124
Since all terms of power and higher have been omitted, the error in the approximation
is on the order of . This situation is shown in Figure 2-14 where the error increases
with increasing values of . Note that the approximation would be exact if the
original function were linear.
Figure 2-14 Errors in Approximating Functions
In design optimization, we are concerned not just with a single independent variable
but rather with a vector of design variables, . Under this condition, the
approximations for the objective and constraints functions become
Eq. 2-125
where a gradient term replaces the first derivative term of Eq. 2-124.
f

X
0
x + ( ) f X
0
( ) =
df
dx
------
X
0
+ x
x
2
x
f f

,
f x ( )
f x
0
x + ( )
x
0
x
0
x +
x
X
F

X
0
x + ( ) F X
0
( ) = F ( )
X
0
+ x
g
j
X
0
x + ( ) g
j
X
0
( ) = g
j
( )
X
0
+ x
Main Index Main Index
103 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
A Simple Linear Design Space
At this point, it is probably worthwhile to take a look at a linearly approximated
design space just to gain a qualitative understanding of the nature of the
approximation. For this example, we can refer to the A Simple Structural Example
on page 20.
Recall that the design variables for the cantilever beam are the base, B, and height, H,
of the beam cross section. If we construct linear approximations to the objective and
constraint equations according to Eq. 2-125, we have
Eq. 2-126
For an initial design at (B= 6, H = 45), the derivatives in the above equation can be
evaluated based on Eq. 1-15 through Eq. 1-17 to yield
Eq. 2-127
The design space resulting from Eq. 2-127 is shown in Figure 2-15, superimposed on
the true design space. From the graph, note that the optimum resulting from this
linear approximation happens to be at a slightly smaller value than the true objective.
The linear approximation has allowed the design to become slightly violated when
compared to the true design space.
V

B
0
B H
0
, H L , + + ( ) V B
0
H
0
L , , ( ) =
V
B
-------
B
0
H
0
,
+ B
V
H
-------
B
0
H
0
,
+ H

B
0
B H
0
H L , + , + ( ) B
0
H
0
L , , ( ) =

B
-------
B
0
H
0
,
+ B

H
-------
B
O
H
0
,
+ H

B
0
B H
0
H L , + , + ( ) B
0
H
0
L , , ( ) =

B
-------
B
0
H
0
,
+ B

H
-------
B
O
H
0
,
+ H
V

B
0
B H
0
H L , + , + ( ) 1.35 = 10
5
2.25 + 10
4
B 3.0 + 10
3
H

B
0
B H
0
H L , + , + ( ) 555.56 = 92.593B 24.691H

B
0
B H
0
H L , + , + ( ) 2.0576 = 0.34294B 0.13717H
Main Index Main Index


104
Figure 2-15 Linearly Approximated Cantilever Beam Design Space
Though not yet an optimal design, the approximate optimum is a useful starting point
for the next approximate optimization cycle.
Move Limits
Since the approximations just discussed are only locally valid, limits are placed on the
amount by which the design is allowed to vary during the approximate optimization.
In MSC.Nastran, move limits are imposed in terms of both allowable changes in
design variables and allowable changes in properties.
Design Variable Limits
Imposing move limits using design variables is natural since the approximate model
is explicit in these quantities. However, move limits on designed properties must also
be considered as well since the analysis model is a function of these properties.
Moreover, these designed properties may be nonlinear functions of the design
variables (Eq. 2-5). A 10% change in a given design variable may equate to a 100% or
Height H (cm)
65
60
55
50
45
40
35
2.5 3 4 5 6 7 7.5
Width B (cm)
50000
100000
125000
150000
175000
V = 200000
H/ B = 12
H = 50
75000
2.54 =

b
700 =
Optimum
Approximate
Optimum

b
X
0

Main Index Main Index

105 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
more change in an analysis model property. Since these large design changes can
easily invalidate the approximate model, move limits on properties must also be
included.
During approximate optimization, each design variable is bounded from above and
below by
Eq. 2-128
where:
These bounds are determined at the beginning of each design cycle according to
Eq. 2-129
where:
The allowable percentage change in the design variable is supplied by DELXV on the
DESVAR Bulk Data entry. This value is optional and if not supplied, defaults to
DELX, which is provided on the DOPTPRM Bulk Data entry. The DELX default is 0.4
or 40% (0.2 or 20% for topology optimization) change in all design variables. The
DXMIN value ensures that the move limits dont become too restrictive when the
design variable value is near zero. The default value for DXMIN is 0.05 (1.0E-5 for
topology optimization), but it, too, can be modified using the DOPTPRM entry.
= the lower bound on the i-th design variable
= the upper bound
MOVE =
XLB = the lower bound on the i-th design variable
XUB = the upper bound
x
i
L
x
i
x
i
U

x
i
L
x
i
U
x
i
L
MAX XLB, x
i
0
MOVE ( ) =
x
i
U
MIN XUB, x
i
0
MOVE + ( ) =
MAX DXMIN, x
i
DELXV ( ) ( )
Main Index Main Index


106
Property Limits
Move limits can be imposed on designed properties as well, but not all properties.
Designed Properties on page 39 has discussed the three types of properties: type-1,
type-2, and type-3. As shown there, type-1 properties are linear in terms of the design
variables (see Eq. 2-2)
while type-2 enable a nonlinear relationship between design variables and properties
(see Eq. 2-5)
and type-3 refers to beam properties that are spawned when using the beam library.
Move limits are always imposed on type-2 and type-3 properties but they are not
imposed on type-1 properties when the property is controlled by a single design
variable
Eq. 2-130
and the user has not imposed bounds on the value of the property.
Move limits are not imposed in this case since the move limit on the design variable
already restricts the change in the property value, i.e., if the DELXV value of Eq. 2-129
imposes a 40% change on the i-th design variable of , the value can never exceed
40% on a given design cycle. Not imposing limits on these properties simplifies the
design optimization task by reducing the number of constraints that have to be
considered.
For the type-1 properties that are a function of more than one design variable and/ or
that have user imposed limits and for type-2 or type-3 properties, bounds are imposed
from below and above the designed properties by
Eq. 2-131
where:
Note: MSC.Nastran 2005 r2 does not place move limits on type-2 and type-3
properties than can change sign. Placing move limits as properties that can
be zero has been shown to create problems
p
j
c
o
= c
i
i

+ x
i
p
j
f X C , ( ) =
p
j
C
o
c
i
x
i
+ =
p
j
p
j
L
p
j
p
j
U

Main Index Main Index
107 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
These bounds are based on percentage changes in the property value at the outset
of the approximate optimization. These bounds are computed as
Eq. 2-132
where:
DELP can be defined on the DOPTPRM entry and has a default of 0.2 for a 20%
maximum property variation. DPMIN can also be defined on the DOPTPRM entry
and has a default value of 0.01.
Figure 2-16 illustrates the situation for a property that is initially close to zero. and
are the move limits resulting from Eq. 2-132. Since these limits are overly
restrictive, the limits based on DPMIN, or and are used instead. Although not
shown in the figure, if PMIN is a numerically greater quantity than , it becomes the
lower bound. Similarly, PMAX will become the upper bound if it is less than .
Figure 2-16 Move Limits on Properties
= the lower bound on the j-th property
= the corresponding upper bound
PMOVE =
PMIN = the lower bound on the property
PMAX = the upper bound
DELP = the maximum allowable percentage change in the property
p
j
L
p
j
U
p
j
0
p
j
L
MAX PMIN p
j
0
PMOVE , ( ) =
p
j
U
MIN PMAX p
j
0
PMOVE + , ( ) =
MAX DPMIN p
j
0
DELP , ( )
p
j
L
p
j
U
p
j
L
p
j
U
p
j
L
p
j
U
1
0
p
j
L
p
j
L
p
j
U
p
j
0
p
j
U
Main Index Main Index


108
Eq. 2-129 and Eq. 2-132 effectively form a box around the current design. This effect
is shown in Figure 2-17 where these move limits are shown for successive design
cycles in a two design variable space. For the first cycle, the approximate optimum is
found to lie at a corner of the box, where the objective is minimized and there are no
active constraints. As a result of the second cycle, one of the constraints is slightly
violated due to errors in the approximation. By the third cycle, a near optimal design
has been found. Of course, the situation is usually more complex than in this simple
design space. The intent of Figure 2-17 is merely to suggest some general features of
the overall process.
Figure 2-17 Sequence of Approximations
Automatic Updates of Move Limits
Parameters related to design variables can be changed using the DOPTPRM entry
(DELX and DXMIN) and the DESVAR entry (DELXV, XLB, XUB). The designed
property limits can be controlled using the DOPTPRM entry (DELP, DPMIN) and the
DVPREL1 or DVPREL2 entries (PMIN, PMAX). This set of constants is used to
recompute the move limits for each design cycle.
1
2
3 4
x
2
x
1
Constraint
Boundaries
F X
0
( )
Main Index Main Index
109 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
At times, the code may automatically adjust these move limits if the design task is
performing poorly. The situation might arise as follows: an approximate problem is
constructed, from which the optimizer determines a corresponding approximate
optimum. Perhaps some of the approximate constraints are critical for this design.
The responses are now evaluated by a finite element analysis, and it is determined that
rather than just critical, these constraints are actually violated. Errors have thus been
detected between the approximate and the true responses.
If this error continues from one design cycle to the next, it can be taken as an indication
that the move limits are probably too wide. Continued constraint violations have an
adverse effect on overall convergence. The move limit-controlling parameters are
updated automatically in MSC.Nastran if the following criteria are satisfied:
1. The current design cycle number is greater than or equal to three.
2. There is at least one violated constraint (violated by more than 2%), and the
level of constraint violation is increasing.
Under these conditions, DELP, DPMIN, DELX, and DXMIN are reduced by one-half
of their current values. The reason for the first condition is that frequently the
optimizer may violate the constraints somewhat as it makes favorable gains in the
objective function in the first few cycles. However, if this condition continues, it may
indicate that the problem is becoming ill-conditioned as a result of excessive move
limits. A corresponding User Warning Message is printed as notification that this
update has occurred (see Modification of Move Limit Parameters on page 358). If
the job is to be restarted, it is recommended that an updated DOPTPRM entry with the
new move limits be included in the restart file.
Implementation of Move Limits
Move limits on independent design variables are applied directly. Since the optimizer
will never propose a value for a design variable outside of its bounds, the lower and
upper move limits computed from Eq. 2-129 are simply input directly to the
optimizer.
Move limits that are imposed on properties are implemented as equivalent
constraints. For the
j-th property in the design model, the equivalent lower and upper bound constraints
are given by
Main Index Main Index


110
Eq. 2-133
where the lower and upper bounds and are given by Eq. 2-132.
Transformation of the Approximate Optimization Task to a
Feasible Design
This section addresses an optimal feature of the approximate optimization task. A
general statement can be made that optimization algorithms perform best when the
design task does not include violated constraints. In fact, some algorithms will fail if
a feasible design does not exist (the algorithms used in MSC.Nastran search for the
best compromise infeasible design so that the transformation to a feasible design is not
a strict requirement). In order to facilitate the use of a general optimization
procedure, such as the ADS code, a simple transformation technique has been
developed to ensure that the optimization task always works in the feasible domain.
The standard optimization task minimizes an objective subject to satisfying a set of
constraints. The transformed problem, designated the Method, involves creating a
design variable that modifies the constraints so that:
and the transforms the objective function so that:
where PENAL is the user defined penalty parameter, is the initial objective
function and is initialized to the maximum initial constraint value. (If this maximum
value is less than CTMIN, the transformation is not applied.) In this way, the
maximum constraint value is never positive while the penalty on the objective acts to
force the value to zero. If there is a feasible design, this technique should lead to it.
If there is not, this will result in a compromise design where the maximum constraint
violation is minimized.
g
L
X ( )
p
j
L
p
j
X ( )
p
j
L
---------------------------- = 0
g
U
X ( )
p
j
X ( ) p
j
U

p
j
U
----------------------------- 0 =
p
j
L
p
j
U

g
j
g =
j
j 1 2 ncon , , = ( )
=

PENAL

0
+

Main Index Main Index

111 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Function Evaluation
The function evaluation proceeds in five steps:
1. The properties are evaluated exactly.
2. Approximate type-1 responses are evaluated.
3. Approximate type-2 responses are evaluated.
4. Approximate type-3 responses are evaluated.
5. The the objective and the constraints are evaluated .
Not all these steps need to be present in a particular design task. Each of these steps is
now described.
Property Evaluation
Given the design variable values, the properties can be computed directly and exactly.
For type-1 properties that are a linear function of the design variables, the properties
are evaluated using a multiply and add operation (see Eq. 2-2). For type-2 properties,
equations that contain constants plus design variable values need to be evaluated on
an individual basis. If beam library dimensions are being designed, it is necessary to
compute the beam properties. Some of these properties are linear in the design
variables and can be computed in a fashion similar to the type-1 properties. Others are
nonlinear, and it is necessary to call the beam library routines to calculate these
properties.
Type-1 Response Evaluation
Three basic types of approximation are available in the computation of type-1
responses:
Direct variable approximations.
Reciprocal variable approximations
Convex linearization
The method used is based on the user specified parameter APRCOD that is discussed
later in the entry, DOPTPRM on page 231.
The formal approximations used in MSC.Nastran are based on simple first-order
Taylor series expansions. The general form of this expansion for a response is
Main Index Main Index


112
Eq. 2-134
The required derivative information is available from the design sensitivity analysis
as outlined in Design Sensitivity Analysis on page 67. Note that the approximate
response, , in Eq. 2-134 is linear in the design variable change . In some
instances, the response actually is a linear function of the design variable and Eq. 2-
134 provides as exact response value. In general, however, we expect some error when
we try to approximate responses that are actually nonlinear in .
Direct Approximations
A direct variable approximation linearizes the function directly in terms of the design
variables. For the j-th response, the direct approximation is written as
Eq. 2-135
where the superscript D indicates a direct approximation.
The quantity represents a total vector move in the design space from the initial
design . The partial derivatives are available from the design sensitivity
analysis.
Reciprocal Approximations
The first-order Taylor series expansions of Eq. 2-134 can alternately be expressed in
terms of reciprocal variables. This choice turns out to be quite useful for those
responses that are inversely proportional to the design variables. This simple idea can
be easily shown in the case of the axially loaded rod element of Figure 2-18 where the
cross-sectional area A is taken as the design variable.
Figure 2-18 Axially Loaded Bar Element
r
j
X
0
x + ( ) r
j
X
0
( ) =
r
j
x
i
-------
X
0 i

+ x
i

r
j
x
0
x + ( ) x
x
r
D
j
X
0
x + ( ) r
j
X
0
( ) =
r
j
x
i
-------
X
0 i

+ x
i

x
X
0
r
j
x
i

L
P
A, E
Main Index Main Index
113 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The axial stress in the bar is equal to P/A, and the displacement is equal to PL/AE. If
these responses are used in the design model, linearizing with respect to the quantity
1/A or 1/E will produce approximations which are exact for both the stress and the
displacement. (The stress is not a function of E so that while the approximation of
stress as a function of E is exact, it is also trivial.) In the general case, of course, these
approximations are not exact due to the static indeterminacy of the structure.
However, for all element types, the proportionality of the stiffness matrices to the
inverse of the sizing quantities forms a reasonable basis for arguing the use of
reciprocal approximations.
Reciprocal approximations can be derived from linear approximations if we first
substitute the intermediate variables . For the approximate constraint of Eq. 2-135,
we get
Eq. 2-136
Setting the intermediate variables to reciprocals of the design variables
Eq. 2-137
yields
Eq. 2-138
where the superscript R indicates a reciprocal approximation.
Convex Linearization
This third approximation method is actually applied on the constraints, but is
discussed here because it bears directly on the method used in the type-1 response
calculation. This method chooses either a direct or reciprocal constraint
approximation depending on which one provides the larger estimation of the
constraint function. In other words, this method chooses the more conservative of the
two approximations.
The direct approximation for the j-th constraint is given in Eq. 2-135 above while the
reciprocal approximation is given in Eq. 2-138.
y
i
r
j
Y
0
y + ( ) r
j
Y
0
( ) =
r
j
y
i
-------
Y
0
y
i
+
y
i
1
x
i
---- =
r
j
R
Y
0
x + ( ) r
j
X
0
( ) =
r
j
x
i
-------
X
0 i

+
x
i
0
x
i
----- x
i
x
i
0
( )
Main Index Main Index


114
Convex linearization is applied on an individual constraint/ design variable basis.
Thus, a combination of direct and reciprocal approximations may be made for a single
constraint.
Since both approximations are readily available (both use the same gradient
information), the choice of which to use can be made by looking at the sign of the
difference between the two for a particular design variable
Eq. 2-139
Since the squared term in the expression is always positive, the choice depends on the
sign of the product:
As stated, convex linearization is only applied to constraints. If the objective function
is obtained from a type-1 response and the convex linearization method has been
chosen, the objective function is calculated using a direct approximation. Also, if a
type-1 response is used in a type-2 or type-3 response, a direct approximation is used
for calculating the type-1 response even when convex linearization is used in the
calculation of constraints directly based on type-1 response.
Special Handling for Eigenvalue/Frequency Response
Eigenvalue (and frequency) responses cannot be reliably approximated using the
either the direct or the reciprocal approximations. This is because the eigenvalue is
sometimes dominated by stiffness terms, which would make it a linear function of a
thickness design variable and sometimes by mass terms, which would make it a linear
function of the inverse of the design variable. MSC.Nastran provides an additional
option in the approximation of eigenvalue responses that is given the name Rayleigh
Quotient Approximation. In a normal response analysis, the eigenvalue can be
expressed using Rayleighs quotient
If
use the direct approximation for since it yields a larger
(more conservative) estimate of constraint value.
If
use the reciprocal approximation since it yields a larger
estimate of constraint value.
g
j
D
X
0
+ x ( ) g
j
R
X
0
x + ( )
x
i
x
i
0
( )
2
x
i
-------------------------
g
j
x
i
--------
X
0
=
1
x
i
----
g
j
x
i
--------
X
0
0
g
i
1
x
i
----
g
j
x
i
--------
X
0
0 <
g
i
Main Index Main Index
115 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-140
where is referred to as the modal strain energy or generalized stiffness and is
referred to as the modal kinetic energy or generalized mass.
Taylor series approximation to the modal strain and kinetic energies can be used to
construct the eigenvalue approximation. It has been found that reciprocal variables
provide a conservative approximation for the strain energies while direct variables are
used for the kinetic energies
Eq. 2-141
Eq. 2-142
And, the eigenvalue approximation by the Rayleigh Quotient is
Eq. 2-143
This Rayleigh Quotient Approximation is the standard technique used in
approximating eigenvalue and normal mode frequency responses. Approximations
using direct and reciprocal approximations are also available, as explained in
DRESP1 on page 244.
Type-2 Response Evaluation
Type-2 responses can be functions of design variables, designed properties, type-1
responses, grid locations, and other type-2 responses. The grid may be a function of
the design variables, so it is first necessary to calculate the grid locations using Eq. 2-
12. It is then a straightforward task to assemble the arguments required by a type-2
response and make the evaluation. When a type-2 response is a function of another
type-2, it is necessary to perform the evaluations in a precise order so that the response
values are available when needed.

T
K

T
M
------------------
U
T
---- = =
U T
U

R
U
0

T K
x
i
-------
i 1 =
n

x
i
x
0i
( )
x
0i
x
i
------- + =
T

D
T
0

T M
x
i
--------
i 1 =
n

x
i
x
0i
( ) + =

RQA
U

R
T

D
------- =
Main Index Main Index


116
An interesting special case occurs when all the design constraints and the objective are
based on type-2 responses and these responses are not a function of the finite element
responses. In this, perhaps academic, scenario the responses can be evaluated exactly
and the optimization task is also exact. Therefore, if no move limits are placed on the
design variables, the final optimal design can be obtained in a single design cycle.
Type-3 Response Evaluation
Type-3 evaluations are identical in form to type-2 evaluations with the exception that
the type-2 evaluations are performed using the internal MSC.Nastran equation
evaluator while the type-3 evaluations are performed using the user supplied external
program. Type-3 responses cannot be a function of type-3 responses, so the recursive
step of the type-2 evaluation is not needed.
Objective and Constraint Value Evaluation
The objective is one of the type-1, 2, or 3 responses so that it is available immediately
once that responses have been evaluated. Constraint evaluations are more involved in
that MSC.Nastran considers four types of constraints:
Constraints on responses
Constraints on dependent design variables
Constraints on properties
Constraints on beam dimensions
The evaluation of each of these types of constraints is now considered in turn.
Constraints on Responses
Once the responses have been evaluated, it is a simple matter to apply Eq. 2-17 to
compute the constraints. An exception to this is when the convex linearization
technique (page 113) is used. In this case, it is necessary to first determine which
approximation concept provides the most conservative constraint value and to then
use that method to evaluate the constraint.
Constraints on Dependent Design Variables
If there are dependent design variables, the lower and upper bounds on these
variables are converted into constraints using
Main Index Main Index
117 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-144
where is the current value of the ith dependent design variable and and
are the bounds imposed on this variable.
These constraints differ from the side constraints of Eq. 1-4 that are applied directly to
the independent design variables by the optimizer.
Constraints on Properties
Property constraints are computed using the relationships given by Eq. 2-133.
Constraints on Beam Dimensions
The beam library calculates beam properties based on beam dimensions. It is
necessary to prevent combinations of dimensions from taking on values that are
physically meaningless. For example, the inner radius of a tube cross section cannot
exceed the outer radius. Table 2-5 documents the constraints imposed on the MSC
supplied cross-sections. Constraints are satisfied when they are < 0.0. (Refer to the
MSC.Nastran Quick Reference Guide for the meaning of the dimensions.)
g
L
DDV
x
i
( )
XLB
i
x
i
( )
XLB
i
------------------------------ =
g
u
DDV
x
i
( )
x
i
XUB
i

XUB
i
-------------------------- =
x
i
XLB
i
XUB
i
Main Index Main Index


118
Table 2-5 Constraints for Beam Library
Section Type Constraint
TUBE DIM2 - DIM1
I DIM4 - DIM2
DIM4 - DIM3
DIM5 +DIM6 - DIM1
CHAN 2
*
DIM4 - DIM2
DIM3 - DIM1
T DIM3 - DIM2
DIM4 - DIM1
BOX 2*DIM4 - DIM1
2*DIM3 - DIM2
CROSS DIM4 - DIM3
H DIM4 - DIM3
T1 DIM4 - DIM1
I1 DIM3 - DIM4
CHAN1 DIM3 - DIM4
Z DIM3 - DIM4
CHAN2 DIM2 - DIM3
2 * DIM1 - DIM4
T2 DIM4 - DIM1
DIM3 - DIM2
BOX1 DIM4 +DIM3 - DIM2
DIM5 +DIM6 - DIM1
HEXA 2 * DIM1 - DIM2
HAT 2 * DIM2 - DIM1
2 * DIM2 - DIM3
Main Index Main Index
119 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Note that the L cross-section type is only available for the PBEAML.
If the user supplies cross-section types that are in addition to the MSC supplied types,
constraints can be specified as a part of the specification (see Adding Your Own
Beam Cross-Section Library in Appendix B). The constraints are applied once for
each designed PBARL while the PBEAML has constraints for each designed section
(possible sections are End A, End B and up to nine intermediate stations). The
constraints are evaluated exactly as listed in Table 2-5 with no normalization.
Gradient Evaluation
The gradient evaluation of Figure 2-11 proceeds in 7 steps:
The gradients of the properties are evaluated.
Gradients of active constraints on type-1 responses are evaluated.
Gradients of active constraints on type-2 responses are evaluated.
Gradients of active constraints on type-3 responses are evaluated.
Gradients of active constraints on dependent design variables are evaluated.
Gradients of active property constraints are evaluated.
Gradients of active beam library constraints are evaluated.
In addition, the gradient of the objective, which is one of the responses is always
calculated.
Not all these steps need to be present in a particular design task. Optimization on
page 97 has provided a definition of active constraints; basically the concept is that the
optimizer selects a reduced set of the retained constraints to deal with when it requires
L DIM3 - DIM2
DIM4 - DIM1
HAT1 DIM3 - DIM1
2* DIM4 - DIM2
2* DIM4 +DIM5 - DMI2
Table 2-5 Constraints for Beam Library (continued)
Section Type Constraint
Main Index Main Index


120
a gradient evaluation. The approximate model takes advantage of this by only
computing the gradients of this reduced set. Each of the gradient steps is now
described.
Gradients of the Properties
Gradients of all the properties are computed with respect to the design variables . This
computation is a simple matter in the case of type-1 properties since they are a linear
function of the design variable (Eq. 2-2) and the gradient is
Eq. 2-145
where the superscript 1 denotes a type-1 property.
Finite difference techniques are applied for the type-2 properties of Eq. 2-5 with
properties being perturbed one design variable at a time to provide
Eq. 2-146
where the notation is meant to imply that the ith design variable is perturbed
while the remaining variables are held at their current values. The gradient of each
type-2 property then is a vector with terms for each independent design variable.
Beam library properties (also called type-3) require chain-rule operation to compute a
property derivative of the form
Eq. 2-147
The second term in the chain rule is obtained from a type-1 gradient as given in Eq. 2-
145. The first term is the sensitivity of the property with respect to the dimension. For
section types in MSC Beam Library, this sensitivity is provided analytically based on
the equations for the beam properties. For example, the area of a tube cross-section is
computed as
Eq. 2-148
Then
p
j
1
x
i
--------- C
j i
=
p
j
2
x
i
---------
f X
0
X
i
C , + ( ) f X
0
x
i
C , ( )
2x
i
---------------------------------------------------------------------------------- =
X
0
x
i
+
p
j
3
x
i
---------
p
j
3
DIM
k
------------------
DIM
k
x
i
------------------
k 1 =
ndi m

=
A DIM1
2
DIM2
2
( ) =
Main Index Main Index
121 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-149
For user supplied cross sections, you have the option of supplying the analytic
gradients or specifying that they are to be computed using central differencing
techniques much along the lines of Eq. 2-146.
Gradients of Type-1 Responses and of Active Constraints from
Type-1 Responses
The gradients of all the type-1 responses are computed with respect to all the design
variables in a two step process. In the first step, the direct sensitivities with respect to
the design variables and the type-2 and type-3 properties are computed. The second
step applies the approximation technique used in the function evaluation to obtain a
gradient consistent with the predicted response
Eq. 2-150
where the first term on the right is available directly from the sensitivity analysis of
Eq. 2-30, as are the
and
terms. The
and
terms are the property gradients just discussed in Eq. 2-148 and Eq. 2-149,
respectively.
Given the gradient of the type-1 response, the calculation of the gradient of the
constraint based on the response depends on the approximation used in calculating
the response. For the direct approximation, the constraint gradient is given by
Eq. 2-151
where, from Eq. 2-17, lower bound constraints have and
while upper bound constraints have and .
A
DIM1
------------------ 2DIM1 =
dr
j
1
dx
i
--------
r
j
1
x
i
--------
r
j
1
p
k
2
---------
p
k
2
x
1
---------
k

r
j
1
p
m
3
----------
p
m
3
x
i
----------
m

+ + =
r
j
1
p
k
2
r
j
1
p
k
3

p
k
2
x
i
p
k
3
x
i

g
k
x
i
---------
1
BND
k
----------------
dr
j
1
dx
i
-------- =
k 2j 1 =
BND
k
GNORM = k 2j = BND
k
GNORM =
Main Index Main Index


122
For the reciprocal approximation, the gradient is
Eq. 2-152
where the subscript and have the same meaning as given above.
If convex linearlization has been selected, the direct gradient of Eq. 2-151 is used if the
product of
while the reciprocal approximation of Eq. 2-152 if this product is less than zero.
Gradients of Type-2 Responses and of Active Constraints on Type-2
Responses
The gradient of a type-2 response is required when the response is the objective, when
the response is constrained directly or when it is used in a type-3 or other type-2
response. The type-2 response can be a function of design variables, designed
properties (both type-1 and type-2 properties), design responses (both type-1 and
type-2 responses) and/ or grid locations. With the exception of the grid locations and
the type-2 responses, the gradients of these components have already been discussed.
The grid location gradient can be obtained directly from the basis vector relationship
of Eq. 2-11. For the sake of this discussion, we can rewrite the type-2 response
equation of Eq. 2-14 into a condensed form of
Eq. 2-153
where ARG can be any argument in the response. A chain rule operation is then
applied to obtain the overall gradient
Eq. 2-154
The is a finite difference change in a component and is the gradient
of the component with respect to the design variable and has already been calculated
( is 1.0 when is . The term is calculated using
g
k
x
i
---------
1
BND
k
---------------
dr
j
1
dx
i
--------
x
i
x
i
0
-----
\ .
|
| |
2
=
k BND
k
x
i
BND
k
---------------
r
j
x
i
------- 0.0 >
r
j
2
f ARG ( ) =
dr
j
2
x
i
--------
f ARG ARG
k
+ ( ) f ARG ARG
k
( )
2ARG
k
-------------------------------------------------------------------------------------------------------
ARG
k
x
i
------------------
k

=
ARG
k
ARG
k
x
i

ARG
k
x
i
ARG
k
x
i
ARG
k
Main Index Main Index
123 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Eq. 2-155
FDCH and FDCHM are 0.001 and , respectively.
The gradients of the constraints based on type-2 two responses are simply
Eq. 2-156
where the subscript and the value have the same meaning as they do for the
type-1 response of Eq. 2-151.
Gradients of Active Type-3 Responses
The gradient of a type-3 responses is required when the response is the objective or
when the response is constrained and the constraint is active. The type-3 response can
be a function of design variables, designed properties (both type-1 and type-2
properties), design responses (both type-1 and type-2 responses) and/ or grid
locations. The same chain rule formulation as that given in Eq. 2-154 for type-2
gradients is applicable for type-3 gradients. Unlike the beam library, analytical
gradients are not supported for the type-3 responses so that the change in the response
due to a change in an argument is calculated using the same central differencing
technique as in Eq. 2-154. Gradients of the constraints based on type-3 responses
follow the method shown in Eq. 2-156.
Gradients of Active Dependent Design Variables
The gradient of an active constraint imposed on the j-th dependent design variable
with respect to the ith independent variable is given by combining Eq. 2-1 with Eq. 2-
144 and differentiating to get
Eq. 2-157
ARG
k
FDCH ARG
k
FDCH ARG
k
FDCHM > ( )
FDCHM otherwise

=
1.0
6
g
k
dx
i
---------
1
BND
k
---------------
dr
j
2
dx
i
-------- =
k BND
k
dg
L
DDV
dx
i
------------------
c
i

XLB
--------------- =
dg
u
DDV

i
------------------
c
i
XUB
---------------- =
Main Index Main Index


124
Gradients of Active Property Constraints
The gradient of an active constraint imposed on the j-th designed property with
respect to the i-th design variable is given by
Eq. 2-158
where the property gradients have been defined in Eq. 2-145., Eq. 2-146, and Eq. 2-
147.
Gradients of Active Beam Library Constraints
From Table 2-5, it is seen that the gradient of the j-th beam library constraint with
respect to the i-th design variable is given by
Eq. 2-159
This is a straight-forward calculation. As an example, suppose a CHAN section type
is being designed and the DIM4 dimension is defined as being equal to . Then
the gradient of the first CHAN constraint of Table 2-5 with respect to is simply
Eq. 2-160
dg
L
P
dx
i
---------
1
p
j
L
----------
p
j
x ( )
x
i
---------------- =
g
u
P
x
i
---------
1
p
u
L
----------
p
j
x ( )
x
i
---------------- =
g
BL
x
i
-------------
g
BL
DIM
j
-----------------
DIM
j
x
----------------- =
0.1 x
5

x
5
g
CH
1
BL
x
5
----------------- 2.0 0.1 2.0 = =
Main Index Main Index
125 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
2.8 Tests for Convergence
Since structural optimization is an iterative process, numerical criteria must be
established to determine when the overall process has converged. There are two levels
at which convergence is tested: the first and lower level is at the optimizer level; the
second and higher level is with respect to the overall design cycles. This section is
concerned with design-cycle level convergence. Convergence at the optimizer level is
discussed in Numerical Optimization in Appendix C.
Convergence of Design Cycles: Hard and Soft
Convergence
Figure 2-1 has indicated that a convergence check is made immediately following the
structural analysis. This is a somewhat simplified representation in that, in reality, two
methods are used to test for convergence with respect to overall design cycles. These
methods are denoted as soft convergence and hard convergence as shown in the
modified version of Figure 2-1 shown in Figure 2-19 which highlights the
convergence aspects of the design cycle loop. Soft convergence is based on the results
of the approximate optimization, while hard convergence is based on finite element
analysis results.
Main Index Main Index


126
Figure 2-19 The Design Cycle Loop with Emphasis on the Convergence Tests
Recall that at the outset of each design cycle, a finite element analysis and a sensitivity
analysis are performed. The approximate model is then constructed followed by an
optimization with respect to this approximation. Once a corresponding approximate
optimum is found, the finite element model, in terms of the design properties and grid
locations, is updated. The proposed design is submitted for another finite element
analysis to compute the updated responses, thus marking the beginning of the next
design cycle.
Hard convergence compares the results of this most recent finite element analysis with
those from the previous design cycle. Since this test compares the exact results (within
the limits of the finite element analysis) from two consecutive analyses, the
conclusions are said to be based on hard evidence. Since this test is conclusive, this is
the default test for determining whether or not to terminate the design-cycle process.
Initial
Design
Improved
Design
Structural
Analysis
Constraint
Screening
Sensitivity
Analysis
Approximate
Model
Optimizer
Finite Element
Analysis
Many Times
Hard
Convergence
Exit
Exit
Soft
Convergence
Y
Y
N
N
Main Index Main Index
127 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Soft convergence compares the design variables and properties output from the
approximate optimization with those of the input to the approximate optimization. If
these variables/ properties have not changed appreciably, another finite element
analysis may not be warranted. For example, if a particular analysis model has a high
solution cost and the optimizer has not changed the design to any significant degree,
it may be permissible to forego the final finite element analysis. The results from the
previous analysis are probably of sufficient validity for design purposes; therefore,
soft convergence may be a preferable and cost effective exit point. Soft convergence is
always tested, but, by default, does not result in program termination. Setting
PARAM SOFTEXIT to YES will result in program termination following a
determination that soft convergence has been achieved.
On the initial cycle of the Design Cycle loop shown in Figure 2-19, the hard
convergence test is skipped and program flow proceeds directly to the sensitivity
analysis and approximate optimization. Once an approximate optimum is obtained, a
soft convergence test is performed to compare this design with the prior design.
If soft convergence has not terminated the design process, a finite element evaluation
of the new, proposed design is performed followed by constraint evaluation and
screening. If the analysis results are not appreciably different from those of the prior
cycle, hard convergence is achieved, and the design cycle process is terminated. If the
hard convergence criteria are not satisfied, the design cycle process continues.
If neither hard nor soft convergence is achieved, the design process will continue until
convergence is indicated or until the maximum allowable number of iterations is met.
This default maximum number of iterations is 5 but can be changed on the DOPTPRM
entry using DESMAX.
The following paragraphs discuss soft and hard convergence decision logic.
Soft Convergence Decision Logic
A schematic of the soft convergence decision logic is shown in Figure 2-20. The result
of the test is a true or false value for the logical variable SOFTCV. The design cycle
terminates only if SOFTCV is TRUE and the parameter SOFTEXIT is YES.
Main Index Main Index


128
Figure 2-20 Soft Convergence Decision Logic
From Figure 2-20, soft convergence is based solely on the relative changes in the
properties and design variables. The definitions of the parameters in the in the figure
appear in Table 2-6 along with their default values. These values may be modified
using the DOPTPRM entry.
Table 2-6 Convergence Criteria Parameters (the P and P-1 Superscripts Refer to
the Current and Previous Design Cycle)
Internal
Variable
Definition Parameters Default
CHGOBJ CONV1 0.001
(1.0E-5 for
topology
optimization)
ACHOBJ CONV2 1.0E-20
CHGPRP CONVPR 0.001
CHGPRP < CONVPR
CHGDV < CONDV
SOFTCV = TRUE SOFTCV = FALSE
Y
Y
N
N
OBJ
P ( )
OBJ
P 1 ( )

OBJ
P 1 ( )
-------------------------------------------------------
OBJ
P ( )
OBJ
P 1 ( )

max
1 i NPROP
P
i
P ( )
P
i
P 1 ( )

P
i
P 1 ( )
---------------------------------------
\ .
|
|
| |
Main Index Main Index
129 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The assumption can be made that if the changes in the design variables and properties
are not appreciable, the objective and the constraints are unchanged and there is
nothing to be gained from continuing.
Hard Convergence Decision Logic
Hard convergence is used to decide whether or not the design cycle process should
continue, providing that the maximum allowable number of design cycles (DESMAX)
has not yet been reached. The satisfaction of hard convergence is always sufficient to
stop the design process.
A flowchart of these criteria is shown in Figure 2-21 with the corresponding
definitions of the terms appearing previously in Table 2-6. Even though this test may
indicate convergence, note that one of three possible conclusions may be reached.
CHGDV CONVDV 0.001
(1.0E-4 for
topology
optimization)
CONMAX GMAX 0.005
Table 2-6 Convergence Criteria Parameters (the P and P-1 Superscripts Refer to
the Current and Previous Design Cycle)
Internal
Variable
Definition Parameters Default
max
1 i NDV
x
i
P ( )
x
i
P 1 ( )

x
i
P 1 ( )
-------------------------------------
\ .
|
|
| |
max
k
g
k
x ( ) { }
Main Index Main Index


130
Figure 2-21 Hard Convergence Decision Logic
The first check is with respect to the relative and absolute changes in the objective
function. The reasoning behind the application of an OR test is based on a
consideration of the objective function magnitude. If the objective is large, such as the
weight of a heavy machine part in kilograms, converging to within a plus or minus
range of a kilogram is probably quite sufficient. Forcing convergence to within a
fraction of a kilogram may be meaningless as well as expensive. On the other hand,
there are situations where the objective function is a very small number (e.g.,
minimization of the difference between analysis and test results). For such cases,
minimization of the absolute change may be more meaningful than the relative
change.
Convergence
Non unique If
CHGPRP > CONVPR or
CHGDV > CONVDV
CHGOBJ < CONV1
or
ACHOBJ < CONV2
Best Compromise,
Infeasible Design
Found
Continue with
Optimization
Y
Y
Y
N
N
N
CONMAX < GMAX
CHGPRP < CONVPR
and
CHGDV < CONVDV
Main Index Main Index
131 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
If the objective has not changed appreciably in either an absolute or a relative sense, a
check is performed to ensure that the maximum constraint value is less than its
maximum allowable. If this criterion is satisfied, then hard convergence is achieved.
Non-Unique Optima
Even if hard convergence has been achieved, the user should check to see if the
relative changes in properties CHGPRP or the relative changes in design variables
CHGDV are satisfied. If they are, then the design process has converged to a unique
design. If not, a non unique design may have been found.
From a design perspective, convergence to a non unique design is extremely useful
information. This would indicate that the optimum properties could be modified by
the engineer with correspondingly little change in the objective function and/ or
constraints. (In other words, the optimal design exhibits low sensitivity with respect
to changes in the design variables.) A design space of this type is shown in Figure 2-
22 where contours of the objective and active constraint boundary have similar
curvatures in the region of the optimum. In situations such as these, the design may
be modified to take advantage of available sheet metal gauges or tube sizes without
violating the active constraint(s) and with correspondingly little change to the
optimum objective. Of course, any proposed changes should be subjected to an
analysis and the results checked carefully to ensure that other performance constraints
are not violated.
Figure 2-22 Non-unique Optimum
Non-unique
Optimum
F 10 =
F 15 =
F 20 =
g x ( )
F x
*
( )
x
2
x
1
Main Index Main Index


132
Compromise Infeasible Design
Returning to the hard convergence decision logic of Figure 2-21, note that the code
only checks the relative changes in properties and design variables if the limit on the
maximum constraint value is not satisfied. The purpose of this check is to determine
whether the design is changing for the case of violated constraints. If the design is still
varying, optimization can continue in order to try to overcome the constraint
violation(s). However, if the design is not changing, then we are at a point in the
design space that represents a best compromise solution among the violated
constraints. This situation is shown in Figure 2-23 where the optimal design is that
which minimizes the sum of the constraint violation.
Figure 2-23 Design Space, No Feasible Solutions
Given the implications of hard convergence, a careful review of the output is advised
in order to determine which convergence conditions have been met. Just because the
process has been terminated does not necessarily imply that a unique, feasible design
has been found. A discussion of the output from both the soft and hard convergence
checks, given in Design Optimization Output starting on page 329, is particularly
relevant to the stopping criteria in that it contains a list of all the possible messages
that can occur at the end of an optimization task.
F X
*
( )
g
1
X ( )
x
2
g
2
X ( )
x
1
Main Index Main Index
133 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
2.9 Special Topics
This final section deals with special topics that are definitely part of MSC.Nastrans
design optmization capability, but do not fit neatly into one of the previous sections.
Discrete Variable Optimization
Most optimization algorithms are based on the assumption that the design variables
or quantities that are modified in search of an optimal design can vary continuously
within a defined range. Such is the case with the gradient-based algorithms that are
the focus of this Users Guide. However, the practicing engineer must choose from
discrete values that most closely match the optimal vector of design variables returned
by the optimizer. The constraints of mass production usually do not afford the
engineer the luxury of specifying a 1.84 mm plate thickness; instead the engineer must
make a practical choice from among several of the nearest available gauge sizes. A
number of subsequent analyses will often need to be performed to verify that the
design goals have been reasonably met and that none of the critical performance
constraints have been violated. Discrete variable optimization seeks to avoid such
tedium by allowing the engineer to specify a list of available gauges beforehand,
leaving the task of selection and validation to the optimizer.
Although some theoretical algorithms promise a high degree of precision (such as the
Branch and Bound Method), most algorithms are not computationally efficient and
therefore, are unsuitable for use with large-scale finite element analysis programs
such as MSC.Nastran. Instead, four practical and cpu-efficient methods have been
implemented:
1. Rounding up to the nearest discrete design variable
2. Rounding off to the nearest discrete design variable
3. Conservative Discrete Design (CDD)
4. Design of Experiments (DOE)
The first two methods simply automate the simple rounding process a user might
employ after a continuous optimization and require no new analyses. For the CDD
method, each variable is independently set to the discrete values that bracket the
continuous variable result. An approximate analysis is carried out for the discrete
variable above the continuous value and one with the discrete variable below the
continuous value. The constraint results of these two analyses are compared and the
discrete variable is chosen that gives the minimum value for the maximum constraint.
Main Index Main Index


134
This is repeated for each design variable so that 2*nddv (where nddv is the number of
design variables that can take on discrete sizes) approximate analyses are carried out
for the CDD approach.
The DOE method implemented in MSC.Nastran makes use of the concept of
orthogonal arrays. As with the CDD method, the postprocessing task is limited to
searching for designs where only the discrete variable values just above and just
below the continuous variable are selected. In this case, however, any combination of
discrete variables can be used. This seemingly straightforward task can still be
difficult as there are

possible vectors that could be analyzed. With a modest 25
variables, functions evaluations would be required for an exhaustive
search. At 10 evaluations per second, this would require over a month of computing
and is clearly impractical. The implementation of DOE employed in MSC.Nastran
employs an exhaustive search when nddv is 16 or less. Above this value, an
Orthogonal Array concept is employed to select

candidate arrays that provide a
representative sampling of the overall design space. Clearly this is an approximation,
but the thinking is that this will provide adequate coverage of the possible discrete
solutions that it will not be far off from the true optimum that would be obtained
from an exhaustive search.
Each of these methods is a postprocessing operation of the continuous, optimal
design. Moreover, the engineer can specify that such discrete variable postprocessing
be applied only to the final design, or to any number of the previous, intermediate
designs that lead up to the final design. Useful design information can often be
obtained by observing the progression of (hopefully) converging optimal discrete
values, because any one of the intermediate solutions may actually result in a more
usable design than the optimal solution itself. Note that computation of these
intermediate discrete solutions has no influence on subsequent designs because the
next design cycle is always based on the previous continuous solution and not the
discrete solution.
Discrete variables can also be mixed with continuous ones, allowing mixed-discrete
optimization. A typical example is that of combined sizing and shape optimization;
gauge sizes must be discrete, while the variables that control shape may vary
smoothly and continuously. Furthermore, both mathematical programming (MP) and
fully stressed design approaches can be used in conjunction with discrete variable
optimization.
2
nddv
3.35 10
7
2
25
( )
2
16
Main Index Main Index
135 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Fully Stressed Design
Fully Stressed Design (FSD) is a concept that has long been regarded as an automated
design algorithm that can quickly produce a design that satisfies key design
requirements. Its implementation in MSC.Nastran complements the long standing
mathematical-programming (MP) approach to automated design that are the focus of
this Users Guide. It utilizes many of the same procedures to produce a quick-look
design at a small fraction of the computational costs. This efficiency can be exploited
by requesting that the redesign take place with literally thousands of independently
designed properties, whereas an MP approach is realistically limited to approximately
one thousand design variables. The method is particularly useful in the design of
aerospace structures where the overriding requirement is that the structural weight be
minimized. Although the proposed design that results from the FSD technique may
not be usable from a manufacturing standpoint because it has neglected key design
considerations, it can be thought of as providing a lower bound estimate on the
amount of structural material required to achieve the imposed design conditions. The
output of the FSD algorithm can also be used as an excellent starting design for a more
general MP design task.
The basic resizing concept of FSD can be described as follows
Eq. 2-161
where:
And the old and new superscripts refer to before and after the resizing.
The implementation of FSD within MSC.Nastran is embedded within the SOL 200
Design Optimization. Design variable and constraint conditions are applied using the
existing Bulk Data entries and Case Control commands that have been developed for
the MP algorithm. Two parameters control the FSD algorithm:
t = designed property
i = index to indicate which property contains the design parameter and the
design response
,
= actual and allowable response quantities, such as stress
= a real number
t
i
new

i

all
( )

t
i
old
=

all
0.0 < 1.0 ( )
Main Index Main Index


136
The FSD algorithm shares much of its user interface and dataflow with the basic
design optimization capability. The capabilities available for FSD are a subset of those
available for general multidisciplinary optimization and include:
1. FSD is applicable to static aeroelastic analyses. Other analyses can be
included in the input and will be evaluated in the analysis associated with
the FSD cycles and included the redesign in any subsequent MP design
cycles.
2. Multiple subcases and multiple boundary conditions are supported.
3. Composite materials are supported.
4. Allowable limits can be placed on element stress and/ or strain.
5. PROD areas and shell thicknesses s (PSHELL, PSHEAR and PCOMP
thicknesses) are supported with FSD.
6. For composites, the thicknesses of individual PCOMP layers can be resized
using the FSD algorithm.
7. All elements sharing a single property ID are sized to a single value that
satisfies all the imposed conditions. Further, if a number of properties are
controlled by a single DESVAR, the single DESVAR value will be sized to
satisfy all imposed conditions.
8. Minimum and maximum size limits are imposed (e.g., design variable
bounds and property limits).
The above list implies the following additional limitations:
1. Beam cross-sections are not designed. Ply orientation angle cannot be used
as a design variable for FSD.
2. If an element is constrained, but the corresponding property for the element
is not resized, the constraint is ignored in the resizing.
3. If a property is designed, but has no imposed constraints, the property value
is unchanged by the resizing algorithm.
FSDALP Relaxation parameter applied in Fully Stressed Design. (Real, 0.0 <
FSDMAX< 1.0, Default = 0.9)
FSDMAX Specifies the number of Fully Stressed Design Cycles that are to be
performed. (Integer, Default = 0)
Main Index Main Index
137 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
4. Shape design variables and material and connectivity properties are not
supported in FSD.
A guideline for the use of the FSD capability is to limit its application to 5-10 design
cycles and follow it up, if practical, with MP design cycles. This is performed
seamlessly in SOL 200 by the resizing algorithm switching from FSD to MP after
FSDMAX design cycles or after the FSD design is considered converged. The MP
algorithms have much more generality in application and are also known to be able to
achieve a lower weight structure. This is not practical when the FSD design task has
thousands of design variables since this is likely to overwhelm the MP algorithm. The
default value of 0.9 for FSDALP is usually adequate to achieve quick convergence.
Lower values, such as 0.5, can be used to achieve a completely converged solution.
Topology Optimization
Unlike sizing and shape optimization, topology optimization finds an optimal
distribution of material, given the package space, loads, and boundary conditions.
These methods have grown rapidly in popularity and application in recent years and
topology optimization methods have been discussed in a large number of
publications. See the material cited in References 2. and 9. in References in
Appendix A for an overview of topology optimization.
The topology optimization capability in MSC.Nastran 2005 is based on the
increasingly popular density approach to topology optimization. In the density
method, Youngs modulus E and density are used as intermediate design variables
for each designable finite element. The actual design variable x is the normalized
density that links Youngs modulus E and density for designable finite elements
using the following relationships
Eq. 2-162
where and are respectively the fully solid Youngs modulus and density. A
penalty factor p is introduced to enforce the design variable to be close to a 0-1
solution when p>1.0. The penalty factor p usually takes values between 2 and 4.
The general topology optimization problem available in MSC.Nastran follows that of
Eq. 1-1 and Eq. 1-2 but the side constraints are set as:

0
x =
E E
0
x
p
=

0
E
0
x
i
1.0 i 1 2 N , , , =
Main Index Main Index


138
is a small positive number to prevent the stiffness matrix singularity.
Although the problem statement is very similar, topology optimization does have a
special TOPVAR Bulk Data entry to define the design variable/ property relationship
and two special responses that address what are viewed as classical topology
optimization capability:
1. a fractional mass response
2. compliance (load times displacement response
Manufacturability Constraints with Topology Optimization
Topology optimization entails selecting finite elements that provide design that best
meets the design requirements while discarding the rest. This can lead to designs that
are theoretically optimal but that are impractical from a manufacturing constraint.
This has been addressed by adding manufacturing constraints to the topology
optimization task. The constraints include minimum member size, symmetry
constraints, casting constraints (also called draw direction constraints), and extrusion
constraints. The user interface that imposes these manufacturability constraints is
provided in the bulk data description of TOPVAR Topological Design Variables
on page 300, while examples are shown in Topology Optimization on page 513.
Minimum Member Size (Constraints)
The minimum member size constraint is mainly used to control the size of members
in topology optimal designs. Avoiding thin members enhances the simplicity of the
design and hence its manufacturability. Minimum member size is more like quality
control than quantity control.
Casting Constraints (Draw Direction Constraints)
Topology optimized designs can contain cavities that are not achievable by casting or
some machining processes. The casting constraints allow users to impose die draw
direction constraints and prevent hollow profiles so that the die can slide in a given
direction.

Main Index Main Index

139 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Extrusion Constraints
It is often desirable to produce a design with a constant cross-section along a given
direction. This constraint is particularly essential for a design manufactured through
an extrusion process. By using the extrusion manufacturing constraints in topology
optimization, constant cross-section designs can be obtained for solid models
regardless of the boundary conditions, loads, and finite element mesh.
Symmetry Constraints
It is often desirable to design a symmetric component or system. However, regular
topology optimization cannot guarantee a perfect symmetric design, even if the
design space, boundary conditions, and loads are symmetric. By using symmetry
constraints in topology optimization, a symmetric design can be obtained regardless
of the boundary conditions or loads. These symmetric constraints can also be used for
irregular finite element meshes.
Advanced Numerical Techniques in SOL 200
This section deals with some specialized techniques that are available in MSC.Nastran
for solving large finite element tasks. The application of these techniques in SOL 200
is only involved with the analysis phase with no specific features for design sensitivity
and optimization. Two techniques, DMP (distributed memory parallel) and ACMS
(automated component modal synthesis) are discussed here briefly to show the
impact they can make of optimization tasks.
Three distributed memory parallel (DMP) techniques are available for SOL 200:
internal parallel for FRRD1 module, frequency domain parallel Lanczos method and
degrees-of-freedom (DOF) domain parallel Lanczos method. All computing nodes
for a DMP job are turned on only in the analysis phase of the design cycle in SOL 200.
During the design sensitivity and optimization phase of a design cycle, the master
node is the only active computing node and all other nodes are idle. Note that the
Adjoint Sensitivity Analysis for Grid Responses on page 88 indicates that it is
possible to solve for the adjoint solution vectors during the analysis. In this case, the
unavailability of DMP during the sensitivity analysis is less significant.
Internal Parallel for FRRD1 Module
FRRD1 is the solution generator for frequency response analysis. For each forcing
frequency, FRRD1 goes through identical steps to produce solution vector(s).
Therefore, the work assigned to each processor is equal to the total number of forcing
Main Index Main Index


140
frequencies (NF) divided by the number of processors (NP). If the results of NF
divided by NP is not an integer, one or more processors will need to pick up one extra
forcing frequency. Hence, the maximum load unbalance among all processors in
internal parallel for FRRD1 module is the solution time for one forcing frequency.
Although this DMP technique is supported in both direct frequency (DFREQ) and
modal frequency (MFREQ), it has the biggest impact for DFREQ jobs.
To request NP number of processors for frequency response SOL 200 job, add
dmp=NP to the job submittal line.
As an illustrative example, a vehicle model with a half million DOFs in the analysis
set has been exercised using a varying number of processors. The direct frequency
analysis has 12 forcing frequencies. A performance study on an IBM SP3 produced
the following results.
Overhead for DMP in FRRD1 occurs at the beginning and the end of FRRD1. Note
that the overhead can be greatly influenced by the workload of the system and
network. The overhead is shown here is to indicate that the overhead can become
significant and the effectiveness of DMP reduces as the number of processors
increases. In this case, the FRRD1 module dominates the overall computing
requirements so that there is a significant payoff from running this module in the
parallel mode.
Note: It is possible to select nodes for the DMP job. The defaults for the hosts are
machine dependent. See hosts in Keywords and Environment
Variables in Appendix B of the MSC.Nastran 2005 r3 Installation and
Operations Guide for more details.
Number of
Processors
Total Elapsed
Time
Speed-up
Elapsed Time
in FRRD1
Overhead for
DMP in
FRRD1
1 252 min. 9 sec. 1 208 min. 4 sec. 0
2 139 min. 48 sec. 1.8 105 min. 44 sec. 9 min. 25 sec.
4 101 min. 29 sec. 2.5 67 min. 16 sec. 16 min. 6 sec.
Main Index Main Index
141 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
Frequency Domain Parallel Lanczos Method
Frequency domain parallel Lanczos in SOL 200 is available for ANALYSIS= MODES
and other analyses that use the modal formulation (ANALYSIS=MFREQ, MTRANS,
MCEIG and FLUTTER) .
The following Executive Control Statement
DOMAINSOLVER MODES (PARTOPT=FREQ)
must be added to the beginning of the Executive Control Section.
Main Index Main Index


142
For example:
NASTRAN buffsize=16385 $
domainsolver modes (partopt=freq) $
SOL 200
Furthermore, the number of processors (NP) must be specified using the dmp=NP
keyword on the submittal line. The Lanczos method must be used.
A vehicle model with a half million DOFs in the analysis set is used to indicate of
benefits of frequency domain parallel Lanczos. This is an MFREQ job with 12 forcing
frequencies. A performance study on an IBM SP3 produced the following results:
In this case, the READ module is less dominant in the overall total job elapsed time,
but there is still a significant speedup with more processors.
DOF Domain Parallel Lanczos Method
The DOF domain parallel Lanczos method is an alternative technique for extracting
eigenvalues in parallel mode. SOL 200 also supports this option for ANALYSIS =
MODES and other analyses that use the modal formulation (MFREQ, MTRANS,
MCEIG, and FLUTTER).
The following Executive Control Statement
DOMAINSOLVER MODES (PARTOPT=DOF) $
must be added to the beginning of the Executive Control Section.
For example:
NASTRAN buffsize=16385 $
domainsolver modes (partopt=dof) $
SOL 200
Number of
Processors
Total Elapsed
Time
Speed-up
Elapsed Time in
READ Module
1 245 min. 27 sec. 1 178 min. 57 sec.
2 217 min. 45 sec. 1.13 133 min. 16 sec.
4 158 min. 43 sec. 1.55 79 min. 35 sec.
Main Index Main Index
143 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The number of processors (NP) must be specified using the dmp=NP keyword on the
submittal line. The Lanczos eigenvalue method must be used.
A vehicle model with a half million DOFs in the analysis set is again used for this
example. This is an MFREQ job and has 12 forcing frequencies. A performance study
on an IBM SP3 produced the following results:
Matrix Domain Automated Component Modal Synthesis
(MDACMS)
MDACMS is an advanced Lanczos solver introduced in MSC.Nastran 2005. You can
find a detailed discussion in ACMS Now Available in the Matrix (DOF) Domain in
the MSC.Nastran 2005 Release Guide. You can use this feature in a SOL 200 run by
specifying the DOMAINSOLVER ACMS (partopt=dof) command. It is available for
Analysis=MODES and/ or =MFREQ and both serial and parallel processing are
supported. A parallel run is activated when DMP = n and n > 1.
This feature has been tested with an NVH optimization task with a 4.2M dofs, for an
eigenanalysis up to 300 Hz that produced 1960 modes. Three separate runs were
performed. The following plot shows that a parallel (4 processor) MDACMS SOL 200
job achieves more than 6 times speedup relative to a Geometric Domain ACMS
(GDACMS) SOL 200 run (reduced time from 37 hours to 6 hours) while the serial
MDACMS SOL 200 run reduces to 14 Hrs for a 2.5 times speedup. It was not possible
to execute this problem at all without the use of some type of advanced numerical
technique.
Number of
Processors
Elapsed Time Speed-up
Elapsed Time
in READ
Module
1 245 min. 27 sec. 1 178 min. 57 sec.
2 228 min. 42 sec. 1.07 149 min. 40 sec.
4 188 min. 49 sec. 1.30 95 min. 3 sec.
Main Index Main Index


144
Superelement Optimization
With few exceptions, design modeling for superelement sensitivity and optimization
is a largely transparent operation in Solution 200. That is, design models spanning
multiple superelements can be expressed using the same tools and methods for non
superelement models discussed in the previous sections. This is a useful tool in any
large-scale design task.
This section introduces the differences in superelement versus non superelement
design modeling. Limitations are summarized at the end of the section.
Supported Superelements
Primary and image superelements only may be used in connection with the design
model. Primary, image as well as external superelements may all be used in the
analysis model.
MDACMS vs. GDACMS in SOL200
14
37
6
32
14
5
0
5
10
15
20
25
30
35
40
GDACMS MDACMS(S) MDACMS(P)
E
l
a
p
s
e
d

T
i
m
e

(
H
o
u
r
)
One Design Cycle One FE Analysis
Main Index Main Index
145 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The external superelement restriction occurs because external superelements are only
known to the code via their structural stiffness, mass, damping, and load matrices.
Not having any of the properties or other modeling data on hand prevents referencing
any of this data in connection with the design model. Although external
superelements may be used in connection with the analysis model, their properties are
assumed to be constant with respect to the design model.
Design Variables in Superelement Design Modeling
You can relate design variables to properties just as in the case with non superelement
models by using either linear relationships specified on DVPREL1 Bulk Data entries
or user-defined equations on DVPREL2 entries. Design variable-to-property relations
may be local to a given superelement or may span superelement boundaries. For
example, a design variable may be related to a particular property entry which is, in
turn, used in a number of superelements in the model.
You can also define design variable-to-grid relationships using DVGRID Bulk Data
entries. A given design variable may be related to grid sets that span superelement
boundaries. A DVGRID entry may reference both interior and exterior grids of
primary superelements. DVGRID entries are the only way to specify shape basis
vectors in superelement design models. The other methods-direct input of shapes,
geometric boundary shapes, and analytic boundary shapes--are not supported.
For secondary superelements, the same property and/ or grid variations prescribed on
the primary superelement also apply to the corresponding image superelements. All
references to external grids of secondary superelements are ignored, however. This
stems from the fact that the superelement mapping matrix is used to compute the
secondary superelement matrices using the primary superelement.
Design Responses in Superelement Design Modeling
You can select design responses using DRESP1 and/ or DRESP2 Bulk Data entries. A
single DRESP1 entry can be used to select a number of design responses by referring
to multiple grid locations, property IDs, or elements. The responses identified on a
single DRESP1 entry can be from multiple superelements.
For weight and volume type responses, the applicable superelement IDs (SEIDi), or
ALL must be specified. Weight or volume responses can be computed for the entire
model or just a subset defined by superelement ID reference. The default is the
Main Index Main Index


146
residual structure, or Superelement 0. By default, only those elements belonging to the
residual superelement are included in weight and volume computations. The design
responses input to DRESP2 equations can span multiple superelements or load cases.
Superelements and Constraint Screening
For superelement sensitivity, the constraint screening criteria are applied on an
individual superelement basis. That is, if a DRESP1 entry lists several element, grid,
or property IDs which span superelement boundaries, the screening criteria will
apply individually to all superelements that share responses listed on this DRESP1
entry. If separate sets of constraint screening parameters, TRS (truncation threshold),
and NSTR (number of retained constraints per region) are required for different
superelements, then separate sets of DRESP1 and DCONSTR entries should be used
to define superelement-specific response groups.
Case Control
The simplest way to incorporate superelements is to use the SUPER = ALL command
in Case Control. This automatically satisfies a number of design optimization Case
Control requirements. You can also use an expanded subcase structure if necessary,
with each subcase pertaining to a superelement or group of superelements. If you
choose the latter approach there are some additional requirements that must be
satisfied:
Each superelement subcase must define an analysis type using the
ANALYSIS = command.
For ANALYSIS = MODES, the eigenvalue constraint must be called out from
each MODES subcase using the DESSUB command.
The METHOD command included in the residual subcase must also be used
in the upstream superelement subcases. This requirement also holds for
Component Modes Synthesis (CMS) and Generalized Dynamic Reduction
(GDR).
Limitations
The limitations that apply to superelement design modeling are summarized as
follows:
Main Index Main Index
147 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
The design model may not reference external superelements. External
superelements can be part of the analysis model but are considered invariant
with respect to changes in the design model.
Image superelements have the same design variations and design responses
as defined on the referenced primary superelement. The exception is
external grids, which are invariant for the image superelements.
Shape basis vectors can only be defined using the DVGRID entry.
Eigenvalue sensitivity is evaluated for the residual superelement. Since the
residual contains the boundary degrees-of-freedom from all superelements,
the computed eigenvalue sensitivities pertain to the entire structure.
Eigenvalues computed during upstream superelement reduction are not
available as design responses. Guyan reduction (the default) is exact for
stiffness properties but only approximate for mass. Thus, you may want to
consider using advanced reduction methods such as component modes
synthesis to improve the sensitivity analysis accuracy.
Synthetic responses defined via DRESP2 and DEQATN entries cannot have
arguments that span superelement boundaries. This restriction is similar to
the non superelement case where responses from different subcases may not
be arguments to a user-defined equation.
Multiple boundary conditions are not permitted in superelement design
modeling. This restriction stems from a similar restriction in superelement
analysis.
The DCONSTR and DSCREEN entries have no provisions for specification
of superelement IDs. If separate response allowables or screening criteria are
to be applied, the DRESP1 entries should define responses that do not span
superelement boundaries.
Special Considerations When Designing One-Dimensional
Bending Elements
The seemingly simple BAR and BEAM straight 1-D bending elements in MSC.Nastran
have some special behaviors that need to be pointed out if they are to be effectively
designed. To begin, there are four property types that are associated with these
elements:
1. The PBAR defines basic bending properties of a CBAR element.
Main Index Main Index


148
2. The PBARL defines dimensions of MSC supplied or user-supplied section
types for the CBAR element.
3. The PBEAM defines bending properties for a more complete 1-D element in
that it allows definition of the shear center, the neutral axis, center of gravity,
warping and taper. The PBEAM property entry is referenced by the
CBEAM. The taper, in particular, makes the design of the PBEAM somewhat
complicated, although MSC has greatly simplified the design model
requirements for the PBEAM in recent releases.
4. The PBEAML defines dimension of MSC provided or user-supplied section
types for the CBEAM element. The availability of these libraries greatly
simplifies the data preparation required from the user. All of the features of
the PBEAM, including taper, are available with the PBEAML.
For all of the discussion of this section, the design is assumed to be performed using
the PNAME option for the fifth field of the DVPREL1 and DVPREL2 entries. The
alternative specification of an FID or and EPT location is deemed obsolete.
PBAR
The design of the PBAR is straightforward. Typically, the users design variables are
cross-sectional dimensions, necessitating the use of type-2 properties in specifying the
properties as a function of the design variables. The following considerations can be
mentioned when designing PBARs:
Only use type-2 properties when they are needed. This is because each type-
2 property is treated as an additional design variable in the design sensitivity
process (this is discussed in Internal Representation of the Sensitivity
Coefficients in MSC.Nastran on pp. 53-54). This places a performance and
numerical accuracy burden on the sensitivity analysis.
Keep in mind that changing a beam dimension affects not only the standard
properties such as area and moments of inertia, but also the stress recovery
locations as well. These latter properties are typically linear in the cross
section dimensions and can therefore be designed using the DVPREL1 entry.
The generality of the PBAR makes it possible to make assumptions about the
relationships among the section properties. One could postulate, for example, a thin-
walled beam where, to a first approximation, all the properties are linear in terms of
the wall thickness. Alternatively, one could select the area of the beam as the design
variable and then specify that the bending moment of inertia is proportional to the
square root of the area.
Main Index Main Index
149 CHAPTER 2
Fundamentals of Design Sensitivity and Optimization in MSC.Nastran
PBARL
The design of the PBARL is the easiest from the users point of view. Typically, the
user specifies designed beam dimensions as linear functions of the design variables
(nonlinear functions of beam dimensions with respect to the design variables are not
permitted) and the rest of the design is performed by the beam library. Appendix B
describes how you can add your own beam cross section types and it provides insight
into the operations that must occur to cause a change in the beam dimension to be
reflected in the beam properties. Briefly, MSC.Nastran determines which properties
of the beam cross section are affected by each design variable and constructs what are
referred to as spawned design properties to be used in the design. These are
properties that have not been explicitly designed by the user, but can be inferred from
the property/ dimension relationship. A determination is made if this
property/ dimension relationship is linear or nonlinear so as to use the less expensive
and more exact linear relationship when it is valid.
A drawback of using the PBARL is that it can produce more designed properties than
may be necessary for the design task at hand. For example, the torsional stiffness of
the BAR may have no effect on the response of the structure and a user might not
include its design when a PBAR is used. The PBARL will always design this property
if it is affected by a designed dimension. Perhaps a better example of this is that stress
recovery locations that are affected by a designed PBARL dimension are always
designed while a user may find it necessary only to define one or two of these points.
In most cases, the design of these extra properties is not a large concern, but it does
become an issue when hundreds of PBARLs are being designed in a large finite
element model.
PBEAM
The design of the PBEAM is complicated by the possibility that the beam can be
tapered. The original requirements for designing the PBEAM were quite involved
and non-intuitive, but MSC has simplified these requirements so that the design of
constant section PBEAMs follows the same basic procedures as those for the PBAR.
Older documentation that refers to the need to, for example, design both ends of a
constant section beam is now obsolete and is superseded by this document.
When designing a tapered beam property, the station is identified in parenthesis. As
examples, specifying A(A) or A on a DVPREL1 or DVPREL2 entry indicates that area
of end A of the beam is to be designed, I1(B) indicates that the plane-1 inertia at end B
is to be designed while D2(6) indicates that the z-location at the D stress recovery point
for the 6th station is being designed.
Main Index Main Index


150
As mentioned, the design of constant section beams is now quite similar to that for the
PBAR. It should be kept in mind that the user input for the PBEAM is used to
determine whether a beam can be tapered. If only END A data are provided on the
PBEAM entry, the beam is regarded as constant section and only END A properties
can be designed. Similarily, if END B or other station data are provided on the
PBEAM entry, the beam is regarded as tapered and care must be taken to explicitly
design each property that is expected to vary. As a final comment, if the PBEAM
contains specification of SO and X/ XB for END B and other stations but blank fields
are used to specify the properties, the beam is still regarded as a constant section. This
input can be used when it is desired to obtain stress output for a constant section beam
at an intermediate station.
PBEAML
The design of the PBEAML is much simpler than that of the PBEAM in that it is only
necessary to indicate which beam dimensions are to be changed and the beam library
performs the remainder of the design task. Appendix B can provide insight into how
this design occurs. As with the PBARL, it should be kept in mind that all the
properties that are affected by the designed dimension will be designed as well,
regardless for whether the user feels this is important. For example, the DIM3
dimension that specifies the height of an H type element affects the C1,D1,E1 and
F1 stress data recovery locations in a linear way and the A,I1,I2,J,K1,K2,CWA and
CWB in a nonlinear fashion. It is likely that some of these properties are not important
in the design or perhaps the stress recovery points are redundant, but they will be
designed in any case. Again, this is not a significant issue unless the models are large
with many designed PBEAML entries.
The PBEAML can be tapered and it is the PBEAML entry that governs whether the
beam is permitted to taper. Like the PBEAM entry, if the PBEAML provides data at
END B and/ or other intermediate stations, it is considered tapered. If there is only
END A data, the beam is considered constant section and only END A dimensions can
be specified on DVPRELi entries. If the PBEAML is tapered, only dimensions
explicitly invoked on DVPRELi entries will be designed. One distinction between the
PBEAM and PBEAML is with regard to taper. As mentioned above, the PBEAM is
considered a constant section if the only inputs at END B and/ or intermediate stations
are SO and X/ XB. For the PBEAML, if any input (including SO and X/ XB) is provided
at END B or an intermediate station, the beam is considered tapered.
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
CHAPTER
3
Developing the Design Model for Sensitivity
and Optimization
I Overview of Design Modeling
I Case Control for Design Optimization
I Defining the Design Variables
I Relating Design Variables to Properties
I Relating Design Variables to Shape Changes
I Designating the Design Responses
I Defining the Objective Function
I Defining the Constraints
Main Index Main Index


152
The previous chapter provided an in-depth description of how Design Sensitivity and
Optimization has been implemented within the MSC.Nastran product. A key concept
in that description was that of the design model (The Design Model on page 39).
The design model is simply a formal statement of allowable changes that can be made
to a structure during the search for an optimal design. It also places limits on these
allowable changes, and defines limits on the structural responses. Constructing the
design model requires good engineering judgment if the design results are to be of any
practical use.
The essential features of the design model are that it must
Define the design variables that may be modified.
Describe the relationship between the design variables and the analysis
model properties and/ or grid locations (for shape optimal design).
Identify and/ or develop responses that are to be used in the design.
Define the objective function, which provides a scalar measure of design
quality.
Place bounds (constraints) limiting the design responses to an acceptable
range.
This chapter describes each of these features in greater detail, with reference to the text
user interface that is used to define the design model. In connection with this chapter,
you will want to refer to Input Data on page 203 that explains the text interface in
detail, Example Problems in Chapter 7, and MSC.Patran Support for Design
Sensitivity and Optimization in Appendix C.
The examples in Chapter 7 should all be available on your system. Running some of
these in connection with the following discussions is a good way to begin to discover
the many design modeling options available to you in MSC.Nastran.
Main Index Main Index
153 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
3.1 Overview of Design Modeling
Given the wide range of problem types that may be addressed using design sensitivity
and optimization, the process of generating a design model is hardly a rote or
mechanical operation. However, most of the operations are sufficiently similar that
one might generate a flowchart outlining the design modeling process as follows:
Main Index Main Index


154
Figure 3-1 Design Modeling Process
The following sections discuss each of these items in greater detail.
Define the analysis disciplines to be used
for Design Optimization
(Executive Control: SOL 200;
Case Control: ANALYSIS = )
Define the design variables
(Bulk Data: DESVAR, DLINK, DDVAL)
Relate the design variables to allowable
structural variations:
for properties (Bulk Data: DVPREL1,
DVPREL2, DVMREL1, DVMREL2,
DVCREL1, DVCREL2)
for shape (Bulk Data: DVGRID,
DVBSHAP, DVSHAP, BNDGRID)
Define the design responses
(Bulk Data: DRESP1, DRESP2, DRESP3)
Define the objective (Case Control: DESOBJ)
and the constraints (Bulk Data: DCONSTR,
DCONADD;
Case Control: DESGLB, DESSUB, DRSPAN)
Provide any necessary parameter overrides
and constraint values
(Bulk Data: DOPTPRM, DSCREEN, DTABLE)
Main Index Main Index
155 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
3.2 Case Control for Design Optimization
Case Control commands in MSC.Nastran provide the overall direction for the analysis
to be carried out as well as the data recovery operations to be performed. In this
context, a powerful feature of design sensitivity and optimization in MSC.Nastran is
that you can specify that a structure be subjected to a number of different analysis
types for a number of various load conditions. The optimizer will consider the results
of all of these analyses simultaneously when proposing an improved design. This
approach is often described as multidisciplinary design optimization.
As an example, you may have a component that is subjected to two static loading
conditions. The component must also satisfy requirements on natural frequency
bounds as determined from a modal analysis. Further, the structure might also be
subjected to transient loading conditions for which peak displacement responses
might be of concern. We can quite easily introduce these different analysis types in
Solution 200 (the solution sequence for design optimization) with the following Case
Control packet:
SOL 200
cend
spc = 100
DESOBJ(MIN) = 15
ANALYSIS = STATICS
subcase 1
DESSUB = 10
displacement = all
stress = all
load = 1
subcase 2
DESSUB = 20
displacement = all
strain(fiber) = all
load = 2
subcase 3
ANALYSIS = MODES
DESSUB = 30
method = 3
subcase 4
ANALYSIS = MTRAN
DESSUB = 40
method = 4
dload = 4
tstep = 10
begin bulk
Main Index Main Index


156
Note some of the features in this example:
Case Control Output Requests in Design Optimization
In addition to the analysis-related Case Control commands, output requests might
also appear, such as:
In general, case control requests are for data recovery only; and are unnecessary for
design optimization. All necessary data recovery is automatically performed based on
the design responses identified in the Bulk Data (identifying the responses is
discussed in Designating the Design Responses on page 188). You do not need to
SOL 200 This is an Executive command indicating Solution 200, the solution
sequence for design sensitivity and optimization, is to be invoked.
ANALYSIS This Case Control command indicates the analysis discipline to be
used for a particular subcase (if appearing above a subcase level, all
subsequent subcases assume the same ANALYSIS command until
changed). ANALYSIS may assume any of the following values:
STATICS Statics
MODES Normal Modes
BUCK Buckling
DCEIG Direct complex eigenanalysis
DFREQ Direct Frequency
MCEIG Modal complex eigneanalysis
MFREQ Modal Frequency
MTRAN Modal Transient
SAERO Steady Aeroelastic
FLUTTER Flutter
DESOBJ Selects the design objective from a response defined in Bulk Data with
DRESP1, DRESP2 or DRESP3.
DESSUB Selects constraints to be applied at a particular subcase level from a
DCONSTR or DCONADD set defined in Bulk Data.
displacement= all
strain(fiber)= all
Main Index Main Index
157 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
explicitly request this data recovery unless you want to view the results. There are
some special cases however that you need to know about where output options do
play a role in specifying design responses:
If various output forms are possible (e.g., strain(strcur) or strain(fiber)),
design optimization will use the defaults unless a Case Control request to the
contrary is provided. To use the STRAIN example, STRAIN(STRCUR)
representation is the default. If STRAIN(FIBER) is provided, data recovery
and design optimization responses are both computed using this form. Two
other instances like this are the VONMISES vs. MAXS option on the
STRESS/ STRAIN commands and the (REAL or IMAG) vs. PHASE option of
many case control commands in a Dynamic Response analysis. Defaults are
listed in the MSC.Nastran Quick Reference Guide.
Shape optimization and superelement optimization each have some
additional Case Control considerations beyond those related to data
recovery. Design Variables in Superelement Design Modeling on
page 145 discusses the Case Control structure for superelement design
models. In addition, Example Problems in Chapter 7 includes examples of
both shape and superelement optimization.
Main Index Main Index


158
3.3 Defining the Design Variables
In design optimization, design variables are the quantities that are modified by the
optimizer during the search for an improved design. In design sensitivity analysis,
rates of change of responses are computed with respect to the design variables.
Defining the design variables is part of the design modeling task of the engineer. Once
defined, the design variables may be used to
Describe analysis model property variations.
Define shape variations.
Add additional terms to the design task that are not explicitly related to the
finite element analysis.
This third category allows you to stretch your imagination when applying Design
Optimization in MSC.Nastran. The availability of type-2 and type-3 responses allows
you to consider design conditions that are not directly related to the finite element
analysis. Although you probably do not want to optimize text book mathematical
programming tasks using MSC.Nastran, it is not outside the range of possibility. On
a more realistic note, the example of Acoustic Optimization on page 478 uses an
added design variable in a novel way to minimize the maximum acoustic pressure in
the analysis.
As the optimizer changes the design variables, it does not know it is changing a
structural model-it is only performing a mathematical search. It is the design engineer
who has defined exactly how the structure will change. For example, a single design
variable might have been used to describe a number of bar cross-sectional areas,
several plate element thicknesses as well as a fillet radius. As the optimizer changes
this single variable, the entire structure changes! Expressing allowable structural
variations with an economy of design variables is part of the challenge of design
modeling.
A Design Variable is defined with a DESVAR Bulk Data entry. From the Bulk Data
listings in Input Data on page 203, notice that the DESVAR entry provides an ID, an
initial value for the variable, bounds (or side constraints), an optional move limit for
the approximate optimization and, for discrete variable optimization, a pointer to a
list of discrete values the design variable can assume. The bounds limit the region of
search with respect to each variable as
Eq. 3-1
The optimizer will never propose a design that violates these bounds.
x
i
L
x
i
x
i
U

Main Index Main Index
159 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
The following DESVAR entries define a pair of variables and with initial values
of 0.05 and 0.03, respectively.
The label fields are strictly to help you in characterizing the variable and you have
complete freedom in the name you specify for the variable. In this case, we can assume
these design variables will be used to define an area and a thickness, although the
DESVAR entry does not actually provide these relations (this is up to the DVPREL1
and DVPREL2 entries). Bounds have also been defined as
Eq. 3-2
The DELXV field, which is 0.2 for the AREA1, limits the amount this design variable
can change during an optimization cycle by 20%. The THICK1 variable has this field
blank so the move limit for this second variable is the value of the DELX parameter
you can provide as a default for all of the variables on the DOPTPRM entry. If you
dont specify the DELX parameter either, the default move limit is 40%.
There is one more optional field on the DESVAR entry that enables you to perform a
discrete optimization task as explained in Special Topics on page 133. This is the
DDVAL attribute on the DESVAR entry and the integer value points to a DDVAL Bulk
Data entry with this ID. The presence of this integer triggers discrete variable
optimization that produces a design with design variable values chosen from the
discrete set of real numbers given on the DDVAL entry.
The entire set of DESVAR entries can be thought of as defining a vector of design
variables or
Eq. 3-3
The quantity n is often referred to as the dimension of the design space. The greater
the dimension the more complicated the optimization task. The optimizers task is to
find the best configuration for these variables.
A question that comes up frequently is: How many design variables can MSC.Nastran
handle in a design task? There is no single answer for this question. The MSC.Nastran
code is written to avoid fixed limits and the design sensitivity and optimization
$DESVAR ID LABEL, XINIT, XLB, XUB, DELXV, DDVAL
$
DESVAR, 10, AREA1, 0.05, 0.01, 0.1, 0.2
DESVAR, 11, THICK1, 0.03, 0.01, 0.08
$
x
10
x
11
0.01 x
10
0.1
0.01 x
11
0.08
X x
1
x
2
x
n
, , , [ ]
T
=
Main Index Main Index


160
capability is no exception to this. Further, the answer depends on the computer
resources you have available, your tolerance for expending these resources and your
time waiting for the results and also on the type of problem you are trying to solve.
Given all that, it is still useful to provide a guideline. Problems with up to a thousand
design variables are considered routine in MSC.Nastran while problems that have
several thousand variables can be considered extreme. In cases where the design task
is easy, (i.e., find the set of design variables that minimizes the weight while
satisfying a single displacement limit), its possible to consider upwards of 10,000
variables.
The above paragraph refers to the DOT and ADS algorithms that are included as part
of the design optimization module. If the design task has several thousand design
variables, we recommend that the topology optimization module also be licensed. As
its name implies, the main purpose for this module is to perform topology
optimization but the BIGDOT optimization algorithm that is made available by this
option has been applied to design tasks with one million design variable.
Design Variables and the Finite Element Model
Recall that the problem we are trying to solve is defined by the basic optimization
problem statement of Eq. 1-1 through Eq. 1-5. Note that the design objective and
constraint functions are expressed as functions of design variables. How is this the
case in structural optimization?
Assume that we are interested in minimizing the total structural mass for a particular
mechanical component. This will be our objective function. We know that the mass
is a function of the properties used to describe the component as well as its shape. This
can be expressed as
Eq. 3-4
where the vector is the collection of properties that describe the model and the
vector is the collection of grid point coordinates.
In the next sections, we will show how the properties and shape of the structure can
be expressed as functions of design variables or
Eq. 3-5
By direct substitution of Eq. 3-5 into Eq. 3-4,
Eq. 3-6
M M P G , ( ) =
P
G
P P X ( ) =
G G X ( ) =
M M P X ( ) G X ( ) , ( ) =
Main Index Main Index
161 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
or simply,
Eq. 3-7
To summarize, as the optimizer changes the design variables, the analysis model
properties and shape will also change as defined by the design model. The modified
properties and shape result in changes to the computed responses that we have used
to define the objective and constraint functions. Based on the modified objective and
constraints, the optimizer can measure the amount of design improvement.
Topology Design Variables
The discussion in this section has been terms of the DESVAR Bulk Data entry that is
used in standard design optimization. For topology optimization, there typically is a
design variable for every finite element and it would be impractical to ask the user to
define each of these individually. Instead, a TOPVAR Bulk Data entry is used as
described in TOPVAR Topological Design Variables on page 300. This entry
identifies a property ID that can be associated with many finite elements. In this case,
design variables are created internally for each of these elements.The initial values
and upper and lower bounds on each of these design variables are obtained from the
TOPVAR entry. The density and theYoungs Modulus of the element are then
governed by the relationships shown in Eq. 2-162.
M M X ( ) =
Main Index Main Index


162
3.4 Relating Design Variables to Properties
For the analysis model to vary as the design variables are changed, its properties
and/ or its shape must be expressed in terms of the design variables. You need to
provide these relations as part of the design model specification. This section
discusses analysis model property changes. Relating Design Variables to Shape
Changes on page 166 discusses shape variations.
Analysis model properties are defined as those quantities which appear as real
numbers on Bulk Data property entries. As discussed in Overview of
Fundamentals on page 34, three different types of properties are supported for
MSC.Nastran design sensitivity and optimization: element properties, material
properties, and connectivity properties. Element properties are items such as plate
thicknesses, area moments of inertia and elastic spring stiffnesses while material
properties are, for example, Youngs modulus and Poissons ratio. Connectivity
properties are on element connectivity entries and include such items as beam offsets
and concentrated masses.
You can relate design variables to analysis model properties in either of two ways:
with linear relations or with MSC.Nastrans equation input capability. A linear
relation is described using one of the DVxREL1 (Design Variable-to-Property
RELation, type-1) Bulk Data entry, where the x is replaced with a P for element
properties, an M for material properties and a C for connectivity properties.
As a review from The Design Model on page 39, linear design variable-to-property
relations are of the form
Eq. 3-8
where is the j-th property expressed as a linear combination of i design variables.
As described as part of the DVPREx1 descriptions (see Bulk Data Entries in
Chapter 4), a special case is provided for the case when C
0
and C
i
(i >1) are 0.0 and the
user wishes to use the actual property value for C
1
. In this case, if the term PVAL is
used as input for C
1
, the actual property value is used as the coefficient.
An equation, or type-2 relation, is defined using a DVxREL2 Bulk Data entry which
references a DEQATN (Design EQuATioN) entry. A property expressed using these
relations can be written as
Eq. 3-9
p
j
C
o
= C
i
i

+ x
i
p
j
p
j
f X C , ( ) =
Main Index Main Index
163 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
where the j-th property is a function of a collection of design variables and constants.
The function is defined using a DEQATN Bulk Data entry.
The same design variables may appear in either type-1 or type-2 relations. The block
diagram of Figure 3-2 may be helpful in understanding the role of the DVPREL1 and
DVPREL2 entries in defining property variations with the DVCREL1, DVMREL1,
DVCREL2, and DVMRLE2 entries having a similar structure.
Figure 3-2 Design Variable-to-Property Relations
The DVPREL1 entry defines linear property variations and requires only design
variable input. All necessary constants are supplied directly on the entry. On the other
hand, the DVPREL2 entry relies on an equation to describe the (generally) nonlinear
property variations. This equation may have design variables (DESVAR) as well as
table constants (DTABLE) as input. As shown in Figure 3-2, the result is again an
analysis model property variation.
In design variable-to-property relations, analysis model properties are referenced by
property entry, rather than on an element-by-element basis. This relation is shown
schematically in Figure 3-3. The property IDs MID1 through MIDm are referenced on
the DVMREL1 and DVMREL2 entries. Since each of these property groups may
contain a large number of elements (EID1 through EIDn), you can control many
elements by using only a small set of DVMREL-type entries. For example, to modify
the material density, all that is required is to reference a MAT1 ID on a DVMREL
entry. All elements that use that material will then change accordingly.
DVPREL2
DEQATN DTABLE
Analysis
Model
Properties
DESVAR
DVPREL1
p
1
{ } C
0
{ } = C
1
[ ] x { } + p
2
{ } f X C , ( ) { } =
Main Index Main Index


164
Figure 3-3 Design Model Reference by Property Entry
You must identify a selected property entry item by its name on the property Bulk
Data Entry, where the name is given in the description of the entry in the
MSC.Nast ran Quick Reference Guide. For example, the area of a beam on a PBAR has
the name A while the thermal expansion coefficient on a MAT1 entry also has the
name A. Table 2-1 through Table 2-3 list the available property names.
To illustrate both type-1 and type-2 design properties consider the following
hypothetical example that calls for designing a plate under the assumption that a
change in the thickness of the plate cause a change in a concentrated mass that is used
to estimate such things as paint wiring and rivets. Specifically suppose the relation
between the mass and stiffness is given as:
and and are 0.1 and 0.5 respectively. The analysis model and the design model
can be written as:
PSHELL 100 1 0.25
CONM2 200 20 0.35
$
DESVAR 10 THICKP 1.0 0.01 10.0
DVPREL1 100 PSHELL 100 T
10 0.25
DVMREL2 200 CONM2 200 M
DESVAR 10
DTABLE A0 A1
DTABLE A0 0.1 A1 0.5
DEQATN 40 MASS(X,A0,A1) = A0 + A1 * SQRT(X)
DVMREL1,2
EID1
EID2
.
.
.
EIDn
MATID1
MATIDm
.
.
.
ass a
0
a
1
t ( )
1 2
+ =
a
0
a
1
Main Index Main Index
165 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
It is not necessary to define constants using the DTABLE entry, but this can enhance
the readability of the DEQATN when it is used in a large model. Also note that it is
not necessary for the value of the designed property be the same as the corresponding
analysis value. In the above example, the CONM2 has an M value of 0.35 while the
designed M value is
The designed property value always take the precedence over the analysis property.
See Special Considerations When Designing One-Dimensional Bending
Elements on page 147 for additional comments on this topic.
0.1 .5(1.0)
1 2
+ 0.6 =
Main Index Main Index


166
3.5 Relating Design Variables to Shape Changes
The Design Model on page 39 has provided a discussion of how shape changes can
be made within MSC.Nastran and has introduced the concept of Basis Vectors and
outlined the four alternative methods that can be used to specify these basis vectors:
Manual grid variation.
Direct input of shapes.
Geometric boundary shapes.
Analytic boundary shapes.
This section enlarges on this earlier discussion by identifying the specific bulk data
entries that are used for each of these options. All of the methods, except for manual
grid variation, use the concept of auxiliary model in some form. Therefore the
auxiliary model concept is first discussed before preceding to the details of the various
methods.
Auxiliary Models in Shape Optimization
An auxiliary model is an additional finite element model used to generate shape basis
vectors. Rather than tediously specifying shape changes on an individual grid-by-grid
basis, auxiliary models are a tool that can be used to simplify this process. An auxiliary
model usually shares the same geometry, element connectivity, and (possibly)
material type as the original structure. However, boundary conditions usually differ.
Applying loads to this structure will result in sets of displacement vectors . When
we think of the translational displacement components of these vectors as
representing individual components of grid motion, we realize that the vectors
can quite conveniently be used as basis vectors for shape optimization. For the direct
input of shapes method, the auxiliary model is exercised in a separate job submittal
and the results input into the shape optimization task. For the geometric boundary
shapes and analytic boundary shapes methods, the auxiliary model is integrated
with the shape optimization task so that the shape vectors are generated is the same
job submittal as is used to perform the optimization task.
As a simple example, consider the cantilever structure of Figure 3-4. The analysis
model is clamped along the left-hand edge and tip-loaded on the right. Suppose we
would like to redesign the shape of the structure by modifying the profile of the lower
edge. In order to do so, we can use an auxiliary model to generate a few characteristic
profiles and use these as our shape basis vectors.
U { }
U { }
Main Index Main Index
167 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Figure 3-4 Simple 2-D Cantilever
Figure 3-5 shows the corresponding auxiliary model and its boundary conditions,
exclusive of the loading. The fixed edges on the left and top sides guarantee that the
left-hand support and the horizontal upper edge will not change during shape
optimization. The support in the x direction along the right edge allows the depth of
the beam to change without changing its length. We can now load the bottom edge in
a way to produce a range of shapes or basis vectors.
Figure 3-5 Auxiliary Model for the Simple 2-D Cantilever
A uniform taper component might be generated by applying SPCDs as shown in
Figure 3-6. Here, the displacements have been plotted to show the basis vector
characteristics. Visualization of these basis vectors prior to their use in connection
with shape optimization is highly recommended. See Appendix C for a discussion of
how MSC.Patran can be used to view shape basis vectors.
B
A
D
C
Main Index Main Index


168
Figure 3-6 Uniform Taper Basis Vector
A more arbitrary shape basis vector could be generated by loading an individual grid
along the lower edge. In order to avoid an inflection point in the resulting deflection
pattern, it is advisable to add some very stiff beams to the bottom edge of the auxiliary
model. The produces a shape as shown in Figure 3-7. A number of these shapes
produced by loading other grids along this edge might comprise a useful set of shape
basis vectors.

Figure 3-7 Point-Load Basis Vector
Modeling Methods (Shape Basis Vector Definition)
Shape optimization in MSC.Nastran assumes the engineer is starting with a
reasonably good design but would like to investigate ways in which the shape of the
part, or even the entire structure, might be modified in order to better meet the design
goals.
Although large shape variations can be investigated in MSC.Nastran, it is important
to note that in a practical design environment, shape changes are often limited by
numerous design considerations. Manufacturability of the part, interference from
neighboring components of the structure, aesthetics, and so on, often limit the degree
to which the structural shape can be modified.
P
Main Index Main Index
169 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Shape optimization in MSC.Nastran allows you to investigate these real world
types of shape variations. It will, for example, let you determine the optimum
placement and size of cutouts to lighten a structure. Unfortunately, it cannot tell you
how many holes to use or otherwise affect the topology of the model.
Four different modeling methods are available in MSC.Nastran. The purpose of each
method is the same: to define a basis vector (or a set of basis vectors) for shape
optimization. The methods differ in terms of their user interfaces and whether or not
the basis vectors are automatically updated on each design cycle. The following
subsections briefly outline each approach, its strong as well as its possible weak
points, and its requirements for use.
Manual Grid Variation
In this method, DVGRID entries alone are used to describe shape changes. For the
simple linear taper of Figure 3-6, the DESVAR/ DVGRID data is defined as follows:
As mentioned, this represents a significant amount of data that must be generated in
some fashion. The benefits, drawbacks and data preparation steps for this method are
as follows:
Benefits
Since this can be considered the lowest-level approach, its strength lies in its
generality. Using DVGRIDs alone, the designer has direct control over every designed
grid point in the model (that is, every grid point whose location is to change during
shape optimization.)
DESVAR 10 LINTABP 1.0 0.01 10.
DVGRID 10 11 1.0 0.0 0.4 0.0
DVGRID 10 12 1.0 0.0 0.2667 0.0
DVGRID 10 13 1.0 0.0 0.1333 0.0
DVGRID 10 15 1.0 0.0 0.8 0.0
DVGRID 10 16 1.0 0.0 0.5333 0.0
DVGRID 10 17 1.0 0.0 0.2667 0.0
DVGRID 10 19 1.0 0.0 1.2 0.0
DVGRID 10 20 1.0 0.0 0.8 0.0
DVGRID 10 21 1.0 0.0 0.4 0.0
DVGRID 10 23 1.0 0.0 1.6 0.0
DVGRID 10 24 1.0 0.0 1.067 0.0
DVGRID 10 25 1.0 0.0 0.5333 0.0
DVGRID 10 27 1.0 0.0 2.0 0.0
DVGRID 10 28 1.0 0.0 1.3333 0.0
DVGRID 10 29 1.0 0.0 0.6667 0.0
Main Index Main Index


170
Drawbacks
In all but the simplest of problems, the data input can be formidable without a
preprocessor. The resultant basis vectors are treated as constant and not updated with
each design cycle, so the risks of mesh distortion are increased.
Data Preparation
1. Define the shape design variables using DESVAR Bulk Data entries.
2. Establish the corresponding grid variations using DVGRID Bulk Data
entries, one entry for every design variable-designed grid pair in the model.
This entry ties design variable changes to changes in the structures shape.
3. Preview the resultant basis vectors (See Appendix C for details on how to do
this graphically) and modify if necessary.
Direct Input of Shapes
This approach greatly simplifies the process of defining shape basis vectors. With this
method, externally-generated vectors are DBLOCATEd and used to define shape
basis vectors. An auxiliary model analysis provides these externally-generated
vectors. These vectors could be the results of a statics analysis, but other real vectors
can also be used. For example, you may wish to use normal modes results. There are
no DVGRID entries required in this method (although you can still use them to
augment the direct input shapes). Instead, the DVSHAP entry is used to relate the
DESVAR to the columns of the results from the auxiliary model analysis. The benefits,
drawbacks and data preparation steps for this method are:
Benefits
Basis vectors for shape optimization can be generated using an external auxiliary
model analysis and DBLOCATEd, allowing for easy generation of shape basis
vectors. In theory, any method of external generation is possible, as long as the
number of degrees-of-freedom in the auxiliary model is the same as in the primary
structure (i.e., G-sets must be equivalent).
Drawbacks
The process is not fully automatic since an auxiliary model analysis must be
performed beforehand, with results saved on the database, and DBLOCATEd for
shape optimization. Since the basis vectors are externally generated, they are not
updated for each design cycle. This may cause mesh distortion problems for large
shape changes.
Main Index Main Index
171 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Data Preparation
1. Define an auxiliary model, perform an analysis, and save the results to the
database using the SCR=NO option on the NASTRAN job submittal
command. This process is the same as a conventional MSC.Nastran analysis
except that the auxiliary model geometry and boundary conditions have
been established with consideration given to the shape redesign goals.
2. DBLOCATE the results for shape optimization using the following FMS
section commands (assuming the master file is file1.MASTER):
ASSIGN FILE1 = file1.MASTER
DBLOCATE DATABLK=(UG/UGD,GEOM1/GEOM1D,GEOM2/GEOM2D),
LOGICAL=FILE1
3. Define shape design variables using DESVAR entries, and correlate these to
the DBLOCATEd basis vectors using DVSHAP Bulk Data entries.
4. Preview the resultant shape basis vectors to check for modeling errors (See
Appendix D).
Geometric Boundary Shapes
In this method, BNDGRD Bulk Data entries are used to define the structures
boundaries and DVGRID entries to furnish the shape variations only over these
boundaries. The linear taper basis vector of Figure 3-6 can therefore be input using:
It is seen that the input preparation is simplified relative to the manual grid variation
input for the same example. The benefits, drawbacks and data preparation for this
method are:
Note: The data blocks UG, GEOM1, and GEOM2, are DBLOCATEDd and renamed
to UGD, GEOM1D, and GEOM2D, respectively. The data block names UGD,
GEOM1D, and GEOM2D cannot be changed.
DESVAR 10 LINTAP 1.0 0.01 10.
DVGRID 10 11 1.0 0.0 0.4 0.0
DVGRID 10 15 1.0 0.0 0.8 0.0
DVGRID 10 19 1.0 0.0 1.2 0.0
DVGRID 10 23 1.0 0.0 1.6 0.0
DVGRID 10 27 1.0 0.0 2.0 0.0
BNDGRID 123 11 15 19 23 27 7 8
9 10 14 18 22 26 30
BNDGRID 1 28 29 30
Note: It is necessary to specify the grid component combinations that are fixed using
the BNGRID entry, but it is not necessary to provide a DVGRID entry when
the grids are fixed to zero.
Main Index Main Index


172
Benefits
Since grid variations need only be specified over the boundaries, data preparation is
greatly simplified. In fact, since DVGRID entries are only written for the boundaries,
many real-world problems can be effectively solved without geometry-based
preprocessors. Internal computation of the shape basis vectors means that these can
be recomputed (updated) for every design cycle, reducing the problems associated
with mesh distortion for large shape changes.
Drawbacks
Describing grid variations over the boundaries may still require significant DVGRID
information, even though orders of magnitude less than the manual grid variation
method.
Data Preparation
1. Define the shape design variables using DESVAR Bulk Data entries.
2. Define corresponding shape variations over the structures boundaries using
DVGRID entries.
3. Define the shape boundary conditions using BNDGRID Bulk Data entries.
The BNDGRID-DVGRID entry combination is analogous to the use of
SPCDs to impose enforced boundary displacements. The DVGRID entry
supplies the enforced variation, much like an SPCD, to the boundary grids
specified on the BNDGRID entry (similar to an SPC entry).
Analytic Boundary Shape
The analytic boundary shapes method can be characterized as being the most
sophisticated of the methods and removes the need for DVGRID generation
completely. Instead, you must generate an auxiliary model that clads the boundary
that is to be redesigned. This can be done in a preprocessor, thereby minimizing the
manual data preparation. The geometry (that is, GRID locations) is shared between
the primary and auxiliary models and it is necessary to include BNGRID entries in the
primary model that identify the boundary in the same way as in the geometry
boundary shapes method. Analytic Boundary Shapes on page 430 provides an
example that uses this method and you should refer to this to get a clear picture of the
requirements for generating the auxiliary models. This method involves separate bulk
data for the auxiliary models, with each model denoted by a BEGIN BULK
AUXMODEL = n command. Unique loads and boundary conditions are applied to the
auxiliary model(s), necessitating separate case control sections for each of the models.
The AUXCASE and AUXMODEL Case Control commands are used to select the
separate bulk data files.
Main Index Main Index
173 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Benefits
The method is very general, and does not require a geometry-based pre- and
postprocessor. The user interface is entirely within the MSC.Nastran environment.
Shape basis vectors are updated on every design cycle, reducing the problems
associated with mesh distortion for large shape changes.
Any combination of static loading available in MSC.Nastran may be used. Point loads
(FORCE), enforced displacements (SPCD), thermal loads (TEMP), pressure loads
(PLOAD), and so on, may all be used to generate the desired shapes.
Drawbacks
Creativity is often required in the selection of loads and boundary conditions for the
auxiliary boundary models. This is listed as a drawback because the creativity
requires that you possess a certain amount of sophistication in model preparation.
You may see this as a benefit in that your creativity can be applied to create the most
appropriate shape basis vectors!
Data Preparation
1. Define auxiliary boundary models using one additional Bulk Data Section
for each model. These sections are identified with the command, BEGIN
BULK AUXMODEL = n where n is the auxiliary boundary model ID.
2. Select loading and boundary conditions in Case Control using AUXCASE
and AUXMODEL = n to define the auxiliary boundary model Case Control
Sections.
3. Define the shape boundary conditions on the actual structure using
BNDGRID Bulk Data entries. In addition to defining the connection points
with the auxiliary boundary model, the BNDGRID entries must define those
boundaries that are invariant during shape optimization.
4. Define the shape design variables using DESVAR entries, and relate these to
the shape basis vectors (which the code will compute) with DVSHAP entries.
Example: Shape Basis Vectors
This example further illustrates the process of defining shape basis vectors for each of
the previously outlined methods, excluding the manual grid variation method for
which data preparation can be quite extensive. Since DVGRIDs are used in the
geometric boundary shapes method anyway, their use is covered here in connection
with that method.
Main Index Main Index


174
Suppose we want to redesign the shape of a hole cutout in a square plate shown in
Figure 3-8. This redesign seeks to minimize the weight, subject to constraints on von
Mises stresses. However, rather than allow only circular variations, we would like to
investigate elliptical shape changes. Since the edge traction force is twice , one
would expect a 2:1 ratio elliptical hole at the optimum, rather than simply a circle. Two
design variables will be used-each one representing an axis length.
Figure 3-8 Plate Quarter Model
Recall that the equation for an ellipse centered at the origin of the x-y plane is given by
Eq. 3-10
where and are the and axis intercepts, respectively, of the ellipse. We want
the optimizer to vary the shape of the ellipse, so a natural choice is to define two
design variables, each representing the changes in and . These quantities can be
defined as
T
x
T
y
0.15m
0.15m
0.05m
Material:
aluminum, 7075-T6 sheet
E 7.2E10 = N m
2

0.33 =
2.8E3 = kg m
3

T
y
1.0E4 = N m
x
2
a
b
-----
y
2
b
2
----- + 1 =
a b x y
a b
Main Index Main Index
175 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
These two entries define two design variables: each with initial values of 1.0, lower
bounds of -20., and upper bounds of +20. The initial design variable values are
somewhat arbitrary as are the lower and upper bounds.
Direct Input of Shapes. Direct input of shapes uses an external auxiliary model to
generate shape basis vectors. The auxiliary model is used to generate a set of
displacement vectors that are then DBLOCATEd and identified as shape basis vectors
in the optimization run. All data for the auxiliary model analysis must be provided for
by the designer in a prior MSC.Nastran run.
Figure 3-9 shows the auxiliary model and its boundary conditions. Its geometry and
connectivity are the same as for the primary structure. The material types are also the
same, although this need not be the case. Note that the chosen boundary conditions
are consistent with our shape redesign goals: outer edges fixed, symmetry along the x
and y axes, and a free edge along the hole boundary.
$ define design variables...
$DESVAR,ID, LABEL, XINIT, XLB, XUB, DELXV
DESVAR, 1, DELA, 1.0, -20., +20.
DESVAR, 2, DELB, 1.0, -20., +20.
Note: From Eq. 2-12 a change in a design variables value causes a change in grid
location. The characteristics of the shape changes depend on the shape basis
vectors. DESVAR entries 1 and 2 just define the presence of basis vector
multipliers that the optimizer is free to vary. The basis vector magnitudes
will govern the corresponding magnitude of shape change.
Main Index Main Index


176
Figure 3-9 External Auxiliary Model
Along the cutout, we will apply two sets of enforced displacements: one an elliptical
variation in the , or , direction, the other an elliptical variation in the , or ,
direction. We can use a geometry-based preprocessor to give us this data or for simple
cases, we can just compute this data by hand. Either way, the following SPCD entries
result:
a x b y
Main Index Main Index
177 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
$
$ "LOAD" 1: ELLIPTICAL X-COMPONENT VARIATION (10MM MAX. VALUE)
SPCD, 101, 1, 1, .01
SPCD, 101, 5, 1, .0098769
SPCD, 101, 6, 1, .0095106
SPCD, 101, 7, 1, .0089101
SPCD, 101, 8, 1, .0080902
SPCD, 101, 9, 1, .0070711
SPCD, 101, 10, 1, .0058779
SPCD, 101, 11, 1, .0045399
SPCD, 101, 12, 1, .0030902
SPCD, 101, 13, 1, .0015644
SPCD, 101, 14, 1, 0.0
SPC1, 102, 123456, 1, 5, 6, 7, 8, 9, +
+, 10, 11, 12, 13, 14
SPCADD, 103, 101, 102
$
$
$ "LOAD 2": ELLIPTICAL Y-COMPONENT VARIATION (10MM MAX. VALUE)
SPCD, 104, 5, 2, .0015644
SPCD, 104, 6, 2, .0030902
SPCD, 104, 7, 2, .0045399
SPCD, 104, 8, 2, .0058779
SPCD, 104, 9, 2, .0070711
SPCD, 104, 10, 2, .0080902
SPCD, 104, 11, 2, .0089101
SPCD, 104, 12, 2, .0095106
SPCD, 104, 13, 2, .0098769
SPCD, 104, 14, 2, .01
SPC1, 105, 123456, 1, 5, 6, 7, 8, 9, +
+, 10, 11, 12, 13, 14
SPCADD, 106, 104, 105
$
Note: The SPCD enforced displacement values correspond to an elliptical variation
along the hole boundary with a maximum value of 0.01. The result of these
load applications are shown in Figure 3-10 and Figure 3-11 (the deformations
have been enlarged for clarity). These displacement vectors are used next to
generate our shape basis vectors.
Main Index Main Index


178
Figure 3-10 Displacement Vector 1
Figure 3-11 Displacement Vector 2
In the optimization run, we will DBLOCATE these displacement vectors and use this
data to define our shape basis vectors. This data can be DBLOCATEd using the
following FMS Section statements:
Main Index Main Index
179 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
where plate33_aux.MASTER is the master DBset file corresponding to the auxiliary
model analysis results shown in Figure 3-10 and Figure 3-11.
Having input this information, we need to identify the design variables that are to act
as multipliers of these basis vectors. This is done using the following DVSHAP Bulk
Data entries:
The first entry defines Design Variable 1 (DVID = 1) as the multiplier of the first
DBLOCATEd displacement vector (COL1 = 1) using the factor 1.0 (SF1) as the constant
of multiplication. This process defines the first shape basis vector. A similar process
on the second DVSHAP entry defines the next shape basis vector.
Note that we now have two basis vectors, one for each of the a and b elliptical
variations of the hole boundary. As the optimizer changes x1 and x2, the grid locations
are varied according to
Eq. 3-11
where is the first DBLOCATEd basis vector (note the 1.0 multiplying factor) and
is the second.
Geometric Boundary Shapes
Rather than externally generating the shape basis vector data as in the previous
approach, this method generates the vectors automatically, based on geometric data
supplied on the boundaries. All of the data preparation and input appear in the
optimization input file; there is no need for a prior auxiliary model analysis.
In addition, with this method the shape basis vectors are updated on each design cycle
as the shape changes. One problem with basis vectors that are not updated is that they
may lead to ill-conditioned meshes for large shape changes. Updating this data on
each design cycle minimizes (but not altogether eliminates) this difficulty.
assign f1 = plate33_aux.MASTER
dblocate datablk=(ug/ugd,geom1/geom1d,geom2/geom2d), logical=f1
$DVSHAP,DVID, COL1, SF1, COL2, SF2, ...
DVSHAP, 1, 1, 1.0
DVSHAP, 2, 2, 1.0
G { } 1.0u
1
1.0u
2
x
1
x
2
)

`


=
u
1
u
2
Main Index Main Index


180
The geometric boundary shapes method begins with a definition of allowable
variations over the boundaries. These variations can be described with a geometry-
based preprocessor, or with Bulk Data alone. Either way, the boundaries are defined
using BNDGRID entries, and the variations defined using DVGRID entries.
The elliptical variations on the boundaries can be defined as follows:
The first set of DVGRID entries are related to Design Variable 1, DELA. Note that the
coefficient magnitudes (COEFF) are the same as the SPCD magnitudes of the previous
example. We observe that this set describes an elliptical shape variation in the x-
direction (N1 = 1.0, N2 = N3 = 0.0), dependent on changes in Design Variable 1. The
second set of DVGRID entries likewise describes an elliptical shape variation in the y-
direction, which is dependent on changes in Design Variable 2.
$ DEFINE DESIGN VARIABLES...
DESVAR, 1, DELA, 1.0, -20., +20.
DESVAR, 2, DELB, 1.0, -20., +20.
$
$ ...AND RELATE THESE TO BOUNDARY VARIATIONS DEFINED USING DVGRIDS:
$DVGRID,DVID, GRID, CID, COEFF, N1, N2, N3
$
$ BASIS VECTOR 1:
DVGRID, 1, 1, , .01, 1.0, 0.0, 0.0
DVGRID, 1, 5, , .0098769,1.0, 0.0, 0.0
DVGRID, 1, 6, , .0095106,1.0, 0.0, 0.0
DVGRID, 1, 7, , .0089101,1.0, 0.0, 0.0
DVGRID, 1, 8, , .0080902,1.0, 0.0, 0.0
DVGRID, 1, 9, , .0070711,1.0, 0.0, 0.0
DVGRID, 1, 10, , .0058779,1.0, 0.0, 0.0
DVGRID, 1, 11, , .0045399,1.0, 0.0, 0.0
DVGRID, 1, 12, , .0030902,1.0, 0.0, 0.0
DVGRID, 1, 13, , .0015644,1.0, 0.0, 0.0
DVGRID, 1, 14, , .0, 1.0, 0.0, 0.0
$
$
$ BASIS VECTOR 2:
DVGRID, 2, 1, , .0, 0.0, 1.0, 0.0
DVGRID, 2, 5, , .0015644,0.0, 1.0, 0.0
DVGRID, 2, 6, , .0030902,0.0, 1.0, 0.0
DVGRID, 2, 7, , .0045399,0.0, 1.0, 0.0
DVGRID, 2, 8, , .0058779,0.0, 1.0, 0.0
DVGRID, 2, 9, , .0070711,0.0, 1.0, 0.0
DVGRID, 2, 10, , .0080902,0.0, 1.0, 0.0
DVGRID, 2, 11, , .0089101,0.0, 1.0, 0.0
DVGRID, 2, 12, , .0095106,0.0, 1.0, 0.0
DVGRID, 2, 13, , .0098769,0.0, 1.0, 0.0
DVGRID, 2, 14, , .01, 0.0, 1.0, 0.0
Main Index Main Index
181 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
BNDGRID entries complete the shape boundary definition process and identify two
categories of grid components:
1. Grid components that change according to data supplied on the DVGRID
entries.
2. Grid components that remain fixed.
The process is similar to the SPC, SPCD combination used to impose enforced
displacements in static analysis. The DVGRID entry supplies the grid motion on the
boundary, much like an SPCD enforced displacement. The boundary grids are
identified using BNDGRID entries, much like an SPC entry that identifies the
corresponding grids and their components.
The first BNDGRID entry identifies all grid components along the plate outer edges.
Since no DVGRIDs have been furnished for these degrees-of-freedom, their
displacements will be fixed to zero. For shape optimization, the plate outer edges will
be invariant.
The second BNDGRID entry identifies components 2 through 6 for those grids that lie
on the x-axis. This allows the x-component to vary during shape optimization,
consistent with the elliptical changes made to the cutout. All other components are
considered fixed since no enforced shape changes for these degrees-of-freedom have
been supplied on DVGRID entries. The third BNDGRID entry furnishes similar
information as the previous entry, only with regard to shape changes along the y-axis.
The last BNDGRID entry identifies components 1 through 6 for the cutout edge grids.
Shape changes are enforced for components 1 and 2 since they have been provided on
DVGRID entries. All other components are considered fixed.
$
$ BOUNDARY CONDITIONS FOR SPCD APPLICATION (PROVIDED VIA DVGRIDS)
$ OUTER PLATE EDGES...
BNDGRID,123456, 2, 15, 16, 17, 18, 3, 19, +
+, 20, 21, 22, 4
$ Y=0 PLANE...
BNDGRID,23456, 23, 24, 25, 26
$ X=0 PLANE...
BNDGRID,13456, 99, 100, 101, 102
$ CUTOUT EDGE...
BNDGRID,123456, 1, 5, 6, 7, 8, 9, 10, +
+, 11, 12, 13, 14
Main Index Main Index
182 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Once the free, fixed, and enforced portions of the shape changes have been defined,
the code interpolates this information to all of the interior degrees-of-freedom. In this
case, the resulting shape basis vectors are identical to those shown in Figure 3-10 and
Figure 3-11.
On subsequent iterations, the shape basis vectors may change slightly because the
interpolated solutions over the interior of the structure will change due to the
modified geometry. Recomputing the basis vectors at the beginning of each design
cycle helps to minimize the problems associated with mesh distortion.
Analytic Boundary Shapes
This method also uses auxiliary models. Here, however, they are usually defined over
the boundaries of the structure, although they can cover the whole structure i
required. The so-called auxiliary boundary models can be an edge of BAR elements
for a two-dimensional redesign or a skin of shell elements for a general, three-
dimensional shape optimization task.
Auxiliary boundary models are used to generate shape variations over the boundaries
of the structure, providing a shortcut to the geometric boundary shapes method. You
can define as many auxiliary boundary models as are necessary using any number
and types of applied static loading to produce sets of desired boundary variations.
The code interpolates these boundary changes to the structures interior, thus
automatically generating the shape basis vectors.
Each auxiliary boundary model requires its own Bulk Data Section. Each model can
be loaded in any number of subcases to produce more than one shape basis vector per
auxiliary boundary model. Load and boundary condition sets are selected using the
auxiliary boundary model Case Control Sections.
The following Bulk Data section provides the auxiliary boundary model definition for
one plate example. This section must appear after the Bulk Data for the primary
model.
Note: The boundary conditions defined by the DVGRID and BNDGRID entries are
only used to generate shape basis vectors. They are not used in connection
with the primary model analysis.
Main Index Main Index
183 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
BEGIN BULK AUXMODEL = 1
$
PARAM, PRGPST, NO
$
$ auxillary model element definition (geometry from primary structure):
$
CBAR, 101, 20, 1, 5, 0., 0., 1.
CBAR, 102, 20, 5, 6, 0., 0., 1.
CBAR, 103, 20, 6, 7, 0., 0., 1.
CBAR, 104, 20, 7, 8, 0., 0., 1.
CBAR, 105, 20, 8, 9, 0., 0., 1.
CBAR, 106, 20, 9, 10, 0., 0., 1.
CBAR, 107, 20, 10, 11, 0., 0., 1.
CBAR, 108, 20, 11, 12, 0., 0., 1.
CBAR, 109, 20, 12, 13, 0., 0., 1.
CBAR, 110, 20, 13, 14, 0., 0., 1.
PBAR, 20, 102, 1.0E-3, 1.0E-12,1.0E-12,1.0E-12
MAT1, 102, 7.2+10, , 0.33, 2.8015+3,1.0-2
$
$ boundary condition set I:
PLOAD1, 160, 101, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 102, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 103, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 104, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 105, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 106, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 107, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 108, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 109, FX, FR, 0., 100., 1., 100.
PLOAD1, 160, 110, FX, FR, 0., 100., 1., 100.
$
$ boundary condition set II:
PLOAD1, 161, 101, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 102, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 103, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 104, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 105, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 106, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 107, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 108, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 109, FY, FR, 0., 100., 1., 100.
PLOAD1, 161, 110, FY, FR, 0., 100., 1., 100.
$
$ temperature distribution:
TEMP, 162, 1, 1., 5, 1., 6, 1.
TEMP, 162, 7, 1., 8, 1., 9, 1.
TEMP, 162, 10, 1., 11, 1., 12, 1.
TEMP, 162, 13, 1., 14, 1.
$
$ spc conditions for both load sets:
SPC, 100, 1, 23456, 0.0
SPC, 100, 14, 13456, 0.0
SPC1, 100, 345, 5, 6, 7, 8, 9, 10, +
+, 11, 12, 13
$
ENDDATA
Main Index Main Index


184
A number of interesting things can be seen in this listing. First, the CBAR entries
define the auxiliary boundary model elements as a ring of bar elements along the
cutout. Note that GRID entries do not appear because grid data is shared with the
primary structure, and additional grids need not be defined. (Of course, if extra grids
are necessary, they can be added as required. If the location of the hole were to change,
one might want to define another grid at the center to use as an attachment point for
rigid elements.)
Two sets of loads are applied. One will result in an elliptical-type variation in the x-
direction, and the other a similar displacement in the y-direction. Uniform loading
with PLOAD1s and a uniform temperature load have been used to generate these
boundary deformations.
The loading and boundary condition sets are selected in the auxiliary boundary model
Case Control Section, which follows the Case Control for the primary model:
The keyword AUXCASE identifies the beginning of the auxiliary model Case Control
Sections. In this example, a single auxiliary boundary model (AUXMODEL = 1) is
used to generate two basis vectors in subcases 10 and 20. The resultant boundary
displacements are shown in Figure 3-12 and Figure 3-13.
$ AUXILIARY MODEL CASE CONTROL:
AUXCASE
AUXMODEL = 1
SUBCASE 10
SUBTITLE = AUXILIARY MODEL 1, LOAD CASE 10
SPC = 100
LOAD = 160
TEMP(LOAD) = 162
DISPLACEMENT = ALL
SUBCASE 20
SUBTITLE = AUXILIARY MODEL 1, LOAD CASE 20
SPC = 100
LOAD = 161
TEMP(LOAD) = 162
DISPLACEMENT = ALL
BEGIN BULK
Main Index Main Index
185 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Figure 3-12 Boundary Displacement Set 1
Figure 3-13 Boundary Displacement Set 2
Once these boundary displacements have been computed, the code interpolates these
displacements to the interior grids. Once again, BNDGRID entries define the degrees-
of-freedom along the boundaries that are allowed to change and those degrees-of-
freedom that are to remain fixed.
Main Index Main Index


186
You will recognize these as being the same set of boundary conditions as were used
in the geometric shapes method. The auxiliary boundary model solutions, combined
with the BNDGRID boundary data, supply sufficient information to the code to use in
interpolating these shape changes to the interior of the structure. The result is a family
of shape basis vectors. Finally, the designer ties these basis vectors to changes in the
set of design variables using the DVBSHAP entries:
Each entry associates a Design Variable with a basis vector resulting from a particular
auxiliary boundary model (AUXMID). The column number (COL) indicates the
auxiliary model displacement solution number. In this example, design variable 1 is
the multiplier (with a 1.0 multiple) of the shape basis vector resulting from a solution
of auxiliary boundary model 1 (AUXMID = 1). But recall that this model has two
subcases: 10 and 20. The COL field on the entry reconciles this. COL1 = 1 implies the
first solution from subcase 10. COL1 = 2 implies the second solution from subcase 20.
Similarly, design variable 2 is the second basis vector multiplier computed from the
second solution (subcase 20) of auxiliary boundary model 1 (AUXMID = 1).
The resultant reduced basis formulation could be written as:
Eq. 3-12
$
$ boundary conditions for basis vector generation:
$ outer plate edges...
BNDGRID,123456, 2, 15, 16, 17, 18, 3, 19, +
+, 20, 21, 22, 4
$ x=0 plane...
BNDGRID,23456, 23, 24, 25, 26
$ y=0 plane...
BNDGRID,13456, 99, 100, 101, 102
$ cutout edge...
BNDGRID,123456, 1, 5, 6, 7, 8, 9, 10, +
+, 11, 12, 13, 14
$DVBSHAP,DVID, AUXMID, COL1, SF1, ...
DVBSHAP,1, 1, 1, 1.0
DVBSHAP,2, 1, 2, 1.0
G { }
1.0U
AUXMID 1 =
SUBCASE 10 =
1.0U
AUXMID 1 =
SUBCASE 20 =
x
1
x
2
)

`


=
Main Index Main Index
187 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
where each column is a displacement solution of the auxiliary model. Each column is
the same size as . The subscripts on displacement solutions indicate the source
of the applied boundary displacements. The first shape basis vector is derived from
the first auxiliary boundary model subcase, while the second shape basis vector is
derived from the second auxiliary model subcase.
G { }
Main Index Main Index


188
3.6 Designating the Design Responses
Before the objective function and constraints can be defined, the analysis responses on
which they depend must be designated. These responses are called design responses
and are specified in the design model using DRESP1, DRESP2, and/ or DRESP3 Bulk
Data entries.
DRESP1 Bulk Data Entries
DRESP1 Bulk Data entries define type-1, or first-level responses. These responses are
available directly from an MSC.Nastran analysis. Structural weight, displacements at
grid points, element stresses, and so on, are all examples of type-1 responses. See the
DRESP1 Bulk Data entry description for a list of available response types in design
optimization. The DRESP1 entries all require input of a response ID, response label,
and response type (e.g., WEIGHT) but the remaining data input is response type
dependent and this entry is therefore discussed at length in Bulk Data Entries on
page 216.
DRESP2 Bulk Data Entries
DRESP2 Bulk Data entries define type-2, or second-level responses. This class of
responses is also referred to as user-defined or synthetic since the DRESP2 utilizes
the equation input feature in MSC.Nastran in a fashion similar to the of the user
defined properties (DVPREL2,DVMREL2 and DVCREL2) discussed in Relating
Design Variables to Properties on page 162. With the DRESP2 entry and its
companion DEQATN entry (Design EQuATioN), responses such as error functions,
stress averages, and so on can be defined. First-level responses, design variables, grid
coordinate values, table constants, designed properties, and even other second-level
responses may all be used as input to these user-defined responses. Figure 3-14 is a
diagram of the DRESP1 and DRESP2 data structure.
Main Index Main Index
189 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Figure 3-14 Second-Level Responses
The DRESP2 entry has a number of features you need to understand to make the most
effective use of this powerful tool. Bulk Data Entries on page 216 provides a more
in-depth discussion, but it is noted for now that one option available on the DRESP2
entry is to invoke a standard type of operation that does not require a DEQATN. For
example, the SSQ option automatically performs a sum of squares operation on the
DRESP2 arguments.
Therefore, you do not need to develop a DEQATN that performs this routine, but
potentially lengthy, summation. Other available standard options are discussed in
Bulk Data Entries on page 216.
To illustrate the difference between DRESP1 and DRESP2 entries, suppose we have a
case where we want to use the x and y static displacement components at a particular
grid point as design responses in our design model. These first-level responses can be
identified using two DRESP1 entries as follows:
$DRESP1,ID, LABEL, RTYPE, PTYPE, REGION, ATTA, ATTB, ATT1, +
$+, ATT2, ...
$
$... X DISPLACEMENT AT GRID 100:
DRESP1, 501, UX100, DISP, , , 1, , 100
$... Y DISPLACEMENT AT GRID 100:
DRESP1, 502, UY100, DISP, , , 2, , 100
Analysis
Results
DRESP1 DEQATN DRESP2
DESVAR
DTABLE
DNODE
DVPREL1
DVCREL1
DVMREL1
DVPREL2
DVCREL2
DVMREL2
SSQ XITEM ( )
i
2
i 1 =
NTERMS

=
Main Index Main Index


190
DRESP1 501 selects the x-component of displacement at Grid 100, and DRESP1 502
selects the corresponding y-component. These displacements may also be used in a
synthetic relation. For example, to express the total x-y plane displacement at Grid 100
as the square root of the sum of displacement squares we could use:
The equation data is supplied on the DEQATN entry, while the DRESP2 entry defines
the arguments to this equation. In this case the arguments are the two first-level
responses DRESP1 501 and 502. DRESP2 520 can now be used as the objective function
or invoked by a constraint.
A standard function can also be used in this case so that the input would be simply:
Efficient Definition of Responses
Since a large number of responses are often used in design, the DRESP1 entry must be
able to efficiently define many responses using few entries. For example, a single
DRESP1 entry can identify the z-component of displacement at grid points 100, 101,
and 102 as follows:
This list can be extended as necessary to any number of grids. A THRU option is not
supported, however.
Element-level responses can be selected using either element IDs (much like the list of
grids in the previous example), or property IDs. By supplying a property ID on a
DRESP1 entry, we identify the element-level response for every element in that
$
$... DEFINITION OF EQUATION
DEQATN 510 U( UX, UY ) = SQRT( UX**2 + UY**2 )
$
$... SYNTHETIC RESPONSE:
DRESP2, 520, U, 510, , , , , , +
+, DRESP1, 501, 502
$
$... SYNTHETIC RESPONSE
DRESP2 520 U RSS
DRESP1 501 502
$DRESP1,ID, LABEL, RTYPE, PTYPE, REGION, ATTA, ATTB, ATT1, +
$+, ATT2, ...
$
DRESP1, 100, UZ, DISP, , , 3, , 100, +
+, 101, 102
Main Index Main Index
191 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
property group. For example, the following selects the axial stress response (the axial
response is selected using a stress item code as explained in Bulk Data Entries on
page 216) for all ROD elements in property groups 150, 160, 170 and 180:
This can generate quite a lot of design data. Figure 3-15 is a schematic diagram of this
hierarchy.
If limits are subsequently placed on these responses using a DCONSTR entry
(discussed in Defining the Constraints on page 201), the result will be a pair of
stress constraints (one for the upper bound and one for the lower bound) for every
element in each of these property groups. When it is further considered that this
constraint may be selected by multiple subcases, it is seen that just a pair of entries can
possibly define hundreds of stress constraints.
Figure 3-15 Identification of Element-Level Responses
These efficiencies in design response identification extend to type-2 and type-3
responses as well. For example, consider the following DRESP1 entries that identify
major and minor principal stresses at surface z1 for PSHELL group 10:
$DRESP1,ID, LABEL, RTYPE, PTYPE, REGION, ATTA, ATTB, ATT1, +
$+, ATT2, ...
$
DRESP1, 250, SIG1, STRESS, PROD, , 2, , 150, +
+, 160, 170, 180
DRESP1
EID1
EID2
.
.
.
EIDn
PID1
PIDm
PID2
.
.
.
Main Index Main Index


192
This pair of entries defines a pair of stresses for every element in this property group.
This could easily result in hundreds of design responses.
Assume we wanted to write the maximum shearing stress as the average of these
responses:
This single DRESP2 entry defines a maximum shear for every element in the group.
Since the underlying first-level responses are element-level, the DRESP2 is as well.
A few restrictions apply to the formulation of second and third-level responses:
1. Subcase-dependent responses may only be used in an equation with other
responses from the same subcase. This implies that these responses cannot
span analysis disciplines.
2. Similarly, responses from separate superelements cannot be combined in an
equation.
3. When combining responses of different types (e.g. stress + strain, etc.), the
responses must be scalar quantities. That is, the corresponding DRESP1
entries must define a single response only. This implies that for
displacements, only a single grid may be listed on the DRESP1 entry; for
element-level responses, the ELEM identifier must be used with only a single
element ID; and so on.
4. A DRESP2 entry cannot self-reference the same DRESP2.
Example Using DRESP1 and DRESP2 Entries
Suppose we would like to include Euler buckling constraints in the design model for
a simple, pin-ended rod elements. A representative element is shown in Figure 3-16.
$DRESP1,ID, LABEL, RTYPE, PTYPE, REGION, ATTA, ATTB, ATT1, +
$+, ATT2, ...
$
$ Major principal stress at surface z1:
DRESP1, 101, SIG1, STRESS, PSHELL, , 7, , 10
$
$ Minor principal stress at surface z1:
DRESP1, 102, SIG2, STRESS, PSHELL, , 8, , 10
$ Input to equation to compute max shears:
DRESP2, 201, MAXS, AVG, , , , , , +
+, DRESP1, 101, 102
Main Index Main Index
193 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Figure 3-16 Pin-ended Rod Element
The critical load which induces the first Euler buckling mode is given by
Eq. 3-13
The design requirements call for the axial load in this ROD element to be greater (i.e.,
less compressive) than this critical load or
Eq. 3-14
which can be rewritten as
Eq. 3-15
Note that the critical load is a function of the area moment of inertia, yet this quantity
is not a part of the analysis model specification for the ROD element. (Since the inertia
of the ROD is not necessary to compute the axial force, the inertia is not needed to
perform the analysis.) If the cross-sectional area of the element is the design variable
and a solid circular section is assumed, then
Eq. 3-16
and from Eq. 3-15
Eq. 3-17
1 2
x
P
A, E, I
L
P
CR

2
EI
L
2
------------ =
P
CR
P
P
CR
----------
1
P L
2

2
EI
------------- 1
A r
2
=
I
r
4
4
--------- = I
A
2
4
------ =
4 P
E
----------
L
A
---
\ .
| |
2
1
Main Index Main Index


194
The buckling constraint is now a function of the cross-sectional area design variable
for this ROD element geometry. The following Bulk Data entries illustrate one way in
which these relations may be implemented:
The two GRID entries along with the CROD and PROD entries specify ROD Element
100 with a length of 10 units along the x-axis and an initial cross-sectional area of 0.5.
The DESVAR entry defines the area design variable with an initial value equal to the
initial cross-sectional area of the ROD. Lower and upper bounds of 0.25 and 0.75,
respectively, are set, representing 50% move limits on the cross-sectional area.
Since the constraint on Euler buckling is applied here for just a single element, the
ELEM identifier is used on the DRESP1 in field 5 along with a 100 in field 9 and a 2
in field 7 to specify that the axial load for Element 100 is to be used in the design
model. The force component is selected by reference to the item codes. (See Item
Codes in Chapter 6 of the MSC.Nastran Quick Reference Guide for these item codes.)
The DRESP2 entry defines the arguments of the Euler buckling equation DEQATN
600. Note that this information is positional; the values are assigned to the argument
list of the equation based on the order of specification in the DRESP2 entry. The order
of the DRESP2 continuation lines is not interchangeable, with the order provided by
the DRESP2 Bulk Data entry description.
This example lends itself to illustrating another feature of the DRESP2 entry: the
ability to reference another DRESP2 entry. The example has been presented in terms
of a single ROD entry and a single buckling constraint. In reality, it is likely that you
will want to impose this type of constraint at many points in the structure and this
could be done by duplicating the design model of the bulk data fragment for each
GRID 1 0. 0. 0.
GRID 2 10. 0. 0.
CROD 100 200 1 2
PROD 200 201 0.5
$ DESIGN MODEL
DESVAR 1 A200 0.5 0.25 0.75
DVPREL1 301 PROD 200 A
1 1.0
DRESP1 410 SA FORCE ELEM 2 100
DRESP2 501 EUL1 600
DESVAR 1
DTABLE L 10.0 E 10.0E7
DRESP1 401
DTABLE L 10.
DEQATN 600 EUL1(A,L,E,P) = -4. * P* L**2/(PI(1) * E * A**2)
$ note the use of PI function (see DEQATN entry description)
DCONSTR 500 501 1.0
Main Index Main Index
195 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
constraint. This entails determining the length of each rod and using it in the equation.
Since the form of the length calculation is shared across all the elements, it may be
desirable to formulate a DRESP2/ DEQATN to just determine the length and to then
use this in the buckling constraint. The bulk data fragment given above could then be
redone as:
DRESP3 Bulk Data Entries
DRESP3 Bulk Data entries define type-3 or external responses. This bears some
similarity to the DRESP2 just described with the important difference that the DRESP3
response is evaluated by invoking an external (to MSC.Nastran) process via an
application programming interface (API). Therefore, instead of invoking an equation
for DRESP2, the DRESP3 entry identifies a GROUP or TYPE for the external response.
The arguments for the DRESP3 are similar to DRESP2 so that the data structure
diagrammed in Figure 3-14 becomes:
GRID 1 0. 0. 0.
GRID 2 10. 0. 0.
CROD 100 200 1 2
PROD 200 201 0.5
$ DESIGN MODEL
DESVAR 1 A200 0.5 0.25 0.75
DVPREL1 301 PROD 200 A
1 1.0
DRESP1 410 SA FORCE ELEM 2 100
DRESP2 411 LENGTH 400
DNODE 1 1 1 2 1 3 2
1 2 2 2 3
DEQATN 400 LENGTH(X1,Y1,Z1,X2,Y2,Z2) = SQRT((X2 - X1)**2 +
(Y2 - y1)**2 + (Z2 - Z1)**2 )
DRESP2 501 EUL1 600
DESVAR 1
DTABLE E
DRESP1 410
DRESP2 411
DTABLE E 1.0E7
DEQATN 600 EUL1(A,E,P,L) = -4. * P * L**2/(PI(1) * E * A**2)
DCONSTR 500 501 1.0
Main Index Main Index


196
Figure 3-17 Third-Level Responses
Note that, unlike the DRESP2, a DRESP3 cannot reference another DRESP3. Also,
there is an additional USRDATA argument than can be used to pass character data
such as names of data files or options to take within the external server.
Since an external response is computed using a user-built server executable, defining
an external response requires additional user actions to establish the association
between the MSC.Nastran program and the response server programs. Five user
actions are required in setting up the external response in a design task:
1. Create a DRESP3 entry to define the response.
2. Write Fortran or C routines and build a server executable based on the server
template routines.
3. Create a CONNECT entry to define an external response group.
4. Create a evaluator connection file to associate the external response group
with the server program.
5. Submit the Nastran job with the gmconn keyword that references to the
evaluator connection file.
Analysis
Results
DRESP1
External
Server
DESVAR
DTABLE
DNODE
DVPREL1
DVCREL1
DVMREL1
DVPREL2
DVCREL2
DVMREL2
DVMREL2
DRESP2
USRDATA
Main Index Main Index
197 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
Descriptions of each user actions are given below, with much of the description
similar to that found in Building and Using the Sample Programs in Chapter 7 of
the MSC.Nastran 2005 r3 Installation and Operations Guide. An example application is
shown in External Response to Include Alternative Buckling Response on
page 504.
Creating a DRESP3 Entry
A DRESP3 entry defines an external response using user-supplied routines. The
response can be used as constraints or as an objective in a design. See DRESP3 on
page 274 for the Bulk Data description.
Build an External Response Server
Copy the entire contents of the install_dir/msc2004/dr3srv directory to your local
directory, where install_dir/msc2004/dr3srv is the system installation directory for the
MSC.Nastran program. Then, modify Fortran routines (r3sgrt, r3svald/ s) to define
your own external responses. Routine, r3sgrt validates the response type names for an
external response group specified on a DRESP3 entry. You are required to assign a
value to NTYPES in the parameter statement for the number of response types for a
given external response group and to assign the response type names to the array,
TYPNAM in the data statement. Routine r3svald/ s computes the desired response
and returns the value. For the short word machine like a typical workstation, use the
r3svald version. Use the r3svals version for the long word machine. You may add your
own routines by including the names of the new object in the makefile. To build the
server program, type the following command: msc2004 ./dr3srv build. If the build
process is successful, an external response server defaulted as dr3serv should be
created. You can change its name in the makefile.
Creating a CONNECT Command
The CONNECT command, specified in the File Management System (FMS) section,
defines the external responses group. Multiple CONNECT entries may be used to
define multiple groups. This command is discussed in File Management on
page 205.
Example:
CONNECT DRESP3 TESTGRP EXTRESP
Main Index Main Index


198
where DRESP3 is the key word for the external response feature. TESTGRP is the
group name of the external response specified on the DRESP3 entry and EXTRESP is
the evaluator name. Currently, there is no distinction between the evaluator of the
external response and the group of the external response.
Creating an Evaluator Connection File
The evaluator connection file defines the association between the external response
group and the corresponding server program.
Format:
Evaluator name, the connection option, the path of the server program
Example for Connection File extconnect:
EXTRESP, -, / net/ harkness/ harkness/ users/ shz/ dr3srv/ dr3serv
where EXTRESP is the evaluator name, symbol, - indicates that the pipe option is used
for the association (i.e., server executable resides in the same computer as the Nastran
program does). The last term is the path name where the server executable, dr3serv
resides.
Submit a Nastran Job with the gmconn Keyword
gmconn is the keyword used on the command line to reference the name of the
evaluator connection file. The connection file is processed at the Nastran initialization
stage to establish a physical link between the Nastran program and the server
program(s).
Example:
nastran myjob src=yes gmconn = extconnect
Guidelines and Limitations for DRESP3
1. Each group of external response can have multiple response types.
2. Each group of external response requires one CONNECT entry.Up to 10
groups are supported including the external Beam Library.
3. The number of characters allowed in the USRDATA field on a DRESP3 entry
can not exceed 32000.
4. The finite difference approach is the only option to evaluate the gradients of
the external responses.
Main Index Main Index
199 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
3.7 Defining the Objective Function
The objective function is a scalar quantity that is either minimized or maximized by
the optimizer. You designate the design objective using the DESOBJ (DESign
OBJective) Case Control command. This command points to a design response
defined on either a DRESP1, DRESP2 or DRESP3 Bulk Data entry which must define
a single scalar response only.
When formulating the design objective, there are a couple of scaling-related issues
that should be kept in mind since they affect overall performance. First, the design
problem should be posed so that the objective function has sufficient sensitivity with
respect to each of the design variables. This is a relative type of consideration and is
somewhat difficult criteria to define in a general sense. However, the idea is as
follows: Suppose a weight design objective is on the order of 1,000 kilograms. If a
structural element in the model is on the order of 1.0E-2 kg or less and is a function of
a single design variable, then varying this quantity by 100% or more does not have any
appreciable effect on the overall weight. There are two alternative remedies for this,
depending on how other design variables affect the objective. If none of the design
variables have a significant effect, you may not have formulated the best objective. In
a weight minimization, this can occur because the calculated weight includes all the
weight in the structure, including fixed weights that are not functions of the design
variables. If these fixed weights dominate the overall weight, you should use a
DRESP2 to subtract out this fixed weight so that the objective becomes a stronger
function of the design variables. In our example, if 998 of the 1,000 kilograms are fixed,
subtracting this number as shown in the following bulk data fragment will result in
an objective that the optimizer will have an easier time in modifying.
The second alternative is to assess whether the design variable has an appreciable
effect on the design in any case. If the variable does not affect the weight significantly
and it has little effect in satisfying a constraint perhaps it should be removed from the
design task or perhaps it can be combined with a number of other variables that are
TITLE = FOCUSSED WEIGHT RESPONSE
DESOBJ = 100
.
.
.
BEGIN BULK
.
.
DRESP1 10 ENTWGHT WEIGHT
DRESP2 100 FWEIGHT 100
DRESP1 10
DEQATN 100 FWEIGHT(ENTWGHT) = ENTWHGT - 998.
Main Index Main Index


200
negligible by themselves but become significant when they vary together. Combining
design variables makes the optimizers task easier and perhaps makes it easier for you
to interpret the results as well. In any case, it is always advisable, prior to
optimization, to perform a design sensitivity analysis to ensure that the objective
function has sufficient sensitivity with respect to the design variables.
A related difficulty can occur if you formulate an objective that has zero as its
minimum. This can happen when, for example, when you are trying to minimize the
error between test results and your finite element analysis using some type of least
squares error function. It could be that the optimizer has no trouble in reducing the
error from 1000.0 to 0.01 but then bounces around between .01 and .005. The
convergence tests of Figure 2-21 would regard this as a very large change and dictate
that further design cycles be made. In fact, this may be an insignificant difference and
the extra design cycles are wasted. In this case, it would be helpful to add a constant
to the error function at a level that you feel is acceptable. For example, if you think
getting to within 0.01 with the error function is a good answer, adding 10.0 to this
function will allow the problem to terminate whenever the change in the error is less
than .01 (this assumes that the default value of CONV1=0.001 is being used).
Alternatively, you could leave the error function unchanged but set the CONV2
parameter to the 0.01 absolute change you regard as indicating a converged solution.
Main Index Main Index
201 CHAPTER 3
Developing the Design Model for Sensitivity and Optimization
3.8 Defining the Constraints
To define constraints in MSC.Nastran, you need to
Identify a first-level response (or group of responses) using a DRESP1 Bulk
Data entry, write a second-level response (or group of second-level responses)
using a DRESP2 Bulk Data entry, or develop a third-level response (or group of
third-level responses) using your own application program.
If the constraint is to be applied to a DRESP2 quantity that references
DRESP1 quantities from different subcases, the DRSPAN case control is
required at the subcase level to identify those responses that span subcases.
Specify the response bounds using a DCONSTR entry.
Optionally collect the DCONSTR entries into a master set using the
DCONADD entry.
Select DCONSTR and/ or DCONADD entry sets in Case Control using either
the DESGLB command for global (e.g., non-subcase dependent) responses
or the DESSUB command for subcase-dependent responses.
In most cases, the constraints are associated with a particular analysis type and
subcase, in which case the DESSUB Case Control command is appropriate. In this
regard, you should know that MSC.Nastran does not allow you to impose subcase
dependent constraints that are on responses that are not consistent with the
ANALYSIS being performed in that subcase. For example, it is a User Fatal error if you
attempt to constrain a LAMA (buckling) response type in an ANALYSIS = STATICS
subcase.
The DESGLB request is used in cases where the response is not associated with a
subcase. The most obvious example is if you want to constrain a WEIGHT or
VOLUME response. If you constrain this response in multiple subcases, you would
get duplicate constraints that cause extra work for the optimizer. Instead, it is
recommended that you apply this constraint before any subcase using the DESGLB
request. Another example where the DESGLB request is appropriate is when you
want to specify a relation among the design variables. An example of this given in
Eq. 1-20 where an aspect ratio limit is imposed on the dimensions of a beam. Since this
response is not a function of the finite element analysis, the DESGLB command is used
to select this constraint.
The DCONADD Bulk Data entry allows you to collect constraints into an overall set
that is selected using the DESSUB or the DESGLB Case Control command. An
example where this is useful is a static design task that has multiple subcases. It is
likely that the strength limits that are imposed on response types such as STRESS,
Main Index Main Index


202
STRAIN, or CFAILURE are applicable in every subcase. At the same time there may
be displacement limits that are unique to each subcase. The availability of the
DCONADD entry makes it possible to have a single set of DCONSTR entries for the
strength limits. These entries are then combined with the displacement limits using
the DCONADD entry.
The DCONSTR entry bounds are used by MSC.Nastran to create a pair of normalized
constraints, one for the lower bound and one for the upper bound, as discussed in
The Design Model on page 39. The DCONSTR entry contains an ID, which is
invoked either directly by a DESGLB/ DESSUB Case Control command or indirectly
through a DCONADD entry.
The DCONSTR entry is typically used to provide upper and lower bounds of the
constraint, but there are two special features that can be useful in frequency response
optimization. The first is that integer IDs can be used in the fields that specify the
bounds. The presence of an integer indicates that the bound is specified by a frequency
dependent table. This enables specification of bound as a function of frequency. The
second feature is that the DCONSTR entry has optional inputs that specify the lowest
and highest frequency for which the constraint is to be applied.
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users GuideLKLK
CHAPTER
4
Input Data
I File Management
I Executive Control
I Case Control Section
I Bulk Data Entries
I Parameters Unique to Design Sensitivity and Optimization
Main Index Main Index


204
This chapter describes all Executive Control, File Management, Case Control, Bulk
Data, and Parameter requirements for design sensitivity and optimization. These
various input formats enable the following:
1. Specification of the applicable analysis discipline(s)
2. Definition of the design model
3. Overriding of the optimizers internal parameters
4. Control of the program flow and results output
Main Index Main Index
205 CHAPTER 4
Input Data
4.1 File Management
Inputs are required in the File Management Section of the input file only in the special
circumstances cited in this section.
Shape Optimization
File Management Section statements are required in shape optimization for the direct
input of shapes method. This approach DBLOCATEs a set of displacement vector data
from a database and uses this data to generate a set of shape basis vectors for design
optimization (see Relating Design Variables to Shape Changes on page 166). The
following statements must be used:
The filename file.MASTER is the name of the database for the auxiliary model
analysis.
The filename file is arbitrary. The UG, GEOM1, and GEOM2 data blocks must be
DBLOCATEd and renamed to UGD, GEOM1D, and GEOM2D, respectively. These
new names cannot be changed since the code looks for the data in these locations
explicitly. See Shape Optimization of a Culvert on page 419 for an example using
this method.
User Defined Beam Libraries
MSC.Nastran contains a beam library feature that allows you to input beam
dimensions for an array of cross section types, such as ROD, I and T. You can augment
or replace MSCs section types with your own sections using an Application
Programming Interface (API) option as outlined in Adding Your Own Beam Cross-
Section Library on page 547. The dimensions of these user defined cross sections can
also be designed. The use of these user defined beam sections is enabled by a
CONNECT file management statement that identifies the evaluator. The form of the
CONNECT statement is:
CONNECT BEAMEVAL group evaluator
For example:
CONNECT BEAMEVAL MYBEAMS MYEVALS
assign f1 = file.MASTER
dblocate datablk=(ug/ugd,geom1/geom1d,geom2/geom2d), logical=f1
Main Index Main Index


206
where group is the group name for the user supplied beam library and is called out
on the PBARL or PBEAML entry and must therefore be eight characters or less.
Evaluator is the evaluator connection file. Each entry in the file uniquely associates
the group name with the server program. You can have multiple connect entries for
multiple user defined beam libraries and also for the External Responses discussed
next.
External Responses
As first discussed in The Design Model on page 39, the DRESP3 Bulk Data entry
allows you to use an Application Programming Interface (API) to compute external
responses that beyond those available using the internal DRESP1 and DRESP2 entries.
The use of these user defined external responses is enabled by a CONNECT in much
the same fashion as for the user defined beam libraries just described. The form of the
CONNECT statement in this cases is:
CONNECT DRESP3 group evaluator
For example:
CONNECT DRESP3 MYRESP MYEXT
where group is the group name for the external response and is called out on the
DRESP3 entry and must therefore be eight characters or less. Evaluator is the name
of the server program and is currently uniquely associated with the group name. You
can have multiple connect entries for multiple external response libraries and also for
the user defined beam cross sections discussed above.
Main Index Main Index
207 CHAPTER 4
Input Data
4.2 Executive Control
Solution 200
In design sensitivity and optimization, the only required Executive Control statement
is the SOL statement:
SOL 200
This states that subDMAP DESOPT is to be invoked, which is the main subDMAP for
design sensitivity and optimization.
Main Index Main Index


208
4.3 Case Control Section
The additions to the Case Control for design optimization are few and do not require
any modification of the Case Control commands already required for analysis.
Case Control commands for design sensitivity and optimization are associated with
the following four tasks:
1. Analysis discipline definition
2. Design task definition
3. Design response characterization
4. Shape basis vector computation
This section discusses each of these tasks individually.
Analysis Discipline Definition
Design optimization in MSC.Nastran is multidisciplinary: a number of different
analyses may be performed in Solution 200, and the results used simultaneously in
optimization.
The analysis types are defined on a subcase basis using the ANALYSIS Case Control
command. It can be set to any of the following values in SOL 200:
The following are a few Case Control issues to be aware of when using the ANALYSIS
command.
1. If ANALYSIS is specified above the subcase level, all subsequent subcases
utilize that analysis type until ANALYSIS is redefined in a later subcase.
STATICS
MODES
BUCKLING
DCEIG
DFREQ
MCEIG
MFREQ
MTRANS
SAERO
FLUTTER
ANALYSIS =
Main Index Main Index
209 CHAPTER 4
Input Data
2. The buckling subcase defined by ANALYSIS = BUCK must also identify a
STATICS subcase that supplies the displacement that is used in the global
buckling analysis. The STATIC subcase is selected using the STATSUB Case
Control command (STATSUB is only used with ANALYSIS = BUCK and is
not available for other analysis types).
3. Multiple boundary conditions may be used in Solution 200 for ANALYSIS =
STATICS, MODES, BUCKLING, SAERO and FLUTTER.
4. Only a single subcase is supported for ANALYSIS = DCEIG,MCEIG and
MTRAN.
5. In general, any number of analysis types can be selected in a single run, but
ANALYSIS=DFREQ and MFREQ cannot both be used in the same run.
6. Table 2-4 indicates which disciplines support multiple subcases and
boundary conditions.
Design Task Definition
The design model definition process in Case Control includes identification of the
design objective function and the design constraint sets.
Design Objective Identification
The design objective is identified in Case Control with the command:
where n is the set identification of a design response on either a DRESP1, DRESP2, or
DRESP3 Bulk Data entry. This response must be a single, scalar quantity.
A DESOBJ command appearing above the subcase level identifies a global response.
Weight and volume are typical examples of global responses. DESOBJ can also be
used at the subcase level if the design goal is to minimize or maximize a subcase-
dependent response.
A DESOBJ command is required in most cases, but there are certain situations where
it is not needed:
1. If only sensitivity results are of interest, a DESOBJ command is not required.
DESOBJ
min
max
\ .
| |
n =
Main Index Main Index


210
2. If Fully Stressed Design is being used, there is not a requirement for a
DESOBJ command. However, it is recommended that the DESOBJ command
be used and that it point to a WEIGHT response type so that the weight
response will be calculated and reported.
Note that if a DESOBJ request is needed for optimization and is not provided, the error
is not detected until optimization is attempted and therefore after an analysis and
sensitivity analysis has been performed. This can be an expensive error if these
analyses consume significant resources.
Design Constraint Identification
Design constraint sets are identified in Case Control by the following commands:
DESGLB = n
DESSUB = n
where n is the set identification number of a DCONSTR or a DCONADD Bulk Data
entry.
DESGLB is used above the subcase level to define a subcase-independent constraint
set. Weight and volume are subcase-independent responses, as are DRESP2 responses
that are not functions of DRESP1 responses (e.g., DRESP2s that are functions of design
variables, table constants, and grid coordinates only).
DESSUB defines subcase-dependent constraint sets at the subcase level. A DESSUB
command remains in effect until replaced with a new DESSUB in a subsequent
subcase.
Subcase Spanning Responses
A special situation occurs when you wish to combine responses from multiple
subcases. For example, you may want to constrain the average displacement across
multiple subcases. The constraint is on a global quantity in this case, so that the
DESGLB command just discussed is used above the subcase level. It is also necessary
to identify the subcase responses that are to be included in the subcase spanning
response. This is done with a DRSPAN command:
DRSPAN = n
The n is the set identification of a previously appearing SET case control command.
Main Index Main Index
211 CHAPTER 4
Input Data
The DRSPAN request occurs at the subcase level and most appear in every subcase
that contains DRESP1 quantities that are to be used in the DRESP2 that spans
subcases.
Design Variable Set Selection
The optional DESVAR Case Control command selects a set of design variables defined
by a SET command. If the command is absent, all DESVAR Bulk Data entries will be
utilized in the design task. Only one DESVAR Case Control command can appear and
it must appear above the subcase level.
The basic purpose of the DESVAR command is to address the scenario where the
design task started with many design variables, but it is desired to focus on a subset.
This could be done by editing the bulk data file to remove the unwanted DESVAR
entries, but this is a tedious process that is avoided using the DESVAR case control
command. It is still necessary to impose rules on what happens when these unneeded
DESVARs are reference on other bulk data entries. These rules include:
If a DVxRELy (e.g., DVPREL2) or DLINK entry references DESVARs that are
all included in the set specified in case control, then that property is
designed.
If a DVxRELy of DLINK entry references DESVARS that not included in the
set specified in case control, then that property is not designed
If a DVxRELy or DLINK entry references some DESVARs that are all
included in the set specified in case control and some that are not, a User
Fatal Message is produced and the run is terminated.
If a DRESP2 or DRESP3 references any DESVARs that are not included in the
set specified in case control, a User Fatal Message is produced and the run is
terminated.
The DESVAR command is:
Format:
DESVAR = n
The n is the set identification of a previously appearing SET Case Control command.
Main Index Main Index


212
Design Sensitivity Print
The meaning and importance of design sensitivity has been discussed at length in
Design Sensitivity Analysis on page 67. The design sensitivity data is quite useful
in its own right and can be printed using the DSAPRT Case Control command.
This command supports a number of options, the first of which indicates whether the
output is to be formatted. The formatting provides labelling information and is
usually the preferred option. The data can be exported to an external file using the
EXPORT option. This enables the use of the sensitivity data in another application.
The frequency of the output is controlled by three parameters: START, BY, and END.
These parameters refer to the design cycle so that START indicates after which design
cycle the print is to begin, BY indicates how often it is to be printed between the
START and END cycles and END identifies the final cycle. The right-hand side of the
command identifies the set of response IDs that are to be printed. The default of ALL
prints all the sensitivities.
Mode Tracking
A normal modes subcase (ANALYSIS = MODES) can constrain the normal modes
eigenvalues or frequencies based on mode number. As the design changes, the order
of the modes may change so that the physical nature of the mode being constrained
would change as well. As discussed in the Mode Tracking paragraphs of
Describers Meaning
All (or blank) All DESVAR Bulk Data entries will be processed.
n Set identification of a set of DESVAR Bulk Data entry identification
numbers. (Integer > 0)
Note: A response sensitivity is available for printing only if the response has been
constrained and then only if the constraint has survived the screening
process described in Constraint Screening on page 64. The DSCREEN
Bulk Data entry described in Bulk Data Entries on page 216 provides a
way of forcing the retention of responses that are of interest.
DSAPRT
FORMATTED
UNFORMATTED
NOPRINT
NOEXPORT
EXPORT
START i = [ ] BY y = [ ] END k = [ ] , , , ,
\ .
|
|
|
| |
ALL
n
NONE
=
Main Index Main Index
213 CHAPTER 4
Input Data
Multidisciplinary Analysis on page 58, MSC.Nastran can be made to track the
modes so that the constraint is being applied to the same physical mode as the design
changes. This is done using a MODTRAK Case Control command:
MODETRAK = n
where n invokes the ID of a MODTRAK Bulk Data entry.
Residual Vectors
The RESVEC Case Control command is used to augment the modes created from a
normal modes analysis with additional shapes tailored to the modal analysis task at
hand. The addition of these modes can greatly improve the accuracy of the finite
element analysis. The RESVEC Case Control command can be used to control the
creation of these additional modes and one of the describers of this command is
relevant here.
ADJLOD/ NOADJLOD is an option on the RESVEC command that indicates whether
adjoint loads should be used to create residual vectors. These loads are derived from
DRESP1 entries that invoke grid responses in a frequency or transient response
analysis. The ADJLOD option (the default) creates these residual vectors and is
recommended, while use of the NOADJLOD option is available to block the inclusion
of adjoint loads in residual vector computations.
Design Response Characterization
Strictly speaking, Case Control output requests are unnecessary in design sensitivity
and optimization. In a number of instances they can, however, be quite useful.
Solution 200 builds its own internal Case Control for data recovery based on the list of
DRESP1 responses identified in the design model. This provision ensures that all
necessary data recovery is always performed for design sensitivity and optimization.
However, if you want to view MSC.Nastran results, Case Control output requests are
necessary. The frequency of this output with respect to design cycle number is
controlled by the Bulk Data parameter NASPRT.
Case Control output commands are sometimes used to resolve design response
ambiguities. For example, element plate stresses can be output using either von Mises
or maximum shear. (The ambiguity arises since both share the same item code ID
referenced on the DRESP1 entry.) The form identified in Case Control is also used for
design sensitivity and optimization. For example,
Main Index Main Index


214
If left unspecified, the defaults will be used (which in this case, is the von Mises
representation).
Dynamic responses can be computed using either a real/ imaginary or a
magnitude/ phase form. Here again, Case Control can be used to define the output
representation. For example,
Shape Basis Vector Computation
Additions to Case Control for shape optimization are only necessary when using the
Analytic Boundary Shapes method.
Recall from Relating Design Variables to Shape Changes on page 166, that the
Analytic Boundary Shapes approach uses additional Bulk Data Sections to describe
the auxiliary boundary models. Each of these additional sections must have a
corresponding Case Control.
Each of the Auxiliary Boundary model Case Control Sections are identified by the
delimiter
AUXCASE
The auxiliary boundary model Case Control Sections must follow the Case Control
Section for the primary model.
Within the auxiliary boundary model Case Control Section, individual auxiliary
models are identified using
AUXMODEL = n
STRESS (VONMISES) = 15 (von Mises stresses used for analysis and optimization)
STRESS (SHEAR) = 15 (maximum shear stresses used for analysis and
optimization)
Note: The SET number (15 in the example above) has no effect on which responses
are used in the design.
STRESS(IMAG) = ALL (real/ imaginary form for analysis and optimization)
STRESS(PHASE) = ALL (magnitude/ phase form for analysis and optimization)
Main Index Main Index
215 CHAPTER 4
Input Data
where n is the auxiliary boundary model ID, referenced in the corresponding
BEGIN BULK = n Bulk Data Section delimiter. An example using AUXCASE and
AUXMODEL is shown in Relating Design Variables to Shape Changes on
page 166.
Generation of a New Bulk Data File
The ECHO Case Control command has an option that is applicable only in design
optimization:
ECHO = PUNCH(NEWBULK)
It is available to provide an unsorted Bulk Data file following a design optimization
task that has updated entries replacing the original entries. For example, the input
XINIT value on a DESVAR entry is replaced by the final value for this design variable
from the design task. Similarly, new thicknesses that are produced will be placed on
the updated PSHELL entry. This file is contained in the output .pch file that is
produced by MSC.Nastran. A complete description of the ECHO Case Control
command is available in ECHO on page 256 of the MSC.Nastran Quick Reference
Guide. Additional information on the output produced by the NEWBULK option can
be found in Special Punch Considerations for Topology Optimization on
page 376.
Main Index Main Index


216
4.4 Bulk Data Entries
The design model is defined in the Bulk Data Section. This definition includes the
design variables, the design variable-to-property relations, shape basis vectors, design
constraints, and so on. This section contains a brief listing and description of all the
Bulk Data entries related to design sensitivity and optimization. This information is
the alphabetical order of the entry name and has been extracted from the complete
listings in the The Bulk Data Section in Chapter 8 of the MSC.Nastran Quick
Reference Guide. For each entry in this section, related entries have also been listed with
appropriately shaded fields to help clarify the interrelations among the various
entries.
BNDGRID
Purpose:
Specifies a list of grid point identification numbers on design boundaries or surfaces
for shape optimization.
Entry Description:
Discussion:
Relating Design Variables to Shape Changes on page 166 identifies four
alternative methods for defining shape basis vectors and two of these methods:
Geometric Boundary Shapes and Analytic Boundary Shapes entail first defining the
shape basis vector on the boundary. This, in turn, requires the definition of which
degrees-of-freedom appear on the boundary. The BNDGRID entry performs this task
BNDGRID
C GP1 GP2 GP3 GP4 GP5 GP6 GP7
GP8 -etc.-
Field Contents
C Component number (any unique combination of integers 1 through 6 with
no embedded blanks).
GPi Shape boundary grid pint identification number.
Main Index Main Index
217 CHAPTER 4
Input Data
by selecting a component set and then identifying the associated grid points.
Analytic Boundary Shapes on page 430 provides an example that applies the
Analytic Boundary Shapes Method and makes use of the BNDGRID entry.
Note: GGrid coordinates that appear on the BNDGRID entry but that are not affected
by the displacement from the auxiliary model are assumed to not change their
location during the shape optimization task.
Main Index Main Index


218
DCONADD
Purpose:
Defines design constraints set as a union of DCONSTR entry sets.
Entry Description:
Associated Entries:
Discussion:
Constraints must be selected in Case Control. The DCONADD entry allows a number
of DCONSTR entry sets to be unified into a single set, making their selection in Case
Control easier.
DCONSTR
Purpose:
Places limits on a design response. When selected in Case Control by either DESGLB
or DESSUB, the DCONSTR sets define the design constraints.
Entry Description:
DCONADD
DCID DC1 DC2 DC3 etc.
Field Contents
DCID Design constraint set identification number. (Integer > 0)
DCi DCONSTR entry identification number. (Integer > 0)
DCONSTR
DCID RID
LALLOW UALLOW LOWFQ HIGHFQ
DCONSTR
DCID RID
LALLOW/
LID
UALLOW/
UID
LOWFQ HIGHFQ
Field Contents
DCID Design constraint set identification number. (Integer > 0)
Main Index Main Index
219 CHAPTER 4
Input Data
Associated Entries:
The IDs on DRESP1, DRESP2 and DRESP3 are the RIDs called out on the DCONSTR
entry.
RID DRESPi entry identification number. (Integer > 0)
LALLOW Lower bound on the response quantity. (Real, Default = -1.0E20)
LID Set identification of a TABLEDi entry that supplies the lower bound
as a function of frequency.
UALLOW Upper bound on the response quantity. (Real, Default = 1.0E20)
UID Set identification of a TABLEDi entry that supplies the upper bound
as a function of frequency.
LOWFQ Low end of frequency range in Hertz. (Real > 0.0, Default = 0.0)
HIGHFQ High end of frequency range in Hertz. (Real > LOWFQ,
Default = 1.0E+20)
DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1
ATT2 -etc.-
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1"
NR1 NR2 -etc.-
DRESP2
NRR1 NRR2
- - - -
DRESP3
ID LABEL GROUP TYPE
REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
DRESP2
NRR1 NRR2
- - - -
Main Index Main Index


220
The DCONADD entry can be used to form the union of a number of DCONSTR sets.
Discussion:
During the optimization process, material may be removed from and/ or added to the
structure to achieve the optimum objective. As this occurs, the response of the
structure to the applied loads changes. Member stresses, joint displacements, natural
frequencies, etc., are altered by the changes which the optimizer is making.
It is important, then, to place bounds on the structural responses so that the optimized
structure responds in a manner that is acceptable to the designer. For example, it is
common to place upper and lower bounds on the member stresses so that the
maximum allowable stresses are not exceeded. Joint displacement, buckling load
factors, and natural frequencies might also be bounded. When a bound is placed on a
structural response, this condition is called a constraint. First, second and third-level
responses may be constrained.
To constrain a given response quantity, a DCONSTR (Design CONSTRaint) entry is
used. This entry, in turn, points to a specific response ID identified on a DRESP1,
DRESP2 or DRESP3 Bulk Data entry. Note that the DCONSTR entry serves only to
place bounds on a response defined elsewhere in the design model data. These
bounds are used to construct normalized constraints for use by the optimizer as
discussed in Defining the Constraints on page 201. The LOWFQ and HIGHFQ
fields are only functional for response types with the FR prefix, for example FRDISP
and excepting RTYPE=FREQ. In these situations the bounds are only applied to the
response if the excitation frequency falls with the range of these two frequencies in Hz.
LOWFQ and HIGHFQ are ignored for other responses types. The upper and lower
bounds can be specified in one of two ways. If real numbers are used in the bound
field, these are the bounds. If integers are used, these point to TABLETi entries that
enable the input of these bounds as a function of frequency.
DCONSTR entry sets (or the DCONADD union of these sets) must be selected in the
Case Control Section to complete the constraint definition process. The Case Control
command DESGLB is used to identify subcase-independent constraints (global
constraints), while the DESSUB Case Control command is used to select subcase-
dependent constraint sets.
DCONADD
DCID DC1 DC2 DC3 -etc.-
Main Index Main Index
221 CHAPTER 4
Input Data
There are two guidelines for the user of the DCONSTR entry that can be noted here.
The first is that it is recommended that zero (and near zero) values be avoided when
specifying LALLOW and UALLOW. There are two reasons for this. The first is that
this causes problem in the normalization as explained in Design Constraints on
page 54. If the constraint limit really is zero (for example it is desired to restrict the
flutter damping to be less than zero), it is recommended that the response be recast
using a DRESP2 entry to offset the response by some amount. For example, if a value
of 1.0 is added to the flutter damping response, it will now be possible to place an
upper bound of 1.0 on the DRESP2 response. The second difficulty with a zero
response is that it can lead to unintended consequences in terms of constraint
screening. Suppose the response is the von Mises stress and it is desired that these
stresses never exceed 30KSI. The UALLOW value is then 30000.0 but it may seem
natural to set LALLOW to 0.0. The unintended consequence of this is that this may
create a critical constraint when the element is essentially unstressed. The optimizer
will try and satisfy this constraint with unpredictable results. In this case, it is
recommended that the LALLOW field be left blank so that the default value of -1.0E20
will be used and this will never be a critical constraint.
The second guideline is that is recommended that equality constraints allow some
tolerance as also discussed in Design Constraints on page 54. For example, if you
are trying to redesign the structure so that the first normal mode is exactly 6.0 Hz., you
should specify limits of LALLOW = 5.9 Hz. and UALLOW = 6.1 Hz. so that the
optimizer has a chance to satisfy both limits simultaneously. If the limits were both set
to 6.0, one or the other constraint would always be violated.
DDVAL
Purpose:
In a discrete optimization task, provide a list of discrete values that a design variable
can assume.
Entry Description:
DDVAL ID
DDVAL1 DDVAL2 DDVAL3 DDVAL4 DDVAL5 DDVAL6 DDVAL7
DDVAL8
-etc.-
Main Index Main Index


222
Associated Entries:
The ID on the DDVAL entry is called out the DDVAL field on the DESVAR entry:
Discussion:
The Discrete Variable Optimization capability discussed in Special Topics on
page 133 enables the designation of design variable values from a discrete set of
values. This selection is based on the conventional continuous optimization that is the
main topic of this Users Guide and provides a means of performing optimization
when the application requires the design variable take on a discrete value (perhaps
from available sheet metal thickness or, in a composite design, an integer number of
plies).
The DDVAL entry provides the allowable set of discrete values and is invoked by one
or more design variables. The real numbers can be input in arbitrary order and the
THRU and BY features provide convenient way of specifying a series of numbers.
If a DESVAR entry does not have a value in the DDVAL field, the design variable is
allowed to vary continuously in a discrete variable optimization task.
If the range of discrete values for a DESVAR entry do not extend out to the XLB and
XUB range specified on the DESVAR entry, the XLB and XUB values are adjusted to
extend only over the discrete values. Similarly, a discrete value that is outside the user
specified XLB/ XUB range will never be selected by the discrete optimization
algorithm.
DEQATN
Purpose:
Defines equations for use in synthetic relations. These equations can be used to define
either second-level responses or second-level design variable-to-property relations.
Field Contents
ID Unique discrete value set identification number. (Integer > 0)
DDVALi Discrete values. (Real)
DESVAR
ID LABEL XINIT XLB XUB DELXV DDVAL
Main Index Main Index
223 CHAPTER 4
Input Data
Entry Description:
Associated Entries:
The equation ID may be referenced in connection with a second-level response on a
DRESP2 entry or in the definition of a synthetic property relation on a DVPREL2
entry.
Discussion:
A unique feature of MSC.Nastran design sensitivity and optimization is that it allows
you to create new response quantities and nonlinear design variable-to-property
relations. This is accomplished by defining equations much like the function
definition procedure of many programming languages.
DEQATN
EQID EQUATION
EQUATION (Cont.)
Field Contents
EQID Unique equation identification number. (Integer > 0)
EQUATION Equation(s). (Character)
DRESP
2
ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1"
NR1 NR2 -etc.-
- - - -
DRESP
2
NRR1 NRR2
DVPREL2
ID TYPE PID FNAME PMIN PMAX EQID
DESVAR
DVID1 DVID2 DVID3 -etc.-
DTABLE
LABL1 LABL2 LABL3 -etc.-
Main Index Main Index


224
The DEQATN entry is used to specify a single equation or a set of nested equations.
The syntax of the expressions follows the FORTRAN language standard except that
all arguments are assumed to be real numbers (no integers). Intrinsic functions, such
as SIN or MAX, may be used.
The equation input arguments are formal arguments that are defined at runtime. The
DRESP2 entry for responses and the DVCREL2, DVMREL2, DVPREL2 entries for
design variable-to-property relations specify the input, or actual, arguments to the
appropriate DEQATN entry. A nested form of equation definition may be used.
Multiple equations can be specified on a single DEQATN entry with the input to the
latter equations based on the results of the previous equations. The return value of the
function call is the last equation in the DEQATN entry.
The power and versatility of the DEQATN, particularly when used with the DRESP2
entry, has given rise to a number of special requests for this entry to extend the ease-
of-use of the DEQATN. The features that have been provided in response to these
requests are explicitly pointed out here because they have primary applicability in
design sensitivity and optimization.
Special Functions
The majority of the intrinsic functions used in the DEQATN are those that are
available in a programming language such as FORTRAN; e.g., ABS for absolute value
and TAN for tangent. However, there are eight functions as shown in Table 4-1 that
have been provided, primarily, to give support in dynamic response design and are
applicable for any relevant analysis. The SUM,RSS, SSQ and AVG functions provide
a shorthand operation for operating on a series of arguments. (Refer to the DRESP2
entry for an alternative means of performing these operation and to the ATTB field of
the DRESP1 entry for the frequency response RTYPEs for another simplified way of
inputting these operations in special cases.) The use of these operators on the
DEQATN is not limited to responses so that you could use the AVG function to find,
for example, the average of a set of design variables). The DB,DBA, INVB and
INVDBA provide a simplified way of dealing with acoustic responses.
Main Index Main Index
225 CHAPTER 4
Input Data
where:
Table 4-1 Special DEQATN Functions
Format Description Mathematical Expression
AVG( ) average
DB(P, PREF) sound
pressure in
decibel
DBA(P, PREF, F) sound
pressure in
decibel
(perceived)
INVDB(DB, PREF) inverse Db
INVDBA(DBA, PREF,
F)
inverse Dba
RSS square root
of sum of
squares
SSQ sum of
squares
SUM summation
X
1
X
2
X
n
, , , ( )
1
n
--- X
i
i 1 =
n

20.0 log
P
PREF
----------------
\ .
| |

20.0 log
P
PREF
---------------
\ .
| |
10.0 + log Ta1 ( ) 10.0 + log Ta2 ( )
10
DB
20.0
---------- log PREF +
\ .
| |
10
DBA 10.0 log Ta1 ( ) 10.0 log Ta2 ( )
20.0
-------------------------------------------------------------------------------------------------------------
\ .
| |
X
1
X
2
S
n
, , , ( )
X
i
2
i 1 =
n

X
1
X
2
S
n
, , , ( )
X
i
2
i 1 =
n

X
1
X
2
S
n
, , , ( )
X
i
i 1 =
n

Main Index Main Index

226
Implied Forcing Frequency Argument
Frequency input to synthetic relations can be supplied in an implicit format that is
activated by:
1. Frequency response-type ("FR" or PSD) arguments for the DRESP1 IDs
invoked by a DRESP2.
2. A DEQATN entry whose input argument list exceeds the number of
arguments defined on the referencing DRESP2 entry.
The situation is perhaps most easily described by way of an example. Assume a simple
average is to be computed based on frequency response-computed displacements that
are "normalized" by their respective forcing frequencies. We can then write:
= structure responses or acoustic pressure
PREF = reference pressure
F = forcing frequency
DB = acoustic pressure in Decibel
DBA = perceived acoustic pressure in Decibel
=
=
= 2.242882e+16
= 1.562339
= 20.598997
= 107.65265
= 737.86223
= 12194.22
X
1
X
2
S
n
P , , , ,
Ta1
K3 F
4

F
2
P2
2
+ ( ) F
2
P3
2
+ ( )
------------------------------------------------------
Ta2
K1 F
4

F
2
P1
2
+ ( )
2
F
2
P4
2
+ ( )
2
----------------------------------------------------------- -
K1
K3
P1
P2
P3
P4
Main Index Main Index
227 CHAPTER 4
Input Data
Normally, the number of input arguments and values specified on the
DEQATN/ DRESP2 entries must match. The rules, however, are relaxed when
frequency response quantities (i.e., those preceded by "FR") are encountered as input
to the equation. In such cases, additional arguments can be defined via implicit
reference to the list of frequency values appearing on the DRESP1 entries, in the
encountered order. In the above example, f1 and f2 are 10. hertz and 20. hertz,
respectively.
The above capability is particularly useful in the case of frequency definitions via the
FREQ3, FREQ4, and FREQ5 entries. Again, a simple example explains this feature.
The presence of FREQ3, FREQ4, or FREQ5 entries is assumed in the following
example:
A DRESP1/ DRESP2 pair of responses will be generated for every forcing frequency,
the value of which is generally unknown at the outset. Note that the third input to the
equation, F, is in excess of the number of arguments defined in the DRESP2 relation.
Because a frequency response displacement is referenced, the code will automatically
assign its value to be the corresponding forcing frequency.
$RESP1, ID, LABEL, RTYPE, PTYPE, REGION, ATTA, ATTB, ATT1, +
$, ATT2, ...
dresp1, 10, fdisp1, frdisp, , , 1, 10., 1001
dresp1, 20, fdisp2, frdisp, , , 1, 20., 1001
$RESP2, ID, LABEL, EQID, REGION, , , , , +
$, DRESP1, NR1, NR2, ..., , , , , +
dresp2, 30, avgfd, 100, , , , , , +
+, dresp1, 10, 20
deqatn 100 avg(d1,d2,f1,f2) = (d1/f1 + d2/f2)*0.5
DRESP1,10,FRGD,FRDISP,,,1,,21
DRESP2,101,DBA,201,
,DTABLE,PREF,,,,,,,
,DRESP1,10
DTABLE,PREF,1.0
DEQATN 201 X(PREF,P,F)=DBA(P,PREF,F)
Main Index Main Index


228
DESVAR
Purpose:
Defines the design variables to be used in design sensitivity and optimization. Design
sensitivity analysis computes the rates of change of design responses with respect to
changes in the design variables. In design optimization, the set of design variables are
the quantities modified by the optimizer in the search for an improved design.
The optional DESVAR Case Control command can be used to specify the set of
DESVAR Bulk Data entries that are to be used in the design task. If the DESVAR Case
Control command is absent, all DESVAR Bulk Data entries will be used.
Entry Description:
Associated Entries:
Design variables may be related to properties on DVPREL1, DVCREL1, DVMREL1 or
DVPREL2, DVCREL2, DVMREL2 entries, related to changes in shape using
DVBSHAP, DVGRID, or DVSHAP entries, linked using DLINK entries or input to
user-defined responses on the DRESP2 entry and external responses using the
DRESP3 entry. The DVCRELi and DVMCRELi entries are not shown here because of
their close similarity with the DVPRELi entries.
DESVAR ID LABEL XINIT XLB XUB DELXV DDVAL
Field Contents
ID Unique design variable identification number. (Integer > 0)
LABEL User-supplied name for printing purposes. (Character)
XINIT Initial value. (Real, XLB < XINIT < XUB)
XLB Lower bound. (Real, default = -1.0E+20)
XUB Upper bound. (Real, default = +1.0E+20)
DELXV
DDVAL
Move limit for the design variable during approximate
optimization. (Real > 0.0)
ID of a DDVAL entry that provides a set of allowable discrete
values. (Blank or integer > 0, Default = blank for continuous design
variables.)
Main Index Main Index
229 CHAPTER 4
Input Data
DVPREL1
ID TYPE PID PMIN PMAX C0
DVID1 COEF1 DVID2 COEF2 DVID3 -etc.-
DVPREL2
ID TYPE PID PMIN PMAX EQID
DESVAR
DVID1 DVID2 DVID3 -etc.-
DTABLE
LABL1 LABL2 LABL3 -etc.-
DVBSHAP
DVID
AUXMOD
COL1 SF1 COL2 SF2 COL3 SF3
DVGRID
DVID GID CID COEFF N1 N2 N3
DVSHAP
DVID COL1 SF1 COL2 SF2 COL3 SF3
DLINK ID DDVID C0 CMULT IDV1 C1 IDV2 C2
IDV3 C3 -etc.-
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1"
NR1 NR2 -etc.-
- - - - - -
DRESP3 ID LABEL GROPU TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - -
USRDATA
GEOMETRY DATA FILE
Main Index Main Index


230
Discussion:
In design optimization, structural properties are changed in order to determine the
optimal value of the objective function. To accomplish this, the optimizer varies the
design variables. Not only must design variables be defined, but they also must be
functionally related to analysis model properties and/ or changes in structural shapes.
Design variables can also be used as design parameters that are unrelated to the
structural properties. In this case, they are invoked by DRESP2 or DRESP3 entries. The
DESVAR entry is used to define an individual design variable.
The DELXV attribute is used to specify the amount a design variable can change
during one optimization cycle. This relates to the The Approximate Model on
page 101 and should be used when it appears that the results from the approximate
analysis are not closely matched by the subsequent exact analysis. DELXV was
provided primarily for shape optimization tasks but is applicable in any context.
The DDVAL field enables the Discrete Variable Optimization capability of Special
Topics on page 133.
Design variables may be defined as members of independent and dependent sets
using DLINK Bulk Data entries, if desired.
DLINK
Purpose:
Imposes a linear relationship among the design variables. The set of all DLINK entries
partitions the design variables into an independent set and a dependent set.
Entry Description:
DLINK ID DDVID CO CMULT IDV1 C1 IDV2 C2
IDV3 C3 -etc.-
Field Contents
ID Unique entry identifier. (Integer > 0)
DDVID Dependent design variable identification number. (Integer > 0)
C0 Constant term. (Real, default = 0.0)
CMULT Constant multiplier. (Real, default = 1.0)
Main Index Main Index
231 CHAPTER 4
Input Data
Associated Entries:
Design variables are identified in DLINK relations by their IDs (both independent and
dependent variables).
Discussion:
A DLINK entry specifies a design variable relationship of the form:
where:
Design variable linking can be used, for example, to ensure structural symmetry or
unify sizing changes across property groups. In this way, you can perform what-if
studies to see how different modeling assumptions affect the derived design. The
efficiency of the design process is usually improved if the number of independent
design variables can be kept to a minimum. Placing a design variable in the dependent
set using a DLINK entry removes it from the independent set.
DOPTPRM
Purpose:
Overrides default values of parameters used in design optimization.
IDVi Independent design variable identification number. (Integer > 0)
Ci Coefficient i (corresponding to IDVi). (Real)
DESVAR ID LABEL XINIT XLB XUB DELXV DDVAL
= dependent design variable
= constant
= constant multiplying coefficient
= independent design variables
x
D
c
o
= c
i
x
i
i

+
x
D
c
o
c
i
x
i
Main Index Main Index


232
Entry Description:
Associated Entries:
None.
Discussion:
There are numerous parameters that control various aspects of the optimization
process itself. While all of these parameters have defaults (the DOPTPRM entry is
optional), the defaults may be changed using the DOPTPRM entry. An overview of
selected parameters, grouped according to their functionality is presented here. See
also the The Bulk Data Section on page 895 of the MSC.Nastran Quick Reference
Guide as it contains the complete list of all optimization parameters that may be
changed.
Choice of Approximation Method (APRCOD)
There are three types of approximation methods to choose from: direct linearization,
mixed method, and convex linearization (see Function Evaluation on page 111).
The mixed method is the default.
Direct linearization (APRCOD = 1) is based on the simple first-order Taylor series
expansion directly in terms of the design variables. The method is often useful for
dynamic response optimization, shape optimization, and optimization tasks that use
basis vector formulations.
The mixed method (APRCOD = 2) uses a combination of direct and reciprocal
approximations, depending on the response type being approximated (see Table 4-6).
DOPTPRM PARAM1
VAL1
PARAM2
VAL2
PARAM3
VAL3
PARAM4
VAL4
PARAM5
VAL5 -etc.-
Field Contents
PARAMi Name of the design optimization parameter. For allowable names,
see the DOPTPRM listing in The Bulk Data Section on page 895
of the MSC.Nastran Quick Reference Guide. (Character)
VALi Value of the parameter. (Real or Integer; see the DOPTPRM listing
in The Bulk Data Section on page 895 of the MSC.Nastran Quick
Reference Guide.)
Main Index Main Index
233 CHAPTER 4
Input Data
Convex linearization (APRCOD = 3) uses either a direct or reciprocal approximation,
depending on which one yields the more conservative approximation. The choice is
made on an individual design variable and individual constraint basis.
Optimization Method (ADSCOD, METHOD)
Optimization Method (OPTCOD, METHOD)
The OPTCOD parameter permits the specification of the optimization code to be used
as shown in Table 4-2. It is seen that MSCADS is the default for shape and sizing
optimization while the familiar DOT code is available as an option. The BIGDOT code
is for topology optimization (that is, if a TOPVAR bulk data entry is present). If you
wish to perform topology optimization but do not have the Topology Optimization
option installed on your computer, you will need to explicitly select either the
MSCADS or DOT optimization code.
The METHOD parameter selects the particular algorithm that is to be used with the
selected OPTCOD as shown in Table 4-3. BIGDOT has only one algorithm, so the
parameter is only relevant for MSCADS or DOT. METHOD=4 is sequential
unconstrained minimization technique (SUMT) and is only available with MSCADS.
METHOD=ijk is provides access to a suite of MSCADS optimization techniques, with
the I specifying the strategy, j the optimizer and k the one-dimensional search method
as shown in Table 4-4. No guidelines are provided at this time in the use of these
alternatives, they are simply made available for knowledgeable users who are
encountering difficulties with one of the basic methods.
Table 4-2 OPTCOD Options Design Optimization Parameters
Name Description, Type and Default Value
OPTCOD OPTCOD (Character; Default = Blank)
= Blank (taken from system cell number 413;
Default=MSCADS for Design Optimization Option
and BIGDOT for Topology Optimization Option)
= MSCADS MSCADS is used
= DOT DOT is used
= BIGDOT BIGDOT is used
Main Index Main Index


234
Table 4-3 METHOD Options
Name Description, Type and Default Value
METHOD Optimization method (Integer > 0; Default = 1)
1 Modified method of feasible directions for both
MSCADS and DOT
2 Sequential linear programming for both MSCADS
and DOT
3 Sequential quadratic programming for both
MSCADS and DOT
4IJK See Table 4-4
Table 4-4 MSCADS Options
Name Description, Type and Default Value
1 Sequential unconstrained minimization using the
exterior penalty function method
2 Sequential unconstrained minimization using the
linear extended interior penalty function method
3 Sequential unconstrained minimization using the
quadratic extended interior penalty function
method
4 Sequential unconstrained minimization using the
cubic extended interior penalty function method
5 Augmented Lagrange multiplier method
6 Sequential linear programming
7 Method of centers
8 Sequential quadratic programming
9 Sequential convex programming
MSCADS optimizer options (Integer 1-5)
1 Fletcher-Reeves algorithm for unconstrained
minimization
Main Index Main Index
235 CHAPTER 4
Input Data
2 Davidson-Fletcher-Powell (DFP) variable metric
method for unconstrained minimization
3 Broydon-Fletcher-Goldfarb-Shanno (BFGS) variable
metric method for unconstrained minimization
4 Method of feasible directions for constrained
minimization
5 Modified method of feasible directions for
constrained minimization
MSCADS one-dimensional search options (integer 1-8)
1 Find the minimum of an unconstrained function
using the Golden Section method
2 Find the minimum of an unconstrained function
using the Golden Section method followed by
polynomial interpolation
3 Find the minimum of an unconstrained function by
first finding bounds and then using polynomial
interpolation
4 Find the minimum of an unconstrained function by
polynomial interpolation/ extrapolation without
first finding bounds on the solution
5 Find the minimum of a constrained function using
the Golden Section method
6 Find the minimum of a constrained function using
the Golden Section method followed by polynomial
interpolation
7 Find the minimum of a constrained function by first
finding bounds and then using polynomial
interpolation
8 Find the minimum of a constrained function by
polynomial interpolation/ extrapolation without
first finding bounds on the solution
Table 4-4 MSCADS Options (continued)
Name Description, Type and Default Value
Main Index Main Index


236
Experience has shown that the MMFD and SQP algorithms give comparable
performance; sometimes MMFD does better and sometimes SQP is best. The SLP
algorithm is regarded as being slightly inferior to the other two, but no definitive
guidelines can be offered. One comment is that if it seems apparent that a particular
algorithm is not performing well, trying a different algorithm may result in improved
performance.
Transformation of the Approximate Optimization Task to a Feasible
Design (PENAL)
The PENAL parameter can be used to transform an infeasible design into a feasible
one using the equations
where the algorithm initializes B to a value equal to the maximum value of (if this
is a negative number, the transformation is not required). is the initial value of the
objective.
This parameter has been shown to be most useful when applied with the ADS
algorithms. PENAL=2.0 is recommended.
Design Cycle Print Controls (P1, P2, P2CR, P2CDDV, P2CP, P2CC,
P2CM, P2CBL, P2CALL, P2RSET)
Insight into the progress of the optimization task can be obtained by requesting prints
of selected information during the optimization. This can potentially produce an
unmanageable amount of data so that a number of DOPTPRM parameters are
available for specifying this print. P1 controls the frequency of the output:
P2 provides a first level control of which design quantities are printed:
P1 = 0 output for initial and optimal designs (default)
= n output for every n-th design cycle
P2 = 0 no output
= 1 objective and design variables (default)
g
j
i
g
j
B j 1 2 ncon , , = ( ) =

1
B PENAL
0
+ =
g
j

0
Main Index Main Index
237 CHAPTER 4
Input Data
The various P2 options can be summed to produce output combinations. For example,
P2 = 15 prints out all available design data. The P2 = 16 option is the result of a special
request and is not directly related to design.
Eight additional parameters control constraint and response prints once these prints
have initially been invoked with the P2 parameter.
P2CALL can be used to provide a default for any or all of the categories. For example,
P2CALL = 10 will print up to 10 of each type of constraint. If there are fewer than ten
constraints of a give type available, only the available number will be printed. Note
that P2CALL applies to each type of constraint individually. No attempt is made to
find the most critical constraints across all types and only print out these constraints.
The remaining P2Cx parameters allow you to selectively specify the number of
constraints printed for that constraint type. If the number specified is fewer than the
number available, a sort is performed to find the P2Cx most critical and only these are
printed. If the number specified exceeds the number of constraints available, no
sorting is performed and all the constraints are printed in the order of increasing
internal constraint ID.
= 2 designed properties
= 4 design constraints
= 8 design responses
= 16 weight as a function of material ID
P2CR Maximum number of Constraints on Responses to be printed
P2CDDV Maximum number of Constraints on Dependent Design Variables to be
printed
P2CP Maximum number of Constraints on Properties to be printed
P2CC Maximum number of Constraints on Connectivity properties to be
printed
P2CM Maximum number of Constraints on Material properties to be printed
P2CBL Maximum number of Constraints on Beam Library dimensions to be
printed
P2CALL Maximum number of Constraints of all categories to be printed
P2RSET SET1 ID for which a set of Response IDs is defined
Main Index Main Index


238
P2RSET can be utilized to print a selected set of responses. P2RSET refers to a SET1
Bulk Data entry that contains a list of DRESPx IDs that are to be printed. Only
retained responses (i.e., those that survived the constraint screening process) are
printed.
The print of responses can now be affected by three DOPTPRM parameters: P2, P2CR
and P2RSET. Figure 4-1 depicts how the three parameters affect the print.
Figure 4-1 Flowchart Showing How P2, P2RSET and P2CR Affect the Print of
Retained Design Responses
Optimizer Level Print Control (IPRINT)
The parameter IPRINT controls the amount of optimizer output. This parameter is
normally turned off (default = 0) by users since the optimizer is used in a black-box
mode. However, sometimes it is useful to follow what is happening within the
optimizer itself. The levels of optimizer print are as follows:
P2 Set for
Response Print
No Response
Print
P2RSET > 0
P2CR > 0
P2RSET > 0
Print all retained
responses in set
P2RSET
Print only
retained
response
associated
with
printed
All retained
responses printed
N
N
N
N
Y
Y
Y
Y
P2RSET = -1
Y
Main Index Main Index
239 CHAPTER 4
Input Data
Note that IPRINT > 3 produces constraint prints. These constraint values are in the
internal constraint order, which is associated with the constraint evaluation and is not
affected by the print order produced with the P2Cx parameters. The internal
constraint ID is still printed when the P2Cx parameters are used, but the constraints
are unlikely to be in ascending order of this ID.
Maximum Number of Mathematical Programming Design Cycles
(DESMAX)
The optimization process is iterative since the optimizer obtains data about the design
space from approximations. The approximate model, constructed based on a detailed
finite element analysis, is used by the optimizer to find an approximate optimum. This
design is resubmitted for another finite element analysis followed by another
approximate optimization. This process is repeated until convergence with respect to
these overall design cycles is reached or until the maximum specified number of
design cycles (DESMAX) is reached. The default value of DESMAX is five. Usually,
good progress toward a optimal design has been made by this point and its a good
time to review the results. However, more cycles are often necessary and the
DESMAX parameter can be adjusted at the users discretion. At other times, it is
desirable to set DESMAX to one so that you can make sure that everything appears to
be in order before committing to a longer run. If the maximum number of design
cycles is reached before convergence is achieved, the problem can always be restarted
from the last design (see Topology Optimization in Chapter 7).
IPRINT Optimizer Printout
0 No output (default)
1 Internal parameters, initial information, and results
2 Same, plus objective function and design variables at each iteration
3 Same, plus constraint values and identification of critical
constraints
4 Same, plus gradients
5 Same, plus search direction
6 Same, plus scaling factors and miscellaneous search information
7 Same, plus one-dimensional search information
Main Index Main Index


240
Fully Stressed Design (FSDMAX, FSDALP)
The Fully Stressed Design (FSD) algorithm discussed in Superelement
Optimization on page 144 has a design loop that is independent from the
mathematical programming loop. The maximum number of FSD design cycles is
specified by FSDMAX, which has a default value of 0 (no FSD cycles). Once the FSD
cycles have been completed (either from convergence of after FSDMAX design cycles),
the algorithm automatically transitions to the mathematical programming algorithm
with DESMAX used to control the maximum number of additional cycles that will be
made. Setting DESMAX to 0 will prevent any mathematical programming cycles.
The FSDALP parameter controls the relaxation factor, , used in the resizing equation
of Eq. 2-161. The default value of 0.9 for FSDALP is usually adequate to achieve quick
convergence. Lower values, such as 0.5, can be used to achieve a completely
converged solution.
Move Limits on the Approximate Optimization (DELP, DPMIN, DELX,
DXMIN)
As the optimizer modifies the design variables, the structures properties and/ or
shape will vary depending on the design model description. As discussed in The
Approximate Model on page 101, move limits need to be placed on the approximate
subproblem for efficiency reasons. These move limits are imposed with respect to
analysis model properties as well as design variables. They can be changed from their
defaults by modifying DELP and DPMIN for properties, and DELX and DXMIN for
design variables. The DELXV attribute on the DESVAR entry allows control in the
movement of an individual design variable. A recommendation on DELP is that its
value should always be less than 1.0. A value of 1.0 or greater could allow the property
to go to zero or even negative, with unpredictable results.
Convergence Criteria (CONV1, CONV2, GMAX, CONVDV, CONVPR)
The parameters CONV1, CONV2, GMAX, CONVDV, and CONVPR are used to test
for overall design cycle convergence. These parameters are used in connection with
tests for both hard and soft convergence. Tests for Convergence on page 125,
describes the types of convergence testing as well as the convergence decision logic.

Main Index Main Index

241 CHAPTER 4
Input Data
Soft convergence compares the design variables and design properties output from
the approximated model optimization with these same values at the beginning of the
design cycle. Soft convergence is not sufficient to terminate the design cycle iterations
unless the parameter SOFTEXIT is set to YES. NO is the default and it can be changed
with a PARAM Bulk Data entry, not with the DOPTPRM entry.
Hard convergence testing compares the analysis results of current design cycle with
those of the previous cycle. This test is a more conclusive test of convergence since it
is based on hard evidence. Hard convergence will always terminate the design cycle
process.
Convergence at the optimizer level can also be controlled using the DOPTPRM entry.
Parameters that can be changed include DABOBJ, DELOBJ, ITMAX, ITRMOP,
ITRMST, and JTMAX.
Finite Difference Perturbations (DELB, STPCSL)
Design sensitivity coefficients are computed in MSC.Nastran using a semianalytic
approach, as explained in Design Sensitivity Analysis on page 67.
The value of the property finite difference move parameter DELB of Eq. 2-47 and
Eq. 2-48 can be changed with the DOPTPRM entry although the default value of 1.0E-
4 gives good results in most cases.
The value of the shape finite difference move parameter STPCSL of Eq. 2-49 can also
be changed with the DOPTPRM entry.
Comparison Between the Design and Analysis Properties (PTOL,
PLVIOL)
The initial value of an analysis property given on a property Bulk Data entry may not
always correspond to the value of the property calculated using the initial values of
the design variables. If the relative difference between the properties differ by more
than PTOL, the program terminates with an error condition. The default of 1.0E35
implies that nearly all discrepancies are tolerated and the initial analysis property
values will be overridden using the design model data. If you want your analysis
models and design models to agree, setting PTOL to a value close to zero will
terminate the run when the two models differ significantly. A related DOPTPRM
parameter is PLVIOL, which can be set to a non-zero integer to allow the run to
continue when the initial designed properties violate the limits imposed on the
Main Index Main Index


242
properties by the PMIN and PMAX values on the DVxRELy entries. Allowing these
violations is not good practice since it puts a burden on the optimizer to satisfy a
violated property constraint that is more properly satisfied when preparing the input.
A final comment on analysis versus design property values is that the DPEPS
parameter entered with a PARAM entry allows you to specify a difference value that
is used to assess whether the analysis properties are close enough to the designed
properties that previously computed results using the analysis properties can be used
in subsequent sensitivity analysis. The default value of DPEPS = 1.0E-4 allows for only
a very small difference so that a new analysis will be performed whenever the design
properties differ from the analysis properties. Setting this parameter to a larger value
allows greater differences and perhaps avoids a finite element analysis. But it must be
realized that the sensitivities, which are computed using the design values, are not
fully consistent with the analysis values which have been calculated using the analysis
values.
Discrete Variable Optimization (DISCOD, DISBEG)
As discussed in Special Topics on page 133, a continuous optimization task can be
followed by an essentially postprocessing operation which determines the values of
the design variables which provides an optimal design while limiting the design
variable values to a discrete set of values specified using the DDVAL entry. Two
DOPTPRM parameters (DISCOD and DISBEG) control the discrete variable
processing. The DISBEG parameter specifies at which mathematical programming
design cycle you want to start performing discrete optimization. The default value of
0 performs discrete optimization only after the continuous optimization process is
complete. The DISCOD parameter selects from one of the four discrete optimization
alternative methods.
DISCOD Value Discrete Optimization Method
1 Design of Experiments (Default)
2 Conservative Discrete Design
3 Round up to the nearest design variable
4 Round off to the nearest design variable
0 Turn off discrete variable processing
Main Index Main Index
243 CHAPTER 4
Input Data
The DISCOD = 0 option facilitates ignoring DDVAL data on the DESVAR entry when
a discrete variable optimization task is not desired.
Constraint Normalization (GSCAL)
Eq. 2-17 indicates how the constraints are normalized and makes reference to a
GSCAL parameter. This parameter can be modified from is default value of 0.001
using the DOPTPRM entry. As the discussion following Eq. 2-17 indicates, it is not
good practice to define constraints in a way that requires the use of GSCAL. Instead,
the response should be recast in a form (perhaps through the use of a DRESP2 entry)
that avoids the small bounds that invoke the use of GSCAL.
Active and Violated Constraints (CT, CTMIN)
The DOT optimizer makes a determination of what constraints it should consider
when performing its task that is over and above the screening actions discussed in
Constraint Screening on page 64. DOPTPRM parameter CT is used to define that
threshold above which a constraint is considered active while CTMIN is used to
identify violated constraints. (See Figure 2-12 and Figure 2-13.) The default values of
-0.03 for CT and 0.003 for CTMIN have been found to be adequate in most cases. When
it appears that the default value is causing the optimizer to neglect constraints that
become critical once the redesign is performed, the CT value can be changed to a value
such as -0.1.
Topology Optimization Variables (TCHECK and TDMIN) and Special
Topology Optimization Settings
Topology optimization has two parameters that are unique to it. Parameter TCHECK
is used to turn on a filtering parameter that is used to prevent a checkerboard pattern
from being produced. The default value of 1 turns on the filtering and is
recommended. TCHECK=0 turns off filtering.
The second parameter, TDMIN, is used to prevent achieving a final design that is
characterized by thin disjoint fibers that are impractical from a manufacturing
standpoint. Design variable elements that are within a distance of TDMIN from an
element with a design variable close to 1.0 are filtered to ensure they are not small.
Note that the value of TDMIN is problem-size dependent but not mesh-size
dependent. The default for TDMIN is 0.0 (no filtering).
Main Index Main Index


244
Tuning of the topology optimization algorithm has indicated the following
parameters benefit from a different default value than is used in standard design
optimization.
DOPTPRM parameter P2 also has a different meaning for topology optimization in
that design variables are printed only when P2 > 8. This is because topology
optimization tasks typically involve thousands of design variables.
DRESP1
Purpose:
Defines direct, or first-level, analysis responses to be used in design sensitivity and
optimization. The responses identified here are those which are directly available
from the analysis results as opposed to second-level responses which are defined
using DRESP2 and DEQATN entries or third-level responses which are defined using
DRESP3 and an external response evaluator.
Entry Description:
Table 4-5 Modified DOPTPRM Default for Topology Optimization
Parameter Topology Default
DESMAX 30
CONV1 1.0E-4
CONVDV 1.0E-4
DELX 0.2
DXMIN 1.0E-5
DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1
ATT2 -etc.-
Field Contents
ID Unique entry identifier. (Integer > 0)
LABEL User-defined label. (Character)
Main Index Main Index
245 CHAPTER 4
Input Data
Associated Entries:
A response identified on a DRESP1 entry may be used either as a constraint or an
objective. For constraints, the response is identified by its ID number on DCONSTR
entries.
For the objective, a first-level response must define a single, scalar response (e.g.,
weight, an eigenvalue, a grid displacement component for a single subcase, and so
on). The objective is defined in Case Control with the command
where n is the DRESP1 entry ID. (Note that the n could optionally point to a DRESP2
or DRESP3 ID.)
A DRESP1 response can also be used as input to synthetic responses defined on
DRESP2 entries or by an external response on a DRESP3. It is referenced by its
response ID.
RTYPE Response type. See the DRESP1 listing in The Bulk Data Section on
page 895 of the MSC.Nastran Quick Reference Guide. (Character)
PTYPE Element flag (PTYPE = ELEM) or property entry name. Used with
element type responses (stress, strain, force, etc.) to identify the
property type, since property entry IDs are not unique across property
types. (Character: ELEM, PBAR, PSHELL, etc.)
REGION Region identifier for constraint screening. (Integer > 0)
ATTA, ATTB,
ATTi
Response attributes. See the DRESP1 listing in Bulk Data entries in
Appendix C. (Integer > 0 or Real or blank)
DCONSTR
DCID RID
LALLOW UALLOW
DRESP2 ID LABEL EQID REGION
- - - -
DTABLE
LABL1 LABL2 -etc.-
DRESP1"
NR1 NR2 -etc.-
- - - - - -
DESOBJ
mi n
max \ .
|
| |
n =
Main Index Main Index


246
If a DRESP1 is to be used on a DRESP2 or DRESP3 that combines responses from two
or more subcases, the DRSPAN Case Control command is required at the subcase
level to indicate this:
DRSPAN = m,n,...
where m and n are DRESP1 entry IDs.
Discussion:
Depending on the particular analysis discipline, responses such as displacements,
stresses, eigenvalues, etc., may be computed. The DRESP1 entry is used to identify
responses that are to be used in connection with design sensitivity and optimization
either as an objective or as a constraint. These responses may also be used as input to
compute a second-level, or synthetic, response on a DRESP2 entry or passed to a third-
level, or external, response on a DRESP3 entry for application in a user defined
response.
An internal Case Control for design sensitivity and optimization is built using the set
of first-level responses. Selecting design responses using DRESP1 entries also ensures
that the appropriate data recovery is performed, regardless of Case Control requests
the user may have supplied. Output requests, if desired, still must be specified with
the appropriate Case Control commands; however, this is for postprocessing
convenience only and is not a requirement for either design sensitivity or
optimization.
Every DRESP1 entry has a unique ID, a user defined label and a response type. The
response type identifies what type of response quantities is of interest. Table 4-6
identifies each of the 34 unique response types, indicates which ANALYSIS discipline
(as defined on the ANALYSIS case control command) they are applicable for, the
approximation method (direct or inverse) that is used when APRCOD=2 and the
default REGION. If a response type is inconsistent with the ANALYSIS command as
given in the table for a particular subcase, it is a User Fatal Error and the run is
terminated.
DRESP3 ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
DRESP1"
NR1 NR2 -etc.-
Main Index Main Index
247 CHAPTER 4
Input Data
The REGION attribute on this entry is used in the constraint screening process
discussed in Constraint Screening on page 64. If this field is left blank, the defaults
of Table 4-6 apply. The N/ A designations in this table refer to scalar responses where
the concept of default REGION is not applicable. If a REGION is set, all responses that
share the REGION ID and have the same response type will be grouped together
during the screening process.
Table 4-6 Response Type Options on the DRESP1 Entry
Response
Types
Associated
ANALYSIS Type
Approximation
Technique
Default Region
WEIGHT Any or None Direct N/ A
VOLUME Any or None Direct N/ A
EIGN MODES Rayleigh Quotient N/ A
CEIG DCEIG or MCEIG Inverse N/ A
FREQ MODES Rayleigh Quotient N/ A
LAMA BUCK Direct* N/ A
DISP STATICS or SAERO Inverse DRESP1 ID
STRAIN STATICS or SAERO Inverse Property ID or N/ A
ESE STATICS or SAERO Inverse Property ID or N/ A
STRESS STATICS or SAERO Inverse Property ID or N/ A
FORCE STATICS or SAERO Direct Property ID or N/ A
SPCFORCE STATICS or SAERO Inverse DRESP1 ID
CSTRAIN STATICS or SAERO Inverse Property ID or N/ A
CSTRESS STATICS or SAERO Inverse Property ID or N/ A
CFAILURE STATICS or SAERO Inverse Property ID or N/ A
CSTRAT STATICS or SAERO Inverse Property ID or N/ A
TOTSE STATICS or SAERO Inverse N/ A
GPFORCE STATICS or SAERO Inverse DRESP1 ID
GPFORCP STATICS or SAERO Inverse DRESP1 ID
FRDISP DFREQ or MFREQ Inverse DRESP1 ID
Main Index Main Index
248 CHAPTER 4
Input Data
PRES DFREQ or MFREQ Inverse DRESP1 ID
FRVELO DFREQ or MFREQ Inverse DRESP1 ID
FRACCL DFREQ or MFREQ Inverse DRESP1 ID
FRSPCF DFREQ or MFREQ Inverse DRESP1 ID
FRSTRE DFREQ or MFREQ Inverse Property ID or
Element ID
FRFORC DFREQ or MFREQ Inverse Property ID or
Element ID
PSDDISP DFREQ or MFREQ Inverse DRESP1 ID
PSDVELD DFREQ or MFREQ Inverse DRESP1 ID
PSDACCL DFREQ or MFREQ Inverse DRESP1 ID
RMSDISP DFREQ or MFREQ Inverse N/ A
RMSVELO DFREQ or MFREQ Inverse N/ A
RMSACCL DFREQ or MFREQ Inverse N/ A
TDISP MTRAN Inverse DRESP1 ID
TVELO MTRAN Inverse DRESP1 ID
TACCL MTRAN Inverse DRESP1 ID
TSPCF MTRAN Inverse DRESP1 ID
TSTRE MTRAN Inverse Property ID or
Element ID
TFORC MTRAN Inverse Property ID or
Element ID
TRIM SAERO Inverse N/ A
STABDER SAERO Inverse N/ A
FLUTTER FLUTTER Inverse DRESP1 ID
COMP** STATICS Inverse N/ A
Table 4-6 Response Type Options on the DRESP1 Entry (continued)
Response
Types
Associated
ANALYSIS Type
Approximation
Technique
Default Region
Main Index Main Index
249 CHAPTER 4
Input Data
The fact that this single entry supports so many different response types make this
entry rather complex, so that this discussion explains each of the response types in
turn.
In order to facilitate this discussion, Table 4-6 is duplicated from the Bulk Data entry
DRESP1 on page 1317 of the MSC.Nastran Quick Reference Guide. The Remarks
that are called out in this table are not given explicitly here, but the information is
provided in detail in the response type descriptions that follow.
FRMASS** ANY Direct N/ A
*Note that the approximation technique for these responses can be charged using
the ATTB field on the DRESP1 entry.
**COMP and FRMASS are only available for topology optimization.
Table 4-7 Design Sensitivity Response Attributes
Response
Type
(RTYPE)
Response Attributes
ATTA (Integer > 0) ATTB ATTI (Integer > 0)
WEIGHT Row Number
(1 < ROW < 6)
See Remark 24.
Column Number
(1 < COL < 6)
SEIDi or All or blank.
See Remark 12.
VOLUME Blank Blank SEIDi or ALL or
blank. See Remark
12.
EIGN Normal Modes
Mode Number
Approximation Code.
See Remark 19.
Blank
CEIG Complex
Eigenvalue Mode
Number
(Integer>0)
ALPHA or OMEGA
(Default=ALPHA)
Blank
Table 4-6 Response Type Options on the DRESP1 Entry (continued)
Response
Types
Associated
ANALYSIS Type
Approximation
Technique
Default Region
Main Index Main Index
250 CHAPTER 4
Input Data
FREQ Normal Modes
Mode Number
See Remark 18.
Approximation Code.
See Remark 19.
Blank
LAMA Buckling Mode
Number
Approximation Code.
See Remark 19.
Blank
DISP Displacement
Component
Blank or Mode
Number
Grid ID
STRAIN Strain Item Code Blank or Mode
Number
Property ID (PID) or
Element ID (EID)
ESE Strain Energy Item
Code
See Remark 21.
Blank or Mode
Number
Property ID (PID) or
Element ID (EID)
STRESS Stress Item Code Blank or Mode
Number
Property ID (PID) or
Element ID (EID)
FORCE Force Item Code Blank or Mode
Number
Property ID (PID) or
Element ID (EID)
SPCFORCE SPC Force
Component
Blank Grid ID
CSTRAIN Strain Item Code LAMINA Number
(Integer; Default = 1)
Property ID (PID) or
Element ID (EID)
CSTRESS Stress Item Code LAMINA Number
(Integer; Default = 1)
Property ID (PID) or
Element ID (EID)
CFAILURE Failure Criterion
Item Code
LAMINA Number
(Integer; Default = 1)
Property ID (PID) or
Element ID (EID)
CSTRAT Composite Stress
Ratio Item Code
LAMINA Number
(Integer; Default = 1)
Property ID (PID) or
Element ID (EID)
Table 4-7 Design Sensitivity Response Attributes
Response
Type
(RTYPE)
Response Attributes
ATTA (Integer > 0) ATTB ATTI (Integer > 0)
Main Index Main Index
251 CHAPTER 4
Input Data
TOTSE
(Total Strain
Energy)
Blank Blank or Mode
Number
SEIDi or All or blank.
See Remark 12.
GPFORCE Grid Point
(see Remark 25)
Blank Element ID
GPFORCP Grid Point
(see Remark 26)
Blank Grid ID connected to
ATTA grid to specify
orientation.
FRDISP Displacement
Component
Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Grid ID
PRES Acoustic Pressure
Component (= 1 or
7)
Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Grid ID
FRVELO Velocity
Component
Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Grid ID
FRACCL Acceleration
Component
Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Grid ID
FRSPCF SPC Force
Component
Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Grid ID
Table 4-7 Design Sensitivity Response Attributes
Response
Type
(RTYPE)
Response Attributes
ATTA (Integer > 0) ATTB ATTI (Integer > 0)
Main Index Main Index
252 CHAPTER 4
Input Data
FRSTRE Stress Item Code Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Property ID (PID) or
Element ID (EID)
FRFORC Force Item Code Frequency Value.
(Blank, Real > 0.0 or
Character) See
Remarks 15 and 20.
Property ID (PID) or
Element ID (EID)
PSDDISP Displacement
Component (see
Remark 27)
Frequency Value.
(Blank, Real > 0.0 or
Character). See
Remarks 15 and 20.
Grid ID
PSDVELO Velocity
Component (see
Remark 27)
Frequency Value
(Blank, Real > 0.0 or
Character). See
Remarks 15 and 20.
Grid ID
PSDACCL Acceleration
Component (see
Remark 27)
Frequency Value.
(Blank, Real > 0.0 or
Character). See
Remarks 15 and 20.
Grid ID
RMSDISP Displacement
Component
RANDPS ID Grid ID
RMSVELO Velocity
Component
RANDPS ID Grid ID
RMSACCL Acceleration
Component
RANDPS ID Grid ID
TDISP Displacement
Component
Time Value. (Blank,
Real, or Character) See
Remarks 16 and 20.
Grid ID
Table 4-7 Design Sensitivity Response Attributes
Response
Type
(RTYPE)
Response Attributes
ATTA (Integer > 0) ATTB ATTI (Integer > 0)
Main Index Main Index
253 CHAPTER 4
Input Data
WEIGHT
The weight response is computed using the same Grid Point Weight Generator utility
that is used when with the PARAM GRDPNT request is used to obtain printed results
regarding the weight. This calculation results in a 6x6 mass matrix computed about
the grid designated by the GRDPNT or about the origin of the basic coordinate system
TVELO Velocity
Component
Time Value. (Blank,
Real, or Character) See
Remarks 16 and 20.
Grid ID
TACCL Acceleration
Component
Time Value. (Blank,
Real, or Character) See
Remarks 16 and 20.
Grid ID
TSPCF SPC Force
Component
Time Value. (Blank,
Real, or Character) See
Remarks 16 and 20.
Grid ID
TSTRE Stress Item Code Time Value. (Blank,
Real, or Character) See
Remarks 16 and 20.
Property ID (PID) or
Element ID (EID)
TFORC Force Item Code Time Value. (Blank,
Real, or Character) See
Remarks 16 and 20.
Property ID (PID) or
Element ID (EID0
TRIM AESTAT or
AESURF Entry ID
Blank Blank
STABDER AESTAT or
AESURF Entry ID
Restraint Flag. (Integer
0 or 1) See Remark 13.
Component
FLUTTER Blank Blank See Remark 14.
COMP Blank (See Remark
28)
Blank Blank
FRMASS Blank (See Remark
28)
Blank Blank
Table 4-7 Design Sensitivity Response Attributes
Response
Type
(RTYPE)
Response Attributes
ATTA (Integer > 0) ATTB ATTI (Integer > 0)
Main Index Main Index
254 CHAPTER 4
Input Data
if the GRDPNT is not defined (or is defined and the grid point does not exist). The
particular weight that is to be used in the response is selected with the ATTA field
specifying which row of the 6x6 matrix field to select and ATTB field identifying the
column. The default values for ATTA and ATTB are 3, resulting in the mass of the
structure in the z direction being the default weight response. The associated
ANALYSIS Type column of Table 4-1 indicates that this response can be invoked
from any ANALYSIS type or above the subcase level before an ANALYSIS type is
specified. It is recommended that the response be invoked above the subcase level
using the DESOBJ Case Control command if it is the response is the objective and the
DESGLB Case Control command if the response is constrained. The response can be
for the entire structure or can be for an individual superelement specified in the ATT1
field. If ATT1 is blank, the total weight of the structure is used for the response. If it is
zero, the weight of the residual is used.
The following bulk data fragment shows how three DRESP1s with response type
WEIGHT can be combined to compute the x and y locations of the center of gravity of
a structure. This enables using the center of gravity value as a design constraint or the
objective.
A final comment on the WEIGHT response is to reiterate the caution on Defining the
Objective Function on page 199 that, since the WEIGHT response gives the weight
of the entire structure, it may be insensitive to design variable changes. Subtracting
out the weight of the undesigned structure using a DRESP2 may improve the
optimizer performance.
$
DRESP1 41 WT33 WEIGHT 3 3
DRESP1 42 W26 WEIGHT 2 6
DRESP1 43 W34 WEIGHT 3 4
$
$ THE CENTER OF GRAVITY (CG)
$
DRESP2 50 CG-X 200
DRESP1 41 43
DRESP2 51 CG-Y 200
DRESP1 41 43
DEQATN 200 F(R1,R2) = R2/R1
$
ENDDATA
Main Index Main Index
255 CHAPTER 4
Input Data
VOLUME
The VOLUME response provides the total volume of all of the structural elements in
the structure based on the geometry of your model. The response can be for the entire
structure or can be for an individual superelement specified in the ATT1 field. If ATT1
is blank, the total weight of the structure is used for the response. If it is zero, the
weight of the residual is used.
EIGN
The EIGN response specifies an eigenvalue from a normal modes analysis as the
response. The FREQ response is also associated with an eigenvalue, but the FREQ
response quantity is in units of Hz. while the EIGN response is applied directly to the
eigenvalue.
The ATTB field of this response can be used to select the approximation technique that
is applied when APRCOD=2 is used. By default, this approximation is performed
using the Rayleigh Quotient approximation of Eq. 2-143. If the ATTB is set to 1, the
direct method is used to approximate the eigenvalue while ATTB=2 is used to specify
the inverse variable approximation The default is felt to be adequate for all cases, with
the additional options available for experimentation.
CEIG
The CEIG response selects a component of a complex eigenvalue. The application for
this response is in areas such as brake squeal or landing gear stability where typically
there is a desire to ensure that the real part of the eigenvalue is positive. The ATTB
field is used to specify the component by specifying ALPHA (real part) or OMEGA
(imaginary part) with ALPHA being the default. If the physical constraint is that the
ALPHA value must be greater than zero, it is recommended that a DRESP2 be used to
offset the DRESP1 response by some quantity in order to avoid a poorly scaled
constraint.
FREQ
The FREQ response is similar to the EIGN response and the comments made above
for that response apply here as well. Specification of design constraints in Hz is
considered as being more intuitive in the typical application. The ATTB options of the
EIGN response are applicable here as well and the default Rayleigh Quotient
Approximation is again considered best.
r
freq
1
2
------ r
eign
=
Main Index Main Index
256 CHAPTER 4
Input Data
LAMA
The LAMA response refers to a buckling eigenvalue and indicates what multiple of
the applied load in a statics analysis would cause the structure to buckle. Typically
then, the response is constrained to be greater than 1.0. The ATTA field specifies the
buckling mode of interest and it is recommended that more than one LAMA response
be applied to preclude buckling behavior. If only a single LAMA response is
constrained, it is likely that the optimization will produce a design where a previously
non-critical mode becomes critical. The ATTB field can again be used to specify the
approximation technique. In this case, the Rayleigh Quotient does not apply. Instead,
ATTB=1 (default) selects the direct method while ATTB=2 select the inverse variable
approximation.
DISP
The DISP response identifies a displacement in a statics or static aeroelastic analysis.
The ATTA field specifies the grid component(s) while the ATTi fields specify the grids
at which the response is to be evaluated. The ATTA field can be any combination of
the digits 1-6 (e.g., 135) and each digit will result in the recovery of a displacement
(e.g., components 1,3 and 5 when ATTA is 135). The ATTB field is used for
ANALYSIS = MODES subcases to identify the mode number. ATTB is ignored for
ANALYSIS = STATICS. Any number of grids can be specified using the ATTi fields.
The grids must exist and there is no support for a THRU option.
STRAIN
The STRAIN response invokes a strain response on a finite element. In this case, the
ATTA field selects the strain item code and the meaning of the ATTi field is dependent
on what has been input in the PTYPE field of the entry. If the PTYPE is ELEM, then
the ATTi data select the element IDs for the response (MSC.Nastran requires that all
element IDs be unique across element types so that there is no ambiguity when
selecting the element IDs.) If the PTYPE selects a property type, such as PROD or
PSHELL, the ATTi field selects property IDs of the type selected with PTYPE. The
ATTB field is used for ANALYSIS = MODES subcases to identify the mode number.
ATTB is ignored for ANALYSIS = STATICS.
Item codes for stress and strain can be found in Item Codes on page 821 of the
MSC.Nastran Quick Reference Guide. An example using the CQUAD4 element type
with corner stresses is shown in Table 4-8.
Main Index Main Index
257 CHAPTER 4
Input Data
Table 4-8 Item Code Example for the QUAD4 Element with Linear Output
Element
Name
(Code)
Real Stresses or Strains Complex Stresses or Strains
Item
Code
Item
Item
Code
Item
Real/Mag. or
Imag./Phase
CQUAD4
(33)
Linear
2 Z1=Fiber distance 1 2 Z1=Fiber distance 1
3
1
Normal x at Z1
3
1
Normal x at Z1 RM
4
1
Normal y at Z1
4
1
Normal x at Z1 IP
5
1
Shear xy at Z1
5
1
Normal y at Z1 RM
6 Shear angle at Z1
6
1
Normal y at Z1 IP
7 Major principal at Z1
7
1
Shear xy at Z1 RM
8 Minor principal at Z1
8
1
Shear xy at Z1 IP
9 von Mises or
maximum shear at Z1
9 Z2=Fiber distance 2
10 Z2=Fiber distance 2
10
1
Normal x at Z2 RM
11
1
NormaI x at Z2
11
1
NormaI x at Z2 IP
12
1
Normal y at Z2
12
1
Normal y at Z2 RM
13
1
Shear xy at Z2
13
1
Normal y at Z2 IP
14 Shear angle at Z2 14 Shear xy at Z2 RM
15 Major principal at Z2 15 Shear xy at Z2 IP
16 Minor principal at Z2
17 von Mises or
maximum shear at Z2
Note: This table is applicable when a STRAIN Case Control command is absent in
case control or present with the CENTER (default) option. As an example
from the table, it is seen that item code 7 selects the major principal stress at
the center Z1 location.
Main Index Main Index
258 CHAPTER 4
Input Data
ESE
The ESE response type identifies an element strain energy response. The ATTB field
is used for ANALYSIS = MODES subcases to identify the mode number. ATTB is
ignored for ANALYSIS = STATICS. The ATTA field selects the element strain energy
item code and the ATTi field selects the applicable elements or properties in the same
manner as just described for the STRAIN response. The item codes are found in
Element Strain Energy Item Codes on page 886 of the MSC.Nastran Quick Reference
Guide. Only item codes 2 and 4 from this table are selectable on a DRESP1 entry and it
is noted from the reference that not all element types are supported with item code 4.
The sensitivity of the ESE response is not computed for design variables which affect
the shape of the model.
STRESS
The STRESS response invokes a stress response in an element. The comments for the
STRAIN response apply to the STRESS response.
FORCE
The FORCE response invokes a force response in an element. The ATTA field in this
case points to item codes contained in Element Force Item Codes on page 865 of the
MSC.Nastran Quick Reference Guide. It is noted from Table 4-6 that the direct form of
approximation is used for the response type. In general, the force in a element is a
weak function of the model properties making it difficult to satisfy constraints
imposed on this response. The ATTB field is used for ANALYSIS = MODES subcases
to identify the mode number. ATTB is ignored for ANALYSIS = STATICS.
SPCFORCE
The SPCFORCE response invokes a single point constraint response at a designated
degree-of-freedom in the model. The ATTA and ATTi field attributes follow the rules
of the DISP response.
CSTRAIN
The CSTRAIN response invokes a strain response in a finite element whose properties
have been provided on a PCOMP entry. The item codes for composite properties
differ so you must make sure you are using the correct element code when requesting
composite responses. Note that corner stresses are not available for composite
elements. Also, composite responses differ from homogeneous element responses in
that the von Mises stress/ strain is not available. The composite responses are derived
Main Index Main Index
259 CHAPTER 4
Input Data
on a layer by layer basis and the ATTB field is used to identify which layer is of
interest. This number corresponds to the value if i in the Ti input on the PCOMP. If the
ATTB field is left blank, the response will be calculated for the first layer.
It is noted that it is possible use the PCOMP entry but still request STRAIN/ STRESS
type responses for the elements by either using PTYPE = PSHELL or by identifying
the elements by their ID.
CSTRESS
The CSTRESS response type is the composite stress equivalent to the CSTRAIN
response for composite strains. The remarks given for the CSTRAIN response type
apply equally to the CSTRESS type.
CFAILURE
The CFAILURE response type allows the specification of a composite failure criterion
in the design of composite structures. The ATTB and ATTi fields for this response are
the same as the CSTRAIN or CSTRESS, but the item codes differ in that the failure item
codes are provide in Element Force Item Codes on page 865 of the MSC.Nastran
Quick Reference Guide. Only item codes 5 (direct stresses) or 7 (interlaminar shear
stress) are supported. Further, the specification of the failure criterion depends on
correctly specifying a number of other items in the bulk data file:
The failure theory to be used (e.g., Tsai-Wu) is selected on the FT field of the
PCOMP entry.
The allowable bonding shear stress is taken from the SB field on the PCOMP
entry.
Stress limits that are used in computing the failure index are taken from the
ST,SC and SS fields on the MATi entry invoked by the PCOMP.
CSTRAT
The composite strength ratio response is a recent addition to the composite output
available when using MSC.Nastran (see Strength Ratio Output for Laminated
Composites in Chapter 5 of the MSC.Nastran 2004 Release Guide). This is a direct
failure indicator and, as such, is an ideal response for the design of composites. Only
item codes 5 and 7 from the composite element force item codes are available for this
response type.
Main Index Main Index
260 CHAPTER 4
Input Data
TOTSE
The TOTSE response is a global response quantity in that the entire structure
contributes to the total strain energy response due to a static load in an
ANALYSIS = STATICS subcase or a normal mode in an ANALYSIS = MODES
subcase. The ATTB field is used to identify the mode number in a normal modes
analysis. The response can be for the entire structure or can be for an individual
superelement specified in the ATT1 field. If ATT1 is blank, the total weight of the
structure is used for the response. If it is zero, the weight of the residual is used.
GPFORCE
The GPFORCE response type allows the specification of a grid point force response in
a statics analysis. The ATTA field can be any combination of the digits 1-6 (e.g., 135)
and each digit will result in the recovery of a grid point force (e.g., components 1, 3,
and 5). Any number of elements can be specified using the ATTi fields. The PTYPE
field is used in this case to indicate which grid the grid point force is associated with.
For standard data recovery in MSC.Nastran, setting NOELOF > 0 allows for a special
type of grid point force printout that is along the edges of the elements. This form of
grid point force is not supported on the DRESP1 entry.
GPFORCP
The GPFORCP response type allows the specification of a grid point force response
that is of the form output when the PARAM NOELOP > 1 is used for standard data
recovery. In this form, the sum of the grid point forces are computed parallel to the
edges of the adjacent elements. The ATTA field identifies the grid point and the ATTi
field(s) identify the orientation of the force by specifying the grid ID(s) that is
connected to the ATTA grid.
FRDISP
The FRDISP response type identifies an displacement in a frequency response analysis
at a particular grid point and component. It is therefore similar to the DISP response
in statics except that the component specified by ATTA can now range from 1 through
12, where components 7-12 refer to the imaginary part of the response or to the phase
angle. It is no longer possible therefore to specify multiple components using the
ATTA field. As indicated in the Case Control description of Case Control Section
on page 208, the selection of whether the REAL/ IMAGINARY or
MAGNITUDE/ PHASE representation is to be used is based on the option selected on
the DISPLACEMENT Case Control command. In the absence of this command, the
Main Index Main Index
261 CHAPTER 4
Input Data
DRESP1 response is based on the default of the REAL/ IMAGINARY option. Multiple
subcases are supported for frequency response analysis and the set of excitation
frequencies, as well as design responses and constraints, can be changed between
subcases. The ATTi fields are used to identify the grid points the responses are to be
extracted from while the ATTB field has a number of options that need to be covered
in detail.
If the ATTB field is blank, the response is evaluated at all the excitation frequencies.
As indicated on the DCONSTR entry, it is possible to specify a range of frequencies
over which the constraint is applied. In this case, the constraints will only be evaluated
over a range of the frequencies. If the excitation frequencies are based on FREQ3,
FREQ4 and/ or FREQ5 Bulk Data entries, the excitation frequencies change as the
design changes so that the frequencies at which the DRESP1 responses are evaluated
will change as well.
If the ATTB field is a real number, the response is calculated at the excitation
frequency (in Hz.) that is nearest to that number.
ATTB field can also be one of six special character strings that have the following
meaning:
SUM X
i
i 1 =
n

=
AVG X
i
n
i 1 =
n

=
SSQ X
i
2
i 1 =
n

=
RSS X
i
2
i 1 =
n

=
MAX Largest value among X
i
i=1 to n ( ) =
MIN Smallest value among X
i
i=1 to n ( ) =
Main Index Main Index
262 CHAPTER 4
Input Data
The is the response at a particular frequency identified by the i subscript and the n
is the number of excitation frequencies in the analysis. These special commands
provide a convenient alternative to the DRESP2/ DEQATN formulation and are
similar to the specialized intrinsic functions discussed with the DEQATN and FUNC
attribute on the DRESP2 entry. A difference is that the use of the ATTB character string
on the DRESP1 applies the function to the particular frequency response results. The
FUNC attribute on the DRESP2 entry is more general in that it applies to all the
arguments while the special DEQATN functions are even more general because the
arguments are explicitly selected.
A special situation exists when the result from a DRESP1 that includes character string
input in the ATTB field is subsequently used in a DRESP2 entry. The DRESP1 in this
case is regarded as a DRESP2 so that the DRESP1 ID should be referenced as a DRESP2
on the subsequent DRESP2. The following bulk data fragment shows how two
DRESP1 entries that provide a RSS response at a particular grid can be summed into
a single response.
PRES
The PRES response identifies an acoustic pressure that is computed in a frequency
response analysis. These pressures are computed for grid points that are in the fluid
of the model and output as component 1 of that grid. The ATTA and ATTi fields are
similar to the other frequency response types except that the ATTA field should only
specify component 1 or 7. The DISP Case Control command selects whether the
REAL/ IMAGINARY or the MAGNITUDE/ PHASE representation of data is used
(input on the PRESSURE Case Control command is ignored). The comments
regarding the ATTB field for the FRACCL response type apply to the PRES response
type as well.
FREQ1 100 0.0 1.0 100
$ DRESP1 41O PROVIDES A SUM OF SQUARES OF THE ACCELERATION OF COMPONENT 2 OF GRID 100
$ OVER THE 101 FREQUENCY POINTS FROM 0.0 TO 100. HZ.
DRESP1 410 ACEL1 FRACCL 2 SSQ 100
$ DRESP1 420 PERFORMS THE SAME FUNCTION FOR GRID 200
DRESP1 420 ACEL2 FRACCL 2 SSQ 200
$ DRESP2 501 USES THE "DRESP2" FLAG TO SUM THE TWO RESPONSES THAT ARE ACTUALLY DRESP1'S
DRESP2 501 SUM
DRESP2 410 420
x
i
Main Index Main Index
263 CHAPTER 4
Input Data
FRVELO
The FRVELO response identifies a velocity in a frequency response analysis at a
particular grid point and component. The comments for the FRACCL response type
apply to the FRVELO response type. The VELO Case Control command selects
whether the REAL/ IMAGINARY or the MAGNITUDE/ PHASE representation of
data is used.
FRACCL
The FRACCL response identifies an acceleration in a frequency response analysis at a
particular grid point and component. The comments for the FRDISP response type
apply to the FRACCL response type. The ACCL Case Control command selects
whether the REAL/ IMAGINARY or the MAGNITUDE/ PHASE representation of
data is used.
FRFORC
The FRFORC response identifies an element force in a frequency response analysis.
The comments regarding the ATTB field for the FRACCL response type apply to the
FRFORC response type as well. The FORCE Case Control command selects whether
the REAL/ IMAGINARY or the MAGNITUDE/ PHASE representation of data is used.
The ATTA and ATTi fields are similar to that described for the FORCE response. In
selecting items codes from Element Force Item Codes on page 865 of the
MSC.Nastran Quick Reference Guide, make sure that the Complex Element Force
columns are used.
FRSTRE
The FRSTRE response identifies an element stress in a frequency response analysis.
The comments regarding the ATTB field for the FRACCL response type apply to the
FRSTRE response type as well. The STRESS Case Control command selects whether
the REAL/ IMAGINARY or the MAGNITUDE/ PHASE representation of data is used.
The ATTA and ATTi fields are similar to that described for the STRAIN response. In
selecting items codes from Element Stress-Strain Item Codes on page 823 of the
MSC.Nastran Quick Reference Guide, make sure that the Complex Element STRESS
columns are used.
Main Index Main Index
264 CHAPTER 4
Input Data
FRSPCF
The FRSPCF response identifies an single point constraint force in a frequency
response analysis. The comments for the FRACCL response type apply to the FRSPCF
response type. The SPCF Case Control command selects whether the
REAL/ IMAGINARY or the MAGNITUDE/ PHASE representation of data is used. A
comment on the FRSPCF response is that even though it appears to be a grid response,
it is not treated as such in the determination of whether adjoint sensitivity analysis is
appropriate (see Design Sensitivity Analysis on page 67). This is because the
required df/ du adjoint load vector of Eq. 2-92 is not trivial to construct for single point
constraint forces.
PSDDISP
The PSDDISP response identifies a root mean square displacement value that is
generated in a frequency response analysis. The form of this response is shown in
Eq. 2-25. The use of the ATTA and ATTi fields are the same as the FRACCL. In this
case, the REAL/ IMAGINARY form of the data is used to construct the RMS value so
the Case Control command does not play a role in selecting the data representation.
The PTYPE field identifies the ID of the RANDPS Bulk Data entry that is used to
define the input power spectrum. The use of the RANDPS enables this response to
span subcases. The PTYPE field supports the same options for specifying the
frequency(ies) of the response as those given with the FRDISP description. If a
PSDDISP response is being applied with RANDPS Bulk Data entries that include
multiple subcases, the DESSUB or DESOBJ request that invokes the RMSDISP
response must be in the first subcase used by the RANDPS entry.
PSDVELO
The PSDVELO response identifies a Power Spectral Density velocity value that is
generated in a frequency response analysis. The comments for the RMSDISP response
are applicable with the PSDVELO response as well.
PSDACCL
The PSDACCL response identifies a Power Spectral Density acceleration value that is
generated in a frequency response analysis. The comments for the PSDDISP response
are applicable with the PSDACCL response as well.
Main Index Main Index
265 CHAPTER 4
Input Data
TDISP
The TDISP response identifies a displacement in a transient response analysis at a
particular grid point and component. It is therefore similar to the DISP response in
statics, including the feature that it possible to specify multiple components using the
ATTA field. The ATTi fields are used to identify the grid points the response is to be
extracted from while the ATTB field has a number of options that need to be covered
in detail.
If the ATTB field is blank, the response is evaluated at all the excitation time steps.
If the ATTB field is a real number, the response is calculated at the excitation time step
that is nearest to that number.
ATTB field can also be one of six special character string discussed with the FRACCL
response except that now the operations are occurring over multiple time steps
instead of multiple frequencies.
TVELO
The TVELO response identifies a velocity in a transient response analysis at a
particular grid point and component. The comments for the TDISP response type
apply to the TVELO response type.
TACCL
The TACCL response identifies an acceleration in a transient response analysis at a
particular grid point and component. The comments for the TDISP response type
apply to the TACCL response type.
TSPCF
The TSPCF response identifies an single point constraint force in a transient response
analysis at a particular grid point and component. The comments for the TDISP
response type apply to the TSPCF response type.
TSTRE
The TSTRE response identifies an element stress in a transient response analysis. In
this case the ATTA field refers to a real stress item code as given in Element Stress-
Strain Item Codes on page 823 of the MSC.Nastran Quick Reference Guide. The
comments on the ATTB field provided with the TDISP description are applicable to
the TSTRE response as well.
Main Index Main Index
266 CHAPTER 4
Input Data
TFORCE
The TFORCE response identifies an element force in a transient response analysis. In
this case the ATTA field refers to a real force item code as given in Element Force
Item Codes on page 865 of the MSC.Nastran Quick Reference Guide. The comments on
the ATTB field provided with the TDISP description are applicable to the TFORCE
response as well.
TRIM
The TRIM response identifies a trim variable in a static aeroelastic response analysis.
This can be used to impose limits on the values these variables can take on during the
analysis. It is only necessary to identify the ID of the AESTAT or AESURF entry that
defines the trim variable using the ATTA field. The MSC.Nastran Users Guide for
Aeroelastic Analysis, V68 provides further information on this response in Aeroelastic
Design Sensitivity and Optimization on page 124.
STABDER
The STABDER response identifies a stability derivative in a static aeroelastic response
analysis. This can be used to impose limits on the values these variables can take on
during the analysis. In addition to the AESTAT/ AESURF entry in the ATTA field, a
single component (1 through 6) is specified using the ATT1 field. The ATTB field is
used to specify whether restrained (ATTB=1) or unrestrained (ATTB=0) form of the
derivative is to be used. The MSC.Nastran Users Guide for Aeroelastic Analysis, V68
provides further information on this response in Aeroelastic Design Sensitivity and
Optimization on page 124 and examples in Chapter 10 of that guide show how the
stability derivative for the roll rate due to the a control surface deflection can be
combined with the stability derivative for the roll rate due to roll to provide a measure
of the achievable roll rate of the vehicle.
FLUTTER
The FLUTTER response identifies a flutter damping value from an aeroelastic flutter
analysis. In this case, the ATTi fields are used to uniquely identify the damping values
of interest. In this case:
ATT1 identifies a SET1 Bulk Data entry that specifies for which modes the
responses are computed. This allows you to pinpoint the flutter roots of
interest and to avoid placing constraints on flutter eigenvalues that are
primarily due to rigid body modes and therefore not affected by changes in
the structure.
Main Index Main Index
267 CHAPTER 4
Input Data
ATT2 identifies an FLFACT entry that specifies the densities at which the
flutter responses are desired. This should be the same set or a subset of the
densities used in the flutter analysis that have been selected on the FLUTTER
Bulk Data entry.
ATT3 identifies an FLFACT entry that specifies the Mach numbers at which
the flutter responses are desired. This should be the same set or a subset of
the Mach numbers used in the flutter analysis that have been selected on the
FLUTTER Bulk Data entry.
ATT4 identifies an FLFACT entry that specifies the velocities at which the
flutter responses are desired. Note that only the PK and PKNL methods of
flutter analysis are supported for this response. This should be the same set
or a subset of the velocities used in the flutter analysis that have been selected
on the FLUTTER Bulk Data entry.
If the PKNL method has been specified on the FLUTTER Bulk Data entry, it must be
specified in the PTYPE field of the DRESP1 entry.
COMP
The COMP response permits the specification of a compliance value as a design
response. Compliance is simply the product of the displacement times the applied
load as is typically used as an objective in a topology optimization design task to
maximize structural stiffness in a static design problem. This response is only
available with topology optimization.
FRMASS
The FRMASS response indicates a target mass fraction that typically is used as a
constraint in a topology optimization task; (e.g., minimize the compliance of the
structure while limiting the mass to 40% of the mass of the original structure). This
response is only available with topology optimization.
DRESP2
Purpose:
Defines equation responses that are used in the design, either as constraints or as an
objective.
Main Index Main Index
268 CHAPTER 4
Input Data
Format:
Example:
1 2 3 4 5 6 7 8 9 10
DRESP2 ID LABEL
EQID or
FUNC
REGION METHOD
C1 C2 C3
DESVAR
DVID1 DVID2 DVID3 DVID4 DVID5 DVID6 DVID7
DVID8 -etc.-
DTABLE
LABL1 LABL2 LABL3 LABL4 LABL5 LABL6 LABL7
LABL8 -etc.-
DRESP1
NR1 NR2 NR3 NR4 NR5 NR6 NR7
NR8 -etc.-
DNODE
G1 C1 G2 C2 G3 C3
G4 C4 etc.
DVPREL1
DPIP1 DPIP2 DPIP3 DPIP4 DPIP5 DPIP6 DPIP7
DPIP8 DPIP9 -etc.-
DVCREL1
DCIC1 DCIC2 DCIC3 DCIC4 DCIC5 DCIC6 DCIC7
DCIC8 DCIC9 -etc.-
DVMREL1
DMIM1 DMIM2 DMIM3 DMIM4 DMIM5 DMIM6 DMIM7
DMIM8 DMIM9 -etc.-
DVPREL2
DPI2P1 DPI2P2 DPI2P3 DPI2P4 DPI2P5 DPI2P6 DPI2P7
DPI2P8 DPI2P9 -etc.-
DVCREL2
DCI2C1 DCI2C2 DCI2C3 DCI2C4 DCI2C5 DCI2C6 DCI2C7
DCI2C8 DCI2C9 -etc.-
DVMREL2 DMI2M1 DMI2M2 DMI2M3 DMI2M4 DMI2M5 DMI2M6 DMI2M7
DMI2M8 DMI2M9
-etc.-
DRESP2
NRR1 NRR2 NRR3 NRR4 NRR5 NRR6 NRR7
NRR8 -etc.-
DRESP2 1 LBUCK 5 3
DESVAR 101 3 4 5 1 205 209
201
DTABLE PI YM L
DRESP1 14 1 4 22 6 33 2
Main Index Main Index
269 CHAPTER 4
Input Data
DNODE
14 1 4 1 22 3
2 1 43 1
DVPREL1
101 102
DVCREL1
201 202
DVMREL1
301
DVPREL2
401 402
DVCREL2
501
DVMREL2
601 602 603
DRESP2 50 51
Field Contents
ID Unique identification number. (Integer > 0)
LABEL User-defined label. (Character)
EQID DEQATN entry identification number. (Integer > 0)
FUNC Function to be applied to the arguments. See Remark 8. (Character)
REGION Region identifier for constraint screening. See Remark 5. (Integer > 0)
METHOD When used with FUNC = BETA, METHOD = MIN indicates a
minimization task while MAX indicates a maximization task. (Default =
MIN)
When used with FUNCT = MATCH, METHOD = LS indicated a least
squares while METHOD = BETA indicated minimization of the
maximum difference. (Default = LS)
Ci Constants used when FUNC = BETA or FUNC = MATCH in
combination with METHOD = BETA. See Remark 8. (Real, Defaults: C1
= 100., C2 = .005)
DESVAR Flag indicating DESVAR entry identification numbers. (Character)
DVIDi DESVAR entry identification number. (Integer > 0)
DTABLE Flag indicating that the labels for the constants in a DTABLE entry
follow. (Character)
LABLj Label for a constant in the DTABLE entry. (Character)
DRESP1" Flag indicating DRESP1 entry identification numbers. (Character)
Main Index Main Index
270 CHAPTER 4
Input Data
Associated Entries:
The DCONSTR entry can be used to place bounds on a DRESP2 response using the
DRESP2 entry ID as a reference:
A DRESP2 response can also be used as input on another DRESP2 entry or on an
external response defined on DRESP3 entries.
NRk DRESP1 entry identification number. (Integer > 0)
DNODE Flag indicating grid point and component identification numbers.
(Character)
Gm Identification number for any grid point in the model. (Integer > 0)
Cm Component number of grid point Gm. (1 < Integer < 3)
DVPREL1 Flag indicating DVPREL1 entry identification number. (Character)
DPIPi DVPREL1 entry identification number. (Integer > 0)
DVCREL1 Flag indicating DVCREL1 entry identification number. (Character)
DCICi DVCREL1 entry identification number. (Integer > 0)
DVMREL1 Flag indicating DVPREL2 entry identification number. (Character)
DMIMi DVMREL1 entry identification number. (Integer > 0)
DVPREL2 Flag indicating DVPREL2 entry identification number. (Character)
DPI2Pi DVPREL2 entry identification number. (Integer > 0)
DVCREL2 Flag indicating DVCREL2 entry identification number. (Character)
DCI2Ci DVCREL2 entry identification number. (Integer > 0)
DVMREL2 Flag indicating DVMREL2 entry identification number. (Character)
DMI2Mi DVMREL2 entry identification number. (Integer > 0)
DRESP2 Flag indicating other DRESP2 entry identification number. (Character)
NRRk DRESP2 entry identification number. (Integer > 0)
DCONSTR
DCID RID
LALLOW UALLOW
DRESP2 ID LABEL EQID REGION
- - - -
Field Contents
Main Index Main Index
271 CHAPTER 4
Input Data
The DESOBJ Case Control command can also reference the DRESP2 entry ID:
Discussion:
It is often desirable to define responses that MSC.Nastran does not calculate directly,
e.g., local buckling criteria. Typically, the response equation is written on a DEQATN
entry. This type of response is referred to as second-level as opposed to first-level
responses which are directly available from the analysis.
The DRESP2 entry defines the input, or actual, arguments to these equations. The
arguments may be design variables, table constants, first-level structural responses,
grid coordinate locations, first and second-level properties and other second level
responses. The order of the arguments in the equation is dictated by the order in which
they appear on the DRESP2. It is therefore expected that the number of arguments on
the DEQATN will equal the number of DRESP2 arguments. The discussion of the
DEQATN on page 222 indicates that the number of DEQATN arguments may
exceed those on the DRESP2. In this case, the missing arguments are implied
frequencies.
The DRESP1 quantities called out on the DRESP2 can belong to separate
superelements or subcases. If the DRESP1 quantities are not in the same subcase, it is
necessary in Case Control to invoke the DRESP2 above the subcase level and to insert
a DRSPAN command in the individual subcases to which the DRESP1 response
belongs. DRESP2 quantities called out on a DRESP2 must belong in the same subcase.
DTABLE
LABL1 LABL2 -etc.-
DRESP2
NRR1 NRR2 -etc.-
- - - - - -
DRESP3
ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
DRESP2
NRR1 NRR2 - - -
DESOBJ
mi n
max \ .
|
| |
n =
Main Index Main Index
272 CHAPTER 4
Input Data
An alternative to referencing an equation ID is to supply one of the available FUNC
character strings:
The meanings of the first six functions is the same as that given for the same functions
described in Special DEQATN Functions on page 225. The arguments used in the
equations are all the arguments of the DRESP2 so that, for example, the AVG function
can be used to find the average of number of design variable values.
The BETA function has been provided to simplify input when the user wishes to
minimize the maximum response. The example in Acoustic Optimization in
Chapter 7 discusses this technique and provides input fragments of doing this with
and without the FUNC = BETA feature.
The DRESP2 with FUNC = BETA is intended to create the following design task
where and are user input values that have default values of 100. and 10.0.
respectively. is used to scale the objective function and is used to offset the
constraint bound from 0.
The quantity is computed so that the maximum constraint for all the response is
equal to another user input value :
Function Description
SUM Sum of the arguments
AVG Average of the arguments
SSQ Sum of the squares of the arguments
RSS Square root of the sum of the squares of the arguments
MAX The largest value of the argument list
MIN The smallest value of the argument list
BETA Minimize the maximum response
MATCH Match analysis results with user specified value
Minimize:
Subject to:
F X

( ) C
1
X

=
g
r
j
X

C
3
--------------------- 0 =
C
1
C
3
C
1
C
3

C
2
Main Index Main Index
273 CHAPTER 4
Input Data
where is the value of the maximum response for the initial design. The default
value for is 0.005, creating a maximum constraint that is just equal to the default
value of DOPTPRM parameter GMAX.
The METHOD field on DRESP2 can be used in conjunction with FUNC = BETA to
designate whether the task is to minimize the maximum response (METH = MIN) or
maximize the minimum response (METH = MAX).
The MATCH function has been provided to simplify input when the user wishes to
match analysis results with given values, such as those obtained in a test. Depending
on the value of METHOD, one of two techniques is used to create the match.
When FUNC = MATCH and METHOD = LS, a least square approach is used.
The technique converts the user input into a response of the following form:
where is a response from a DRESP1 and is a user defined target response.
Typically, this response would be designated as the objective function to be
minimized but it could also be constrained.
When FUNC = BETA and METHOD = BETA, a beta approach is applied to minimize
maximum deviation in a way similar to the FUNC = BETA method described above.
In this case, the DRESP2 results in a design task to minimize a spawned design
variable:
subject to:
Because can become small, it is necessary to offset the constraint in a fashion similar
to the BETA method given above.
g
max
r
j max
X

( ) C
3
c
2
= =
g
max
C
2
R
mat ch
r
j
r
j
T

r
j
T
-----------------
\ .
|
|
| |
2
J 1 =
m

=
r
j
r
j
T
F X

( ) C
1
X

=
X

r
j
r
j
T
( )
r
j
T
---------------------- X

j 1 2 m , , =
X

Main Index Main Index

274 CHAPTER 4
Input Data
Define:
and then determine and ,, the maximum and minimum values of .
The quantity can then be determined from user specified values of and using
the following equation:
In this case, the DRESP2 can only be invoked as the objective and cannot be
constrained.
DRESP3
Purpose:
Defines an external response using user-supplied routines. The response can be used
as a constraint or as an objective.
Entry Description:
1 2 3 4 5 6 7 8 9 10
DRESP3
ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 DVID3 DVID4 DVID5 DVID6 DVID7
DVID8 -etc.-
DTABLE
LABL1 LABL2 LABL3 LABL4 LABL5 LABL6 LABL7
LABL8 -etc.-
DRESP1
NR1 NR2 NR3 NR4 NR5 NR6 NR7
NR8 -etc.-
DNODE
G1 C1 G2 C2 G3 C3
G4 C4 etc.
DVPREL1
DPIP1 DPIP2 DPIP3 DPIP4 DPIP5 DPIP6 DPIP7
DPIP8 DPIP9 -etc.-
DVCREL1

DCIC1 DCIC2 DCIC3 DCIC4 DCIC5 DCIC6 DCIC7

r
j
r
j
r
j
T
( )
r
j
T
---------------------- =
R
2max
R
2min
r
j 2
C
2
C
3
Max R
2max
R
2mi n
, ( ) X

C
3
------------------------------------------------------------------------- C
2
=
Main Index Main Index
275 CHAPTER 4
Input Data
Example:
DCIC8 DCIC9 -etc.-
DVMREL1

DMIM1 DMIM2 DMIM3 DMIM4 DMIM5 DMIM6 DMIM7

DMIM8 DMIM9 -etc.-
DVPREL2
DPI2P1 DPI2P2 DPI2P3 DPI2P4 DPI2P5 DPI2P6 DPI2P7
DPI2P8 DPI2P9 -etc.-
DVCREL2

DCI2C1 DCI2C2 DCI2C3 DCI2C4 DCI2C5 DCI2C6 DCI2C7

DCI2C8 DCI2C9 -etc.-
DVMREL2

DMI2M1 DMI2M2 DMI2M3 DMI2M4 DMI2M5 DMI2M6 DMI2M7

DMI2M8 DMI2M9 -etc.-
DRESP2
NRR1 NRR2 NRR3 NRR4 NRR5 NRR6 NRR7
NRR8 -etc.-
USRDATA

Character String Data

-etc.-
DRESP3 1 LBUCK
TAILWNG
BUCK
DESVAR 101 3 4 5 1 205 209
201
DTABLE PI YM L
DRESP1
14 1 4 22 6 33 2
DNODE
14 1 4 1 22 3
2 1 43 1
DVPREL1
101 102
DVCREL1
201 202
DVMREL1
301
DVPREL2
401 402
DVCREL2
501
DVMREL2
601 602 603
DRESP2
50 51
USRDATA
1234 Material Constants
Main Index Main Index
276 CHAPTER 4
Input Data
Field Contents
ID Unique identification number. (Integer > 0)
LABEL User-defined label. (Character)
GROUP Group of external response that this type belongs to. (Character)
TYPE External response type. (Character)
DESVAR Flag indicating DESVAR entry identification numbers. (Character)
DVIDi DESVAR entry identification number. (Integer > 0)
DTABLE Flag indicating that the labels for the constants in a DTABLE entry
follow. (Character)
LABLj Label for a constant in the DTABLE entry. (Character)
DRESP1" Flag indicating DRESP1 entry identification numbers. (Character)
NRk DRESP1 entry identification number. (Integer > 0)
DNODE Flag signifying that the following fields are grid points.
Gm Grid point identification number. (Integer > 0)
Cm Degree-of-freedom number of grid point GM. (1 < Integer < 3)
DVPREL1 Flag indicating DVPREL1 entry identification number. (Character)
DPIPi DVPREL1 entry identification number. (Integer > 0)
DVCREL1 Flag indicating DVCREL1 entry identification number. (Character)
DCICi DVCREL1 entry identification number. (Integer > 0)
DVMREL1 Flag indicating DVPREL2 entry identification number. (Character)
DMIMi DVMREL1 entry identification number. (Integer > 0)
DVMPREL2 Flag indicating DVPREL2 entry identification number. (Character)
DPI2Pi DVPREL2 entry identification number. (Integer > 0)
DVCREL2 Flag indicating DVCREL2 entry identification number. (Character)
DCI2Ci DVCREL2 entry identification number. (Integer > 0)
DVMREL2 Flag indicating DVMREL2 entry identification number. (Character)
DMI2Mi DVMREL2 entry identification number. (Integer > 0)
DRESP2 Flag indicating other DRESP2 entry identification number. (Character)
Main Index Main Index
277 CHAPTER 4
Input Data
Associated Entries:
The DCONSTR entry can be used to place bounds on a DRESP3 response using the
DRESP3 entry ID as a reference:
The DESOBJ Case Control command can also reference the DRESP3 entry ID.
Discussion:
As discussed in Design Responses on page 53, the DRESP3 entry enables the
consideration of responses that are not direct MSC.Nastran quantities and that cannot
be formulated from MSC.Nastran quantities through the use of the DRESP2 entry.
Instead, the DRESP3 invokes an API (application programming interface) to compute
the response using a combination of MSC.Nastran supplied data and user supplied
data and algorithms. It is seen that the input format for the DRESP3 is very similar to
the DRESP2 in that it permits a variety of types of information to be used as
arguments. The DRESP3 does not use the DEQATN, so there is no EQID identified.
Instead, the GROUP name refers to an FMS CONNECT entry (See File
Management on page 205) that defines the API evaluator while the TYPE argument
identifies the particular response type that is to be used to evaluate the response.
Another difference from the DRESP2 is the provision for another type of data
designed USRDATA. This character string data provides a convenient way of
communicating to the API and can include, for example, a file name that needs to be
opened or some special instructions. The server program parses this character string
and takes the appropriate action. As with the DRESP2, it is necessary to use the
DRSPAN Case Control command when the DRESP1 quantities come from more than
one subcase.
NRRk DRESP2 entry identification number. (Integer > 0)
USRDATA Flag indicating user input data. (Character)
DCONSTR
DCID RID
LALLOW UALLOW
Field Contents
DESOBJ
mi n
max \ .
|
| |
n =
Main Index Main Index


278
DSCREEN
Purpose:
Defines override information necessary to screen the constraints for temporary
deletion.
Entry Description:
Associated Entries:
The regions for constraint screening are automatically established by default. These
defaults can be overridden though and new ones can be established by defining new
regions on the DRESP1, DRESP2 or DRESP3 entries.
DSCREEN
RTYPE TRS NSTR
Field Contents
RTYPE Response type for which the screening criteria apply. (Character)
TRS Truncation threshold. (Real; Default = -0.5)
NSTR Maximum number of constraints to be retained per region per load
case. (Integer > 0; Default = 20)
DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1
ATT2 -etc.-
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
DTABLE
LABL1 LABL2 -etc.-
DRESP3 ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
-
- - -
DRESP1
NR1 NR2 -etc.-
DRESP2
NRR1 NRR2
Main Index Main Index
279 CHAPTER 4
Input Data
Discussion:
To reduce the costs associated with the sensitivity analysis and to reduce the size of
the optimization problem, many of the noncritical constraints are dynamically deleted
during the optimization process. Default values of parameters that control the
screening process can be overridden using DSCREEN entries. Since these defaults are
often satisfactory, including DSCREEN entries is often unnecessary. One application
where it has utility is when you are performing a sensitivity analysis and want to
make sure that the sensitivities of all the responses of interest are provided. The TRS
value can then be set to a large negative number (say -1000.) and the NSTR value can
be set to a number equal to or greater than the number of responses.
The RTYPE refers to the RTYPE used on the DRESP1. For the DRESP2s, the applicable
RTYPE is EQUA. For the DRESP3s, the applicable RTYPE for constraint screening is
DRESP3. Constraint screening is performed in two stages. In the first step, deletion,
the constraint values that fall below a certain threshold level are deleted from the
constraint set. The default threshold level is -0.5, but this can be overridden by
specifying a TRS value on the DSCREEN entry. In MSC.Nastran, the normalized
constraints are considered violated if they are positive quantities. The default TRS
states that only those constraints that are within 50% of their critical values are
retained.
The second step, regionalization, sorts the remaining constraints and retains up to a
prescribed maximum per region per load case. The default region specifications are
given in Table 4-6 as a function of Response type. NSTR, or the number of constraints
to be retained per region, has a default value of 20, which again can be overridden. The
NSTR value can not be used for response types WEIGHT, VOLUME, EIGN, and
LAMA.
DTABLE
Purpose:
Defines a table of constants to be used in conjunction with DEQATN equations.
Entry Description:
DTABLE LABL1 VALU1 LABL2 VALU2 LABL3 VALU3 LABL4 VALU4
LABL5 VALU5 LABL6 VALU6 -etc.-
Main Index Main Index


280
Associated Entries:
DRESP2, DRESP3 and DVPREL2, DVCREL2, DVMREL2 entries list the input
arguments to equations that define synthetic response and property relations,
respectively. These input arguments may include table constants defined on the
DTABLE entry.
Field Contents
LABLi Label for the constant. (Character)
VALUi Value of the constant. (Real)
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
DTABLE
LABL1 LABL2 -etc.-
DRESP1"
NR1 NR2 -etc.-
- - - - - -
DRESP3 ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
DTABLE
LABL1 LABL2 -etc.-
- - - -
DVPREL2
ID TYPE PID FID PMIN PMAX EQID
DESVAR
DVID1 DVID2 DVID3 -etc.-
DTABLE
LABL1 LABL2 LABL3 -etc.-
DVCREL2
ID TYPE EID
CPNAME
CPMIN CPMAX EQID
DESVAR
DVID1 DVID2 DVID3 -etc.-
DTABLE
LABL1 LABL2 LABL3 -etc.-
DVMREL2
ID TYPE MID
MINAME
MPMIN MPMAX EQID
DESVAR
DVID1 DVID2 DVID3 -etc.-
DTABLE
LABL1 LABL2 LABL3 -etc.-
Main Index Main Index
281 CHAPTER 4
Input Data
Discussion:
Constants used in equations can either be built into the DEQATN entry when the
equation is defined or passed in as arguments. Building-in the values of constants may
be inconvenient and can prevent one equation from easily being used in different
contexts.
The DTABLE entry allows these constants to be stored in a table and then used in the
equations as necessary. On the DTABLE entry, a constant is given a name and a value.
When the DTABLE argument list is defined on a DVPREL2 or other entry, the
constant is referenced by its name.
Only one DTABLE entry may appear in the Bulk Data.
DVBSHAP
Purpose:
Defines a shape basis vector as a linear combination of analytic boundary shapes
solutions and assigns a design variable to the result.
Entry Description:
Associated Entries:
Design variables must first be defined using DESVAR entries.
DVBSHAP
DVID
AUXMOD
COL1 SF1 COL2 SF2 COL3 SF3
Field Contents
DVID Design variable identification number of a DESVAR entry.
(Integer > 0)
AUXMOD Auxiliary model identification number. (Integer > 0)
Coli Load sequence identification number from AUXMODEL Case
Control command. (Integer > 0)
SFi Scaling factor for load sequence identification number. (Real;
Default = 1.0)
DESVAR
ID LABEL XINIT XLB XUB DELXV
Main Index Main Index


282
Discussion:
In shape optimization, shape basis vectors relate the changes in a design variable to
changes in grid locations. Analytical Boundary Shapes are calculated in an auxiliary
model as described in Relating Design Variables to Shape Changes on page 166.
The DVBSHAP entry is then used to identify the auxiliary model and then apply scale
factors to a series of solution vectors. If multiple DVBSHAP entries share the same
design variable ID, the shapes are added.
DVCREL1
Purpose:
Defines the relation between a connectivity property and design variables.
Entry Description:
DVCREL1
ID TYPE EID
CFNAME
CPMIN CPMAX C0
DVID1 COEF1/
PVAL
DVID2 COEF2 DVID3 COEF3 -etc.-
Field Contents
ID Unique identification number. (Integer > 0)
TYPE Name of an element connectivity entry, such as CBAR,
CQUAD, etc. (Character)
EID Element Identification number. (Integer > 0)
CPNAME Name of connectivity property, such as X1, X2, X3,
ZOFFS, etc. (Character)
CPMIN Minimum value allowed for this property. If CPNAME references
a connectivity property that can only be positive, then the default
value of CPMIN is 1.0E-20. Otherwise, it is -1.0E35. (Real)
CPMAX Maximum value allowed for this property. (Real; Default =
1.0E+20)
C0 Constant term of relation. (Real; Default = 0.0)
DVIDi DESVAR entry identification number. (Integer > 0)
Main Index Main Index
283 CHAPTER 4
Input Data
Associated Entries:
Design variables are referenced on DVCREL1 entries by their DESVAR-defined IDs.
The DVCREL1 can be included in DRESP2 or DRESP3 entries:
Discussion:
A connectivity property is one that has been defined on a Bulk Data entry that
begins with the letter C. The design of these properties enables you to design things
such as beam offsets and concentrated masses. A connectivity property can be
expressed as a linear combination of design variables as
COEFi Coefficient of linear relation. (Real)
PVAL Flag to indicate COEF2 is to be set to the current connectivity
property. (Character = PVAL)
DESVAR ID LABEL XINIT XLB XUB DELXV
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - - - -
DVCREL1

DCIC1 DCIC2 -etc.-

DRESP3 ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
- - - -
- - - - - -
DVCREL1

DCIC1 DCIC2 -etc.-

p
1
c
0
= c
1
x
1
c
2
x
2
c
n
x
n
+ + + +
Main Index Main Index


284
Both independent as well as dependent design variables can appear in this relation
(see the DLINK entry). This form is especially useful in the sense that it can be used to
express not only a simple one-to-one correspondence between a design variable and
a property, but more complex reduced basis formulations as well.
The CPNAME is a character string that identifies the required property value by using
the same name as that given in the MSC.Nast ran Quick Reference Guide. Table 2-3 of
this Users Guide identifies the properties that are available for design.
It is seen that limits can be placed on the property by specifying CPMIN and CPMAX
values. This is in addition to the limits that are provided by the XLB/ XUB fields on the
DESVAR entries. If possible, the limits on the DESVAR entries should be more
restrictive than those on the DVCREL1 entries. This is because the optimizer will
never allow a design variable limit to be violated, but it may allow a property limit to
be violated if this seems to help satisfy some other violated constraint. Violating a
connectivity constraint could have disastrous results if it produces an element that is
physically meaningless.
It is seen that there is a certain ambiguity in the computation of the connectivity
property value in that it is the product of the user defined coefficients and the design
variable value. A guideline is that the coefficients should be specified in a way that
allows the design variable values to be near unity. The optimizer performs best when
all the design variables are in the same range and, in particular, when they are all on
the order of 1.0.
If the term PVAL is entered in the COEF1 field, this is a flag that the COEF1 value is
to be obtained from the corresponding property on the connectivity entry. This
simplifies the process of completing this entry when the desire is to have a one-to-one
correspondence between the designed variable and the designed property. The
assumption is that the DESVAR entry identified by DVID1 is 1.0.
DVCREL2
Purpose:
Defines a connectivity property by reference to an equation defined on a DEQATN
entry.
Main Index Main Index
285 CHAPTER 4
Input Data
Entry Description:
Associated Entries:
To define a synthetic property relation, the DVCREL2 entry identifies a DEQATN
entry and declares the design variable (DESVAR) and any necessary table constant
(DTABLE) arguments.
DVCREL2
ID TYPE EID
CPNAME
CPMIN CPMAX EQID
DESVAR
DVID1 DVID2 -etc.-
DTABLE
LABl1 LABL2 -etc.-
Field Contents
ID Unique identification number. (Integer > 0)
TYPE Name of an element connectivity entry, such as CBAR,
CBEAM, etc. (Character)
EID Element Identification number. (Integer > 0)
CPNAME Connectivity property name, such as Z1, ZOFFS, etc.
(Character)
CPMIN Minimum value allowed for this property. If CPNAME references
a connectivity property that can only be positive, then the default
value of CPMIN is 1.0E-20. Otherwise, it is -1.0E35. (Real)
CPMAX Maximum value allowed for this property. (Real; Default = 1.0E20)
EQID DEQATN entry identification number. (Integer > 0)
DESVAR DESVAR flag. Indicates that the IDs of DESVAR entries follow.
(Character)
DVIDi DESVAR entry identification number. (Integer > 0)
DTABLE DTABLE flag. Indicates that the IDs for the constants in a DTABLE
entry follow. This field may be omitted if there are no constants
involved in this relation. (Character)
LABi Label for constant on the DTABLE entry. (Integer > 0)
DEQATN
EQID EQUATION
EQUATION (Cont.)
Main Index Main Index


286
The DVCREL2 entries can be included in DRESP2 or DRESP3 entries.
Discussion:
For some applications, the linear design variable-to-property relations provided by
the DVCREL1 entry may not be sufficient. In these cases, equations may be defined to
relate design variables to properties in a nonlinear fashion. The DVCREL2 entry
provides the arguments to an equation defined on a DEQATN entry. These arguments
may be design variables defined on DESVAR entries and table constants defined on a
DTABLE entry.
Every property defined on a DVCREL2 entry becomes an independently varying
property internally in MSC.Nastran. Design sensitivities are first computed with
respect to these independently varying properties, and then the chain rule applied to
relate these sensitivities to changes in the design variables. Care should be taken to not
use the DVCREL2 entry when the relationship between the design variable and the
DESVAR ID LABEL XINIT XLB XUB DELXV
DTABLE LABL1 VALU1 LABL2 VALU2 LABL3 VALU3 LABL4 VALU4
LABL5 VALU5 LABL6 VALU6 -etc.-
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - - - -
DVCREL2

DCIC1 DCIC2 -etc.-

DRESP3
ID LABEL
GROUP
TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
- - - -
- - - - - -
DVCREL2

DCIC1 DCIC2 -etc.-

Main Index Main Index
287 CHAPTER 4
Input Data
connectivity property is linear; the DVCREL1 entry should be used in this case. This
is because the DVCREL2 entry requires the added design sensitivity calculations and
could therefore impact the performance.
DVGRID
Purpose:
Defines design variable-to-grid coordinate relations for shape sensitivity and
optimization.
Entry Description:
Associated Entries:
The DESVAR entry defines a design variable that can be used to describe grid
variations as well as other design relations. The GRID entry (not shown) provides the
baseline locations for the designed grid.
Discussion:
Changes in grid coordinate values are expressed as functions of design variables
according to the relation
DVGRID DVID GID CID COEFF N1 N2 N3
Field Contents
DVID DESVAR entry identification number. (Integer > 0)
GID Grid point identification number. (Integer > 0)
CID Coordinate system identification number. (Integer > 0; Default = 0)
COEFF Multiplier of the vector defined by Ni. (Real; Default = 0.0)
Ni Components of the vector measured in the coordinate system
defined by CID. (Real; at least one )
DESVAR ID LABEL XINIT XLB XUB DELXV
Ni 0.0
Main Index Main Index


288
where changes in the i-th grid point are expressed in terms of changes in the j-th
design variable. If the number of design variables is less than the number of grid
points, this relation is called a reduced basis formulation. Reduced basis formulations
are discussed in Fundamentals of Design Sensitivity and Optimization in
MSC.Nastran on page 33.
The DVGRID entry defines the particular grid point, design variable, and components
of each vector. Multiple references to the same grid point and design variable
pair result in vectorial addition of the corresponding .
Other shape basis vector definition methods utilize DVBSHAP and DVSHAP Bulk
Data entries.
DVMREL1
Purpose:
Defines the relation between a material property and design variables.
Entry Description:
DVMREL1
ID TYPE MID MFNAME MPMIN
CMPMAX
C0
DVID1 COEF1/
'PVAL
DVID2 COEF2 DVID3 COEF3 -etc.-
Field Contents
ID Unique identification number. (Integer > 0)
TYPE Name of a material property entry, such as MAT1, MAT2, etc.
(Character)
MID Material Identification number. (Integer > 0)
MPNAME Name of material property, such as E, RHO, etc. (Character)
MPMIN Minimum value allowed for this property. If MPNAME references
a material property that can only be positive, then the default
value of MPMIN is 1.0E-20. Otherwise, it is -1.0E35. (Real)
G
i
{ } T { }
i j
j

= x
j

T { }
ij
T { }
ij
Main Index Main Index
289 CHAPTER 4
Input Data
Associated Entries:
Design variables are referenced on DVMREL1 entries by their DESVAR-defined IDs.
The DVMREL1 can be included in DRESP2 or DRESP3 entries:
Discussion:
A material property is a real number that has been defined on a MATi Bulk Data entry.
The design of these properties allows you to vary such key parameters as Youngs
Modulus and material density. A material property can be expressed as a linear
combination of design variables as
MPMAX Maximum value allowed for this property. (Real; Default =
1.0E+20)
C0 Constant term of relation. (Real; Default = 0.0)
DVIDi DESVAR entry identification number. (Integer > 0)
COEFi Coefficient of linear relation. (Real)
PVAL Flag to indicate COEF1 is to be set to the current material property.
(Character = PVAL)
DESVAR ID LABEL XINIT XLB XUB DELXV
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - - - -
DVMREL1

DMIM1 DMIM2 -etc.-

DRESP3 ID LABEL
GROUP
TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - - - -
DVMREL1

DMIM1 DMIM2 -etc.-

Main Index Main Index


290
Both independent as well as dependent design variables can appear in this relation
(see the DLINK entry). This form can be used to express not only a simple one-to-one
correspondence between a design variable and a property, but more complex reduced
basis formulations as well.
The MPNAME is a character string that identifies the required property value by
using the same name as that given in the MSC.Nast ran Quick Reference Guide.
Table 2-2 of this Users Guide identifies the properties that are available for design.
It is seen that limits can be placed on the property by specifying MPMIN and MPMAX
values. This is in addition to the limits that are provided by the XLB/ XUB fields on the
DESVAR entries. If possible, the limits on the DESVAR entries should be more
restrictive than those on the DVPREL1 entries. This is because the optimizer will never
allow a design variable limit to be violated, but it may allow a property limit to be
violated if this seems to help satisfy some other violated constraint. Violating a
material property constraint could have disastrous results if it produces a material
that is physically meaningless.
It is seen that there is a certain ambiguity in the computation of the material property
value in that it is the product of the user defined coefficients and the design variable
value. A guideline is that the coefficients should be specified in a way that allows the
design variable values to be near unity. The optimizer performs best when all the
design variables are in the same range and, in particular, when they are all on the
order of 1.0.
If the term PVAL is entered in the COEF1 field, this is a flag that the COEF1 value is
to be obtained from the corresponding property on the material entry. This simplifies
the process of completing this entry when the desire is to have a one-to-one
correspondence between the designed variable and the designed property. The
assumption is that the DESVAR entry identified by DVID1 is 1.0.
DVMREL2
Purpose:
Defines a connectivity property by reference to an equation defined on a DEQATN
entry.
p
1
c
0
= c
1
x
1
c
2
x
2
c
n
x
n
+ + + +
Main Index Main Index
291 CHAPTER 4
Input Data
Entry Description:
Associated Entries:
To define a synthetic property relation, the DVMREL2 entry identifies a DEQATN
entry and declares the design variable (DESVAR) and any necessary table constant
(DTABLE) arguments.
DVMREL2
ID TYPE EID
MFNAME
MPMIN
MPMAX
EQID
DESVAR
DVID1 DVID2 -etc.-
DTABLE
LABL1 LABL2 -etc.-
Field Contents
ID Unique identification number. (Integer > 0)
TYPE Name of a material property entry, such as MAT1, MAT2, etc.
(Character)
MID Material Identification number. (Integer > 0)
MPNAME Name of material property, such as E, RHO, etc. (Character)
MPMIN Minimum value allowed for this property. If MPNAME references
a material property that can only be positive, then the default value
of MPMIN is 1.0E-20. Otherwise, it is -1.0E35. (Real)
MPMAX Maximum value allowed for this property. (Real; Default =
1.0E+20)
EQID DEQATN entry identification number. (Integer > 0)
DESVAR DESVAR flag. Indicates that the IDs of DESVAR entries follow.
(Character)
DVIDi DESVAR entry identification number. (Integer > 0)
DTABLE DTABLE flag. Indicates that the IDs for the constants in a DTABLE
entry follow. This field may be omitted if there are no constants
involved in this relation. (Character)
LABi Label for a constant on the DTABLE entry. (Integer > 0)
DEQATN
EQID EQUATION
EQUATION (Cont.)
Main Index Main Index


292
The DVMREL2 entries can be included in DRESP2 or DRESP3 entries.
Discussion:
For some applications, the linear design variable-to-property relations provided by
the DVMREL1 entry may not be sufficient. In these cases, equations may be defined
to relate design variables to material properties in a nonlinear fashion. The DVMREL2
entry provides the arguments to an equation defined on a DEQATN entry. These
arguments may be design variables defined on DESVAR entries and table constants
defined on a DTABLE entry.
Every property defined on a DVMREL2 entry becomes an independently varying
property internally in MSC.Nastran. Design sensitivities are first computed with
respect to these independently varying properties, and then the chain rule applied to
relate these sensitivities to changes in the design variables. Care should be taken to not
use the DVMREL2 entry when the relationship between the design variable and the
DESVAR ID LABEL XINIT XLB XUB DELXV
DTABLE LABL1 VALU1 LABL2 VALU2 LABL3 VALU3 LABL4 VALU4
LABL5 VALU5 LABL6 VALU6 -etc.-
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - - - -
DVMREL2

DMI2M1 DMI2M2 -etc.-

DRESP3 ID LABEL
GROUP
TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
- - - - - -
DVMREL2

DMI2M1 DMI2M2 -etc.-

Main Index Main Index
293 CHAPTER 4
Input Data
material property is linear; the DVMREL1 entry should be used in this case. This is
because the DVMREL2 entry requires the added design sensitivity calculations and
could therefore impact the performance.
DVPREL1
Purpose:
Defines the relation between an analysis model property and design variables.
Format:
Example:
1 2 3 4 5 6 7 8 9 10
DVPREL1
ID TYPE PID
PNAME/
FID
PMIN PMAX C0
DVID1 COEF1/
PVAL
DVID2 COEF2 DVID3 -etc.-
DVPREL1
12 PBAR 612 6 0.2 3.0
4 0.25 20 20.0 5 0.3
Field Contents
ID Unique identification number. (Integer > 0)
TYPE Name of a property entry, such as PBAR, PBEAM, etc.
(Character)
PID Property entry identification number. (Integer > 0)
PNAME/ FlD Property name, such as T, A, or field position of the property
entry, or word position in the element property table of the analysis
model. (Character or Integer 0)
PMIN Minimum value allowed for this property. If FID references a stress
recovery location, then the default value for PMIN is -1.0 + 35. PMIN
must be explicitly set to a negative number for properties that may be
less than zero (for example, field ZO on the PCOMP entry). (Real;
Default = 1.0E-20)
PMAX Maximum value allowed for this property. (Real; Default = 1.0E+20)

Main Index Main Index

294
Associated Entries:
Design variables are referenced on DVPREL1 entries by their DESVAR-defined IDs.
The DVPREL1 IDs can be included in DRESP2 or DRESP3 entries:
Discussion:
An analysis model property can be expressed as a linear combination of design
variables as
C0 Constant term of relation. (Real; Default = 0.0)
DVIDi DESVAR entry identification number. (Integer > 0)
COEFi Coefficient of linear relation. (Real)
PVAL Flag to indicate COEF1 is to be set to the current property value
(Character = PVAL)
DESVAR ID LABEL XINIT XLB XUB DELVXV
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - - - -
DVPREL1

DPIP1 DPIP2 -etc.-

DRESP3 ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
- - - - - -
DVPREL1

DPIP1 DPIP2 -etc.-

Field Contents
p
1
c
0
= c
1
x
1
c
2
x
2
c
n
x
n
+ + + +
Main Index Main Index
295 CHAPTER 4
Input Data
Both independent as well as dependent design variables can appear in this relation
(see the DLINK entry). This form can be used to express not only a simple one-to-one
correspondence between a design variable and a property, but more complex reduced
basis formulations as well.
The PNAME is a character string that identifies the required property value by using
the same name as that given in the MSC.Nast ran Quick Reference Guide. Table 2-1 of
this Users Guide identifies the properties that are available for design.
Limits can be placed on the property by specifying PMIN and PMAX values. This is
in addition to the limits that are provided by the XLB/ XUB fields on the DESVAR
entries. If possible, the limits on the DESVAR entries should be more restrictive than
those on the DVPREL1 entries. This is because the optimizer will never allow a design
variable limit to be violated, but it may allow a property limit to be violated if this
seems to help satisfy some other violated constraint. Violating a material property
constraint could have disastrous results if it produces a material that is physically
meaningless.
Note that there is a certain ambiguity in the computation of the property value in that
it is the product of the user defined coefficients and the design variable value. A
guideline is that the coefficients should be specified in a way that allows the design
variable values to be near unity. The optimizer performs best when all the design
variables are in the same range and, in particular, when they are all on the order of 1.0.
If the term PVAL is entered in the COEF1 field, this is a flag that the COEF1 value is
to be obtained from the corresponding property entry. This simplifies the process of
completing this entry when the desire is to have a one-to-one correspondence between
the designed variable and the designed property. The assumption is that the DESVAR
entry identified by DVID1 is 1.0.
Guidelines for the PCOMP Entry
Table 2-1 indicates that all the real values on the PCOMP entry can be designed. Two
of these merit further discussion
The ZO property indicates the offset from the reference plane to the bottom surface. If
this is left blank, the value defaults to half the total thickness. It is not necessary to
design the ZO quantity if the user input value is blank and the half thickness value is
acceptable since the ZO value adjusts to half the total thickness in this case.
The ORIENTi field is a difficult property to design since the composite materials vary
as the fourth power of the quantity. Suggested guidelines are to:
Main Index Main Index


296
Limit the allowable movement of the designed orientation to no more than
+20
o
.
Avoid design variable values of 0.0 because it is difficult to move away from
this value. It is better to apply an offset such as so that an of 1.0
gives a zero orientation angle.
DVPREL2
Purpose:
Defines a structural property by reference to an equation defined on a DEQATN entry.
Entry Description:
DVPREL2
ID TYPE PID
FNAME
PMIN PMAX EQID
DESVAR
DVID1 DVID2 -etc.-
DTABLE
LABL1 LABL2 -etc.-
Field Contents
ID Unique identification number. (Integer > 0)
TYPE Name of a property entry, such as PBAR, PBEAM, etc. (Character)
PID Property entry identification number. (Integer > 0)
PNAME/ FID Property name, such as T, A, or field position of the property
entry, or word position in the element property table of the analysis
model. (Character or Integer 0)
PMIN Minimum value allowed for this property. If FID references a stress
recovery location field, then the default value for PMIN is -1.0+35.
PMIN must be explicitly set to a negative number for properties that
may be less than zero (for example, field ZO on the PCOMP entry).
(Real; Default = 1.E-20)
PMAX Maximum value allowed for this property. (Real; Default = 1.0E20)
EQID DEQATN entry identification number. (Integer > 0)
DESVAR DESVAR flag. Indicates that the IDs of DESVAR entries follow.
(Character)
DVIDi DESVAR entry identification number. (Integer > 0)

i
1 X
i
+ = X
i

Main Index Main Index

297 CHAPTER 4
Input Data
Associated Entries:
To define a synthetic property relation, the DVPREL2 entry identifies a DEQATN
entry and declares the design variable (DESVAR) and any necessary table constant
(DTABLE) arguments.
The DVPREL2 entries can be included in DRESP2 or DRESP3 entries.
DTABLE DTABLE flag. Indicates that the LABLs for the constants in a DTABLE
entry follow. This field may be omitted if there are no constants
involved in this relation. (Character)
LABLi Label for a constant on the DTABLE entry. (Integer > 0)
DEQATN
EQID EQUATION
EQUATION (Cont.)
DESVAR ID LABEL XINIT XLB XUB DELXV
DTABLE LABL1 VALU1 LABL2 VALU2 LABL3 VALU3 LABL4 VALU4
LABL5 VALU5 LABL6 VALU6 -etc.-
DRESP2 ID LABEL EQID REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
DRESP1
NR1 NR2 -etc.-
- - - - - -
DVPREL2
DP2P1 DP2P2 -etc.-
DRESP3 ID LABEL GROUP TYPE REGION
DESVAR
DVID1 DVID2 -etc.-
- - - -
- - - - - -
DVPREL2
DP2P1 DP2P2 -etc.-
Field Contents
Main Index Main Index


298
Discussion:
For some applications, the linear design variable-to-property relations provided by
the DVPREL1 entry may not be sufficient. In these cases, equations may be defined to
relate design variables to properties in a nonlinear fashion. The DVPREL2 entry
provides the arguments to an equation defined on a DEQATN entry. These arguments
may be design variables defined on DESVAR entries and table constants defined on a
DTABLE entry.
Every property defined on a DVPREL2 entry becomes an independently varying
property internally in MSC.Nastran. Design sensitivities are first computed with
respect to these independently varying properties, and then the chain rule applied to
relate these sensitivities to changes in the design variables. Care should be taken to not
use the DVPREL2 entry when the relationship between the design variable and the
property is linear; the DVPREL1 entry should be used in this case. This is because the
DVPREL2 entry requires the added design sensitivity calculations and could therefore
impact the performance.
DVSHAP
Purpose:
Defines a shape basis vector by relating a design variable identification number
(DVID) to columns of a displacement matrix.
Format:
1 2 3 4 5 6 7 8 9 10
DVSHAP
DVID COL1 SF1 COL2 SF2 COL3 SF3
Field Contents
DVID Design variable identification number on the DESVAR entry.
(Integer > 0)
COLi Column number of the displacement matrix. See Remark 2.
(1 < Integer < maximum column number in the displacement matrix.)
SFi Scaling factor applied to the COLi-th column of the displacement matrix.
(Real; Default = 1.0)
Main Index Main Index
299 CHAPTER 4
Input Data
Associated Entries:
The design variables must first be defined using DESVAR entries.
The displacement vector are DBLOCATEd using File Management Statements as
shown in File Management on page 205.
Discussion:
In shape optimization, shape basis vectors relate the changes in a design variable to
changes in grid locations. (See basis vectors in shape optimization in Relating
Design Variables to Shape Changes on page 166.)
The DVSHAP entry is applied to solution vectors that are output in a separate
analysis. The DVSHAP entry is used to apply scale factors to a series of solution
vectors. If multiple DVSHAP entries share the same design variable ID, the shapes are
added.
MODTRAK
Purpose:
Specifies parameters for mode tracking in design optimization (SOL 200).
Format:
DESVAR
ID LABEL XINIT XLB XUB DELXV
1 2 3 4 5 6 7 8 9 10
MODTRAK
SID
LOWRNG HIGHRNG MTFILTER
Field Contents
SID Sets identification number that is selected in the Case Control Section
with the MODTRAK command. (Integer; No Default)
LOWRNG Lowest mode number in range to search. (Integer > 0, Default = 0. If
nonzero, LOWRNG < HIGHRNG.)
Main Index Main Index


300
Associated Entries:
The MODTRAK entry is invoked by the Case Control command:
MODTRAK = n
Discussion:
The concept of mode tracking is presented in Multidisciplinary Analysis on
page 58. Mode tracking is available only in normal modes analysis (ANALYSIS =
MODES). The MODTRAK Bulk Data entry specifies a range of modes (by sequence
number) that are to be tracked and a threshold value MTFILTER that is used to
designate whether modes correlate. The default value of 0.9 is applied to normalized
results of the triple matrix product of Eq. 2-29 and is usually adequate.
TOPVAR Topological Design Variables
Purpose:
Defines a topology design region for topology optimization.
Format:
HIGHRNG Highest mode number in range to search. (Integer > 0,
Default =number of eigenvalues extracted. If nonzero,
LOWRNG< HIGHRNG.)
MTFILTER Filtering parameter used in mode cross-orthogonality check. (Real,
Default = 0.9)
1 2 3 4 5 6 7 8 9 10
TOPVAR ID LABEL PTYPE XINIT XLB DELXV POWER PID
SYM CID MSi MSi MSi
CAST CID DDi DIE
EXT CID EDi
TDMIN TV
Field Contents
Main Index Main Index
301 CHAPTER 4
Input Data
Example:
1 2 3 4 5 6 7 8 9 10
TOPVAR 2 PS1 PSOLID 0.3 10
SYM 5 XY ZX
CAST 5 X 2
TDMIN 0.6
Field Contents
ID Unique topology design region identification number. (Integer > 0)
LABEL User-supplied name for printing purpose. (Character)
PTYPE Property entry name. Used with PID to identify the elements to be designed.
(Character: PBAR, PSHELL, PSOLID, etc.)
XINIT Initial value. (Real, XLB < XINIT). Typically, XINIT is defined to match the
mass target constraint, so the initial design does not have violated constraints.
For example, if the mass target is 30% on DRESP1=FRMASS, then it is
recommended XINIT=0.3.
XLB Lower bound to prevent the singularity of the stiffness matrix. (Real; Default =
0.001)
DELXV Fractional change allowed for the design variable during approximate
optimization. (Real > 0.0; Default = 0.2. See Remark 3).
POWER A penalty factor used in the relation between topology design variables and
element Youngs modulus. (Real > 1.0; Default = 3.0). 2.0 < POWER < 5.0 is
recommended.
PID Property entry identifier (Integer > 0)
SYM Indicates that this line defines symmetry constraints.
CID Rectangular coordinate system ID used for specifying manufacturing
constraints. (Blank or Integer > 0; Default = 0 = the basic coordinate system)
MSi Mirror symmetry plane. (Character, XY, YZ, or ZX)
CAST Indicates that this line defines casting constraints (i.e., die draw direction
constraints). See Remarks 6 and 7.
Main Index Main Index


302
Discussion:
Topology optimization is a special version of optimization that seeks to find the
optimal distribution of material based on some criterion for given space, loads, and
boundary conditions. The TOPVAR entry identifies the finite elements that are to
participate in the design task. Every element that is a member of the PTYPE/ PID pair
given of the TOPVAR has its own design variable that drives the Youngs modulus
and density of the element. The desired case is that the design variable will be driven
to either its upper bound value of 1.0 (retain the element) or its lower bound values
close to zero (discard the element). In addition to specifying the PTYPE/ PID
information, the entry provides an initial value for the variable upper and lower
bound , DELXV parameter to specify how rapidly the design variable can change, and
POWER parameter that is the exponent used in Eq. 2-162 (as given in Topology
Optimization on page 137). Typically, the default values for these parameters are
adequate.
The PTYPEs than can be invoked by the TOPVAR include PROD, PBAR, PBARL,
PBEND, PBEAM, PBEAML, PSHELL, PSHEAR, PSOLID, and PWELD. All
designated element properties must refer to a MAT1 entry; therefore, a PCOMP
cannot be designed in topology optimization.
DDi Draw Direction. DDi=X, Y, Z or X-, Y-, Z- for a single die option (DIE=1) where
X-, Y-, Z- indicates the opposite direction of X, Y, and Z respectively. DDi=X,
Y, and Z for two die option (DIE =2) (Character)
DIE Die Options. (Blank or integer 1 or 2; Default = 1)
= 1 (or blank). A single die will be used and the die slides in the given draw
direction (i.e., material grows from the bottom in the draw direction)
= 2. Two dies will be used and the dies split apart along the draw direction (i.e.,
material grows from the splitting plane in opposite direction along the axis
specified by the draw direction DDi. The splitting plane is determined by
optimization)
EXT Indicates that this line defines extrusion constraints (i.e., enforce constant cross-
section).
EDi Extrusion direction. (Character, X, Y, or Z)
TDMIN Indicates that this line defines a minimum member size.
TV Minimum member size. (Real > 0.0)
Field Contents
Main Index Main Index
303 CHAPTER 4
Input Data
The optional continuation lines for this entry permit the specification of
manufacturability constraints as defined as given in the Topology Optimization on
page 137, the same question applies for the citation given in the previous paragraph).
Manufacturability constraints are specified in a rectangular (CORD1R and CORD2R)
coordinate system and only one system can be specified for a given TOPVAR entry.
One, two or three different mirror symmetry planes can present (such as MS1=XY,
MS2=YZ, and MS3=ZX). Casting (CAST) and Extrusion (EXT) manufacturability
constraints can be applied to PTYPE=PSOLID only. Casting constraints cannot be
combined with extrusion constraints for the same TOPVAR entry.
Some symmetry constraint types can be combined with casting or extrusion
constraints. The referenced coordinate system CID must be the same for the combined
constraints. Some possible combinations are:
For EXT constraints, possible combinations are (Edi=X, MSi=XY and/ or
ZX), (EDi=Y, MSi=YZ and/ or XY), (EDi=Z, MSi=ZX and/ or YZ),
For CAST constraints, possible combinations are (DDi=X or -X, MSi=XY
and/ or ZX), (DDi=Y or -Y, MSi=YZ and/ or XY), (DDi=Z or -Z, MSi=ZX
and/ or YZ)
For the two dies option (DIE=2), the splitting plane is optimized. For a single die
DIE=1, the parting plane is the bottom surface of the designed part in the draw
direction.
TDMIN is a dimensional quantity with a guideline that it be set to at least three times
a representative element dimension. Without a TDMIN continuation line, the
minimum member size constraint is taken from the specification of TDMIN parameter
on the DOPTPRM entry. This option is applied on 2 and 3 D elements only. Minimum
member size constraints can be used with SYM, CAST, and EXT constraints.
The casting constraints may have difficulty dealing with a design model that has one
or more non-smoothed boundary surfaces to be designed. It is recommended to use
smooth surfaces for topology designed boundary surfaces.
Topology Optimization on page 513 contains a number of simple Topology
Optimization examples that demonstrate various manufacturability constraints.
Main Index Main Index


304
4.5 Parameters Unique to Design Sensitivity and
Optimization
MSC.Nastran allows you to specify a number of parameters for the input of scalar
values and for requesting special features. These are in addition to the optimization
control possible with Bulk Data entries such as DOPTPRM and DSCREEN. This
section is intended to be a complete quick reference for all such Case Control and Bulk
Data parameters that are associated with Design Sensitivity and Optimization. Refer
to Parameters on page 631 of the MSC.Nastran Quick Reference Guide for a
description of all MSC.Nastran parameters.
AUTOADJ Select the adjoint sensitivity analysis.
YES Selection is based on a determination as whether the
adjoint or the direct method is most efficient. (Default)
NO Direct method is used.
CDIF Controls the selection of finite difference scheme used in
sensitivity analysis.
YES Selects central differences. (Default for shape
optimization)
NO Selects forward differences. (Default for property
optimization only)
DESPCH Controls the frequency of updated DESVAR and GRID Bulk Data
entry output to the PUNCH file.
N < 0 No output.
N = 0 Final design cycle. (Default)
N > 0 Every N-th, as well as final design cycles.
DESPCH1 Specifies the amount of data to be written to the .pch file.
N = 0 Write no data.
N = + 1 Write property entries that are designed.
N = + 2 Write all property entries of a given type when one or
more property of that type is designed.
N = + 4 Write new DESVAR and DRESP1 entries.
N = + n Write combined quantities from summing values.
Main Index Main Index
305 CHAPTER 4
Input Data
Note: A positive value of N results in large field formats
while a negative value gives small field formats.
DPEPS Controls overriding of the analysis value by the design model.
(Default = 1.0E-4)
If all the differences between the analysis and design model
properties are less than DPEPS, the differences are ignored and it
is possible to reuse analysis results from a previous run. Setting
this parameter to a large value is not recommended since it can
result in poor sensitivity values.
DSNOKD Indicates if the differential stiffness effect is to be included in
buckling sensitivity analysis.
1.0 Include differential stiffness.
0.0 Do not include differential stiffness. (Default)
DSZERO Specifies the minimum absolute value for printout of design
sensitivities.
IUNIT IUNIT specifies the FORTRAN unit number on which the
DSCMCOL table and the DSCM2 matrix in SOL 200 will be
written. (See the OUTPUT2 and OUTPUT4 module descriptions
in the MSC.Nast ran 2005 r3 DMAP Programmers Guide.)
(Default = 11)
NASPRT Controls the frequency of MSC.Nastran output.
-1 No output.
0 Output on initial and final design cycles. (Default)
N Output every N-th iteration and also prior to exit.
OPTEXIT Instructs the program to exit at one of six predetermined exit
points.
0 No user-defined exit (default).
N > 0
Exit at one of the locations 1 through 7 .
SENSUOO If set to yes, pseudo-displacements are recovered to the o-set
while including the terms of Eq. 2-76.
SOFTEXIT Determines whether to terminate design cycles if soft
convergence is indicated.
1 N 7 ( )
Main Index Main Index


306
NO Do not stop if soft convergence is indicated. (Default)
YES Terminate design cycles if soft convergence is
achieved.
SOLADJC Indicates if adjoint solution vectors are to be calculated during the
analysis.
-1 Do not calculate adjoint solution vectors during the
analysis.
>0 For ANALYSIS = DFREQ, the adjoint vectors will be
calculated during the solution if the conditions given
in the full description in Parameters on page 631 of
the MSC.Nastran Quick Reference Guide are met.
(Default)
UPDTBSH Controls the update of the boundary shapes in the analytic
boundary shapes method for shape optimization.
NO Do not update the boundary shapes. (Default)
YES Update the boundary shapes.
Note: Regardless of the value of UPDTBSH, shape basis
vectors are still updated for every design cycle
(interpolation to the interior grids). UPDTBSH only
controls updates on the boundary shapes.
XYUNIT n Specifies the unit number for storage of design
sensitivity data in comma-separated value format.
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
CHAPTER
5
Solution Sequences
I Solution 200 Program Flow
I Design Sensitivity and Optimization subDMAPs
I Design Sensitivity and Optimization Modules
Main Index Main Index


308
For most applications, it is unlikely that an extensive understanding of the SOL 200
DMAP sequence or the design optimization modules will be required. However, a
general knowledge of the overall program flow, the subDMAPs and the module
functions may be of help when evaluating design optimization messages, user
warning messages, or error messages. Advanced users may wish to alter the DMAP
solution sequence to perform customized tasks, but the user is cautioned that the
multidisciplinary nature of SOL 200 makes it a particularly difficult Solution Sequence
to understand
This chapter provides a high level overview of the SOL 200 sequence and a brief
description of the key subDMAPs that are used. A short description is also provided
of those modules which are unique to design optimization. Many of the module
names begin with the letters DOM (which stands for design optimization module) or
DSA (for design sensitivity analysis). More complete information on the modules and
on data blocks associated with design sensitivity and optimization can be found in the
MSC.Nast ran 2005 r3 DMAP Programmers Guide.
Main Index Main Index
309 CHAPTER 5
Solution Sequences
5.1 Solution 200 Program Flow
Design sensitivity analysis and optimization are available in Solution Sequence 200.
The following flowchart describes the organization of the DESOPT subDMAP that is
the driver for Solution 200 as well as listing some of the more significant modules and
subDMAPs invoked.
Each of the boxes in this high-level flowchart describes a major program phase as well
as the design sensitivity and/ or optimization modules used to accomplish that
particular task. These module names are included inside each of the boxes. If program
flow is directed to a particular subDMAP, each name is listed alongside each of the
boxes.
The circled numbers 1 through 7 appearing in the flowchart indicate the locations of
seven user-defined exit points. Any one of these exits can be taken by setting the
OPTEXIT parameter. Other normal program terminations include both hard and soft
convergence and maximum number of design cycles.
Main Index Main Index


310
Figure 5-1 Solution 200 Sequence Flowchart
DRESPi Preprocessing that does not
require an analysis (DOPR3)
Begin Design Cycle Loop
Collect Analysis Results, Generate
Enforced Boundary Displacements
for Auxiliary Structure Analysis AXMDRV, DSAJ
MSC.Nastran Input File Processing
for Primary
and All Auxiliary Models
AXMDRV, AXMPR1
1
Generate Nonrepetitive Tables Used in
Design Sensitivity and Analysis.
Analysis Model Override by Design Model
DOPR1, DOPR2, DOPR4, DOM11
EXITOPT
IFPL
PREDOM
Case Control Partitioning
Based on Analysis Type
MDCASE
PHASE0 Operations, Restart Checking for
Primary and Auxiliary Models (for Shape)
AXMDRV, AXMPR2
DESINIT,
SETSOLAP
PHASE0
Design Optimization Data Initialization
DMPR, SDSA, SDSB
PRESENS
If AMBS
or GMBS
If AMBS
Static Analysis for All Auxiliary Models,
Loop over Auxiliary Models, Boundary
Conditions (FEAOPT = Analysis)
BNDSHP
YES
NO
YES
NO
Auxiliary Model
Boundary Shapes
Geometry Model
Boundary Shapes

GMBS

AMBS
FEA
L
o
o
p

o
n

D
e
s
i
g
n

C
y
c
l
e
Main Index Main Index
311 CHAPTER 5
Solution Sequences
Figure 5-1 Solution 200 Sequence Flowchart (continued)
DRESPi Processing
2
Finite Element Analysis and Data Recovery
for All Analysis Disciplines, Superelements,
and Subcases. Basis Vector Solution
(for Shape Optimization)
Hard Convergence and
Not Discrete
Optimization
Pre-Sensitivity Operations. for
Shape Basis Vector Initialization
DOPR2, DOPR4
FEA
NO
Shape Variables
Scale Shape Variables
DRESPi Depends on
Analysis
SCSHBV
PRESENS
Hard Convergence Check
DOM12
Design Constraint Evaluation and Screening
DSAD, DSAR, DSPRM, WEIGHT
DESCON
SDR2STAT
Design Cycle = 1 or
FSD
Data Recovery
FEA
L
o
o
p

o
n

D
e
s
i
g
n

C
y
c
l
e
YES
YES
NO
NO
YES
Print Results DOM10
Main Index Main Index


312
Figure 5-1 Solution 200 Sequence Flowchart (continued)
Design Constraint Evaluation and
Screening for MP, DSAD, DSAR,
DSPRM, WEIGHT
Exit and Print
Sensitivities
DSTAP2, DOM6, SDSC
Compute Necessary Response
Sensitivities for All Retained Constraints
DSAF, DSAH, DSAL, DSAW, DSDVRG,
DSFLTF, DSFLTE, DSVG1, DSVG4,
FSD is done and Math
Programming(MP) is to be
Performed
FEA
DESAERDR,
PSLGDV
L
o
o
p

o
n

D
e
s
i
g
n

C
y
c
l
e
DESCON
SDR2STAT
EXIT
6, 3
LAST CYC = TRUE
Hard
Convergence
NO
4
RESPSEN,
ADJSENS,
ADJSENF,
SDR2STAT,
FLUTSENS,
CEIGSENS,
EXITOPT
YES
EXITOPT
Generate Solutions for
Sensitivity Analysis
FSD
YES
YES
OPTEXIT = 3 or (OPTEXIT = 6
and Design Cycle = 2) or Hard
Convergence and no Discrete
Optimization and no Design
Sensitivity
FSD
RESIZING
DOPFS
YES
F
S
D

P
a
t
h
Generate Pseudo Loads
Main Index Main Index
313 CHAPTER 5
Solution Sequences
Figure 5-1 Solution 200 Sequence Flowchart (continued)
End Design
Cycle Loop
If Soft
Convergence
Data Recovery
and Exit
(FEAOPT =
DATARECOVERY")
NO
YES
FEA, *
EXITOPT
Design Optimization
Output DOM10
Soft Convergence Check
DOM12
5
Analysis Model Updates
(Auxiliary + Primary)
AXMPR2
UPDATE
7
L
o
o
p

o
n

D
e
s
i
g
n

C
y
c
l
e
Discrete
Optimization
Perform Discrete
Optimization
NO
YES
*OPTEXIT = 7 includes computation of sensitivities for the current design cycle (all
operations up to DOM9).
Optimization with Respect
to Approximate Models
DOM9
F
S
D

P
a
t
h
Output Updated Element
and Grid Entries
DOM11
Main Index Main Index


314
5.2 Design Sensitivity and Optimization SubDMAPS
The Solution 200 Sequence presented in the flowchart of Figure 5-1 is made up of a
number of subDMAPS. Table 5-1 is a list of the design sensitivity and optimization
subDMAPs that are contained in the overall Solution Sequence. The list includes the
name of the subDMAP, the name of the subDMAP that calls it and a brief description
of the function of the subDMAP.
Table 5-1 SubDMAPs in Solution 200 (SOL 200)
SubDMAP
Name
Calling
SubDMAP
Description
ADJSENS RESPSEN Calculation of adjoint sensitivity for statics analysis
ADJSENF RESPSEN Calculation of adjoint sensitivity for frequency
response analysis
BNDSHP DESOPT For the geometric and analytic boundary shape
methods, generate dummy case control records for
shape loads and appends them to the primary
model case control. Output enforced displacement
vectors corresponding to the boundary shapes.
CEIGSENS RESPSEN Compute sensitivities for complex eigenvalues
DESAERDR PSLGDV Performs special preparation operations needed in
the computation of static aeroelastic sensitivity
analysis
DESCON DESOPT Evaluates response values, the objective and
constraints. Performs constraint screening and load
case deletion, thereby setting up for sensitivity
analysis.
DESINIT DESOPT Performs miscellaneous initialization tasks prior to
design sensitivity and optimization. These tasks
include setting up data blocks that guide design
sensitivity when there are superelements.
DESOPT Driver This is the driver subDMAP for SOL 200 and
performs the functions listed in Figure 5-1.
DISOPTS DESOPT Performs discrete variable optimization
Main Index Main Index
315 CHAPTER 5
Solution Sequences
DSTATRS FEA Performs static sensitivity analysis
EXITOPT DESOPT Performs cleanup operations at the end of a design
sensitivity and optimization task. This includes
such tasks as fulfilling requests for design
sensitivity data, writing final design results, storing
data to the punch file and the output database(s).
FEA DESOPT This is the workhorse subDMAP for performing
analysis and data recovery. It also performs the
numerical analysis related tasks for sensitivity
analysis, such as solving for pseudo-displacements.
As such it calls a number of subDMAPs that are not
listed here but that perform each of the distinct
analyses (statics, normal modes, buckling, etc.) that
are required. Data recovery is performed in a
separate call to FEA as is sensitivity analysis.
FEAMODE FEA Performs normal modes analysis
FEAFREQ FEA Performs frequency response analysis
FLUTSENS RESPSEN Computes sensitivities for flutter eigenvalues
GENDV2P PSLGDV/
RESPSEN
Generates bookkeeping matrices when there are
multiple boundary conditions.
INERSEN DSTATRS Computes duddot for inertia relief sensitivity and
pseudo-loads for the l-set (See Inertia Relief
Sensitivity Analysis on page 76).
PREDOM DESOPT Performs preprocessing operations for the designed
properties and shapes. This creates invariant tables
and matrices that are used throughout the design
sensitivity and optimization task. Also sets up
parameters to control the print of design sensitivity
information.
Table 5-1 SubDMAPs in Solution 200 (SOL 200) (continued)
SubDMAP
Name
Calling
SubDMAP
Description
Main Index Main Index


316
PRESENS DESOPT Performs preprocessing operations for design
responses. In the typical case, this can be done prior
to the design cycle loop, but certain situations (such
as mode tracking and support of
FREQ3,FREQ4,FREQ5 entries) necessitate calling
this subDMAP within the design cycle loop.
PSLGDV DESOPT Generates the pseudo-loads vectors for statics, static
aeroelasticity and dynamic response analysis. Also
computes the sensitivity of normal modes
eigenvalues.
RESPSEN DESOPT Generates the DSCM sensitivity matrix which is
used to compute design sensitivities. The
subDMAP invokes CEIGSENS and FLUTSENS for
special sensitivity calculations and performs the
major operations for weight, volume and buckling
sensitivities.
SAERSENS FEA Generates the basic sensitivity information requires
for static aeroelastic subcases.
SCSHBV DESOPT Scales the shape basis vectors when performing
perturbation analysis.
SDR2STAT DESCON/
RESPSEN
Performs static data recovery.
SETDROPT DESOPT Specifies parameters that drive data recovery
operations during design sensitivity and
optimization.
UPDATE DESOPT In shape optimization, creates key data blocks for
the next design cycle based on the updated
geometry.
XMODTRK FEAMODE Performs mode tracking.
Table 5-1 SubDMAPs in Solution 200 (SOL 200) (continued)
SubDMAP
Name
Calling
SubDMAP
Description
Main Index Main Index
317 CHAPTER 5
Solution Sequences
5.3 Design Sensitivity and Optimization Modules
A number of the functional modules are specific to the design sensitivity and
optimization capability. Table 5-2 names each of these modules, lists the name(s) of
the subDMAPS that call the modules and provides a brief description of each of them.
Additional information can be obtained by referring to the MSC.Nast ran 2005 r3
DMAP Programmers Guide.
Table 5-2 Design Sensitivity and Optimization Modules
Module
Name
Calling
SubDMAPs
Description
AXMDRV BNDSHP
DESOPT
IFP2
(Auxiliary model driver) drives the auxiliary model
looping (for shape optimization using auxiliary
boundary models) by reading the auxiliary model
list and outputting the next sequential auxiliary
model ID (AUXMID). The auxiliary model loop flag
AMLPFL is FALSE if this is the last auxiliary model
in the list.
AXMPR1 IFPL Builds the auxiliary model list AMLIST used in the
auxiliary model boundary shapes method for shape
optimization.
AXMPR2 DESOPT Generates complete geometry (grid) data for the
auxiliary boundary models by merging with the
geometry of the primary structure. It also generates a
Case Control data block CASEVEC used later to
generate a geometry partitioning vector. This is used
if additional grids have been defined by the engineer
in the auxiliary boundary model.
DISOPT DISOPTS Performs discrete variable optimization.
DOM6 EXITOPT Computes the design sensitivity coefficient matrix
DSCM2 for output purposes. This module is
invoked only if sensitivities are to be printed.
Otherwise, basic design sensitivity information is
available to the optimizer via matrix DSCM.
DOM9 DESOPT Performs the optimization task using the previously
created approximating functions.
Main Index Main Index


318
DOM10 DESOPT
DISOPTS
EXITOPT
Provides printed output of the results of the
approximate optimization. This output includes the
values of the proposed new design variables and the
estimated values of the objective and constraint
functions. The amount of output is determined by
the parameters P1 and P2 on the DOPTPRM Bulk
Data entry.
DOM11 DESOPT
DISOPTS
Updates the element property table (EPT), grid
coordinates (COORDO), and geometry data blocks
(GEOM1N) in accordance with the new vector of
design variables.
DOM12 DESOPT
DISOPTS
EXITOPT
Performs both hard and soft convergence testing,
outputs the results of these tests, updates the design
optimization history table, and prints out the
summary of design cycle history. It is called for
every normal exit condition in Solution 200.
DOPFS DESOPT Performs fully stressed design when a non-zero
value of FSDMAX is provided on the DOPTPRM
Bulk Data entry.
DOPR1 PREDOM Sets up nonrepetetive tables for design variables,
design variable linking, DTABLE constants,
designed property attributes, initial design variable
and property values, constraint screening data, and
optimization parameters.
DOPR2 DESOPT
PREDOM
SCSHBV
Performs pre-processing operations for the shape
basis vectors (sensitivity and optimization). It is
called in PREDOM for direct input of basis vectors,
and DESOPT for the AMBS, GMBS and DVGRID
options. Creates the shape basis vectors in basic
coordinates as well as the necessary design variable
correlation data.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
Main Index Main Index
319 CHAPTER 5
Solution Sequences
DOPR3 PRESENS Generates tables that are used to compute both first-
and second-level responses, constraints, and the
objective function. Since these tables are typically
invariant, they are usually formed prior to the first
design cycle. However, dataflow needs sometimes
dictate that the module be called following the
analysis of the first design cycle and, in rarer cases,
after every analysis.
DOPR4 DESOPT
PREDOM
Expands tables DTOS2J and DTOS4J to account for
DVPREL2 properties, which are treated as internal
design variables. Design variable numbering in the
resultant DTOS2K and DTOS4K tables is in terms of
the external design variables (DESVAR input) and
the internal (DVPREL2) variables.
DOPR5 DESOPT Generates updated DTOS2 and DTOS4 tables from
DTOS2K and DTOS4K using the new vector of
design variables XINIT.
DOPR6 PSLGDV Shape optimization perturbations.
DOPRAN PRESENS Performs pre-processing operations when there are
random root mean square 00 power spectral density
responses.
DSABO PSLGDV In the semianalytic sensitivity analysis approach
used in MSC.Nastran, perturbed structural matrices
must be generated. DSABO constructs tables to
which later modules refer when building property-
related variational terms. These tables are primarily
in the form of perturbed element summary tables
and design sensitivity processing tables.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
Main Index Main Index


320
DSAD DESCON Evaluates first, second and third level responses and
then computes and screens constraints that are
functions of these responses. Also evaluates the
objective, performs load case deletion and makes a
determination if the adjoint method of sensitivity
analysis is appropriate. If the adjoint method is
chosen, the adjoint loads and associated tables are
produced to drive adjoint sensitivity analysis. Case
control records are created to drive data recovery
during the sensitivity analysis
DSADJ RESPSEN Calculates terms for the sensitivity matrix for the
adjoint method. Separate calls produce contributions
from mass, damping and stiffness perturbations.
DSAE PSLGDV For purposes of computational efficiency, the design
variables are split into two sets. The first set consists
of those design variables that are related to
structural properties, while the second set consists of
variables related to grid coordinates. DSAE merges
the tables and data blocks associated with these
groups of design variables so that later modules can
compute the perturbed structural configurations.
DSAF RESPSEN Generates an updated element summary table
(ESTDCN) for recovery of responses in design
sensitivity analysis. It contains the subset of
elements necessary to recover element-level
responses and incorporates the effects of constraint
deletion.
DSAH PSLGDV
RESPSEN
Forms a number of miscellaneous tables and data
blocks, including Case Control that are used in the
formation of the design response sensitivities.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
Main Index Main Index
321 CHAPTER 5
Solution Sequences
DSAJ BNDSHP Converts grid variations defined on the boundaries
of the structure (geometry model boundary solution
for shape optimization) to a shape basis vector. This
in turn is used to define enforced displacements on
the primary structure, for purposes of interpolation
to the interior grids of the mesh.
DSAL RESPSEN Actually computes the matrix of partial sensitivity
coefficient DSCM given the baseline and perturbed
responses, and corresponding design variable data.
The form of the terms in DSCM is . This
form is converted to in modules DOM6 and
DOM9.
DSAM PSLGDV Same concept as the DSABO module but with
respect to structural shape variations. Computes
perturbed basic grid point data.
DSAN PSLGDV Generates Design Sensitivity Table 1 DSPT1 used in
DSVG1 and DSVG2 to generate the pseudo-load
vectors.
DSAP PSLGDV In direct and modal frequency analysis, DSAP sums
the variational terms that are functions of , ,
and to form the pseudo-load vector.
DSAPRT EXITOPT Prints formatted sensitivity results as directed by the
DSAPRT Case Control command.
DSAR DESCON
FEA
PSLGDV
Splits the modal transient solution containing data
for displacement, velocity, and acceleration onto
separate data blocks: one data block each for
displacement, velocity, and acceleration for all
solution time steps.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
r DELB
r x
K B
M
Main Index Main Index


322
DSARLP DESAEPR
SAERSENS
Calculates pseudo-trim vectors that are used in
calculating the sensitivities of aeroelastic stability
derivatives. Also sets parameters used in static
aeroelastic sensitivity analysis.
DSARME DESCON Calculates random root mean square and power
spectral density responses.
DSARMG RESPSEN Calculates gradients of random root mean square
and power spectral density.
DSARSN SAERSENS In a static aeroelastic design task, computes the delta
response values for stability derivatives and trim
variables.
DSAW RESPSEN Computes the delta volume according to the changes
in the design variables in connection with the
semianalytic sensitivity analysis. Also creates tables
that are used in the calculation of the delta weight.
DSFLTE FLUTSENS
CEIGSENS
In a complex eigenvalue or flutter sensitivity
analysis, obtains the eigenvectors needed in the
analysis and develops a table of parameters that are
applied in the DSFLTF module.
DSFLTF FLUTSENS
CEIGSENS
Completes the complex eigenvalue of flutter
sensitivity analysis and places the results in an
output table.
DSPRM DESCON
DESOPT
PRESENS
PSLGDV
RESPSEN
Outputs parameters used for DMAP flow control.
Depending on the types of responses required,
parameters are output to indicate whether or not
response evaluation is required or whether
sensitivities are to be recovered.
DSTAP2 EXITOPT Generates the DSCMCOL correlation table
describing the design response column order for the
DSCM2 matrices.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
Main Index Main Index
323 CHAPTER 5
Solution Sequences
DSVG1 CEIGSENS
FLUTSENS
PSLGDV
RESPSEN
(Design Sensitivity Vector Generator 1) computes
terms in the pseudo-load vectors for sensitivity
analysis that depend on , , and . For
normal modes, buckling, complex eigenvalues and
complex eigenvalue sensitivities, performs a triple
matrix product using the eigenvectors and the
perturbed element matrices.
DSVG2 PSLGDV Design Sensitivity Vector Generator 2. This module
computes the thermal load contribution to the
pseudo-load vector.
DSVGP4 PSLGDV Generates a perturbed multipoint constraint
transformation matrix for rigid body element
sensitivity analysis.
DSVGP5 PSLGDV
RESPSEN
Matrix utility used in rigid body sensitivity analyses
to perform multiple multiply and add operations
when the second matrix in the multiply is fixed
while the first matrix and the add term can vary.
MDCASE DESOPT
PRESENS
PSLGDV
RESPSEN
Partitions the original input case control onto
individual Case Control records based on analysis
type. It also outputs parameters that are used at the
DMAP level to direct analysis for the various
disciplines.
MODTRK XMODTRK Performs mode tracking based on the MODTRAK
Bulk Data entry. If necessary, eigenvectors and
eigenvalues are reordered to be consistent with the
original design.
SDSA DESINIT Partitions the design model entries table (EDOM) on
superelement. The result, EDOMS, is a superelement
partition of the original EDOM.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
K B M
Main Index Main Index


324
SDSB DESINIT Builds the superelement design sensitivity
processing data block DSLIST that is used to direct
the pseudo-load and response sensitivity
superelement loops. Its structure is similar to SLIST,
used in superelement analysis.
SDSC EXITOPT Prints the DSCMCOL correlation table that identifies
the column-response correlation in the design
sensitivity coefficient matrix DSCM2.
SHPCAS BNDSHP Generates a new Case Control data block to drive
solutions for basis vectors in shape optimization.
The initial shape data is supplied as enforced
displacements on the boundary of the structure and
then interpolated to the interior grids via a static
solution of the primary structure. SHPCAS generates
the Case Control data block CASENEW, which is
used in connection with this solution. It also
generates a partitioning vector CVEC to extract the
appropriate solution vectors from the analysis
results.
WEIGHT DESCON Computes the structural volume for use in design
optimization. If there is a need for a weight response,
the OPWG table which has been formed in the
GPWG table is converted into a 6x6 matrix. When
P2=15 has been set on the DOPTRM Bulk Data entry,
this module is used to produce a table that contains
the structural weight for each material ID in the
model.
Table 5-2 Design Sensitivity and Optimization Modules (continued)
Module
Name
Calling
SubDMAPs
Description
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
CHAPTER
6
Output Features and Interpretation
I Output-Controlling Parameters
I Design Optimization Output
I Design Sensitivity Output
I Punch Output
I Postprocessing Output
Main Index Main Index


326
An MSC.Nastran run will produce a number of output files and the goal of this
chapter is to explain the features of these files that are unique to Design Sensitivity and
Optimization. Results shown in the .f06 file are emphasized, but attention is also given
to results in the .pch (punch) file, a special comma separated values file and in two
possible output database files: .op2 and .xdb.
Since design optimization is an iterative process, a significant amount of output may
be generated depending on the quantity of data requested for each design cycle and
the number of cycles performed. There are a few parameters that affect the level of
detail and the frequency of output (for design cycle-dependent data). Although
covered elsewhere in this guide, these parameters are summarized in this chapter in
Output-Controlling Parameters on page 327.
This is followed by Design Optimization Output on page 329 that serves to
document the printed output that is unique to design optimization through the use of
the output from a design optimization problem (DSOUG4.DAT) in the Test Problem
Library (TPL) supplied with your MSC.Nastran installation tapes. This example
introduces most of the frequently encountered types of design optimization output,
within the context of an actual problem. The section also documents prints that occur
for some of the specialized features of design optimization, such as Fully Stressed
Design. The Design Sensitivity Output on page 364 is a similar output example for
design sensitivity analysis. Punch output is described in Design Punch Output on
page 374 while several postprocessing options are described in Postprocessing
Output on page 380.
Main Index Main Index
327 CHAPTER 6
Output Features and Interpretation
6.1 Output-Controlling Parameters
Parameters that affect either the frequency or level of detail of the output are discussed
here briefly. Some of them (namely, IPRINT, P1, P2, and P2xxxx) are specified on the
DOPTPRM Bulk Data entry described in Bulk Data Entries on page 216 while the
remainder are input using the PARAM Bulk Data entry. More information on many
of these same pararmeters can be obtained from Parameters Unique to Design
Sensitivity and Optimization on page 304.
DESPCH Controls the frequency updated Bulk Data entry output to the punch file.
The default is for the final design cycle only, but can be changed to every
n-th design cycle (plus the last) with this parameter.
DESPCH1 Controls the amount and form of updated Bulk Data entry output to the
punch file. The amount refers to updated properties, DESVAR, DRESP1
and GRID entries. The form selects between short and large field
formats.
IPRINT Controls the level of printout available directly from the optimizer. Its
value, which may range from 0 to 7, is set on the DOPTPRM entry. An
increasing value provides increasing levels of detail. See the Bulk Data
entry, DOPTPRM on page 231.
NASPRT Governs the frequency of MSC.Nastran analysis results output. Output
may be requested for the first and last designs only, the first and every
n-th design, or turned off completely with this Bulk Data parameter.
OPTEXIT When set to a value of 4 or 7, provides output of the unformatted design
sensitivity coefficient matrix DSCM2. This Bulk Data parameter may
assume any value from 1 to 7 indicating exit at any one of seven
predefined locations in the solution sequence. However, only
OPTEXIT=4 and 7 generate DSCM2 (design sensitivity coefficient)
output. This parameter has been superseded by the DSAPRT Case
Control command.
P1 Controls the frequency of design cycle output. By default, initial and
final optimization summaries are output but can be changed to every n-
th cycle using this parameter.
P2 Controls how much design cycle information is output at the intervals
determined by P1.
P2CR Maximum number of Constraints on Responses to be printed.
Main Index Main Index


328
P2CDDV Maximum number of Constraints on Dependent Design Variables to be
printed.
P2CP Maximum number of Constraints on Properties to be printed.
P2CC Maximum number of Constraints on Connectivity properties to be
printed.
P2CM Maximum number of Constraints on Material properties to be printed.
P2CBL Maximum number of Constraints on Beam Library dimensions to be
printed.
P2CALL Maximum number of Constraints of all categories to be printed.
P2RSET SET1 ID for which a set of Response IDs is defined.
POST This Bulk Data parameter is used to identify the data base (.xdb or .op2)
to be used for storing pre- and postprocessing data and, in the case of the
.op2 file, what postprocessor is to be used. In the context of design
optimizaton, PARAM POST-1 is the most useful in that it creates data
blocks that can be used to perform xy plotting in MSC.Patran as shown
in Postprocessing Output on page 380.
Main Index Main Index
329 CHAPTER 6
Output Features and Interpretation
6.2 Design Optimization Output
This section presents design sensitivity and optimization output in the context of one
of the example problems (DSOUG4) from MSC.Nastrans test problem library that is
also presented as the Stiffened Plate on page 412. This problem is chosen because
it contains most of the output commonly encountered in a typical optimization
example. The best approach is for the reader to locate this problem in the test problem
library and run it on his/ her installation. The output from the actual problem can then
be referenced while reading this section. The results shown here were generated using
MSC.Nastran 2004 on an Intel-based PC running Linux. Your output should be quite
similar, but there may be slight differences.
DSOUG4 Output
Assuming you have run this example on your installation and have the output file
available, you will notice the following table appearing shortly after the Bulk Data
echo:
Main Index Main Index


330
Note that this example has two sets of properties that are being compared: the
properties that are defined on a property Bulk Data entry and designed using a
DVPRELi Bulk Data entry and the connectivity properties that are defined on
connectivity bulk data entries (those that begin with the letter c) and that are
designed using the DVCRELi Bulk Data entries. If there had been designed material
properties in the example, this would have caused the print of an additional table that
is a COMPARSION BETWEEN INPUT MATERIAL PROPERTY VALUES FROM
ANALYSIS AND DESIGN MODELS. This comparison data is only printed once, at
the beginning of the design task and it cannot be turned off.
The point of these tables is to allow you to check if the design values in your model
are what you expect and to see if they differ from the analysis values. If any of the
analysis model properties differ from the design model description, the design model
will be used to override the analysis model. This table simply reports on any
differences found and provides notification if this override took place. The design
Note: Analysis
and Design
Values.
The design value
will be used
when the two
differ.
----- COMPARISON BETWEEN INPUT PROPERTY VALUES FROM ANALYSIS AND DESIGN MODELS -----
-----------------------------------------------------------------------------------------------------------------------------
PROPERTY PROPERTY PROPERTY ANALYSIS DESIGN LOWER UPPER DIFFERENCE SPAWNING
TYPE ID NAME VALUE VALUE BOUND BOUND FLAG FLAG
-----------------------------------------------------------------------------------------------------------------------------
PBAR 3 A 9.600000E-01 9.600000E-01 1.000000E-15 1.000000E+20 NONE *
PBAR 3 I1 1.207200E+00 1.207200E+00 1.000000E-15 1.000000E+20 NONE *
PBAR 3 I2 9.812000E-01 9.812000E-01 1.000000E-15 1.000000E+20 NONE *
PBAR 3 J 3.200000E-03 3.200000E-03 1.000000E-15 1.000000E+20 NONE *
PBAR 3 C1 1.500000E+00 1.500000E+00 -1.000000E+35 1.000000E+20 NONE *
PBAR 3 D1 -1.500000E+00 -1.500000E+00 -1.000000E+35 1.000000E+20 NONE *
PBAR 3 E1 -1.500000E+00 -1.500000E+00 -1.000000E+35 1.000000E+20 NONE *
PBAR 3 F1 1.500000E+00 1.500000E+00 -1.000000E+35 1.000000E+20 NONE *
PBAR 3 K1 5.833333E-01 5.833333E-01 -1.000000E-03 1.000000E+20 NONE *
PBAR 3 K2 3.750000E-01 3.750000E-01 -1.000000E-03 1.000000E+20 NONE *
PBARL 3 DIM2 1.000000E-01 1.000000E-01 N/A N/A NONE
PSHELL 1 T 1.500000E-01 1.500000E-01 1.000000E-02 1.000000E+20 NONE
1. THE DIFFERENCE FLAG IS USED TO CHARACTERIZE DIfFERENCES BETWEEN ANALYSIS AND DESIGN MODEL PROPERTIES:
IF THE FLAG IS NONE, THEN THERE IS NO SIGNIFICANT DIFFERENCE BETWEEN THE TWO VALUES.
IF THE FLAG IS WARNING, THEN THE USER IS ADVISED THAT DIFFERENCES EXIST AND THE DESIGN MODEL
IS BEING USED TO OVERRIDE THE ANALYSIS MODEL.
IF THE FLAG IS FATAL, THEN THE DIFFERENCES ARE GREATER THAN 1.00000E+35 AND THE RUN WILL BE TERMINATED.
2. THE SPAWNING FLAG (*) INDICATES THAT THE SPAWNED PROPERTY IS DERIVED EITHER FROM THE BEAM CROSS SECTION
LIBRARY OR FROM A PBEAM ENTRY. THE PROPERTY ID FOR THE SPAWNED PROPERTY IS IDENTICAL TO ITS PARENT.
----- COMPARISON BETWEEN INPUT CONNECTIVITY PROPERTY VALUES FROM ANALYSIS AND DESIGN MODELS -----
---------------------------------------------------------------------------------------------------------------
ELEMENT ELEMENT CONNECTIVITY ANALYSIS DESIGN LOWER UPPER DIFFERENCE
TYPE ID NAME VALUE VALUE BOUND BOUND FLAG
---------------------------------------------------------------------------------------------------------------
CBAR 31 W3A 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 31 W3B 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 32 W3A 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 32 W3B 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 33 W3A 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 33 W3B 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 34 W3A 1.575000E+00 1.575000E+00 N/A N/A NONE
CBAR 34 W3B 1.575000E+00 1.575000E+00 N/A N/A NONE
1. THE DIFFERENCE FLAG IS USED TO CHARACTERIZE DIfFERENCES BETWEEN ANALYSIS AND DESIGN MODEL PROPERTIES:
IF THE FLAG IS NONE, THEN THERE IS NO SIGNIFICANT DIFFERENCE BETWEEN THE TWO VALUES.
IF THE FLAG IS WARNING, THEN THE USER IS ADVISED THAT DIFFERENCES EXIST AND THE DESIGN MODEL
IS BEING USED TO OVERRIDE THE ANALYSIS MODEL.
Main Index Main Index
331 CHAPTER 6
Output Features and Interpretation
value column contains the property values computed from the initial design variable
values. In this example, no differences were found for any of the properties defined
on property entries or on connectivity properties. Note that the first table lists a series
of PBAR properties that have not been explicitly designed in the input file. Instead, the
second beam dimension on the PBARL entry (which corresponds to the thickness of a
HAT beam cross section) is designed and this causes 10 PBAR properties to change,
including four stress recovery points. These derived or spawned properties are
indicated by an asterisk in the final field of the table.
The upper and lower bounds that are printed in these tables correspond to the PMIN
and PMAX values provided on the DVPREL1 and DVCREL1 entries. The spawned
properties have default values applied to these bounds. The DIM2 property of the
PBARL and all of the material properties have bounds of N/ A (not applicable)
indicating that no bounds are imposed on these properties that are linear functions of
a single design variable.
If the DIFFERENCE FLAG has a FATAL value, this indicates that the analysis model
and design model differ by such a large degree that the design task will be stopped.
This should not happen in normal circumstances, but the discussion of the PTOL
parameter in the Comparison Between the Design and Analysis Properties portion
of the DOPTPRM on page 231 description shows how the allowable difference
between the analysis and designed property values can be made arbitrarily small.
Once the analysis and design models have been compared, Solution 200 proceeds with
an analysis of the model to gather baseline response data. A DMAP information
message from subDMAP FEA serves notice that a static analysis has been initiated:
Since this is the first analysis and the parameter NASPRT is at its default value of zero,
full data recovery is performed based on Case Control output requests. Since
displacements and stresses have been requested, we see the familiar MSC.Nastran
static analysis output (abbreviated here):
^^^ USER INFORMATION MESSAGE 9051 (FEA)
^^^ STATIC ANALYSIS INITIATED. DESIGN CYCLE NUMBER= 1
Main Index Main Index


332
This is followed by the DMAP information message:
If other analyses are performed (normal modes, dynamic response, etc.), similar
DMAP information messages and data recovery output will follow.
With the finite element analysis complete, the code performs a hard convergence
check when this is the second or greater design cycle. Since this is just the first cycle
though, DOM12 (the module that performs the convergence tests) simply reports on
the maximum constraint value:
Recall that these constraint values refer to the normalized constraints constructed
internally in MSC.Nastran. Since the maximum constraint value is positive, we can
characterize this initial design as infeasible. Furthermore, its value of 0.21914 indicates
a constraint violation of around 22%. We will see shortly how to determine which
constraint is responsible for this maximum value.
LOAD CONDITION 1 SUBCASE 1

D I S P L A C E M E N T V E C T O R

POINT ID. TYPE T1 T2 T3 R1 R2 R3
10000 G 0.0 0.0 0.0 -1.950617E-02 1.827125E-03 0.0
10001 G 1.318095E-03 -8.015184E-05 0.0 2.840014E-02 2.743307E-05 0.0
10002 G 2.886389E-03 -2.524653E-04 0.0 3.445658E-02 4.750498E-06 0.0
10003 G 5.111539E-03 -8.099531E-05 0.0 2.841208E-02 -3.201039E-05 0.0
10004 G 8.218050E-03 1.826257E-03 0.0 -1.948978E-02 -1.827407E-03 0.0
10100 G 0.0 -3.837221E-04 0.0 1.325681E-02 -1.977302E-02 0.0
.. .. ..
.. .. ..
.. .. ..
.. .. ..

LOAD CONDITION 2 SUBCASE 2

S T R E S S E S I N Q U A D R I L A T E R A L E L E M E N T S ( Q U A D 4 )
ELEMENT FIBER STRESSES IN ELEMENT COORD SYSTEM PRINCIPAL STRESSES (ZERO SHEAR)
ID. DISTANCE NORMAL-X NORMAL-Y SHEAR-XY ANGLE MAJOR MINOR VON MISES
1 -7.500000E-02 -1.891894E+03 3.042163E+01 3.228667E+02 80.7160 8.320031E+01 -1.944673E+03 1.987579E+03
7.500000E-02 -2.076745E+03 3.314972E+02 -7.038036E+02 -74.8469 5.220972E+02 -2.267345E+03 2.568504E+03
2 -7.500000E-02 -2.019023E+03 -2.120160E+02 -1.346214E+02 -85.7627 -2.020418E+02 -2.028997E+03 1.935900E+03
7.500000E-02 -1.100828E+03 5.068430E+02 -9.807557E+02 -64.6692 9.710898E+02 -1.565075E+03 2.216372E+03
3 -7.500000E-02 -5.511049E+02 -5.202870E+02 -3.804567E+02 -46.1596 -1.549273E+02 -9.164646E+02 8.496614E+02
7.500000E-02 -7.571406E+02 -6.938679E+01 -7.407589E+02 -57.4509 4.034217E+02 -1.229949E+03 1.473673E+03
4 -7.500000E-02 7.372604E+02 -1.338301E+02 -9.432302E+02 -32.6072 1.340649E+03 -7.372183E+02 1.824604E+03
7.500000E-02 -5.660520E+02 6.675811E+01 1.285017E+03 51.9163 1.073750E+03 -1.573044E+03 2.305746E+03
^^^ USER INFORMATION MESSAGE 9052 (FEA)
^^^ STATIC ANALYSIS COMPLETED. DESIGN CYCLE NUMBER= 1
THIS IS THE FIRST ANALYSIS - NO CONVERGENCE CHECK
-------------------------------------------------------
MAXIMUM VALUE OF CONSTRAINTS : 2.1913E-01
-------------------------------------------------------
Main Index Main Index
333 CHAPTER 6
Output Features and Interpretation
The next set of design related prints in the output is controlled by the P1 and P2
parameters just discussed in Output-Controlling Parameters on page 327. The
example has set P1=1 and P2=15 so that all available design data are printed. An
abbreviated version of these prints is presented here interspersed in the text. In this
case, the prints follow the initial analysis so there is no redesign information available
to print. Instead, initial data, as indicated by the following prints, is presented.
:
The DESIGN OBJECTIVE print provides attributes of the objective and its initial
value. This is followed by a table DESIGN VARIABLES that contains information
included on the DESVAR, including the initial values of the design variables. Two IDs
are provided each design variable: an internal ID that indicates the position of the
design variable in the design task and the user provided DESVAR ID. Dependent
design variables that are defined using DLINK entries always follow the independent
********************************************************************
* *
* *
* D E S I G N O P T I M I Z A T I O N *
* *
* *
********************************************************************
***********************************************
* *
* I N I T I A L A N A L Y S I S *
* *
***********************************************
******* ANALYSIS RESULTS BASED ON THE INITIAL DESIGN *******
----- DESIGN OBJECTIVE -----
----------------------------------------------------------------------------------------------------
INTERNAL TYPE MINIMIZE
RESPONSE OF OR SUPERELEMENT SUBCASE
ID RESPONSE LABEL MAXIMIZE ID ID VALUE
----------------------------------------------------------------------------------------------------
1 DRESP1 WEIGHT MINIMIZE 0 0 6.9618E+00
----- DESIGN VARIABLES -----
---------------------------------------------------------------------------------------------------------
INTERNAL DESVAR LOWER UPPER
ID ID LABEL BOUND VALUE BOUND
---------------------------------------------------------------------------------------------------------
1 1 T-PLATE 1.0000E-03 1.5000E-01 1.0000E+01
2 2 T-BOX 1.0000E-03 1.0000E+00 1.0000E+01
Main Index Main Index


334
design variables. The identity of these design variables is useful when attempting to
analyze the output from the optimizer algorithm and when interpreting unformatted
design sensitivity prints.
The designed properties are presented next and it is again seen that a distinction is
made in the type of property in that standard properties are presented in a table
separate from connectivity properties (and material properties if any are present in the
design). Also spawned beam properties are designated as having a SECPRO type of
property. The initial value of the property is bracketed by its lower and upper bounds.
Again, properties that are a linear function of a single design variable and have no
user designed limits do not have bounds.
The property prints are followed by constraint prints and there can be up to four types
of constraints printed. The first table shows the constraints on the design responses.
These are the constraints that are imposed on DRESP1/ 2/ 3 responses. The constraints
each have an internal ID and this is useful in analyzing the detailed optimization
prints discussed on page 343. The internal constraint ID is followed by a DCONSTR
----- DESIGNED PROPERTIES -----
----------------------------------------------------------------------------------------------------------
PROPERTY PROPERTY PROPERTY TYPE OF LOWER UPPER
TYPE ID NAME PROPERTY BOUND VALUE BOUND
----------------------------------------------------------------------------------------------------------
PSHELL 1 T DVPREL1 1.0000E-02 1.5000E-01 1.0000E+20
PBARL 3 DIM2 DVPREL1 N/A 1.0000E-01 N/A
PBAR 3 A SECPRO 1.0000E-15 9.6000E-01 1.0000E+20
PBAR 3 I1 SECPRO 1.0000E-15 1.2072E+00 1.0000E+20
PBAR 3 I2 SECPRO 1.0000E-15 9.8120E-01 1.0000E+20
PBAR 3 J SECPRO 1.0000E-15 3.2000E-03 1.0000E+20
PBAR 3 C1 SECPRO -1.0000E+35 1.5000E+00 1.0000E+20
PBAR 3 D1 SECPRO -1.0000E+35 -1.5000E+00 1.0000E+20
PBAR 3 E1 SECPRO -1.0000E+35 -1.5000E+00 1.0000E+20
PBAR 3 F1 SECPRO -1.0000E+35 1.5000E+00 1.0000E+20
PBAR 3 K1 SECPRO -1.0000E-03 5.8333E-01 1.0000E+20
PBAR 3 K2 SECPRO -1.0000E-03 3.7500E-01 1.0000E+20
----- DESIGNED CONNECTIVITY PROPERTIES ----
----------------------------------------------------------------------------------------------------------
ELEMENT ELEMENT PROPEETY TYPE OF LOWER UPPER
TYPE ID NAME PROPERTY BOUND VALUE BOUND
----------------------------------------------------------------------------------------------------------
CBAR 31 W3A DVCREL1 N/A 1.5750E+00 N/A
CBAR 32 W3A DVCREL1 N/A 1.5750E+00 N/A
CBAR 33 W3A DVCREL1 N/A 1.5750E+00 N/A
CBAR 34 W3A DVCREL1 N/A 1.5750E+00 N/A
CBAR 31 W3B DVCREL1 N/A 1.5750E+00 N/A
CBAR 32 W3B DVCREL1 N/A 1.5750E+00 N/A
CBAR 33 W3B DVCREL1 N/A 1.5750E+00 N/A
CBAR 34 W3B DVCREL1 N/A 1.5750E+00 N/A
Main Index Main Index
335 CHAPTER 6
Output Features and Interpretation
ID that was used in the definition of the constraint and then has the internal response
ID. This internal ID can be used to identify the response that resulted in the constraint,
as will be seen in a moment when we discuss how to trace the identity of the response
that results in the maximum constraint. The response type (RTYPE on the DRESP1
entry) is then given, as is a flag indicating whether the constraint arose from a lower
bound of upper bound limit. The region ID used in the screening process is followed
by the subcase ID and finally the constraint value.
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE
ID ID ID TYPE FLAG ID ID VALUE
---------------------------------------------------------------------------------------------------------
1 10 2 STRESS UPPER 2 1 -4.2961E-01
2 10 3 STRESS UPPER 2 1 -3.9654E-01
3 10 4 STRESS UPPER 2 1 -4.2961E-01
4 10 5 STRESS UPPER 2 1 -3.9654E-01
5 10 6 STRESS UPPER 2 1 -4.6636E-01
6 10 7 STRESS UPPER 2 1 -4.5379E-01
7 10 8 STRESS UPPER 2 1 -4.6636E-01
8 10 9 STRESS UPPER 2 1 -4.5379E-01
9 10 11 STRESS UPPER 1 2 -3.8670E-01
10 10 12 STRESS UPPER 1 2 2.1914E-01**
11 10 13 STRESS UPPER 1 2 -3.0252E-01
12 10 14 STRESS UPPER 1 2 7.9566E-02
13 30 10 DISP UPPER 14 2 -3.3261E-01
----- CONSTRAINTS ON DESIGNED PROPERTIES -----
-------------------------------------------------------------------------------------------------------
INTERNAL PROPERTY PROPERTY L/U CYCLE
ID ID NAME FLAG LIMIT VALUE
-------------------------------------------------------------------------------------------------------
14 1 T LOWER 7.5000E-02 -1.0000E+00
16 3 A LOWER 4.8000E-01 -1.0000E+00
17 3 I1 LOWER 6.0360E-01 -1.0000E+00
18 3 I2 LOWER 4.9060E-01 -1.0000E+00
19 3 J LOWER 1.0000E-15 -3.2000E+12
20 3 C1 LOWER 7.5000E-01 -1.0000E+00
21 3 D1 LOWER -2.2500E+00 -3.3333E-01
22 3 E1 LOWER -2.2500E+00 -3.3333E-01
23 3 F1 LOWER 7.5000E-01 -1.0000E+00
24 3 K1 LOWER 2.9167E-01 -1.0000E+00
25 3 K2 LOWER 1.8750E-01 -1.0000E+00
26 1 T UPPER 2.2500E-01 -3.3333E-01
28 3 A UPPER 1.4400E+00 -3.3333E-01
29 3 I1 UPPER 1.8108E+00 -3.3333E-01
30 3 I2 UPPER 1.4718E+00 -3.3333E-01
31 3 J UPPER 1.3200E-02 -7.5758E-01
32 3 C1 UPPER 2.2500E+00 -3.3333E-01
33 3 D1 UPPER -7.5000E-01 -1.0000E+00
34 3 E1 UPPER -7.5000E-01 -1.0000E+00
35 3 F1 UPPER 2.2500E+00 -3.3333E-01
36 3 K1 UPPER 8.7500E-01 -3.3333E-01
37 3 K2 UPPER 5.6250E-01 -3.3333E-01
----- CONSTRAINTS ON DESIGNED BEAM LIBRARY DIMENSIONS -----
---------------------------------------------------------------------------------------------------------
INTERNAL PBARL GROUP TYPE CONSTRAINT
ID ID NO. VALUE
---------------------------------------------------------------------------------------------------------
54 3 MSCBML0 HAT 1 -2.8000E+00
55 3 MSCBML0 HAT 2 -1.8000E+00
The maximum constraint is
associated with Internal
Response ID = 12
Main Index Main Index


336
The header for the DESIGN CONSTRAINT ON RESPONSES also includes the
notation: MAXIMUM RESPONSE CONSTRAINTS ARE MARKED WITH **. In this
case, we see that the twelfth internal constraint has a value of 0.21913 (the value given
from the hard convergence check above) and that the corresponding internal response
ID of this maximum constraint is also 12. To see which response this is, we skip ahead
to the table on page 338. It is seen that the twelfth internal response is a stress response
in a CBAR element 34 and that the stress occurs in the second subcase and is
component 7 (the maximum stress at End A of the element). It is also seen that the
violated constraint results from the stress being greater than the allowable value of
25,000.
Returning to the constraint printout, the next type of constraint that could be printed
is those resulting from constraints placed on dependent design variables. In this
example, there are no dependent design variables, so there is no print for this type of
constraint. There are constraints applied to the properties however and these occur
next in the printout. The internal constraint numbering continues from the response
constraints and the table then lists the property ID and name, an indication as to
whether the constraint comes from a lower or upper bound, the cycle limit and then
the constraint value. The cycle limit is the limit placed on the property for the current
approximate model and, as explained in the Move Limits on page 104, this limit is
computed using the current value of the property, the user applied limits and the
DELP and DPMIN parameters. A comparison of the list of constrained properties in
this printout with the designed properties of the previous fragment indicates that not
all designed properties are constrained. As the Move Limits on page 104 has
indicated, a property is not constrained when
1. it is a linear function of a single design variable, and
2. the user has not imposed any limits on the property.
These two criteria are satisfied for the DIM2 PBARL dimension and for all the
connectivity properties. The thickness of the PSHELL is a linear function of a single
design variable, but it has been limited to be greater than 0.01, hence the move limit
constraints.
The final constraint type is associated with the beam library. It is seen that there are
two of these constraints and that the are applied to the PBARL with PID=3 and that
the section type is a HAT. Referring to Constraints for Beam Library on page 118,
it is seen that the two constraints are to insure that twice the thickness (DIM2) never
Main Index Main Index
337 CHAPTER 6
Output Features and Interpretation
exceeds the height (DIM1) of the HAT or the center (DIM3) of the HAT. In this case,
the initial thickness is 0.1, the (undesigned) height is 3.0 and the width is 2.0, resulting
in constraint values of -2.8 and -1.8, respsectively.
Note that the maximum constraint that is marked with the ** is only for the constraints
that derive from the responses. In this case, this happens to be the maximum response
across all response types, but it is possible to have situations where the maximum
constraint comes from one of the other constraint types and it will not be flagged.
The final P1/ P2 associated prints are associated with the responses themselves. Each
response type has specialized formats to provide the information and it is not felt to
be helpful to describe each of the possible formats. Each of the responses does have an
internal ID and we have already seen how this can be used to link the constraints on
the responses to the responses themselves. The DRESP1 responses are sorted by
superelement, by subcase and by response type. If there there are DRESP2 responses,
they follow the DRESP1 responses for a given superelelement/ subcase and DRESP3
responses follow the DRESP2s.
Main Index Main Index


338
After the initial design information has been printed, two messages appear in the
output to indicate that a sensitivity analysis has been performed.
----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
----- WEIGHT RESPONSE -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ROW COLUMN LOWER UPPER
ID ID LABEL ID ID BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
1 15 W 3 3 N/A 6.9618E+00 N/A
I N I T I A L A N A L Y S I S S U B C A S E = 1
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
2 3 S13 2 9 N/A 1.4268E+04 2.5000E+04
3 3 S13 3 9 N/A 1.5089E+04 2.5000E+04
4 3 S13 14 9 N/A 1.4268E+04 2.5000E+04
5 3 S13 15 9 N/A 1.5089E+04 2.5000E+04
6 6 S16 2 17 N/A 1.3343E+04 2.5000E+04
7 6 S16 3 17 N/A 1.3658E+04 2.5000E+04
8 6 S16 14 17 N/A 1.3343E+04 2.5000E+04
9 6 S16 15 17 N/A 1.3658E+04 2.5000E+04
I N I T I A L A N A L Y S I S S U B C A S E = 2
----- DISPLACEMENT RESPONSES -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER UPPER
ID ID LABEL ID NO. BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
10 14 D2 10203 3 N/A 1.9938E-02 3.0000E-02
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
11 1 SBARA 33 7 N/A 1.5107E+04 2.5000E+04
STATIC ANALYSIS OF A STIFFENED PLATE DSOUG4 AUGUST 23, 2001 MSC.NASTRAN 4/ 9/01 PAGE 37

SUBCASE 1
I N I T I A L A N A L Y S I S S U B C A S E = 2
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
12 1 SBARA 34 7 N/A 3.0347E+04 2.5000E+04
13 2 SBARB 32 14 N/A 1.7291E+04 2.5000E+04
14 2 SBARB 33 14 N/A 2.6764E+04 2.5000E+04
This is the response that
results in the largest
constraint value.
{
^^^ USER INFORMATION MESSAGE 9051 (FEA)
^^^ STATIC SENSITIVITY ANALYSIS INITIATED. DESIGN CYCLE NUMBER= 1
^^^ USER INFORMATION MESSAGE 9052 (FEA)
^^^ STATIC SENSITIVITY ANALYSIS COMPLETED. DESIGN CYCLE NUMBER= 1
Main Index Main Index
339 CHAPTER 6
Output Features and Interpretation
Unless the DSAPRT Case Control command has been used to requests sensitivities for this
design cycle, no prints are produced during the sensitivity calculations. Sensitivity prints are
discussed in the next section: Design Sensitivity Output on page 364. After the sensitivity
analysis has been performed, all of the information necessary to perform an optimization is
available.
The parameter IPRINT, which can be set using the DOPTPRM entry, controls the amount of
diagnostic output from the optimizer. With the default zero value, no output is generated
during the numerical optimization phase. In this example, IPRINT has been set to 7. This is the
highest available value and has been used here to illustrate the available prints but is generally
to be avoided because it produces extensive print information, even for simple examples like the
one shown here. An abbreviated form of the output is presented and discussed here, primarily
using annotations within the output.
Main Index Main Index


340
DESIGN OPTIMIZATION TOOLS
(C) COPYRIGHT, 2000
VANDERPLAATS R&D
ALL RIGHTS RESERVED, WORLDWIDE
VERSION 5.3
CONTROL PARAMETERS
OPTIMIZATION METHOD, METHOD = 1
NUMBER OF DECISION VARIABLES, NDV = 2
NUMBER OF CONSTRAINTS, NCON = 37
PRINT CONTROL PARAMETER, IPRINT = 7
GRADIENT PARAMETER, IGRAD = 1
GRADIENTS ARE SUPPLIED BY THE USER
THE OBJECTIVE FUNCTION WILL BE MINIMIZED
-- SCALAR PROGRAM PARAMETERS
REAL PARAMETERS
1) CT = -3.00000E-02 8) DX2 = 2.00000E-01
2) CTMIN = 3.00000E-03 9) FDCH = 1.00000E-03
3) DABOBJ = 6.96180E-04 10) FDCHM = 1.00000E-04
4) DELOBJ = 1.00000E-03 11) RMVLMZ = 4.00000E-01
5) DOBJ1 = 1.00000E-01 12) DABSTR = 6.96180E-04
6) DOBJ2 = 1.39236E+00 13) DELSTR = 1.00000E-03
7) DX1 = 1.00000E-02
INTEGER PARAMETERS
1) IGRAD = 1 6) NGMAX = 4 11) IPRNT1 = 1
2) ISCAL = 2 7) IGMAX = 0 12) IPRNT2 = 2
3) ITMAX = 40 8) JTMAX = 20 13) JWRITE = 0
4) ITRMOP = 2 9) ITRMST = 2 14) MAXINT = 2147418112
5) IWRITE = 6 10) JPRINT = 0 15) NSTORE = 1754
STORAGE REQUIREMENTS
ARRAY DIMENSION MINIMUM DESIRED MAXIMUM USED
WK 3471501 383 400 2083 2083
IWK 147 147 147
-- INITIAL VARIABLES AND BOUNDS
LOWER BOUNDS ON THE DECISION VARIABLES (XL-VECTOR)
1) 7.50000E-02 5.00000E-01
DECISION VARIABLES (X-VECTOR)
1) 1.50000E-01 1.00000E+00
UPPER BOUNDS ON THE DECISION VARIABLES (XU-VECTOR)
1) 2.25000E-01 1.50000E+00
-- INITIAL FUNCTION VALUES
OBJ = 6.9618
The DOT algorithm is supplied by
Vanderplaats R& D.
Modified method of feasible directions
METHOD can be set on the DOPTPRM entry.
IGRAD = 1 indicates that gradients are
supplied to the optimizer. These are
MSC.Nastran Gradient Evaluations.
DOT can adjust its memory requirements
to make do with what is available.
Initial objective
Initial design variable data
Main Index Main Index
341 CHAPTER 6
Output Features and Interpretation
CONSTRAINT VALUES (G-VECTOR)
1) -4.29609E-01 -3.96550E-01 -4.29609E-01 -3.96550E-01 -4.66369E-01
6) -4.53803E-01 -4.66369E-01 -4.53803E-01 -3.86718E-01 2.19133E-01
11) -3.02530E-01 7.95523E-02 -3.32617E-01 -1.00000E+00 -1.00000E+00
16) -1.00000E+00 -1.00000E+00 -3.20000E+12 -1.00000E+00 -3.33333E-01
21) -3.33333E-01 -1.00000E+00 -1.00000E+00 -1.00000E+00 -3.33333E-01
26) -3.33333E-01 -3.33333E-01 -3.33333E-01 -7.57576E-01 -3.33333E-01
31) -1.00000E+00 -1.00000E+00 -3.33333E-01 -3.33333E-01 -3.33333E-01
36) -2.80000E+00 -1.80000E+00
***** WARNING - ONE OR MORE CONSTRAINT VALUES EXCEEDS 10.0 IN MAGNITUDE.
MAXIMUM MAGNITUDE = 3.20000E+12
IT IS SUGGESTED TO DIVIDE ALL G VALUES WITH AN ABSOLUTE VALUE
GREATER THAN 1.0 BY A CONSTANT OF THAT ORDER OF MAGNITUDE.
E.G. IF ABS(G(1)) = 70.3, DIVIDE G(1) BY 10.0 OR 100.0
THIS SHOULD IMPROVE EFFICIENCY AND RELIABILITY OF THE OPTIMIZATION.
-- BEGIN CONSTRAINED OPTIMIZATION: MFD METHOD
-- ITERATION NUMBER 1
SCALING INFORMATION
DESIGN VARIABLE SCALE FACTORS
1) 1.00000E+00 1.00000E+00
THERE ARE 0 ACTIVE CONSTRAINTS AND 2 VIOLATED CONSTRAINTS
CONSTRAINT NUMBERS
10 12
GRADIENT OF THE OBJECTIVE FUNCTION (DF-VECTOR)
1) 2.83103E+01 2.66084E+00
GRADIENT OF CONSTRAINT NUMBER 10
1) -1.18229E+00 -1.03914E+00
GRADIENT OF CONSTRAINT NUMBER 12
1) -1.26042E+00 -8.86792E-01
THERE ARE 0 ACTIVE SIDE CONSTRAINTS
K-T PARAMETERS, BETA = 1.00000E+00 MAX. RESIDUAL = 1.74248E+01
SEARCH DIRECTION (S-VECTOR)
1) 1.00000E+00 9.23963E-01
RETAINED ACTIVE/VIOLATED CONSTRAINT NUMBERS
10 12
DG(I)/DALPHA VECTOR
1) -2.14242E+00 -2.07978E+00
PROPOSED ALPHA = 7.50000E-02
* * * * * BEGIN ONE-DIMENSIONAL SEARCH * * * * *
Initial constraints. These are all the
constraints including property and beam
library constraints.
This message can usually be ignored when
the large constraint magnitude is negative.
Gradients of the objective and the constraints DOT
considers important. In this case, the two important
constraints are both violated.
One-dimensional search direction
Change in the important constraints
for a unit step in the search direction
Main Index Main Index


342
FIND BRACKETS ON MINIMUM
ALPHA = 7.50000E-02
DECISION VARIABLES (X-VECTOR)
1) 2.25000E-01 1.06930E+00
OBJ = 9.26919E+00
CONSTRAINT VALUES (G-VECTOR)
1) -7.67431E-01 -7.42637E-01 -7.67431E-01 -7.42637E-01 -7.94028E-01
6) -7.78704E-01 -7.94028E-01 -7.78704E-01 -4.62983E-01 9.26787E-02
11) -3.78145E-01 -4.09346E-02 -4.08364E-01 -2.00000E+00 -1.13551E+00
16) -1.12712E+00 -1.12773E+00 -3.90677E+12 -1.00279E+00 -3.34263E-01
21) -3.34263E-01 -1.00279E+00 -9.92978E-01 -9.95181E-01 0.00000E+00
26) -2.88164E-01 -2.90959E-01 -2.90757E-01 -7.04033E-01 -3.32404E-01
31) -9.97211E-01 -9.97211E-01 -3.32404E-01 -3.35674E-01 -3.34940E-01
36) -2.78614E+00 -1.78614E+00
ALPHA = 1.96353E-01
DECISION VARIABLES (X-VECTOR)
1) 2.25000E-01 1.18142E+00
OBJ = 9.56595E+00
CONSTRAINT VALUES (G-VECTOR)
1) -7.68584E-01 -7.43162E-01 -7.68584E-01 -7.43162E-01 -7.94459E-01
6) -7.79102E-01 -7.94459E-01 -7.79102E-01 -5.04827E-01 4.63437E-04
11) -4.30340E-01 -1.19625E-01 -4.57685E-01 -2.00000E+00 -1.35391E+00
16) -1.32979E+00 -1.33160E+00 -5.25680E+12 -1.00729E+00 -3.35763E-01
21) -3.35763E-01 -1.00729E+00 -9.81572E-01 -9.87353E-01 0.00000E+00
26) -2.15362E-01 -2.23403E-01 -2.22799E-01 -6.01758E-01 -3.30904E-01
31) -9.92711E-01 -9.92711E-01 -3.30904E-01 -3.39476E-01 -3.37549E-01
36) -2.76372E+00 -1.76372E+00
BRACKETS, XL = 0.00000E+00 XU = 1.96353E-01
ALPHA = 1.35676E-01
DECISION VARIABLES (X-VECTOR)
1) 2.25000E-01 1.12536E+00
OBJ = 9.41775E+00
CONSTRAINT VALUES (G-VECTOR)
1) -7.68037E-01 -7.42913E-01 -7.68037E-01 -7.42913E-01 -7.94254E-01
6) -7.78913E-01 -7.94254E-01 -7.78913E-01 -4.84949E-01 4.42723E-02
11) -4.05544E-01 -8.22420E-02 -4.34299E-01 -2.00000E+00 -1.24484E+00
16) -1.22892E+00 -1.23010E+00 -4.54871E+12 -1.00504E+00 -3.35013E-01
21) -3.35013E-01 -1.00504E+00 -9.87282E-01 -9.91272E-01 0.00000E+00
26) -2.51720E-01 -2.57026E-01 -2.56635E-01 -6.55400E-01 -3.31653E-01
31) -9.94960E-01 -9.94960E-01 -3.31653E-01 -3.37573E-01 -3.36243E-01
36) -2.77493E+00 -1.77493E+00
--- RESULTS OF ONE-DIMENSIONAL SEARCH
ALPHA = 1.96353E-01
OBJ = 9.56595E+00
DECISION VARIABLES (X-VECTOR)
1) 2.25000E-01 1.18142E+00
CONSTRAINT VALUES (G-VECTOR)
1) -7.68584E-01 -7.43162E-01 -7.68584E-01 -7.43162E-01 -7.94459E-01
6) -7.79102E-01 -7.94459E-01 -7.79102E-01 -5.04827E-01 4.63437E-04
11) -4.30340E-01 -1.19625E-01 -4.57685E-01 -2.00000E+00 -1.35391E+00
16) -1.32979E+00 -1.33160E+00 -5.25680E+12 -1.00729E+00 -3.35763E-01
21) -3.35763E-01 -1.00729E+00 -9.81572E-01 -9.87353E-01 0.00000E+00
26) -2.15362E-01 -2.23403E-01 -2.22799E-01 -6.01758E-01 -3.30904E-01
31) -9.92711E-01 -9.92711E-01 -3.30904E-01 -3.39476E-01 -3.37549E-01
36) -2.76372E+00 -1.76372E+00
* * * * * END OF ONE-DIMENSIONAL SEARCH * * * * *
CALCULATED ALPHA = 1.96353E-01
Steps in the 1-D search
1-D search results
Main Index Main Index
343 CHAPTER 6
Output Features and Interpretation
MANY LINES HAVE BEEN DELETED FOR CLARITY
-- OPTIMIZATION IS COMPLETE
NUMBER OF ITERATIONS = 6
CONSTRAINT TOLERANCE, CT = -7.50000E-03
THERE ARE 3 ACTIVE CONSTRAINTS AND 0 VIOLATED CONSTRAINTS
CONSTRAINT NUMBERS
2 4 10
THERE ARE 0 ACTIVE SIDE CONSTRAINTS
TERMINATION CRITERIA
MAXIMUM K-T RESIDUAL = 0.00000E+00 IS LESS THAN 1.00000E-03
MAXIMUM POSSIBLE MOVE = 0.00000E+00 IS LESS THAN 1.00000E-05
-- OPTIMIZATION RESULTS
OBJECTIVE, F(X) = 6.80534E+00
DECISION VARIABLES, X
ID XL X XU
1 7.50000E-02 1.08386E-01 2.25000E-01
2 5.00000E-01 1.38715E+00 1.50000E+00
CONSTRAINTS, G(X)
1) -4.50848E-02 -1.17188E-06 -4.50848E-02 -1.17188E-06 -9.06751E-02
6) -8.11571E-02 -9.06751E-02 -8.11571E-02 -4.65630E-01 -2.73055E-03
11) -4.23451E-01 -9.52752E-02 -4.41598E-01 -4.45142E-01 -1.75192E+00
16) -1.69206E+00 -1.69687E+00 -8.47233E+12 -1.01551E+00 -3.38503E-01
21) -3.38503E-01 -1.01551E+00 -9.60506E-01 -9.72896E-01 -5.18286E-01
26) -8.26924E-02 -1.02647E-01 -1.01045E-01 -3.58157E-01 -3.28163E-01
31) -9.84490E-01 -9.84490E-01 -3.28163E-01 -3.46498E-01 -3.42368E-01
36) -2.72257E+00 -1.72257E+00
FUNCTION CALLS = 52
GRADIENT CALLS = 5
******************************************
***** CPU Statistics for Module DOM9 *****
******************************************
Pre-processing = 9.999990E-03 seconds
Function evaluation = 9.999990E-03 seconds
Gradient evaluation = 0.000000E+00 seconds
Optimization - DOT = 1.099999E-01 seconds
post-processing = 9.999990E-03 seconds
---------------------------------------------
Total = 1.399999E-01 seconds
Final results for
the approximate
optimization
The optimization required 52 function evaluations
and five gradient evaluations. These are performed
with the approximate model.
Timing data to see where the time
has gone
The approximation task is complete
Main Index Main Index


344
The above annotated listing provides insight into the types of prints that are available,
but the actual prints you will see for the IPRINT parameter will be a function of the
level of print requested and the METHOD selected on the DOPTPRM Bulk Data entry.
The IPRINT data can be summarized to say that the approximate optimization task
has resulted in a design whose objective decreased from 6.9618 to 6.8053 while the
maximum constraint values decreased from 0.21914 (violated) to -1.1719E-4
(satisfied).
These conclusions can be seen more readily in the summary output for the first design
cycle, which is also produced by the P1 = 1, P2 = 15 parameters on the DOPTPRM
entry. This is the same type of output as we discussed starting on page 333 except that
now we have data available from the input and the output of the approximate
optimization problem. The input values are identical to the initial values for the
design and the tables of data have the same number of rows so only an abbreviated
set of these tables are shown here. This summary begins with the identification of the
design cycle number and this is followed by listing of the design objective and the
design variables:
********************************************************************
* *
* *
* D E S I G N O P T I M I Z A T I O N *
* *
* *
********************************************************************
*******************************************
* *
* D E S I G N C Y C L E 1 *
* *
*******************************************
***** OPTIMIZATION RESULTS BASED ON THE APPROXIMATE MODEL *****
----- DESIGN OBJECTIVE -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL TYPE MINIMIZE
RESPONSE OF OR SUPERELEMENT SUBCASE INPUT OUTPUT
ID RESPONSE LABEL MAXIMIZE ID ID VALUE VALUE
---------------------------------------------------------------------------------------------------------------------------
1 DRESP1 WEIGHT MINIMIZE 0 0 6.9618E+00 6.8053E+00
----- DESIGN VARIABLES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DESVAR LOWER INPUT OUTPUT UPPER
ID ID LABEL BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
1 1 T-PLATE 1.0000E-03 1.5000E-01 1.0839E-01 1.0000E+01
2 2 T-BOX 1.0000E-03 1.0000E+00 1.3871E+00 1.0000E+01
Main Index Main Index
345 CHAPTER 6
Output Features and Interpretation
Note that the first design variable has decreased while the second has increased.
The listing of the Designed Properties shows the corresponding changes in the
properties and this is particularly useful for the beam properties where the thickness
change in the HAT cross-section produces changes in the properties that are not a
linear function of the design variable.
----- DESIGNED PROPERTIES -----
---------------------------------------------------------------------------------------------------------------------------
PROPERTY PROPERTY PROPERTY TYPE OF LOWER INPUT OUTPUT UPPER
TYPE ID NAME PROPERTY BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
PSHELL 1 T DVPREL1 1.0000E-02 1.5000E-01 1.0839E-01 1.0000E+20
PBARL 3 DIM2 DVPREL1 N/A 1.0000E-01 1.3871E-01 N/A
PBAR 3 A SECPRO 1.0000E-15 9.6000E-01 1.3209E+00 1.0000E+20
PBAR 3 I1 SECPRO 1.0000E-15 1.2072E+00 1.6249E+00 1.0000E+20
PBAR 3 I2 SECPRO 1.0000E-15 9.8120E-01 1.3231E+00 1.0000E+20
PBAR 3 J SECPRO 1.0000E-15 3.2000E-03 8.4723E-03 1.0000E+20
PBAR 3 C1 SECPRO -1.0000E+35 1.5000E+00 1.5116E+00 1.0000E+20
PBAR 3 D1 SECPRO -1.0000E+35 -1.5000E+00 -1.4884E+00 1.0000E+20
PBAR 3 E1 SECPRO -1.0000E+35 -1.5000E+00 -1.4884E+00 1.0000E+20
PBAR 3 F1 SECPRO -1.0000E+35 1.5000E+00 1.5116E+00 1.0000E+20
PBAR 3 K1 SECPRO -1.0000E-03 5.8333E-01 5.7181E-01 1.0000E+20
PBAR 3 K2 SECPRO -1.0000E-03 3.7500E-01 3.6992E-01 1.0000E+20
----- DESIGNED CONNECTIVITY PROPERTIES ----
---------------------------------------------------------------------------------------------------------------------------
ELEMENT ELEMENT PROPERTY TYPE OF LOWER INPUT OUTPUT UPPER
TYPE ID NAME PROPERTY BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
CBAR 31 W3A DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 32 W3A DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 33 W3A DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 34 W3A DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 31 W3B DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 32 W3B DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 33 W3B DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
CBAR 34 W3B DVCREL1 N/A 1.5750E+00 1.5542E+00 N/A
Main Index Main Index


346
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE INPUT OUTPUT
ID ID ID TYPE FLAG ID ID VALUE VALUE
---------------------------------------------------------------------------------------------------------------------------
1 10 2 STRESS UPPER 2 1 -4.2961E-01 -4.5085E-02
2 10 3 STRESS UPPER 2 1 -3.9655E-01 -1.1719E-06**
3 10 4 STRESS UPPER 2 1 -4.2961E-01 -4.5085E-02
4 10 5 STRESS UPPER 2 1 -3.9655E-01 -1.1719E-06**
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE INPUT OUTPUT
ID ID ID TYPE FLAG ID ID VALUE VALUE
---------------------------------------------------------------------------------------------------------------------------
5 10 6 STRESS UPPER 2 1 -4.6637E-01 -9.0675E-02
6 10 7 STRESS UPPER 2 1 -4.5380E-01 -8.1157E-02
7 10 8 STRESS UPPER 2 1 -4.6637E-01 -9.0675E-02
8 10 9 STRESS UPPER 2 1 -4.5380E-01 -8.1157E-02
9 10 11 STRESS UPPER 1 2 -3.8672E-01 -4.6563E-01
10 10 12 STRESS UPPER 1 2 2.1913E-01** -2.7305E-03
11 10 13 STRESS UPPER 1 2 -3.0253E-01 -4.2345E-01
12 10 14 STRESS UPPER 1 2 7.9552E-02 -9.5275E-02
13 30 10 DISP UPPER 14 2 -3.3262E-01 -4.4160E-01
----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
----- WEIGHT RESPONSE -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ROW COLUMN LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ID BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
1 15 W 3 3 N/A 6.9618E+00 6.8053E+00 N/A
D E S I G N C Y C L E = 1 S U B C A S E = 1
----- STRESS RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
2 3 S13 2 9 N/A 1.4260E+04 2.3873E+04 2.5000E+04
3 3 S13 3 9 N/A 1.5086E+04 2.5000E+04 2.5000E+04
4 3 S13 14 9 N/A 1.4260E+04 2.3873E+04 2.5000E+04
5 3 S13 15 9 N/A 1.5086E+04 2.5000E+04 2.5000E+04
6 6 S16 2 17 N/A 1.3341E+04 2.2733E+04 2.5000E+04
7 6 S16 3 17 N/A 1.3655E+04 2.2971E+04 2.5000E+04
8 6 S16 14 17 N/A 1.3341E+04 2.2733E+04 2.5000E+04
9 6 S16 15 17 N/A 1.3655E+04 2.2971E+04 2.5000E+04
D E S I G N C Y C L E = 1 S U B C A S E = 2
----- DISPLACEMENT RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
10 14 D2 10203 3 N/A 2.0021E-02 1.6752E-02 3.0000E-02
----- STRESS RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
11 1 SBARA 33 7 N/A 1.5332E+04 1.3359E+04 2.5000E+04
12 1 SBARA 34 7 N/A 3.0478E+04 2.4932E+04 2.5000E+04
13 2 SBARB 32 14 N/A 1.7437E+04 1.4414E+04 2.5000E+04
14 2 SBARB 33 14 N/A 2.6989E+04 2.2618E+04 2.5000E+04
Main Index Main Index
347 CHAPTER 6
Output Features and Interpretation
The listing of Constraints on Responses shows the decrease in the maximum
constraint value and that none of the other constraint became violated from the
redesign.
The listing of the Design Responses also shows the input and output values; note that
the responses that initially exceed their bounds on input are now at their limits. It is
important to note that the constraints and responses in the preceding output are only
approximate, because they are based on the approximate model (see Type-1
Response Evaluation on page 111). The optimizer has been able to achieve a
reduction in the peak approximate stresses. These and all other necessary structural
responses will be updated again on the next design cycle with a full finite element
analysis, and the process will be repeated again. The output Design Variable and
Designed Property values are exact and are not changed by the finite element analysis.
Once the approximate optimization is complete for this cycle, a soft convergence test
is performed, and the results are reported in the following table:
It is interesting to note that while the design variables only changed by 39%, there was
a 165% change in one of the properties. This again indicates the nonlinear nature of the
relationship between beam properties and the design variables and could be an
indication that too much change is being allowed in the property values. (The
allowable change is controlled, in part, by the DELP parameter which has been set to
0.5 in this case. The property change exceeded the 50% allowable change because the
the DPMIN parameter came into play to allow more change in the torsional stiffness,
J, property that has a small initial value. See Move Limits on page 104.
Even if soft convergence had been achieved (which it was not), this test will not
terminate the design cycles unless the Bulk Data parameter SOFTEXIT had been set to
YES (NO is the default).
A finite element analysis begins the new design cycle as is apparent from the
following DMAP information messages:

****************************************************************************************
INSPECTION OF CONVERGENCE DATA FOR THE OPTIMAL DESIGN WITH RESPECT TO APPROXIMATE MODELS
(SOFT CONVERGENCE DECISION LOGIC)
****************************************************************************************
MAXIMUM OF RELATIVE PROP. CHANGES 1.6476E+00 MUST BE LESS THAN 1.0000E-03
--- AND ---
MAXIMUM OF RELATIVE D.V. CHANGES 3.8715E-01 MUST BE LESS THAN 1.0000E-03
****************************************************************************************
^^^ USER INFORMATION MESSAGE 9051 (FEA)
^^^ STATIC ANALYSIS INITIATED. DESIGN CYCLE NUMBER= 2
Main Index Main Index


348
With two successive finite element analyses available (one from the current design
cycle and one from the previous), a hard convergence test can now be performed.
Even though convergence has not yet been achieved, the test results are reported
nonetheless.

Note that convergence has failed on all counts. The objective has changed more than
is permitted, the maximum constraint (based on the exact analysis) is violated and the
design variables and properties have changed significantly. This is not surprising for
the first redesign task.
Insight into the validity of the approximate model can be obtained by presenting some
of the design optimization results from the second design cycle.
CONVERGENCE NOT ACHIEVED YET (HARD CONVERGENCE DECISION LOGIC)
----------------------------------------------------------------------------------------------
RELATIVE CHANGE IN OBJECTIVE : 2.2447E-02 MUST BE LESS THAN 1.0000E-03
OR ABSOLUTE CHANGE IN OBJECTIVE : 1.5627E-01 MUST BE LESS THAN 1.0000E-20
--- AND ---
MAXIMUM CONSTRAINT VALUE : 7.2485E-02 MUST BE LESS THAN 5.0000E-03
(CONVERGENCE TO A FEASIBLE DESIGN)
--- OR ---
MAXIMUM OF RELATIVE PROP. CHANGES : 1.6476E+00 MUST BE LESS THAN 1.0000E-03
AND MAXIMUM OF RELATIVE D.V. CHANGES : 3.8715E-01 MUST BE LESS THAN 1.0000E-03
(CONVERGENCE TO A BEST COMPROMISE INFEASIBLE DESIGN)
----------------------------------------------------------------------------------------------
Main Index Main Index
349 CHAPTER 6
Output Features and Interpretation
********************************************************************
* *
* *
* D E S I G N O P T I M I Z A T I O N *
* *
* *
********************************************************************
*******************************************
* *
* D E S I G N C Y C L E 2 *
* *
*******************************************
***** OPTIMIZATION RESULTS BASED ON THE APPROXIMATE MODEL *****
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE INPUT OUTPUT
ID ID ID TYPE FLAG ID ID VALUE VALUE
---------------------------------------------------------------------------------------------------------------------------
5 10 7 STRESS UPPER 2 1 7.2485E-02** 2.5982E-03**
6 10 8 STRESS UPPER 2 1 -3.9151E-01 -4.2190E-01
7 10 9 STRESS UPPER 2 1 -4.2905E-01 -4.6044E-01
8 10 10 STRESS UPPER 2 1 -1.6393E-02 -8.4054E-02
9 10 11 STRESS UPPER 2 1 -8.9584E-03 -7.5725E-02
10 10 12 STRESS UPPER 2 1 -3.1821E-01 -3.5271E-01
11 10 13 STRESS UPPER 2 1 -4.6496E-01 -4.9209E-01
12 10 14 STRESS UPPER 2 1 -3.9258E-01 -4.2968E-01
13 10 15 STRESS UPPER 2 1 -4.6496E-01 -4.9209E-01
14 10 16 STRESS UPPER 2 1 -3.9258E-01 -4.2968E-01
15 10 17 STRESS UPPER 2 1 -4.2905E-01 -4.6044E-01
16 10 18 STRESS UPPER 2 1 -1.6393E-02 -8.4054E-02
17 10 19 STRESS UPPER 2 1 -8.9584E-03 -7.5725E-02
18 10 20 STRESS UPPER 2 1 -3.1821E-01 -3.5271E-01
19 20 2 DISP UPPER 13 1 3.3911E-02 -7.5347E-02
20 10 22 STRESS UPPER 1 2 -4.8499E-01 -4.6741E-01
21 10 23 STRESS UPPER 1 2 -4.2154E-02 -1.4844E-06
22 10 24 STRESS UPPER 1 2 -4.4244E-01 -4.1976E-01
23 10 25 STRESS UPPER 1 2 -1.2974E-01 -9.4884E-02
24 30 21 DISP UPPER 14 2 -4.6866E-01 -4.4608E-01
Main Index Main Index


350
A review of the DESIGN CONSTRAINTS ON RESPONSES shows a constraint with a
positive value when the approximate model results for the first design cycle indicated
that all constraints have been satisfied. The maximum constraint violation from the
exact analysis is only 7.2% so the approxmate values were not that far off. Further
comparisons of the RESPONSES IN THE DESIGN MODEL from the two design cycles
gives further information on the validity of the approximate model. The output
----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
----- WEIGHT RESPONSE -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ROW COLUMN LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ID BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
1 15 W 3 3 N/A 6.8055E+00 6.7267E+00 N/A
D E S I G N C Y C L E = 2 S U B C A S E = 1
----- DISPLACEMENT RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
2 13 D1 10302 3 N/A 1.0339E-01 9.2465E-02 1.0000E-01
----- STRESS RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
3 3 S13 2 9 N/A 2.5650E+04 2.3953E+04 2.5000E+04
4 3 S13 3 9 N/A 2.6812E+04 2.5065E+04 2.5000E+04
5 3 S13 4 9 N/A 1.5212E+04 1.4452E+04 2.5000E+04
6 3 S13 14 9 N/A 2.5650E+04 2.3953E+04 2.5000E+04
7 3 S13 15 9 N/A 2.6812E+04 2.5065E+04 2.5000E+04
8 3 S13 16 9 N/A 1.5212E+04 1.4452E+04 2.5000E+04
9 6 S16 1 17 N/A 1.4274E+04 1.3489E+04 2.5000E+04
10 6 S16 2 17 N/A 2.4590E+04 2.2899E+04 2.5000E+04
11 6 S16 3 17 N/A 2.4776E+04 2.3107E+04 2.5000E+04
12 6 S16 4 17 N/A 1.7045E+04 1.6182E+04 2.5000E+04
13 6 S16 5 17 N/A 1.3376E+04 1.2698E+04 2.5000E+04
14 6 S16 8 17 N/A 1.5186E+04 1.4258E+04 2.5000E+04
15 6 S16 9 17 N/A 1.3376E+04 1.2698E+04 2.5000E+04
16 6 S16 12 17 N/A 1.5186E+04 1.4258E+04 2.5000E+04
17 6 S16 13 17 N/A 1.4274E+04 1.3489E+04 2.5000E+04
18 6 S16 14 17 N/A 2.4590E+04 2.2899E+04 2.5000E+04
19 6 S16 15 17 N/A 2.4776E+04 2.3107E+04 2.5000E+04
20 6 S16 16 17 N/A 1.7045E+04 1.6182E+04 2.5000E+04
0
D E S I G N C Y C L E = 2 S U B C A S E = 2
----- DISPLACEMENT RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
21 14 D2 10203 3 N/A 1.5940E-02 1.6618E-02 3.0000E-02
----- STRESS RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
22 1 SBARA 33 7 N/A 1.2875E+04 1.3315E+04 2.5000E+04
23 1 SBARA 34 7 N/A 2.3946E+04 2.5000E+04 2.5000E+04
24 2 SBARB 32 14 N/A 1.3939E+04 1.4506E+04 2.5000E+04
25 2 SBARB 33 14 N/A 2.1757E+04 2.2628E+04 2.5000E+04
Main Index Main Index
351 CHAPTER 6
Output Features and Interpretation
WEIGHT response from the first design cycle 6.8053) is nearly identical with the input
for the second design cycle (6.8055). This should not be too surprising since the weight
is a linear function of the design variables in this case so there should not be any
approximation in this response. The grid and element responses differ and it is seen,
in fact, that the set of retained responses differ somewhat for the two design cycles. A
response that is retained in both design cycles is the von Mises stress at the top center
(item code 9) of CQUAD4 element with ID=15. For the first design cycle, this is the
fifth internal response and the output value is 25,000. At the second design cycle, this
is the seventh internal response and the input value is 26,812. This is a fairly significant
discrepancy, but recall that the thickness of the plate at which the stress is being
measured has been changed from 0.15000 to 0.10832.
This is an appropriate place to show the use of DOPTPRM parameters that can affect
the prints that occur based on the P1 and P2 parameters. For this simple case, the
amount of print is modest, but one can imagine larger models could produce an
overwhelming amount of printout. Suppose the DOPTPRM entry at the end of
Listing 7-8 is replaced with:
The P2CR parameter is now requesting that only the five most critical constraints on
response quantities are to be printed while the P2RSET = 10 points to a SET1 entry that
indicates that only retained responses for DRESP1 IDs 1 and 2 are to be printed. The
results from the two parameters are shown:
$
$...Optional override of optimization parameters:
$
DOPTPRM IPRINT 7 DESMAX 20 DELP 0.5 P1 1
P2 15 P2CR 5 PRSET 10
SET1 10 1 2
Main Index Main Index


352
As requested, the five largest response constraints are printed and they have been
sorted based on the input (to the approximate optimization task) constraint values
with the most critical occurring first. For the design responses, the four retained
responses associated with DRESP1 = 1 or 2 are printed.
In the case of test problem DSOUG4, the design cycle proceeds for three design cycles
when the following soft convergence results are presented:
Note that neither of the soft convergence criteria are satsified in this case. However,
the following hard convergence check indicates the problem has converged.
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE INPUT OUTPUT
ID ID ID TYPE FLAG ID ID VALUE VALUE
---------------------------------------------------------------------------------------------------------------------------
2 10 4 STRESS UPPER 2 1 6.9280E-02** 2.3837E-03**
5 10 7 STRESS UPPER 2 1 6.9280E-02** 2.3837E-03**
19 20 2 DISP UPPER 13 1 2.8612E-02 -7.5818E-02
1 10 3 STRESS UPPER 2 1 2.3290E-02 -4.1707E-02
4 10 6 STRESS UPPER 2 1 2.3290E-02 -4.1707E-02
.......
.......
----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
D E S I G N C Y C L E = 2 S U B C A S E = 2
----- STRESS RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER INPUT OUTPUT UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
22 1 SBARA 33 7 N/A 1.2822E+04 1.3198E+04 2.5000E+04
23 1 SBARA 34 7 N/A 2.4068E+04 2.5000E+04 2.5000E+04
24 2 SBARB 32 14 N/A 1.3950E+04 1.4446E+04 2.5000E+04
25 2 SBARB 33 14 N/A 2.1785E+04 2.2547E+04 2.5000E+04
****************************************************************************************
INSPECTION OF CONVERGENCE DATA FOR THE OPTIMAL DESIGN WITH RESPECT TO APPROXIMATE MODELS
(SOFT CONVERGENCE DECISION LOGIC)
****************************************************************************************
MAXIMUM OF RELATIVE PROP. CHANGES 2.1100E-03 MUST BE LESS THAN 1.0000E-03
--- AND ---
MAXIMUM OF RELATIVE D.V. CHANGES 2.1100E-03 MUST BE LESS THAN 1.0000E-03
****************************************************************************************
Main Index Main Index
353 CHAPTER 6
Output Features and Interpretation
In this case, the relative change in the objective is less than the criterion and the
maximum constraint value is also less than the criterion so convergence has been
achieved. The absolute change in the objective did not meet the strenuous criterion,
but only one of the objective criteria needs to be satisfied. In this case, the changes in
the design variables and the designed properties are greater than the criteria, but it
was not necessary to invoke these criteria in this case.
Following convergence, additional data recovery may occur and final design
optimization data, as indicated in the following fragments, are printed as governed by
the P2 parameter. In this case, it is necessary to only print the value from the final
analysis since that is the only new information that is available (approximate model
results have already been printed).
***** NORMAL CONVERGENCE CRITERIA SATISFIED ***** (HARD CONVERGENCE DECISION LOGIC)
**************************************************************************************
CONVERGENCE ACHIEVED BASED ON THE FOLLOWING CRITERIA
(HARD CONVERGENCE DECISION LOGIC)
RELATIVE CHANGE IN OBJECTIVE 9.9732E-04 MUST BE LESS THAN 1.0000E-03
OR ABSOLUTE CHANGE IN OBJECTIVE 6.7086E-03 MUST BE LESS THAN 1.0000E-20
--- AND ---
MAXIMUM CONSTRAINT VALUE 1.2891E-05 MUST BE LESS THAN 5.0000E-03
(CONVERGENCE TO A FEASIBLE DESIGN)
--- OR ---
MAXIMUM OF RELATIVE PROP. CHANGES 2.1100E-03 MUST BE LESS THAN 1.0000E-03
AND MAXIMUM OF RELATIVE D.V. CHANGES 2.1100E-03 MUST BE LESS THAN 1.0000E-03
(CONVERGENCE TO A BEST COMPROMISE INFEASIBLE DESIGN)
**************************************************************************************
Main Index Main Index


354
********************************************************************
* *
* *
* D E S I G N O P T I M I Z A T I O N *
* *
* *
********************************************************************
*******************************************
* *
* F I N A L A N A L Y S I S *
* *
*******************************************
******* ANALYSIS RESULTS BASED ON THE FINAL DESIGN *******
----- DESIGN OBJECTIVE -----
----------------------------------------------------------------------------------------------------
INTERNAL TYPE MINIMIZE
RESPONSE OF OR SUPERELEMENT SUBCASE
ID RESPONSE LABEL MAXIMIZE ID ID VALUE
----------------------------------------------------------------------------------------------------
1 DRESP1 WEIGHT MINIMIZE 0 0 6.7333E+00
----- DESIGN VARIABLES -----
---------------------------------------------------------------------------------------------------------
INTERNAL DESVAR LOWER UPPER
ID ID LABEL BOUND VALUE BOUND
---------------------------------------------------------------------------------------------------------
1 1 T-PLATE 1.0000E-03 1.1264E-01 1.0000E+01
2 2 T-BOX 1.0000E-03 1.3136E+00 1.0000E+01
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE
ID ID ID TYPE FLAG ID ID VALUE
---------------------------------------------------------------------------------------------------------
1 10 3 STRESS UPPER 2 1 -4.4438E-02
2 10 4 STRESS UPPER 2 1 1.2891E-05**
3 10 5 STRESS UPPER 2 1 -4.2312E-01
4 10 6 STRESS UPPER 2 1 -4.4438E-02
5 10 7 STRESS UPPER 2 1 1.2891E-05**
6 10 8 STRESS UPPER 2 1 -4.2312E-01
7 10 9 STRESS UPPER 2 1 -4.6178E-01
8 10 10 STRESS UPPER 2 1 -8.6493E-02
9 10 11 STRESS UPPER 2 1 -7.8124E-02
10 10 12 STRESS UPPER 2 1 -3.5420E-01
11 10 13 STRESS UPPER 2 1 -4.9346E-01
12 10 14 STRESS UPPER 2 1 -4.3102E-01
13 10 15 STRESS UPPER 2 1 -4.9346E-01
14 10 16 STRESS UPPER 2 1 -4.3102E-01
15 10 17 STRESS UPPER 2 1 -4.6178E-01
16 10 18 STRESS UPPER 2 1 -8.6493E-02
17 10 19 STRESS UPPER 2 1 -7.8124E-02
18 10 20 STRESS UPPER 2 1 -3.5420E-01
19 20 2 DISP UPPER 13 1 -7.7224E-02
20 10 22 STRESS UPPER 1 2 -4.6803E-01
21 10 23 STRESS UPPER 1 2 -5.8016E-04
22 10 24 STRESS UPPER 1 2 -4.2017E-01
23 10 25 STRESS UPPER 1 2 -9.5600E-02
24 30 21 DISP UPPER 14 2 -4.4653E-01
Main Index Main Index
355 CHAPTER 6
Output Features and Interpretation

----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
----- WEIGHT RESPONSE -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ROW COLUMN LOWER UPPER
ID ID LABEL ID ID BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
1 15 W 3 3 N/A 6.7333E+00 N/A
F I N A L A N A L Y S I S S U B C A S E = 1
----- DISPLACEMENT RESPONSES -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER UPPER
ID ID LABEL ID NO. BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
2 13 D1 10302 3 N/A 9.2278E-02 1.0000E-01
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
3 3 S13 2 9 N/A 2.3889E+04 2.5000E+04
4 3 S13 3 9 N/A 2.5000E+04 2.5000E+04
5 3 S13 4 9 N/A 1.4422E+04 2.5000E+04
6 3 S13 14 9 N/A 2.3889E+04 2.5000E+04
7 3 S13 15 9 N/A 2.5000E+04 2.5000E+04
8 3 S13 16 9 N/A 1.4422E+04 2.5000E+04
9 6 S16 1 17 N/A 1.3456E+04 2.5000E+04
10 6 S16 2 17 N/A 2.2838E+04 2.5000E+04
11 6 S16 3 17 N/A 2.3047E+04 2.5000E+04
12 6 S16 4 17 N/A 1.6145E+04 2.5000E+04
13 6 S16 5 17 N/A 1.2663E+04 2.5000E+04
14 6 S16 8 17 N/A 1.4224E+04 2.5000E+04
15 6 S16 9 17 N/A 1.2663E+04 2.5000E+04
16 6 S16 12 17 N/A 1.4224E+04 2.5000E+04
17 6 S16 13 17 N/A 1.3456E+04 2.5000E+04
18 6 S16 14 17 N/A 2.2838E+04 2.5000E+04
19 6 S16 15 17 N/A 2.3047E+04 2.5000E+04
20 6 S16 16 17 N/A 1.6145E+04 2.5000E+04
----- DISPLACEMENT RESPONSES -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER UPPER
ID ID LABEL ID NO. BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
21 14 D2 10203 3 N/A 1.6604E-02 3.0000E-02
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
22 1 SBARA 33 7 N/A 1.3299E+04 2.5000E+04
23 1 SBARA 34 7 N/A 2.4985E+04 2.5000E+04
24 2 SBARB 32 14 N/A 1.4496E+04 2.5000E+04
25 2 SBARB 33 14 N/A 2.2610E+04 2.5000E+04
Main Index Main Index


356
After this final design cycle print, the summary of the design cycle history is given:
From this summary we see that an optimal design was found after 3 design cycles and
4 finite element analyses (initial analysis plus one from each design cycle.) As was also
seen in the hard convergence output, the final design is feasible.
The objective function history traces the progress made by the optimizer during
successive optimizations with respect to approximate models. The fractional error of
approximation is a measure of the error in the approximated objective function versus
the true objective. The true objective is computed from the analysis at the beginning
of the next design cycle. Here, the error is negligible since the weight objective is a
linear function of the design variables. For objectives based on nonlinear structural
responses (virtually any response other than weight or volume), the error may be
more significant. If these approximation errors are large, tighter move limits may be
warranted.
Note also that, as a measure of approximation quality, the fractional error of
approximation is incomplete at best. Even though the error in the objective function
is small, errors in constraints can be much greater.
0
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 4
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 3
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 6.961800E+00 2.191328E-01
1 6.805342E+00 6.805526E+00 -2.697548E-05 7.248523E-02
2 6.726673E+00 6.726640E+00 4.962151E-06 3.852422E-03
3 6.733349E+00 6.733348E+00 1.416345E-07 1.289062E-05
---------------------------------------------------------------------------------------------------------------
1 STATIC ANALYSIS OF A STIFFENED PLATE DSOUG4 JUNE 9, 2003 MSC.NASTRAN 6/ 1/03 PAGE 117

0
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | T-PLATE | 1.5000E-01 : 1.0839E-01 : 1.1241E-01 : 1.1264E-01 :
2 | 2 | T-BOX | 1.0000E+00 : 1.3871E+00 : 1.3136E+00 : 1.3136E+00 :
Main Index Main Index
357 CHAPTER 6
Output Features and Interpretation
The design variable history as a function of design cycle is listed at the end of the
summary. The external design variable ID and the label columns are taken directly
from the DESVAR Bulk Data entries. The internal design variable ID column simply
refers to the order of internal sort on the design variables. (The internal and external
design variable sort may differ if DLINK entries are present. Independent design
variables are first sorted in ascending numerical order, followed by the dependent
design variables, again in ascending order.)
A final message is printed which indicates why the optimization task was stopped. In
this case, this message says:
indicating that the design task was successfully completed.
Other possible messages for the end of the optmization task and a brief explanation
for each is:
RUN TERMINATED DUE TO MAXIMUM NUMBER OF DESIGN
CYCLES =
The user prescribed maximum number of design cycles FSDMAX +
DESMAX have been carried out without achieving convergence.
RUN TERMINATED DUE TO PARAMETER OPTEXIT = n
The OPTEXIT parameter permits exitting the design tasks at a number of
points as documented in Appendix C.
RUN TERMINATED DUE TO SOFT CONVERGENCE AT CYCLE
NUMBER = Parameter SOFTEXIT has been set to YES and soft convergence
was achieved
RUN TERMINATED DUE TO HARD CONVERGENCE TO A BEST
COMPROMISE INFEASIBLE DESIGN AT CYCLE NUMBER =
The optimizer could not find a feasible design and it could make no further
progress, therefore the run was terminated (see Compromise Infeasible
Design on page 132.
RUN TERMINATED DUE TO CASE CONTROL COMMAND
DSAPRT(END=SENSITIVITY)
The DSAPRT command has requested a design sensitivity only run.
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 3.
Main Index Main Index


358
RUN TERMINATED DUE TO POOR GEOMETRY OR POOR
PROPERTIES. SEE FATAL MESSAGES PRINTED ABOVE FOR
INFORMATION
The redesign has produced a finite element model that violates finite element
modeling rules. For example, a mesh has distorted elements or an element
property has taken on a value that is not allowed, such as a negative shell
thickness.
RUN TERMINATED DUE TO MODE TRACKING FAILURE. SEE USER
FATAL MESSAGE 6677 (MTFRD) PRINTED ABOVE
The mode tracking algorithm (see Mode Tracking on page 62) was unable
to do its job and the run was therefore terminated.
Modification of Move Limit Parameters
As explained in the Automatic Updates of Move Limits on page 108, the
optimization algorithm sometimes reduces the move limit parameters by a factor of
2.0. When this occurs, a message like the following appears in the .f06 file:
This message never appeared in the DSOUG4.F06 output discussed in this section, but
still requires explanation. It is seen that four move limits have been revised and
guidance is provided for continuing the design task in a new run using the final
design variable values obtained in the current run. There are several guidelines
contained in the above listing. The first is that the new values should be used if the
design task is continued. The thinking is that the original move limits were too loose,
********************************************************
* *
* USER WARNING MESSAGE *
* *
* IF YOU WANT TO CONTINUE THE DESIGN AFTER THIS JOB *
* IS COMPLETED, YOU MUST INCLUDE A REVISED DOPTPRM *
* BULK DATA ENTRY IN THE BULK DATA SECTION WITH THE *
* FOLLOWING ITEMS MODIFIED AS SHOWN: *
* *
* DELP = 1.0000E-01 *
* DPMIN = 5.0000E-03 *
* DELX = 5.0000E-01 *
* DXMIN = 2.5000E-02 *
* *
********************************************************
********************************************************
* *
* IF A DELXV IS SPECIFIED ON A DESVAR BULK DATA ENTRY, *
* UPDATED DELXV VALUES ARE PRESENT ON THE DESVAR *
* ENTRIES CONTAINED IN THE PUNCH FILE. *
* *
* *
* (NOTE: THERE MAY BE MORE THAN ONE MESSAGE LIKE *
* THIS. THE LAST ONE IN THIS RUN SHOULD *
* BE LOCATED AND USED.) *
* *
********************************************************
Main Index Main Index
359 CHAPTER 6
Output Features and Interpretation
so the reduced values are preferred. The statement that YOU MUST INCLUDE A
REVISED... is somewhat misleading in that the new job will run with the original
values for the parameters. Changing these limits is really left to the users judgement.
The second guideline points out that DEXV values included on DESVAR entries in the
punch file (see Design Punch Output on page 374) reflect the reduction in the
individual move limits. It may be that these move limits are now too restrictive and it
will be necessary to manually adjust the DELXV values on the DESVAR entries.
Special Prints for Fully Stressed Design
Output
The output produced by the FSD algorithm is very similar to that just presented for
the MP algorithm. One significant difference is in the results that are printed at each
design cycle. With MP, these results are printed after an approximate optimization
and before a reanalysis has occurred. With FSD, there is no approximate analysis
because the optimizer is never called. The output is deferred until a new analysis has
been performed. This is indicated in the output by the printing of the following
message:
*******OPTIMIZATION RESULTS BASED ON AN EXACT ANALYSIS *********
instead of the message:
*******OPTIMIZATION RESULTS BASED ON THE APPROXIMATE MODEL****
The absence of approximate results also affects the SUMMARY OF THE DESIGN
CYCLE HISTORY that occurs at the end of the run. The following printout shows this
table resulting from a design task that included five FSD cycles followed by MP cycles
to convergence. Note that the number of FSD cycles appears at the top of the table, and
that no approximate objective values and therefore no fractional error are output
for the FSD cycles.
Main Index Main Index


360
Special Prints for Discrete Variable Optimization
Output
The output produced by the discrete variable processing methods is similar to the
continuous optimization results. There is no change for results printed at each
continuous optimization design cycle. At each discrete variable processing cycle, the
feasibility of a new discrete solution is checked based on both approximate and exact
reanalysis, a similar approach to the one taken for continuous variable optimization.
The approximate analysis is based on the approximate model and is labeled Soft
Feasible if all the constraints are satisfied. The results of the exact analysis of the
discrete design is labeled Hard Feasible if all the constraints are satisfied. Soft
Infeasible and Hard Infeasible indicate that at least one constraint is violated in
either the approximate or exact analyses, respectively.
.
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 10
NUMBER OF FULLY STRESSED DESIGN CYCLES COMPLETED 5
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 4
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------
INITIAL 4.828427E+00 -3.234952E-01
1 FSD 2.668171E+00 N/A 4.203515E-02
2 FSD 2.502887E+00 N/A 8.190254E-02
3 FSD 2.541077E+00 N/A 6.268603E-02
4 FSD 2.593662E+00 N/A 4.396826E-02
5 FSD 2.635597E+00 N/A 3.135078E-02
6 2.709053E+00 2.709045E+00 2.640250E-06 3.502930E-04
7 2.702080E+00 2.702064E+00 6.000030E-06 7.404297E-04
8 2.695884E+00 2.695881E+00 1.061257E-06 2.517090E-03
9 2.694964E+00 2.694966E+00 -7.962130E-07 2.553125E-03
***** A SOFT FEASIBLE DISCRETE SOLUTION FOUND (SOFT FEASIBILITY DISCRETE SOLUTION CHECK LOGIC) *****
MAXIMUM CONSTRAINT VALUE : -3.6252E-02 MUST BE LESS THAN 5.0000E-03
*************************************************************************************************************
***** A HARD FEASIBLE DISCRETE SOLUTION FOUND (HARD FEASIBILITY DISCRETE SOLUTION CHECK LOGIC) *****
MAXIMUM CONSTRAINT VALUE : 4.7029E-03 MUST BE LESS THAN 5.0000E-03
***************************************************************************************************************
Main Index Main Index
361 CHAPTER 6
Output Features and Interpretation
At each discrete variable optimization cycle, the design objective, design variables,
designed properties, and design constraints are presented after the preceding
continuous design cycle. For the discrete cycle, there is no input value since the input
values are the output values of the preceding continuous design cycle. Discrete results
are indicated by the printing of the following message. (Note that these results are
based on a full finite element analysis.)
The existing SUMMARY OF DESIGN CYCLE HISTORY is modified to include
these new discrete designs. An example of this new output follows. The summary
table also shows the number of discrete processing analyses completed and whether
a soft or hard feasible discrete solution is obtained. Note that the cycle numbers
flagged with a D correspond to the discrete results that have been computed based
on the preceding continuous design.
************************************************************
* *
* D I S C R E T E D E S I G N C Y C L E 6D *
* *
************************************************************
***** OPTIMIZATION RESULTS BASED ON AN EXACT ANALYSIS *****
Main Index Main Index


362
A final message is printed that indicates why the discrete optimization task was
terminated. For example,
indicating that the design task was successfully completed due to hard convergence
of the corresponding continuous design task and a hard feasible discrete design was
found.
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT CONVERGENCE ACHIEVED)
( HARD FEASIBLE DISCRETE SOLUTION OBTAINED)
( SOFT FEASIBLE DISCRETE SOLUTION OBTAINED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 7
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 6
NUMBER OF DISCRETE PROCESSING ANALYSES COMPLETED 2
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
-----------------------------------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM BJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
-----------------------------------------------------------------------------------------------------------------------------------------
INITIAL 4.740178E+02 2.907700E-02
1 3.957753E+02 3.924499E+02 8.473461E-03 7.261696E-04
2 2.943152E+02 2.943384E+02 -7.900563E-05 7.262860E-04
3 1.962019E+02 1.962255E+02 -1.202970E-04 6.985210E-04
4 9.808932E+01 9.811287E+01 -2.400495E-04 6.985210E-04
5 4.883356E+00 4.905637E+00 -4.541855E-03 6.985210E-04
5D 4.93356E+00 5.056371E+00 -4.38652 E-03 9.285210E-04
6 4.905637E+00 4.905637E+00 0.000000E+00 6.985210E-04
6D 4.92156E+00 5.041347E+00 -4.28652 E-03 7.253256E-04
----------------------------------------------------------------------------------------------------------------------------------------
DESIGN VARIABLE HISTORY
-----------------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5
-----------------------------------------------------------------------------------------------------------------------------------------
1 | 1 | LAYER1 | 1.0000E+00 : 1.2734E+00 : 1.3084E+00 : 1.2966E+00 : 1.2829E+00 : 1.2671E+00
2 | 2 | E | 1.0000E+00 : 9.9815E-01 : 9.9593E-01 : 9.9645E-01 : 1.0015E+00 : 1.0146E+00
3 | 3 | RHO | 1.0000E+00 : 8.0000E-01 : 6.0000E-01 : 4.0000E-01 : 2.0000E-01 : 2.0000E-01
4 | 4 | NU | 1.0000E+00 : 1.2000E+00 : 1.4400E+00 : 1.7280E+00 : 2.0736E+00 : 2.4883E+00
---------------------------------------------------------------------------------------------------------------------------------------
-----------------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL |
DV. ID. | DV. ID. | LABEL | : 5D 6: 6D :
-----------------------------------------------------------------------------------------------------------------------------------------
1 | 1 | LAYER1 | : 1.2000E+00 : 1.2671E+00 : 1.2000E+00 :
2 | 2 | E | : 1.0000E+00 : 1.0146E-01 : 1.0000E-01
3 | 3 | RHO | : 2.0000E+00 : 1.0000E-01 : 1.0000E-01 :
4 | 4 | NU | : 2.5000E+00 : 1.2886E+00 : 1.5000E+00 :
---------------------------------------------------------------------------------------------------------------------------------------
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 15.
AND HARD FEASIBLE DISCRETE DESIGN OBTAINED
Main Index Main Index
363 CHAPTER 6
Output Features and Interpretation
Special Prints for Topology Optimization
The majority of the prints for topology optimization follow that for standard
optimization. An exception is when design variables are to be printed. Even small
topology optimization tasks are likely to have thousands of design variables.
Therefore, the Comparison Between Input Property Variables from Analysis and
Design Models print never occurs and the design variable print that normally occurs
if P2 = 1 is suppressed unless P2>8. Similarly, the final Design Variable History print
does not occur unless P2 > 8 is specified.
Main Index Main Index


364
6.3 Design Sensitivity Output
Design Optimization Output on page 329 has provided an explanation of the
output that is produced in an optimization task. It is also possible to obtain output
from a sensitivity analysis and that is the topic of this subchapter. Recall from Design
Sensitivity Analysis on page 67 that a sensitivity coefficient is defined as a response
rate of change for a corresponding design variable change. (The resultant partial
derivative thus gives the slope of the j-th response function for the current design in
the i-th design variable dimension.) As has been mentioned several times, sensitivity
information is valuable in its own right since it provides a quantitative indication of
how much a design response will change due to a change in a design variable.
The primary means of requesting design sensitivity output is with the DSAPRT Case
Control command as detailed in the Case Control Section on page 208 and also in
Appendix D. This command provides tables of data that allow you to easily identify
which response and design variable are associated with the sensitivity. This form of
data is referred to as formatted sensitivities and can be contrasted with
unformatted sensitivities that simply provide a matrix print of the DSCM2 matrix
that contains the sensitivity coefficients as
Eq. 6-1
where i is the row order and j is the column order.
Unformatted sensitivity data prints can be activated by setting the Bulk Data
parameter OPTEXIT to 4 or 7, but this capability has been superseded by the DSAPRT
command. All the functionality, and much more, of the OPTEXIT parameter is
available with the DSAPRT command. Both the formatted and unformatted forms of
sensitivity data are presented here because the unformatted form has utility in that it
provides a more compact form of the the data and it is ideal for those who would like
to couple MSC.Nastran-computed sensitivities with their own postprocessors and/ or
optimizers.
Formatted Design Sensitivity
Examples of the formatted sensitivity output are presented in the following listings
that are based on the DSOUG4 example of Design Optimization Output on
page 329 with the following Case Control command:
DSAPRT(START=1,END=LAST)=ALL
DSCM2 ( )
i j
r
j
x
i
------- =
Main Index Main Index
365 CHAPTER 6
Output Features and Interpretation
DSAPRT used to provide formatted sensitivity information for the initial and final
design.
The first print is for the initial design:
****************************************************************************
* *
* D E S I G N S E N S I T I V I T Y M A T R I X O U T P U T *
* *
* *
* R E S P O N S E S E N S I T I V I T Y C O E F F I C I E N T S *
* *
****************************************************************************
--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 15 RESPONSE TYPE= WEIGHT SEID= 0
RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
6.9618E+00 1 T-PLATE 2.8310E+01 2 T-BOX 2.6608E+00

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 14 RESPONSE TYPE= DISP GRID ID= 10203 COMP NO= 3 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 2.0022E-02 1 T-PLATE -2.3718E-02 2 T-BOX -1.6799E-02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 1 RESPONSE TYPE= STRESS ELEM ID= 33 COMP NO= 7 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 1.5332E+04 1 T-PLATE -2.2852E+04 2 T-BOX -1.1791E+04

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 1 RESPONSE TYPE= STRESS ELEM ID= 34 COMP NO= 7 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 3.0478E+04 1 T-PLATE -2.9687E+04 2 T-BOX -2.5978E+04

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 2 RESPONSE TYPE= STRESS ELEM ID= 32 COMP NO= 14 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 1.7437E+04 1 T-PLATE -1.8750E+04 2 T-BOX -1.4722E+04

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 2 RESPONSE TYPE= STRESS ELEM ID= 33 COMP NO= 14 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 2.6989E+04 1 T-PLATE -3.1380E+04 2 T-BOX -2.2170E+04

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 2 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.4260E+04 1 T-PLATE -1.6849E+05 2 T-BOX -3.3482E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 3 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.5087E+04 1 T-PLATE -1.7292E+05 2 T-BOX -1.5709E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 14 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.4260E+04 1 T-PLATE -1.6849E+05 2 T-BOX -3.3482E+02
Main Index Main Index


366
The order of the data for the formatted prints can differ from that given for the
RESPONSES IN THE DESIGN MODEL listing since a different sort is used to organize
the data. For formatted sensitivities, the order of the data is:
Responses that share the superelement, response type, grid/ element ID and
component have a single header with the subcase and frequency time/ step supplied
with the response value. Response ID order corresponds to that given by Table 1 of
the DRESP1 Bulk Data entry (see Bulk Data Entries on page 893 of the MSC.Nastran
Quick Reference Guide).
The example of the listing does not contain any type2 or type 3 responses. If any are
present, they are sorted:

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 15 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.5087E+04 1 T-PLATE -1.7292E+05 2 T-BOX -1.5709E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 2 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.3341E+04 1 T-PLATE -1.6374E+05 2 T-BOX -1.2157E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 3 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.3655E+04 1 T-PLATE -1.6230E+05 2 T-BOX -1.0364E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 14 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.3341E+04 1 T-PLATE -1.6374E+05 2 T-BOX -1.2157E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 15 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.3655E+04 1 T-PLATE -1.6230E+05 2 T-BOX -1.0364E+02
First by superelement ID , then by
Response type, then by
External Response ID, then by
Grid/ Element ID, then by
Component, then by
Subcase, then by
Frequency/ Time Step
Main Index Main Index
367 CHAPTER 6
Output Features and Interpretation
The extensive labelling makes the interpretation straightforward, so only a limited
amount of explanation is given here. The first response in the table is for
RTYPE=WEIGHT. The value of the response is given as 6.9618 and the sensitivity with
respect to the first design variable (which has a label T-PLATE) is 28.310. This means
that if the value of the first design variable changes by a unit amount, the weight will
increase by 28.310. The sensitivity of the weight to the second design variable is 2.6608,
indicating that a unit change in this design variable increases the weight by a much
smaller amount than the first design variable.
If we scan down the table of sensitivity coefficients, we come to a header associated
with DRESP1 ID=2, Response Type=Stress, Element ID=33 and Component =14. Note
that the response value in this case is 26,764 which is greater than the 25,000 stress
allowable that is imposed in this example. In this case, the sensitivity with respect to
the first design variable is -31,771. This implies that a change of 0.0555 in this design
variable will reduce the stress to the allowable value. Using the weight sensitivity, this
corresponds to a weight increase of 1.572. For the second design variable, the stress
sensitivity is -21,953, indicating that the design variable would have to change by
0.0763 for a weight increase of only 0.203. This simplified analysis indicates that a
good design strategy is to increase the second design variable while decreasing the
second. If we look at the results of the first optimization cycle on page 345, it is seen
that this is exactly what the optimizer did in this situation, taking the first design
variable from 0.15 to 0.1087 and the second from 1.0 to 1.3347.
The listing of the design sensitivites for the final design are given below. It is left as an
exercise to the reader to use the sensitivities to determine that there is no direction that
can be found that will decrease the weight while keeping all the constraints within
their limits.
First by superelement ID, then by
External response ID, then by
Subcase, then by
Frequency/ time step
Main Index Main Index


368
.
****************************************************************************
* *
* D E S I G N S E N S I T I V I T Y M A T R I X O U T P U T *
* *
* *
* R E S P O N S E S E N S I T I V I T Y C O E F F I C I E N T S *
* *
****************************************************************************


--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 15 RESPONSE TYPE= WEIGHT SEID= 0
RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
6.7333E+00 1 T-PLATE 2.8302E+01 2 T-BOX 2.6249E+00
1 STATIC ANALYSIS OF A STIFFENED PLATE DSOUG4 FEBRUARY 11, 2003 MSC.NASTRAN 2/10/03 PAGE 121

0

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 13 RESPONSE TYPE= DISP GRID ID= 10302 COMP NO= 3 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 9.2276E-02 1 T-PLATE -2.4314E+00 2 T-BOX -1.1260E-03

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 14 RESPONSE TYPE= DISP GRID ID= 10203 COMP NO= 3 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 1.6605E-02 1 T-PLATE -1.8024E-02 2 T-BOX -1.0560E-02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 1 RESPONSE TYPE= STRESS ELEM ID= 33 COMP NO= 7 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 1.3300E+04 1 T-PLATE -2.5575E+04 2 T-BOX -7.5762E+03

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 1 RESPONSE TYPE= STRESS ELEM ID= 34 COMP NO= 7 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 2.4986E+04 1 T-PLATE -2.0807E+04 2 T-BOX -1.6104E+04

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 2 RESPONSE TYPE= STRESS ELEM ID= 32 COMP NO= 14 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 1.4496E+04 1 T-PLATE -1.7772E+04 2 T-BOX -9.0430E+03

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 2 RESPONSE TYPE= STRESS ELEM ID= 33 COMP NO= 14 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
2 2.2611E+04 1 T-PLATE -2.8783E+04 2 T-BOX -1.3908E+04
--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 2 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.3889E+04 1 T-PLATE -3.9550E+05 2 T-BOX -3.1901E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 3 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.5000E+04 1 T-PLATE -4.0365E+05 2 T-BOX -9.6785E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 4 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.4423E+04 1 T-PLATE -1.7590E+05 2 T-BOX -1.5196E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 14 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.3889E+04 1 T-PLATE -3.9550E+05 2 T-BOX -3.1901E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 15 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.5000E+04 1 T-PLATE -4.0365E+05 2 T-BOX -9.6785E+01
-4.8121E+02
Main Index Main Index
369 CHAPTER 6
Output Features and Interpretation
Unformatted Design Sensitivity
Unformatted design sensitivity output consists of two parts: a table listing the column
order of the DSCM2 matrix of Eq. 6-1, followed by a print of the DSCM2 matrix.
--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 2 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 3 RESPONSE TYPE= STRESS ELEM ID= 16 COMP NO= 9 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.4423E+04 1 T-PLATE -1.7590E+05 2 T-BOX -1.5196E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 1 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.3456E+04 1 T-PLATE -1.8969E+05 2 T-BOX
1 2.2838E+04 1 T-PLATE -3.8908E+05 2 T-BOX -5.6129E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 3 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.3047E+04 1 T-PLATE -3.8458E+05 2 T-BOX -8.4729E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 4 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.6146E+04 1 T-PLATE -2.0122E+05 2 T-BOX -3.5959E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 5 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.2663E+04 1 T-PLATE -1.7426E+05 2 T-BOX -9.5901E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 8 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.4224E+04 1 T-PLATE -2.0781E+05 2 T-BOX 3.4494E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 9 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.2663E+04 1 T-PLATE -1.7426E+05 2 T-BOX -9.5901E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 12 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.4224E+04 1 T-PLATE -2.0781E+05 2 T-BOX 3.4494E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 13 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.3456E+04 1 T-PLATE -1.8969E+05 2 T-BOX -4.8121E+02

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 14 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.2838E+04 1 T-PLATE -3.8908E+05 2 T-BOX -5.6129E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 15 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 2.3047E+04 1 T-PLATE -3.8458E+05 2 T-BOX -8.4729E+01

--------------------------------------------------------------------------------------------------------------------------------
DRESP1 ID= 6 RESPONSE TYPE= STRESS ELEM ID= 16 COMP NO= 17 SEID= 0
SUBCASE RESP VALUE DESIGN VARIABLE COEFFICIENT DESIGN VARIABLE COEFFICIENT
--------------------------------------------------------------------------------------------------------------------------------
1 1.6146E+04 1 T-PLATE -2.0122E+05 2 T-BOX -3.5959E+01
Main Index Main Index


370
The rows of DSCM2 are sorted on ascending independent design variable IDs, given
on the DESVAR Bulk Data entries. Row 1 thus corresponds to the independent design
variable with the lowest ID, row 2 the independent design variable with the second
lowest ID and so on. There are a number of places where this order is specified with
the best one perhaps the DESIGN VARIABLE HISTORY summary that was shown at
the end of the DSOUG4 Output on page 329.
DSCM2 column order is given in a correlation table. For OPTEXIT = +4 or 7, the
formatted table contents are printed in the output file. This output indicates the
particular column and response correlation in the DSCM2 matrix. For OPTEXIT = -4,
this correlation information is output as the DSCMCOL table. See the MSC.Nast ran
2005 r3 DMAP Programmers Guide for details.
Design Sensitivity Output Example
This example is taken from the Test Problem Library, problem number D200X5.
Modified by the addition of PARAM,OPTEXIT,4 to the Bulk Data Section. Thus, only
sensitivity analysis and not optimization will be performed.
D200X5 contains eleven design variables: three independent and eight dependent.
The independent design variables are used as multipliers of basis functions, which
have been implemented using DLINK entries.
The problem includes weight, displacement, and stress responses, with two separate
loading conditions. In short, enough of the basic components are present to give a
fairly complete description of the output format in design sensitivity analysis.
The following is an abbreviated listing of the DSCM2 matrix output:
Main Index Main Index
371 CHAPTER 6
Output Features and Interpretation
Note that the size of DSCM2 is 3 rows by 61 columns. Each row corresponds to a
single independent design variable. Since design variables 9, 10, and 11 are
independent, rows 1, 2, and 3 correspond, respectively, to these quantities. Each
r
3
x
3
---------
u
z 29 ,
x
3
----------------
\ .
|
| |
SUBCASE1
=
Matrix
Dimension
See the following
Correlation Table
matrix density
rows columns zero terms
rows columns
-----------------------------------------------------------------------------
\ .
| |
= 100
MATRIX DSCM2 (GINO NAME 101 ) IS A REAL 61 COLUMN X 3 ROW RECTANG MATRIX.
0COLUMN 1 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 4.0000E+01 2.2500E+01 1.5938E+01
0COLUMN 2 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 3.2871E+01 1.9345E+01 1.3255E+01
0COLUMN 3 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 3.2871E+01 1.9345E+01 1.3255E+01
0COLUMN 4 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 2.5733E+04 2.9702E+04 3.2985E+04
0COLUMN 5 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 2.5733E+04 2.9702E+04 3.2985E+04
0COLUMN 6 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -1.5881E+05 -1.5696E+05 -1.5544E+05
0COLUMN 7 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -1.5881E+05 -1.5696E+05 -1.5544E+05
. . . .
. . . .
. . . .
. . . .
0COLUMN 39 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -2.0435E+05 -2.9491E+04 -5.0688E+03
0COLUMN 40 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 2.1453E+00 1.4776E+00 1.1223E+00
0COLUMN 41 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) 2.1453E+00 1.4776E+00 1.1223E+00
0COLUMN 42 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -1.7913E+04 -1.7736E+04 -1.7589E+04
0COLUMN 43 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -1.7913E+04 -1.7736E+04 -1.7589E+04
0COLUMN 44 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -1.7913E+04 -1.7736E+04 -1.7589E+04
0COLUMN 45 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -1.7913E+04 -1.7736E+04 -1.7589E+04
. . . .
. . . .
. . . .
. . . .
0COLUMN 61 ROWS 1 THRU 3 --------------------------------------------------
ROW
1) -2.1265E+04 -1.0639E+04 -5.4152E+03
0THE NUMBER OF NON-ZERO TERMS IN THE DENSEST COLUMN = 3
0THE DENSITY OF THIS MATRIX IS 100.00 PERCENT.
Main Index Main Index


372
column corresponds to a particular response; we can note that the sensitivities of 61
responses have been computed. The response order is listed in the correlation table,
discussed shortly.
Before moving on to the correlation table, we should note that the density of this
matrix is 100%, that is, no zero terms are present. We can conclude that every
response will undergo some finite change for a change in any of the independent
design variables.
If this density were less than 100%, a closer inspection of the design model
formulation might be warranted. A response that is not a function of any of the design
variables (hence, a zero sensitivity coefficient), might indicate a design modeling
error. Quite often, these types of errors will show up as a null row or column of
DSCM2. A null column indicates a response which is not a function of any of the
design variables. A null row, on the other hand, indicates that changing a design
variable will have no effect whatsoever on any of the design responses. Both of these
errors are usually easily addressed, and underscore the significance of using design
sensitivities as a method for model checkout.
Column order in the design sensitivity coefficient matrix is given by the following
abbreviated correlation table:
----- IDENTIFICATION OF COLUMNS IN THE DESIGN SENSITIVITY -----
----- MATRIX THAT ARE ASSOCIATED WITH DRESP1 ENTRIES -----
----- WEIGHT/VOLUME RESPONSES -----
------------------------------------------
COLUMN DRESP1 RESPONSE
NO. ENTRY ID TYPE
------------------------------------------
1 35 WEIGHT
----- STATICS RESPONSES -----
----------------------------------------------------------------------------------------------------------
COLUMN DRESP1 RESPONSE GRID/ELM COMPONENT SUB PLY
NO. ENTRY ID TYPE ID NO. CASE NO.
----------------------------------------------------------------------------------------------------------
2 33 DISP 9 3 1
3 34 DISP 29 3 1
4 1 STRESS 1 7 1
5 1 STRESS 11 7 1
6 2 STRESS 1 9 1
7 2 STRESS 11 9 1
. . . . . .
. . . . . .
. . . . . .
. . . . . .
39 32 STRESS 18 17 1
40 33 DISP 9 3 2
41 34 DISP 29 3 2
42 2 STRESS 1 9 2
43 2 STRESS 11 9 2
44 4 STRESS 1 17 2
45 4 STRESS 11 17 2
. . . . . .
. . . . . .
. . . . . .
. . . . . .
61 20 STRESS 15 17 2
Main Index Main Index
373 CHAPTER 6
Output Features and Interpretation
Note that each column is associated with a particular response component for a given
subcase. For example, column 3 of DSCM2 contains the sensitivities of the z-
component displacement for grid 29, subcase 1. Column 41 contains the sensitivities
of the same displacement component for subcase 2.
In general, responses are grouped first by superelement, then by subcase, and finally
by DRESP1 or DRESP2 order. The responses are sorted by individual response type.
This order is given in Table 1 of the DRESP1 Bulk Data entry description (see Bulk
Data Entries on page 893 of the MSC.Nastran Quick Reference Guide).
Main Index Main Index


374
6.4 Design Punch Output
The term punch file refers to a output file whose origins can now be considered to
be associated with the infancy of digital computing. This file contains outputs from
the MSC.Nastran analysis that can be conveniently used as input to another computer
analysis. In the early days of computer analysis, these results could be punched onto
computer cards so that they could be read in, via a card reader, as input to another
analysis. This may seem quaint now that card punches and card readers are obsolete,
but the punch file retains considerable utility for postprocessing of MSC.Nastran
results. The file is particularly useful in the design sensitivity and optimization context
because it produces output in the format of the MSC.Nastran bulk data entries. This
provides a convenient way of capturing the results of the design process process and
using these results in a subsequent analysis to either punch fragments of the Bulk Data
file or to punch a complete new input Bulk Data file. This section provides an
overview of the options that are available for punched output.
Punch Parameters
As briefly explained in Output-Controlling Parameters on page 327, the DESPCH
parameter indicates how often design Bulk Data information is printed to the punch
file while DESPCH1 specifies both the format of the data (small field versus large
field) and what data are to be punched. Typically, the default of DESPCH=0 is
adequate since this provides information for the final design cycle and it is typically
only these data that are of interest in subsequent analyses. However, setting DESPCH
to a positive integer n produces punched output at every nth design cycle and the last
design cycle. This could be useful if a review of the design cycle history indicates that
an intermediate design actually is better than the final design. DESPCH can also be set
to a negative number, in which case no design data is punched.
The DESPCH1 parameter controls the format and the amount of data that are
punched. Positive values of this parameter produce Bulk Data entries in the large
field format; i.e, each item in the entry occupies 16 spaces. Data in the large field
format can be difficult to interpret visually if you are not used to it, but this may be the
only way to obtain sufficient numerical accuracy to make the design results usable.
Negative values of DESPCH1 produce results in the standard small field (i.e., 8
spaces) format.
The Bulk Data entries that are punched are determined by the absolute value of the
DESPCH1 parameter as indicated by Table 6-1.
Main Index Main Index
375 CHAPTER 6
Output Features and Interpretation
Some comments on this table are:
If any GRID location is being designed, the complete set of GRID points will
be punched whenever DESPCH1 is nonzero
DESPCH1=1 typically produces fewer entries than DESPCH1=2 and is ideal
when your bulk data file has the designed properties in a segregated part of
the file. DESPCH1=2 supports the case where the designed properties are
sprinkled thoughout the input file and it is simpler to globally replace all the
properties, even those that have not changed
The DRESP1 punch is for situations where mode tracking has been utilized.
The original design model may be imposing a limit on the 3rd mode, but the
redesign has placed this physical mode in the 5th position. The punched
DRESP1 will indicate that the 5th mode is to be designed. All DRESP1 entries
that are of RTYPE=EIGN or RTYPE=FREQ will be punched in this situation.
Combined quantities can be obtained by summing the three values in the
table. For example DESPCH1=6, which is the default, punches all properties,
the DESVARs and any DRESP1 entries. Meaningful combined values are 5
and 6.
Example for DESPCH
Figure 6-1 is an example of the information found in the .pch based on a DESPCH
request when a PCOMP Bulk Data entry has plies that are being designed. The
DESIGN CYCLE number is provided in a banner for the file and this is followed by
annotations for each type of data that follows. In this case DESVAR entries are
provided in the large field format and this is followed by values of the designed
PCOMP entry. A feature of PCOMP analysis in MSC.Nastran is that the PCOMP data
is actually converted to equivalent PSHELL/ MAT2 entries. These updated entries are
Table 6-1
DESPCH1 MEANING
0 Nothing is punched
1 Designed property entries (e.g., PROD or MAT1) are punched
2 All properties of a given type are punched when one or more of
that propety type is designed.
4 Updated DESVAR and DRESP1 entries are punched
Main Index Main Index


376
also provided in the .pch file, but are commented out so that the entire file can be
included in a bulk data file and there will be no difficulty with duplicate property IDs.
In certain applications, it may be desirable to remove the PCOMP data and remove the
comments from the PSHELL/ MAT2 Bulk Data entries.
Figure 6-1 Sample of Punched Output from PARAM DESPCH
Special Punch Considerations for Topology Optimization
The DESPCH parameter has a separate function in Topology Optimization. In this
case, PARAM DESPCH N writes output to a jobname.des file that lists the element ID
and the design variable value for every n-th design cycle. As shown in
Postprocessing Output on page 380, this file can be brought into MSC.Patran for
two-dimensional problems to show the topology optimization results. Setting
DESPCH 0 (default) produces design variable values for the final design cycle only.
$ **************************************************
$ * *
$ * DESIGN CYCLE NUMBER = 9 *
$ * *
$ **************************************************
$
$
$ UPDATED DESIGN MODEL DATA ENTRIES
$
DESVAR * 1PLY90 8.22658688E-02 1.99999996E-02+D 1V
*D 1V 3.00000000E+02 5.00000000E-01
DESVAR * 2PLYM45 2.85679966E-01 1.99999996E-02+D 2V
*D 2V 3.00000000E+02 5.00000000E-01
DESVAR * 3PLY45 1.99999996E-02 1.99999996E-02+D 3V
*D 3V 3.00000000E+02 5.00000000E-01
DESVAR * 4PLY0 8.38311464E-02 1.99999996E-02+D 4V
*D 4V 3.00000000E+02 5.00000000E-01
$
$ UPDATED ANALYSIS MODEL DATA ENTRIES
$
PCOMP* 1 -2.54759588E-03 0.00000000E+00 5.00000000E+03*
* HILL 0.00000000E+00 0.00000000E+00 SYM*
* 1 4.44235717E-04 9.00000000E+01 YES*
* 1 1.54267182E-03 4.50000000E+01 YES*
* 1 1.08000000E-04 -4.50000000E+01 YES*
* 1 4.52688197E-04 0.00000000E+00 YES
$ Spawned PSHELL, MAT2 entries from PCOMP 1
$ PSHELL* 1 100000001 5.09519130E-03 200000001*
$ * 1.00000000E+00 300000001 1.00000000E+00 0.00000000E+00*
$ * -2.54759588E-03 2.54759565E-03 0
$ MAT2* 100000001 8.43634500E+06 3.43898675E+06 2.56559100E+06*
$ * 8.37588350E+06 2.56559075E+06 3.73030525E+06 5.09999990E+00*
$ * 0.00000000E+00 0.00000000E+00 0.00000000E+00 0.00000000E+00*
$ * 0.00000000E+00 0.00000000E+00 0.00000000E+00 0.00000000E+00
$ * 0
$ MAT2* 200000001 4.85598850E+06 3.05657825E+06 2.49240750E+06*
$ * 1.27210570E+07 2.49240725E+06 3.34789700E+06 5.09999990E+00*
$ * 0.00000000E+00 0.00000000E+00 0.00000000E+00 0.00000000E+00*
$ * 0.00000000E+00 0.00000000E+00 0.00000000E+00 0.00000000E+00
$ * 0
$ MAT2* 300000001 7.44932688E+05 0.00000000E+00 0.00000000E+00*
$ * 9.02194062E+05 0.00000000E+00 0.00000000E+00 5.09999990E+00*
$ * 0.00000000E+00 0.00000000E+00 0.00000000E+00 0.00000000E+00*
$ * 0.00000000E+00 0.00000000E+00 0.00000000E+00 0.00000000E+00
$ * 0
Main Index Main Index
377 CHAPTER 6
Output Features and Interpretation
Special Punch in the Case of Shape Optimization
As described above, if shape optimization is being performed, a complete set of GRID
Bulk Data entries will be punched every n-th design cycle where n is set using
PARAM DESPCH. A special case occurs when the new mesh has become distorted to
the extent that the resulting finite elements cannot be generated.
In this case, an invalid mesh message is produced by the EMG modals and, if it has
not already been punched, the grid data for the last design whose mesh was valid as
punched. This allows the user to start up again from this last valid mesh while
perhaps changing some move parameters so that the mesh distortions do not occur.
New Bulk Data File
As indicated in the Case Control Section on page 208, the ECHO command can be
used to cause the writing of a new Bulk Data file after a design task has been
completed. Figure 6-3 displays a portion of the file that is produced if the DSOUG4
example of Design Optimization Output on page 329 includes the command
ECHO = SORT,PUNCH(NEWBULK)
A banner indicates the beginning of the updated Bulk Data file. It is apparent that
much of the file does not require any change. The first new Bulk Data entries are the
CBAR entries that utilize the new beam offset values. Properties and DESVAR entries
are also changed, while the responses and constraints do not. It is now very
convenient to make additional MSC.Nastran runs by including this new Bulk Data file
in a new job submittal. The new run may continue the SOL 200 design task or it could
perform a completely new type of analysis, such as nonlinear.
Main Index Main Index


378
Figure 6-2 Punch File Produced Using ECHO=PUNCH(NEWBULK)
$ *************************************************************
$ * *
$ * ENTIRE NEW BULK DATA DECK WITH UPDATED ENTRIES INSERTED *
$ * *
$ *************************************************************
$
BEGIN BULK
PARAM POST -1
PARAM NASPRT 1
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL:
$------------------------------------------------------------------------
$
GRID 10000 0.0 0.0 0.0
GRID 10001 2.5 0.0 0.0
......
......
GRID 10402 5.0 10.0 0.0
GRID 10403 7.5 10.0 0.0
GRID 10404 10.0 10.0 0.0
$
CQUAD4 1 1 10000 10001 10101 10100
CQUAD4 2 1 10001 10002 10102 10101
......
......
CQUAD4 15 1 10302 10303 10403 10402
CQUAD4 16 1 10303 10304 10404 10403
$
CBAR* 31 3 10200 10201*
* 0.00000000E+00 1.00000000E+00 0.00000000E+00 1*
* 0 0 0.00000000E+00 0.00000000E+00*
* 1.55632222E+00 0.00000000E+00 0.00000000E+00 1.55632222E+00
CBAR* 32 3 10201 10202*
* 0.00000000E+00 1.00000000E+00 0.00000000E+00 1*
* 0 0 0.00000000E+00 0.00000000E+00*
* 1.55632222E+00 0.00000000E+00 0.00000000E+00 1.55632222E+00
CBAR* 33 3 10202 10203*
* 0.00000000E+00 1.00000000E+00 0.00000000E+00 1*
* 0 0 0.00000000E+00 0.00000000E+00*
* 1.55632222E+00 0.00000000E+00 0.00000000E+00 1.55632222E+00
CBAR* 34 3 10203 10204*
* 0.00000000E+00 1.00000000E+00 0.00000000E+00 1*
* 0 0 0.00000000E+00 0.00000000E+00*
* 1.55632222E+00 0.00000000E+00 0.00000000E+00 1.55632222E+00
$
PSHELL* 1 1 1.12644322E-01 1*
* 1.00000000E+00 0 8.33333313E-01 0.00000000E+00*
* 0 *
*
PSHELL* 2 1 2.00000003E-01 1*
* 1.00000000E+00 0 8.33333313E-01 0.00000000E+00*
* 0 *
*
PBARL* 3 1 MSCBML0 HAT *
* *
* 3.00000000E+00 1.31358892E-01 2.00000000E+00 8.99999976E-01*
* 0.00000000E+00
$
MAT1 1 1.0E+7 0.33 0.283
$
FORCE 1 10004 2000.0 1.0 0.0 0.0
......
......
PARAM AUTOSPC YES
$
$------------------------------------------------------------------------
$ DESIGN MODEL: $------------------------------------------------------------------------
$
$...DEFINE THE DESIGN VARIABLES:
$
$DESVAR ID LABEL XINIT XLB XUB DELXV
DESVAR * 1T-PLATE 1.12644322E-01 1.00000005E-03+D 1V
*D 1V 1.00000000E+01 1.00000000E+00
DESVAR * 2T-BOX 1.31358886E+00 1.00000005E-03+D 2V
*D 2V 1.00000000E+01 1.00000000E+00
Main Index Main Index
379 CHAPTER 6
Output Features and Interpretation
Figure 6-3 Punch File Produced Using ECHO=PUNCH(NEWBULK)
$
$...RELATE THE DESIGN VARIABLES TO ANALYSIS MODEL PROPERTIES
$ (LINEAR RELATIONS SO USE DVPREL1)
$
$...EXPRESS SHELL THICKNESSES AS FUNCTIONS OF X1 X2:
$DVPREL1 ID TYPE PID FID PMIN PMAX C0 +
$+ DVIDD1 COEF1 DVID2 COEF2 ...
DVPREL1 1 PSHELL 1 T 0.01 +DP1
+DP1 1 1.0
$
$...EXPRESS BOX THICKNESS AS A FUNCTION OF X2:
DVPREL1 3 PBARL 3 DIM2
2 0.1
$. EXPRESS BOX OFFSET LOCATIONS AS A FUNCTION OF PLATE THICKNESS AND
$ FIXED BOX DIMENSIONS
DVCREL1 10 CBAR 31 W3A 1.5
1 0.5
DVCREL1 20 CBAR 32 W3A 1.5
1 0.5
DRESP1 13 D1 DISP 3 10302
DRESP1 14 D2 DISP 3 10203
DRESP1 15 W WEIGHT
$
$...PLACE BOUNDS ON THE RESPONSES:
$
$DCONSTR DCID RID LALLOW UALLOW
DCONSTR 10 1 -25000. 25000.
DCONSTR 10 2 -25000. 25000.
DCONSTR 10 3 -25000. 25000.
DCONSTR 10 6 -25000. 25000.
DCONSTR 20 13 -0.1 0.1
DCONSTR 30 14 -0.03 0.03
$
$DCONADD DCID DC1 DC2 ...
$ SUMMED CONSTRAINT SET FOR SUBCASE 1
DCONADD 100 10 20
$ SUMMED CONSTRAINT SET FOR SUBCASE 2
DCONADD 200 10 30
$
$...OPTIONAL OVERRIDE OF OPTIMIZATION PARAMETERS:
$
DOPTPRM IPRINT 7 DESMAX 20 DELP 0.5 P1 1 +
+ P2 15
$ (DELP=0.5 ALLOWS LARGER MOVES THUS OVERCOMING CONSTRAINT
$ VIOLATIONS QUICKER)
$
ENDDATA
Main Index Main Index


380
6.5 Postprocessing Output
This final section in the chapter discusses some of the postprocessing options that are
available from SOL 200 analyses. In particular, the ability of MSC.Patran to
postprocess optimization results is discussed. The immediate discussion is of a file
that contains results data in a format that allows in input into a spreadsheet, thereby
providing access to the spreadsheets plotting capability.
Comma Separated Values File
The File Management on page 205 indicates how a file can be assigned for the
special purpose of producing a file that can be read into a spreadsheet utility such as
Microsoft Excel. These results can then be manipulated using the spreadsheet utilities
and, in particular they can be plotted. The file is designated a Comma Separated
Values (or CSV) because the data are delimited by commas so that the spreadsheet can
readily understand them. In addition to the ASSIGN statement such as:
ASSIGN USERFILE=DSOUG4.CSV , STATUS=NEW, FORM=FORMATTED,
UNIT=52
it is necessary to include a PARAM, XYUNIT, 52 in the Bulk Data file. The DSAPRT
command is required to produce formatted sensitivity data. Design cycle history data,
including the objective, maximum constraint and design variable values are all
available in this file. Figure 6-4 shows four plots that can be derived from the CSV file
that is produced when the DSOUG4 file of the Stiffened Plate on page 412 is used.
The first three plots show:
1. The sensitivity of the weight as a function of the two design variables
2. The sensitivity of the stress as a function of the two design variables
3. The ratio of the stress sensitivities to the weight sensitivities for the same two
design variables.
This third chart is derived in Microsoft Excel by dividing the data of the second chart
by that of the first. This then shows in pictures the point made in words on page 367:
the most effective way to achieve the desired stress limit is to reduce the first design
variable while increasing the second.
The fourth chart in Figure 6-4 depicts the values of the design variables as a function
of the design cycle. Similar plots could be produced for the objective and the
maximum constraint value.
Main Index Main Index
381 CHAPTER 6
Output Features and Interpretation
Figure 6-4 Plots Produced from a Comma Separated Values File
Output from PARAM POST
This section discusses the postprocessing support that is available in MSC.Patran for
optimization results from MSC.Nastran. There are two alternatives for creating
output databases for MSC.Patran:
PARAM POST 0 in the MSC.Nastran bulk data file results in an .xdb file that
can be attached in an MSC.Patran session.
PARAM POST -1 results in an .op2 file that is read onto the MSC.Patran data
base.
weight sensitivities
0.00E+00
5.00E+00
1.00E+01
1.50E+01
2.00E+01
2.50E+01
3.00E+01
1 2
Design Variable
S
e
n
s
i
t
i
v
i
t
y
Series1
stress sensitivities
-4.00E+04
-3.00E+04
-2.00E+04
-1.00E+04
0.00E+00
1.00E+04
1 2
design variable
s
e
n
s
i
t
i
v
i
t
i
e
s
Series1
Series2
Stress Sensitivity Divided by Weight Sensitivity
-9.00E+03
-8.00E+03
-7.00E+03
-6.00E+03
-5.00E+03
-4.00E+03
-3.00E+03
-2.00E+03
-1.00E+03
0.00E+00
1 2
Design Variable
R
a
t
i
o
delta stress / delta
weight
0.00E+00
2.00E-01
4.00E-01
6.00E-01
8.00E-01
1.00E+00
1.20E+00
1.40E+00
1.60E+00
1 2 3 4
Plate Thickness
Box Thickness
a. Weight Sensitivities b. Stress Sensitivities
c. Stress/ Weight Sensitivities
d. Design Variables
Main Index Main Index


382
Both of these options support display of finite element results as a function of design
cycle. (It is necessary to set PARAM NASPRT 1 to have data saved at each design cycle
and only data requested in case control is saved.) Therefore, it is possible to
graphically depict, for example, the stress patterns or the deformations in the
designed structure at each design cycle.
The options differ in that the .op2 file contains some datablocks that are that specific
to SOL 200. These datablocks are documented in Section 4.4 of the MSC.Patran
MSC.Nastran Preference Guide, Volume 1: Structural Analysis, which is available online
when running MSC.Patran. In particular, a GEOMIN datablock permits the display of
shape basis vectors and new shapes as a function of the design cycle. Another
datablock named R1TABRG allows the display of which elements and grid responses
are retained for the design task (see Constraint Screening on page 64). For element
responses, elements with retained responses are identified in a on-off manner with
retained elements marked in white while the non-retained elements are black. For grid
responses, the direction of a marker at the grid point indicates which component of
the grid response is retained.
The .op2 file also contains information on the design variables, the maximum
constraint values and the objective function which are stored as global variables.
These data are handled in a special way: when the .op2 file that contains these data is
read into the MSC.Patran database, three XY plots are generated but not posted:
1. Objective function vs. design cycle
2. Maximum constraint vs. design cycle
3. Design Variable vs. design cycle.
Figure 6-5 compares displacement results for the first subcase at the initial and final
design cycles for the design task given in Cantilevered Plate on page 401. Note the
scales on the side of the two plots which indicates that the maximum displacement has
increased from 0.218 to 2.00, which is equal to one of the the design limits for this
problem. Figure 6-6 shows XY plots that are automatically generated in MSC.Patran
that show the design objective and the three design variable values as a function of the
design cycle.
Main Index Main Index
383 CHAPTER 6
Output Features and Interpretation
Figure 6-5 Displacement Patterns for the Initial and Final Design Object
Main Index Main Index


384
Figure 6-6 Representative XY Plot from MSC.Patran
a. Design variables as a function of design cycle.
b. Design objective as a function of design cycle.
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
CHAPTER
7
Example Problems
I Three-Bar Truss
I Vibration of a Cantilevered Beam (Turner's Problem)
I Cantilevered Plate
I Stiffened Plate
I Shape Optimization of a Culvert
I Analytic Boundary Shapes
I Dynamic Response Optimization
I Twenty-Five Bar Truss, Superelement and Discrete Variable Optimization
I Design Optimization with Composite Materials with Fully Stressed Design
I Acoustic Optimization
I RMS Response
I Transient Dynamic Optimization
I External Response to Include Alternative Buckling Response
I Topology Optimization
I Continuing the Design Process in a Subsequent Job
Main Index Main Index


386
This chapter includes a wide range of examples intended to highlight and illustrate
the Design Sensitivity and Optimization features of MSC.Nastran. Though by no
means exhaustive, these seventeen examples cover most major features including
both Property and Shape Optimization, Dynamic Response Optimization,
Superelements, Composites, Acoustic Responses, Discrete Variables, Fully Stressed
Design, External Responses and Restarts. Aeroelastic Optimization examples are
presented in the MSC.Nast ran Aeroelast ic Analysis Users Guide.
Running these examples on your own machine is a good way to familiarize yourself
with some of the capabilities in MSC.Nastran. The iterative nature of optimization is
such that different versions of MSC.Nastran and different computer platforms could
result in different results from those shown here. It is expected that these differences
will be small. All of these files are available in a directory delivered with your system.
The install_dir/ msc2004/ nast directory on UNIX and install_dir\ msc2004\ nast
directory on Windows contain these directories, news, documentation, and sample
problems for MSC.Nastran.
The installation of MSC.Nastran at your site includes the desopt directory, which is
shaded in the above figure. It contains all the input files associated with this chapter.
If you are unable to find these files, you should contact your System Administrator for
assistance.
Another good way to learn is by modifying these files. You might try adding Shape
variables to a Property Optimization example, adding analysis disciplines, modifying
the constraints, and so on. Most of these files are relatively small, and should not incur
tpl demo rfa del misc
archive sssalter doc
instest.dat
news.txt
relnotes linstat thermal dynamics desopt
nast
error.lis
Main Index Main Index
387 CHAPTER 7
Example Problems
any great CPU-related costs. In any event, the time spent learning with simple
examples can more than pay for itself when faced with larger, more complex
problems.
Note: The results shown in this guide were obtained from MSC.Nastran 2001 and
2004 running on Linux and Windows platforms. Results from running these
cases on different platforms or versions of MSC.Nastran may result in
somewhat different results, but they should be similar.
Main Index Main Index


388
7.1 Three-Bar Truss
A common task in design optimization is to reduce the mass of a structure subjected
to several load conditions. Figure 7-1 shows a simple three-bar truss that must be
built to withstand two separate loading conditions. Note that these two loads subject
the outer truss members to both compressive as well as tensile loads. Due to the
loading symmetry, we expect the design to be symmetric as well. As an exercise, we'll
show how to enforce this symmetry using design variable linking.
Figure 7-1 Three-Bar Truss
An important, but often overlooked consideration is that the optimization capability
in MSC.Nastran is multidisciplinary. That is, the final optimal design is the result of a
simultaneous consideration of all analysis disciplines across all subcases. In this case,
the optimal three-bar truss design will satisfy the load requirements for both statics
subcases, which is to be expected. (If, for example, a normal modes or buckling
subcase were to be added, the resultant design would have to not only satisfy the static
strength requirements, but also constraints on eigenvalues. As an exercise you may
wish to try adding an eigenvalue constraint.)
3
2
1
3
2
10 10
1
A1
A3
10
4
X
Y
20,000 lbs
(Subcase 2)
20,000 lbs
(Subcase 1)
Main Index Main Index
389 CHAPTER 7
Example Problems
Analysis Model Description
Design Model Description
The input data for this problem is given in Listing 7-1. Grid, element, and load data
are assigned based on the data supplied in the figure. The two separate static load
cases are defined in Solution 200 Case Control using the parameter ANALYSIS =
STATICS.
Listing 7-1 Input File for DSOUG1
ID MSC DSOUG1 $
TIME 10 $
SOL 200 $ OPTIMIZATION
CEND
TITLE = SYMMETRIC THREE BAR TRUSS DESIGN OPTIMIZATION - DSOUG1
SUBTITLE = BASELINE - 2 CROSS SECTIONAL AREAS AS DESIGN VARIABLES
ECHO = SORT
SPC = 100
DISP = ALL
STRESS = ALL
DESOBJ(MIN) = 20 $ (DESIGN OBJECTIVE = DRESP ID)
DESSUB = 21 $ DEFINE CONSTRAINT SET FOR BOTH SUBCASES
ANALYSIS = STATICS
SUBCASE 1
LABEL = LOAD CONDITION 1
LOAD = 300
SUBCASE 2
LABEL = LOAD CONDITION 2
Three rod (pin-jointed) structure in the x-y plane
Symmetric structural configuration with respect to the y axis
Two distinct 20,000 lb load conditions
Material: E = 1.0E7 psi
Weight density = 0.1 lbs/ in
3
Objective: Minimization of structural weight
Design variable: Cross-sectional areas A
1
, A
2
, and A
3
Design variable linking such that A
3
= A
1
Constraints: Allowable stress: Tensile 20,000 psi
Compressive = -15,000 psi
Displacement: + 0.2 at grid 4 in x end of y directions
Main Index Main Index


390
LOAD = 310
BEGIN BULK
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL
$------------------------------------------------------------------------
$
$ GRID DATA
$ 2 3 4 5 6 7 8 9 10
GRID 1 -10.0 0.0 0.0
GRID 2 0.0 0.0 0.0
GRID 3 10.0 0.0 0.0
GRID 4 0.0 -10.0 0.0
$ SUPPORT DATA
SPC1 100 123456 1THRU3
$ ELEMENT DATA
CROD 1 11 1 4
CROD 2 12 2 4
CROD 3 13 3 4
$ PROPERTY DATA
PROD 11 1 1.0
PROD 12 1 2.0
PROD 13 1 1.0
MAT1 1 1.0E+7 0.33 0.1
$ EXTERNAL LOADS DATA
FORCE 300 4 20000. 0.8 -0.6
FORCE 310 4 20000. -0.8 -0.6
$
$------------------------------------------------------------------------
$ DESIGN MODEL
$------------------------------------------------------------------------
$
$...DESIGN VARIABLE DEFINITION
$DESVAR ID LABEL XINIT XLB XUB DELXV(OPTIONAL)
DESVAR 1 A1 1.0 0.1 100.0
DESVAR 2 A2 2.0 0.1 100.0
DESVAR 3 A3 1.0 0.1 100.0
$
$...IMPOSE X3=X1 (LEADS TO A3=A1)
$DLINK ID DDVID CO CMULT IDV1 C1 IDV2 C2 +
$+ IDV3 C3 ...
DLINK 1 3 0.0 1.0 1 1.00
$
$...DEFINITION OF DESIGN VARIABLE TO ANALYSIS MODEL PARAMETER RELATIONS
$DVPREL1 ID TYPE PID NAME PMIN PMAX C0 +
$+ DVID1 COEF1 DVID2 COEF2 ...
DVPREL1 10 PROD 11 A
1 1.0
DVPREL1 20PROD 12 A
2 1.0
DVPREL1 30 PROD 13 A
3 1.0
$
$...STRUCTURAL RESPONSE IDENTIFICATION
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1 +
$+ ATT2 ...
Main Index Main Index
391 CHAPTER 7
Example Problems
DRESP1 20 W WEIGHT
DRESP1 21 U4 DISP 12 4
DRESP1 23 S1 STRESS PROD 2 11
12 13
$...CONSTRAINTS
$DCONSTR DCID RID LALLOW UALLOW
DCONSTR 21 21 -0.20 0.20
DCONSTR 21 23 -15000. 20000.
$
$...OPTIMIZATION CONTROL:
$
DOPTPRM DESMAX 10 DELP 0.5 P1 1 P2 15
$
$.......2.......3.......4.......5.......6.......7.......8.......9.......0
ENDDATA
Turning to the design model description, we see three design variables are used to
control the individual rod element cross-sectional areas. The set of DESVAR,
DVPREL1 entries define the relations:
Eq. 7-1
Note that the initial design variable values are equal to the initial rod cross-sectional
areas. With the multiplying coefficients of 1.0 on the DVPREL1 entries, the design
model and the initial analysis model properties agree. Thus, no design model
override takes place.
Since the model is symmetric and the two loading cases act to impose the same critical
design loads on each side of the plane of symmetry, we expect the optimizer to yield
a symmetric final design with A1 = A3. We may want to enforce some type of
symmetry ourselves, perhaps to address ease-of-manufacture concerns. This can be
done using the DLINK entry to explicitly link design variables together. Although
strictly not necessary, DLINK 1 is used to define the relation
Eq. 7-2
The value of now depends on . The optimizer now only needs to consider two
independent design variables and instead of three. This not only simplifies the
problem, but also reduces the cost of the sensitivity analysis. These cost savings can
be significant in a larger problem.
An alternate method of linking would have been placing rod elements 11 and 13 in the
same property group. The DLINK entry would then be unnecessary since a single
design variable could control the areas of all elements of the group. You may want to
try this formulation yourself to see that the results are indeed identical. Alternatively,
A
i
1.0X
i
= i 1 2 3 , , =
X
3
1.0X
1
=
X
3
X
1
X
1
X
2
Main Index Main Index


392
you may wish to remove the DLINK entry to assure yourself that MSC.Nastran will
produce a symmetric design in this case even when the design model allows for
.
DRESP1 entries define the design responses: displacement and stress responses
which are to be constrained and the weight response that is to be used as the objective.
Note that the objective and constraints are called out in Case Control. The DESOBJ
command points to the weight response defined on DRESP1 20, while the DESSUB
command identifies DCONSTR set 21. Since this appears above the subcase level, the
constraint bounds will be applied to both subcases.
A DOPTPRM entry is used to override some of the default optimization parameters.
DESMAX, or the maximum number of design cycles to be performed, has been
increased from the default of 5 to 10. (If convergence is indicated, fewer than 10 cycles
may be performed.) DELP has also been increased from its default of 0.2 to 0.5. This
allows any analysis model parameter to undergo changes of up to 50% on each design
cycle. As was discussed in Move Limits on page 104, this provides move limits on
the approximate model. And finally, IPRINT, P1, and P2 have all been increased from
their defaults to allow more diagnostic output.
Turning to the results of the optimization, we see from the hard convergence decision
logic output that the problem converges to a feasible design since all constraints are
satisfied at the optimum. Note that the properties and design variables have still
changed appreciably from their values on the previous design cycle. This suggests
that we may have some flexibility in choosing final dimensions for our design without
greatly affecting the overall weight.
The summary of design cycle history and the design variable history tables both
indicate that convergence was achieved in six iterations. A total of seven finite
element analyses were performed, one for the initial analysis and one in connection
with each of the design cycles. Also listed in the summary of design cycle history is
the fractional error of approximation in the objective function, which compares the
approximate objective function on exit from the optimizer with the true value
computed from the subsequent finite element analysis. This is an indication of the
quality of the objective function approximation. Since the weight is linear in the
design variables , we expect the linearized weight objective to
be an exact approximation; this indeed is seen to be the case, except for a small amount
of numerical "noise".
X
1
X
3

W L
1
X
1
L
2
X
2
L
3
X
3
+ + = ( )
Main Index Main Index
393 CHAPTER 7
Example Problems
Caution should be used when interpreting the data appearing the the fractional error
of approximation column. One might be tempted to use the fractional error of
approximation information as justification for increasing the move limits (DELP and
DELX on the DOPTPRM entry, and/ or DELXV on the DESVAR entry) in order to
achieve convergence earlier. This might not yield the expected results. In this case,
the linear objective function was approximated linearly, which is exact. However, the
stress constraints in this problem are proportional to the inverse of the cross sectional
areas, and are thus nonlinear in the design variables. Too large of a move limit could
lead to increased approximation errors. Here, the mixed method of approximation
was used, which we know to be pretty good at approximating sizing-type stress
constraints. Thus, increasing move limits to 50% is reasonable, as it will probably save
us the cost of a few additional finite element analyses. A recommendation would be
to stick with the defaults, unless something is known about the problem beforehand
that would indicate enhanced efficiency by changing them.
Main Index Main Index


394
Listing 7-2 Selected Output for DSOUG1
***** NORMAL CONVERGENCE CRITERIA SATISFIED ***** (HARD CONVERGENCE DECISION LOGIC)
**************************************************************************************
CONVERGENCE ACHIEVED BASED ON THE FOLLOWING CRITERIA
(HARD CONVERGENCE DECISION LOGIC)
RELATIVE CHANGE IN OBJECTIVE 7.1533E-04 MUST BE LESS THAN 1.0000E-03
OR ABSOLUTE CHANGE IN OBJECTIVE 1.9295E-03 MUST BE LESS THAN 1.0000E-20
--- AND ---
MAXIMUM CONSTRAINT VALUE 1.9983E-03 MUST BE LESS THAN 5.0000E-03
(CONVERGENCE TO A FEASIBLE DESIGN)
--- OR ---
MAXIMUM OF RELATIVE PROP. CHANGES 6.1480E-02 MUST BE LESS THAN 1.0000E-03
AND MAXIMUM OF RELATIVE D.V. CHANGES 6.1480E-02 MUST BE LESS THAN 1.0000E-03
(CONVERGENCE TO A BEST COMPROMISE INFEASIBLE DESIGN)
**************************************************************************************
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 7
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 6
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 4.828427E+00 -3.234952E-01
1 3.006489E+00 3.006791E+00 -1.004647E-04 -3.267090E-03
2 2.821740E+00 2.821782E+00 -1.487063E-05 -1.902090E-02
3 2.734679E+00 2.734640E+00 1.412391E-05 -7.078516E-03
4 2.703527E+00 2.703542E+00 -5.467625E-06 2.142578E-04
5 2.697373E+00 2.697375E+00 -6.187238E-07 1.870606E-03
6 2.695444E+00 2.695446E+00 -7.076191E-07 1.998340E-03
---------------------------------------------------------------------------------------------------------------
1 SYMMETRIC THREE BAR TRUSS DESIGN OPTIMIZATION - DSOUG1 SEPTEMBER 7, 2001 MSC.NASTRAN 4/ 9/01 PAGE 76
BASELINE - 2 CROSS SECTIONAL AREAS AS DESIGN VARIABLES
0
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | A1 | 1.0000E+00 : 7.0990E-01 : 7.8283E-01 : 8.1260E-01 : 8.2586E-01 : 8.2957E-01 :
2 | 2 | A2 | 2.0000E+00 : 9.9888E-01 : 6.0759E-01 : 4.3627E-01 : 3.6765E-01 : 3.5099E-01 :
3 | 3 | A3 | 1.0000E+00 : 7.0990E-01 : 7.8283E-01 : 8.1260E-01 : 8.2586E-01 : 8.2957E-01 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | A1 | 8.3652E-01 :
2 | 2 | A2 | 3.2941E-01 :
3 | 3 | A3 | 8.3652E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 6.
Main Index Main Index
395 CHAPTER 7
Example Problems
7.2 Vibration of a Cantilevered Beam (Turner's Problem)
This problem was originally published by M.J. Turner (see Reference 10.). The
problem is to design a minimum weight structure while constraining the fundamental
natural frequency to be at or above 20 Hz. The beam is symmetric about Z = 0 and
made up of a shear web having top and bottom caps that are modeled with rod
elements. Turner's original design model consisted of piecewise linear bar cross-
sectional areas and web thicknesses; however, we will just approximate this as a step
function model with uniform cross-sectional rod elements and uniform thickness
shear elements within each of three bays.
Figure 7-2 Cantilever Beam Vibration Model
Analysis Model Description
Web is modeled by QUAD4 elements
Caps are modeled by ROD elements
Material: E = 1.03E7 psi
Poisson ratio = 0.3
Nonstructural mass: Lumped masses at top and bottom nodes, 15 lbs each
20 20 20
6
A1
A2 A3
X
Z
t2 t3 t1
Weight density 0.1 in
3
=
Main Index Main Index


396
Design Model Description
Normal modes analysis is requested in Solution 200 by setting the Case Control
parameter ANALYSIS = MODES. In order to determine a minimum weight design
subject to a lower-bound constraint on fundamental frequency, the thicknesses of the
three shear panels and the six rod cross-sectional areas are allowed to vary. These
variations are described using six design variables; three for the web thicknesses and
three for the rod cross-sectional areas (symmetry is imposed on the upper and lower
ROD element areas.) See Listing 7-3 for details.
Listing 7-3 Input File for DSOUG2
ID MSC DSOUG2 $
TIME 10
SOL 200 $ OPTIMIZATION
CEND
TITLE = VIBRATION OF A BEAM. DSOUG1
SUBTITLE = TURNER'S PROBLEM
ECHO = UNSORT
DISP = ALL
STRESS = ALL
METHOD = 1
ANALYSIS = MODES
DESOBJ(MIN) = 10 $ OBJECTIVE FUNCTION DEFINITION
DESSUB = 10 $ CONSTRAINT SET SELECTION
$
BEGIN BULK
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL:
$------------------------------------------------------------------------
$
EIGRL 1
GRID 1 0.0 0.0 -3.0 123456
GRID 2 20.0 0.0 -3.0 2456
GRID 3 40.0 0.0 -3.0 2456
GRID 4 60.0 0.0 -3.0 2456
GRID 5 0.0 0.0 3.0 123456
GRID 6 20.0 0.0 3.0 2456
GRID 7 40.0 0.0 3.0 2456
GRID 8 60.0 0.0 3.0 2456
$
CROD 1 201 5 6
CROD 2 202 6 7
Objective: Minimization of structural weight
Design variables: Cross-sectional areas of rod elements (symmetry imposed on
the top and bottom elements)
Constraints: Fundamental bending frequency at or above 20 Hz
Main Index Main Index
397 CHAPTER 7
Example Problems
CROD 3 203 7 8
CROD 7 201 1 2
CROD 8 202 2 3
CROD 9 203 3 4
PROD 201 1 1.0 0.0
PROD 202 1 1.0 0.0
PROD 203 1 1.0 0.0
$
CQUAD4 4 204 1 2 6 5
CQUAD4 5 205 2 3 7 6
CQUAD4 6 206 3 4 8 7
PSHELL 204 1 0.2
PSHELL 205 1 0.2
PSHELL 206 1 0.2
$
CONM2 10 2 15.0
CONM2 11 3 15.0
CONM2 12 4 15.0
CONM2 14 6 15.0
CONM2 15 7 15.0
CONM2 16 8 15.0
$
MAT1 1 1.03E7 0.3 0.1
PARAM WTMASS 0.002588
PARAM GRDPNT 1
$
$------------------------------------------------------------------------
$ DESIGN MODEL:
$------------------------------------------------------------------------
$
$...Define the design variables
$DESVAR ID LABEL XINIT XLB XUB DELXV
$
DESVAR 1 A1 1.0 0.01 100.0
DESVAR 2 A2 1.0 0.01 100.0
DESVAR 3 A3 1.0 0.01 100.0
DESVAR 4 T1 1.0 0.001 10.0
DESVAR 5 T2 1.0 0.001 10.0
DESVAR 6 T3 1.0 0.001 10.0
$
$...Relate the design variables to analysis model properties
$DVPREL1 ID TYPE PID FID PMIN PMAX C0 +
$+ DVIDD1 COEF1 DVID2 COEF2 ...
$
DVPREL1 1 PROD 201 A
1 1.0
DVPREL1 2 PROD 202 A
2 1.0
DVPREL1 3 PROD 203 A
3 1.0
DVPREL1 4 PSHELL 204 T
4 0.2
DVPREL1 5 PSHELL 205 T
5 0.2
DVPREL1 6 PSHELL 206 T
6 0.2
Main Index Main Index


398
$
$...Identify the analysis responses to be used in the design model
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1 +
$+ ATT2 ...
$
DRESP1 1 W WEIGHT
DRESP2 10 STWGHT 20
DTABLE FIXWGT
DRESP1 1
DEQATN 20 STWGT(fixwgt,TOTWGT) = TOTWGT - FIXWGT
DTABLE FIXWGT 90.
DRESP1 2 F1 FREQ 1
$
$ Identify the constraints
$DCONSTR DCID RID LALLOW UALLOW
DCONSTR 10 2 20.
$
$...Optional override of design optimization parameters:
DOPTPRM IPRINT 2 DESMAX 10 DELP 0.5 P1 1
P2 15
$
PARAM POST -1
ENDDATA
The lower-bound constraint on the first mode is 20 Hz. The DRESP1 entry with ID=2
is used to designate the first mode and DCONSTR 10 imposes the lower bound of 20.
Hz. Note that the FREQ RTYPE has been used on the DRESP1, allowing the constraint
specifcation in Hz. If the EIGN RTYPE had been used, the constraint would have to be
specified on the eigenvalue value. Since there is no upper bound imposed on the
response, the upper bound field is left blank on the DCONSTR entry.
The objective to be minimized is the structural weight. The DRESP1 entry with ID=1
selects and the WEIGHT response type and, since the ATTA and ATTB fields are left
blank, the total structural weight in the Z direction is designated. Because the total
weight includes the fixed weight of the lumped masses, a DRESP2 is used to subtract
this 90. pound fixed weight from the total weight. This subtraction step is not a
requirement for this problem, but it does serve to highlight the structural weight and
improves the ability of the optimizer to focus on reducing the weight that it has
influence over.
Turning to the optimization results of Listing 7-4, we note that hard convergence has
been achieved after six design cycles, requiring a total of seven finite element analyses.
We note that a feasible design has been achieved as the maximum constraint value is
less than the allowable maximum of GMAX=0.005. The slight constraint violation
results in a final first natural frequency of 19.941 Hz. as opposed to the 20.0 Hz.
requirement. This is considered more than adequate for engineering purposes, but it
should be clear that the GMAX limit could be reduced or the DCONSTR bound could
Main Index Main Index
399 CHAPTER 7
Example Problems
be increased to produce a final design that is closer to or above the 20 Hz. requirement.
It is seen that the structural weight has been reduced from the 19.20 pounds of the
original design to 6.973 pounds.
Main Index Main Index


400
Listing 7-4 Design History Output for DSOUG2
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 7
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 6
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.920000E+01 -2.947325E-01
1 1.065987E+01 1.066022E+01 -3.220598E-05 -7.206659E-02
2 7.744514E+00 7.744980E+00 -6.008964E-05 -5.249882E-03
3 6.947861E+00 6.948044E+00 -2.635353E-05 6.191349E-03
4 6.985199E+00 6.985268E+00 -9.829910E-06 2.629376E-03
5 6.972710E+00 6.972694E+00 2.188363E-06 2.970886E-03
6 6.972694E+00 6.972694E+00 0.000000E+00 2.970886E-03
---------------------------------------------------------------------------------------------------------------
0
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | A1 | 1.0000E+00 : 7.6546E-01 : 8.5520E-01 : 8.2450E-01 : 8.8108E-01 : 8.7456E-01 :
2 | 2 | A2 | 1.0000E+00 : 5.0000E-01 : 3.4241E-01 : 4.3515E-01 : 4.4017E-01 : 4.3576E-01 :
3 | 3 | A3 | 1.0000E+00 : 5.0000E-01 : 2.5000E-01 : 1.2501E-01 : 9.6605E-02 : 1.0639E-01 :
4 | 4 | T1 | 1.0000E+00 : 4.9996E-01 : 3.1470E-01 : 2.3823E-01 : 2.2935E-01 : 2.2438E-01 :
5 | 5 | T2 | 1.0000E+00 : 4.9949E-01 : 2.4975E-01 : 2.2405E-01 : 1.9731E-01 : 1.9497E-01 :
6 | 6 | T3 | 1.0000E+00 : 4.9987E-01 : 2.4994E-01 : 1.2497E-01 : 1.2079E-01 : 1.2475E-01 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | A1 | 8.7456E-01 :
2 | 2 | A2 | 4.3576E-01 :
3 | 3 | A3 | 1.0639E-01 :
4 | 4 | T1 | 2.2438E-01 :
5 | 5 | T2 | 1.9497E-01 :
6 | 6 | T3 | 1.2475E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 6.
Main Index Main Index
401 CHAPTER 7
Example Problems
7.3 Cantilevered Plate
This section uses a simple cantilevered plate example to illustrate two important
optimization features:
Multidisiplinary analysis and design
Reduced basis vectors
The multidisciplinary aspect of the example refers to the fact that a total of four
subcases with three different analysis types (STATICS, MODES and MFREQ) are
analyzed and each has their own set of design conditions. Reduced basis formulations
were introduced in Designed Properties on page 39, as an efficient way to reduce
the number of independent design variables. This example shows how the DVPREL1
entries can be used to implement these reduced basis formulations.
Other features, such as constraint screening and optimization parameters are also
illustrated and commented upon during the description of the input data file.
Suppose we wish to determine the optimum thickness distribution of the cantilever
plate in Figure 7-3 such that the structural mass is minimized. Two separate static
loading conditions are imposed: the first is a tip loading condition as shown in the
figure; the second is a uniform pressure loading in the -Z direction. Constraints are
placed on maximum allowable tip displacement and von Mises stresses at the upper
surface of the first row of elements along the length of the cantilever. A normal modes
analysis is also performed and a constraint is placed on first natural frequency.
Finally, a frequency response analysis is performed that uses a modal solution based
on the modes computed in the normal modes analysis. The loading is the same tip
loading that is applied in the first static subcase except that it is applied across a
frequency range extending from 0.0 to 10. Hz. and at a magnitude that is 10% of the
statically applied load.
Main Index Main Index


402
Figure 7-3 Cantilever Plate Thickness Design
Analysis Model Description
Design Model Description
2 x 8 array of CQUAD4 elements
Material: E = 1.0E+6 psi
Poisson ratio = 0.33
Two static load conditions:
Tip loading:
Pressure loading:
One normal modes analysis
One modal frequency analysis
Tip loading:
Structural damping of 10%
Two 500 lb loads in -z direction
Two 50 lb loads in z-direction
Objective: Structural weight minimization
Design variables: Basis functions: Constant, linear and quadratic in
x direction to describe plate
element thicknesses
Constraints:
10
40
t1 t2 t3 t4 t5 t6 t7 t8
p2
Z
Y
p1
X
Weight density 0.1 in
3
=
Uniform at 7.5 lb in
2

Main Index Main Index

403 CHAPTER 7
Example Problems
The reduced basis functions used here consist of three basis vectors: constant, linear,
and quadratic. The thickness distribution can be written as a linear combination of
these basis vectors as:
Eq. 7-3
The eight individual plate thicknesses (one for each station along the length of the
cantilever) are functions of three independent design variables , and .
Although DLINK entries could be used to create these relationships of (you may want
to try this exercise on your own), this example applies the relations of Eq. 7-3 directly
on the DVPREL1 entries.
Listing 7-5 begins with four subcases that define the design task. Subcases 1 and 2
perform static analysis and impose differing load conditions while sharing the
imposed design constraints (DESSUB=10). Subcase 3 performs a normal modes
analysis with the METHOD=100 identififying the EIGRL Bulk Data entry that
specifies the eigenvalue requests while the DESSUB=30 points to the DCONSTR that
imposes a limit on the first natural frequency. Subcase 4 performs a modal frequency
response analysis (ANALYSIS=MFREQ) and uses the same METHOD=100
specification for the eigenanalyis. For the fourth subcase, the PHASE part of the
DISP(PHASE,SORT2)=100 acts to signify that the magnitude/ phase represenations is
to be used not only in the printed and stored results, but also when for the FRDISP
request on DRESP1 ID=230. The SORT2 request and the 100 have no effect on the on
the design response.
Listing 7-5 Input File for DSOUG3
ID MSC DSOUG3 $
TIME 10
ANALYSIS = STATICS
ANALYSIS = MODES
ANALYSIS = MFREQ
Tip Displacement 2.0
von Mises Stress 29ksi
First natural frequency 1.5Hz.
Magnitude of tip displacement 2.0
t
1
t
2
t
3
.
t
8
)



`

1
1.0
1.0
1.0
.
1.0
)



`




=
2
1.0
0.875
0.75
.
0.125
)



`

3
1.0
0.7656
0.5625
.
0.0156
)



`




+ +

1

2

3
Main Index Main Index


404
SOL 200 $ OPTIMIZATION
CEND
TITLE = CANTILEVERED PLATE - DSOUG3
SUBTITLE = REDUCED BASIS FORMULATION - MULTIDISPLINARY OPTIMIZATION
SPC = 100
DISP = ALL
STRESS = ALL
DESOBJ(MIN) = 35 $ OBJECTIVE FUNCTION DEFINITION
SUBCASE 1
ANALYSIS = STATICS
LABEL = LOAD CONDITION 1
LOAD = 300
DESSUB = 10
SUBCASE 2
ANALYSIS = STATICS
LABEL = LOAD CONDITION 2
LOAD = 310
DESSUB = 10
SUBCASE 3
ANALYSIS = MODES
LABEL = NORMAL MODES ANALYSIS
STRESS = NONE
METHOD = 100
DESSUB = 30
SUBCASE 4
ANALYSIS = MFREQ
METHOD = 100
SET 100 = 9,19,29
DISP(PHASE,SORT2) = 100
STRESS = NONE
LABEL = MODAL FREQUENCY ANALYSIS
LOADSET = 2000
DLOAD = 1000
DESSUB = 40
FREQ = 1000
BEGIN BULK
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL:
$------------------------------------------------------------------------
$
MAT1 51 1.0E+7 0.33 0.1 +M2
+M2 50000. 50000. 29000.
SPC1 100 123456 1 11 21
GRID 1 0. -5. 0.
GRID 2 5. -5. 0.
GRID 3 10. -5. 0.
GRID 4 15. -5. 0.
GRID 5 20. -5. 0.
GRID 6 25. -5. 0.
GRID 7 30. -5. 0.
GRID 8 35. -5. 0.
GRID 9 40. -5. 0.
GRID 11 0. 0. 0.
GRID 12 5. 0. 0.
GRID 13 10. 0. 0.
Main Index Main Index
405 CHAPTER 7
Example Problems
GRID 14 15. 0. 0.
GRID 15 20. 0. 0.
GRID 16 25. 0. 0.
GRID 17 30. 0. 0.
GRID 18 35. 0. 0.
GRID 19 40. 0. 0.
GRID 21 0. 5. 0.
GRID 22 5. 5. 0.
GRID 23 10. 5. 0.
GRID 24 15. 5. 0.
GRID 25 20. 5. 0.
GRID 26 25. 5. 0.
GRID 27 30. 5. 0.
GRID 28 35. 5. 0.
GRID 29 40. 5. 0.
$
CQUAD4 1 1 1 2 12 11
CQUAD4 2 2 2 3 13 12
CQUAD4 3 3 3 4 14 13
CQUAD4 4 4 4 5 15 14
CQUAD4 5 5 5 6 16 15
CQUAD4 6 6 6 7 17 16
CQUAD4 7 7 7 8 18 17
CQUAD4 8 8 8 9 19 18
CQUAD4 11 1 11 12 22 21
CQUAD4 12 2 12 13 23 22
CQUAD4 13 3 13 14 24 23
CQUAD4 14 4 14 15 25 24
CQUAD4 15 5 15 16 26 25
CQUAD4 16 6 16 17 27 26
CQUAD4 17 7 17 18 28 27
CQUAD4 18 8 18 19 29 28
PSHELL 1 51 3.0 51 51
PSHELL 2 51 2.640625 51
PSHELL 3 51 2.3125 51
PSHELL 4 51 2.015625 51
PSHELL 5 51 1.75 51
PSHELL 6 51 1.515625 51
PSHELL 7 51 1.3125 51
PSHELL 8 51 1.140625 51
$ 2 3 4 5 6 7 8 9 10
FORCE 300 9 500. 0.0 0.0 -1.0
FORCE 300 29 500. 0.0 0.0 -1.0
PLOAD2 310 7.5 1 THRU 8
PLOAD2 310 7.5 11 THRU 18
$
EIGRL 100 10
$
LSEQ 2000 28 300
RLOAD1 1000 28 29
TABLED129
0.0 0.1 1000.00.1ENDT
FREQ1 1000 0.0 0.10 100
PARAM G 0.10
$
$------------------------------------------------------------------------
Main Index Main Index


406
$ DESIGN MODEL:
$------------------------------------------------------------------------
$
$...DEFINE THE DESIGN VARIABLES
$DESVAR ID LABEL XINIT XLB XUB DELXV
DESVAR 10 ALPH1 1.0
DESVAR 20 ALPH2 1.0
DESVAR 30 ALPH3 1.0
$
$...EXPRESS ANALYSIS MODEL PROPERTIES LINEARLY IN TERMS OF
$ DESIGN VARIABLES
$DVPREL1 ID TYPE PID FID PMIN PMAX C0
$+ DVID1 COEF1 DVID2 COEF2 ...
DVPREL1 1 PSHELL 1 T
10 1.0 201.000301.0
DVPREL1 2 PSHELL 2 T
10 1.0 200.875300.7656
DVPREL1 3 PSHELL 3 T
10 1.0 200.750300.5625
DVPREL1 4 PSHELL 4 T
10 1.0 200.625300.3906
DVPREL1 5 PSHELL 5 T
10 1.0 200.500300.250
DVPREL1 6 PSHELL 6 T
10 1.0 200.375300.1406
DVPREL1 7 PSHELL 7 T
10 1.0 200.250300.0625
DVPREL1 8 PSHELL 8 T
10 1.0 200.125300.0156
$
$...IDENTIFY THE DESIGN RESPONSES
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1
$+ ATT2 ...
$ STATIC VON MISES STRESS IN SHELL ELEMENTS
DRESP1 2 S12 STRESS ELEM 9 1
2 3 4 5 6 7 8
$ STATIC DISPLACEMENT AT THE TIP
DRESP1 33 D1 DISP 3 19
$ FIRST NATURAL FREQUENCY
DRESP1 130 FFREQ FREQ 1
$ DEFINE THE RESPONSES IN THE DYNAMIC RESPONSE ANALYSIS
DRESP1 230 TDISP FRDISP 3 19
$
$...DEFINE THE WEIGHT RESPONSE TO BE USED AS THE OBJECTIVE FUNCTION:
DRESP1 35 W WEIGHT
$
$...DEFINE THE STATIC DESIGN CONSTRAINTS
$DCONSTR DCID RID LALLOW UALLOW
DCONSTR 10 2 -29000. 29000.
DCONSTR 10 33 -2. 2.
$...DEFINE THE NORMAL MODES DESIGN CONSTRAINT
DCONSTR30 130 1.5
$...DEFINE THE FREQUENCY RESPONSE DESIGN CONSTRAINTS
DCONSTR40 230 2.0
$
$...OVERRIDE OPTIMIZATION PARAMETER DEFAULTS (OPTIONAL)
Main Index Main Index
407 CHAPTER 7
Example Problems
DOPTPRM DESMAX 20 P11P215
$
PARAM NASPRT 1
PARAM POST -1
ENDDATA
Selected results from the .f06 file from this run are shown in Listing 7-6 and Listing 7-
7. The first set of data shows the print of final design information, including the
objective, the design variable values, the final property values, and the final responses.
As expected, the final design has the thickest elements and the root and the thickness
decrease monotonically along the span of the plate towards the tip.
The final responses show that only one response is close to its limiting value: the
displacement of the structure under the first static load. This single constraint
condition is therefore dictating the design of the entire structure to produce the
thickness distribution shown above.
Main Index Main Index


408
Listing 7-6 Final Results for DSOUG3
********************************************************************
* *
* *
* D E S I G N O P T I M I Z A T I O N *
* *
* *
********************************************************************
*******************************************
* *
* F I N A L A N A L Y S I S *
* *
*******************************************
******* ANALYSIS RESULTS BASED ON THE FINAL DESIGN *******
----- DESIGN OBJECTIVE -----
----------------------------------------------------------------------------------------------------
INTERNAL TYPE MINIMIZE
RESPONSE OF OR SUPERELEMENT SUBCASE
ID RESPONSE LABEL MAXIMIZE ID ID VALUE
----------------------------------------------------------------------------------------------------
1 DRESP1 WEIGHT MINIMIZE 0 0 3.7018E+01
----- DESIGN VARIABLES -----
---------------------------------------------------------------------------------------------------------
INTERNAL DESVAR LOWER UPPER
ID ID LABEL BOUND VALUE BOUND
---------------------------------------------------------------------------------------------------------
1 10 ALPH1 -1.0000E+20 3.6149E-01 1.0000E+20
2 20 ALPH2 -1.0000E+20 8.4961E-01 1.0000E+20
3 30 ALPH3 -1.0000E+20 2.1597E-01 1.0000E+20
----- DESIGNED PROPERTIES -----
----------------------------------------------------------------------------------------------------------
PROPERTY PROPERTY PROPERTY TYPE OF LOWER UPPER
TYPE ID NAME PROPERTY BOUND VALUE BOUND
----------------------------------------------------------------------------------------------------------
PSHELL 1 T DVPREL1 1.0000E-15 1.4271E+00 1.0000E+20
PSHELL 2 T DVPREL1 1.0000E-15 1.2702E+00 1.0000E+20
PSHELL 3 T DVPREL1 1.0000E-15 1.1202E+00 1.0000E+20
PSHELL 4 T DVPREL1 1.0000E-15 9.7685E-01 1.0000E+20
PSHELL 5 T DVPREL1 1.0000E-15 8.4028E-01 1.0000E+20
PSHELL 6 T DVPREL1 1.0000E-15 7.1046E-01 1.0000E+20
PSHELL 7 T DVPREL1 1.0000E-15 5.8739E-01 1.0000E+20
PSHELL 8 T DVPREL1 1.0000E-15 4.7106E-01 1.0000E+20
Main Index Main Index
409 CHAPTER 7
Example Problems
The summary of design cycle history in Listing 7-7 indicates that the initial design is
well within the imposed design requirements. This is indicated by the N/ A that is
shown in the MAXIMUM VALUE OF CONSTRAINT column which indicates that
----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
----- WEIGHT RESPONSE -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ROW COLUMN LOWER UPPER
ID ID LABEL ID ID BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
1 35 W 3 3 N/A 3.7018E+01 N/A
F I N A L A N A L Y S I S S U B C A S E = 1
----- DISPLACEMENT RESPONSES -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER UPPER
ID ID LABEL ID NO. BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
2 33 D1 19 3 -2.0000E+00 -2.0053E+00 N/A
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
3 2 S12 5 9 N/A 1.4778E+04 2.9000E+04
4 2 S12 6 9 N/A 1.4697E+04 2.9000E+04
F I N A L A N A L Y S I S S U B C A S E = 2
----- DISPLACEMENT RESPONSES -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER UPPER
ID ID LABEL ID NO. BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
5 33 D1 19 3 N/A 1.7815E+00 2.0000E+00
0 SUBCASE 1
F I N A L A N A L Y S I S S U B C A S E = 2
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
6 2 S12 2 9 N/A 1.4565E+04 2.9000E+04
F I N A L A N A L Y S I S S U B C A S E = 3
----- NORMAL MODE RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE MODE LOWER UPPER
ID ID LABEL NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
7 130 FFREQ 1 1.5000E+00 1.7120E+00 N/A
F I N A L A N A L Y S I S S U B C A S E = 4
----- FREQUENCY DISPLACEMENT RESPONSES -----
---------------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE GRID COMPONENT LOWER UPPER
ID ID LABEL ID NO. FREQUENCY BOUND VALUE BOUND
---------------------------------------------------------------------------------------------------------------------------
8 230 TDISP 19 3 1.6000E+00 N/A 1.1551E+00 2.0000E+00
9 230 TDISP 19 3 1.7000E+00 N/A 1.8253E+00 2.0000E+00
10 230 TDISP 19 3 1.8000E+00 N/A 1.2531E+00 2.0000E+00
Main Index Main Index


410
none of the constraints exceeded the -0.5 value that is the default for designating a
constraint as being active. After six iterations, a minimum weight design has been
achieved that satisfies all constraints and has reduced the structural weight from
78.375 to 37.018.
Listing 7-7 Design History for DSOUG3
Figure 7-4 depicts results from the two static subcases for the final design that have
produced MSC.Patran. This figure is representative of the many results can be
produced from results in the .op2 database that was generated based on the PARAM
POST -1 in the input data file of Listing 7-5. Other items in the input data file that
affect the amount of data written to the output database are the case control
commands, such as DISP=ALL and STRESS=ALL and PARAM NASPRT 1 that is
used to output data recovery results at every design cycle.
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 7
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 6
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 7.843750E+01 N/A
1 6.267220E+01 6.266749E+01 7.517695E-05 N/A
2 5.006994E+01 5.006892E+01 2.041863E-05 -4.491552E-01
3 4.000198E+01 4.000354E+01 -3.909719E-05 -1.562115E-01
4 3.661429E+01 3.661575E+01 -4.000583E-05 3.589535E-02
5 3.701775E+01 3.701763E+01 3.194575E-06 2.625108E-03
6 3.701763E+01 3.701763E+01 0.000000E+00 2.625108E-03
---------------------------------------------------------------------------------------------------------------
DESIGN VARIABLE HISTORY
---------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
---------------------------------------------------------------------------------------------------------------------------------
1 | 10 | ALPH1 | 1.0000E+00 : 8.0208E-01 : 6.4330E-01 : 5.1593E-01 : 3.6167E-01 : 3.6149E-01 :
2 | 20 | ALPH2 | 1.0000E+00 : 7.9649E-01 : 6.3445E-01 : 5.0541E-01 : 8.3706E-01 : 8.4961E-01 :
3 | 30 | ALPH3 | 1.0000E+00 : 7.9459E-01 : 6.3135E-01 : 5.0164E-01 : 2.0803E-01 : 2.1597E-01 :
---------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
---------------------------------------------------------------------------------------------------------------------------------
1 | 10 | ALPH1 | 3.6149E-01 :
2 | 20 | ALPH2 | 8.4961E-01 :
3 | 30 | ALPH3 | 2.1597E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 6.
Main Index Main Index
411 CHAPTER 7
Example Problems
Figure 7-4 Representative Displacement and Stress Results
for the Final Design of the Cantilevered Beam
X
Y
Z
X
Y
Z
X
Y
Z
X
Y
Z
X
Y
Z
X
Y
Z
X
Y
Z
X
Y
Z
Main Index Main Index


412
7.4 Stiffened Plate
An effective way to keep the number of independent design variables to a minimum
is by grouping designed elements by property type. A smaller set of independent
design variables decreases the cost associated with the sensitivity analysis, allows the
optimizer to perform more efficiently, and makes interpretation of the final results
much easier.
A simple example is shown in Figure 7-5 and includes a plate with a hat stiffener. The
design goal is to reduce the weight of the stiffened panel subject to stress and
displacement constraints under two separate static load conditions. The thickness of
the plate and the thickness of the hat stiffener are allowed to vary. The boundary
condition creates a simply supported condition with the plate also restrained in the x
direction along x=0.0. The first load case includes both uniaxial tension in the x-
direction and a vertical pressure load in the z-direction. The second load case is a
concentrated load applied in the +z direction at grid 10203, which is directly under the
hat. The example illustrates how the beam library can be utilized to simplify the
modelling and design tasks and how the beam offset relations can be adjusted as the
structural properties change.
Figure 7-5 Plate with a HAT Stiffener
10
5.
10
x
1
x
2
10
------
X
Y
Z
Main Index Main Index
413 CHAPTER 7
Example Problems
Analysis Model Description
The input file for this example is shown in Listing 7-8.
Listing 7-8 Input File for DSOUG4
ID MSC DSOUG4 $
TIME 10
SOL 200 $ OPTIMIZATION
CEND
$
TITLE = STATIC ANALYSIS OF A STIFFENED PLATE DSOUG4
ECHO = BOTH
DISP = ALL
STRESS = ALL
SPC = 1
ANALYSIS = STATICS
DSAPRT(START=1,END=LAST)=ALL
DESOBJ(MIN) = 15 $ OBJECTIVE FUNCTION DEFINITION
$ (MIN IS THE DEFAULT)
SUBCASE 1
LABEL = LOAD CONDITION 1
LOAD = 1
DESSUB = 100 $ CONSTRAINT DEFININITION
SUBCASE 2
LABEL = LOAD CONDITION 2
LOAD = 2
DESSUB = 200 $ CONSTRAINT DEFININITION
BEGIN BULK
param post -1
PARAM NASPRT 1
$
$----------------------------------------------------------------
$ ANALYSIS MODEL:
16 CQUAD4 elements to model uniform thickness base plate
4 CBAR elements to model the hat stiffener. The HAT type is selected on the PBARL
entry.
Material: E = 1.0E7 psi
Poisson ratio =
0.33
Loadings: Subcase 1: In-plane load of Nx = 1000 lbs/ in
Uniform pressure load of 50 psi over the plate in the
positive z-direction
Subcase 2: Lumped vertical load of 10,000 lbs at Grid 10203
Weight density 0.283 lbs

in
3
=
Main Index Main Index


414
$----------------------------------------------------------------
$
GRID 10000 0.0 0.0 0.0
GRID 10001 2.5 0.0 0.0
GRID 10002 5.0 0.0 0.0
GRID 10003 7.5 0.0 0.0
GRID 10004 10.0 0.0 0.0
GRID 10100 0.0 2.5 0.0
GRID 10101 2.5 2.5 0.0
GRID 10102 5.0 2.5 0.0
GRID 10103 7.5 2.5 0.0
GRID 10104 10.0 2.5 0.0
GRID 10200 0.0 5.0 0.0
GRID 10201 2.5 5.0 0.0
GRID 10202 5.0 5.0 0.0
GRID 10203 7.5 5.0 0.0
GRID 10204 10.0 5.0 0.0
GRID 10300 0.0 7.5 0.0
GRID 10301 2.5 7.5 0.0
GRID 10302 5.0 7.5 0.0
GRID 10303 7.5 7.5 0.0
GRID 10304 10.0 7.5 0.0
GRID 10400 0.0 10.0 0.0
GRID 10401 2.5 10.0 0.0
GRID 10402 5.0 10.0 0.0
GRID 10403 7.5 10.0 0.0
GRID 10404 10.0 10.0 0.0
$
CQUAD4 1 1 10000 10001 10101 10100
CQUAD4 2 1 10001 10002 10102 10101
CQUAD4 3 1 10002 10003 10103 10102
CQUAD4 4 1 10003 10004 10104 10103
CQUAD4 5 1 10100 10101 10201 10200
CQUAD4 6 1 10101 10102 10202 10201
CQUAD4 7 1 10102 10103 10203 10202
CQUAD4 8 1 10103 10104 10204 10203
CQUAD4 9 1 10200 10201 10301 10300
CQUAD4 10 1 10201 10202 10302 10301
CQUAD4 11 1 10202 10203 10303 10302
CQUAD4 12 1 10203 10204 10304 10303
CQUAD4 13 1 10300 10301 10401 10400
CQUAD4 14 1 10301 10302 10402 10401
CQUAD4 15 1 10302 10303 10403 10402
CQUAD4 16 1 10303 10304 10404 10403
$
CBAR 31 3 10200 10201 0.0 1.0 0.0
1.575 1.575
CBAR 32 3 10201 10202 0.0 1.0 0.0
1.575 1.575
CBAR 33 3 10202 10203 0.0 1.0 0.0
1.575 1.575
CBAR 34 3 10203 10204 0.0 1.0 0.0
1.575 1.575
$
PSHELL 1 1 0.15 1
PSHELL 2 1 0.2 1
Main Index Main Index
415 CHAPTER 7
Example Problems
PBARL 31 hat
3.0 0.1 2.0 0.9
$
MAT1 1 1.0E+7 0.33 0.283
$
FORCE 1 10004 2000.0 1.0 0.0 0.0
FORCE 1 10104 2000.0 1.0 0.0 0.0
FORCE 1 10204 2000.0 1.0 0.0 0.0
FORCE 1 10304 2000.0 1.0 0.0 0.0
FORCE 1 10404 2000.0 1.0 0.0 0.0
FORCE 2 10203 10000.0 0.0 0.0 1.0
PLOAD2 1 50. 1 THRU 16
$
SPC1 1 1236 10000
SPC1 1 136 10100 10300 10400
SPC1 1 36 10001 10002 10003 10004 10104
SPC1 1 36 10401 10402 10403 10404 10304
SPC1 1 3 10204
SPC1 1 13 10200
SPC1 1 6 10101 10102 10103 10104
SPC1 1 6 10301 10302 10303 10304
$
PARAM GRDPNT 1
PARAM WTMASS 0.00259
PARAM AUTOSPC YES
$
$-----------------------------------------------------------------
$ DESIGN MODEL:
$-----------------------------------------------------------------
$
$...Define the design variables:
$
$DESVAR ID LABEL XINIT XLB XUB DELXV
DESVAR 1 T-PLATE 0.15 0.001 10.0
DESVAR 2 T-BOX 1.0 0.001 10.0
$
$...Relate the design variables to analysis model properties
$ (linear relations so use DVPREL1)
$
$...Express shell thicknesses as functions of x1 x2:
$DVPREL1 ID TYPE PID FID PMIN PMAX C0 +
$+ DVIDD1 COEF1 DVID2 COEF2 ...
DVPREL1 1 PSHELL 1 T 0.01 +DP1
+DP1 1 1.0
$
$...Express HAT THICKNESS as a function of x2:
DVPREL1 3 PBARL 3 DIM2
2 0.1
$. EXPRESS BOX OFFSET LOCATIONS AS A FUNCTION OF PLATE THICKNESS AND
$ FIXED HAT DIMENSIONS
DVCREL110 CBAR 31 W3A 1.5
1 0.5
DVCREL120 CBAR 32 W3A 1.5
1 0.5
DVCREL130 CBAR 33 W3A 1.5
1 0.5
Main Index Main Index


416
DVCREL140 CBAR 34 W3A 1.5
1 0.5
DVCREL150 CBAR 31 W3B 1.5
1 0.5
DVCREL160 CBAR 32 W3B 1.5
1 0.5
DVCREL170 CBAR 33 W3B 1.5
1 0.5
DVCREL180 CBAR 34 W3B 1.5
1 0.5
$...Identify the design responses:
$
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1 +
$+ ATT2 ...
DRESP1 1 SBARA STRESS PBAR 7 3
DRESP1 2 SBARB STRESS PBAR 14 3
DRESP1 3 S13 STRESS PSHELL 9 1
DRESP1 6 S16 STRESS PSHELL 17 1
DRESP1 13 D1 DISP 3 10302
DRESP1 14 D2 DISP 3 10203
DRESP1 15 W WEIGHT
$
$...Place bounds on the responses:
$
$DCONSTR DCID RID LALLOW UALLOW
DCONSTR 10 1 -25000. 25000.
DCONSTR 10 2 -25000. 25000.
DCONSTR 10 3 -25000. 25000.
DCONSTR 10 6 -25000. 25000.
DCONSTR 20 13 -0.1 0.1
DCONSTR 30 14 -0.03 0.03
$
$DCONADD DCID DC1 DC2 ...
$ summed constraint set for subcase 1
DCONADD 100 10 20
$ summed constraint set for subcase 2
DCONADD 200 10 30
$
$...Optional override of optimization parameters:
$
DOPTPRM IPRINT 7 DESMAX 20 DELP 0.5 P1 1 +
+ P2 15
$
ENDDATA
Design Model Description
Objective: Structural weight minimization
Design
variables:
Plate thickness and hat stiffener thickness
Constraints:
Main Index Main Index
417 CHAPTER 7
Example Problems
The DVPREL1 entries express the analysis model properties in terms of design
variables:
Eq. 7-4
The third relationship adjusts the beam offset of the for the hat section to be one-half
the height of the hat stiffener plus one half the design plate thickness.
In order for a structural response to be used either as an objective or a constraint, it
first must be identified on a DRESPi Bulk Data entry. The DRESP1 entries 1 and 2, for
example, identify the maximum stress at ends A and B of a bar element. (Plot codes
are used in the ATTA fields of the DRESP1 entry. For the BAR element, Item 7 is the
maximum stress at end A, and Item 14 is the maximum stress at end B. (See Item
Codes on page 821 of the MSC.Nastran Quick Reference Guide for a list of these plot
codes.) The ATTi fields on these entries identify the property ID; here it is 3. DRESP1
number 1 indicates that the maximum end A stress for every BAR element in PBAR
group 3 is to be used in the design model for a total of four such responses. DRESP1
number 2 identifies similar information for end B of the BAR elements.
Similarly, DRESP1 entries 3 and 6 identify the von Mises stresses at Z1 and Z2 for half
the elements in the the plate (only one side of the plate needs the responses in this case
due to the symmetry of the structure and the loading about line y=5.0. Separate
DRESP1 entries identify displacement responses at grid points. These responses are
constrained by the bounds set using a corresponding set of DCONSTR entries. The
DCONADD entry is used to accumulate the design constraints that are appropriate
for each subcase. Note that the stress constraints are used in both subcases while a
distinct displacement constraint is applied for the two loading conditions.
Strength: von Mises stress < 25,000 psi at the centroid of plate
elements
Maximum stress < 25,000 psi at both ends of rod elements.
Displaceme
nt:
Vertical displacement at grid point 10302 for Subcase 1 <
0.1 inches
Vertical displacement at grid point 10203 for Subcase 2 <
0.03 inches
t
plate
1.0 X
1
=
t
hat
0.1 X
2
=
w3 1.5 0.5 X
1
+ =
Main Index Main Index


418
Extensive output for this example is presented in Design Optimization Output on
page 329. The SUMMAY OF DESIGN CYCLE HISTORY is repeated here Listing 7-9
and it is seen that the optimization task required 3 design cycles with 4 finite element
analyses. Note that it was possible to both overcome design constraints while
reducing the weight objective function. The MAXIMUM VALUE OF CONSTRAINT
column shows the original design had a response that exceeded its limit by over 21%
while the final design has a maximum violation that is much less than the 0.5% by the
GMAX parameter. The weight decreases from 6.962 to 6.733.
Listing 7-9 Design History Output for DSOUG4

0
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 5
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 4
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 6.961800E+00 2.191384E-01
1 6.793607E+00 6.793794E+00 -2.751337E-05 7.363109E-02
2 6.726397E+00 6.726370E+00 3.969880E-06 3.968594E-03
3 6.733279E+00 6.733277E+00 2.832720E-07 1.492187E-05
4 6.733277E+00 6.733277E+00 0.000000E+00 1.492187E-05
---------------------------------------------------------------------------------------------------------------
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | T-PLATE | 1.5000E-01 : 1.0832E-01 : 1.1240E-01 : 1.1264E-01 : 1.1264E-01 :
2 | 2 | T-BOX | 1.0000E+00 : 1.3833E+00 : 1.3136E+00 : 1.3136E+00 : 1.3136E+00 :
Main Index Main Index
419 CHAPTER 7
Example Problems
7.5 Shape Optimization of a Culvert
This example considers shape optimization using the direct input of shapes method.
In order to use this method, sets of displacement vectors are first generated using a
separate auxiliary model analysis. These displacement vectors are then DBLOCATEd
in a subsequent design optimization run and used to generate a set of shape basis
vectors. The optimizer then seeks to find the best combination of these shape basis
vectors.
Problem Description
Figure 7-6 shows the finite element model of one half of a symmetric culvert structure
(symmetry exists with respect to the y-axis.) The structure is made of steel, and has
been modeled using CQUAD4 elements in plane strain. This example has been taken
from Shape Optimal Design Using Isoparametric Elements (see Reference 1.).
Figure 7-6 Initial Culvert Design
With the bottom surface fixed, pressure loads are applied on the top surface. The
design task is to minimize the volume of the structure by changing the shape of the
initially circular hole, subject to von Mises stress constraints over the interior.
X
Y
Z 1 2 3 4
5 6 7
8
9 10 11
12
13
14
15
16 17 18 19
20
21
22
23 24 25 26
27
28
29
30 31 32 33
34 35 36 37 38 39 40
1 2 3
4 5 6
7 8 9
10
11
12
13
14
15
16
17
18
19 20 21
22 23 24
25 26 27
X
Y
Z X
Y
Z
2.73+04
5.89+03
X
Y
Z
Main Index Main Index


420
Modeling Considerations
The process of shape basis vector generation is closely associated with the nature of
the design problem itself. For example, since the goal here is weight minimization, the
shape basis vectors must clearly be able to reduce the volume of the structure. Were
they to simply redistribute material, the weight would not change and the optimizer
would be able to make little progress.
Furthermore, since the redesign is subject to von Mises stresses, we can use the initial
stress distribution to help us in the selection of appropriate basis vectors. This is also
shown in Figure 7-6. Here we note that the maximum von Mises stresses occur
around the lower portion of the circular hole. The stresses are not equally distributed
either. This tells us that a circular boundary does not provide the optimal design.
Thus, our shape basis vectors must contain other than just radial components.
We also need to decide how many design variables (or shape basis vectors) are
needed. The decision is not unique because the selection can be affected by
experience, various functional and manufacturing requirements, or aesthetic
requirements. However, the shape basis vectors should yield as much generality as
possible, consistent with our design goals.
Creation of Auxiliary Model, and Generation of Basis
Vectors
The top left frame in Figure 7-7 shows the auxiliary model geometry. Note that it has
the same geometry as the primary structure (Figure 7-6), but with different boundary
and loading conditions. The input data file for this model is given in Listing 7-10.
.
Main Index Main Index
421 CHAPTER 7
Example Problems
Figure 7-7 Auxiliary Model
Outside edges of the culvert are fixed in the auxiliary model to satisfy straight edge
requirements. Grid points 1, 2, 3 on the bottom are allowed to move in the x-direction
and grid points 13, 20, 27 on the symmetry line can move along the y-direction. Grids
5, 9, 14, 15, and 16 on the hole boundary are allowed to move as well. Along this hole
boundary, six CBAR elements have been added to help smooth the applied loading
effects. (It is important to allow z-rotations along this boundary.)
To generate a set of displacements to be used as basis vectors, we can statically load
each grid in a direction normal to the boundary as shown in Figure 7-7. Note from
Listing 7-10 that this can be performed in Solution 101 (or in Solution 200, with
OPTEXIT = 2). This will result in seven displacement vectors, one corresponding to
X
Y
Z X
Y
Z
X
Y
Z
1.52-02
X
Y
Z
X
Y
Z 1.54-02 X
Y
Z
X
Y
Z
2.68-03
X
Y
Z
Main Index Main Index


422
each load case. Three of these vectors are also shown in Figure 7-7: the first with a load
applied at grid 13, the second at grid 16, and the third at grid 1. The database is saved
through the use of the SCR=NO or SCR=MINI keyword when submitting the
auxiliary model job.
Listing 7-10 Auxiliary Model for the DSOUG5 Example (File DSOUG5A)
ID MSC, DSOUG5A $
$ This is the external auxiliary model file for the shape
$ optimization of a culvert example. This deck must be
$ run first and the results saved to the datbase before the
$ culvert example, dsoug5.dat, is run.
TIME 10
SOL 101
CEND
TITLE=Culvert Example Using External Auxiliary Model DSOUG5
SUBTITLE=The External Auxiliary Model
SPC=25
$
$ seven load cases
$
SUBCASE 1
LOAD=100
DISP=ALL
SUBCASE 2
LOAD=101
DISP=ALL
SUBCASE 3
LOAD=102
DISP=ALL
SUBCASE 4
LOAD=103
DISP=ALL
SUBCASE 5
LOAD=104
DISP=ALL
SUBCASE 6
LOAD=105
DISP=ALL
SUBCASE 7
LOAD=106
DISP=ALL
BEGIN BULK
PARAM,POST,0
param,newseq,-1
$
$ The same GRID and CQUAD4 entries as the primary structure
$
GRID, 1,, 3.00000, 0.00000,.00
GRID, 2,, 4.00000, 0.00000,.00
GRID, 3,, 5.00000, 0.00000,.00
. . . . . .
. . . . . .
(see optimization input files)
Main Index Main Index
423 CHAPTER 7
Example Problems
. . . . . .
. . . . . .
GRID, 39,, 2.50000, 5.19600,.00
GRID, 40,, 3.00000, 5.19600,.00
CQUAD4, 1,101, 1, 2, 6, 5
CQUAD4, 2,101, 2, 3, 7, 6
CQUAD4, 3,101, 3, 4, 8, 7
. . . . . .
. . . . . .
(see optimization input files)
. . . . . .
. . . . . .
CQUAD4, 25,101, 30, 31, 38, 37
CQUAD4, 26,101, 31, 32, 39, 38
CQUAD4, 27,101, 32, 33, 40, 39
PSHELL,101,102,.44
MAT1,102,2.+7,,.3
$
$ Additional CBAR elements maintain smoothness of the circular boundary
$
CBAR,31,1,13,14,,1.0
CBAR,32,1,14,15,,1.0
CBAR,33,1,15,16,,1.0
CBAR,34,1,16, 9,,1.0
CBAR,35,1, 9, 5,,1.0
CBAR,36,1,5 , 1,,1.0
PBAR 1 102 20.0 1.0 1.0
$
$ Seven load cases
$
FORCE,100,13,0,1.e5,0.,1.,0.
FORCE,101,14,0,1.e5,0.259,.9659
FORCE,102,15,0,1.e5,0.5,0.866,0.0
FORCE,103,16,0,1.e5,1.,1.,0.
FORCE,104,9,0,1.e5,0.866,0.5,0.0
FORCE,105,5,0,1.e5,0.9659,0.259
FORCE,106,1,0,1.e5,1.,0.,0.
$
$ Boundary conditions satisfy functional and manufacturing requirements
$
SPC1,25,345,1,THRU,40
SPC1,25,6,2,THRU,4
SPC1,25,6,6,THRU,8
SPC1,25,6,10,THRU,12
SPC1,25,6,17,THRU,19
SPC1,25,6,20,THRU,26
SPC1,25,6,27,THRU,33
SPC1,25,6,34,THRU,40
SPC1,25,12,33,THRU,40
SPC1,25,12,4,8,12,19,26
SPC1,25,1,13,20,27
SPC1,25,2,1,2,3
ENDDATA
Main Index Main Index


424
Design Optimization Input
In the optimization run, the displacement matrix containing the seven displacement
vectors from the Auxiliary Model analysis is retrieved using the DBLOCATE
statement. The optimization input file is shown in Listing 7-11.
Listing 7-11 Input File for DSOUG5
$ DSOUG5.DAT
$ This is the input deck for the shape optimization of a culvert
$ example. It DBLOCATE's data from a prior auxiliary model run,
$ file DSOUG5A.DAT. That job must be run first and the results
$ saved to the database before this one is run.
$
$ FMS section for retrieving the auxiliary displacement matrix
$ For Linux/Unix platforms, the MASTER file name must be lower case.
$
ASSIGN F1_AUX='dsoug5a.MASTER'
dblocate datablk=(ug/ugd,geom1/geom1d,geom2/geom2d) ,
logical=f1_aux
ID MSC, DSOUG5 $
SOL 200 $
TIME 100
CEND
TITLE=CULVERT EXAMPLE USING EXTERNAL AUXILIARY STRUCTURE DSOUG5
SUBTITLE=THE PRIMARY STRUCTURE
ANALYSIS = STATICS
SPC=25
LOAD=1
DISP=ALL
STRESS=all
DESSUB = 10
desobj = 5
BEGIN BULK
PARAM,POST,-1
param,cdif,no
PARAM,NEWSEQ,-1
GRID, 1,, 3.00000, 0.00000,.00
GRID, 2,, 4.00000, 0.00000,.00
GRID, 3,, 5.00000, 0.00000,.00
GRID, 4,, 6.00000, 0.00000,.00
GRID, 5,, 2.89464, 0.78478,.00
GRID, 6,, 3.79369, 0.75885,.00
GRID, 7,, 4.69274, 0.73293,.00
GRID, 8,, 5.59178, 0.70700,.00
GRID, 9,, 2.60164, 1.49178,.00
GRID, 10,, 3.46229, 1.46585,.00
GRID, 11,, 4.32293, 1.43993,.00
GRID, 12,, 5.18357, 1.41400,.00
GRID, 13,, 0.00000, 3.00000,.00
GRID, 14,, 0.78478, 2.89464,.00
GRID, 15,, 1.49178, 2.60164,.00
GRID, 16,, 2.12100, 2.12100,.00
GRID, 17,, 3.00578, 2.12100,.00
GRID, 18,, 3.89057, 2.12100,.00
Main Index Main Index
425 CHAPTER 7
Example Problems
GRID, 19,, 4.77535, 2.12100,.00
GRID, 20,, 0.00000, 3.73200,.00
GRID, 21,, 0.68985, 3.66176,.00
GRID, 22,, 1.32785, 3.46643,.00
GRID, 23,, 1.91400, 3.14600,.00
GRID, 24,, 2.67052, 3.14600,.00
GRID, 25,, 3.42704, 3.14600,.00
GRID, 26,, 4.18357, 3.14600,.00
GRID, 27,, 0.00000, 4.46400,.00
GRID, 28,, 0.59493, 4.42888,.00
GRID, 29,, 1.16393, 4.33122,.00
GRID, 30,, 1.70700, 4.17100,.00
GRID, 31,, 2.33526, 4.17100,.00
GRID, 32,, 2.96352, 4.17100,.00
GRID, 33,, 3.59178, 4.17100,.00
GRID, 34,, 0.00000, 5.19600,.00
GRID, 35,, 0.50000, 5.19600,.00
GRID, 36,, 1.00000, 5.19600,.00
GRID, 37,, 1.50000, 5.19600,.00
GRID, 38,, 2.00000, 5.19600,.00
GRID, 39,, 2.50000, 5.19600,.00
GRID, 40,, 3.00000, 5.19600,.00
CQUAD4, 1,101, 1, 2, 6, 5
CQUAD4, 2,101, 2, 3, 7, 6
CQUAD4, 3,101, 3, 4, 8, 7
CQUAD4, 4,101, 5, 6, 10, 9
CQUAD4, 5,101, 6, 7, 11, 10
CQUAD4, 6,101, 7, 8, 12, 11
CQUAD4, 7,101, 9, 10, 17, 16
CQUAD4, 8,101, 10, 11, 18, 17
CQUAD4, 9,101, 11, 12, 19, 18
CQUAD4, 10,101, 13, 14, 21, 20
CQUAD4, 11,101, 14, 15, 22, 21
CQUAD4, 12,101, 15, 16, 23, 22
CQUAD4, 13,101, 20, 21, 28, 27
CQUAD4, 14,101, 21, 22, 29, 28
CQUAD4, 15,101, 22, 23, 30, 29
CQUAD4, 16,101, 27, 28, 35, 34
CQUAD4, 17,101, 28, 29, 36, 35
CQUAD4, 18,101, 29, 30, 37, 36
CQUAD4, 19,101, 16, 17, 24, 23
CQUAD4, 20,101, 17, 18, 25, 24
CQUAD4, 21,101, 18, 19, 26, 25
CQUAD4, 22,101, 23, 24, 31, 30
CQUAD4, 23,101, 24, 25, 32, 31
CQUAD4, 24,101, 25, 26, 33, 32
CQUAD4, 25,101, 30, 31, 38, 37
CQUAD4, 26,101, 31, 32, 39, 38
CQUAD4, 27,101, 32, 33, 40, 39
FORCE 1 34 0 1250. -1.
FORCE 1 35 0 2500. -1.
FORCE 1 36 0 2500. -1.
FORCE 1 37 0 2500. -1.
FORCE 1 38 0 2500. -1.
FORCE 1 39 0 2500.00 -1.
FORCE 1 40 0 1250. -1.
Main Index Main Index


426
PSHELL,101,102,.44
MAT1,102,2.+7,,.3,0.731-3
SPC1,25,3456,1,THRU,40
SPC1,25,12,1,THRU,4
SPC1,25,1,13,20,27,34
$
$ design model
$
desvar 1 b1 3. -1.e6 1.e6 .25
desvar 2 b2 3. -1.e6 1.e6 .25
desvar 3 b3 3. -1.e6 1.e6 .25
desvar 4 b4 3. -1.e6 1.e6 .25
desvar 5 b5 3. -1.e6 1.e6 .25
desvar 6 b6 3. -1.e6 1.e6 .25
desvar 7 b7 3. -1.e6 1.e6 .25
$
$ A DVSHAP entry defines a shape basis vector by associating one design
$ variable to a dblocated displacement vector.
$
dvshap 1 1 66.773
dvshap 2 2 117.35
dvshap 3 3 216.33
dvshap 4 4 443.55
dvshap 5 5 220.89
dvshap 6 6 115.69
dvshap 7 7 65.669
dresp1 5 volume volume
dresp1 2 von-mis stress pshell 9 101
DCONSTR 10 2 3.100e4
doptprm DESMAX 25 APRCOD 1
param,nasprt,1
ENDDATA
Since each shape basis vector is defined in terms of a single displacement vector, seven
DESVAR and DVSHAP entries are used to define seven shape basis vectors. The
seven DVSHAP entry scaling factors have been selected such that the maximum
component of each shape basis vector is unity. For example, DVSHAP entry number
1 identifies the first shape basis vector in terms of the first DBLOCATE'd displacement
vector (1 in field 3). The 1 in field 2 indicates that design variable number 1 is to be the
multiplier of this vector. Furthermore, since the maximum component of this vector
is 1.0/ 66.773 (as determined from a manual inspection of the data), the scaling factor
has been given as 66.773, effectively unit normalizing the vector.
Each design variable acts as a multiplying coefficient of a shape basis vector. Initial
values are somewhat arbitrary and have been selected as 3.0 here. However, when
combined with 20% allowable move limits (0.2 in field 7 of the DESVAR entries), it can
be seen that part of the reason for the choice is that these initial values provide for
Main Index Main Index
427 CHAPTER 7
Example Problems
reasonable move limits on the initial design. Lower and upper limits of -1.0E6 and
+1.0E6 indicate that the design variables are to be considered effectively unbounded
during optimization.
Figure 7-5 shows a representation of the shapes at the end of design cycles 1, 2, 3 and
5 represented as deformation relative to the original grid. Viewing these in
MSC.Patran can provide guidance in the qualitative behavior of the shape changes as
a function of design cycle.
Figure 7-8 Designed Shapes for the Culvert as a Function of Design Cycle
X
Y
Z
3.28-01
0. X
Y
Z
X
Y
Z
1.25+00
0. X
Y
Z
X
Y
Z
1.15+00
0. X
Y
Z X
Y
Z
1.08+00
0. X
Y
Z
Design Cycle 3 Design Cycle 5
Design Cycle 1 Design Cycle 2
Main Index Main Index


428
Listing 7-12 shows that the optimization task was completed in five design cycles and
that the volume was decreased from 7.215 to 5.853, a decrease of 18.9% in volume (and
weight) while still satisfying the stress limits.
Listing 7-12 Design History Output for DSOUG5
The final shape is plotted in the left half of Figure 7-9 while the final stress contour is
plotted in the right half of the same figure.
0
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 6
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 5
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 7.214704E+00 -1.021607E-01
1 6.714619E+00 6.694779E+00 2.963468E-03 1.423551E-02
2 6.156898E+00 6.163190E+00 -1.020956E-03 -1.690115E-02
3 5.842054E+00 5.839865E+00 3.747831E-04 1.827577E-02
4 5.857889E+00 5.857942E+00 -9.035413E-06 1.642956E-03
5 5.852680E+00 5.852699E+00 -3.258921E-06 2.008821E-03
---------------------------------------------------------------------------------------------------------------
1 CULVERT EXAMPLE USING EXTERNAL AUXILIARY STRUCTURE DSOUG5 SEPTEMBER 14, 2001 MSC.NASTRAN 4/ 9/01 PAGE 95
THE PRIMARY STRUCTURE
0
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | B1 | 3.0000E+00 : 3.7500E+00 : 4.6875E+00 : 4.5473E+00 : 3.7210E+00 : 2.7907E+00 :
2 | 2 | B2 | 3.0000E+00 : 3.7500E+00 : 4.6875E+00 : 4.8695E+00 : 6.0869E+00 : 7.3548E+00 :
3 | 3 | B3 | 3.0000E+00 : 3.7466E+00 : 4.6833E+00 : 5.4171E+00 : 4.7637E+00 : 4.2786E+00 :
4 | 4 | B4 | 3.0000E+00 : 2.9473E+00 : 2.8163E+00 : 3.0271E+00 : 3.0831E+00 : 3.0930E+00 :
5 | 5 | B5 | 3.0000E+00 : 3.7500E+00 : 4.2426E+00 : 3.5757E+00 : 3.6450E+00 : 3.8176E+00 :
6 | 6 | B6 | 3.0000E+00 : 3.7500E+00 : 4.6870E+00 : 5.3197E+00 : 5.2333E+00 : 4.9588E+00 :
7 | 7 | B7 | 3.0000E+00 : 3.7500E+00 : 4.6875E+00 : 5.8594E+00 : 5.7831E+00 : 5.9582E+00 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 5.
Main Index Main Index
429 CHAPTER 7
Example Problems
Figure 7-9 Final Culvert Design
Main Index Main Index


430
7.6 Analytic Boundary Shapes
This example illustrates the use of the analytic boundary shapes method in shape
optimal design. In this method, the entire modeling task can be written using the
MSC.Nastran input file alone, without the need for a modeling pre- and
postprocessor. Analytic Boundary Shape on page 172 includes a checklist for
setting up the design model using this method. You may want to refer to that section
in connection with this example.
To use this method, you need to define auxiliary models over the boundaries of the
structure. When constrained and loaded, these boundary models produce static
deformations that can be used to describe shape variations over the boundaries. The
code then interpolates this information to the interior grids, resulting in basis vectors
for shape optimization. A static analysis is used for this interpolation.
Problem Description
Figure 7-10 shows the initial structure. It is a simple cantilever, modeled with eighty
solid elements, fixed at the support and tip-loaded at the free end. The design goal is
to minimize the structure's weight subject to constraints that the element von Mises
stresses must be less than 200. We'll investigate minimizing the weight by tapering the
cantilever's shape. The initial stress distribution is shown in Figure 7-11. The
maximum stress is 183. and that there are large regions where the stress is
considerably less than the 200. limit.
Main Index Main Index
431 CHAPTER 7
Example Problems
Figure 7-10 Solid Cantilever
X
Y
Z
25.00
123456
50.00
123456
25.00
123456
123456
123456
123456
123456
X
Y
Z
Main Index Main Index


432
Figure 7-11 Solid Cantilever -- Initial Stress Distribution
Since the cantilever is oriented and loaded in an x-z plane, removing material from the
upper and lower surfaces is an effective way to reduce the weight, tapering the
cantilever from its root to its tip. Basis vectors describing this characteristic shape can
be easily generated using the analytic boundary shapes method. In this
Boundary Shape Changes Using Auxiliary Boundary
Models
Figure 7-12 shows the auxiliary boundary models that can be used to generate these
shapes. These disjoint models, one for the upper surface and one for the lower, are
built using QUAD4 elements.
X
Y
Z
1.83+02
5.15+00
X
Y
Z
Main Index Main Index
433 CHAPTER 7
Example Problems
Figure 7-12 Auxiliary Boundary Models
In the example, a total of six boundary shapes are computed using the auxiliary
model:
Unit enforced displacements are imposed at the root and the tip of the top
plate, resulting in a uniform variation along the top surface.
A unit displacement is enforced at the tip of the top plate while the root is
constrained in translation but allowed to rotate about the y-axis. This
produces a shape that is linear along the top surface.
An enforced is applied at the tip while the root is fully constrainted to
produce a cubic displacement along the top surface. Note that a quadratic
displacement could be produced by applying a moment at the tip of the
plate. This is left as an exercise for the reader.
The remaining three shapes are produced by imposing similar conditions on
the bottom surface while holding the top surface fixed.
X
Y
Z
Main Index Main Index


434
Listing 7-13 is an abbreviated input file for the combined analysis and optimization of
the primary structure, and the auxiliary boundary model specification. The complete
file can be found in the Test Problem Library as DSOUG6.DAT. The primary model
definition is for static analysis and is convertional, so its description will be omitted
here. The design model portion of the primary structure will be described shortly.
Main Index Main Index
435 CHAPTER 7
Example Problems
Listing 7-13 Selected Portions of Input File DSOUG6
ID MSC, DSOUG6 $
TIME 600 $
SOL 200 $
CEND
TITLE = CANTILEVERED BEAM - HEXA **** DSOUG6 ****
SUBTITLE = SHAPE OPTIMIZATION WITH ANALYTIC BOUNDARY SHAPES
DESOBJ = 15
DESSUB = 100
SUBCASE 100
ANALYSIS = STATICS
SPC = 1
LOAD = 1
DISPLACEMENT(plot) = ALL
stress(plot,corner) = all
AUXCASE
TITLE = AUXILIARY MODEL 1
AUXMODEL = 1
DISP(plot) = all
SUBCASE 200
SPC = 200
LOAD = 220
LABEL = LOWER - CONSTANT
SUBCASE 300
SPC = 300
LOAD = 330
LABEL = UPPER - CONSTANT
SUBCASE 400
SPC = 400
LOAD = 440
LABEL = LOWER - LINEAR
SUBCASE 500
SPC = 500
LOAD = 550
LABEL = UPPER -LINEAR
SUBCASE 600
SPC = 600
LOAD = 660
LABEL = LOWER - FLEXIBLE
SUBCASE 700
SPC = 700
LOAD = 770
LABEL = UPPER - FLEXIBLE
BEGIN BULK
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL:
$------------------------------------------------------------------------
$
PARAM AUTOSPC YES
PARAM POST -1
PARAM GRDPNT 0
PARAM MAXRATIO1.0E+8
$
GRID 1 0 0.0 0.0 0.0 0
Main Index Main Index


436
GRID 2 0 1.0 0.0 0.0 0
GRID 3 0 2.0 0.0 0.0 0
GRID 4 0 3.000 0.0 0.0 0
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
GRID 162 0 7.0 2.000 4.000 0
GRID 163 0 8.0 2.000 4.000 0
GRID 164 0 9.000 2.000 4.000 0
GRID 165 0 10.000 2.000 4.000 0
$GRDSET 456
CHEXA 1 1 1 2 13 12 34 35+EA 1
+EA 1 46 45
CHEXA 2 1 2 3 14 13 35 36+EA 2
+EA 2 47 46
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
CHEXA 80 1 120 121 132 131 153 154+EA 80
+EA 80 165 164
MAT1* 1 2.0680E+05 0.28999999166+MA 1
*MA 1 1.00000000 1.169999996E-05 +MB 1
*MB 1 1500000.00 1500000.00 68000.00 +MC 1
*MC 1
PSOLID 1 1 0 0 0 0
SPC 1 1 123456 0.0
SPC 1 12 123456 0.0
SPC 1 23 123456 0.0
SPC 1 34 123456 0.0
SPC 1 45 123456 0.0
SPC 1 56 123456 0.0
SPC 1 67 123456 0.0
SPC 1 78 123456 0.0
SPC 1 89 123456 0.0
SPC 1 100 123456 0.0
SPC 1 111 123456 0.0
SPC 1 122 123456 0.0
SPC 1 133 123456 0.0
SPC 1 144 123456 0.0
SPC 1 155 123456 0.0
SPC1 1 456 1 THRU 165
FORCE 1 143 0 0.5 0.0 0.0 -50.0
FORCE 1 154 0 1.0 0.0 0.0 -50.0
FORCE 1 165 0 0.5 0.0 0.0 -50.0
$
$------------------------------------------------------------------------
$ DESIGN MODEL:
$------------------------------------------------------------------------
$
PARAM,DESPCH,1
PARAM,NASPRT,1
$
$DESVAR,ID, LABEL, XINIT, XLB, XUB, DELXV
DESVAR 1 UPPERC 1.0 .00 7.00 0.1
DESVAR 2 LOWERC 1.0 .00 7.00 0.1
Main Index Main Index
437 CHAPTER 7
Example Problems
DESVAR 3 UPPERL 1.0 .00 7.00 0.1
DESVAR 4 LOWERL 1.0 .00 7.00 0.1
DESVAR 5 UPPERF 1.0 .00 7.00 0.1
DESVAR 6 LOWERF 1.0 .00 7.00 0.1
$
$DVBSHAP,DVID, AUXMID, COL1, SF1, COL2, SF2, ...
DVBSHAP 1 1 1 1.0
DVBSHAP 2 1 2 1.0
DVBSHAP 3 1 3 1.0
DVBSHAP 4 1 4 1.0
DVBSHAP 5 1 5 1.0
DVBSHAP 6 1 6 1.0
$
$DLINK, ID, DDVID, CO, CMULT, IDV1, C1, IDV2, C2, +
$+, IDV3, C3, ...
$LINK 1 2 1.0 1 1.0
$
$ BOUNDARY CONDITIONS FOR SHAPE INTERPOLATIONS:
$
$ ---TOP SURFACE:
$BNDGRID,C, GP1, GP2, GP3, GP4, GP5, GP6, GP7, +
$+, GP8, ...
BNDGRID 123 133 THRU 165
$
$ ---BOTTOM SURFACE:
BNDGRID 123 1 THRU 33
$
$ ---EXTERIOR SURFACES - INTERPOLATION IN X&Z DIRECTION ONLY:
BNDGRID 2 34 35 36 37 38 39 40
41 42 43 44
BNDGRID 2 56 57 58 59 60 61 62
63 64 65 66
BNDGRID 2 67 68 69 70 71 72 73
74 75 76 77
BNDGRID 2 89 90 91 92 93 94 95
96 97 98 99
BNDGRID 2 100 101 102 103 104 105 106
107 108 109 110
BNDGRID 2 122 123 124 125 126 127 128
129 130 131 132
$
$ ---TIP END:
BNDGRID 1 44 55 66 77 88 99 110
121 132
$
$ ---FIXED END:
BNDGRID 1 34 45 56 67 78 89 100
111 122
$
$ FORMULATE WEIGHT-BASED SYNTHETIC RESPONSE: F = 1.E5*W
DRESP1 1 WEIGHT WEIGHT
DRESP2 15 WE1000 1
+ DRESP1 1
DEQATN 1 F(A)=100000.*A
$
$ CONSTRAINTS ON VON MISES STRESSES:
Main Index Main Index


438
DRESP1 2 STRESS STRESS PSOLID 13 1
DRESP1 3 STRESS STRESS PSOLID 34 1
DRESP1 4 STRESS STRESS PSOLID 55 1
DRESP1 5 STRESS STRESS PSOLID 76 1
DRESP1 6 STRESS STRESS PSOLID 97 1
DRESP1 7 STRESS STRESS PSOLID 118 1
DRESP1 8 STRESS STRESS PSOLID 139 1
DRESP1 9 STRESS STRESS PSOLID 160 1
DRESP1 10 STRESS STRESS PSOLID 181 1
DCONSTR 100 2 200.
DCONSTR 100 3 200.
DCONSTR 100 4 200.
DCONSTR 100 5 200.
DCONSTR 100 6 200.
DCONSTR 100 7 200.
DCONSTR 100 8 200.
DCONSTR 100 9 200.
DCONSTR 100 10 200.
$
$ OVERRIDE OF OPTIMIZATION PARAMETERS
DOPTPRM DESMAX 20 P1 1 P2 15METHOD3
$
$------------------------------------------------------------------------
$ AUXILIARY BOUNDARY MODEL(S):
$------------------------------------------------------------------------
$
BEGIN BULK AUXMODEL=1
PARAM,PRGPST,NO
PARAM MAXRATIO1.0E+8
PARAM,AUTOSPC,YES
$
$ LOWER SURFACE:
CQUAD4 1000 2 1 2 13 12 0.0
CQUAD4 1001 2 2 3 14 13 0.0
CQUAD4 1002 2 3 4 15 14 0.0
CQUAD4 1003 2 4 5 16 15 0.0
CQUAD4 1004 2 5 6 17 16 0.0
CQUAD4 1005 2 6 7 18 17 0.0
CQUAD4 1006 2 7 8 19 18 0.0
CQUAD4 1007 2 8 9 20 19 0.0
CQUAD4 1008 2 9 10 21 20 0.0
CQUAD4 1009 2 10 11 22 21 0.0
CQUAD4 1010 2 12 13 24 23 0.0
CQUAD4 1011 2 13 14 25 24 0.0
CQUAD4 1012 2 14 15 26 25 0.0
CQUAD4 1013 2 15 16 27 26 0.0
CQUAD4 1014 2 16 17 28 27 0.0
CQUAD4 1015 2 17 18 29 28 0.0
CQUAD4 1016 2 18 19 30 29 0.0
CQUAD4 1017 2 19 20 31 30 0.0
CQUAD4 1018 2 20 21 32 31 0.0
CQUAD4 1019 2 21 22 33 32 0.0
$
$ UPPER SURFACE:
CQUAD4 950 2 133 144 145 134 0.0
CQUAD4 951 2 134 145 146 135 0.0
Main Index Main Index
439 CHAPTER 7
Example Problems
CQUAD4 952 2 135 146 147 136 0.0
CQUAD4 953 2 136 147 148 137 0.0
CQUAD4 954 2 137 148 149 138 0.0
CQUAD4 955 2 138 149 150 139 0.0
CQUAD4 956 2 139 150 151 140 0.0
CQUAD4 957 2 140 151 152 141 0.0
CQUAD4 958 2 141 152 153 142 0.0
CQUAD4 959 2 142 153 154 143 0.0
CQUAD4 960 2 144 155 156 145 0.0
CQUAD4 961 2 145 156 157 146 0.0
CQUAD4 962 2 146 157 158 147 0.0
CQUAD4 963 2 147 158 159 148 0.0
CQUAD4 964 2 148 159 160 149 0.0
CQUAD4 965 2 149 160 161 150 0.0
CQUAD4 966 2 150 161 162 151 0.0
CQUAD4 967 2 151 162 163 152 0.0
CQUAD4 968 2 152 163 164 153 0.0
CQUAD4 969 2 153 164 165 154 0.0
$
MAT1 11 2.1E+5 0.8E+5 0.3 0.00
PSHELL 2 11 0.20 11 0.0
$ USE ENFORCED DISPLACEMENTS TO CREATE A UNIFORM TRANSLATION ON THE
$ LOWER SURFACE
SPC1 200 123 1 12 23
SPC1 200 123 11 22 33
SPC1 200 123456 34 THRU 165
SPCD 220 1 3 1.0 12 3 1.0
SPCD 220 23 3 1.0
SPCD 220 11 3 1.0 22 3 1.0
SPCD 220 33 3 1.0
$ USE ENFORCED DISPLACEMENTS TO CREATE A UNIFORM TRANSLATION ON THE
$ UPPER SURFACE
SPC1 300 123 133 144 155
SPC1 300 123 143 154 165
SPC1 300 123456 1 THRU 132
SPCD 330 133 3 -1.0 144 3 -1.0
SPCD 330 155 3 -1.0
SPCD 330 143 3 -1.0 154 3 -1.0
SPCD 330 165 3 -1.0
$ USE ENFORCED DISPLACEMENTS TO CREATE A LINEAR TRANSLATION ON THE
$ LOWER SURFACE (FREE UP THE 5 ROTATION AT THE ROOT)
SPC1 400 12346 1 12 23
SPC1 400 12 11 22 33
SPC1 400 123456 34 THRU 165
SPCD 440 11 3 1.0 22 3 1.0
SPCD 440 33 3 1.0
SPC1 400 3 11 22 33
$ USE ENFORCED DISPLACEMENTS TO CREATE A LINEAR TRANSLATION ON THE
$ UPPER SURFACE (FREE UP THE 5 ROTATION AT THE ROOT)
SPC1 500 12346 133 144 155
SPC1 500 12 143 154 165
SPC1 500 123456 1 THRU 132
SPCD 550 143 3 -1.0 154 3 -1.0
SPCD 550 165 3 -1.0
SPC1 500 3 143 154 165
$ USE ENFORCED DISPLACEMENT AT TIP AND RESTRAINED ROOT TO CREATE A
Main Index Main Index


440
$ FLEXIBLE SHAPE ON THE LOWER SURFACE
SPC1 600 123456 1 12 23
SPC1 600 12 11 22 33
SPC1 600 123456 34 THRU 165
SPCD 660 11 3 1.0 22 3 1.0
SPCD 660 33 3 1.0
SPC1 600 3 11 22 33
$ USE ENFORCED DISPLACEMENT AT TIP AND RESTRAINED ROOT TO CREATE A
$ FLEXIBLE SHAPE ON THE UPPER SURFACT
SPC1 700 123456 133 144 155
SPC1 700 12 143 154 165
SPC1 700 123456 1 THRU 132
SPCD 770 143 3 -1.0 154 3 -1.0
SPCD 770 165 3 -1.0
SPC1 700 3 143 154 165
ENDDATA
Turning first to the auxiliary boundary model specification, we see that the boundary
model is defined in a special Bulk Data Section, appearing after the Bulk Data for the
primary model. The statement, BEGIN BULK AUXMODEL = 1, is used to indicate the
beginning of this section. This section essentially defines three components of the
model: its connectivity, its loads, and its boundary conditions. This is just like any
other MSC.Nastran model for static analysis.
The auxiliary boundary model connectivity is defined using CQUAD4 elements. The
geometry should not be redefined, since the primary model geometry is used. (If
necessary, additional grids can be included. For example, an additional grid may be
necessary for use as a rigid element connection point. This grid should then appear in
this Bulk Data Section.)
The structure is constrained using six SPC sets that create the six boundary shapes
identified above and use enforced motion (SPCDs) to supply the load. The first set of
the SPC/ SPCD conditions is explained here to help understand this concept. First the
grids at the root and tip of the lower plate are constrained in the translational degrees
of freedom as part of SPC SET 200. All of the remaining grids in the structure that are
not on the lower plate are constrained using SPC SET 200 in all six degrees of freedom
so that these grid locations will not move. SPCD SET 220 is then used to apply a unit
enforced displacement at the root and the tip in the z-direction. Note that the points
on the interior of the bottom plate are not constrained and are therefore free to move.
In this case the whole plate will move as a rigid body in the z-direction, thereby
providing the boundary shape. The Case Control for these models appears after the
primary model Case Control, in a section beginning with the label, AUXCASE. Each
subcase selects one of the six SPCD-defined loads, using a different boundary
condition for each. Static loading is always assumed for these auxiliary model
analyses.
Main Index Main Index
441 CHAPTER 7
Example Problems
The results of these six analyses are sets of six displacement vectors can be used to
describe the shape changes over the Primary Structure's boundary. However, these
boundary displacements must still be interpolated over the structure's interior in
order to form shape basis vectors for the entire structure. This is achieved in the
primary model section of the Bulk Data.
Shape Changes over the Interior
BNDGRID entries provide the boundary conditions for the shape interpolation steps.
The relation between BNDGRID entries and the auxiliary boundary model solutions
is similar to that of SPC's and SPCD's. The auxiliary boundary model solutions are
imposed as enforced displacements on the primary structure. This is like an SPCD.
These displacement degrees of freedom must be present on BNDGRID entries, as with
SPC entries. Degrees of freedom listed on the BNDGRID entries are either enforced
or fixed, depending on whether or not an auxiliary boundary model solution exists for
them. All other degrees of freedom are considered "free", and are solved for in the
interpolation.
The BNDGRID data in this example are seen to be in three sets. The first set is applied
to the grids along the top and bottom surface where it is seen that components 1,2 and
3 are identified, which means that these displacements will be determined from the
auxiliary model. The second set is applied to grids that are on the side faces of the
structure (excluding those at the top and bottom corners). These grids are restrained
from moving in the y-direction so that the sides of the structure do not bulge out or
contract. These side grids can move in the x and z directions. The third set of grids is
made up of points at the root and tip of the structure. Here, the shape is not allowed
to move in the x direction while the x and z directions are left free.
Figure 7-13 shows the six deformations that result on the primary model when the
displacement from the auxiliary model are applied.
Main Index Main Index


442
Figure 7-13 Shape Basis Vectors Shown as Deformations on the Primary Model
Modeling Summary
To summarize, auxiliary boundary models are created using additional Bulk Data
Sections, labeled using BEGIN BULK AUXMODEL = n. We can have as many of these
sections as necessary to suit our auxiliary boundary model needs. Boundary
conditions and loads are applied with AUXCASE-labelled Case Control Sections. The
resultant boundary deformations are then interpolated to the interior of the primary
structure using BNDGRID entries to define the boundaries. The resulting total
displacement vectors can now be combined in any way to yield basis vectors for shape
optimization.
Shape Basis Vectors
The shape basis vectors are defined using DVBSHAP Bulk Data entries. DVBSHAP
entries 1 through 6 relate design variables 1 through 6, respectively, to the six
displacement solutions in a one-to-one fashion. Thus, each shape basis vector is
simply 1.0 times each of the resultant displacement solutions.
X
Y
Z
1.00+00
0.
X
Y
Z
X
Y
Z
1.00+00
0.
X
Y
Z
X
Y
Z
1.00+00
0.
X
Y
Z
X
Y
Z
1.00+00
0.
X
Y
Z
X
Y
Z
1.00+00
0.
X
Y
Z
X
Y
Z
1.00+00
0.
X
Y
Z
X
Y
Z
1.00+00
0.
X
Y
Z
Main Index Main Index
443 CHAPTER 7
Example Problems
Design Task
The design task is to modify the shape of the structure to find the minimum weight
structure that satisfies the limit that the von Mises cannot exceed 200. It is seen that a
DRESP2 is used to magnify the weight by 1.0E5. This DRESP2 is then selected as the
objective. This step is not necessary in this case, but it does provide the optimizer with
a sizable objective. Nine DRESP1 entries are used to identify von Mises stress
responses at each corner, plus the center, of each CHEXA element. An upper bound
limit of 200. is applied to these responses using DCONSTR entries. From the
DOPTPRM entry, it is seen that the Sequential Quadratic Programming method of
optimization is selected using METHOD=3.
Optimization Results
Listing 7-14 shows the summary of design cycle history from the output file. In
thirteen design cycles, the objective has been reduced from 8.0 E+6 to 6.01 E+6. The
final shape is shown in Figure 7-14, and the corresponding stress distribution in
Figure 7-15 . It is seen that the final design is nearly symmetric and that much of the
stress along the top and bottom surfaces is at the limit of 200.
Main Index Main Index


444
Listing 7-14 Summary of Design Cycle History for DSOUG6
0 SUBCASE 100
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 13
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 12
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 8.000000E+06 -1.038307E-02
1 7.622278E+06 7.622278E+06 0.000000E+00 4.371643E-05
2 7.254322E+06 7.254322E+06 0.000000E+00 -1.355743E-04
3 6.847802E+06 6.847802E+06 0.000000E+00 -1.314545E-04
4 6.399672E+06 6.399672E+06 0.000000E+00 -1.700592E-04
5 6.285760E+06 6.285760E+06 0.000000E+00 1.011421E-02
6 6.212054E+06 6.212054E+06 0.000000E+00 1.214539E-02
7 6.171776E+06 6.171776E+06 0.000000E+00 8.254166E-03
8 6.119370E+06 6.119370E+06 0.000000E+00 8.320237E-03
9 6.082725E+06 6.082725E+06 0.000000E+00 6.020203E-03
10 6.039925E+06 6.039925E+06 0.000000E+00 6.038284E-03
11 6.008418E+06 6.008418E+06 0.000000E+00 4.509506E-03
12 6.006243E+06 6.006243E+06 0.000000E+00 -3.089905E-05
---------------------------------------------------------------------------------------------------------------
1 AUXILIARY MODEL 1 FEBRUARY 13, 2003 MSC.NASTRAN 2/12/03 PAGE 336

0 SUBCASE 100
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | UPPERC | 1.0000E+00 : 1.0070E+00 : 1.0028E+00 : 9.9867E-01 : 9.9434E-01 : 9.8921E-01 :
2 | 2 | LOWERC | 1.0000E+00 : 1.0070E+00 : 1.0028E+00 : 9.9867E-01 : 9.9434E-01 : 9.8921E-01 :
3 | 3 | UPPERL | 1.0000E+00 : 1.1000E+00 : 1.2100E+00 : 1.3310E+00 : 1.4641E+00 : 1.6105E+00 :
4 | 4 | LOWERL | 1.0000E+00 : 1.1000E+00 : 1.2100E+00 : 1.3310E+00 : 1.4641E+00 : 1.6105E+00 :
5 | 5 | UPPERF | 1.0000E+00 : 1.1000E+00 : 1.2100E+00 : 1.3310E+00 : 1.4641E+00 : 1.3989E+00 :
6 | 6 | LOWERF | 1.0000E+00 : 1.1000E+00 : 1.2100E+00 : 1.3310E+00 : 1.4641E+00 : 1.3177E+00 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | UPPERC | 9.8574E-01 : 9.8194E-01 : 9.7848E-01 : 9.7548E-01 : 9.7295E-01 : 9.7055E-01 :
2 | 2 | LOWERC | 9.8573E-01 : 9.8193E-01 : 9.7846E-01 : 9.7546E-01 : 9.7254E-01 : 9.7019E-01 :
3 | 3 | UPPERL | 1.7423E+00 : 1.8656E+00 : 1.9817E+00 : 2.0808E+00 : 2.1755E+00 : 2.2567E+00 :
4 | 4 | LOWERL | 1.7697E+00 : 1.8848E+00 : 1.9997E+00 : 2.0975E+00 : 2.1899E+00 : 2.2698E+00 :
5 | 5 | UPPERF | 1.2590E+00 : 1.1331E+00 : 1.0198E+00 : 9.1782E-01 : 8.2604E-01 : 7.4344E-01 :
6 | 6 | LOWERF | 1.1859E+00 : 1.0673E+00 : 9.6060E-01 : 8.6454E-01 : 7.7808E-01 : 7.0027E-01 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 12 : 13 : 14 : 15 : 16 : 17 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | UPPERC | 9.6874E-01 :
2 | 2 | LOWERC | 9.6842E-01 :
3 | 3 | UPPERL | 2.3159E+00 :
4 | 4 | LOWERL | 2.3280E+00 :
5 | 5 | UPPERF | 6.6909E-01 :
6 | 6 | LOWERF | 6.3025E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 12.
Main Index Main Index
445 CHAPTER 7
Example Problems
Figure 7-14 Solid Cantilever -- Final Shape
X
Y
Z
P
R
2
Z
R
1
X
Y
Z
Main Index Main Index


446
Figure 7-15 Solid Cantilever -- Final Stress Distribution
X
Y
Z
P
R
2
Z
R
1
1.99+02
1.11+01
X
Y
Z
Main Index Main Index
447 CHAPTER 7
Example Problems
7.7 Dynamic Response Optimization
This example demonstrates structural optimization when the structural loads are
frequency dependent. The system considered is a flat rectangular plate clamped on
three edges and free along the fourth, as shown in Figure 7-16. The problem
investigates minimization of the mean square response of the transverse displacement
at the midpoint of the free edge, while constraining the volume of the structure (and
hence, weight) to be equal to that of the initial design. A pressure loading with an
amplitude of is applied across a frequency range of 20.0 to 200.0 Hz. A small
amount of frequency-dependent modal damping has also been included.
Figure 7-16 Pressure-Loaded Flat Plate
Figure 7-17 shows the finite element representation. Due to symmetry conditions,
only half of the structure needs to be modeled. Ten design variables are related to ten
plate element property group thicknesses. The first such group is shown in the figure
as the shaded "ring" of elements. Subsequent design variables control the thicknesses
of subsequent rings of elements up to the tenth design variable which controls the
single element connected to GRID1110.
1.0 in
f
in
2

(Uniform Pressure)
X
Grid 1110
Y
Main Index Main Index


448
Figure 7-17 Clamped-Free Plate
The dynamic excitation is applied from 20. to 200. Hz. The objective is to minimize the
root mean square (RMS) of transverse displacements at GRID 1110. MSC.Nastran
provides several alternative ways of computing RMS quantities and three of these are
illustrated here. In the first variation, the SQRT and SSQ arguments in the DEQATN
functions and used to minimize an objective of the form:
Eq. 7-5
The above notation indicates the square of the displacement responses are summed
over 1.0 Hz intervals from 20.0 to 100.0 Hz, and over 2.0 Hz intervals from 102.0 to
200.0 Hz. The square root of this quantity, therefore, provides a close approximation
to the rms response over the frequency range. This particular form was selected
simply for illustrative purposes and to highlight the versatility of the DEQATN
feature.
The second variation uses the RSS function directly on the DRESP1 entry. In this
method, a root sum of squares is performed for all the excitation frequencies in the
analysis. When this response is chosen as the objective, the resulting objective function
is of the form:
Symmetric
Boundary Conditions
Free
Edge
Grid 1110
x
y
t
1
x
1
=
mi n u
z 1110 ,
i
( )
2
i 20 =
100

2 u
z 1110 ,
2i
( )
2
i 51 =
100

+ =
Main Index Main Index
449 CHAPTER 7
Example Problems
Eq. 7-6
The third variation selects the RMSDISP response type on the DRESP1 entry. The
input power spectral input is specified to be a unit value over the excitation range and
the excitations frequencies are select so that the response value computed from the
RMSDISP response is virtually identical to that of Eq. 7-7.
Fifteen modes were retained for the modal representation. This number was selected
based on a convergence study which indicated good sensitivity information was
obtained with this amount. A guideline recommendation is to retain at least twice as
many modes for modal sensitivity and optimization studies as would be appropriate
for an analysis. Of course, the resultant sensitivities should be checked for accuracy
and the number of modes increased, if necessary.
The input file Listing 7-15 shows the portion of the input that is common for all three
versions of the example. It can be seen that the thickness distribution of the initial
design is a constant 0.08 inches. The corresponding initial volume is 8.0 . The
volume constraint is imposed as follows: the volume response is first identified on
DRESP1 number 201, bounds are placed on this response by DCONSTR number 10,
which is in turn selected as a global constraint in Case Control (global, since volume
is a subcase-independent response) using the DESGLB Case Control command. This
results in:
Eq. 7-7
Listing 7-15 Common Portion of Input File DSOUG7
ID MSC DSOUG7A $
TIME 200
SOL 200
CEND
TITLE = Synthesis of Responses across Different Frequencies: DSOUG7A
SET 10 = 1110
DISPL(PHASE,SORT1) = 10 $ MAGNITUDE/PHASE REPRESENTATION FOR RESPONSE
$ ANALYSIS AS WELL AS SENSITIVITY ANALYSIS
desglb = 10
subcase 1
SPC = 100
DLOAD = 700
FREQ = 740
METHOD = 500
ANALYSIS = MFREQ
sdamping = 2000
DESOBJ = 1
mi n u
z 1110 ,
i
( )
2
i 20 =
200

=
in
3
7.99 vol 8.01
Main Index Main Index


450
output
disp(plot,phase) = 10
output(xyout)
cscale 2.0
ymax=4.0
plotter = nastran
ytitle = displacement at grid 1110
xyplot disp / 1110(t3)
$
BEGIN BULK
PARAM WTMASS .002588
$------------------------------------------------------------------------------
$ ANALYSIS MODEL
$------------------------------------------------------------------------------
$
$...GRID AND SPC DATA:
$
GRDSET 6
GRID 100 0. 0. 0.
= *(1)= *(1.) = =
=9
GRID 200 0. 1.0 0.
= *(1) = *(1.) = =
=9
GRID 300 0. 2.0 0.
= *(1) = *(1.) = =
=9
GRID 400 0. 3.0 0.
= *(1) = *(1.) = =
=9
GRID 500 0. 4.0 0.
= *(1) = *(1.) = =
=9
GRID 600 0. 5.0 0.
= *(1) = *(1.) = =
=9
GRID 700 0. 6.0 0.
= *(1) = *(1.) = =
=9
GRID 800 0. 7.0 0.
= *(1) = *(1.) = =
=9
GRID 900 0. 8.0 0.
= *(1) = *(1.) = =
=9
GRID 1000 0. 9.0 0.
= *(1) = *(1.) = =
=9
GRID 1100 0. 10.0 0.
= *(1) = *(1.) = =
=9
$
SPC1 100 123456 100 101 102 103 104 105 +
+ 106 107 108 109 110 200 300 400 +
+ 500 600 700 800 900 1000 1100
SPC1 100 246 1101 1102 1103 1104 1105 1106 +
Main Index Main Index
451 CHAPTER 7
Example Problems
+ 1107 1108 1109
SPC1 100 246 1110
$
$...ELEMENT DEFINITION AND PROPERTIES:
$ (ELEMENTS GROUPED BY PID SINCE THICKNESS OF ALL ELEMENTS IN A GROUP
$ ARE TO BE AFFECTED BY A SINGLE DESIGN VARIABLE)
$
MAT1 150 1.0E7 0.3 0.1
$...ELEMENT GROUP 1:
CQUAD4 101 1 100 101 201 200
= *(100) = *(100) *(100) *(100) *(100)
=8
CQUAD4 102 1 101 102 202 201
= *(1) = *(1) *(1) *(1) *(1),
=7
PSHELL 1 150 .08 150
$...ELEMENT GROUP 2:
CQUAD4 202 2 201 202 302 301
= *(100) = *(100) *(100) *(100) *(100)
=7
CQUAD4 203 2 202 203 303 302
= *(1) = *(1) *(1) *(1) *(1),
=6
PSHELL 2 150 .08 150
$...ELEMENT GROUP 3:
CQUAD4 303 3 302 303 403 402
= *(100) = *(100) *(100) *(100) *(100)
=6
CQUAD4 304 3 303 304 404 403
= *(1) = *(1) *(1) *(1) *(1),
=5
PSHELL 3 150 .08 150
$...ELEMENT GROUP 4:
CQUAD4 404 4 403 404 504 503
= *(100) = *(100) *(100) *(100) *(100)
=5
CQUAD4 405 4 404 405 505 504
= *(1) = *(1) *(1) *(1) *(1),
=4
PSHELL 4 150 .08 150
$...ELEMENT GROUP 5:
CQUAD4 505 5 504 505 605 604
= *(100) = *(100) *(100) *(100) *(100)
=4
CQUAD4 506 5 505 506 606 605
= *(1) = *(1) *(1) *(1) *(1),
=3
PSHELL 5 150 .08 150
$...ELEMENT GROUP 6:
CQUAD4 606 6 605 606 706 705
= *(100) = *(100) *(100) *(100) *(100)
=3
CQUAD4 607 6 606 607 707 706
= *(1) = *(1) *(1) *(1) *(1),
=2
PSHELL 6 150 .08 150
Main Index Main Index


452
$...ELEMENT GROUP 7:
CQUAD4 707 7 706 707 807 806
= *(100) = *(100) *(100) *(100) *(100)
=2
CQUAD4 708 7 707 708 808 807
= *(1) = *(1) *(1) *(1) *(1),
=1
PSHELL 7 150 .08 150
$...ELEMENT GROUP 8:
CQUAD4 808 8 807 808 908 907
= *(100) = *(100) *(100) *(100) *(100)
=1
CQUAD4 809 8 808 809 909 908
= *(1) = *(1) *(1) *(1) *(1),
PSHELL 8 150 .08 150
$...ELEMENT GROUP 9:
CQUAD4 909 9 908 909 1009 1008
= *(100) = *(100) *(100) *(100) *(100)
CQUAD4 910 9 909 910 1010 1009
PSHELL 9 150 .08 150
$...ELEMENT GROUP 10:
CQUAD4 1010 10 1009 1010 1110 1109
PSHELL 10 150 .08 150
$
$...EIGENVALUE EXTRACTION INFORMATION - 15 RETAINED MODES
$
EIGRL 500 15 0
$
$...FREQUENCY DEPENDENT LOADING DATA: (OSCILLATORY PRESSURE LOAD)
$
RLOAD1 700 730 800
PLOAD2 730 1.0 101 THRU 110
PLOAD2 730 1.0 201 THRU 210
PLOAD2 730 1.0 301 THRU 310
PLOAD2 730 1.0 401 THRU 410
PLOAD2 730 1.0 501 THRU 510
PLOAD2 730 1.0 601 THRU 610
PLOAD2 730 1.0 701 THRU 710
PLOAD2 730 1.0 801 THRU 810
PLOAD2 730 1.0 901 THRU 910
PLOAD2 730 1.0 1001 THRU 1010
TABLED1 800 +
+ 0.0 1.0 1.0E3 1.0 ENDT
$
tabdmp1 2000
0.0 0.20 1000.0 0.20 endt
$
$------------------------------------------------------------------------------
$ DESIGN MODEL
$------------------------------------------------------------------------------
$
$...SPECIFY DESIGN VARIABLES RELATE LINEARLY TO PLATE THICKNESS
DESVAR 1 T1 .08 .001 1.0
DESVAR 2 T2 .08 .001 1.0
DESVAR 3 T3 .08 .001 1.0
DESVAR 4 T4 .08 .001 1.0
Main Index Main Index
453 CHAPTER 7
Example Problems
DESVAR 5 T5 .08 .001 1.0
DESVAR 6 T6 .08 .001 1.0
DESVAR 7 T7 .08 .001 1.0
DESVAR 8 T8 .08 .001 1.0
DESVAR 9 T9 .08 .001 1.0
DESVAR 10 T10 .08 .001 1.0
$
$...RELATE DESIGN VARIABLES TO PLATE THICKNESSES
DVPREL1 101 PSHELL 1 T .01
1 1.0
DVPREL1 102 PSHELL 2 T .01
2 1.0
DVPREL1 103 PSHELL 3 T .01
3 1.0
DVPREL1 104 PSHELL 4 T .01
4 1.0
DVPREL1 105 PSHELL 5 T .01
5 1.0
DVPREL1 106 PSHELL 6 T .01
6 1.0
DVPREL1 107 PSHELL 7 T .01
7 1.0
DVPREL1 108 PSHELL 8 T .01
8 1.0
DVPREL1 109 PSHELL 9 T .01
9 1.0
DVPREL1 110 PSHELL 10 T .01
10 1.0
$
DRESP1 201 VOLUME VOLUME
include 'dsoug7_deq.inc'
$include 'dsoug7_rss.inc'
$include 'dsoug7_rms.inc'
$
$
DCONSTR 10 201 7.99 8.01
doptprm desmax 40 p1 1 p2 8 conv1 0.01
$.......2.......3.......4.......5.......6.......7.......8.......9.......0
$
PARAM POST-1
ENDDATA
The three versions of the objective are developed by including one of three files. The
first is named dsoug7_deq.inc and is given in Listing 7-16. The RMS response is the
objective function. The underlying transverse displacements are first identified by
DRESP1 entries 20 through 100 and 102 through 200 (note the use of Bulk Data
replicators). These identify the transverse component of displacement at GRID 1110
across the frequency range of interest. These first-level responses are then used as
input to DEQATN 1 via DRESP2 number 1, which defines the equation input. The
resultant response is then defined as the objective function using the Case Control
DESOBJ command. The DEQATN uses the SSQ and SQRT functions to create the
objective given by Eq. 7-5.
Main Index Main Index


454
Listing 7-16 Include File dsoug7-deq.inc
FREQ1 740 20. 1. 179
DRESP1 20 g1110L FRDISP 3 20.0 1110
= *(1) = = = = = *(1.0) =
=79
DRESP1 102 G1110H FRDISP 3 102.0 1110
= *(2) = = = = = *(2.0) =
=48
$
DRESP2 1 UZ2 1
DRESP1 20 21 22 23 24 25 26
27 28 29 30 31 32 33
34 35 36 37 38 39 40
41 42 43 44 45 46 47
48 49 50 51 52 53 54
55 56 57 58 59 60 61
62 63 64 65 66 67 68
69 70 71 72 73 74 75
76 77 78 79 80 81 82
83 84 85 86 87 88 89
90 91 92 93 94 95 96
97 98 99 100 102 104 106
108 110 112 114 116 118 120
122 124 126 128 130 132 134
136 138 140 142 144 146 148
150 152 154 156 158 160 162
164 166 168 170 172 174 176
178 180 182 184 186 188 190
192 194 196 198 200
$
DEQATN 1 UZ2(U20,u21,u22,u23,u24,u25,U26,U27,U28,U29,U30,
U31,U32,U33,U34,U35,U36,U37,U38,U39,U40,
U41,U42,U43,U44,U45,U46,U47,U48,U49,U50,
U51,U52,U53,U54,U55,U56,U57,U58,U59,U60,
U61,U62,U63,U64,U65,U66,U67,U68,U69,U70,
U71,U72,U73,U74,U75,U76,U77,U78,U79,U80,
U81,U82,U83,U84,U85,U86,U87,U88,U89,U90,
U91,U92,U93,U94,U95,U96,U97,U98,U99,U100,
U102,U104,U106,U108,U110,U112,U114,U116,U118,U120,
U122,U124,U126,U128,U130,U132,U134,U136,U138,U140,
U142,U144,U146,U148,U150,U152,U154,U156,U158,U160,
U162,U164,U166,U168,U170,U172,U174,U176,U178,U180,
U182,U184,U186,U188,U190,U192,U194,U196,U198,U200)
=sqrt(ssq(U20,u21,u22,u23,u24,u25,U26,U27,U28,U29,U30,
U31,U32,U33,U34,U35,U36,U37,U38,U39,U40,
U41,U42,U43,U44,U45,U46,U47,U48,U49,U50,
U51,U52,U53,U54,U55,U56,U57,U58,U59,U60,
U61,U62,U63,U64,U65,U66,U67,U68,U69,U70,
U71,U72,U73,U74,U75,U76,U77,U78,U79,U80,
U81,U82,U83,U84,U85,U86,U87,U88,U89,U90,
U91,U92,U93,U94,U95,U96,U97,U98,U99,U100) + 2.0 *
ssq( U102,U104,U106,U108,U110,U112,U114,U116,U118,U120,
U122,U124,U126,U128,U130,U132,U134,U136,U138,U140,
U142,U144,U146,U148,U150,U152,U154,U156,U158,U160,
U162,U164,U166,U168,U170,U172,U174,U176,U178,U180,
U182,U184,U186,U188,U190,U192,U194,U196,U198,U200))
Main Index Main Index
455 CHAPTER 7
Example Problems
The input for the second version is given in Listing 7-17 and it is seen that the RSS
specification of the ATTB field of the DRESP1 with ID=1 is sufficient to generate a root
mean square response. In this case, all 181 frequencies contribute to the response.
Listing 7-17 Include File dsoug7-rss.inc
The unique input for the third version is given in . The excitation frequencies have
been changed so that 2 Hz intervals are taken at excitation frequencies about 100. Hz.
In this case, a DRESP1 with an RMSDISP response type is used. The ATTB field of the
DRESP1 points to the RANDPS entry with ID=20. This random input power spectrum
has a value of 1.0 across the range of excitation frequencies, with the net result that the
RMSDISP is virtually identical to the DEQATN of the first variation given by Eq. 7-5.
Listing 7-18 Include File dsoug7-rms.inc
Printed output results for the three options are given in Listing 7-19, Listing 7-20, and
Listing 7-21. Figure 7-18 plots the final thickness distributions for the three variations.
The results are seen to be very similar for the different methods. The RMS value
decreases from 15.20 to around 12.0. All methods converged in the five design cycles.
$
FREQ1 740 20. 1. 180
DRESP1 1 g1110L FRDISP 3 rss 1110
$
FREQ1 740 20. 1. 79
FREQ1 740 102. 2. 49
DRESP1 1 G1110L RMSDISP 3 20 1110
RANDPS20 1 1 1.0 20
TABRND120
0.0 1.0 1000.01.0 ENDT
Main Index Main Index


456
Listing 7-19 Design History Results Using the DEQATN Option
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.519667E+01 -1.249123E-03
1 1.443932E+01 1.365850E+01 5.716709E-02 -1.247873E-03
2 1.324028E+01 1.271585E+01 4.124225E-02 -1.214774E-03
3 1.252116E+01 1.248591E+01 2.822777E-03 -1.217750E-03
4 1.232194E+01 1.216405E+01 1.298024E-02 -1.230133E-03
5 1.195737E+01 1.212859E+01 -1.411712E-02 -1.233466E-03
---------------------------------------------------------------------------------------------------------------
1 SYNTHESIS OF RESPONSES ACROSS DIFFERENT FREQUENCIES: DSOUG7A SEPTEMBER 19, 2001 MSC.NASTRAN 4/ 9/01 PAGE 87

0 SUBCASE 1
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | T1 | 8.0000E-02 : 9.6000E-02 : 9.3140E-02 : 1.0635E-01 : 8.9943E-02 : 1.0793E-01 :
2 | 2 | T2 | 8.0000E-02 : 7.9898E-02 : 8.9400E-02 : 8.3661E-02 : 9.3010E-02 : 7.7354E-02 :
3 | 3 | T3 | 8.0000E-02 : 6.4675E-02 : 7.2220E-02 : 7.0797E-02 : 7.1599E-02 : 6.7977E-02 :
4 | 4 | T4 | 8.0000E-02 : 6.5901E-02 : 5.2721E-02 : 4.4020E-02 : 4.4555E-02 : 4.2334E-02 :
5 | 5 | T5 | 8.0000E-02 : 7.3891E-02 : 6.0025E-02 : 5.1219E-02 : 4.7529E-02 : 4.7742E-02 :
6 | 6 | T6 | 8.0000E-02 : 7.9610E-02 : 7.1892E-02 : 7.0936E-02 : 7.5427E-02 : 7.3412E-02 :
7 | 7 | T7 | 8.0000E-02 : 8.5374E-02 : 8.6322E-02 : 8.9028E-02 : 9.5125E-02 : 9.4058E-02 :
8 | 8 | T8 | 8.0000E-02 : 9.6000E-02 : 1.1396E-01 : 1.2035E-01 : 1.2918E-01 : 1.3052E-01 :
9 | 9 | T9 | 8.0000E-02 : 9.6000E-02 : 1.1520E-01 : 1.2517E-01 : 1.3843E-01 : 1.4504E-01 :
10 | 10 | T10 | 8.0000E-02 : 9.6000E-02 : 1.1520E-01 : 1.2082E-01 : 1.2812E-01 : 1.3241E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 5.
Main Index Main Index
457 CHAPTER 7
Example Problems
Listing 7-20 Design History Results for the RSS Option on DRESP1
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.519907E+01 -1.249123E-03
1 1.444129E+01 1.366040E+01 5.716482E-02 -1.247873E-03
2 1.324232E+01 1.271818E+01 4.121153E-02 -1.245088E-03
3 1.252379E+01 1.248799E+01 2.866141E-03 -1.240851E-03
4 1.232415E+01 1.216508E+01 1.307595E-02 -1.236614E-03
5 1.195864E+01 1.212917E+01 -1.405937E-02 -1.237628E-03
---------------------------------------------------------------------------------------------------------------
1 SYNTHESIS OF RESPONSES ACROSS DIFFERENT FREQUENCIES: DSOUG7 SEPTEMBER 19, 2001 MSC.NASTRAN 4/ 9/01 PAGE 450

0 SUBCASE 1
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | T1 | 8.0000E-02 : 9.6000E-02 : 9.3151E-02 : 1.0633E-01 : 8.9940E-02 : 1.0793E-01 :
2 | 2 | T2 | 8.0000E-02 : 7.9913E-02 : 8.9366E-02 : 8.3660E-02 : 9.2985E-02 : 7.7328E-02 :
3 | 3 | T3 | 8.0000E-02 : 6.4681E-02 : 7.2203E-02 : 7.0783E-02 : 7.1560E-02 : 6.7921E-02 :
4 | 4 | T4 | 8.0000E-02 : 6.5895E-02 : 5.2716E-02 : 4.3998E-02 : 4.4495E-02 : 4.2223E-02 :
5 | 5 | T5 | 8.0000E-02 : 7.3882E-02 : 6.0032E-02 : 5.1232E-02 : 4.7626E-02 : 4.7890E-02 :
6 | 6 | T6 | 8.0000E-02 : 7.9598E-02 : 7.1914E-02 : 7.0961E-02 : 7.5457E-02 : 7.3469E-02 :
7 | 7 | T7 | 8.0000E-02 : 8.5362E-02 : 8.6328E-02 : 8.9035E-02 : 9.5133E-02 : 9.4091E-02 :
8 | 8 | T8 | 8.0000E-02 : 9.6000E-02 : 1.1396E-01 : 1.2036E-01 : 1.2920E-01 : 1.3054E-01 :
9 | 9 | T9 | 8.0000E-02 : 9.6000E-02 : 1.1520E-01 : 1.2518E-01 : 1.3846E-01 : 1.4506E-01 :
10 | 10 | T10 | 8.0000E-02 : 9.6000E-02 : 1.1520E-01 : 1.2083E-01 : 1.2814E-01 : 1.3242E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 5.
Main Index Main Index


458
Listing 7-21 Design History for the RMS Option
0 SUBCASE 1
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 6
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 5
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.518897E+01 -1.249123E-03
1 1.442682E+01 1.368694E+01 5.405684E-02 -1.248587E-03
2 1.326691E+01 1.269624E+01 4.494826E-02 2.155018E-04
3 1.238254E+01 1.212222E+01 2.147395E-02 2.113242E-04
4 1.198582E+01 1.198919E+01 -2.805536E-04 2.107274E-04
5 1.184377E+01 1.188050E+01 -3.091925E-03 2.132339E-04
---------------------------------------------------------------------------------------------------------------
1 SYNTHESIS OF RESPONSES ACROSS DIFFERENT FREQUENCIES: DSOUG7 SEPTEMBER 19, 2001 MSC.NASTRAN 4/ 9/01 PAGE 56

0 SUBCASE 1
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | T1 | 8.0000E-02 : 9.6000E-02 : 9.4803E-02 : 1.0653E-01 : 9.1509E-02 : 1.0526E-01 :
2 | 2 | T2 | 8.0000E-02 : 8.1651E-02 : 8.6642E-02 : 8.0803E-02 : 9.0622E-02 : 7.9363E-02 :
3 | 3 | T3 | 8.0000E-02 : 6.5678E-02 : 7.1309E-02 : 6.5509E-02 : 7.1306E-02 : 6.3366E-02 :
4 | 4 | T4 | 8.0000E-02 : 6.5819E-02 : 5.2655E-02 : 4.2124E-02 : 3.7225E-02 : 3.4805E-02 :
5 | 5 | T5 | 8.0000E-02 : 7.2886E-02 : 6.0233E-02 : 4.8187E-02 : 5.0442E-02 : 5.4060E-02 :
6 | 6 | T6 | 8.0000E-02 : 7.8117E-02 : 7.2950E-02 : 6.9732E-02 : 7.4214E-02 : 7.5995E-02 :
7 | 7 | T7 | 8.0000E-02 : 8.3734E-02 : 8.7547E-02 : 9.5598E-02 : 9.6369E-02 : 9.7185E-02 :
8 | 8 | T8 | 8.0000E-02 : 9.4436E-02 : 1.1332E-01 : 1.3599E-01 : 1.3745E-01 : 1.3842E-01 :
9 | 9 | T9 | 8.0000E-02 : 9.6000E-02 : 1.1540E-01 : 1.3847E-01 : 1.4329E-01 : 1.4723E-01 :
10 | 10 | T10 | 8.0000E-02 : 9.6000E-02 : 1.1520E-01 : 1.3824E-01 : 1.4108E-01 : 1.4347E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 5.
Main Index Main Index
459 CHAPTER 7
Example Problems
Figure 7-18 Final Thickness Distributions
Figure 7-19 shows the initial and final response data for the displacements used to
formulate the objective for the first variation of this problem (the other variations have
a similar form). Note that the integral of the final curve is indeed less than that of the
original curve. In addition, the response peak has shifted somewhat, from a value of
3.623 at 58 Hz to a value of 3.017 at 53 Hz, indicating a slight decrease in overall
structural stiffness.
0.00E+00
2.00E-02
4.00E-02
6.00E-02
8.00E-02
1.00E-01
1.20E-01
1.40E-01
1.60E-01
1.80E-01
1 2 3 4 5 6 7 8 9 10
Design Variable
T
h
i
c
k
n
e
s
s
DEQATN RSS on DRESP1 RMSDISP
Main Index Main Index


460
Figure 7-19 Frequency-Dependent Displacements
Main Index Main Index
461 CHAPTER 7
Example Problems
7.8 Twenty-Five Bar Truss, Superelement and Discrete
Variable Optimization
This problem, often seen in the design optimization literature, calls for a minimum
weight structure subject to member stress, Euler buckling, and joint displacement
constraints. The structure is shown in Figure 7-20. The formulation of the buckling
constraints is a good example of constructing normalized constraints based on user-
defined structural responses.
In addition, this problem will be substructured in order to illustrate superelement
optimization and the final design will be selected from a user specified list of discrete
variables.
Figure 7-20 Twenty-Five Bar Truss
Analysis Model Description
Three-dimensional truss tower
Symmetric with respect to x-y plane and y-z plane
Materials: E = 1.0E7 psi
Two distinct loading conditions
10
1
2
4
8
9
7
5
6
3
Superelement 1
Superelement 0
Y
X
Weight density 0.1 in
3
=
Main Index Main Index


462
Design Model Description
Euler buckling occurs when the magnitude of a member's compressive stress is
greater than a critical stress which, for the first buckling mode of a pin-connected
member, is
Eq. 7-8
We will prescribe thin-walled tubular members with a diameter-to-thickness ratio of
10, as indicated in the sketch in Figure 7-21.
Figure 7-21 Thin-Walled Tube Cross-Section
Minimization of structural weight
Design
variables:
Cross-sectional areas linked to eight independent design variables
Constraints:
Allowable
stress:
Tensile = 40,000 psi
Compressive = -40,000 psi
Displacement: +0.35 inches at grid 1 and 2 for all translational degrees
of freedom
Euler buckling constraints for compressive members assuming tubular
section diameter to thickness ratio of 10

b
P
b
A
------
1
A
---

2
EI
L
2
------------
\ .
|
| |
= =
r
i
r
o
D
avg
r
o
r
i
+ ( )
t D
avg
10 =
Main Index Main Index
463 CHAPTER 7
Example Problems
With this formulation, the area and inertia terms of Eq. 7-8 become
Eq. 7-9
Eq. 7-10
The critical buckling stress of Eq. 7-8 can therefore be written as:
Eq. 7-11
It is also desirable to impose a factor of safety on the design for buckling so that the
buckling response constraint can be written as:
Eq. 7-12
where is the factor of safety.
The input data file for this example is shown in Listing 7-22 and needs only limited
description. Eight independent design variables (DESVAR 1 through 8) are used to
represent eight ROD average diameters (the diameter of the midsurface of the thin-
walled tube). A single list of discrete variables is invoked on each DESVAR entry to
limit the allowable values of the design variables to 0.1, 0.5, or integer values between
1.0 and 100.0. DVPREL2 entries are used to relate the design variables to the rod areas
using the relationship of Eq. 7-9.
Listing 7-22 Input File for Example DSOUG8
ID MSC DSOUG8 $
TIME 10
SOL 200 $ OPTIMIZATION
CEND
TITLE = OPTIMAL SIZING OF A 25-BAR TRUSS - DSOUG8
SUBTITLE = EIGHT INDEPENDENT ROD DIAMETERS
ECHO = UNSORT
OLOAD = ALL
DISP = ALL
SPCFORCE = ALL
ELFORCE = ALL
STRESS = ALL
SPC = 100
ANALYSIS = STATICS $
A 2 r
o
2
r
i
2
( ) D
avg
t
D
avg
( )
2
10
------------------------ = = =
I

4
--- r
o
4
r
i
4
( ) 0.013 D
avg
4
= =

2
E
L
2
A
---------- I
0.13
2
ED
avg
2
L
2
------------------------------------ = =
r
F
s

b
----------
7.69L
2
F
s

2
ED
avg
2
------------------------------- 1.0 = =
F
s
Main Index Main Index


464
DESOBJ(MIN) = 15 $ OBJECTIVE FUNCTION DEFINITION
DESSUB = 12 $ CONSTRAINT DEFININITION
SUBCASE 1
LABEL = LOAD CONDITION 1
LOAD = 300
SUBCASE 2
LABEL = LOAD CONDITION 2
LOAD = 310
BEGIN BULK
$
$-------------------------------------------------------------------------------
$ ANALYSIS MODEL
$-------------------------------------------------------------------------------
$
SESET 1 1 2
GRDSET 456
MAT1 1 10.0E6 0.1 +M1
+M1 25000. 25000.
SPC1 100 123 7 THRU 10
GRID 1 -37.5 0.0 200.0
GRID 2 37.5 0.0 200.0
GRID 3 -37.5 37.5 100.0
GRID 4 37.5 37.5 100.0
GRID 5 37.5 -37.5 100.0
GRID 6 -37.5 -37.5 100.0
GRID 7 -100.0 100.0 0.0
GRID 8 100.0 100.0 0.0
GRID 9 100.0 -100.0 0.0
GRID 10 -100.0 -100.0 0.0
CROD 1 1 1 2
CROD 2 2 1 4
CROD 3 2 2 3
CROD 4 2 1 5
CROD 5 2 2 6
CROD 6 3 2 4
CROD 7 3 2 5
CROD 8 3 1 3
CROD 9 3 1 6
CROD 10 4 3 6
CROD 11 4 4 5
CROD 12 5 3 4
CROD 13 5 5 6
CROD 14 6 3 10
CROD 15 6 6 7
CROD 16 6 4 9
CROD 17 6 5 8
CROD 18 7 4 7
CROD 19 7 3 8
CROD 20 7 5 10
CROD 21 7 6 9
CROD 22 8 6 10
CROD 23 8 3 7
CROD 24 8 5 9
CROD 25 8 4 8
$
PROD 1 1 2.0 0.0
PROD 2 1 2.0 0.0
PROD 3 1 2.0 0.0
PROD 4 1 2.0 0.0
PROD 5 1 2.0 0.0
PROD 6 1 2.0 0.0
PROD 7 1 2.0 0.0
Main Index Main Index
465 CHAPTER 7
Example Problems
PROD 8 1 2.0 0.0
$
FORCE 300 1 1.0 1000. 10000. -5000.
FORCE 300 2 1.0 0. 10000. -5000.
FORCE 300 3 1.0 500. 0. 0.
FORCE 300 6 1.0 500. 0. 0.
FORCE 310 1 1.0 0. 20000. -5000.
FORCE 310 2 1.0 0. -20000. -5000.
$
$-------------------------------------------------------------------------------
$ DESIGN MODEL
$-------------------------------------------------------------------------------
$
$...DEFINE THE DESIGN VARIABLES
$DESVAR ID LABEL XINIT XLB XUB DELXV DDVAL
$
DESVAR 1 D1 2.0 0.01 100.0 10
DESVAR 2 D2 2.0 0.01 100.0 10
DESVAR 3 D3 2.0 0.01 100.0 10
DESVAR 4 D4 2.0 0.01 100.0 10
DESVAR 5 D5 2.0 0.01 100.0 10
DESVAR 6 D6 2.0 0.01 100.0 10
DESVAR 7 D7 2.0 0.01 100.0 10
DESVAR 8 D8 2.0 0.01 100.0 10
DDVAL 10 0.1 0.5
1.0 THRU 100. BY 1.0
$
$...RELATE THE DESIGN VARIABLES TO THE ANALYSIS MODEL PROPERTIES
$DVPREL2 ID TYPE PID FID PMIN PMAX EQID
$ DESVAR DVID1
$
DVPREL2 1 PROD 1 A 100
DESVAR 1
DVPREL2 2 PROD 2 A 100
DESVAR 2
DVPREL2 3 PROD 3 A 100
DESVAR 3
DVPREL2 4 PROD 4 A 100
DESVAR 4
DVPREL2 5 PROD 5 A 100
DESVAR 5
DVPREL2 6 PROD 6 A 100
DESVAR 6
DVPREL2 7 PROD 7 A 100
DESVAR 7
DVPREL2 8 PROD 8 A 100
DESVAR 8
$
$
$...EQUATIONS USED TO DEFINE TYPE 2 PROPERTIES
$DEQATN EQUID F() = ...
$
DEQATN 100 AREA(DAVG) = PI(1) * DAVG**2 / 10.0;
$
$...IDENTIFY THE RESPONSES TO BE USED IN THE DESIGN MODEL
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1 +
$+ ATT2 ...
$
DRESP1 1 S1 STRESS PROD 2 1
DRESP1 2 S2 STRESS PROD 2 2
DRESP1 3 S3 STRESS PROD 2 3
Main Index Main Index


466
DRESP1 4 S4 STRESS PROD 2 4
DRESP1 5 S5 STRESS PROD 2 5
DRESP1 6 S6 STRESS PROD 2 6
DRESP1 7 S7 STRESS PROD 2 7
DRESP1 8 S8 STRESS PROD 2 8
DRESP1 9 D1 DISP 123 1
DRESP1 12 D4 DISP 123 2
DRESP1 15 W WEIGHT
$
$...FORMULATE THE SECOND LEVEL RESPONSES (HERE SIMPLE EULER BUCKLING)
$DRESP2 ID LABEL EQID REGION
$+ DESVAR DVID1 DVID2 ...
$+ DTABLE LABEL1 LABEL2 ...
$+ DRESP1 NR1 NR2 ...
$+ DNODE NID1 DIR1 NID2 DIR2 ...
$
DRESP2 16 SC1 1
DESVAR 1
DTABLE FS E L1
DRESP1 1
DRESP2 17 SC2 1
DESVAR 2
DTABLE FS E L2
DRESP1 2
DRESP2 18 SC3 1
DESVAR 3
DTABLE FS E L3
DRESP1 3
DRESP2 19 SC4 1
DESVAR 4
DTABLE FS E L4
DRESP1 4
DRESP2 20 SC5 1
DESVAR 5
DTABLE FS E L5
DRESP1 5
DRESP2 21 SC6 1
DESVAR 6
DTABLE FS E L6
DRESP1 6
DRESP2 22 SC7 1
DESVAR 7
DTABLE FS E L7
DRESP1 7
DRESP2 23 SC8 1
DESVAR 8
DTABLE FS E L8
DRESP1 8
$
$...EQUATIONS USED TO DEFINE SECOND LEVEL RESPONSES
$DEQATN EQUID F() = ...
$
DEQATN 1 NUM(DAVG,FS,E,L,SIGMA) = 7.69 * L**2 * SIGMA;
DENOM = (PI(1) * DAVG)**2*E;
BUCKLING = -FS * NUM / DENOM
$
$...TABLE CONSTANTS
$DTABLE LABEL1 VALUE1 LABEL2 VALUE2 LABEL3 VALUE3 LABEL4 VALUE4
$+ LABEL5 VALUE5 ...
$
DTABLE L1 75.00 L2 130.50 L3 106.80 L4 75.00
L5 75.00 L6 181.14 L7 181.14 L8 133.46
Main Index Main Index
467 CHAPTER 7
Example Problems
E 1.0E7 FS 1.25
$
$...DEFINE THE DESIGN CONSTRAINTS
$CONSTR DCID RID LALLOW UALLOW
$
DCONSTR 10 1 -40000. 40000.
DCONSTR 10 2 -40000. 40000.
DCONSTR 10 3 -40000. 40000.
DCONSTR 10 4 -40000. 40000.
DCONSTR 10 5 -40000. 40000.
DCONSTR 10 6 -40000. 40000.
DCONSTR 10 7 -40000. 40000.
DCONSTR 10 8 -40000. 40000.
DCONSTR 10 9 -0.35 0.35
DCONSTR 10 10 -0.35 0.35
$
DCONSTR 11 16 1.0
DCONSTR 11 17 1.0
DCONSTR 11 18 1.0
DCONSTR 11 19 1.0
DCONSTR 11 20 1.0
DCONSTR 11 21 1.0
DCONSTR 11 22 1.0
DCONSTR 11 23 1.0
$
$...COMBINE THE TWO CONSTRAINT SETS
$ (EQUIVALENT TO JUST PUTTING ALL INTO THE SAME SET TO BEGIN WITH)
DCONADD 12 10 11
$
$...OVERRIDE OPTIMIZATION PARAMETER DEFAULTS:
$
DOPTPRM IPRINT 3 DESMAX 15 DELP 0.5 P1 1
P2 15
ENDDATA
$.......2.......3.......4.......5.......6.......7.......8.......9.......0
Axial stresses and grid displacements are identified on DRESP1 entries 1 through 12
and are constrained using DCONSTR entries with an ID of 10. Additionally, the axial
rod stresses are used as input in the definition of the Euler buckling responses
DRESP2s 16 through 23), which all reference DEQATN 1. Note that element lengths
are included in these relations via constants defined on the DTABLE entry. This
simplifies the input since the element lengths do not need to be hard-coded on eight
different DEQATN entries. The DTABLE entry also defines the factor of safety at
125%. DCONSTR entries place bounds on the buckling response. Since the lower
bound is not of interest (e.g., tensile stresses on the elements will not induce Euler
buckling), we leave the lower bound blank on the DCONSTR entry.
Turning to the superelement analysis model, in Case Control the SUPER = ALL
command is recommended, rather than an explicit data recovery subcase for each
superelement. In Bulk Data, the SESET has been used to place grids 1 and 2 on the
interior of superelement 1.
Main Index Main Index


468
As far as the superelement aspects of the design model are concerned, the WEIGHT
response of DRESP1 ID 15 has an ALL to indicate the total structural weight across
all superelements is to be computed and used as the objective Function. "ALL" is the
default for this attribute so that even this adjustment in the design model to
accommodate superelements was not strictly necessary.
Final results for the design task are shown in Listing 7-23. Note that the design
required five finite analyses: an initial design, three continuous redesigns, and a final
discrete design. The initial design had a weight of 795.9 and was infeasible by a large
amount (the maximum constraint value is 6.12). After eight redesigns, the weight
increased to 814.1 and the design is feasible. A single discrete variable optimization is
performed and the weight increases to 1007.3 while the maximum constraint
decreases. The rather large increase in the objective of over 200. caused by the discrete
design may motivate consideration of additional discrete sizes.
Main Index Main Index
469 CHAPTER 7
Example Problems
Listing 7-23 Design History Results for DSOUG8

***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT FEASIBLE DISCRETE DESIGN OBTAINED)
(HARD FEASIBLE DISCRETE DESIGN OBTAINED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 10
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 8
NUMBER OF DISCRETE PROCESSING ANALYSES COMPLETED 1
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 4.155959E+02 6.115134E+00
1 7.781863E+02 7.782357E+02 -6.352635E-05 2.631261E-01
2 8.140873E+02 8.140758E+02 1.409526E-05 7.928371E-03
3 8.038315E+02 8.038285E+02 3.796529E-06 5.822778E-03
4 7.995189E+02 7.995188E+02 7.633987E-08 8.440852E-03
5 7.976517E+02 7.976515E+02 3.060743E-07 6.628275E-03
6 7.964160E+02 7.964161E+02 -1.532745E-07 5.273938E-03
7 7.961834E+02 7.961834E+02 0.000000E+00 2.780318E-03
8 7.958543E+02 7.958543E+02 0.000000E+00 2.032995E-03
8D 1.007305E+03 1.007323E+03 -1.769270E-05 -1.106629E-01
---------------------------------------------------------------------------------------------------------------

DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | D1 | 2.0000E+00 : 1.9659E+00 : 1.3901E+00 : 9.8293E-01 : 6.9502E-01 : 4.9142E-01 :
2 | 2 | D2 | 2.0000E+00 : 2.8775E+00 : 2.9790E+00 : 2.9664E+00 : 2.9533E+00 : 2.9436E+00 :
3 | 3 | D3 | 2.0000E+00 : 2.7629E+00 : 2.8609E+00 : 2.8687E+00 : 2.8763E+00 : 2.8824E+00 :
4 | 4 | D4 | 2.0000E+00 : 1.9405E+00 : 1.3722E+00 : 9.6946E-01 : 6.8615E-01 : 4.8526E-01 :
5 | 5 | D5 | 2.0000E+00 : 1.9567E+00 : 1.3836E+00 : 1.1831E+00 : 1.1844E+00 : 1.1898E+00 :
6 | 6 | D6 | 2.0000E+00 : 2.5337E+00 : 2.5074E+00 : 2.4997E+00 : 2.4902E+00 : 2.4859E+00 :
7 | 7 | D7 | 2.0000E+00 : 3.0888E+00 : 3.2761E+00 : 3.2802E+00 : 3.2871E+00 : 3.2916E+00 :
8 | 8 | D8 | 2.0000E+00 : 2.8033E+00 : 2.9788E+00 : 2.9769E+00 : 2.9744E+00 : 2.9746E+00 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 8D : 9 : 9D :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | D1 | 3.4747E-01 : 3.1663E-01 : 2.8519E-01 : 1.0000E-01 :
2 | 2 | D2 | 2.9386E+00 : 2.9420E+00 : 2.9408E+00 : 3.0000E+00 :
3 | 3 | D3 | 2.8850E+00 : 2.8831E+00 : 2.8838E+00 : 3.0000E+00 :
4 | 4 | D4 | 3.4316E-01 : 2.4276E-01 : 1.6463E-01 : 1.0000E-01 :
5 | 5 | D5 | 1.1940E+00 : 1.1935E+00 : 1.1928E+00 : 2.0000E+00 :
6 | 6 | D6 | 2.4837E+00 : 2.4831E+00 : 2.4826E+00 : 3.0000E+00 :
7 | 7 | D7 | 3.2930E+00 : 3.2917E+00 : 3.2922E+00 : 4.0000E+00 :
8 | 8 | D8 | 2.9735E+00 : 2.9752E+00 : 2.9744E+00 : 3.0000E+00 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 8.
AND HARD FEASIBLE DISCRETE DESIGN OBTAINED
Main Index Main Index


470
7.9 Design Optimization with Composite Materials with
Fully Stressed Design
In the optimal design of laminated composite structures, individual ply thicknesses
and orientations are often selected as design quantities. Figure 7-22 shows a simple
wing model that has an aluminum substructure and wing cover skins that are made
from composite materials. These cover skins have layers in the 0,90,+45 and -45 degree
directions relative to material coordinate system that is aligned with the 50% chord of
the wing structure. Variations of this model have appeared numerous times in the
design optimization literature. Johnson and Neill provide a representative description
(see Reference 4.).
Figure 7-22 Intermediate Complexity Wing Model
In the variation of the design task given here, the wing structure is subjected to two
static load cases and a normal modes analysis is also performed. The Fully Stressed
Design (FSD) technique is employed to find a good starting design for the complete
design task, which requires a mathematical programming approach. As discussed in
Fully Stressed Design on page 135, the FSD technique can only deal with static
Main Index Main Index
471 CHAPTER 7
Example Problems
loads and a limited subset of design responses and designed properties. The design
task is to minimize the wings structural weight while satisfying the following
strength and stiffness considerations:
The input data file is much too large to be listed here in its entirety so that only
fragments are shown here in Listing 7-24. The complete input file can be found in the
MSC.Nastran TPL with the name dsoug9.dat. The finite element model consists of 109
GRIDSs, 39 CRODs that link top and bottom structural grids, 55 CSHEARs that model
the wing ribs and spars and 2 CTRIA3s and 62 CQUAD4s that model the upper and
lower wing skins. 39 CONM2s model the nonstructural mass of the wing and 102
MPCs are used to connect the grids associated with the mass model structural model.
The design model contains 153 DESVARs. A single DESVAR controls the areas of all
the rods in the model while a second DESVAR controls the thickness of all the ribs.
The thicknesses of each of the 23 CSHEARs that model the spars are independently
designed. The remaining 128 design variables control the thicknesses of the four
layers of the composite cover skins. For the design concept employed in this example,
unequal number of plus and minus 45 degree plys are allowed but the top and bottom
surfaces are constrained to have the same thickness values. Each of the 32 finite
elements that model a surface is independently designed. Clearly, there are any
number of ways to specify the design model for this structure and the one shown may
not be the most practical or desirable. You may wish to try decoupling the top and
bottom surfaces for example or to see what the weight penalty is for imposing a
restriction that the number of +45 plies must equal the number of -45 plies.
Analysis and design with composite materials is a advanced feature in MSC.Nastran
and it is felt worthwhile to point out some of the special features that were used in this
example. The first is that the SYM option has been used for the LAM attribute on the
PCOMP Bulk Data entry. With this option, it is only necessary to model half the layup
with the other half added in such a way that there is a equal amount of material above
and below the midplane of the laminate. In an actual layup, it would be necessary to
Composite Plies Strain is below failure point (see the CFAILURE
discussion on page 474).
Posts in Substructure
Webs in Substructures
Second Natural Frequency
5.7 10
4
6.7 10
4

3.9 10
4
3.9 10
4

f
2
40. Hz
Main Index Main Index


472
specify the order of the plies in terms of individual laminae. This ordering is not
possible here and it is seen that the zero plies are all lumped at the extreme locations
in the laminate with the 90, +45 and -45 directions following so that the -45 plies are at
the midplane.
The objective is to minimize the weight and a DRESP2 is used to subtract 115. pounds
from the weight computed using the DRESP1 with RTYPE=WEIGHT. The 115. value
is a an estimate of the weight of the concentrated masses that is not affected by
changes in design. Removing this fixed weight from the optimizer allows the
algorithm to focus on the weight it can affect.
The DVPREL1s that specify the ply thicknesses multiply the DESVAR value by
.00125. The intent here is to make the DESVAR value a count of the required number
of plies under the assumption that each laminae has a thickness of .00250. Discrete
variables could have been used in this example to make the design variables an integer
ply count, but this was not done.
It is also necessary to take care in specifying the location of the bottom surface of the
laminate (Z0 on the PCOMP entry). The grid locations are taken to specify the OML
(outer mold line) of the wing that has been specified to achieve the desired
aerodynamic shape and performance. The structural material must therefore lie
within the OML so that Z0 must be designed to be equal to the total thickness of the
laminate. Also note that the direction of the bottom surface is governed by the order
in which the element is numbered so that care must be taken to make this order
consistent with the design intent. The Z0 quantity must be designed using a DVPREL1
entry so that the bottom surface changes as the individual plies change. Note that the
-.0025 factor accounts for the fact that the total thickness is twice the thickness given
on the PCOMP entry.
Main Index Main Index
473 CHAPTER 7
Example Problems
Listing 7-24 Input File Fragments for DSOUG9
ID MSC, DSOUG9 $
TIME 200 $
SOL 200 $
CEND
TITLE =INTERMEDIATE COMPLEXITY WING **STAT & EIGN CONSTRAINTS* DSOUG9
SUBTIT = QUAD4 ELEMENTS WITH 153 DESIGN VARIABLES ** CMS ***
LABEL = COMPOSITE STRUCTURE WITH FIBER ORIENTATIONS (0,90,+45,-45)
ECHO = SORT
SPC = 1
MPC = 200
DESOBJ= 1001
DESSUB= 1
SUBCASE 1
ANALYSIS=STATICS
METHOD=10
LOAD = 1
SUBCASE 2
ANALYSIS=STATICS
METHOD=10
LOAD = 2
SUBCASE 3
DESSUB = 3
ANALYSIS=MODES
METHOD=10
BEGIN BULK
DCONSTR 1 1 1.0
DCONSTR 1 2 1.0
DCONSTR 1 3 1.0
DCONSTR 1 4 1.0
DCONSTR 1 11 1.0
DCONSTR 1 12 1.0
DCONSTR 1 13 1.0
DCONSTR 1 14 1.0
DCONSTR 1 21 -5.7+4 6.7+4
DCONSTR 1 31 -3.9+4 3.9+4
DCONSTR 3 200 40.0
DESVAR 33 RIBS 0.21 0.20
DESVAR 34 SHEAR1 0.21 0.20
DESVAR 56 SHEAR23 0.21 0.20
DESVAR 57 POSTS 0.21 0.20
DESVAR 1101 CTRIA31 1.000 1.00000
DESVAR 1432 CQUAD32 1.001 1.00000
DOPTPRM FSDMAX 10 DESMAX 20 DELP 0.50 DPMIN 0.001
delx 0.49 p1 1 p2 15
DRESP1 1 STRAIN1 CFAILUREPCOMP 5 1 10001
DRESP1 2 STRAIN1 CFAILUREPCOMP 5 2 10001
DRESP1 3 STRAIN1 CFAILUREPCOMP 5 3 10001
DRESP1 4 STRAIN1 CFAILUREPCOMP 5 4 10001
DRESP1 11 STRAIN1 CFAILUREPCOMP 5 1 30031
30001 30002 30003 30004 30005 30006 30007 30008
30009 30010 30011 30012 30013 30014 30015 30016
30017 30018 30019 30020 30021 30022 30023 30024
30025 30026 30027 30028 30029 30030
DRESP1 12 STRAIN1 CFAILUREPCOMP 5 2 30031
30001 30002 30003 30004 30005 30006 30007 30008
DRESP1 21 STRESS STRESS PROD 2 10001
DRESP1 31 STRESS STRESS PSHEAR 3 40001
40002 40003 40004 40005 40006 40007 40008 40009
40010 40011 40012 40013 40014 40015 40016 40017
40018 40019 40020 40021 40022 40023 40024DRESP1 101 W WEIGHT
ALL
Main Index Main Index


474
Finally, the design task specifies a constraint on a CFAILURE response. The correct
specification of this response requires a combination of inputs that are described here.
On the DRESP1, the CFAILURE response type is invoked for each of the designed
properties for each of the PCOMPS on the upper surface. Item code 5 is specified,
which from Table 6-2 in the MSC.Nastran Quick Reference Guide is seen to be a failure
mode for direct stresses. The type of the failure mode is specified as TSAI (for Tsai-Wu
theory) on the PCOMP entry. Table 13-1 in Composites in Chapter 13 of the
MSC.Nastran Reference Manual gives the formula for the Tsai-Wu criteria as:
DRESP2 1001 STRWGT 1001
DRESP1 101
DEQATN 1001 STRWGT(TOTAL) = TOTAL - 115.
DRESP1 200 FREQ FREQ 2 1
DVPREL1 33 PSHEAR 40001 T .02
33 0.1
DVPREL1 57 PROD 10001 A .0001
57 0.1
DVPREL1 1101 PCOMP 10001 T1
1101 0.00125
DVPREL1 1102 PCOMP 30001 T1
1102 0.00125
DVPREL1 4101 PCOMP 10001 Z0
1101 -.00250 1201 -.00250 1301 -.00250 1401 -
.00250
DVPREL1 4102 PCOMP 30001 Z0
1102 -.00250 1202 -.00250 1302 -.00250 1402 -
.00250
DYNRED 100 300.0
EIGRL 10 4 0
FORCE 1 1 0 1.0 205.0 -7380.0 926.0
FORCE 1 2 0 1.0 -205.0 7380.0 926.0
GRID 1 63.5000 90.0000 1.1250
GRID 2 63.5000 90.0000 -1.1250
MAT1 10 1.05E+7 4.04E+6 0.30000 0.10000 0.00000 0.00000 0.00000
6.70E+4 5.70E+4 3.90E+4
MAT8 70 1.85E+7 1.60E+6 0.25000 0.65E6 0.05500
0.0 0.0 100. 1.15E+5 1.15E+5 1.15E+5 1.15E+5 1.0E+15
MAT8 72 1.85E+7 1.60E+6 0.25000 0.65E6 0.05500
0.0 0.0 100.
MPC 200 1 1 1.0 201 1 -.8924
201 5 -1.0040 201 2 -.4512
201 4 .5076
PARAM WTMASS 0.00259
PCOMP 10001 -.0105 0.0 0.65E6 TSAI SYM
70 0.500 0.0 YES 70 0.500 90. YES
70 0.500 45. YES 70 0.500 -45. YES
PCOMP 12001 -.0105 0.0 SYM
72 0.500 0.0 YES 72 0.500 90. YES
72 0.500 45.0 YES 72 0.500 -45. YES
PROD 10001 10 1.00
PSHEAR 40001 10 1.0
SPC1 1 6 201 THRU 217
SPC1 1 123456 79 THRU 88
ENDDATA
Main Index Main Index
475 CHAPTER 7
Example Problems
The values are specified on the material entry, MAT8 in this
example.
The DOPTPRM Bulk Data entry is used to specify that up to 10 FSD cycles are to be
performed followed by up to 20 mathematical programming cycles, where up to
refers to the fact that both the FSD and MP algorithms will terminate before the
specified number of cycles if the design is judged to have converged.
Listing 7-25 shows design history data for the example while Figure 7-23 shows some
of the same data in xy plot form. It is seen from Listing 7-25 that ten cycles of FSD were
performed followed by six MP cycles. The FSD algorithm did an effective job of
coming up with a good starting design for the MP algorithm in that the maximum
constraint value is reduced from 570 to near 1 (note the log scale in Listing 7-25). The
final FSD design has a weight of 75.16 lbs but is infeasible. The final design is feasible
and has a weight of 78.1 lbs. For this simple example, the CPU time for each design
cycle was approximately 2 secs. for each FSD cycles and 8 secs. for each MP cycle on a
computer using an MSC/ Linus operating system, illustrating the performance
advantage of the FSD method.
1
X
t
-----
1
X
c
-------
\ .
| |

1
1
Y
t
-----
1
Y
c
-------
\ .
| |

2

1
2
X
t
X
c
------------

2
2
Y
t
Y
c
-----------

12
2
S
2
-------- 2F
12

2
FI = + + + + +
X
t
X
c
Y
t
Y
c
S and F
12
, , , ,
Main Index Main Index


476
Listing 7-25 Design Cycle History for DSOUG9
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 17
NUMBER OF FULLY STRESSED DESIGN CYCLES COMPLETED 10
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 6
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 5.392062E+01 5.705082E+02
1 FSD 5.783698E+01 N/A 2.566903E+02
2 FSD 6.309813E+01 N/A 1.161377E+02
3 FSD 6.789255E+01 N/A 5.556812E+01
4 FSD 7.195250E+01 N/A 2.609151E+01
5 FSD 7.492827E+01 N/A 1.171095E+01
6 FSD 7.603766E+01 N/A 5.958905E+00
7 FSD 7.610654E+01 N/A 2.610097E+00
8 FSD 7.636430E+01 N/A 1.177912E+00
9 FSD 7.595876E+01 N/A 9.161955E-01
10 FSD 7.516133E+01 N/A 1.998079E+00
11 7.968182E+01 7.968294E+01 -1.397905E-05 2.672843E-01
12 7.883505E+01 7.883524E+01 -2.322635E-06 1.180389E-01
13 7.833035E+01 7.833054E+01 -2.337600E-06 6.503105E-03
14 7.831383E+01 7.831381E+01 1.948416E-07 2.219319E-03
15 7.812685E+01 7.812686E+01 -1.953078E-07 3.269792E-03
16 7.809853E+01 7.809856E+01 -3.907573E-07 2.132654E-04
Main Index Main Index
477 CHAPTER 7
Example Problems
Figure 7-23 Selected Design Cycle Results for DSOUG9
Main Index Main Index


478
7.10 Acoustic Optimization
Acoustic Optimization uses acoustic pressures as a design response. These are
computed from a solution of the coupled fluid-structure interaction problem. An
optimal design can thus be found based not only on a consideration of acoustic
pressures, but structural responses as well.
This example considers a closed box with fluid elements on the interior. An acoustic
source is located at one end of the box, with a transducer located at the opposite end.
The design goal is to modify the thicknesses of the box walls such that the peak
acoustic pressure at the transducer is minimized without increasing the weight of the
box.
The box geometry and property groups of thicknesses to be modified are shown in
Figure 7-24. Six design variables are to be related to six of these property groups (the
third property group in Figure 7-24 remains fixed.) The model consists of 1000
structural elements and 2000 fluid elements.
Figure 7-24 Acoustic Box Showing Portions Designed by Each Design Variable
(Prop 3 is Fixed)
Main Index Main Index
479 CHAPTER 7
Example Problems
An interesting feature of this, as well as many other dynamic response problems, is
that a number of response peaks may exist. In addition, during the course of design
optimization these peaks may not only increase or decrease, but shift as well. This
example presents a useful way of addressing this problem.
Figure 7-25 shows an arbitrary pressure distribution as a function of frequency, .
Three frequencies of interest are shown as , , and . What we would like to do is
to minimize the maximum pressure response over all of these points.
Figure 7-25 Minimization of Response Peaks
We can choose a design variable to represent a peak threshold, here shown in the
figure as (is just a constant of proportionality to facilitate scaling the threshold).
The difference between and the pressure distribution must be a positive quantity
at all frequencies of interest. Thus, constraints can be written which require , , and
to be positive distances. The design objective can then be to minimize , or
minimize
Eq. 7-13
subject to:
Eq. 7-14
P f ( )
f
1
f
2
f
3
c
1
c
2
c
3
P f ( )
x
f
1
f
2
f
3
f
x
x
c
1
c
2
c
3
x
x
P f 1 ( ) x 0
P f 2 ( ) x 0
P f 3 ( ) x 0
Main Index Main Index


480
As the optimizer decreases the threshold, the constraints ensure that the peaks are
reduced as well. Any number of these constraints can be written to cover all
frequency ranges of interest (or, in the case of transient analysis, time steps).
Due to the size of this problem, Listing 7-26 is an excerpt of the input file consisting
mainly of the design model entries, as well as some the analysis entries. Analysis
modeling for Acoustics is covered in the Additional Topics on page 555 of the
MSC.Nastran Reference Guide.
Listing 7-26 Fragments from Input Data File DSOUG10
ID MSC, DSOUG10 $
TIME 9999
SOL 200 $ MODAL FREQUENCY RESPONSE
CEND
TITLE = DESIGN OPTIMIZATION WITH ACOUSTICS
SUBTITLE = ACOUSTIC AND STRUCTURAL ELEMENTS
LABEL = BOXAE1.DAT
SET 20 = 11280
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
ECHO = SORT(PARAM,EIGC,EIGRL,FREQ,DESVAR,DCONSTR,DRESP1,DRESP2,DEQATN,
DVPREL1)
SPC = 1
DESGLB = 5
DISP(PHASE) = 20
METHOD(STRUCTURE) = 20
METHOD(FLUID) = 30
$
FREQUENCY = 200
DLOAD = 100
DESSUB = 10
DESOBJ = 100
ANALYSIS = MFREQ
$
BEGIN BULK
$
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
EIGRL 20 300.
EIGRL 30 15. 200. 0 105.
$
$ SOUND PRESSURE LEVEL
PARAM,RMS,YES
$ REFERENCE PRESSURE FOR DB AND DBA
PARAM,PREFDB,2.-5
$
PARAM AUTOSPC NO
$
$ FLUID/STRUCTURE INTERFACE
ACMODL,DIFF , , , ,0.01
$
$ STRUCTURAL DAMPING
PARAM,G,0.02
$
PARAM POST -1
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
RLOAD1,100,101,,,102
DAREA,101,1288,3,100.
Main Index Main Index
481 CHAPTER 7
Example Problems
TABLED1,102
,0.,1.,1000.,1.,ENDT
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
FREQ1 200 40. 4.0 41
FREQ 200 97.5 102.5
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$ SENSITIVITY
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$
$
DESVAR,1,P1,0.02493,0.0001,1.
DESVAR,2,P2,0.01953,0.0001,1.
DESVAR,4,P4,0.02047,0.0001,1.
DESVAR,5,P5,0.02596,0.0001,1.
DESVAR,6,P6,0.02175,0.0001,1.
DESVAR,7,P7,0.02426,0.0001,1.
$
DESVAR 8 BETA 1.0 0.001
$
DVPREL1,1,PSHELL,1,T,0.0001
,1,1.
DVPREL1,2,PSHELL,2,T,0.0001
,2,1.
DVPREL1,4,PSHELL,4,T,0.0001
,4,1.
DVPREL1,5,PSHELL,5,T,0.0001
,5,1.
DVPREL1,6,PSHELL,6,T,0.0001
,6,1.
DVPREL1,7,PSHELL,7,T,0.0001
,7,1.
$
DRESP2 100 BETA 100
DESVAR 8
DEQATN 100 OBJ(BETA) = 10000.0 * BETA
DRESP1 1 PRESS FRDISP 1 11280
DRESP1 2 WEIGHT WEIGHT
DCONSTR 5 2 2890. 2910.
DRESP2 11 PRESBET 10
DESVAR 8
DRESP1 1
DEQATN 10 F(BETA,PRES) = 100.0 * BETA - PRES + 100.
DCONSTR 10 11 100.
$
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
$
$
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$ STRUCTURAL MODEL
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$
PSHELL 1 1 .02493 1
PSHELL 2 1 .01953 1
PSHELL 3 1 .100 1
PSHELL 4 1 .02047 1
PSHELL 5 1 .02596 1
PSHELL 6 1 .02175 1
PSHELL 7 1 .02426 1
$
GRID 1 0.0 0.0 0.0
GRID 2 2. 0.0 0.0
Main Index Main Index


482
GRID 3 2. 0.0 1.
GRID 1293 1.5 .5 1.
GRID 1294 1.6 .5 1.
GRID 1295 1.7 .5 1.
$
CQUAD4 1 1 1 9 29 28
CQUAD4 2 1 9 10 30 29
CQUAD4 998 3 1138 1139 296 317
CQUAD4 999 3 1139 1140 275 296
CQUAD4 1000 3 1140 897 5 275
$
$ THIS SECTION CONTAINS THE LOADS,CONSTRAINTS, AND CONTROL BULK DATA ENTRIES
$
$
SPC 1 1 123 0.0
SPC 1 2 123 0.0
SPC 1 986 4
SPC 1 987 4
SPC 1 975 4
$
$
MAT1 1 2.+11 .3 7600.
$
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$ ACOUSTIC MODEL
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
$
$-------2-------3-------4-------5-------6-------7-------8-------9-------10------
GRID 10001 0.0 0.0 0.0 -1
GRID 10002 2. 0.0 0.0 -1
GRID 10003 2. 0.0 1. -1
GRID 11279 1. .5 .1 -1
GRID 11280 1. .5 0.0 -1
GRID 11281 1. .6 1. -1
GRID 12539 2. 1. .3 -1
GRID 12540 2. 1. .2 -1
GRID 12541 2. 1. .1 -1
$
CHEXA 10001100 10004 10126 10127 10009 10018 10137 +
+ 10138 10019
CHEXA 10002100 10009 10127 10128 10010 10019 10138 +
+ 10139 10020
CHEXA 11999100 12411 12530 12531 12412 12422 12540 +
+ 12541 12423
CHEXA 12000100 12412 12531 12532 12413 12423 12541 +
+ 10006 12424
$
PSOLID,100,100,,,,1,PFLUID
$
MAT10,100,,1.293,200.
DOPTPRM DESMAX 30 P1 1 P2 15 APRCOD1
GMAX 0.015 DELX 0.5 CONV1.005
$
ENDDATA
Main Index Main Index
483 CHAPTER 7
Example Problems
Turning to the listing, we see that the weight budget is established as a global
constraint. The weight response is defined on DRESP1 number 2; bounds are placed
on this response with DCONSTR number 5, which is in turn referenced in Case
Control by the DESGLB command. The bounds on weight allow less than one third
of one percent variation from the initial weight. Allowing some variation is usually
preferred over an equality constraint since it not only yields a better conditioned
problem, but is often consistent with our design goals as well.
The objective function is defined as a constant times design variable number 8 using
the DRESP2 number 100 entry. This refers to DEQATN 100 which defines
Eq. 7-15
This is defined as the objective function by the DESOBJ 100 Case Control command.
The constraints on response peaks are defined by first identifying the pressure
responses themselves with DRESP1 entry number 1. This designates the "1"
component of displacement at grid 11280 as the response. Since this grid is part of the
fluid model, the response is a pressure with units of . Since the ATTB field is
blank on this DRESP1 entry, the response is computed for all output frequencies.
These pressure responses are used as input to DEQATN 10 via DRESP2 number 11.
These entries, in combination with DCONSTR number 10, identify constraints on
pressure response of the form:
Eq. 7-16
Note that this is similar to Eq. 7-14, with the constant added to avoid a bound of
zero on the constraint. It should be clear that the constants have been chosen to scale
the objective and responses to values that would minimize numerical difficulties. In
this example, and are set at 100. This is in line with a nominal peak pressure.
Listing 7-27 and Figure 7-26, and Figure 7-27 and Figure 7-28 present optimization
results. The design cycle history of Listing 7-27 shows the objective has decreased
from 10000.0 to 282.8 in thirteen iterations. A comparison of the plots of Figure 7-26
and Figure 7-27 indicates that although the objective appears to have reached its
minimum after eleven design cycles, the design variables continue to change. This
indicates that the design space is quite flat for this case, giving the designer some
freedom in the selection of the design variable values. The benefit of the procedure is
dramatically demonstrated in Figure 7-28, which depicts a reduction in the peak
F 10000. = x
8

N m
2

k
1
x
8
P f ( ) k
2
+ k
2

k
2
k
1
k
2
Main Index Main Index


484
pressure from 205 to 3 . The technique of minimizing the maximum
pressure has resulted in frequency response function that has three peaks near the
maximum value whereas the initial function has a single sharp peak.
N/ mm
2
N/ mm
2
Main Index Main Index
485 CHAPTER 7
Example Problems
Listing 7-27 Design Cycle History for DSOUG10
Main Index Main Index


486

***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 14
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 13
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.000000E+04 1.047036E+00
1 5.000000E+03 5.000000E+03 0.000000E+00 3.324837E-02
2 5.000115E+03 5.000115E+03 0.000000E+00 5.938492E-03
3 2.500058E+03 2.500058E+03 0.000000E+00 1.071139E-01
4 2.484153E+03 2.484153E+03 0.000000E+00 1.760712E-02
5 1.863115E+03 1.863115E+03 0.000000E+00 1.761840E-06
6 1.397336E+03 1.397336E+03 0.000000E+00 1.843811E-02
7 1.048002E+03 1.048002E+03 0.000000E+00 -1.774032E-06
8 7.860015E+02 7.860015E+02 0.000000E+00 -1.774032E-06
9 5.360015E+02 5.360015E+02 0.000000E+00 5.554199E-04
10 2.860015E+02 2.860015E+02 0.000000E+00 7.388764E-03
11 2.843468E+02 2.843468E+02 0.000000E+00 8.308182E-03
12 2.823511E+02 2.823511E+02 0.000000E+00 5.078888E-03
13 2.828331E+02 2.828331E+02 0.000000E+00 2.645645E-03
---------------------------------------------------------------------------------------------------------------
1 DESIGN OPTIMIZATION WITH ACOUSTICS FEBRUARY 13, 2003 MSC.NASTRAN 2/12/03 PAGE 331
ACOUSTIC AND STRUCTURAL ELEMENTS
0 BOXAE1.DAT
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | P1 | 2.4930E-02 : 2.2428E-02 : 2.2154E-02 : 1.7599E-02 : 2.2632E-02 : 2.4025E-02 :
2 | 2 | P2 | 1.9530E-02 : 1.5055E-02 : 1.4713E-02 : 1.5493E-02 : 1.6190E-02 : 1.4141E-02 :
3 | 4 | P4 | 2.0470E-02 : 2.9199E-02 : 2.9896E-02 : 3.7955E-02 : 4.2955E-02 : 4.3238E-02 :
4 | 5 | P5 | 2.5960E-02 : 2.0812E-02 : 2.0825E-02 : 2.4126E-02 : 2.9151E-02 : 3.0608E-02 :
5 | 6 | P6 | 2.1750E-02 : 3.1750E-02 : 3.2472E-02 : 3.7073E-02 : 3.3682E-02 : 3.5170E-02 :
6 | 7 | P7 | 2.4260E-02 : 1.9444E-02 : 1.8950E-02 : 1.1453E-02 : 6.4526E-03 : 4.0505E-03 :
7 | 8 | BETA | 1.0000E+00 : 5.0000E-01 : 5.0001E-01 : 2.5001E-01 : 2.4842E-01 : 1.8631E-01 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | P1 | 2.8256E-02 : 2.8926E-02 : 2.8926E-02 : 2.8925E-02 : 2.8923E-02 : 3.1006E-02 :
2 | 2 | P2 | 1.3790E-02 : 1.3760E-02 : 1.3760E-02 : 1.3760E-02 : 1.3758E-02 : 8.7585E-03 :
3 | 4 | P4 | 3.8190E-02 : 3.8408E-02 : 3.8408E-02 : 3.8407E-02 : 3.8406E-02 : 3.9782E-02 :
4 | 5 | P5 | 3.3954E-02 : 3.4772E-02 : 3.4772E-02 : 3.4771E-02 : 3.4770E-02 : 3.4336E-02 :
5 | 6 | P6 | 3.1546E-02 : 3.2732E-02 : 3.2732E-02 : 3.2731E-02 : 3.2729E-02 : 3.5156E-02 :
6 | 7 | P7 | 3.0006E-03 : 6.5768E-04 : 6.5768E-04 : 6.5925E-04 : 6.6553E-04 : 6.7114E-04 :
7 | 8 | BETA | 1.3973E-01 : 1.0480E-01 : 7.8600E-02 : 5.3600E-02 : 2.8600E-02 : 2.8435E-02 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 12 : 13 : 14 : 15 : 16 : 17 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | P1 | 2.6006E-02 : 2.1006E-02 :
2 | 2 | P2 | 1.0261E-02 : 1.0188E-02 :
3 | 4 | P4 | 4.4782E-02 : 4.9782E-02 :
4 | 5 | P5 | 3.4393E-02 : 3.4391E-02 :
5 | 6 | P6 | 3.6662E-02 : 3.9233E-02 :
6 | 7 | P7 | 6.6993E-04 : 6.7202E-04 :
7 | 8 | BETA | 2.8235E-02 : 2.8283E-02 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 13.
Main Index Main Index
487 CHAPTER 7
Example Problems
Figure 7-26 Objective Function versus Design Cycle
Main Index Main Index


488
Figure 7-27 Design Variables versus Design Cycle
Main Index Main Index
489 CHAPTER 7
Example Problems
Figure 7-28 Acoustic Optimization Sound Pressure Levels: Initial and Final
Distributions
Main Index Main Index


490
7.11 RMS Response
This example simulates a car subjected to random loading that is fully correlated. The
goal of this analysis is to minimize the RMS displacement at the center of gravity (cg)
of the vehicle due to the random loading. A sketch of the simple vehicle representation
is given below.
The goal of the optimization task is to minimize the RMS displacement of the vehicle
CG while constraining the stress. Design variables are the stiffness and damping
(spring and shock absorber) of the suspension. Numerical values for the design limits
are:
The complete input data file is given in Listing 7-28. Two subcases are required in this
case -- one for the front wheel and one for the back wheel. A unit enforced
displacement is applied to each wheel using the enforced motion method.
Listing 7-28 Input Data File for DSOUG11
$
$ DSOUG11.DAT - RANDOM INPUT AT WHEELS - ENFORCED MOTION
$ DIRECT METHOD
$ STRESS CONSTRAINTS AT DISCRETE FREQ
$ MINIMIZE ACCE AT GRID PT 5
$ CROSS SPECTRUM
$ USING NEW ENFORCED MOTION
$
120
1 5 2
3 4
y
x
100 K 1000 < <
1.0 B 10. < <
4500
Main Index Main Index
491 CHAPTER 7
Example Problems
ID MSC, DSOUG11 $
SOL 200 $
CEND
TITLE = SIMPLE CAR WITH RANDOM INPUT
SPC = 100
FREQUENCY = 130
STRESS(PHASE) = ALL
DISP(PHASE) = ALL
DESOBJ= 101
ANALYSIS = DFREQ
DESSUB = 800
$
$ THIS RANDOM CASE CONTROL CALLOUT IS NEEDED ONLY
$ FOR THE XYPLOT REQUEST WHEN USING OPTIMIZATION
$
RANDOM = 1000
SUBCASE 1
DLOAD = 111
SUBCASE 2
DLOAD = 112
$
OUTPUT (XYPLOT)
XTITLE = FREQUENCY (HZ)
YTITLE = DISP PSD AT GRID PT 5
XYPUNCH DISP PSDF /5(T2)
$
BEGIN BULK
$
$ CAR BODY
$
GRID 1 0. 0. 0.
GRID 2 120. 0. 0.
GRID 5 60. 0. 0.
$
$ WHEELS
$
GRID 3 0. -10. 0.
GRID 4 120. -10. 0.
$
CBAR 5 11 1 5 0. 1. 0.
CBAR 6 11 5 2 0. 1. 0.
$
PBAR 11 12 1.E2 1.E3 1.E3
100. 0. -100. 0.
MAT1 12 3.E+7 7.8E-4
$
$ CONSTRAIN TO A PLANAR PROBLEM
$
SPC1 100 1345 1 2 5
SPC1 100 13456 3 4
SPC1,100,2,3,4
$
CONM2 10 1 2.5
CONM2 15 2 2.5
CONM2 20 5 5.0
CONM2 25 3 2.5
Main Index Main Index


492
CONM2 26 4 2.5
$
CBUSH 100 1000 1 3 0
CBUSH 200 1000 2 4 0
PBUSH 1000 K 200.
B 2.
$
RLOAD2 111 222 444DISP
SPCD 222 3 2 1.0
TABLED1 444
0. 1. 100. 1. ENDT
$
RLOAD2 112 223 444DISP
SPCD 223 4 2 1.0
$
FREQ2 130 0.1 2. 20
$
$ DEFINE THE INPUT PSD
$
RANDPS 1000 1 1 1. 0. 145
RANDPS 1000 2 2 1. 0. 145
RANDPS 1000 1 2 1. 0. 146
RANDPS 1000 1 2 0. 1. 147
TABRND1 145
.1 .1 5. 1. 10. .05 ENDT
$
TABRND1 146
0.1000 0.0990 0.1096 0.1005 0.1202 0.1022 0.1318 0.1040
0.1445 0.1059 0.1585 0.1079 0.1738 0.1101 0.1905 0.1124
0.2089 0.1147 0.2291 0.1172 0.2512 0.1197 0.2754 0.1221
0.3020 0.1246 0.3311 0.1269 0.3631 0.1289 0.3981 0.1305
0.4365 0.1314 0.4786 0.1315 0.5248 0.1304 0.5754 0.1277
0.6310 0.1228 0.6918 0.1151 0.7586 0.1038 0.8318 0.0879
0.9120 0.0665 1.0000 0.0382 1.0965 0.0020 1.2023 -0.0434
1.3183 -0.0986 1.4454 -0.1636 1.5849 -0.2372 1.7378 -0.3159
1.9055 -0.3934 2.0893 -0.4593 2.2909 -0.4984 2.5119 -0.4910
2.7542 -0.4149 3.0200 -0.2508 3.3113 0.0072 3.6308 0.3364
3.9811 0.6685 4.3652 0.8819 4.7863 0.8223 5.2481 0.3435
5.7544 0.2965 6.3096 0.6842 6.9183 0.5743 7.5858 0.0891
8.3176 0.2820 9.1201 0.1964 10.0000 0.0063 ENDT
$
TABRND1 147
0.1000 0.0142 0.1096 0.0159 0.1202 0.0177 0.1318 0.0198
0.1445 0.0221 0.1585 0.0248 0.1738 0.0279 0.1905 0.0313
0.2089 0.0352 0.2291 0.0397 0.2512 0.0448 0.2754 0.0506
0.3020 0.0572 0.3311 0.0648 0.3631 0.0734 0.3981 0.0832
0.4365 0.0944 0.4786 0.1070 0.5248 0.1212 0.5754 0.1371
0.6310 0.1547 0.6918 0.1741 0.7586 0.1951 0.8318 0.2173
0.9120 0.2401 1.0000 0.2625 1.0965 0.2830 1.2023 0.2993
1.3183 0.3084 1.4454 0.3061 1.5849 0.2875 1.7378 0.2467
1.9055 0.1775 2.0893 0.0749 2.2909 -0.0631 2.5119 -0.2318
2.7542 -0.4160 3.0200 -0.5848 3.3113 -0.6898 3.6308 -0.6687
3.9811 -0.4624 4.3652 -0.0505 4.7863 0.4968 5.2481 0.8888
5.7544 0.8037 6.3096 0.3101 6.9183 -0.2722 7.5858 -0.5008
8.3176 -0.2390 9.1201 0.0928 10.0000 0.0496 ENDT
$
Main Index Main Index
493 CHAPTER 7
Example Problems
$ -----------------------------------------------------------
$
$ DESIGN MODEL
$
DESVAR,1,K2,1.,0.5,5.
DESVAR,2,B2,1.,0.5,5.
$
DVPREL1,101,PBUSH,1000,K2,100.
,1,200.
DVPREL1,201,PBUSH,1000,B2,1.
,2,2.0
$
DRESP1,101,MING5T2,RMSDISP,,,2,1000,5
$
DCONSTR,800,801,,4500.
DCONSTR,800,802,,4500.
DRESP1,801,E5PTC,FRSTRE,PBAR,,12,,11
DRESP1,802,E5PTE,FRSTRE,PBAR,,14,,11
$
DOPTPRMDESMAX 20 P1 1 P2 14 METHOD 3
$
ENDDATA
The stress constraints are applied to the stresses resulting from these unit
displacements in the individual subcases while the RMS response is calculated by
integrating the results from the two subcases as given by Eq. 2-26 on page 60.
Unlike a non-optimization run, where the RANDPS is called out by the RANDOM
Case Control command, the optimization run references the RANDPS entries through
the ATTB field (field 8) on the DRESP1 entry. There are four RANDPS entries for this
case. The first and second RANDPS entries represent the same autospectrum applied
at the front and back wheel, respectively. The third and fourth RANDPS entries
represent the real and imaginary parts of the cross spectrum between the front and
back wheels. Note that these spectra are applied the same way as in a non-
optimization run. The RTYPE field (field 4) on the DRESP1 entry specifies that an
RMS displacement is to be used by specifying RMSDISP. In this example, we use the
rms displacement (RMSDISP) as the objective function.
Listing 7-29 shows the design history for this example while Figure 7-29 shows the
objective as a function of design cycle and Figure 7-30 shows the design variable
values. We see that the damping variable goes to its upper bound while the stiffness
is reduced relative to its initial value. Figure 7-31 shows the initial and final PSD plots
. The figures also shows that it is the area under these two curves that is reduced as
the objective RMS displacement decrease from 1.906 to 0.627 inches.
Main Index Main Index


494
Listing 7-29 Design History

***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 11
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 10
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.905989E+00 -1.686030E-01
1 3.156076E-01 2.041587E+00 -8.454106E-01 1.185777E-01
2 -3.101647E+00 2.027636E+00 -2.529686E+00 -9.756684E-02
3 1.684823E+00 1.540025E+00 9.402309E-02 -3.569878E-01
4 -6.815850E-01 1.263362E+00 -1.539501E+00 N/A
5 8.680997E-01 9.153620E-01 -5.163237E-02 N/A
6 8.145639E-01 8.076525E-01 8.557389E-03 N/A
7 7.562498E-01 7.334383E-01 3.110227E-02 N/A
8 6.907170E-01 6.701388E-01 3.070729E-02 N/A
9 6.399633E-01 6.239330E-01 2.569246E-02 N/A
10 6.239330E-01 6.239330E-01 0.000000E+00 N/A
---------------------------------------------------------------------------------------------------------------

0
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | K2 | 1.0000E+00 : 1.2000E+00 : 9.6004E-01 : 8.9541E-01 : 7.1633E-01 : 5.7306E-01 :
2 | 2 | B2 | 1.0000E+00 : 1.2000E+00 : 1.4369E+00 : 1.7243E+00 : 2.0692E+00 : 2.4830E+00 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | K2 | 5.0000E-01 : 5.0000E-01 : 5.0000E-01 : 5.0000E-01 : 5.0000E-01 :
2 | 2 | B2 | 2.9796E+00 : 3.5755E+00 : 4.2906E+00 : 5.0000E+00 : 5.0000E+00 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 10.
1 SIMPLE CAR WITH RANDOM INPUT JUNE 19, 2002 MSC.NASTRAN 4/ 9/01 PAGE 139
Main Index Main Index
495 CHAPTER 7
Example Problems
Figure 7-29 Objective Function
Figure 7-30 Design Variables
Main Index Main Index


496
Figure 7-31 Comparison of PSDF DISP
Main Index Main Index
497 CHAPTER 7
Example Problems
7.12 Transient Dynamic Optimization
This simple example demonstrates the ability to perform structural optimization in
conjunction with a transient analysis. The example is a portal frame modeled with
three bar elements that is excited by an enforced motion at its base. This is a simple
representation of a civil engineering structure subjected to a seismic load.
The frame is subjected to a horizontal ground motion in the form of a half sine pulse:
The design task is to minimize the weight of the structure while imposing limits on
the rms structural displacements and stresses as well as on the first natural frequency:
The frame is modeled using three finite element bars (symmetry of the problem could
have been used to limit this to two elements). This example has been adapted from a
PhD thesis (see Cassis, J.H., in Reference 3.) and has some non-standard modeling
techniques that are included here because they illustrate the use of DVPREL2 and
DRESP2 entries. In particular, the two design variables for the problem are the
bending inertia of the horizontal and vertical beams. The area of the beams is
computed using:
and the stress at the end of the bars is computed as the ratio of the bending moment
divided by the section modulus, where the section modulus is given by:
The transient response was performed over a period of 250 milliseconds and
responses were retained at every 5 milliseconds. An RMS response is computed by
using the RSS option for the ATTB field on the DRESP1 entries for transient
displacement and transient force response and then dividing the result by 51.0 (the
number of samples used to compute the RSS).
u
e
t ( ) 1.0 30t ( ) sin t 30.0 =
f
i
9.55 t 3 >
RMS Displacement 0.1in.
RMS Stress 3000 psi
A 0.465 I =
S 60.6I 84000 + 290 =
Main Index Main Index


498
The input file for this example is shown in Listing 7-30. It is seen that there are two
subcases, with the first subcase performing a modal analysis while the second
performs the modal transient analysis (since the same METHOD entry is used for both
subcases and the boundary condition does not change, only one modal analysis is
performed). The enforced displacement is applied using the SPCD entries invoked by
a a TLOAD2 entry to specify the spatial location of the enforced motion and the
temporal variation is specified directly on the TLOAD2 entry.
Listing 7-30 Input Data File for DSOUG12
ID MSC, DSOUG12 $
TIME 60
SOL 200
DIAG 8 $
CEND
TITLE= TRANSIENT DYNAMIC OPTIMIZATION - DSOUG12
METHOD = 300
DISPL = ALL
DESOBJ = 1000
SPC = 100
SUBCASE 2
DESSUB = 2
ANALYSIS = MODES
SUBCASE 3
DESSUB = 3
ANALYSIS = MTRAN
STRESS = ALL
FORCE = ALL
DLOAD = 310
TSTEP = 335
BEGIN BULK
$-----------------------------------------------------------------------
$ ANALYSIS MODEL
$-----------------------------------------------------------------------
$
GRID 1 0. 0. 0. 23456
GRID 2 0. 180. 0. 345
GRID 3 180. 180. 0. 345
GRID 4 180. 0. 0. 23456
$
$ VERTICAL COLUMNS
CBAR 1 101 1 2 -1. 0. 0.
CBAR 2 101 4 3 -1. 0. 0.
$ HORIZONTAL BAR
CBAR 3 102 2 3 0. 1. 0.
$
$ EIGENVALUE EXTRACTION METHOD
EIGRL 300 10
MAT1 200 30.E6 0.33 0.28
PARAM WTMASS 2.588E-3
PBAR 101 200 7.92 290.10.
PBAR 102 200 7.92 290. 10.
SPC1 100 1 1 4
$ TRANSIENT LOADING DATA
$...SPECIFY HALF-SINE WAVE ENFORCED DISPLACEMENT BEGINNING AT T=0 :
SPCD 320 1 1 1.0
Main Index Main Index
499 CHAPTER 7
Example Problems
SPCD 320 4 1 1.0
TLOAD2 310 320 DISP 0. .10472 4.77465 -90.
0.0 0.0
$
TSTEP 335 250 0.001 5
$-----------------------------------------------------------------------
$ DESIGN MODEL
$-----------------------------------------------------------------------
$
DESVAR 1 X1 1.0 .1 10.
DESVAR 2 X2 1.0 0.1 10.
$
$...DV TO I RELATIONS:
DVPREL1 10 PBAR 101 I1 1.0 0.0
1 1200.
DVPREL1 11 PBAR 102 I1 1.0 0.0
2 1200.
$
$...DV TO A RELATIONS:
DVPREL2 12 PBAR 101 A 0.1 200
DESVAR 1
DVPREL2 13 PBAR 102 A 0.1 200
DESVAR 2
DEQATN 200 AREA(X) = 0.465*SQRT(1200. * x)
$
$...DEFINE WEIGHT OBJECTIVE:
DRESP1 1 W WEIGHT
dresp2 1000 sweig 1000
dresp1 1
deqatn 1000 sweig(weight) = weight/100.
$
$...FREQUENCY CONSTRAINT:
DRESP1 2 L1 FREQ 1
DCONSTR 2 2 9.55
$
$...TRANSIENT DISPLACEMENT CONSTRAINTS:
DRESP1 3 UX2 TDISP 1 rss 2
DRESP2 30 RMS 30
DRESP2 3
DEQATN 30 RMS(RSS) = RSS / 51.
DCONSTR 3 30 0.10
$
$...TRANSIENT BENDING STRESS CONSTRAINTS:
$......VERTICAL COLUMN, STRESSES AT ENDS A AND B:
DRESP1 5 M1A1 TFORC ELEM 2 rss 1
DRESP1 6 M1B1 TFORC ELEM 4 rss 1
DRESP2 10 SIGA1 210
DESVAR 1
DRESP2 5
DRESP2 11 SIGB1 210
DESVAR 1
DRESP2 6
$....HORIZONTAL BEAM, STRESS AT END A
DRESP1 7 M2A1 TFORC PBAR 2 rss 102
DRESP2 12 SIGA2 210
DESVAR 2
DRESP2 7
DEQATN 210 BSIG(X,M) = M/( SQRT(72720. *X + 84000.) - 290. )/ 51.
dscreen equa -10.
dscreen disp -10.
DCONSTR 3 10 +3.E3
DCONSTR 3 11 +3.E3
Main Index Main Index


500
DCONSTR 3 12 +3.E3
$
DOPTPRM DESMAX 20IPRINT7p1 1 p2 15
param post -1
ENDDATA
The initial values of the two DESVARs are set to 1.0 with the beam bending inertias
being set to 1200. times the design variable value. The inertia values are linear in the
design variable and can therefore be input on a DVPREL1 entry. The area, however,
is nonlinear in the design variable as indicated by . This requires DVPREL2 entries.
The responses are standard, but the RMS responses can perhaps benefit from a
description. DRESP1 3 selects a TDISP (transient displacement) response at GRID 2,
component 1. The RSS in the ATTB field for this entry indicates that the responses that
are computed from this DRESP1 are to be converted into a RSS value. The output of
this process is scaled by 51., reflecting the 51 time steps that went into the RSS
calculation. Note that the DRESP2 that is used to provide this RMS responses invokes
the RSS values using a DRESP2 label on its continuation even though the RSS value
was created using a DRESP1. This is a special rule for type-1 responses that use the
character inputs in the ATTB field: subsequent references to the response must
identify it as a DRESP2. The limit of 0.1 in. on the RMS displacement is imposed as an
upper bound on the DRESP2 value. Three DRESP1 entries create RSS values for the
bending moments at both ends of the one of the vertical beams and at the left end of
the horizontal beam (symmetry considerations make it unnecessary to impose
constraints on the left side of the structure). This value is converted into an RMS stress
value by applying the relationship of and again dividing by 51.0.
Selected output is shown for this job in Listing 7-31. Attention is directed to the
fragment that prints the final response values. The RSS response type responses create
some special results that are explained here. The output fragment begins with prints
of the final response values that are produced by the DRESP1 entries. Data are shown
for only two of the time steps, but the actual output file contains these data for all 51
time steps. This is followed by a print of the RETAINED DRESP2 RESPONSES. It is
seen that the first four DRESP2 responses have the same response label as the DRESP1
values and that the Equation ID for all of them is RSS. These responses are the RSS
values that are computed from the DRESP1 transient responses. This is followed by
the final values for four actual DRESP2 responses that are constrained in the design
task. It is seen that two of the stress responses and the displacement response are very
close to their limit values.
Main Index Main Index
501 CHAPTER 7
Example Problems
The design history show that a converged design is obtained in 18 cycles. The initial
weight is 7835. pounds (this includes 5400. pounds of non-structural mass) while the
final weight is 7513. The design variable history indicates that the bending inertias of
the vertical bars have been reduced by about a third while the horizontal bar is slightly
reduced. It is also seen that the design fluctuates significantly and this is felt to be
representative of problems of this type.
Listing 7-31 Selected Output for the Transient Dynamic Optimization
of the Portal Frame (DSOUG12)
----- MODAL TRANSIENT FORCE RESPONSES -----
-------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT COMPONENT LOWER UPPER
ID ID LABEL ID NO. TIME BOUND VALUE BOUND
-------------------------------------------------------------------------------------------------------------
202 5 M1A1 1 2 2.4500E-01 N/A -1.0052E+06 N/A
203 6 M1B1 1 4 2.4500E-01 N/A 7.8041E+05 N/A
204 7 M2A1 3 2 2.4500E-01 N/A 7.8041E+05 N/A
----- MODAL TRANSIENT FORCE RESPONSES -----
-------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT COMPONENT LOWER UPPER
ID ID LABEL ID NO. TIME BOUND VALUE BOUND
-------------------------------------------------------------------------------------------------------------
206 5 M1A1 1 2 2.5000E-01 N/A -2.5050E+05 N/A
207 6 M1B1 1 4 2.5000E-01 N/A 1.9598E+05 N/A
208 7 M2A1 3 2 2.5000E-01 N/A 1.9598E+05 N/A

---- RETAINED DRESP2 RESPONSES ----
----------------------------------------------------------------------------------------------------------
INTERNAL DRESP2 RESPONSE EQUATION LOWER UPPER
ID ID LABEL ID BOUND VALUE BOUND
----------------------------------------------------------------------------------------------------------
1 1000 SWEIG 1000 N/A 7.6037E+01 N/A
2 3 UX2 RSS N/A 4.6063E+00 N/A
3 5 M1A1 RSS N/A 9.9974E+06 N/A
4 6 M1B1 RSS N/A 8.5369E+06 N/A
5 7 M2A1 RSS N/A 1.5054E+07 N/A
6 30 RMS 30 -1.0000E+20 9.0320E-02 1.0000E-01
7 10 SIGA1 210 -1.0000E+20 2.3520E+03 3.0000E+03
8 11 SIGB1 210 -1.0000E+20 2.0084E+03 3.0000E+03
9 12 SIGA2 210 -1.0000E+20 2.9452E+03 3.0000E+03
Main Index Main Index


502
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 13
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 12
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 7.835541E+01 -2.878003E-02
1 7.713461E+01 7.713528E+01 -8.704015E-06 -1.274490E-01
2 7.551521E+01 7.551541E+01 -2.626805E-06 1.188135E-02
3 7.558096E+01 7.558089E+01 8.075474E-07 -5.690023E-03
4 7.538185E+01 7.538213E+01 -3.643545E-06 -1.888045E-02
5 7.502102E+01 7.502102E+01 0.000000E+00 3.817041E-02
6 7.554443E+01 7.554432E+01 1.413892E-06 -2.673690E-02
7 7.504337E+01 7.504343E+01 -8.133312E-07 4.212101E-02
8 7.558733E+01 7.558722E+01 1.413090E-06 1.004394E-03
9 7.557368E+01 7.557362E+01 7.066720E-07 3.501766E-02
10 7.586059E+01 7.586061E+01 -3.017137E-07 1.459245E-02
11 7.605100E+01 7.605096E+01 5.015975E-07 -2.337809E-03
12 7.603665E+01 7.603665E+01 0.000000E+00 -1.826733E-02
---------------------------------------------------------------------------------------------------------------
1 TRANSIENT DYNAMIC OPTIMIZATION - DSOUG12 JUNE 11, 2003 MSC.NASTRAN 6/10/03 PAGE 448

0 SUBCASE 3
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | X1 | 1.0000E+00 : 8.0000E-01 : 7.2362E-01 : 6.6304E-01 : 6.9027E-01 : 6.8597E-01 :
2 | 2 | X2 | 1.0000E+00 : 1.1254E+00 : 9.0033E-01 : 1.0603E+00 : 9.4500E-01 : 8.7015E-01 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 6 : 7 : 8 : 9 : 10 : 11 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | X1 | 7.0256E-01 : 6.9055E-01 : 7.2474E-01 : 7.4406E-01 : 7.6334E-01 : 7.7360E-01 :
2 | 2 | X2 | 9.5525E-01 : 8.6500E-01 : 9.1469E-01 : 8.6896E-01 : 8.9363E-01 : 9.1597E-01 :
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | 12 : 13 : 14 : 15 : 16 : 17 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | X1 | 7.6164E-01 :
2 | 2 | X2 | 9.3886E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 12.
Main Index Main Index
503 CHAPTER 7
Example Problems
Figure 7-32 displays initial and final time histories for the horizontal displacement at
grid 2.
Figure 7-32 Initial and Final Transient Response
Main Index Main Index


504
7.13 External Response to Include Alternative Buckling
Response
This example demonstrates how the DRESP3 Bulk Data entry can be applied to
include a buckling criteria that is not available directly from MSC.Nastran. The
application is to a bar loaded longitudinally as shown in Figure 7-33. This particular
design example is so simple that it could be solved using hand calculations. However,
the contribution of this example is not in the complexity of the problem solved but in
its illustration of the process you must follow to apply the external response
capability.
Figure 7-33 Pinned-Pinned Column Buckling
The design task is to find the area of the rod that minimizes the weight while satisfying
constraints on the longitudinal stress and on the buckling behavior. The stress
constraint is
while the critical buckling stress is computed using a combination of Johnson and
Euler buckling criteria, with the selection of the criterion to use based on the
slenderness ratio of the rod (see Peery and Azar in Reference 7.).
Eq. 7-17
Eq. 7-18
P P
L L 40'' = =
r 2 =
P 200,000 =

co
+80,000 =
L
80,000 100,000

cr
1

co
L ( )
2
4
2
E
------------------------------- = L
2E

co
---------

cr

2
E
L ( )
2
---------------------- = L
2E

co
---------
Main Index Main Index
505 CHAPTER 7
Example Problems
where:
The fact that the failure criteria can branch based on a combination of structural
properties and responses makes it an ideal application for the external response
capability. Listing 7-32 shows the input file for this example. It is seen that there are
two subcases with the first subcase performing a statics analysis and imposing
constraints on the stress in the element and on the combined buckling failure criteria.
The second subcase performs a buckling analysis and places a lower bound of 1.0 on
the first buckling eigenvalue. This is redundant with the Euler buckling condition of
the first subcase and therefore provides a check on the external response calculation.
Listing 7-32 Input Data File for the DRESP3 Example (DSOUG13)
CONNECT DRESP3 TESTGRP EXTRESP
DIAG 8,15 $
SOL 200
CEND
TITLE = BUCKLING TEST CASE WITH AN EXTERNAL RESPONSE - DSOUG13
SUBTITLE = JOHNSON/EULER BUCKLING CASE
ECHO = SORT
SPC = 100
DESOBJ = 20
SUBCASE 1
DESSUB = 1
LABEL = LOAD CONDITION 1
LOAD = 300
ANALYSIS = STATICS
DISP = ALL
STRESS = ALL
SUBCASE 2
DESSUB = 2
ANALYSIS = BUCK
METHOD = 1
DISP = ALL
LABEL = BUCKLING FACTORS
$
BEGIN BULK
PARAM,POST,-1
$------------------------------------------------------------------------
$ ANALYSIS MODEL
= critical buckling stress
= effective length
= radius of gyration
= empirically determined column yield stress

cr
L

co
Main Index Main Index


506
$------------------------------------------------------------------------
$
$ GRID DATA
$ 2 3 4 5 6 7 8 9 10
GRDSET 345
GRID 1 0.0 0.0 0.0
GRID 2 8.0 0.0 0.0
GRID 3 16.0 0.0 0.0
GRID 4 24.0 0.0 0.0
GRID 5 32.0 0.0 0.0
GRID 6 40.0 0.0 0.0
GRID 10 0.0 0.0 100.0 123456
$
$ ELEMENT AND MATERIAL DATA
CBAR 1 10 1 2 10
CBAR 2 10 2 3 10
CBAR 3 10 3 4 10
CBAR 4 10 4 5 10
CBAR 5 10 5 6 10
MAT1 1 3.0E7 0.33 0.1
$ PROPERTY DATA
PBARL 10 1 MSCBML0ROD
1.0
$ BOUNDARY CONDITION DATA
SPC1 100 1 1
SPC1 100 2 1 6
$ EXTERNAL LOADS DATA
FORCE 300 6 -2.0E5 1.0
$ BUCKLING ANALYSIS DATA
EIGRL 1 .05 4
$
$------------------------------------------------------------------------
$ DESIGN MODEL
$------------------------------------------------------------------------
$
DESVAR 1 RG1.00.0110.0
DOPTPRM P215 DESMAX 20 DELP 0.5 GMAX 0.01
CONVDV 0.01 CONVPR 0.02P11
DRESP1 20 W WEIGHT
DRESP1 23 S1 STRESS PBAR 7 10
DRESP1 24 S1 STRESS PBAR 8 10
DRESP1 25 S1 STRESS PBAR 6 10
DRESP332 JOHNSONTESTGRPEULJOH
DESVAR1
DTABLEL E SIGMAC
DRESP125
$
DCONSTR1 23 100000.
DCONSTR1 24 -80000.
DCONSTR1 32 1.0
DTABLEE 30.0E6L 40.0 SIGMAC8.0E4
Main Index Main Index
507 CHAPTER 7
Example Problems
DVPREL110 PBARL 10 DIM1
1 1.0
$
$ DESIGN FOR BUCKLING EIGENVALUE
DRESP1 1 BUCK1 LAMA 1
DCONSTR2 1 1.0
PARAM DSNOKD1.0
ENDDATA
The listing shows the DRESP3 entry that provides the buckling response. It is seen that
the inputs include the design variable value, constants that define E, the rod length
and and the axial stress in the rod. TESTGRP is specified on the DRESP3 entry as
the external response GROUP and EULJOH as the TYPE. A CONNECT file
management statement specifies a DRESP3 application, TESTGRP as the group, and
EXTRESP as the external evaluator.
Two subroutines are modified to perform the server task. As explained in DRESP3
Bulk Data Entries on page 195, the entire contents of the / MSC2004/dr3srv/ directory
need to be copied to your local directory. The two subroutines that require
modification are members of that directory. The first is entitled R3SGRT and is shown
in Listing 7-33. R3SGRT is used to specify the types of responses that are available
from the server. This is used to make sure that the user input TYPE is supported by
the server. In this case, there is only one type of response supported and it is EULJOH.
The second subroutine is entitled R3SVALD (shown in Listing 7-35) and performs the
response evaluation based on the user input arguments. (Note that the D suffix on this
subroutine name is used when running on a short word; i.e., 32 bit, machine and that
the subroutine named R3SVALS should be modified when a long word machine is
used). In this case, there are four real arguments being passed and their order is given
on the DRESP3 entry. The output of the subroutine is the DR3VAL argument and is
calculated using the formulas given in Eq. 7-17 and Eq. 7-18.

co
Main Index Main Index


508
Listing 7-33 The R3SGRT Subroutine
Listing 7-34 The R3SVALD Subroutine
SUBROUTINE R3SVALD(GRPID,TYPNAM,
. NITEMS,ARGLIS,
. NSIZE, ARGVAL,
. NWRDA8,ARGCHR,
. DR3VAL,ERROR)
C ----------------------------------------------------------------------
C
C PURPOSE: COMPUTE THE EXTERNAL RESPONSE
SUBROUTINE R3SGRT(GRPID,TYPNAM,ERROR)
C ----------------------------------------------------------------------
C
C PURPOSE: VERIFY THE EXTERNAL RESPONSE TYPE
C
C GRPNAM: INPUT INTEGER - GROUP ID
C TYPNAM INPUT CHARACTER*8 - NAME OF EXTERNAL RESPONSE TYPE
C ERROR INPUT/OUTPUT INTEGER -ERROR CODE FOR THE CALL.
C > 0 = PROCESSING ERROR
C
C METHOD
C MATCH THE USER INPUT: TYPNAM WITH THE LIST OF AVAILABLE
C EXTERNAL RESPONSE TYPES. IF NO MATCH IS FOUND, SET ERROR CODE.
C
C CALLED BY
C R3CGRT
C
C NOTE:
C THE WRITER OF THIS ROUTINE IS RESPONSIBLE FOR SPECIFYING
C NTYPES AND R3TYPE.
C ----------------------------------------------------------------------
C
C VARIABLES PASSED IN
C
INTEGER GRPID, ERROR
CHARACTER*8 TYPNAM
C
C LOCAL VARIABLES
C
INTEGER NTYPES, BADTYP
PARAMETER(NTYPES=1)
CHARACTER*8 R3TYPE(NTYPES)
C
DATA BADTYP/7554/
DATA R3TYPE/'EULJOH '/
ERROR = 0
DO 100 ITYPE = 1, NTYPES
IF (TYPNAM .EQ. R3TYPE(ITYPE)) THEN
GOTO 200
END IF
100 CONTINUE
ERROR = BADTYP
200 CONTINUE
C
RETURN
END
Main Index Main Index
509 CHAPTER 7
Example Problems
C
C GRPID INPUT INTEGER - GROUP ID
C TYPNAM INPUT CHARACTER*8 - NAME OF EXTERNAL RESPONSE TYPE
C NITEMS INPUT INTEGER - DIMENSION OF ARRAY ARGLIS
C NSIZE INPUT INTEGER - DIMENSION OF ARRAY ARGVAL
C NWRDA8 INPUT INTEGER - DIMENSION OF CHARACTER ARRAY ARGCHR
C ARGLIS INPUT INTEGER - ARRAY OF NO. OF ITEMS FOR EACH ARGUMENT
C ARGVAL INPUT DOUBLE - ARRAY OF ALL ARGUMENT VALUES
C ARGCHR INPUT CHARACTER*8 - ARRAY OF CHARACTERS
C DR3VAL OUTPUT DOUBLE - VALUE OF THE EXTERNAL RESPONSE
C ERROR INPUT/OUTPUT INTEGER -ERROR CODE FOR THE CALL.
C 0 = PRINT ERROR MESSAGES
C 1 = DO NOT PRINT ERROR MESSAGES.
C
C METHOD
C A)SET UP VARIOUS PARAMETERS FROM THE ARGUMENT LIST
C B)EVALUATE THE EXTERNAL RESPONSE BASED ON THE GIVEN TYPNAM
C C)RETURN BADTYP ERROR IF TYPNAM IS NOT MATCHED HERE.
C
C CALLED BY
C SENDR3SVALD
C ----------------------------------------------------------------------
C
C VARIABLES PASSED IN
C
CHARACTER*8 TYPNAM, ARGCHR(NWRDA8)
INTEGER GRPID, NITEMS, NSIZE, ARGLIS(NITEMS), ERROR, NWRDA8
DOUBLE PRECISION ARGVAL(NSIZE), DR3VAL
C
C
C LOCAL VARIABLES
C
INTEGER BADTYP
DOUBLE PRECISION PI, FAC, FACT, SLNDER
DOUBLE PRECISION R,L,E,SIGMA,SIGMAC, RGYRA
C
DATA BADTYP /7554/
C
C THE USER-SUPPLIED EQUATION TO DEFINE THE EXTERNAL RESPONSE
C SIGMA = DRESP1, R=DESVAR, L, E AND SIGMAC = DTABLE CONSTANTS
C
C EULER : EULER= -SIGMA * (L/ RGYRA ) **2 / (PI**2 * E)
C RGYRA = R / 2.0
C
C JOHNSON: JOHNSON = -SIGMA / (SIGMAC * FACTOR )
FACTOR = 1. - SIGMAC * (L/RGYRA)**2 /(4 * PI**2 * E)
ERROR = 0
C
C SET UP PARAMETERS FOR VARIOUS ARGUMENT ITEMS
C
PI = 3.14159
PI2 = PI * PI
IF (TYPNAM .EQ. 'EULJOH ') THEN
R = ARGVAL(1)
L = ARGVAL(2)
E = ARGVAL(3)
SIGMAC = ARGVAL(4)
SIGMA = ARGVAL(5)
RGYRA = R / 2.0
SLNDER = L / RGYRA
FACT = PI * SQRT(2.0D0 * E / SIGMAC)
IF ( SLNDER .LE.FACT) THEN
C JOHNSON CRITERION
FAC = 1.0D0 - SIGMAC * (SLNDER) ** 2 /(4.0D0 * PI2 * E )
DR3VAL = -SIGMA / (SIGMAC * FAC)
ELSE
C EULER CRITERION
DR3VAL = -SIGMA * SLENDER**2 / (PI2 * E)
Main Index Main Index


510
END IF
ELSE
ERROR = BADTYP
END IF

RETURN
END
Once the two files have been modified, the next step is to build the server command
using the command:
msc2004 ./ dr3srv build
This will create the external response server with the name dr3serv.
A final file that is required is the one that indicates where the server program resides
and, in this case, is named evalconnect and has the form:
EXTRESP,-,/ usr/ ehj/ dsoug/ dr3srv/ dr3serv
where EXTRESP is the evaluator name given on the CONNECT file management
statement and the - connect option in this case indicates the server exists on any
network mounted computer and this is followed by the name of the file which
contains the server.
Now that all the files have been assembled, the job is submitted using the gmconn
option; for example:
nastran dsoug13 scr=yes gmconn=evalconnect
Selected results for this simple design task are presented in Listing 7-35. The printout
first gives results from the final analysis and it is seen that the single design variable
that the user has specified to be the radius of the ROD input on the PBARL has
generated nine designed properties for the bar cross section. The RESPONSE IN THE
DESIGN MODEL output indicates that the stress in the bars are well below their
compression allowable and that the final global buckling response is 1.0863. Since this
is above the 1.0 limit that has been placed on this response, we can infer that the
Johnson criterion is the critical design constraint. This is confirmed by the fact that all
of the retained DRESP3 responses are at their upper bound limit. The design history
summary indicates it was necessary to increase the rod radius from 1.0 to 1.1072 to
satisfy this design requirement and that this was accomplished in two design cycles.
Main Index Main Index
511 CHAPTER 7
Example Problems
Listing 7-35 Final Results in the Buckling Design Task (DSOUG13)
*******************************************
* *
* F I N A L A N A L Y S I S *
* *
*******************************************
******* ANALYSIS RESULTS BASED ON THE FINAL DESIGN *******
----- DESIGN OBJECTIVE -----
----------------------------------------------------------------------------------------------------
INTERNAL TYPE MINIMIZE
RESPONSE OF OR SUPERELEMENT SUBCASE
ID RESPONSE LABEL MAXIMIZE ID ID VALUE
----------------------------------------------------------------------------------------------------
1 DRESP1 WEIGHT MINIMIZE 0 0 1.5405E+01
----- DESIGN VARIABLES -----
---------------------------------------------------------------------------------------------------------
INTERNAL DESVAR LOWER UPPER
ID ID LABEL BOUND VALUE BOUND
---------------------------------------------------------------------------------------------------------
1 1 RG 1.0000E-02 1.1072E+00 1.0000E+01
----- DESIGNED PROPERTIES -----
----------------------------------------------------------------------------------------------------------
PROPERTY PROPERTY PROPERTY TYPE OF LOWER UPPER
TYPE ID NAME PROPERTY BOUND VALUE BOUND
----------------------------------------------------------------------------------------------------------
PBARL 10 DIM1 DVPREL1 1.0000E-15 1.1072E+00 1.0000E+20
PBAR 10 C1 DVPREL1 -1.0000E+35 1.1072E+00 1.0000E+20
PBAR 10 D2 DVPREL1 -1.0000E+35 1.1072E+00 1.0000E+20
PBAR 10 E1 DVPREL1 -1.0000E+35 -1.1072E+00 1.0000E+20
PBAR 10 F2 DVPREL1 -1.0000E+35 -1.1072E+00 1.0000E+20
PBAR 10 A SECPRO 1.0000E-15 3.8512E+00 1.0000E+20
PBAR 10 I1 SECPRO 1.0000E-15 1.1803E+00 1.0000E+20
PBAR 10 I2 SECPRO 1.0000E-15 1.1803E+00 1.0000E+20
PBAR 10 J SECPRO 1.0000E-15 2.3605E+00 1.0000E+20
----- DESIGN CONSTRAINTS ON RESPONSES -----
(MAXIMUM RESPONSE CONSTRAINTS MARKED WITH **)
---------------------------------------------------------------------------------------------------------
INTERNAL INTERNAL
INTERNAL DCONSTR RESPONSE RESPONSE L/U REGION SUBCASE
ID ID ID TYPE FLAG ID ID VALUE
---------------------------------------------------------------------------------------------------------
1 1 2 STRESS LOWER 1 1 -3.5085E-01
2 1 3 STRESS LOWER 1 1 -3.5085E-01
3 1 4 STRESS LOWER 1 1 -3.5085E-01
4 1 5 STRESS LOWER 1 1 -3.5085E-01
5 1 6 STRESS LOWER 1 1 -3.5085E-01
6 1 1 EXTERNAL UPPER 32 1 2.7899E-03**
7 1 2 EXTERNAL UPPER 32 1 2.7899E-03**
8 1 3 EXTERNAL UPPER 32 1 2.7899E-03**
9 1 4 EXTERNAL UPPER 32 1 2.7899E-03**
10 1 5 EXTERNAL UPPER 32 1 2.7899E-03**
11 2 12 LAMA LOWER 0 2 -8.6330E-02
Main Index Main Index


512
----------------------------------------------------------------------
| R E S P O N S E S IN D E S I G N M O D E L |
----------------------------------------------------------------------
(N/A - BOUND NOT ACTIVE OR AVAILABLE)
----- WEIGHT RESPONSE -----
------------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ROW COLUMN LOWER UPPER
ID ID LABEL ID ID BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------
1 20 W 3 3 N/A 1.5405E+01 N/A
F I N A L A N A L Y S I S S U B C A S E = 1
----- STRESS RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE ELEMENT VIEW COMPONENT LOWER UPPER
ID ID LABEL ID ELM ID NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
2 24 S1 1 8 -8.0000E+04 -5.1932E+04 N/A
3 24 S1 2 8 -8.0000E+04 -5.1932E+04 N/A
4 24 S1 3 8 -8.0000E+04 -5.1932E+04 N/A
5 24 S1 4 8 -8.0000E+04 -5.1932E+04 N/A
6 24 S1 5 8 -8.0000E+04 -5.1932E+04 N/A
7 25 S1 1 6 N/A -5.1932E+04 N/A
8 25 S1 2 6 N/A -5.1932E+04 N/A
9 25 S1 3 6 N/A -5.1932E+04 N/A
10 25 S1 4 6 N/A -5.1932E+04 N/A
11 25 S1 5 6 N/A -5.1932E+04 N/A
F I N A L A N A L Y S I S S U B C A S E = 2
----- BUCKLING LOAD RESPONSES -----
-----------------------------------------------------------------------------------------------------------
INTERNAL DRESP1 RESPONSE MODE LOWER UPPER
ID ID LABEL NO. BOUND VALUE BOUND
-----------------------------------------------------------------------------------------------------------
12 1 BUCK1 1 1.0000E+00 1.0863E+00 N/A
---- RETAINED DRESP3 RESPONSES ----
------------------------------------------------------------------------------------------------------------------
INTERNAL DRESP3 RESPONSE GROUP TYPE LOWER UPPER
ID ID LABEL NAME NAME BOUND VALUE BOUND
------------------------------------------------------------------------------------------------------------------
1 32 JOHNSON TESTGRP EULJOH N/A 1.0028E+00 1.0000E+00
2 32 JOHNSON TESTGRP EULJOH N/A 1.0028E+00 1.0000E+00
3 32 JOHNSON TESTGRP EULJOH N/A 1.0028E+00 1.0000E+00
4 32 JOHNSON TESTGRP EULJOH N/A 1.0028E+00 1.0000E+00
5 32 JOHNSON TESTGRP EULJOH N/A 1.0028E+00 1.0000E+00
0 SUBCASE 2
***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
(SOFT CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 3
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 2
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 1.256637E+01 4.017631E-01
1 1.539844E+01 1.539775E+01 4.471774E-05 3.486395E-03
2 1.540468E+01 1.540468E+01 1.238162E-07 2.789855E-03
---------------------------------------------------------------------------------------------------------------
0 SUBCASE 2
DESIGN VARIABLE HISTORY
----------------------------------------------------------------------------------------------------------------------------------
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
----------------------------------------------------------------------------------------------------------------------------------
1 | 1 | RG | 1.0000E+00 : 1.1069E+00 : 1.1072E+00 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 2.
Main Index Main Index
513 CHAPTER 7
Example Problems
7.14 Topology Optimization
This section provides a series of topology optimization examples to illustrate the
concept of topology optimization in general and then manufacturability constraints in
general. Two different simple models are used for these illustrations. The first is
referred to in the topology optimization literature as the MBB beam and is shown in
Figure 7-34 in its undesigned form:
Figure 7-34 The MBB Beam Before Optimization
The figure shows half of the total structure that is cantilevered at its ends and has a
transverse load applied at the bottom center location. The design task is to minimize
structural compliance (displacement times applied load) with a mass target 0.5 (i.e.,
remove 50% of the material). The structure is modeled with 4800 CQUAD4 elements
and the goal is to find the 2400 elements that create the minimum deflection under the
applied loading. Listing 7-36 contains the fragment of TPL file t opexe3.dat that is
related to topology optimization. It is seen that the DESOBJ entry invokes a DRESP1
with RTYPE=COMP. This is the compliance response that is to be minimized. There
is only one constraint and it is invoked by the DESGLB response to place an upper
bound of the fractional mass of 0.5. The original mass is determined from the
thickness given on the PSHELL entry times the density of the material and the area of
the beam. It is seen that the initial TOPVAR value is 0.5, which creates a mass value
that just satisfies the constraint condition.
Listing 7-36 Input File for the MBB Beam with No Manufacturability Constraints
DESOBJ = 1
DESGLB = 1
SUBCASE 1
$ Subcase name : Default
SUBTITLE=Default
Main Index Main Index


514
SPC = 2
LOAD = 2
ANALYSIS = STATICS
BEGIN BULK
DCONSTR 1 2 .5
TOPVAR, 1 , Tshel, Pshell, .5, , , , 1
DRESP1 1 COMPL COMP
DRESP1 2 FRMASS FRMASS
Figure 7-35 shows the topology optimized result that has been smoothed and
remeshed using MSC.Patran 2005 r2. The result is plausible in that it appears to select
a design that is a number of trusses strategically located to best transfer the load to the
boundary. From a practical standpoint, this design is of limited value because it
contains a number of small members that complicate the production of the suggested
design.
Figure 7-35 The MBB Beam After Optimization but Without Manufacturability
Constraints
In order to overcome the small member problem, the example is rerun with a
minimum member size constraint. The input data for this example related to topology
optimization with minimum member size is given in Listing 7-37. The minimum
member size value is defined by the parameter TDMIN = 0.5 on the DOPTPRM entry.
The half model is six units in width and two units in height and the TDMIN value is
in these units. The input data file is t opexe3a.dat
Listing 7-37 Input File for the MBB Beam with Minimum Member Size
Constraints
DESOBJ = 1
DESGLB = 1
SUBCASE 1
$ Subcase name : Default
SUBTITLE=Default
SPC = 2
LOAD = 2
ANALYSIS = STATICS
BEGIN BULK
DOPTPRM, TDMIN, 0.5
DCONSTR 1 2 .5
TOPVAR, 1 , Tshel, Pshell, .5, , , , 1
Main Index Main Index
515 CHAPTER 7
Example Problems
DRESP1 1 COMPL COMP
DRESP1 2 FRMASS FRMASS
Figure 7-36 shows the topology optimized result with minimum member size
TDMIN=0.5. Compared the design shown in Figure 7-34, this design is obviously
much simpler and there are no tiny members.
Figure 7-36 The MBB Beam After Optimization with Manufacturability
Constraints
A final example for the MBB imposes symmetry constraints. The half model has its x
axis along the length of the beam, starting at the boundary and its y-axis in the vertical
direction, starting at the bottom surface. The use of a half model naturally imposes
symmetry about the x-centerline of the beam but loading applied on the bottom
surface creates topology optimized designs that are not symmetric about the y-
centerline. For the purposes of this example additional symmetries are imposed along
the quarter-span of the beam and as well as about the vertical center. Imposing these
symmetries in this case is perhaps not meaningful, but it does provide a simple
illustration of the technique.
To apply symmetric constraints on designed properties, it is necessary to create a
reference coordinate system using a rectangular coordinate system. The relevant
input data for this example is given in Listing 7-38 and is extracted from TPL file
t opexe3b.dat . In this example, grid 10001 (location x=3, y=1, and z=0) is defined as the
origin. Grid 10002 (x=3, y=1, and z=1) lies on the z-axis, and grid 1003 (x=4, y=1, and
z=0) lies in the x-z plane. CORD1R CID=1 defines a reference coordinate system. A
continuation line SYM enforces the property PSHELL=1 to be symmetric about the
planes YZ and ZX in the reference coordinate system CID=1. In addition, a minimum
member size TDMIN=0.15 is applied.
Main Index Main Index


516
Listing 7-38 Input File for the MBB Beam with Minimum Member Size and
Symmetry Constraints
DESOBJ = 1
DESGLB = 1
SUBCASE 1
$ Subcase name : Default
SUBTITLE=Default
SPC = 2
LOAD = 2
ANALYSIS = STATICS
BEGIN BULK
CORD1R 1 10001 10002 10003
GRID 10001 3. 1. 0.0
GRID 10002 3. 1. 1.0
GRID 10003 4. 1. 0.0
TOPVAR, 1 , Tshel, Pshell, , , , , 1
, SYM , 1 , YZ , ZX
, TDMIN, 0.15
DRESP1 1 COMPL COMP
DRESP1 2 FRMASS FRMASS
DCONSTR 1 2 .5
Listing 7-37 shows the topology optimal result with symmetric constraints and
minimum member size.
Figure 7-37 The MBB Beam After Optimization with Manufacturability and
Symmetry Constraints
The MBB example is a two-dimensional structure, but topology optimization is
frequently imposed on a three-dimensional structure made up of solid elements.
Here, manufacturing constraints become particularly important to enable the
manufacture of the final design. A series of examples is given here that all start from
a solid cantilevered 3-D beam shown in Figure 7-38 load at its tip by a pair of twisting
forces. 2048 CHEXA elements are used for this model. The objective is to minimize the
structural compliance with mass target of 0.3 (i.e., remove 70% material).
Main Index Main Index
517 CHAPTER 7
Example Problems
Figure 7-38 The Torsional Beam Before Optimization
The extrusion constraints are used to enforce a constant cross-section design along the
given extrusion direction. In this first example with the Torsional Beam, an extrusion
constraint is imposed along the z-axis. Listing 7-39 shows input data of TPL file
topexe5.dat related to imposing this constraint on this model.
Listing 7-39 Input File for the Torsional Beam with Extrusion Constraints
DESOBJ = 1
DESGLB = 1
SUBCASE 1
$ Subcase name : Default
SUBTITLE=Default
ANALYSIS = STATICS
SPC = 2
LOAD = 2
$ Direct Text Input for this Subcase
BEGIN BULK
DRESP1 2 Frmass FRMASS
DRESP1 1 COMPL COMP
DCONSTR 1 2 .3
TOPVAR, 1 , TSOLID, PSOLID, .3, , , , 1
, EXT , , Z
PSOLID 1 1 0
Figure 7-39 shows the topology optimized result with extrusion constraints. It is
obvious that the design has a constant cross-section along the z-axis.
Main Index Main Index


518
Figure 7-39 The Torsional Beam with Extrusion Constraints along the Z-Axis
A second example with the torsional beam is used to demonstrate the combination of
one die casting manufacturability constraints and symmetry constraints. Listing 7-40
shows input data of TPL file topexe5a.dat related to imposing casting and symmetry
constraints on this model. The casting constraint with one die option enforces the
requirement that material can only be added to the region by filling up in the given
draw direction from the bottom (or, stated another way, the voids extend from the top
surface and do not reappear in the die direction). To apply casting constraints and
symmetric constraints on designed properties, a reference rectangular coordinate
system CID=1 is defined. A CAST continuation line defines casting constraints in
the Y direction and one die is a default option. The SYM continuation line defines
symmetric constraints about the YZ plane.
Listing 7-40 Input File for the Torsional Beam with Casting and Symmetry
Constraints
DESOBJ = 1
DESGLB = 1
SUBCASE 1
$ Subcase name : Default
SUBTITLE=Default
ANALYSIS = STATICS
SPC = 2
LOAD = 2
$ Direct Text Input for this Subcase
BEGIN BULK
DRESP1 2 Frmass FRMASS
DRESP1 1 COMPL COMP
DCONSTR 1 2 .3
CORD1R 1 5 167 7
PSOLID 1 1 0
Main Index Main Index
519 CHAPTER 7
Example Problems
TOPVAR, 1 , TSOLID, PSOLID, .3, , , , 1
, CAST, 1 , Y
, SYM, 1 , YZ
PSOLID 1 1 0
Figure 7-40 shows the topology optimized result with one die casting constraints. It is
observed that the design material is added by filling up in the Y direction from the
bottom. In addition, the design is symmetric about the YZ plane in the reference
coordinate system CID=1.
Figure 7-40 The Torsional Beam with a Casting Constraint along the y-axis and
a symmetry constraint about the YZ plane.
A final torsional beam example demonstrates two die casting manufacturability
constraints. The input for two die casting constraints is similar to the one die option
just shown. The difference is that a 2 is selected for the DIE field on the TOPVAR entry.
The input data related to imposing two die casting constraints is given in Listing 7-41
and this is extracted from TPL file topexe5b.dat.
Listing 7-41 Input File for the Torsional Beam with Two Casting and One
Symmetry Constraints
DESOBJ = 1
DESGLB = 1
SUBCASE 1
$ Subcase name : Default
SUBTITLE=Default
ANALYSIS = STATICS
SPC = 2
LOAD = 2
Main Index Main Index


520
$ Direct Text Input for this Subcase
BEGIN BULK
DRESP1 2 Frmass FRMASS
DRESP1 1 COMPL COMP
DCONSTR 1 2 .3
CORD1R 1 5 167 7
PSOLID 1 1 0
TOPVAR, 1 , TSOLID, PSOLID, .3 , , , , 1
, CAST, 1 , Y, 2
, SYM , 1 , YZ
PSOLID 1 1 0
Figure 7-41 shows the topology optimized result with two die casting constraints. It is
observed that the design material grows from the splitting plane in opposite
directions along the y-axis specified in the reference coordinate system CID=1. The
splitting plane is determined by optimization.
Figure 7-41 The Torsional Beam with a Two Casting Constraints along and a
Symmetry constraint about the Y-Axis.
Main Index Main Index
521 CHAPTER 7
Example Problems
7.15 Continuing the Design Process in a Subsequent Job
This final example section provides some guidance on continuing a design task across
multiple job submittals. The preferred situation would be that design information
gathered on an initial run be available in a subsequent run with a minimal amount of
recalculation. This is related to the powerful feature in MSC.Nastran referred to as
automatic restarts (see Automatic Restarts, (see RESTART on page 102 of the
MSC.Nastran Quick Reference Guide, Chapter 2, File Management Statements).
However, design optimization has specialized techniques in this area that differ from
those available in the other Solution Sequences. These techniques require more user
intervention so that it is felt valuable to provide this section with design optimization
examples so that you can benefit from the available capabilities. At the same time,
there is limited support for the standard Restart capability of MSC.Nastran and this
support is also described by example. This section therefore focuses on three separate
scenarios:
1. Continuing a property optimization design task
2. Continuing a shape optimization design task
3. Restarting from a previous analysis
Continuing a Property Optimization Design Task
The basic principle to recognize when continuing an optimization task is that, in SOL
200, the design model data overrides the analysis model (shape optimization is a
special case, as discussed in the next subsection). This feature can be used to our
advantage to continue from a previously obtained design.
As a design cycle is performed, PARAMeters DESPCH and DESPCH1 can be used to
save critical design information to the .pch file that is created during the run. By
default, DESVAR and property results are written to the .pch file following the last
design cycle. It is a simple matter then to delete the existing DESVAR entries and
insert the new DESVAR entries from the punch file to continue the design process.
We can illustrate this process using the Three-Bar Truss example of DSOUG1.
Listing 7-1 on page 389 does not contain any punch parameters, with the result that
the .pch file produced by this run contains only final DESVAR and PROD properties
as shown in Listing 7-42. Note that, by default, this output is given using the double
field format so that adequate data precision is obtained. (Small field format can be
obtained by using negative integers for the DESPCH1 parameter.) The DESVAR
entries from the punch file are used in place of the original DESVAR entries to
Main Index Main Index


522
produce the data file shown in Listing 7-43. It is not necessary (but does not hurt) to
include the PROD entries since the DESVAR values will be used to override the
original properties. Some other changes have been made in the file of Listing 7-43: the
DLINK entry has been removed so that all three rod areas are now being designed and
the DOPTPRM entry has been changed to that the SQP algorithm (METHOD=3) is
used for optimization and a direct approximation (APRCOD=1) is used for the
approximate model calculations.
Listing 7-42 The Punch File Produced by DSOUG1 in Section 7.1
Listing 7-43 Three Bar Truss Updated with File from Listing 7-42
ID MSC DSOUG14 $
TIME 10 $
SOL 200 $ OPTIMIZATION
CEND
TITLE = SYMMETRIC THREE BAR TRUSS DESIGN OPTIMIZATION - DSOUG14
SUBTITLE = CONTINUATION - 3 CROSS SECTIONAL AREAS AS DESIGN VARIABLES
ECHO = SORT
SPC = 100
DISP = ALL
STRESS = ALL
DESOBJ(MIN) = 20 $ (DESIGN OBJECTIVE = DRESP ID)
DESSUB = 21 $ DEFINE CONSTRAINT SET FOR BOTH SUBCASES
ANALYSIS = STATICS
SUBCASE 1
LABEL = LOAD CONDITION 1
LOAD = 300
SUBCASE 2
$
$ *************************************************************
$ * *
$ * CONTINUOUS DESIGN CYCLE NUMBER = 6 *
$ * *
$ *************************************************************
$
$
$ UPDATED DESIGN MODEL DATA ENTRIES
$
DESVAR * 1A1 8.36519659E-01 1.00000001E-01+D 1V
*D 1V 1.00000000E+02 1.00000000E+00
DESVAR * 2A2 3.29410642E-01 1.00000001E-01+D 2V
*D 2V 1.00000000E+02 1.00000000E+00
DESVAR * 3A3 8.36519659E-01 1.00000001E-01+D 3V
*D 3V 1.00000000E+02 1.00000000E+00
$
$ UPDATED ANALYSIS MODEL DATA ENTRIES
$
PROD* 11 1 8.36519659E-01 0.00000000E+00*
* 0.00000000E+00 0.00000000E+00
PROD* 12 1 3.29410642E-01 0.00000000E+00*
* 0.00000000E+00 0.00000000E+00
PROD* 13 1 8.36519659E-01 0.00000000E+00*
* 0.00000000E+00 0.00000000E+00
Main Index Main Index
523 CHAPTER 7
Example Problems
LABEL = LOAD CONDITION 2
LOAD = 310
BEGIN BULK
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL
$------------------------------------------------------------------------
$
$ GRID DATA
$ 2 3 4 5 6 7 8 9 10
GRID 1 -10.0 0.0 0.0
GRID 2 0.0 0.0 0.0
GRID 3 10.0 0.0 0.0
GRID 4 0.0 -10.0 0.0
$ SUPPORT DATA
SPC1 100 123456 1THRU3
$ ELEMENT DATA
CROD 1 11 1 4
CROD 2 12 2 4
CROD 3 13 3 4
$ PROPERTY DATA
PROD 11 1 1.0
PROD 12 1 2.0
PROD 13 1 1.0
MAT1 1 1.0E+7 0.33 0.1
$ EXTERNAL LOADS DATA
FORCE 300 4 20000. 0.8 -0.6
FORCE 310 4 20000. -0.8 -0.6
$
$------------------------------------------------------------------------
$ DESIGN MODEL
$------------------------------------------------------------------------
$
$...DESIGN VARIABLE DEFINITION
$DESVAR ID LABEL XINIT XLB XUB DELXV(OPTIONAL)
$
$ UPDATED DESIGN MODEL DATA ENTRIES
$
DESVAR * 1A1 8.36519659E-01 1.00000001E-01+D 1V
*D 1V 1.00000000E+02 1.00000000E+00
DESVAR * 2A2 3.29410642E-01 1.00000001E-01+D 2V
*D 2V 1.00000000E+02 1.00000000E+00
DESVAR * 3A3 8.36519659E-01 1.00000001E-01+D 3V
*D 3V 1.00000000E+02 1.00000000E+00
$
$...DEFINITION OF DESIGN VARIABLE TO ANALYSIS MODEL PARAMETER RELATIONS
$DVPREL1 ID TYPE PID NAME PMIN PMAX C0 +
$+ DVID1 COEF1 DVID2 COEF2 ...
DVPREL1 10 PROD 11 A
1 1.0
DVPREL1 20PROD 12 A
2 1.0
DVPREL1 30 PROD 13 A
3 1.0
$
$...STRUCTURAL RESPONSE IDENTIFICATION
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1 +
$+ ATT2 ...
DRESP1 20 W WEIGHT
DRESP1 21 U4 DISP 12 4
DRESP1 23 S1 STRESS PROD 2 11
12 13
Main Index Main Index


524
$...CONSTRAINTS
$DCONSTR DCID RID LALLOW UALLOW
DCONSTR 21 21 -0.20 0.20
DCONSTR 21 23 -15000. 20000.
$
$...OPTIMIZATION CONTROL:
$
DOPTPRM DESMAX 10 METHOD3APRCOD1
$
$.......2.......3.......4.......5.......6.......7.......8.......9.......0
ENDDATA
The design cycle history for the new file is shown in Listing 7-44. It is seen that there
is a small change in the design in this case in terms of the objective function. There is
a larger change in the design variables and note that it took ten iterations to achieve
this small change, perhaps because of the less accurate approximations that are
obtained using APRCOD=1. It is reassuring to see that the first and third design
variables remain identical even though there is no longer an explicit linking between
the two variables.
Listing 7-44 Design Cycle History for the Continued Design Task

***************************************************************
S U M M A R Y O F D E S I G N C Y C L E H I S T O R Y
***************************************************************
(HARD CONVERGENCE ACHIEVED)
NUMBER OF FINITE ELEMENT ANALYSES COMPLETED 3
NUMBER OF OPTIMIZATIONS W.R.T. APPROXIMATE MODELS 2
OBJECTIVE AND MAXIMUM CONSTRAINT HISTORY
---------------------------------------------------------------------------------------------------------------
OBJECTIVE FROM OBJECTIVE FROM FRACTIONAL ERROR MAXIMUM VALUE
CYCLE APPROXIMATE EXACT OF OF
NUMBER OPTIMIZATION ANALYSIS APPROXIMATION CONSTRAINT
---------------------------------------------------------------------------------------------------------------
INITIAL 2.695446E+00 1.998340E-03
1 2.698603E+00 2.698606E+00 -1.148534E-06 9.824219E-04
2 2.698332E+00 2.698332E+00 -1.767155E-07 7.113281E-04
---------------------------------------------------------------------------------------------------------------
1 SYMMETRIC THREE BAR TRUSS DESIGN OPTIMIZATION - DSOUG14 DECEMBER 18, 2001 MSC.NASTRAN 4/ 9/01 PAGE 39
CONTINUATION - 3 CROSS SECTIONAL AREAS AS DESIGN VARIABLES
0
DESIGN VARIABLE HISTORY
-----------------------------------------------------------------------------------------------------------------------------
-----
INTERNAL | EXTERNAL | |
DV. ID. | DV. ID. | LABEL | INITIAL : 1 : 2 : 3 : 4 : 5 :
-----------------------------------------------------------------------------------------------------------------------------
-----
1 | 1 | A1 | 8.3652E-01 : 8.6093E-01 : 8.4220E-01 :
2 | 2 | A2 | 3.2941E-01 : 2.6353E-01 : 3.1623E-01 :
3 | 3 | A3 | 8.3652E-01 : 8.6093E-01 : 8.4220E-01 :
*** USER INFORMATION MESSAGE 6464 (DOM12E)
RUN TERMINATED DUE TO HARD CONVERGENCE TO AN OPTIMUM AT CYCLE NUMBER = 2.
Main Index Main Index
525 CHAPTER 7
Example Problems
Even though this example did not show a benefit from a subsequent submittal, it does
illustrate several reasons why this type of resubmittal could be beneficial. These
reasons and others are given in the following list:
1. Investigate the effect of optimization parameters.
2. Enable the ability to pause and examine the results, possibly adjusting the
optimization parameters. This is particularly useful in large problems where
it would be prudent to limit the number of design cycles until confidence is
gained that the task is proceeding satisfactorily.
3. Investigate the effects of linking options.
4. Focus in on an intermediate design that is of interest. (This requires setting
DESPCH=1 so that design results are captured after each design cycle).
5. Apply more analysis types and/ or subcases to the final design from the
initial design task.
6. Use the previously computed optimal design with modified design
requirements.
An alternative use of the .pch file is to take the final property values from this file and
create a data file that can be used for other types of analyses. For example, nonlinear
analyses that cannot be performed in SOL 200 could now be run in order to check this
behavior.
Continuing a Shape Optimization Design Task
The procedures are somewhat different for a shape optimization task since, in this
case, the DESVAR values do not cause an override of the grid locations. Recall from
Eq. 2-12 on page 50 that the design change is the product of the shape basis vector and
the change in the design variable values. If the technique of using the final DESVAR
values in a resubmittal were applied, there would be no change in the shape relative
to the initial design since the new design task knows nothing about the original
DESVAR values. For shape optimization then, it is necessary to input the final grid
locations for the new job submittal. Again, the .pch file can be used for this purpose
since the new GRID bulk data entries are provided in the this file. The DESPCH
parameter specifies the frequency of this output with the default punching the grid
bulk data entries for the final design. (There is a special case when the optimization
creates a mesh that provides distorted elements, making it impossible to proceed. In
this case, the punch file contains both the last good mesh and the final distorted mesh.)
Main Index Main Index


526
Restarting from a Previous Analysis
The second class of restarts is based on MSC.Nastran's automatic restart capability. In
design optimization, this works in conjunction with the predefined Solution 200 exit
points, OPTEXIT 1 through 6. Recall that setting this parameter in Bulk Data will
terminate the program at one of these six exit points.
Automatic restart for design optimization is somewhat limited since only the analysis
restart logic is in place. That is, only data associated with the analysis results are used
on restart, and all design optimization-related processes are repeated. For example,
restarts from an OPTEXIT point of 4 or greater will result in a repeat of the constraint
evaluation and sensitivity analysis phases. This recalculation occurs regardless of
whether the design model has changed or not. You are cautioned that it makes no
sense perform this type of restart when changing the design model because the
analysis results from the previous run and no longer valid.
To illustrate this procedure, suppose we wanted to perform a baseline analysis for the
structure of TPL file DSOUG1 (the first example in this chapter), compute and display
the design responses and then pause to examine these results before restarting. The
results of the finite element analysis are saved to the database by including SCR=NO
in the job submittal command.
The only difference between the original three-bar truss and the input of Listing 7-45
is the Bulk Data parameter OPTEXIT. In this case, OPTEXIT is set to 3 which instructs
the code to exit after design constraint evaluation and screening.
Listing 7-45 Three Bar Truss with OPTEXIT=3 (DSOUG1C)
$
ID MSC DSOUG1 $
TIME 10 $
SOL 200 $ OPTIMIZATION
CEND
TITLE = SYMMETRIC THREE BAR TRUSS DESIGN OPTIMIZATION - DSOUG1
SUBTITLE = BASELINE - 2 CROSS SECTIONAL AREAS AS DESIGN VARIABLES
ECHO = SORT
SPC = 100
DISP = ALL
STRESS = ALL
DESOBJ(MIN) = 20 $ (DESIGN OBJECTIVE = DRESP ID)
DESSUB = 21 $ DEFINE CONSTRAINT SET FOR BOTH SUBCASES
ANALYSIS = STATICS
SUBCASE 1
LABEL = LOAD CONDITION 1
LOAD = 300
SUBCASE 2
LABEL = LOAD CONDITION 2
LOAD = 310
BEGIN BULK
Main Index Main Index
527 CHAPTER 7
Example Problems
$
$------------------------------------------------------------------------
$ ANALYSIS MODEL
$------------------------------------------------------------------------
$
$ GRID DATA
$ 2 3 4 5 6 7 8 9 10
GRID 1 -10.0 0.0 0.0
GRID 2 0.0 0.0 0.0
GRID 3 10.0 0.0 0.0
GRID 4 0.0 -10.0 0.0
$ SUPPORT DATA
SPC1 100 123456 1THRU3
$ ELEMENT DATA
CROD 1 11 1 4
CROD 2 12 2 4
CROD 3 13 3 4
$ PROPERTY DATA
PROD 11 1 1.0
PROD 12 1 2.0
PROD 13 1 1.0
MAT1 1 1.0E+7 0.33 0.1
$ EXTERNAL LOADS DATA
FORCE 300 4 20000. 0.8 -0.6
FORCE 310 4 20000. -0.8 -0.6
$
$------------------------------------------------------------------------
$ DESIGN MODEL
$------------------------------------------------------------------------
$
$...DESIGN VARIABLE DEFINITION
$DESVAR ID LABEL XINIT XLB XUB DELXV(OPTIONAL)
DESVAR 1 A1 1.0 0.1 100.0
DESVAR 2 A2 2.0 0.1 100.0
DESVAR 3 A3 1.0 0.1 100.0
$
$...IMPOSE X3=X1 (LEADS TO A3=A1)
$DLINK ID DDVID CO CMULT IDV1 C1 IDV2 C2 +
$+ IDV3 C3 ...
DLINK 1 3 0.0 1.0 1 1.00
$
$...DEFINITION OF DESIGN VARIABLE TO ANALYSIS MODEL PARAMETER RELATIONS
$DVPREL1 ID TYPE PID NAME PMIN PMAX C0 +
$+ DVID1 COEF1 DVID2 COEF2 ...
DVPREL1 10 PROD 11 A
1 1.0
DVPREL1 20PROD 12 A
2 1.0
DVPREL1 30 PROD 13 A
3 1.0
$
$...STRUCTURAL RESPONSE IDENTIFICATION
$DRESP1 ID LABEL RTYPE PTYPE REGION ATTA ATTB ATT1 +
$+ ATT2 ...
DRESP1 20 W WEIGHT
DRESP1 21 U4 DISP 12 4
DRESP1 23 S1 STRESS PROD 2 11
12 13
$...CONSTRAINTS
$DCONSTR DCID RID LALLOW UALLOW
Main Index Main Index


528
DCONSTR 21 21 -0.20 0.20
DCONSTR 21 23 -15000. 20000.
$
$...OPTIMIZATION CONTROL:
$
DOPTPRM DESMAX 10 DELP 0.5 P1 1 P2 15
PARAM OPTEXIT 3
$
$.......2.......3.......4.......5.......6.......7.......8.......9.......0
ENDDATA
After the finite element and design results are examined and verified, the decision is
made to continue with optimization in the restart run. This file is shown in Listing 7-
46.
Listing 7-46 Restart File for Three Bar Truss (DSOUG1R)
RESTART, VERSION=1, KEEP
ID MSC DSOUG1 $
TIME 10 $
DIAG 8,15
SOL 200 $ OPTIMIZATION
CEND
TITLE = SYMMETRIC THREE BAR TRUSS DESIGN OPTIMIZATION - DSOUG1
SUBTITLE = BASELINE - 2 CROSS SECTIONAL AREAS AS DESIGN VARIABLES
ECHO = SORT
SPC = 100
DISP = ALL
STRESS = ALL
DESOBJ(MIN) = 20 $ (DESIGN OBJECTIVE = DRESP ID)
DESSUB = 21 $ DEFINE CONSTRAINT SET FOR BOTH SUBCASES
ANALYSIS = STATICS
SUBCASE 1
LABEL = LOAD CONDITION 1
LOAD = 300
SUBCASE 2
LABEL = LOAD CONDITION 2
LOAD = 310
BEGIN BULK
/ 28
ENDDATA
The RESTART command is required, and indicates that the first version on the
database is to be used (VERSION=1) and all versions are to be retained (KEEP). If we
always intend to use the last version on the database for restarts and delete the rest,
we can instead use,
RESTART, VERSION=LAST, NOKEEP
By specifying VERSION=1, KEEP, we always have the first version to fall back on.
The original database can be selected in one of two ways. At job submittal, the
database can be selected with a command such as:
Main Index Main Index
529 CHAPTER 7
Example Problems
nastran dsoug1r scr=n dbs=dsoug1c
alternatively, an assign statement can be included in the data file:
ASSIGN MASTER=dsoug1c.MASTER
In Listing 7-46, the ` / ,28 ' deletes the OPTEXIT parameter assignment from the Bulk
Data, which is entry number 28 in the sorted bulk data of the original run. It is usually
a good idea to request a listing of the sorted Bulk Data, especially if subsequent
restarts are anticipated. Sorted bulk data is the default, but is explicit with the Case
Control command, ECHO=SORT or ECHO=BOTH.
The optimization results from this restart are identical to those given in Listing 7-2 on
page 394 and are not repeated here. The contents of the .F04 file include statements
such as the following, which indicate the initial analysis was skipped.
10:50:54 0:03 237.0 0.0 3.1 0.0 PHASE1A 70 (S)SEMG BEGN
*56* SEMG NOT EXECUTED.
10:50:54 0:03 239.0 0.0 3.1 0.0 PHASE1A 134 (S)SELG BEGN
*56* SELG NOT EXECUTED.
Main Index Main Index


530
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
APPENDIX
A
Nomenclature, Glossary of Terms, and
References
I Nomenclature
I Glossary of Terms
I References
Main Index Main Index


532
A.1 Nomenclature
Vector dot product
x Vector cross product
Scalar multiplication
Absolute value of the quantity a
Matrix of damping terms in the finite element analysis
Objective function
Lower-bound constraint
Upper-bound constraint
Direct variable approximation for the j-th constraint
Reciprocal variable approximation for the j-th constraint
j-th constraint function
Approximation of the j-th constraint function
k-th equality constraint
Identity matrix
Stiffness matrix in the finite element analysis
Differential stiffness matrix
Mass matrix in the finite element analysis
j-th property value
Upper move limit on j-th property
j-th type-1 property (CDVPREL1)
j-th type-2 property (CDVPREL2)


a
B [ ]
F X ( )
g
L
g
U
g

j D
g

j R
g
j
X ( )
g

j
X ( )
h
k
X ( )
I [ ]
K [ ]
K
d
[ ]
M [ ]
p
j
p
j
U
p
j
1
p
j
2
Main Index Main Index
533 APPENDIX A
Nomenclature, Glossary of Terms, and References
Lower move limit on j-th property
Vector of initial property values
Load vector in the finite element analysis
j-th response value
Lower bound of the j-th structural response
Upper bound on the j-th structural response
j-th type-1 response (DRESP1)
j-th type-2 response (DRESP2)
j-th type-3 response (DRESP3)
Vector of analysis model responses (displacements, stresses, eigenvalue,
etc.)
Search vector in the optimizations one-dimensional search
Matrix of constant coefficients in reduced basis formulation where M >
N
Vector of displacements
Vector of intermediate variables used in formal approximations
i-th design variable
Lower-bound side constraint on i-th design variable
Upper-bound side constraint on i-th design variable
Vector of design variables
Initial value of vector of design variables (this characterizes the initial
design).
p
j
L
p
0
{ }
P { }
r
j
r
j
L
r
j
U
r
j
1
r
j
2
r
j
3
r { }
S
T [ ]
MXN
U { }
Y
x
i
x
i
L
x
i
U
X
X
o
Main Index Main Index


534
Vector of design variables at the optimal design
Transpose of a vector
Transpose of a matrix
One-dimensional search parameter
Optimal value of one-dimensional search parameter for the current
search direction
Partial derivatives with respect to the i-th design variable
Finite difference move, delta
Adjoint solution vector
Lagrange multiplier used in the Kuhn-Tucker conditions
Sensitivity coefficient for the i-th design variable and j-th response
Gradient operator
n-th eigenvector
Modal transformation matrix
Mass density
Frequency of oscillation of a dynamic system (rad/ s)
Vector of modal responses
X
*
{ }
T
[ ]
T

*
S

x
i
-------

x
i
------- ,

ij

n
{ }
[ ]

{ }
Main Index Main Index
535 APPENDIX A
Nomenclature, Glossary of Terms, and References
A.2 Glossary of Terms
Active Constraint is a constraint whose numerical value is near zero. In MSC.Nastran,
the constant CT (default = -0.03) is used to measure whether a constraint is
active or inactive. A constraint whose value is less than CT is deemed inactive,
and if its value is greater than CT, it is considered active. This information is
used at the optimizer level in connection with the numerical search process.
See Violated Constraints.
Adjoint Sensitivity Analysis performs sensitivity analysis by first determining an
adjoint load vector that is the change the response quantity due to a change in
the solution vector. This can be compared with the Direct Sensitivity Analysis
methods. The two methods give comparable sensitivity results but the efforts
required in their computation can differ dramatically.
Analysis Discipline is used in Solution 200 to refer to the available analysis types:
Statics, Normal Modes, Buckling, Direct and Modal Complex Eigenanalysis,
Direct and Modal Frequency, Modal Transient, Static Aeroelasticity and
Flutter.
Analysis Model defines the geometry, element connectivity, material properties, and
loads associated with an MSC.Nastran finite element analysis. The analysis
model may be overridden by the design model.
Approximate Model is an approximate representation of the design space that is used
by the optimizer to determine an approximate optimum. Since the
approximate model contains a reduced number of constraints and is explicit in
the design variables, it lends itself to efficient solution by a numerical
optimizer.
Approximate Optimization is used to refer to the optimization with respect to
approximate design spaces.
Approximation Concepts include design variable linking, constraint regionalization
and deletion, and formal approximations. All of these approximations are
available in MSC.Nastran. These tools allow numerical optimizers to be
coupled effectively with structural analysis codes where the responses are
usually implicit functions of the design variables. Solving the resultant
approximate subproblem yields an approximate optimum.
Main Index Main Index


536
Auxiliary Boundary Model is similar to an Auxiliary Model for generation of shape
basis vectors, though defined just on the boundary of the finite element model.
It may be, for example, an assembly of bar elements along the edge of a two-
dimensional structure, or a group of plate elements over the surface of a three-
dimensional model. It is used to generate the boundary portion of the shape
basis vector from which the motion of interior points, and the remaining
portion of the basis vector, is interpolated.
Auxiliary Model (or Auxiliary Structure) is a finite element model used to generate
grid variations for purposes of shape sensitivity or optimization. It usually
shares geometry and connectivity with the original (or primary) structure, but
with different loads, boundary conditions, and perhaps material types. The
resultant deformations are used to generate a basis vector for shape sensitivity
and optimization.
Basic Optimization Problem Statement is a formal definition of the optimization
task. This problem statement is invoked by the engineer when constructing the
design model and by the optimizer when searching for an optimal design. See
Eq. 1-1 through Eq. 1-5.
Basis Vector is a collection of constant coefficients used to relate changes in structural
properties to changes in design variables. They may be used in connection with
either property optimization or shape optimization. When used for shape
optimization, they are commonly referred to as Shape Basis Vectors.
Beam Library contains standard cross section definitions for relating beam
dimensions to beam properties. The MSC supplied section types can be
augmented by user defined section types and all of the members of the beam
library can be designed.
Beta Method is an optimization technique that minimizes an artificial design variable
to minimize a maximum response. See Acoustic Optimization on page 478.
Central Difference Sensitivities rely on finite difference perturbations to compute
approximate derivatives, or sensitivities, of continuous functions. Central
Difference Approximations, which perturb the design variables in both a
forward and backward direction, yield greatly improved derivative
approximations when compared to first order differencing schemes (forward
and backward differences are first order.)
Main Index Main Index
537 APPENDIX A
Nomenclature, Glossary of Terms, and References
Continuous Design Variables can assume any value in a specified range and are to
be distinguished from discrete design variables.
Constraint (Design Constraint) is defined as an inequality which must be satisfied in
order to indicate a feasible design. MSC.Nastran convention is that negative
constraint values are deemed satisfied while positive values are violated.
Constraint Deletion is the process which temporarily removes constraints for a given
design cycle. Once a constraint is removed, its sensitivities do not need to be
computed, and the optimizer does not need to consider the constraint during
the approximate optimization. Later design cycles may include constraints that
previously were deleted.
Constraint Regionalization refers to the grouping of constraints by type in
preparation for constraint deletion. Regionalization is done in conjunction with
constraint deletion to limit the number of constraints that need to be considered
in a particular region of the structure.
Constraint Screening is an approximation concept which seeks to reduce the number
of constraints in the optimization problem. This reduction is based on the
assumption that the entire constraint set is likely to contain redundant design
information. Both regionalization and deletion are used to group and screen
constraints for temporary deletion. Constraints which are temporarily deleted
may be retained on subsequent design cycles.
Convergence is the term to indicate that the design process has reached a final state
and can be terminated. This concept applies to both the Design Cycle and the
Design Iteration processes. The converged design can be feasible or infeasible.
CSV (Comma Separated Values) File is a specialized output from design
optimization which produces design optimization data in a comma separated
values format that can be read into a spreadsheet and used to produce
graphical displays of optimization results.
Dependent Design Variable is a design variable defined on a DESVAR entry which
is linearly dependent on one or more independent design variables. This
dependence is defined by DLINK Bulk Data entries.
Main Index Main Index


538
Design Cycle refers to the process which invokes the optimizer once for each design
cycle. Within a design cycle a finite element analysis is performed, an
approximate problem is created, the optimizer is invoked, and convergence is
tested. A number of design cycles may need to be completed before overall
convergence is achieved.
Design Iteration is the optimizer level process of determining a search direction from
available gradient information, and performing a one-dimensional search in
this direction. A number of these iterations may be performed before an
approximate optimum is found.
Design Model defines the design variables, objective, and constraints. In the design
model, design variables are used to express permissible analysis model
variations, and design responses are used as the objective and constraint
functions. The optimizer uses the information in the design model to propose
improved designs.
Design Optimization refers to the automated redesign process that attempts to
minimize (or maximize) a specific quantity subject to limits or constraints on
other responses. This process is often referred to as Structural Optimization in
this guide to indicate the application of these methods to structural redesign
tasks.
Designed Properties are those items that can be changed in the design optimization
process. Properties can be those found on MSC.Nastran property entries (such
as the thickness of a plate), those found on material entries (such as Youngs
modulus) and those found on connectivity entries (such as a beam offset).
Variations in shape are not considered Designed Properties.
Design Sensitivity Analysis is the process which is used to determine rates of change
of structural responses with respect to changes in design variables. The
resulting partial derivatives or design sensitivity coefficients can be
used directly to perform parametric design studies or to input to a numerical
optimizer for design optimization.
Design Sensitivity Coefficient (see Design Sensitivity Analysis).
r
j
x
i

Main Index Main Index

539 APPENDIX A
Nomenclature, Glossary of Terms, and References
Design Space is the n-dimensional region over which the objective and constraints are
defined, where n is the number of independent design variables. The range of
the design variables is limited by upper and lower bounds on the variables and
it is within this space that the optimizer searches for an optimal design while
simultaneously trying to satisfy all of the imposed constraints.
Design Variable Linking allows a design variable to be expressed as a linear function
of other, independent, design variables. This linking must be explicitly defined
by the engineer. It is one of the components in the family of approximation
concepts. Design variable linking can be used to enforce symmetry, ensure
equivalent element properties, and so on. It can also improve the convergence
characteristics of the problem by reducing the number of independent design
variables that must be considered by the optimizer.
Designed Coordinates are grid coordinates that are allowed to vary in shape
optimization. If a designed grid is to only move in the x-y plane, for example,
then it will have only two designed coordinates x and y. The z coordinate for
that grid will not be designed. More specifically, the designed coordinates
correspond to nonzero entries in the basis vector matrix for shape optimization.
Designed Grids are those grids whose locations vary in shape optimization. Grids on
a moving boundary would thus be designed, as would be grids on the interior
of the structure which are updated. Grids on a fixed boundary (one that does
not change during shape optimization) would not be Designed Grids. See also
Designed Coordinates.
Direct Input of Shapes refers to the method of DBLOCATEing displacement vectors
that have been generated using an external Auxiliary Model. Once input, they
may be used for shape sensitivity and optimization.
Direct Sensitivity Analysis performs sensitivity analysis by first determining a
pseudo load vector that is the sensitivity of the system matrix times solution
vector. This can be compared with the Adjoint Sensitivity Analysis methods.
The two methods give comparable sensitivity results but the efforts required in
their computation can differ dramatically.
Discrete Design Variables are design variables that can only take on one of the
specified discrete values over a specified range, as opposed to continuous
variables.
Main Index Main Index


540
External Responses in MSC.Nastran are those responses that are computed by an
external process and included with the MSC.Nastran optimization task either
as they objective or by placing constraints on the response values.
Feasible Design is a design which satisfies all of the constraints. A feasible design may
not be optimal, but an optimal design must be feasible.
Finite Differences are a numerical method for approximating derivatives. The
method derives from the Taylor series expansions of arbitrary functions. First-
forward finite differences (see Eq. 2-35) and central differences (see Eq. 2-36)
are frequently used in MSC.Nastrans semianalytic sensitivity analysis to
provide low-cost derivative approximations.
First-level Response is a response that is directly available from an MSC.Nastran
analysis. These responses are identified for use in the design model with
DRESP1 entries and is to be distinguished from second and third level
responses.
Flat Design Space is a term use to designate the situation where significant changes
in design variable values produce limited changes in the objective and
constraint values. In this situation, the designer has latitude in selecting the
final design parameters other than those produced by the computational
optimization process without incurring a significant penalty.
Formal Approximations are the Taylor series expansions of the implicit responses
used in connection with design optimization. The resultant explicit
approximations are used by the optimizer to find a corresponding approximate
optimum.
Forward (and Backward) Sensitivities perform finite difference sensitivity analysis
by perturbing the design variable in a single direction. These methods are
faster, but less accurate, than central differencing techniques.
Fully Stressed Design is a non-gradient based resizing algorithm that is used to resize
element thicknesses or areas so as to produce a design where each designed
property is subjected to its maximum allowable stress (or strain) under at least
one loading condition.
Main Index Main Index
541 APPENDIX A
Nomenclature, Glossary of Terms, and References
Global Optimum is the term used to designate the optimum design that is the best
among all positive relative minima. The design optimization process as
implemented in MSC.Nastran can lead to a relative, as opposed to, global
optimum. This can be investigated by starting the design task from different
points in the design space and comparing the optimal designs.
Gradient is defined as a vector of partial derivatives. See Eq. 1-10. Physically, the
gradient of a function points in the direction of most rapidly increasing
function values.
Gradient-based Numerical Optimizer is any optimizer that uses function gradient
information to search for an optimal design. The optimizer in MSC.Nastran is
of this class.
Grid Perturbation refers to a small change in a grid point coordinate for a
corresponding infinitesimal design variable change. These small variations are
used in differencing schemes for design sensitivity analysis.
Grid Variation refers to a finite change in a grid point coordinate for a corresponding
change in a design variable. This results in a finite change of structural shape.
Hard Convergence refers to the design cycle convergence test that compares the
results of two consecutive finite element analyses. The terminology is chosen to
indicate that the test is based on hard, or conclusive, evidence.
Independent Design Variable is a design variable defined on a DESVAR entry that is
not expressed as a function of any other design variable. All design variables
are independent unlike they are made dependent by a DLINK entry. In design
optimization, the optimizer varies the independent variables directly.
Infeasible Design is a design that violates one or more constraints.
Kuhn-Tucker Conditions provide the formal definition of an optimal design (see
Numerical Optimization in Appendix D). These conditions are implicitly
evaluated in MSC.Nastran in connection with convergence detection at the
optimizer level.
Load Case Deletion represents a special case of constraint deletion where none of the
constraints in a given subcase survive the constraint deletion process. In this
circumstance, the direct method of sensitivity analysis can eliminate the load
case from further consideration, thereby decreasing the amount of effort
required in the sensitivity analysis.
Main Index Main Index


542
Mode Tracking refers to the technique that is employed to allow the user to design a
particular physical mode shape (such as the first torsion mode) rather than a
specific mode number. The mode number for the physical mode can switch and
the mode tracking permits the design task to switch as well.
Move Limits serve to limit the region in which the optimizer may search during the
current approximate optimization. Since the approximate subproblem is only
valid in the region of the current design (see Move Limits on page 104),
move limits serve to restrict the search in this approximate design space to
avoid numerical ill-conditioning due to poor approximations. Move limits are
placed both on the design variables and on the designed properties.
Nonunique Optimal Design is an optimum design at which the objective function
and all active constraints are relatively insensitive to changes in the design
variables. See Figure 2-22 in Tests for Convergence on page 125, for a
qualitative explanation of the situation. This is also referred to as a Flat Design
Space.
Objective Function (Design Objective) is the function that the optimizer seeks to
minimize (or maximize). The objective function must be a continuous function
of the design variables.
One-Dimensional Search is the search process at the optimizer level in which all
changes to the vector of design variables are characterized by changes in a
single-scalar parameter, . See Search Direction, and Eq. 1-12.
Optimum Design is defined as a point in the design space for which the objective
function is minimized and the Kuhn-Tucker Conditions are satisfied. By
definition, the optimum design is feasible. It is possible that the optimum
design is a relative minima in the design space, other optimal designs can exist.
Primary Model (or Primary Structure) refers to the finite element model used for
analysis. The distinction arises when Auxiliary Models are used for shape
optimization. See the definitions for Auxiliary Models, and Auxiliary
Boundary Models.
Property Optimization defines a design task in which the properties describing the
analysis model are allowed to vary. Also frequently referred to as sizing
optimization, quantities such as plate thicknesses, bar element moments of
inertia, composite ply orientations, material properties and concentrated mass
values are modified in search of an improved design.

Main Index Main Index

543 APPENDIX A
Nomenclature, Glossary of Terms, and References
Pseudo-Load Vector refers to the right-hand side vector in the sensitivity analysis
equations. See Design Sensitivity Analysis in Chapter 2.
Reduced Basis Formulation uses a small number of design variables to describe
changes to a large number of analysis model properties or grid coordinates.
Essentially, each design variable acts as a multiplier of a (constant) vector. The
linear superposition of these so-called basis vectors represents the total design
changes. See Eq. 2-3.
Search Direction (Search Vector) an n-dimensional vector where n is the number of
independent design variables, that characterizes a particular search path used
by the optimizer. The optimizer may search along a number of different paths
in connection with a given design cycle. Each of these searches is termed a
design iteration. Upon completion of these iterations, an approximate
optimum is at hand. This approximate optimization process is repeated in
subsequent design cycles.
Second Level Responses, also known as synthetic responses, are formulated by the
engineer using MSC.Nastrans equation input capability. These responses are
defined using DRESP2 and DEQATN Bulk Data entries.
Semianalytic Sensitivity Analysis refers to an approximate solution of the analytic
sensitivity equations. With this approach, the appropriate system equations are
differentiated and the structural matrices approximated using finite
differences. The resulting approximate set of equations is then explicitly solved
for the response derivatives. See Design Sensitivity Analysis in Chapter 2,
to see how this approach is used for the various analysis disciplines.
Shape Basis Vector is a basis vector used in connection with shape sensitivity and
optimization. The vector is a collection of directions along which designed
grids can move. Each element of the vector represents the magnitude of change
in a grid coordinate due to a unit change in a design variable. See the definition
for Basis Vector.
Shape Design Variable is a design variable used to describe structural shape
variations. Theoretically, shape design variables act as multipliers of shape
basis vectors. Physically, the variables can represent, for example, a radius, a
coordinate location defining the center of a cutout or the percentage addition
of taper to a control arm.
Main Index Main Index


544
Shape Optimization defines a design task in which boundaries and connection point
locations are allowed to change in search of an improved design.
Shape Sensitivity computes the rate of change of structural responses with respect to
shape-defining design variables. The grid coordinate changes are expressed in
terms of design variable changes using the methods of Relating Design
Variables to Shape Changes in Chapter 3.
Side Constraint is an upper or a lower bound on a design variable. Since the optimizer
never searches outside of these bounds, these side constraints define the limits
of the design space.
Soft Convergence refers to the design cycle convergence test that compares the design
variables an properties that are output from an approximate optimization task
with those used as input to approximate optimization task. This test, while not
as conclusive as hard convergence, may be a suitable test to indicate that
further redesign cycles are unlikely to produce an improved design.
Synthetic Response is another term for Second Level Response.
Third Level Responses are defined on DRESP3 Bulk Data entries and are also referred
to as External Responses.
Type-1 Design Variable-to-Property Relations express analysis model properties as
linear functions of design variables. They are defined on DVPREL1 Bulk Data
entries.
Type-2 Design Variable-to-Property Relations express analysis model properties in
terms of design variables using MSC.Nastrans equation input capability. They
are defined on DVPREL2 Bulk Data entries and use equations defined on
DEQATN entries.
Violated Constraint is, strictly speaking, any constraint whose value is greater than
zero. In MSC.Nastran, the constant CTMIN (default = 0.003) is used to provide
a numerical zero. Thus, up to one-half of one percent constraint violation is
tolerated before a constraint is considered violated. This improves the
numerical conditioning of the problem since trying to precisely satisfy an
inequality relation is often a waste of computational resources.
Main Index Main Index
545 APPENDIX A
Nomenclature, Glossary of Terms, and References
A.3 References
1. Belegundu, A.D. and Rajan, S.P., Shape Optimal Design Using
Isoparametric Elements, Proceedings, 29th AIAA/ ASME/ ASEE/ AMS/ ASC
Structures, Dynamics and Materials Conference, pp. 696-701, Williamsburg,
VA, April 1988.
2. Bendsoe, M.P. and Sigmund, O., Topology Optimization Theory, Methods, and
Applications, Springer, 2003.
3. Cassis, J.H., Optimum Design of Structures Subjected to Dynamic Loads, UCLA-
ENG-7451, June 1974.
4. Johnson, E.H. and Neill, D.J., Automated Structural Optimization System
(ASTROS), Volume III, Application Manual, AFWAL TR-88-3028, December
1988, pp. 133-191.
5. Kelly, J. E., The Cutting Plane Method for Solving Convex Programs,
J.SIAM, Vol. 8, pp. 702-712, 1960.
6. Nelson, R.B., Simplified Calculations of Eigenvector Derivatives, AIAA
Journal, Vol. 14, No. 9, pp. 1201-1205.
7. Peery, D.J. and Azar, J.J., Aircraft Structures, McGraw-Hill Book Company,
1982, pp 328-341.
8. Powell, M. J. D., Algorithms for Nonlinear Constraints that Use Lagrangian
Functions, Math. Prog., Vol. 14, No. 2, pp. 224-248, 1978.
9. Rozvany, G.I.N., Bendsoe, M.P., and Kirsch, U., Layout Optimization of
Structures, Appl. Mech. Rev., 48, 1995, pp. 41-119.
10. Turner, M.J., "Design of Minimum Mass Structures with Specified Natural
Frequencies", AIAA Journal Vol. 5, No. 3, March 1967.
11. Vanderplaats, G. N., ADS -- A FORTRAN Program for Automated Design
Synthesis -- Version 1.10, NASA Contractor Report 177985, NASA Langley
Research Center, Hampton, Virginia, 1985.
12. Vanderplaats, G. N. and Sugimoto, H., Application of Variable Metric
Methods to Structural Synthesis, Engineering Computations, Vol. 2, No. 2,
June 1985.
13. Vanderplaats, G. N., Numerical Optimization Techniques for Engineering
Design: With Applications, 3rd Edition, Vanderplaats Research and
Development, Inc., Colorado Springs, CO, 1999.
Main Index Main Index


546
14. Vanderplaats, G. N., Very Large Scale Optimization, presented at the 8th
AIAA/ USAF/ NASA/ ISSMO Symposium at Multidisciplinary Analysis
and Optimization, Long Beach, CA, September 6-8, 2000.
15. DOT, Design Optimization Tools, User Manual, Version 5, Vanderplaats
Research and Development, Inc., Colorado Springs, CO, 1999.
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
APPENDIX
B
Adding Your Own Beam Cross-Section
Library
I Introduction
I BSCON Subroutine
I BSBRPD Subroutine
I BSGRQ Subroutine
I BSBRT Subroutine
I BSBRID Subroutine
I BSBRGB Subroutine
I BSBRCD Subroutine
I BSMSG Subroutine
I Linking Your Library to MSC.Nastran
I Example of Building and Linking a Beam Server
Main Index Main Index


548
B.1 Introduction
The standard cross sections provided by MSC should be adequate in the majority of
cases. If these standard sections are not adequate for your purposes, you can add your
own library of cross sections to suit your needs. To add your own library, you need to
write a few simple subroutines in FORTRAN and link them to MSC.Nastran through
inter-process communications. This process requires writing and/ or modifying up to
eight basic subroutines.
1. BSCON -- Defines the number of dimensions for each of the section types.
2. BSBRPD -- Calculates section properties based on section dimensions.
3. BSGRQ -- Defines NSECT, the number of section types, and NDIMAX, the
maximum number of dimensions (including nonstructural mass) required
by any of the sections.
4. BSBRT -- Provides the name, number of dimensions and number of design
constraints for each section type.
5. BSBRID -- Provides information for the calculation of gradients of section
properties with respect to section dimensions.
6. BSBRGD -- Calculates any nonlinear gradients of section properties with
respect to section dimensions.
7. BSBRCD -- Defines constraints in the design of section dimensions.
8. BSMSG -- A utility routine; handles errors that occur in the beam library.
BSCON and BSBRPD are always required. BSGRQ, BSBRT, and BSBRID are required
if you wish to perform sensitivity and/ or optimization tasks using the beam library.
BSBRGD is required if you are providing nonlinear analytical sensitivities in the
design task, and BSBRCD is an optional routine that can be provided to help the
optimizer to stay within physical design constraints. BSMSG handles any error
messages you feel are appropriate.
This appendix describes each of these basic routines in turn and provides MSC
Softwares version of each. Routines that are called by these basic routines are also
described with adequate examples to allow you to construct your own library. All the
example routines shown are for the 32-bit machines, such as IBM RISC/ System 6000.
Main Index Main Index
549 APPENDIX B
Adding Your Own Beam Cross-Section Library
For 64-bit machines such as CRAY C90, all the routine names that end with D
should be changed to end with S, and all real variables must be designated as single
precision instead of double precision. Therefore, the naming convention for routines
on 64-bit machines are BSCON, BSBRPS, BSGRQ, BSBRT, BSPRIS, BSBRGS, BSBRCS,
and BSMSG.
Main Index Main Index


550
B.2 BSCON Subroutine
This routine provides the number of fields in the continuation lines to be read from
the Bulk Data entries PBARL and PBEAML for each cross section in the library. The
value of the ENTYP variable may be 0, 1, or 2. When ENTYP=0, the value returned is
the number of DIMi. When ENTYP=1, the value returned includes both the DIMi and
NSM fields. The value of 1 is used for PBARL only. When ENTYP = 2, the value
returned includes the DMIi, NSM, SO, and XIXB fields for 11 different stations. The
value of 2 applies to PBEAML only.
The calling sequence and example routine for the standard MSC library is given
below.
Listing B-1 BSCON Subroutines
SUBROUTINE BSCON(GRPID,TYPE,ENTYP,NDIMI,ERROR)
C
C ----------------------------------------------------------------------
C Purpose
C To get the dimension information for 'MSC' types
C
C Arguments:
C
C GRPID input INTEGER INTEGER ID OF THIS GROUP OR GROUP NAME.
C TYPE input CHARACTER*8 TYPE OF CROSS SECTION
C ENTYP input INTEGER - 1: PBARL, 2:PBEAML
C NDIMI outpt NUMBER OF INPUT FIELDS FOR THE 'ENTYP' CROSS SECTION
C
C Called by BCCON
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT REAL (A-H,O-Z)

INTEGER ENTYP,GRPID,ERROR
CHARACTER*8 TYPE

C Local variables
INTEGER NAM(2)
DATA NAM/4HBSCO,4HN /

C Default to 'nothing wrong'
ERROR = 0

C Dimensions vary with section type
IF ( TYPE.EQ.'ROD ') THEN
NDIMI = 1
ELSEIF( TYPE.EQ.'TUBE ' .OR. TYPE.EQ.'BAR ') THEN
NDIMI = 2
Main Index Main Index
551 APPENDIX B
Adding Your Own Beam Cross-Section Library
ELSEIF( TYPE.EQ.'HEXA ') THEN
NDIMI = 3
ELSEIF( TYPE.EQ.'BOX ' .OR. TYPE.EQ.'T ' .OR.
+ TYPE.EQ.'L ' .OR.
+ TYPE.EQ.'CHAN ' .OR. TYPE.EQ.'CROSS ' .OR.
+ TYPE.EQ.'H ' .OR.
+ TYPE.EQ.'I1 ' .OR. TYPE.EQ.'T1 ' .OR.
+ TYPE.EQ.'CHAN1 ' .OR.
+ TYPE.EQ.'Z ' .OR. TYPE.EQ.'CHAN2 ' .OR.
+ TYPE.EQ.'T2 ' .OR. TYPE.EQ.'HAT ') THEN
NDIMI = 4
ELSEIF( TYPE.EQ.'HAT1 ') THEN
NDIMI = 5
ELSEIF( TYPE.EQ.'I ' .OR. TYPE.EQ.'BOX1 ') THEN
NDIMI = 6
ELSE
C Is the return here necessary ?
ERROR = 5150
RETURN
ENDIF

C Specify ?
IF (ENTYP.EQ.1) NDIMI = NDIMI+1
IF (ENTYP.EQ.2) NDIMI = (NDIMI+3)*11

C All done
RETURN
END
Main Index Main Index


552
B.3 BSBRPD Subroutine
Finite element analysis requires section properties such as area, moment of inertia,
etc., instead of section dimension. Therefore, the dimensions specified on PBARL and
PBEAML need to be converted to equivalent properties usually specified on PBAR
and PBEAM entries. The images of all these entries are stored in EPT datablock as
records.
BSBRPD subroutine is the interface of your properties evaluator with MSC.Nastran.
You may use your own naming convention for the subroutines that calculate the
cross-section properties from the dimensions. The calling tree used for the MSC
standard library is shown in Figure B-1.
Figure B-1 Calling Tree Generate Property Data
The BSBRPD calls the bar evaluator routine MEVBRD for the BAR element and the
beam evaluator routine MEVBMD for the BEAM element. The evaluators in turn call
the routines for each section. The routines are named BRXXPD and BMXXPD, where
XX is a two-letter identifier for the section. For example, the routines for the TUBE
section are called BRTUPD and BMTUPD. The details for various routines are given
in Listing B-2.
Listing B-2 BSBRPD Subroutines
SUBROUTINE BSBRPD(GRPID,ENTYP,TYPE,IDI,NID,IDO,NIDO,DIMI,NDIMI,
+ DIMO,NDIMO,ERROR)
C=======================================================================
C PURPOSE:
C 1.SET INTEGER PART
C 2.DETECT HOW MANY SECTIONS
BRTUPD
BRBXPD
MEVBRD
BMTUPD BMBXPD
BSBRPD
MEVBMD
OTHER
SECTIONS
Main Index Main Index
553 APPENDIX B
Adding Your Own Beam Cross-Section Library
C 3.UNIFORM OR LINEAR BEAM
C 4.CHECK WHAT ENTITY TYPES SUBROUTINE TO BE CALLED
C (1:PBARL ;2:PBEAML)
C-----------------------------------------------------------------------
C CALLED BY:
C BCBRP
C-----------------------------------------------------------------------
C CALLS:
C MEVBRD ,MEVBMD
C-----------------------------------------------------------------------
C IMPLICIT DECLARATIONS
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)
C.......................................................................
C EXPLICIT DECLARATIONS
INTEGER ENTYP,ERROR,GRPID
CHARACTER*8 TYPE
C.......................................................................
C LOCAL VARIABLES
INTEGER NAM(2)
C-----------------------------------------------------------------------
C DIMENSION STATEMENTS
INTEGER IDI(NID),IDO(NIDO)
DOUBLE PRECISION DIMI(NDIMI),DIMO(NDIMO)
C-----------------------------------------------------------------------
C DATA
DATA NAM/4HBSBR,4HPD /
C=======================================================================
C=== ENTYP=1, FOR PBARL; 2, FOR PBEAML
IF (ENTYP.EQ.1) THEN
CALL MEVBRD(GRPID,TYPE,IDI,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO
+ ,ERROR)
END IF
IF (ENTYP.EQ.2) THEN
CALL MEVBMD(GRPID,TYPE,IDI,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO
+ ,ERROR)
END IF
C-----------------------------------------------------------------------
RETURN
END
MEVBRD and MEVBMD Subroutines
MEVBRD and MEVBMD are the brancher routines for the various sections, and
convert the section dimensions to section properties for BAR and BEAM elements.
Your may rename these routines as you like or move the function of these routines to
BSBRPD. These routines call the BRXXPD or BMXXPD routines where XX is the two-
Main Index Main Index


554
letter keyword for various section types. The MEVBRD routine for the MSC standard
library is given in Listing B-3. The MEVBMD routine is not shown but can be
developed by using the BMXXPD convention.
Listing B-3 MEVBRD Subroutine
SUBROUTINE MEVBRD(GRPID,TYPE,ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,
+ NDIMO,ERROR)
C=======================================================================
C Purpose
C Call the default type subroutine to convert PBARL to PBAR
C
C Arguments
C
C GRPID input int ID of group
C TYPE input char Type of cross section
C DIMI input flt Dimension values of cross section
C NDIMI input int Size of DIMI array
C DIMO output flt Properties of cross section
C NDIMO output flt Size of DIMO array
C ERROR output int Type of error
C
C Method
C Call the subroutine with respect to the section type
C
C Called by
C BSBRPD
C
C CALLS
C BRRDPD,BRTUPD,BRBRPD,BRBXPD,BRIIPD,BRTTPD,BRLLPD,BRCHPD
C-----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER ERROR,GRPID,ID(NID),IDO(NIDO)
CHARACTER*8 TYPE
DOUBLE PRECISION DIMI(NDIMI),DIMO(NDIMO)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HMEVB,4HRD /
C=======================================================================

C Clear the output array before usage
CALL ZEROD ( DIMO, NDIMO )
CALL ZEROI ( IDO, NIDO )
ERROR = 0
C
Main Index Main Index
555 APPENDIX B
Adding Your Own Beam Cross-Section Library
IF ( TYPE.EQ.ROD ) THEN
CALL BRRDPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.TUBE ) THEN
CALL BRTUPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.BAR ) THEN
CALL BRBRPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.BOX ) THEN
CALL BRBXPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.I ) THEN
CALL BRIIPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.T ) THEN
CALL BRTTPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
c10939ELSEIF( TYPE.EQ.L ) THEN
c10939 CALL BRLLPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.CHAN ) THEN
CALL BRCHPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.CROSS ) THEN
CALL BRCRPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.H ) THEN
CALL BRHHPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.T1 ) THEN
CALL BRT1PD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.I1 ) THEN
CALL BRI1PD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.CHAN1 ) THEN
CALL BRC1PD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.Z ) THEN
CALL BRZZPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.CHAN2 ) THEN
CALL BRC2PD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.T2 ) THEN
CALL BRT2PD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.BOX1 ) THEN
CALL BRB1PD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.HEXA ) THEN
CALL BRHXPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.HAT ) THEN
CALL BRHTPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSEIF( TYPE.EQ.HAT1 ) THEN
CALL BRCLPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
ELSE
ERROR = 5150
END IF
C-----------------------------------------------------------------------
RETURN
END
Main Index Main Index


556
BRXXPD and BMXXPD Subroutines
The purpose of the BRXXPD and BMXXPD routines is to calculate the properties from
the section dimensions. For each cross section, subroutines are required to convert the
images of PBARL and PBEAML records to the images of PBAR and PBEAM records
in the EPT datablock.
BRTUPD Subroutine
BRTUPD is an example routine that shows how to convert PBARL EPT record to
PBAR EPT record for the Tube section. First, the details for the PBARL and PBAR
record are shown, and then the routine itself is given.
PBARL Record
The PBARL record in the EPT datablock is a derived from the PBARL Bulk Data entry
and is given in Table B-1.
Table B-1 PBARL (9102, 91, 52)
Word Name Type Description
1 PID I Property identification number.
2 MID I Material identification number.
3 Group Char Group Name.
4 Group Char Group Name.
5 TYPE Char4 Cross-section Type.
6 TYPE Char4 Cross-section Type
7 Dim1 RS Dimension1.
8 Dim2 RS Dimension2.
n+7-1 Dim n RS Dimension n (note that the final
dimension is the nonstructural mass).
n+7 Flag I -1. Flag indicating ed of cross-section
dimensions.
Main Index Main Index
557 APPENDIX B
Adding Your Own Beam Cross-Section Library
PBAR Record
The PBAR record in the EPT datablock is derived from the PBAR Bulk Data entry and
consists of 19 words. It is a replica of the Bulk Data entry, starting with PID field. The
word 8 in the record is set to 0.0 since the field 9 in the first line of the PBAR Bulk Data
entry is not used. The details of the PBAR record are given in Table B-2.
Table B-2 PBAR (52, 20, 181)
Word Name Type Description
1 PID I Property identification number.
2 MID I Material identification number.
3 A RS Area of cross-section.
4 I1 RS Area moment of inertia for
bending in plane 1.
5 I2 RS Area moment of inertia for
bending in plane 2.
6 J RS Torsional constant.
7 NSM RS Nonstructural mass per unit
length.
8 FE RS Not used. Set of 0.0.
9 C1 RS Stress recovery location.
10 C2 RS Stress recovery location.
11 D1 RS Stress recovery location.
12 D2 RS Stress recovery location.
13 E1 RS Stress recovery location.
14 E2 RS Stress recovery location.
15 F1 TS Stress recovery location.
16 F2 RS Stress recovery location.
17 K1 RS Area factor of shear for plane 1.
18 K2 RS Area factor of shear for plane 2.
19 I12 RS Area product of inertia.
Main Index Main Index


558
Listing B-4 BRTUPD Subroutine
SUBROUTINE BRTUPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
C=======================================================================
C Purpose:
C Convert PBARL(entity type : TUBE) to PBAR
C
C Arguments:
C
C ID input int Array of values PID, MID contained in PBARL
C entries
C NID input int Size of ID array, NID=2 for PBARL entry
C IDO output int Array of integer values contained in PBAR
C entries
C NIDO output int Size of IDO array, NIDO=2 for PBAR entry
C DIMI input flt Array of dimension values for cross-section
C (SEE FIG. 5 IN MEMO SSW-25, REV. 4, DATE 8/16/94)
C NDIM input int Size of DIMI array
C DIMO output flt Array of property values for cross-section
C NDIMO output int Size of DIMO array
C ERROR output int Type of error
C
C DISCRIPTION FOR DIMO ARRAY:
C DIMO (1) = A
C DIMO (2) = I1
C DIMO (3) = I2
C DIMO (4) = J
C DIMO (5) = NSM
C DIMO (6) = FE
C DIMO (7) = C1
C DIMO (8) = C2
C DIMO (9) = D1
C DIMO (10) = D2
C DIMO (11) = E1
C DIMO (12) = E2
C DIMO (13) = F1
C DIMO (14) = F2
C DIMO (15) = K1
C DIMO (16) = K2
C DIMO (17) = I12
C
C Method
C
C Called by:
C MEVBRD
C-----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER ID(NID),IDO(NIDO),ERROR
Main Index Main Index
559 APPENDIX B
Adding Your Own Beam Cross-Section Library
DOUBLE PRECISION DIMI(NDIMI),DIMO(NDIMO)

C Local variables
INTEGER NAM(2)

C Nastran common blocks
COMMON /CONDAD/PI

C Local data
DATA NAM/4HBRTU,4HPD /
C=======================================================================
C
C=== WRITE THE PART OF INTEGER
DO 30 II = 1,NID
IDO(II) = ID(II)
30 CONTINUE
C
DIM1 = DIMI(1)
DIM2 = DIMI(2)
DIMO(1) = PI*(DIM1*DIM1-DIM2*DIM2)
DIMO(2) = PI*(DIM1**4-DIM2**4)/4.D0
DIMO(3) = DIMO(2)
DIMO(4) = PI*(DIM1**4-DIM2**4)/2.D0
DIMO(5) = DIMI(3)
DIMO(6) = 0.D0
DIMO(7) = DIM1
DIMO(8) = 0.D0
DIMO(9) = 0.D0
DIMO(10) = DIM1
DIMO(11) = -DIM1
DIMO(12) = 0.D0
DIMO(13) = 0.D0
DIMO(14) = -DIM1
DIMO(15) = 0.5D0
DIMO(16) = 0.5D0
DIMO(17) = 0.D0

IF ( DIMI(1).LE.DIMI(2) ) error = 5102

C-----------------------------------------------------------------------
RETURN
END
BMTUPD Subroutine
BMTUPD is an example routine that shows how to convert PBEAM EPT record to
PBEAM EPT record for the Tube section. First, the details of the PBEAML and the
PBEAM records are shown, and then the routine itself is given.
Main Index Main Index


560
PBEAML Record
The PBEAML record on the EPT datablock is a derived from the PBEAML Bulk Data
entry and is given in Table B-3.
PBEAM Record
The PBEAM record in EPT datablock consists of 197 words. The first five words and
the last 16 words are common to all the 11 stations. Each of the 11 stations have their
own 16 unique words. The details of the PBEAM record are given in Table B-4.
Table B-3 PBEAML (9202, 92, 53)
Word Name Type Description
1 PID I Property ID.
2 MID I Material ID.
3 Char Group Name.
4 Char Group Name.
5 Char4 Cross-section Type.
6 Char4 Cross-section Type.
7 RS Stress output request flag, 1=yes, =no.
8 RS X/ XB - parametric location of the station.
9 RS Dimension 1.
10 RS Dimension 2.
n+9-1 RS Dimension n (note that the final dimension is
the nonstructural mass).
n+9 I -1. Flag indicating end of cross-section
dimensions.
Words 7 through 11 repeat 10 times.
Table B-4 PBEAM (5402, 54, 262)
Word Name Type Description
1 PID I Property identification number.
2 MID I Material identification number.
Main Index Main Index
561 APPENDIX B
Adding Your Own Beam Cross-Section Library
3 N I Number of intermediate stations.
4 CCF I Constant cross-section flag. 1 = constant, 2 =
variable.
5 X RS Unused.
6 SO RS Stress output request. 1.0 = yes, 0.0 = no.
7 XXB RS Parametric location of the station. Varies
between 0. and 1.0.
8 A RS Area.
9 I1 RS Moment of inertia for bending in plane 1.
10 I2 RS Moment of inertia for bending in plane 2.
11 I12 RS Area product of inertia.
12 J RS Torsional constant.
13 NSM RS Nonstructural mass.
14 C1 RS Stress recovery location.
15 C2 RS Stress recovery location.
16 D1 RS Stress recovery location.
17 D2 RS Stress recovery location.
18 E1 RS Stress recovery location.
19 E2 RS Stress recovery location.
20 F1 RS Stress recovery location.
21 F2 RS Stress recovery location.
Words 6 through 21 repeat 10 times.
182 K1 RS Area factor for shear for plane 1.
183 K2 RS Area factor for shear for plane 2.
184 S1 RS Shear-relief coefficient for plane 1.
185 S2 RS Shear-relief coefficient for plane 2.
Table B-4 PBEAM (5402, 54, 262)
Word Name Type Description
Main Index Main Index


562
Listing B-5 BMTUPD Subroutine.
SUBROUTINE BMTUPD(ID,NID,IDO,NIDO,DIMI,NDIMI,DIMO,NDIMO,ERROR)
C=======================================================================
C Purpose
C Convert PBEAML(entity type : TUBE) to PBEAM
C
C Arguments
C
C ID input int Contain the integer information PID, MID
C NID input int Size of ID array, NID = 2
C DIMI input flt Dimension values of cross section
C ( See FIG.5 IN MEMO SSW-25, REV. 4, DATE 8/16/94)
C NDIMI input int Size of DIMI array
C IDO output int Contain the integer information PID,MID,N,CCF
C NIDO output int Size of IDO array, NIDO = 4
186 NSIA RS Nonstructural mass moment of inertia at
end A.
187 NSIB RS Nonstructural mass moment of inertia at
end B.
188 CWA RS Warping coefficient for end A.
189 CWB RS Warping coefficient for end B.
190 M1A RS Y-coordinate of center of gravity for
nonstructural mass at end A.
191 M2A RS Z-coordinate of center of gravity for
nonstructural mass at end A.
192 M1B RS Y-coordinate of center of gravity for
nonstructural mass at end B.
193 M2B RS Z-coordinate of center of gravity for
nonstructural mass at end B.
194 N1A RS Y-coordinate for neutral axis at end A.
195 N2A RS Z-coordinate for neutral axis at end A.
196 N1B RS Y-coordinate for neutral axis at end B.
197 N2B RS Z-coordinate for neutral axis at end B.
Table B-4 PBEAM (5402, 54, 262)
Word Name Type Description
Main Index Main Index
563 APPENDIX B
Adding Your Own Beam Cross-Section Library
C DIMO output flt Properties of cross section
C NDIMO output int Size of DIMO array
C ERROR output int Type of error
C
C Discription for DIMO array
C DIMO (1) = X
C DIMO (2) = SO
C DIMO (3) = XXB
C DIMO (4) = A
C DIMO (5) = I1
C DIMO (6) = I2
C DIMO (7) = I12
C DIMO (8) = J
C DIMO (9) = NSM
C DIMO (10) = C1
C DIMO (11) = C2
C DIMO (12) = D1
C DIMO (13) = D2
C DIMO (14) = E1
C DIMO (15) = E2
C DIMO (16) = F1
C DIMO (17) = F2
C DIMO(2) thru DIMO(17) repeat 11 times
C DIMO (178) = K1
C DIMO (179) = K2
C DIMO (180) = S1
C DIMO (181) = S2
C DIMO (182) = NSIA
C DIMO (183) = NSIB
C DIMO (184) = CWA
C DIMO (185) = CWB
C DIMO (186) = M1A
C DIMO (187) = M2A
C DIMO (188) = M1B
C DIMO (189) = M2B
C DIMO (190) = N1A
C DIMO (191) = N2A
C DIMO (192) = N1B
C DIMO (193) = N2B
C
C Method
C Simply calculate the properties and locate that data
C to the image of PBEAM entries
C Called by
C MEVBMD
C-----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER ERROR,ID(NID),IDO(NIDO)
Main Index Main Index


564
DOUBLE PRECISION DIMI(NDIMI),DIMO(NDIMO)

C Local variables
INTEGER NAM(2)

C NASTRAN common blocks
COMMON /CONDAD/PI

C Local data
DATA NAM/4HBMTU,4HPD /
C=======================================================================
C
C=== WRITE THE PART OF INTEGER
DO 30 II = 1,NID
IDO(II) = ID(II)
30 CONTINUE
C-----------------------------------------------------------------------
C=== DETECT HOW MANY STATION , CONSTANT OR LINEAR BEAM.
C-----------------------------------------------------------------------
ISTATC = NDIMI/11
DO 35 II = 0,10
NW = II*ISTATC
IF (DIMI(3+NW).EQ.0.D0) THEN
call bmsido( ido(3), ido(4),ii)
GO TO 40
END IF
35 CONTINUE
C If we got here , we have all the stations, so set the values
C
if( dimi(2+9*istatc).eq. dimi(2+10*istatc) ) then
ii = 9
else
ii = 10
endif
ido(3) = ii
ido(4) = 0
c
40 continue
DIMO(1) = 0.D0
DO 100 L1 = 0,10
LC = 16*L1
NW = L1*ISTATC
IF (DIMI(3+NW+ISTATC).EQ.0.D0) LC = 160
DIM1 = DIMI(3+NW)
DIM2 = DIMI(4+NW)

IF ( DIM1.LE.DIM2 ) ERROR = 5102

DIMO(2+LC) = DIMI(1+NW)
DIMO(3+LC) = DIMI(2+NW)
DIMO(4+LC) = PI*(DIM1*DIM1-DIM2*DIM2)
Main Index Main Index
565 APPENDIX B
Adding Your Own Beam Cross-Section Library
DIMO(5+LC) = PI*(DIM1**4-DIM2**4)/4.D0
DIMO(6+LC) = DIMO(5+LC)
DIMO(7+LC) = 0.D0
DIMO(8+LC) = PI*(DIM1**4-DIM2**4)/2.D0
DIMO(9+LC) = DIMI(5+NW)
DIMO(10+LC) = DIM1
DIMO(11+LC) = 0.D0
DIMO(12+LC) = 0.D0
DIMO(13+LC) = DIM1
DIMO(14+LC) = -DIM1
DIMO(15+LC) = 0.D0
DIMO(16+LC) = 0.D0
DIMO(17+LC) = -DIM1
IF (LC.EQ.160) GO TO 110
100 CONTINUE
110 DIMO(178) = 0.5D0
DIMO(179) = 0.5D0
C-----------------------------------------------------------------------
300 RETURN
END
Main Index Main Index


566
B.4 BSGRQ Subroutine
For the optimization of PBARL or PBEAML entries, you need to provide overall
information on the number of section types in your library and the maximum number
of fields on the continuation lines. For the PBARL, the maximum number is the
maximum number of dimensions plus 1 for the non-structural mass field. For the
PBEAML, it is the number of words per station times the eleven possible stations. The
count for the number of station words is the maximum number of dimensions plus
three additional words for SO,X/ XB and NSM. BSGRQ is only required if you wish to
perform sensitivity or optimization with the section dimensions. The calling sequence
and example subroutine for the MSCBML1 library is shown in
Listing B-6. It is seen that NDIMAX is set to 7 for the PBARL and 99 for the PBEAML.
Listing B-6 BSGRQ Routine
SUBROUTINE BSGRQ(GRPID,ENTYP, NSECT,NDIMAX,ERROR)
C
C ======================================================================
C PURPOSE:
C GET DIMENSION INFORMATION OF BEAM CROSS SECTION OF MSCBML0 TYPES
C ----------------------------------------------------------------------
C ARGUMENTS:
C
C INPUT :
C GRPID - GROUP NAME
C ENTYP - INTEGER - Entity Type.
c (1: PBARL, 2:PBEAML )
C OUTPUT :
C NSECT - NUMBER OF DIFFERENT SECTION TYPES
C NDIMAX - MAXIMUM NUMBER OF DIMENSION FOR ANY SECTION TYPE
C ERROR - INDICATES IF AN ERROR HAS OCCURRED. THE CODE RETURNED
C INDICATES THE TYPE OF ERROR
C
C REASSIGNED :
C ----------------------------------------------------------------------
C CALLED BY:
C BCGRQ routine
C-----------------------------------------------------------------------
C
integer entyp
C
if ( entyp .eq. 1 ) then
nsect = 21
Main Index Main Index
567 APPENDIX B
Adding Your Own Beam Cross-Section Library
ndimax = 7
else if ( entyp .eq. 2 ) then
nsect = 21
ndimax = 99
endif
C
RETURN
END
Note: The arguments GRPID and ERROR are not used. GRPID is reserved for future
use. You may use the ERROR argument to send an error code which could
later be used to print an error message.
Main Index Main Index


568
B.5 BSBRT Subroutine
The BSBRT routine provides the name, number of dimension, and the number of
constraints for each section in the library. As an example, the name of the tube section,
shown in Figure B-2, in the MSC standard library is TUBE, the number of
dimension for the tube section is three (OUTER RADIUS, INNER RADIUS, and NSM)
and there is one physical constraint. This subroutine is shared by the PBARL and
PBEAML. For the PBEAML, the numbers apply to each situation.
The physical constraint is that the inner radius (DIM2) cannot be greater than outer
radius (DMI1). It is necessary to specify the constraints so that the optimization of the
section dimension in SOL 200 does not result into an inconsistent shape.
Figure B-2 Tube Section
This routine is called during the optimization process. The calling sequence and
example routine is given in Listing B-7.
Listing B-7 BSBRT Subroutine
SUBROUTINE BSBRT(GRPID,ENTYP,TYPE,NDIM,NCONST,NSECT,ERROR)
C
C ======================================================================
C Purpose
C Get type dimensioning information for default sections
Y
Z
Dim1
Dim2
F
E
D
C
Tube
r
0
DIM1 =
r
i
DIM2 =
Main Index Main Index
569 APPENDIX B
Adding Your Own Beam Cross-Section Library
C
C Arguments
C
C GRPID input integer The ID of group name
C ENTYP input integer 1: PBARL, 2: PBEAML
C TYPE output character*8 Arrays for cross section types
C NDIM output integer Number of dimensions for isect
C section type
C NCONST output integer Number of dimensional constraints
C imposed by isect section type
C NSECT input integer Number of sections
C ERROR output integer Type of error
C
C Method
C Transfers dimension information for various sections
C
C Called by
C BCBRT routine
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER GRPID,ENTYP,NDIM(NSECT),NCONST(NSECT)
CHARACTER*8 TYPE(NSECT)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBSBR,4HT /
C ======================================================================

GRPID = 1
C ENTYP = 1

TYPE(1) = 'ROD'
NDIM(1) = 2
NCONST(1) = 0
C
TYPE(2) = 'TUBE'
NDIM(2) = 3
NCONST(2) = 1
C
TYPE(3) = 'BAR'
NDIM(3) = 3
NCONST(3) = 0
C
TYPE(4) = 'BOX'
NDIM(4) = 5
NCONST(4) = 2
Main Index Main Index


570
C
TYPE(5) = 'I'
NDIM(5) = 7
NCONST(5) = 3
C
TYPE(6) = 'T'
NDIM(6) = 5
NCONST(6) = 2
C
TYPE(7) = 'L'
NDIM(7) = 5
NCONST(7) = 2
C
TYPE(8) = 'CHAN'
NDIM(8) = 5
NCONST(8) = 2
C
TYPE(9) = 'CROSS'
NDIM(9) = 5
NCONST(9) = 1
C
TYPE(10) = 'H'
NDIM(10) = 5
NCONST(10) = 1
C
TYPE(11) = 'T1'
NDIM(11) = 5
NCONST(11) = 1
C
TYPE(12) = 'I1'
NDIM(12) = 5
NCONST(12) = 1
C
TYPE(13) = 'CHAN1'
NDIM(13) = 5
NCONST(13) = 1
C
TYPE(14) = 'Z'
NDIM(14) = 5
NCONST(14) = 1
C
TYPE(15) = 'CHAN2'
NDIM(15) = 5
NCONST(15) = 2
C
TYPE(16) = 'T2'
NDIM(16) = 5
NCONST(16) = 2
C
TYPE(17) = 'BOX1'
NDIM(17) = 7
Main Index Main Index
571 APPENDIX B
Adding Your Own Beam Cross-Section Library
NCONST(17) = 2
C
TYPE(18) = 'HEXA'
NDIM(18) = 4
NCONST(18) = 1
C
TYPE(19) = 'HAT'
NDIM(19) = 5
NCONST(19) = 2
C
TYPE(20) = 'HAT1'
NDIM(20) = 6
NCONST(20) = 3

C
RETURN
END
Main Index Main Index


572
B.6 BSBRID Subroutine
The BSBRID subroutine is required if optimization is to be performed. Its function is
to provide information required in the calculation of the sensitivities (gradients) of the
bar and beam properties with respect to their dimensions.
Two basic types of information are provided. The first is the SENTYP array, which
indicates how each section property varies as a function of each dimension. Values in
the SENTYP array can be either: 0 for no variation; 1 for a linear variation; 2 for a
nonlinear variation; 3 for an unknown variation. The SENTYP=3 option is to be used
when you know that the property is a function of the design dimension, but analytical
gradient information is not being provided using the BSBRGD subroutine. In this case,
MSC.Nastran will calculate the gradients for you using central differencing
techniques.
The second piece of information is the ALIN array. This array provides any linear
sensitivity data. For example, the C1 stress recovery location for the TUBE section is
1.0 * DIM1 so that this sensitivity of this stress recovery point with respect to the first
dimension is 1.0.
You may use your own naming convention for the subroutines that specify the section
sensitivity data. The calling tree used for the MSC.Nastran standard library is shown
in Figure B-3.
Figure B-3 Calling Tree to Generate Sensitivity Data
BSBRID calls the bar evaluator routine MSBRID/ S for the BAR element and
MSBMID/ S for the BEAM element. The evaluators in turn call the routines for each
section. The routines are named BRxxID or BMxxID, where xx is a two-letter identifier
BRTUID
BRBXID
MSBRID
BMTUID BMBXID
BSBRPD
MSBMD
OTHER
SECTIONS
Main Index Main Index
573 APPENDIX B
Adding Your Own Beam Cross-Section Library
for the section. For example, the routine for the bar TUBE section is called BRTUID
while the beam TUBE is called BMTUID. Listings of BSBRID, MSBRID, BRTUID and
BMTUID are given in Listing B-8, Listing B-9, Listing B-10, and Listing B-11,
respectively. While MSBMID is not shown, it can be developed by calling the BMXXID
subroutines in place of the BRXXID of MSBRID.
Listing B-8 BSBRID Subroutine
SUBROUTINE BSBRID(GRPID,ENTYP,SECTON,SENTYP,ALIN,NDIM,NPROP,
1 ERROR)
C =====================================================================
C Purpose
C set up section dependent information for a particular cross
C section type
C
C Arguments
C
C GRPID input integer ID of the group
C ENTYP input integer 1: PBARL, 2: PBEAML
C SECTON input character*8 Section type
C SENTYP output integer Type of sensitivity, 0: invariant,
C 1: linear, 2: nonlinear
C ALIN output double Matrix providing the linear
C factors for sensitive relationships
C NDIM input integer Number of dimensions
C NPROP input integer Number of properties
C ERROR output integer Type of error
C
C Method
C Simply transfer control based on entry type
C
C Called by
C BCBRID
C
C Calls
C MSBRID, MSBMID
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER ENTYP,ERROR,GRPID,SENTYP(NPROP,NDIM)
CHARACTER*8 SECTON
DOUBLE PRECISION ALIN(NPROP,NDIM)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBSBR,4HID /
Main Index Main Index


574
C ======================================================================
GRPID = 1
IF(ENTYP.EQ.1) THEN
CALL MSBRID(SECTON,SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE if ( entyp .eq. 2 ) then
CALL MSBmID(SECTON,SENTYP,ALIN,NDIM,NPROP,ERROR)
else
ERROR = 5400
END IF

C
RETURN
END
MSBRID Subroutine
MSBRID is a brancher routine for providing information on the calculation of
sensitivities for each of the bar types. You may rename this routine as you like or move
its function to BSBRID. MSBRID calls the BRXXID routines, where XX is the two-letter
keyword for various section types.
Listing B-9 MSBRID Subroutine
SUBROUTINE MSBRID(SECTON,SENTYP,ALIN,NDIM,NPROP,ERROR)
C =====================================================================
C Purpose
C To set up section dependent information for PBARL cross section
C
C Arguments
C
C SECTON input character*8 Name of section type
C SENTYP output integer Type of sensitivity, 0: invariant,
C 1: linear, 2: nonlinear
C ALIN output double Matrix providing the linear factors
C for sensitive relationships
C NDIM input integer No. of dimensions
C NPROP input integer No. of properties in EPT data block
C ERROR output integer Type of error
C
C Method
C Simply transfer the section dependent information of
C the 19 kinds
C Called by
C BSBRID
C
C Calls
C BRRDID, BRBRID, BRTUID, BRBXID, BRIIID, BRLLID, BRTTID,
C BRCHID, BRCRID, BRHHID, BRT1ID, BRI1ID, BRC1ID, BRZZID,
C BRC2ID, BRT2ID, BRB1ID, BRHXID, BRHTID, BRCLID
C ZEROI, ZEROD (Nastran utility)
Main Index Main Index
575 APPENDIX B
Adding Your Own Beam Cross-Section Library
C ----------------------------------------------------------------------

IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
CHARACTER*8 SECTON
INTEGER ERROR,SENTYP(NPROP,NDIM)
DOUBLE PRECISION ALIN(NPROP,NDIM)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HMSBR,4HID /
C ======================================================================
C
CALL ZEROI( SENTYP(1,1), NPROP*NDIM )
CALL ZEROD( ALIN(1,1), NPROP*NDIM )
ERROR = 0

C
IF(SECTON.EQ.ROD) THEN
CALL BRRDID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.BAR) THEN
CALL BRBRID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.TUBE) THEN
CALL BRTUID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.BOX) THEN
CALL BRBXID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.I) THEN
CALL BRIIID(SENTYP,ALIN,NDIM,NPROP,ERROR)
c10939ELSE IF(SECTON.EQ.L) THEN
c10939 CALL BRLLID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.T) THEN
CALL BRTTID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.CHAN) THEN
CALL BRCHID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.CROSS) THEN
CALL BRCRID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.H) THEN
CALL BRHHID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.T1) THEN
CALL BRT1ID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.I1) THEN
CALL BRI1ID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.CHAN1) THEN
CALL BRC1ID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.Z) THEN
CALL BRZZID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.CHAN2) THEN
Main Index Main Index


576
CALL BRC2ID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.T2) THEN
CALL BRT2ID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.BOX1) THEN
CALL BRB1ID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.HEXA) THEN
CALL BRHXID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.HAT) THEN
CALL BRHTID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE IF(SECTON.EQ.HAT1) THEN
CALL BRCLID(SENTYP,ALIN,NDIM,NPROP,ERROR)
ELSE
ERROR = 5410
END IF

C
RETURN
END
BRTUID Subroutine
BRTUID is an example routine that shows how to define the sensitivity type of each
of the bar properties for the tube and the subset of the sensitivities that are linear.
Listing B-10 BRTUID Subroutine
SUBROUTINE BRTUID(SENTYP,ALIN,NDIM,NPROP,ERROR)
C ======================================================================
C Purpose
C To set up section dependent information for rod cross section
C
C Arguments
C
C SENTYP output integer Type of sensitivity, 0: invariant,
C 1: linear, 2: nonlinear
C ALIN output double Matrix providing the linear factors for
C sensitive relationships
C NDIM input integer No. of dimensions
C NPROP input integer No. of properties in EPT data block
C ERROR output integer Type of error
C
C Method
C Simply provides the information
C
C Called by
C MSBRID
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
Main Index Main Index
577 APPENDIX B
Adding Your Own Beam Cross-Section Library
INTEGER ERROR,SENTYP(NPROP,NDIM)
DOUBLE PRECISION ALIN(NPROP,NDIM)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBRTU,4HID /
C ======================================================================
C
ALIN( 7,1) = 1.0D0
ALIN(10,1) = 1.0D0
ALIN(11,1) = -1.0D0
ALIN(14,1) = -1.0D0
ALIN( 5,3) = 1.0D0
C
SENTYP( 1,1) = 2
SENTYP( 1,2) = 2
SENTYP( 2,1) = 2
SENTYP( 2,2) = 2
SENTYP( 3,1) = 2
SENTYP( 3,2) = 2
SENTYP( 4,1) = 2
SENTYP( 4,2) = 2
SENTYP( 7,1) = 1
SENTYP(10,1) = 1
SENTYP(11,1) = 1
SENTYP(14,1) = 1
SENTYP( 5,3) = 1

C
RETURN
END
BMTUID Subroutine
BMTUID is an example routine that shows how to define the sensitivity type of each
of the beam properties for the tube and the subset of the sensitivities that are linear.
Listing B-11 BMTUID Subroutine
SUBROUTINE BMTUID(SENTYP,ALIN,NDIM,NPROP,ERROR)
C ======================================================================
C Purpose
C To set up section dependent information for TUBE cross section
C
C Arguments
C
C SENTYP output integer Type of sensitivity, 0: invariant,
C 1: linear, 2: nonlinear
Main Index Main Index


578
C ALIN output double Matrix providing the linear factors for
C sensitive relationships
C NDIM input integer No. of dimensions
C NPROP input integer No. of properties in EPT data block
C ERROR output integer Type of error
C
C Method
C Simply provides the information
C
C Called by
C MSBMID
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER ERROR,SENTYP(NPROP,NDIM)
DOUBLE PRECISION ALIN(NPROP,NDIM)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBMTU,4HID /
C ======================================================================
C
C Tube dimensions for the beam:
C 1 - s0
C 2 - x/xb
C 3 - outer radius
C 4 - inner radius
C 5 - nsm

ALIN(10,3) = 1.0D0
ALIN(13,3) = 1.0D0
ALIN(14,3) = -1.0D0
ALIN(17,3) = -1.0D0
ALIN( 9,5) = 1.0D0
C
SENTYP( 4,3) = 2
SENTYP( 4,4) = 2
SENTYP( 5,3) = 2
SENTYP( 5,4) = 2
SENTYP( 6,3) = 2
SENTYP( 6,4) = 2
SENTYP( 8,3) = 2
SENTYP( 8,4) = 2
SENTYP(10,3) = 1
SENTYP(13,3) = 1
SENTYP(14,3) = 1
SENTYP(17,3) = 1
Main Index Main Index
579 APPENDIX B
Adding Your Own Beam Cross-Section Library
SENTYP( 9,5) = 1
C
100 continue
C call mprtd ( alin, alin, nprop,ndim )
C ---------------------------------------------------------------------
-
RETURN
END
Main Index Main Index


580
B.7 BSBRGB Subroutine
The BSBRGD subroutine is required if optimization is to be performed and analytical
sensitivities are needed (SENTYPE = 2 in subroutine BSBRID). Its function is to
provide the nonlinear gradients of the bar properties with respect to the bar
dimensions. You may use your own naming convention for the subroutines that
calculate the section gradients from the dimensions. The calling tree used for the
MSC.Nastran standard library is shown in Figure B-4.
Figure B-4 Calling Tree to Generate Nonlinear Sensitivity Data
BSBRGD calls the bar evaluator routine MSBRGD/ S for the BAR element and
MSBMGD/ S for the BEAM element. The evaluator, in turn, calls the routines for each
section. The routines are named BRxxGD or BMxxGD where xx is a two-letter
identifier for the section. For example, the routine for the TUBE section is called
BRTUGD. Listings of BSBRGD, MSBRGD, BRTUGD and BMTUGD are given in
Listing B-12, Listing B-13, Listing B-14, and Listing B-15, respectively. MSBMGD is
not shown but can be developed by calling MBxxGD subroutines in place of the
BRxxGD of MSBRGP.
Listing B-12 BSBRGD Subroutine
SUBROUTINE BSBRGD(GRPID,ENTYP,TYPE,DIMI,NDIMI,ANONL,NPROP,ERROR)
C ======================================================================
C Purpose
C To get the nonlinear factors of sensitivities for default sections
C
C Arguments
C
BRTUGD
BRBXGD
MSBRGD
BMTUGD BMBXGD
BSBRGD
MSBMD
OTHER
SECTIONS
Main Index Main Index
581 APPENDIX B
Adding Your Own Beam Cross-Section Library
C GRPID input integer ID of group name
C ENTYP input integer 1: PBARL, 2: PBEAML
C TYPE input character*8 Type name of cross-section
C DIMI input double Array from EPT record
C NDIMI input integer Number of dimensions (Plus NSM)
C ANONL output double Array providing the nonlinear factors
C for sensitivity relationships
C NPROP input integer Number of properties in PBAR entries
C ERROR output integer Type of error
C
C Method
C Simply transfer control based on entry types
C Called by
C BCBRGD
C
C Calls
C MSBRGD
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER GRPID,ENTYP,NDIMI,NPROP,ERROR
CHARACTER*8 TYPE
DOUBLE PRECISION ANONL(NPROP,NDIMI), DIMI(NDIMI)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBSBR,4HGD /
C ======================================================================

GRPID = 1
IF(ENTYP .EQ. 1) THEN
CALL MSBRGD(TYPE,DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE if ( entyp .eq. 2 ) then
CALL MSBMGD(TYPE,DIMI,NDIMI,ANONL,NPROP,ERROR)
else
ERROR = 5300
END IF
C ----------------------------------------------------------------------
RETURN
END
Main Index Main Index


582
MSBRGD Subroutine
MSBRGD is a brancher routine for providing information on the calculation of
nonlinear gradients for each of the bar types. You may rename this routine as you like
or move its function to BSBRGD. MSBRGD calls the BRxxGD routines, where xx is the
two-letter keyword for various section types.
Listing B-13 MSBRGD Subroutine
SUBROUTINE MSBRGD(TYPE,DIMI,NDIMI,ANONL,NPROP,ERROR)
C
C ======================================================================
C Purpose
C To get the nonlinear factors of sensitivities for PBARL entries
C
C Arguments
C
C TYPE input character*8 Type name of cross-section
C DIMI input double Array from EPT record
C NDIMI input integer Number of dimensions (Plus NSM)
C ANONL output double Array providing the nonlinear factors
C for sensitivity relationships
C NPROP input integer Number of properties in PBAR entries
C ERROR output integer Type of error
C
C Method
C Simply transfer information based on cross-section type
C Called by
C BSBRGD
C
C Calls
C BRRDGD,BRBRGD,BRBXGD,BRTUGD,BRIIGD,BRTTGD,BRLLGD,BRCHGD
C BRCRGD,BRHHGD,BRT1GD,BRT2GD,BRI1GD,BRC1GD,BRC2GD,BRZZGD
C BRHXGD,BRB1GD,BRHTGD,ZEROD(Nastran utility)
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER NDIMI,NPROP,ERROR
CHARACTER*8 TYPE
DOUBLE PRECISION ANONL(NPROP,NDIMI), DIMI(NDIMI)

C Local variables
INTEGER NAM(2)

C Local data
DATA NAM/4HMSBR,4HGD /
C ======================================================================

Main Index Main Index
583 APPENDIX B
Adding Your Own Beam Cross-Section Library
CALL ZEROD( ANONL(1,1), NPROP*NDIMI )
ERROR = 0

IF(TYPE.EQ.ROD) THEN
CALL BRRDGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.BAR) THEN
CALL BRBRGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.BOX) THEN
CALL BRBXGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.TUBE) THEN
CALL BRTUGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.I) THEN
CALL BRIIGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
c10939ELSE IF(TYPE.EQ.L) THEN
c10939 CALL BRLLGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.T) THEN
CALL BRTTGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.CHAN) THEN
CALL BRCHGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.CROSS) THEN
CALL BRCRGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.H) THEN
CALL BRHHGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.T1) THEN
CALL BRT1GD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.I1) THEN
CALL BRI1GD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.CHAN1) THEN
CALL BRC1GD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.Z) THEN
CALL BRZZGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.CHAN2) THEN
CALL BRC2GD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.T2) THEN
CALL BRT2GD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.BOX1) THEN
CALL BRB1GD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.HEXA) THEN
CALL BRHXGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.HAT) THEN
CALL BRHTGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE IF(TYPE.EQ.HAT1) THEN
CALL BRCLGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
ELSE
ERROR = 5310
ENDIF

C
RETURN
END
Main Index Main Index


584
BRTUGD Subroutine
BRTUGD is a BAR example routine that shows how to define the nonlinear gradients
of the TUBE section as a function of the outer and inner radii of the tube.
Listing B-14 BRTUGD Subroutine
SUBROUTINE BRTUGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
C ======================================================================
C Purpose
C To get the nonlinear factors of sensitivities for TUBE section
C
C Arguments
C
C DIMI input double Array of EPT records (Dimi+NSM)
C NDIMI input integer Number of dimensions (Plus NSM)
C ANONL output double Array providing the nonlinear factors
C for sensitivity relationships
C NPROP input integer Number of properties in PBAR entries
C ERROR output integer Type of error
C
C Method
C Simply calculates the nonlinear factors
C Called by
C MSBRGD
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER NDIMI,NPROP,ERROR
DOUBLE PRECISION ANONL(NPROP,NDIMI), DIMI(NDIMI)

C Local variables
INTEGER NAM(2)

C Nastran common blocks
COMMON /CONDAD/ PI

C Local data
DATA NAM/4HBRTU,4HGD /
C ======================================================================

DIM1 = DIMI(1)
DIM2 = DIMI(2)
PDIM13 = PI*DIM1**3
PDIM23 = PI*DIM2**3
C
ANONL(1,1) = 2*PI*DIM1
ANONL(1,2) = -2*PI*DIM2
ANONL(2,1) = PDIM13
Main Index Main Index
585 APPENDIX B
Adding Your Own Beam Cross-Section Library
ANONL(2,2) = -PDIM23
ANONL(3,1) = PDIM13
ANONL(3,2) = -PDIM23
ANONL(4,1) = 2*PDIM13
ANONL(4,2) = -2*PDIM23

C
RETURN
END
BMTUGD Subroutine
BMTUGD is a BEAM example routine that shows how to define the nonlinear
gradients of the beam TUBE section as a functional inner and outer radii of the tube.
Listing B-15 BMTUGD Subroutine
SUBROUTINE BMTUGD(DIMI,NDIMI,ANONL,NPROP,ERROR)
C
C ======================================================================
C Purpose
C To get the nonlinear factors of sensitivities for TUBE section
C
C Arguments
C
C DIMI input double Array of EPT records (Dimi+NSM)
C NDIMI input integer Number of dimensions (Plus NSM)
C ANONL output double Array providing the nonlinear factors
C for sensitivity relationships
C NPROP input integer Number of properties in PBEAM entries
C ERROR output integer Type of error
C
C Method
C Simply calculates the nonlinear factors
C Called by
C MSBMGD
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER NDIMI,NPROP,ERROR
DOUBLE PRECISION ANONL(NPROP,NDIMI), DIMI(11 * NDIMI)

C Local variables
INTEGER NAM(2)

C Nastran common blocks
COMMON /CONDAD/ PI

C Local data
Main Index Main Index


586
DATA NAM/4HBMRD,4HGD /
C ======================================================================

C Tube dimensions for the beam:
C 1 - s0
C 2 - x/xb
C 3 - outer radius
C 4 - inner radius
C 5 - nsm
C items 1 through 5 repeat for 10 additional stations

ndims = 5
do 20 isect = 0, 10
irows = isect * 16 + 1
idim = isect * ndims
idim3 = idim + 3
idim4 = idim + 4
C special handling puts end b data
if ( isect .lt. 10 .and. dimi(idim3 + ndims) .eq. 0 ) then
irows = 161
endif
C
DIM1 = DIMI(idim3)
DIM2 = DIMI(idim4)
PDIM13 = PI*DIM1**3
PDIM23 = PI*DIM2**3
C
ANONL(irows+ 3,3) = 2*PI*DIM1
ANONL(irows+ 3,4) = -2*PI*DIM2
ANONL(irows+ 4,3) = PDIM13
ANONL(irows+ 4,4) = -PDIM23
ANONL(irows+ 5,3) = PDIM13
ANONL(irows+ 5,4) = -PDIM23
ANONL(irows+ 7,3) = 2*PDIM13
ANONL(irows+ 7,4) = -2*PDIM23

C
if ( irows .eq. 161) go to 30
20 continue
30 continue
C Call mprtd ( nonlin,anonl,nprop,ndimi )

C
RETURN
END
Main Index Main Index
587 APPENDIX B
Adding Your Own Beam Cross-Section Library
B.8 BSBRCD Subroutine
The BSBRCD subroutine allows you to place constraints on values the beam
dimensions can take during a design task. It is not needed unless optimization is used
and, even then, is available only to impose conditions on the dimensions to keep the
optimization process from selecting physically meaningless dimensions.
For example, the optimizer might select a TUBE design with the inner radius greater
than the outer radius because this allows for a negative are and therefore a negative
weight (something a weight minimization algorithm loves!). These constraints are not
the same as the PMIN and PMAX property limits that are imposed on the DVPREL1
entry. Instead, these are constraints that occur between or among section dimensions.
A DRESP2 entry could be used to develop the same design constraints, but the
subroutine reduces the burden on the user interface, the primary goal of the beam
library project. The calling tree used for the MSC.Nastran standard library is shown in
Figure B-5.
Figure B-5 Calling Tree to Generate Beam Dimension Constraints
BSBRCD call the bar evaluator routine MSBRCD or the beam evaluator routine
MSBMCD. The evaluator in turn calls the routines for each section that requires
constraints. The routines are named BRxxCD or BMxxCD, where xx is a two-letter
identifier for the section. For example, the routine for the TUBE section is called
BRTUCD. Listings for BSBRCD, MSBRCD, BRTUCD and BMTUCD are given in
Listing B-16, Listing B-17, Listing B-18, and Listing B-19, respectively. MSBMCD is
not shown, but can be developed by calling the BMxxCD subroutines in place of
BRxxCD of MSBRCD.
BRTUCD
BRBXCD
MSBRCD
BMTUCD BMBXCD
BSBRCD
MSBMCD
OTHER
SECTIONS
Main Index Main Index


588
Listing B-16 BSBRCD Subroutine
SUBROUTINE BSBRCD(GRPID,ENTYP,TYPE,AFACT,NCONST,NDIMI,ERROR)
C
C ======================================================================
C Purpose
C To get constraint information for default types
C
C Arguments
C
C GRPID input integer ID of the group name
C ENTYP input integer 1: PBARL, 2: PBEAML
C TYPE input character*8 Section type
C NCONST input integer Number of constraints
C for the section type
C NDIMI input integer Number of dimensions
C for the section type
C AFACT output double The factor for the NDIMI dimension in
C the constraint relation. Dimensions are
C NCONST by NDIMI.
C ERROR output integer type of error
C
C Method
C Simply transfers control based on PBARL or PBEAML entries
C
C Called by
C BCBRCD
C
C Calls
C MSBRCD
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
CHARACTER*8 TYPE
INTEGER GRPID,ENTYP,NCONST,NDIMI,ERROR
DOUBLE PRECISION AFACT(NCONST,NDIMI)

C Local Variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBSBR,4HCD /
C ======================================================================
C
GRPID = 1
IF(ENTYP.eq.1) THEN
CALL MSBRCD(TYPE,AFACT,NCONST,NDIMI,ERROR)
else
CALL MSBMCD(TYPE,AFACT,NCONST,NDIMI,ERROR)
Main Index Main Index
589 APPENDIX B
Adding Your Own Beam Cross-Section Library
END IF
C ----------------------------------------------------------------------
RETURN
END
MSBRCD Subroutine
MSBRCD is a brancher routine for providing information on the calculation of
gradients for each of the bar types. You may rename this routine as you like or move
its function to BSBRCD. MSBRCD calls the BRxxCD routines, where xx is the two-
letter keyword for various section types.
Listing B-17 MSBRCD Subroutine
SUBROUTINE MSBRCD(TYPE,AFACT,NCONST,NDIMI,ERROR)
C
C ======================================================================
C Purpose
C To get constraint information for PBARL types
C
C Arguments
C
C TYPE input character*8 Section type
C AFACT output double The factor for the NDIMI dimension in
C the constraint relation. Dimensions are
C NCONST by NDIMI.
C NCONST input integer Number of constraints
C for the section type
C NDIMI input integer Number of dimensions
C for the section type
C ERROR output integer type of error
C
C Method
C Simply transfers control based on group name
C
C Called by
C BSBRCT
C
C Calls
C BRTUCD, BRBXCD, BRIICD, BRTTCD, BRLLCD, BRCHCD, BRCRCD,
C BRHHCD, BRT1CD, BRI1CD, BRC1CD, BRZZCD, BRC2CD, BRT2CD,
C BRB1CD, BRHXCD, BRHTCD, BRCLCD,
C ZEROD (Nastran utility)
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
CHARACTER*8 TYPE
INTEGER NCONST,NDIMI,ERROR
Main Index Main Index


590
DOUBLE PRECISION AFACT(NCONST,NDIMI)

C Local Variables
INTEGER NAM(2)

C Local data
DATA NAM/4HMSBR,4HCD /
C ======================================================================

CALL ZEROD( AFACT(1,1), NCONST*NDIMI )
ERROR = 0

IF(TYPE.EQ.TUBE) THEN
CALL BRTUCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.BOX) THEN
CALL BRBXCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.I) THEN
CALL BRIICD(AFACT,NCONST,NDIMI,ERROR)
c10939ELSE IF(TYPE.EQ.L) THEN
c10939 CALL BRLLCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.T) THEN
CALL BRTTCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.CHAN) THEN
CALL BRCHCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.CROSS) THEN
CALL BRCRCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.H) THEN
CALL BRHHCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.T1) THEN
CALL BRT1CD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.I1) THEN
CALL BRI1CD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.CHAN1) THEN
CALL BRC1CD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.Z) THEN
CALL BRZZCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.CHAN2) THEN
CALL BRC2CD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.T2) THEN
CALL BRT2CD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.BOX1) THEN
CALL BRB1CD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.HEXA) THEN
CALL BRHXCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.HAT) THEN
CALL BRHTCD(AFACT,NCONST,NDIMI,ERROR)
ELSE IF(TYPE.EQ.HAT1) THEN
CALL BRCLCD(AFACT,NCONST,NDIMI,ERROR)
ELSE
ERROR = 5500
END IF
Main Index Main Index
591 APPENDIX B
Adding Your Own Beam Cross-Section Library

C
RETURN
END
BRTUCD Subroutine
BRTUCD is an example routine that shows how to define the constraints for a bar
section. This example routine is for the TUBE section and imposes a single constraint
that -DIM1 + DIM2 < 0.0, where DIM1 is the outer radius and DIM2 is the inner radius
of the tube. The constraints should always be specified so that the specified linear
combination of dimensions is less or equal to zero when the constraint is satisfied.
Listing B-18 BRTUCD Subroutine
SUBROUTINE BRTUCD(AFACT,NCONST,NDIMI,ERROR)
C ======================================================================
C Purpose
C To get constraint information for TUBE type
C
C Arguments
C
C AFACT output double The factor for the NDIMI dimension in
C the constraint relation. Dimensions are
C NCONST by NDIMI.
C NCONST input integer Number of constraints
C for the section type
C NDIMI input integer Number of dimensions
C for the section type
C ERROR output integer type of error
C
C Method
C Simply transfers constraint information
C
C Called by
C MSBRCD
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
INTEGER NCONST,NDIMI,ERROR
DOUBLE PRECISION AFACT(NCONST,NDIMI)

C Local Variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBRTU,4HCD /
C ======================================================================
Main Index Main Index


592
C
IF(NCONST.NE.1) THEN
ERROR = 5502
RETURN
END IF
C
AFACT(1,1) = -1.0D0
AFACT(1,2) = 1.0D0

C
RETURN
END
BMTUCD Subroutine
BMTUCD is an example routine that shows how to define the constraints for a beam
section This example routine is for the TUBE section and imposes a single constraint
that -DIM1 + DIM2 < 0.0, where DIM1 is the outer radius and DIM2 is the inner radius
of the tube. The constraints should always be specified so that the specified linear
combination of dimensions is less or equal to zero when the constraint is satisfied.
Listing B-19 BMTUCD Subroutine
SUBROUTINE BMTUCD(AFACT,NCONST,NDIMI,ERROR)
C
C ======================================================================
C Purpose
C To get constraint information for TUBE type
C
C Arguments
C
C AFACT output double The factor for the NDIMI dimension in
C the constraint relation. Dimensions are
C NCONST by NDIMI.
C NCONST input integer Number of constraints
C for the section type
C NDIMI input integer Number of dimensions
C for the section type
C ERROR output integer type of error
C
C Method
C Simply transfers constraint information
C
C Called by
C MSBMCD
C ----------------------------------------------------------------------
IMPLICIT INTEGER (I-N)
IMPLICIT DOUBLE PRECISION (A-H,O-Z)

C Calling sequence arguments
Main Index Main Index
593 APPENDIX B
Adding Your Own Beam Cross-Section Library
INTEGER NCONST,NDIMI,ERROR
DOUBLE PRECISION AFACT(NCONST,NDIMI)

C Local Variables
INTEGER NAM(2)

C Local data
DATA NAM/4HBMTU,4HCD /
C ======================================================================
C
IF(NCONST.NE.1) THEN
ERROR = 5502
RETURN
END IF
C
AFACT(1,3) = -1.0D0
AFACT(1,4) = 1.0D0

C
RETURN
END
Main Index Main Index


594
B.9 BSMSG Subroutine
The Error handling is performed by a subroutine called bsmsg.f. This routine has the
following parameters.
SUBROUTINE BSMSG(GRPID,ERRCOD,MXLEN,Z,ERROR)
C ======================================================================
C Purpose
C To handle the error message of MSCBML0 group
C ----------------------------------------------------------------------
C Arguments
C
C GRPID input int ID of group or group name
C ERRCOD input int Error message number if any found
C MXLEN output int Maximum length of the message that can be passed
C Z output char Array to contain the message return
C ERROR output int The code returned indicates the type of error
C
C Reassigned
C
C Called by
C BCMSG routine
C-----------------------------------------------------------------------
The purpose of this subroutine is to return an error String associated with an error
code. The error codes are to be returned by the other 7 BS... routines and, as such,
the BSMSG routine is used as the repository of all of the error messages for the Beam
Library applications.
For example, suppose that BSBRPD application returns an error code of 5103 when
a certain error condition occurs. Then, the Beam Library Client routines will expect
that there will be String returned from the BSMSG routine, which corresponds to
this error code. These error messages will be printed in the *.f06 file to guide the user
as to what the error could have been and how to fix it. The string is limited to be no
more than 139 characters.
The following is an example of BSMSG code construct:
IF(ERRCOD .EQ. 5104) THEN
Z(1:MXLEN) = 'this is a User specified error message...'//
& ' Message can be up to 139 character long..'
ERROR = ERRCOD
ELSE IF ...
...
...
END
Main Index Main Index
595 APPENDIX B
Adding Your Own Beam Cross-Section Library
We recommend that you use the example Beam Server files as a template to generate
your BSMSG routines.
Main Index Main Index


596
B.10 Linking Your Library to MSC.Nastran
Once you have created the eight BS... routines, these routines may be linked with
MSC.Nastran Beam Server Library to build a beam server executable. An example
user-defined beam server has been delivered with Version 69 of MSC.Nastran. It is
highly recommended that you study, build and use the example Beam Server before
you build your own version of the beam server.
The MSC.Nastran special library contains a main routine as well as the
communications routines that allow MSC.Nastran to communicate with the user-
defined beam server.
The MSC.Nastran user may connect up to 10 beam servers in a single job execution.
This connection is made using the concept of evaluator groups described in the
remainder of this section. For each group, the user specifies on the PBEAML/ PBARL
entry referring to an external beam server, MSC.Nastran will start and communicate
with the beam server.
You may define as many beam evaluators as required using the CONNECT FMS
commands. Only 10 of these evaluators, however, may be referenced in the groups on
the PBARL or PBEAML Bulk Data entries.
The PBARL/ PBEAML entries specifies a GROUP name on the fourth field of the
first entry. This group name is associated with an Evaluator class using a Connect
command in the FMS section. Finally, the Evaluator class is associated with an
executable using the MSC.Nastran configuration file specified via the gmconn
keyword on the nastran command line.
The following example shows the mappings mentioned:
1. Group is referenced on the PBARL/ PBEAML entry (or entries).
PBARL,39,6,LOCSERV,I_SECTION (specify the Group name).
2. The Group is associated with an Evaluator class.
CONNECT,BEAMEVAL,LOCSERV,EXTBML (Associate the Group name
with an Evaluator class.)
3. The external evaluator connection file associates the Evaluator class with
a server executable. The following statement must be specified in the
connection file:
EXTBML,-,beam_server_pathname
Main Index Main Index
597 APPENDIX B
Adding Your Own Beam Cross-Section Library
4. Refer to the external evaluator connection file on the command line using the
gmconn keyword.
nastran myjob...gmconn=external_evaluator_pathname
5. In the example Beam Library section note that the standard input
(FORTRAN unit 5) or standard output (FORTRAN unit 6) are not used as
these I/ O channels are reserved by the Inter-Process Communications (IPC)
subsystem.
Main Index Main Index


598
B.11 Example of Building and Linking a Beam Server
As an example of building and linking a beam server executable, the sample beam
server will be modified. Complete instructions on building and using a beam server
are provided in the MSC.Nastran Installation and Operations Guide for each release of
MSC.Nastran.
Make a copy of the beam server sample source.
Edit the source for the BRTUPD subroutine; this routine describes the
equations that convert the PBARL dimensions into the standard PBAR
dimensions for a tube cross section.
Add an extra multiplication of 3.0 to the DIMO(2) equation to increase the
calculated moments of inertia.
Since the formulation of this bar section has been changed, the sensitivities for
optimization will also change. Rather than calculate what the new sensitivities should
be, the MSC.Nastran can calculate them using central differencing techniques. To
permit this, edit the source file for the BRTUID subroutine and change all occurrences
of SENTYP = 2 to SENTYP = 3.
Build your new beam server using the instructions detailed in the Installation and
Operations Guide for your system. Once you have built the beam server executable,
you must create an external evaluator connection to point to your executable.
Typically, this file would be kept in the users home directory, but for this example it
will remain in the current directory. Edit the new file bmconfig.fil. Put the following
line in the file:
LOCBMLS,-,pathname
where LOCBMLS is the evaluator referenced in the SAMPLE data file included with
the beam server. Remember, this file can contain references to any number of beam
servers.
To run the sample job, type in the following command:
nastran sample scr=yes bat=no gmconn=bmconfig.fil
Common problems which may occur when attempting to run an external beam library
job are generally indicated in the F06 by USER FATAL MESSAGE 6498. If this message
includes the text No such group defined, the PBARL/ PBEAML selected a group not
Main Index Main Index
599 APPENDIX B
Adding Your Own Beam Cross-Section Library
defined on a CONNECT entry. If UFM 6498 includes the text No such evaluator
class, either the gmconn keyword was not specified or the CONNECT entry
selected an evaluator not defined in the configuration file.
If the job was successful, you can look at the Design Variable History and see that the
results for the variable mytubeor are different than the results for tubeor. These
variables refer to the other radius of tube sections from equivalent models. One model
used the MSC.Nastran tube section while the other used the tube section in your
modified beam server.
Main Index Main Index


600
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide
APPENDIX
C
MSC.Patran Support for Design Sensitivity
and Optimization
I Tools Concepts and Definitions
I The Tools Menu
I Tools Commands
I Submitting a Job from MSC.Patran for SOL 200
I Supported OUTPUT2 Result and Model Quantities
Main Index Main Index


602
The material in this appendix has been taken almost directly from the MSC.Patran
documentation; the source of the documents is contained in the following table.
Please consult the MSC.Patran documentation for the most complete and up-to-date
description of this capability. These fragments are included here for convenience and
to provide an introduction to the MSC.Patran support for design sensitivity and
optimization.
Table C-1 MSC.Patran Documentation Sources
Appendix
Section
MSC.Patran
Document
Chapter/
Section
D.1 Basic Functions 10.1
D.2 Basic Functions 10.2
D.3 Basic Functions 10.3
D.4 Preference Guides
MSC.Nastran
4.4
Main Index Main Index
603 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
C.1 Tools Concepts and Definitions
The Tools menu commands in MSC.Patran provide easy access to a number of self-
contained add-on MSC.Patran features, as well as to some specialized application
modules. Two of these commands that are related to Design Sensitivity and
Optimization are: Model Variables and Design Study.
Design Studies and Design Optimization
Finite element analysis is seldom conducted as a single-run operation; frequently the
process goes through several iterations. In each iteration the model may be
tweaked, some dimensions or other properties of the model modified, and the
analysis repeated until an optimal design is attained.
As a simple example, you may run a linear static plane stress problem with a certain
thickness assigned to the elements. If the stress and displacement outcomes are well
within the acceptable range, you may decide to reduce the element thickness, thereby
decreasing the weight of the object, and run the analysis again. You may continue
refining the model through several iterations, until the weight reaches a desirable
minimized value without compromising the stress or other criteria.
This procedure can be automated through a series of design studies and, ultimately,
design optimization.
Design Studies
A design study is a named event in which you specify the following:
the objective of the design study and of optimization -- what do you want to
achieve through multiple iterations of the analysis process? Although there
may be a number of possible objectives, in the majority of structural analysis
problems the objective of optimization is to minimize the weight of the
model.
the parameters of the model -- which dimensions or properties may vary in
order to achieve the optimal design?
the constraints placed on the design study -- what condition is a limiting
factor in the optimization process? For example, in the plane stress problem
cited before, as the plate thickness is reduced to decrease the weight, the
thickness value must not become so small as to cause the plate to deform
beyond a given acceptable limit.
Main Index Main Index


604
Variables
Variables are those parameters, or properties, whose magnitude will be modified in
the process of studying the solutions that can improve the design. The variable may
be some dimension, an element property (e.g., plate thickness, beam cross-section), or
a material property. To perform studies for the purpose of improving and optimizing
a design through iterative solutions, you must parameterize the model, that is, identify
and label variables and set up possible relationships between them.
When you define a model variable, a corresponding field may be created as well. This
field is a linear function of the variable and it becomes available throughout
MSC.Patran. Thus, if desired, it can be used to define additional model properties.
Any changes made to the variable will also change the properties dependent on the
field.
Main Index Main Index
605 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
C.2 The Tools Menu
The commands of the Tools menu invoke various self-contained add-on MSC.Patran
features, as well as the specialized application modules.
Menu Conventions
A menu item with ellipses (...) attached to it calls up an additional dialog box in which
you enter further data.
The Tools menu functionalities are described below in the order in which they appear.
List Creates and manipulates lists with which
you can reference a number of entities for
input in commands that accommodate
multiple entity selection.
Mass Properties Calculates mass properties either of entire
geometric and finite element models or of
their subregions. Displays graphic symbols
related to mass properties and outputs
report files and summary spreadsheets.
Beam Library Creates and manages dimensioned cross
section shapes to be associated with beam
elements. Calculates section properties for
any selected beam cross section.
Tools
List
Mass Properties...
Beam Library...
Model Variables...
Design Study...
Rebar Definitions...
MSC.Fatigue...
Laminate Modeler
Analysis Manager...
Design Optimization Related
Main Index Main Index


606
Model Variables Creates and manages variable parameters
to be used in design studies and design
optimization.
Design
Optimization
Related
Design Study Creates and manages design study cases in
which the functions necessary for design
optimization are defined.
Rebar Designer Creates and displays Abaqus beam shapes
in MSC.Patran.
MSC.Fatigue Invokes the MSC.Fatigue application
module.
MSC.Laminate
Modeler
Implements the MSC.Laminate Modeler
application.
MSC.Patran
Analysis Manager
Calls up the MSC.Patran Analysis
Manager.
Main Index Main Index
607 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
C.3 Tools Commands
The following is a detailed description of the commands and dialog boxes that are
referenced in the Tools menu for Design Optimization.
Tools>Model Variables
Model variables provide the foundation for design studies. They are the parameters,
or properties, whose magnitude will be modified in the process of performing
iterative solutions. A variable may be a dimension value, an element property (e.g.,
plate thickness), or a material property. Element property in MSC.Patran can include
what Table 2-3 on page 43 refers to as connectivity properties.
Creating Model Variables
Variables defined for a geometric or finite element model fall into the following
categories:
Value -- a non-specific attribute value associated with the model;
for example, a dimension.
Property -- an element property; for example, element thickness.
Material -- a material property; for example, density.
Beam dimension -- a selected dimension in a previously defined beam cross
section; for example, the inner diameter of a tube.
Note: It is also possible to specify the final three categories as being designed on
the form that is used to specify the property, material, or beam dimension.
Details on this alternative technique are given following the description of
the Element Properties as Variables on page 609.
Main Index Main Index


608
Values as Variables
Action Lists the action that can be performed. In addition
to creating new variables, you can also show
(verify), modify, and delete existing variables.
Existing Variables Displays the names of previously defined variables.
Variable Name Names the new variable. The unique name may be a letter, a
symbol, or a text string, up to 32 characters long.
Description Provides a space where you can describe the nature of the
variable. This optional entry is especially helpful if you used
an abbreviation or symbol for the name of the variable and
may not remember in the future what the intent was (for
example, which dimension was designated as L1?).
Analysis Value Shows the initial value of the variable. Enter a real, scalar
number.
Create
Referencing Field
If ON, creates a field for the model variable.
Main Index Main Index
609 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Element Properties as Variables
Existing Variables Displays the names of previously defined variables.
Variable Name Names the new variable. The unique name may be a letter, a
symbol, or a text string, up to 32 characters long.
Description Provides a space where you can describe the nature of the
variable. If no description is entered, MSC.Patran will create a
description; it specifies the names of the Property Set and
Property Name from which the variable was created.
Main Index Main Index


610
Note that this technique for specifying the element property as a variable creates a
one-to-one relationship between the model variable and the property. There is an
alternative method of linking the design variables to the properties that provides the
full capability of the DVPREL1 entry (see Relating Design Variables to Properties
(Ch. 3 p. 162). The alternative method entails the following steps:
1. Create one or more variables using the Values as Variables form on page 608.
2. Create an additional field that provides the relationship between the design
variable(s) and the property (e.g., ).
3. On the relevant property form, invoke the new field for the real scalar
property value.
Analysis Value Shows the initial value of the variable. Enter a real, scalar
number.
Dimension
Type
These two entries act together as a filter that facilitates finding
the element property that will be designated as the variable.
The search will be limited to the properties associated with
those elements for which the named dimensionality and type
apply.
Select Property Set Lists the names of the Property Sets defined for the designated
element type. Pick the set that contains the property you wish
to parameterize.
Select Psets by
Elements
Selects a Property Set associated with an element you pick on
the screen (this can be helpful if you do not remember the
pairing of property sets and elements).
Select Property
Name
Displays a specific property contained in the named Property
Set.
PROP 0.5 X1 2.0 X2 + + =
Main Index Main Index
611 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Beam Dimensions as Variables
Existing Variables Displays the names of previously defined variables.
Variable Name Names the new variable. The unique name may be a letter, a
symbol, or a text string, up to 32 characters long.
Description Provides a space where you can describe the nature of the
variable. If no description is entered, MSC.Patran will create a
description; it will identify the name of the element property
set and specific element property (e.g., Thickness) from which
the variable was created.
Analysis Value Shows the initial value of the variable. Enter a real, scalar
number.
Main Index Main Index


612
See the note at the end of the Element Properties as Variables section on page 609 for
an alternative way of specifying beam dimensions that enables specifying the beam
dimensions as a linear function of one or more variables.
Type This entry acts as a filter that facilitates finding the beam
section for which the variable is applied. The beam library
will be searched for the specified beam cross section type.
Select Beam
Section
Lists the names of previously defined beam cross sections.
Pick the one that contains the property you wish to
parameterize.
Select Beams
by Elements
Selects a beam cross section associated with an element you
pick on the screen (this can be helpful if several beam
elements exist each with their own cross sections and you do
not remember which name goes with which element).
Select Beam
Dimension
Lists all dimensions pertaining to the named beam section.
Select a beam dimension which will become the variable.
Main Index Main Index
613 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Material Properties as Variables
Existing Variables Displays the names of previously defined variables.
Variable Name Names the new variable. The unique name may be a letter, a
symbol, or a text string, up to 32 characters long.
Description Provides a space where you can describe the nature of the
variable. If no description is entered, MSC.Patran will create a
description; it will identify the name of the material property
set and specific element property (e.g., Density) from which
the variable was created.
Main Index Main Index


614
See the note at the end of the Element Properties as Variables section on page 609 for
an alternative way of specifying material properties that enables specifying the
properties as a linear function of one or more variables.
Displaying Variables
With the Model Variables>Show command you can display and verify any or all
variables and their values.
Analysis Value Shows the initial value of the variable. Enter a real, scalar
number.
Category This entry acts as a filter that facilitates finding the property
that will be designated as the variable. The search will be
limited to the material to which the category type applies
(e.g., Isotropic).
Select Material Lists the names of previously defined materials. Pick the
material that contains the property you wish to parameterize.
Select Materials
by Elements
Selects a material associated with an element you pick on the
screen (this can be helpful if you do not remember which
material goes with which element).
Select Property
Name
Selects a specific property contained in the named material
definition. Also enters its current value into the Analysis
Value textbox.
Main Index Main Index
615 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
After you picked the variable(s) you want to verify, press Apply and the variable
information will be displayed in a spreadsheet.
Pick any of the properties and its description will appear in the lower left corner of the
spreadsheet.
Modifying Variables
In this dialog box you can modify the initial value of a variable as well as the
description attached to it. The names of variables cannot be changed.
Select Variables
to Show
Displays all previously defined variables. Pick one or several
to show in detail.
Select by Elements Selects the properties associated with elements you pick on
the screen.
Main Index Main Index


616
Deleting Variables
When you delete a model variable, all fields and other properties associated with it
will be deleted from the database.
If a finite element property is defined as a function of a deleted variable, the current
value will be used for the property entry rather than leaving the property blank.
Select Variable
to Modify
Lists all variables contained in the database. Pick the variable
you want to modify.
Variable Name Displays the name you selected. This name cannot be
changed.
Description Displays the original description that you can edit or replace.
Modified Value Enter the new value you want to assign to the variable. All
field values and relationships that depend on the value of this
parameter will be updated when the new value is applied.
Main Index Main Index
617 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Tools>Design Studies
Before you can initiate a design study, you have to:
consider the objective, or goal of the design study and of optimization.
Although there may be a number of possible objectives, in the majority of
structural analysis problems the objective of optimization is to minimize the
weight of the model. This is the only objective supported in MSC.Patran
parameterize the model. Define the variable dimensions or properties that
may change in order to achieve the optimal design.
decide the constraints to place on the design, that is, define the conditions that
are considered as a limiting factor in the optimization process.
Select Variables
to Delete
Lists all variables contained in the database. Pick one or more
variables that you want to delete.
Select by Elements Select an element whose property is a variable that you want
to delete.
Apply/Close Pick Apply to complete the deletion. Pick Close to cancel the
delete action and exit from the application.
Main Index Main Index


618
Creating Design Studies
Action Lists the action that can be performed.
Note that with Create you can also modify a design study.
Main Index Main Index
619 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Object Displays the object of the action.
Current Design
Study
Shows the name of the current design study.
Existing Design
Study
Displays the names of existing design studies. Pick the name
of an existing design study if you want to modify it.
Design Study Name Provides a textbox where you can enter a name for a new
design study. The name may consist of any number of alpha-
numeric characters. special symbols and spaces are not
acceptable.
Description Provides a textbox where you can enter a detailed description
of the design study.
Make Current Makes the selected design study current. Only those design
parameter values that originate in a current design study can
be used to update finite element model properties.
You can modify a design study even if it is not current (the
toggle is not on), but the parameter values of the model will
not be updated. Modifications to a design study that is not
current will not affect a current design study.
Design Study Setup
Design
Variables
Displays the Design Variables dialog box where you can
review and modify design variables (see Design Variables
on page 620).
Design
Objective
Displays the Design Objective dialog box (see Design
Objective on page 623).
Design
Constraints
Displays the Design Constraints dialog box where you can
create or modify design constraints (see Design Constraints
on page 624).
Apply/Close Saves or discards input. The design study will be saved only if
you press Apply.
Main Index Main Index


620
Design Variables
A design variable is a model variable that is used in a design study and in
optimization. When you press Design Variables in the Design Study dialog box, a
spreadsheet will display a summary of variables and their attributes. Because initially
every model variable is also considered a design variable, the spreadsheet will contain
all variables defined in a database.
With the exception of the parameter names and their default initial values, all other
data in the spreadsheet may be modified.
The first column contains the name of the variable. Subsequent column headings and
their contents are as follows:
Main Index Main Index
621 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Design Variable A YES entry in this column confirms that the model variable is
also a design variable. If a variable is not to be used in the
design study, you can deselect it by clicking on the word
YES. All variable information for this item will be removed,
indicating that the variable is not a design variable, even
though it is listed in the spreadsheet. To toggle back to YES,
click in the field again.
You can pick several consecutive items in this column at the
same time by dragging the cursor down the column.
A button will then appear at the top of the
spreadsheet to show that the current status is YES.
Click on YES, to display the choices, pick NO,
then press the Enter (or Return) key, and the items will be
deselected. The button also works in the reverse (NO to YES).
Analysis Value Displays the value you assigned to the variable when you
defined it. This value cannot be changed here.
Design Value Provides a field where the value of a design variable will be
placed, if you define one that is different from the value of the
model variable.
When you click in this field, a
textbox appears at the top of the
spreadsheet. Type a new value then
press Enter (or Return).
If you specify a new design value, the original lower and
upper boundary values will no longer be valid. Therefore, you
will see the following message:
Press Yes to acknowledge the warning. The new value will be
entered in the Design Value category and one or more of the
boundary values will change accordingly.
Main Index Main Index


622
Design Cycle Select
In this dialog box you can update the design variables in the study with the solution
values output by the solver and stored in the results database.
Lower Bound Displays the lowest value the design variable may assume
during the design study/ design optimization process. By
default, this boundary is set to 10% below the base value. To
change the boundary value, click in the appropriate field then
enter the new value in the Input New Value textbox.
Upper Bound Displays the highest value the design variable may assume
during the design study/ design optimization process. By
default, this boundary is set to 10% above the base value. To
change the boundary value, click in the appropriate field then
enter the new value in the Input New Value textbox.
Move Limit Specifies the change allowed in a design variable in each
design cycle. Default is 1.0, the equivalent of 100%.
Design Cycle Select Calls up the Design Cycle Select dialog box.
Defaults Resets all entries to their original value. Keeps the dialog box
on the screen.
OK/Cancel OK accepts all variables. Cancel closes the form.
Main Index Main Index
623 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
To apply the results of a particular cycle of an analysis run in the design study, pick
the name of the Run then pick a Design Cycle. The design variable values of that run
and cycle will be shown. Press OK to accept the new values. When the Design
Variables spreadsheet is redisplayed, the variables will be updated to the new values.
Design Objective
Design objective is the goal of the optimization process. Currently, the only available
option for design objective is the goal of minimizing the entire weight of the model.
Runs Displays the names of the result cases, or runs, that were
output by iteratively performing analysis with different
values of the variables.
Design Cycles Presents the list of design cycles (subcases).
Design Variables Shows the design variables.
Values Displays the values of the design variables for a selected
design cycle.
Main Index Main Index


624
At the present time (MSC.Patran 2001) none of the items in this dialog box are
selectable for a modified input.
Design Constraints
Design constraints provide certain restrictions, or limits, to ensure that as the
optimization process advances toward achieving the design objective, other design
conditions do not become compromised. As an example, you may decrease the cross-
sectional area of a beam to minimize its weight, but not to the extent where the axial
stress generated in the beam increases beyond a specified value.
The nature of a design constraint depends, primarily, on the type of analysis solution
you are performing. At the present time, the available options are linear static and
normal modes.
Constraints for Linear Static Solutions
In linear static analysis the constraint responses may be displacements, stresses,
strains, and forces.
Solution Displays Global as the only choice for the extent of
optimization. The objective of minimizing the weight extends
globally, that is, for the entire structure.
Response Displays Weight as the only choice for the goal of
optimization is minimizing the weight.
Existing Objectives Displays Total Weight; the predefined--and only--name of
the objective function.
Min/Max Displays minimize; the only action that the optimizer is
enabled to perform on the response entity.
Main Index Main Index
625 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Displacement as Constraint
Action Specifies Create or Delete. With the Create action you can also
modify existing constraints.
Solution Names the type of analysis you are running. The options are
Linear Static, or Normal Modes.
Main Index Main Index


626
Response Displays the type of attribute that may be
selected as a constraint.
Existing Constraints Lists the names of all constraints previously defined for the
current design study.
Constraint Name Names the current constraint. Provides a default name with a
prefix that is based on the response type (for displacements
the name will begin with DISP). If desired, you may replace
the default name with any other name.
To modify a constraint, pick its name here.
Constraint Region Identifies the domain over which the constraint will be
applied. The only applicable domain for displacements is the
finite element model, more specifically, the nodes in the
model.
Select Node Provides a textbox for node selection. Pick nodes individually,
screen-select a number of nodes, or type in node IDs.
Displacement
Component
Shows the components of the displacement constraint. Select a
translational (TX, TY, TZ) or rotational (RX, RY, RZ)
displacement component. Alternatively, you may select
Magnitude, the resultant displacement, that will
automatically create an MSC.Nastran DRESP2/ DEQATN
entry pair.
Lower Bound/
Upper Bound
Displays textboxes where you can enter the lowest and
highest acceptable values of the constraint.
The upper bound value must be higher than the lower bound
value.
Do not enter zero for either bound value. Either provide a
non-zero value or leave the field empty.
Main Index Main Index
627 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Stress as Constraint
Constraint Name Displays the default name which begins with the prefix
STRESS.
Constraint Region Provides two choices for the domain over which the
constraint will be applied. You may select the entire finite
element model or restrict the selection to certain elements for
which a set of properties has been previously defined.
Therefore the options are FEM and PSET (property set).
If you pick PSET, a databox will display the names of existing
selectable properties. Alternately, you can pick the elements
to which the property has been assigned.
Main Index Main Index


628
Element Dimension/
Element Type
Displays the options for element dimensionality and type.
Pick the attributes that apply to the elements in your model.
Whatever you select for dimension (1D, 2D, or 3D) will
dictate what the selections are in the element type box.
The nature of further options available in the remainder of
the dialog box depends on what you select for these two
attributes.
Select Finite Element Pick the finite elements for which the constraints apply.
Stress Component Identifies the stress component whose value acts as the
constraint. The contents of the options form will vary,
depending on the selected element dimensionality and type.
Examples of Stress Component Options:
Beam (1D) Plate/ Shell (2D) Solid Element (3D)
Main Index Main Index
629 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Location on
Element
Specifies the location of the stress constraint on the element.
1D beam, bar, or bend elements -- either end of the
element.
Other 1D elements -- option is not applicable
2D plates and shells:
3D solid elements -- corners or center.
Lower Bound/
Upper Bound
Displays textboxes where you can enter the lowest and
highest acceptable values of the constraint.Keep in mind that
the upper bound value must be higher than the lower bound
value.
Main Index Main Index


630
Strain as Constraint
All items in this dialog box are the same and require the same input as those in the
dialog box for stress design constraints.
Main Index Main Index
631 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Force as Constraint
Main Index Main Index


632
Element Dimension/
Element Type
Displays the options for element dimensionality and type.
Pick these attributes as they apply to the elements in your
model.
Whatever you select for dimension (1D or 2D) will dictate
what the selections are in the element type box. The nature
of further options available in the remainder of the dialog
box depends on what you select for these two attributes.
Force Component Identifies the force component whose value acts as the
constraint. The contents of the options form will vary,
depending on the selected element dimensionality and type.
Examples of Force Component Options:
Location
on Element
Specifies the location of the force constraint on the element.
1D beam, bar, or bend elements -- either end of the
element.
Other 1D elements -- option is not applicable
2D plates and shells -- element corners or center.
2D solid elements -- element corners or edges.
Lower Bound/
Upper Bound
Displays textboxes where you can enter the lowest and
highest acceptable values of the constraint.
Keep in mind that the upper bound value must be higher
than the lower bound value.
Beam (1D) Plate/ Shell (2D) Shear Panel (2D)
Main Index Main Index
633 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Composite Stress/Composite Strain as Constraint
Stress/Strain
Component
Identifies the stress or strain component whose value acts as
the constraint.
Main Index Main Index


634
Constraints for Normal Modes
In modal analysis the only possible constraint response is frequency.
Ply Number Specifies the layer on the PCOMP entry to which the
constraint is to be applied.
Lower Bound/
Upper Bound
Displays textboxes where you can enter the lowest and
highest acceptable values of the constraint. Keep in mind that
the upper bound value must be higher than the lower bound
value.
Frequency Mode
Number
The number of the mode shape that supplies the frequency
constraint.
Lower Bound/
Upper Bound
For modal analysis the unit of boundary values is Hertz
(Hz.). Keep in mind that the upper bound value must be
higher than the lower bound value.
Even if the design goal is to drive the design exactly to a
certain frequency, do not enter the same upper and lower
bounds. Make them slightly different to allow the
optimization solver some freedom.
Main Index Main Index
635 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
C.4 Submitting a Job from MSC.Patran for SOL 200
An optimization job is submitted from MSC.Patran by selecting the Analysis toggle on
the main menu and selecting Optimize the Action.
Analysis
Action: Optimize
Object: Entire Model
Method: Full Run
Code:
Type:
Study:
MSC.Nastran
Structural
Available Jobs
Job Name
test3
Job Description
MSC.Nastran job
Translation Parameters...
Optimization Parameters...
Direct Text Input...
Subcases...
Subcase Select...
Analysis Manager...
Apply
created on 09-Oct-97
List of already existing jobs.
Sets up the menu for Optimization Analysis.
Brings up the Optimization Parameters menu for Glob
Optimization Analysis Parameters.
Brings up the Subcases form that creates a subcase w
solution type can be changed at the time of subcase
creation.
Brings up the Subcase Select Optimization form to se
subcases with different solution sequences.
Main Index Main Index


636
Optimization Parameters
This form appears when the Optimization Parameters button is selected on the
Analysis/ Optimize form. It is used to define optimization parameter for the job.
Main Index Main Index
637 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Optimization Parameters
Automatic Constraints. Use Shell Normals.
Mass Calculation: Lumped Data Deck Echo: None
Tolerance Angle = 20. Plate Rz Stiffness Factor = 0.0
Maximum Printed Lines = 999999999
Maximum Run Time = 600
Wt.-Mass Conversion = 1.0 Node i.d. for Wt. Gener. =
Maximum Number of Design cycles (DESMAX) = 5
Design Data to be Printed (P2):
Objective and design variables. Properties.
Constraints. Responses.
Print Design Data (P1) every n-th cycle where n = 0
Print Analysis Results (NASPRT) every n-th cycle where n = 0
Relative Objective Convergence (CONV1) = 0.001
Absolute Objective Convergence (CONV2) = 1e-20
Relative Convergence on Design Variables (CONVDV) = 0.001
Relative Convergence on Properties (CONVPR) = 0.01
Fractional Property Change (DELP) = 0.2
Fractional Design Variable Change (DELX) = 1.0
Minimum Property Move Limit (DPMIN) = 0.01
Minimum Design Variable Move Limit (DXMIN) = 0.05
Apply Defaults Cancel
These are the criteria for the objective function, design variable, and design properties which an
optimizer uses to determine whether the design optimization process has converged or not.
Relaxing these criteria, in general, may reduce the number of design cycles for an optimization job
to terminate.
During any optimization design cycle, the change in the properties and design variables are limited to
maintain a good approximate model. Parameters DELP and DELX are used to specify such a move limit.
DPMIN and DXMIN are used to provide a minimum change to avoid numerical difficulties.
Parameter P2 controls
what type of design
data or optimization
results are printed in
an f06 file. Parameter
P1 controls how often
design data are
written. Default
(P1=0) prints initial
results and final
results (if an
optimization task is
performed).
Main Index Main Index


638
Subcases
This form appears when the Subcases button is selected on the Analysis/ Optimize
form.
Subcase Parameters
This form appears when the Subcase Parameters button is selected on the Subcases
form.
Subcases
Solution Type: 101 LINEAR STATIC
Available Subcases
Default
Subcase Name
Subcase Description
Available Load Cases
Default
Constraints in Current Subcase
Subcase Options
Subcase Parameters...
Output Requests...
Direct Text Input...
Select Superelements...
Apply Delete Cancel
List of the associated constraints with the selected
subcase.
Displays all the available subcase based on
the solution type selection.
The subcase name that is being created.
Subcase description which can be up
to 256 characters long.
List of available loadcases.
Displays the Subcase Parameters form. For more
information see Subcase Parameters.
Main Index Main Index
639 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
Subcase Select Optimize
This form appears when the Subcase Select button is selected on the
Analysis/ Optimize form. This form is used to select a sequence of subcases associated
with an optimization analysis job.
Subcase Parameters
Inertia Relief
Alternate Reduction
OK Cancel
Indicates that an alternate method of performing the static
condensation is desired. The PARAM, ALTRED,YES command is
included if selected and if Database Run is also selected.
Indicates that the inertia relief flags are to be set by including the PARAM, INREL,-1
command. This flag can only be chosen if Database Run is selected and Cyclic
Symmetry is disabled. If inertia relief is selected, a node-ID for weight generation must
be selected. A PARAM, GRDPNT and a SUPORT command will be written to the input
file using the same node-ID selected for weight generation. The SUPORT card will
specify all 6 degrees of freedom.
Main Index Main Index


640
Subcase Select
Current Job: test3
Solution Type: 101 LINEAR STATIC
Subcases Available:
Default
Select All Unselect All
Subcases Selected:
OK Clear Cancel
Used to filter the subcases by
their solution type
List of subcases based on the
solution type setting.
List of the selected
subcases.
101 Default
103 Default
N NN
Main Index Main Index
641 APPENDIX C
MSC.Patran Support for Design Sensitivity and Optimization
C.5 Supported OUTPUT2 Result and Model Quantities
The following table indicates results quantities that can be loaded into the MSC.Patran
database during results translation from MSC. Nastran and that are specific to design
sensitivity and optimization. The Primary and Secondary Labels are items selected from
the postprocessing menus. The Type indicates whether the results are Scalar, Vector,
or Tensor, and determines which postprocessing techniques are available to view the
results quantity. Data Block indicates which MSC. Nastran OUTPUT2 data block the
data comes from. The Description gives a brief discussion about the results quantity,
such as if it is only for certain element types, and what Output Request selection will
generate this data block. For design optimization, general results can be loaded as a
function of design cycle.
Primary Label Secondary Label Type Data Block Description
Shape Change None Vector GEOMIN In a shape optimization run,
this is the new shape
displayed as a deformation
of the original shape.
Active
Constraints
Element Stress Scalar R1TABRG Element stress
Element Strain Scalar R1TABRG Element strain
Element Force Scalar R1TABRG Element force
Element Ply Failure Scalar R1TABRG Element ply failure
Translational
Displacement
Vector R1TABRG Nodal translational
displacement
Rotational
Displacement
Vector R1TABRG Nodal rotational
displacement
Translational Velocity Vector R1TABRG Nodal translational velocity
Rotational Velocity Vector R1TABRG Nodal rotational velocity
Translational
Acceleration
Vector R1TABRG Nodal translational
acceleration
Rotational
Acceleration
Vector R1TABRG Nodal rotational
acceleration
Translational SPC Vector R1TABRG Nodal translational SPC
force
Rotational SPC Vector R1TABRG Nodal rotational SPC force
Main Index Main Index


642
Global Variables. In addition to standard results quantities, a number of Global
Variables can be created. This table outlines Global Variables that may be created
related to design sensitivity and optimization. Global Variables are results quantities
where one value is representative of the entire model.
XY Plots. For results from MSC. Nastran design optimization solution 200 runs,
three XY Plots are generated, but not posted, when the Read OUTPUT2 option is
selected:
1. Objective Function vs. Design Cycle.
2. Maximum Constraint Value vs. Design Cycle.
3. Design Variable vs. Design Cycle.
These plots can be viewed under the XY Plot option. When they are initially posted,
you will have to expand their windows to view them properly.
Label Type Data Block Description
Design Cycle S Oxxx Cycle number in an optimization run
(SOL 200).
Design Variable S DESTAB
HISADD
Design Variable for optimization
(Label from DESTAB, value from
HISADD).
Maximum
Constraint Value
S HISADD Maximum constraint value for
optimization.
Objective
Function
S HISADD Objective function for optimization.
Main Index Main Index
MSC.Nastran Design Sensitivity and Optimization Users Guide CNumerical Optimization
APPENDIX
D
Numerical Optimization
I Basic Optimization
I The Modified Feasible Direction Algorithm
I Sequential Linear Programming
I Sequential Quadratic Programming
I Summary
Main Index Main Index


644
The basic features of numerical optimizers were introduced in Getting Started in
Chapter 1. That general overview is probably sufficient in most situations to
appreciate the tasks performed by the optimizer, to understand how to pose the
design problem for effective solution, and to interpret most of the optimization-
related output.
This section completes the discussion begun in Getting Started in Chapter 1 and is
intended for the intermediate- to advanced-level reader who wishes to know more
about the implementation of the optimizer used in MSC.Nastran. This section has
largely been excerpted from the DOT, Design Optimization Tools Users Manual,
Version 5.0, from Vanderplaats Research & Development, Inc.
(http:/ / www.vrand.com).
The processes described here are a part of the code and are not subject to change by
the engineer. However, many of those methods use numerical constants in connection
with convergence detection, one-dimensional search control, and so on. Defaults are
provided for all of these internal optimizer parameters, but these may be overridden
using the DOPTPRM Bulk Data entry. This section presents the advanced-level
information necessary to make these override decisions. This appendix contains an
overview of the basics of mathematical programming as applied to optimization
tasks, followed by specifics on each of those algorithms that are used in DOT.
Main Index Main Index
645 APPENDIX D
Numerical Optimization
D.1 Basic Optimization
Getting Started in Chapter 1 introduced the Basic Optimization Problem
Statement. Recall that this provides the definition of the design variables, objective,
and constraints as
minimize objective Eq. D-1
subject to inequality constraints Eq. D-2
equality constraints Eq. D-3
side constraints Eq. D-4
where:
design variables Eq. D-5
It is within the design space defined by the above problem statement that the
optimizer searches for a best design.
Chapter 1 also introduced the analogy of a search along the side of a hill for the point
of lowest elevation to describe some of the basic processes used by gradient-based
numerical optimizers. At that time, we introduced the idea of a one-dimensional
search and the steepest descent algorithm as a particular method of establishing this
search direction. However, we also noted that this method, in addition to not being
particularly efficient, cannot address those situations in which one or more constraints
were active. In this section we discuss how these issues are resolved by the optimizer
in MSC.Nastran and move on to consider some of the other details of the search
process as well.
Recall that the design variable vector update can be written as
Eq. D-6
where is the initial vector of design variables, is the search vector, and is the
search parameter. Eq. D-6 represents a one-dimensional search since the update on
depends only on the single scalar parameter . is the value of that yields the
optimal design in the direction defined by . In other words, to continue with the hill
analogy, we have either encountered a fence or are at the bottom of a valley in the
direction of . Finding completes the first iteration in the design process.
F X ( )
g
j
X ( ) 0 j 1 n
g
, , =
h
k
X ( ) 0 k 1 n
h
, , = =
x
i
I
x
i
x
i
u
i 1 n , , =
X x
1
x
2
x
n
, , , { } =
X
1
X
0
=
*
S
1
+
X
0
S
1

X
1

*

S
S
1

*
Main Index Main Index


646
If we were at the bottom of a valley in the first search direction, we could just repeat
the process of finding a new steepest descent direction and keep moving. In practice,
there is a better choice of direction called a conjugate direction, which we will discuss
shortly. However, regardless of the search direction used, the concept is the same. We
find a search direction down the hill and move in that direction as far as possible.
Now, assume instead that we (or the optimizer) have encountered a fence. In order to
make any further improvement in the design, a new search direction

must be
found that continues to reduce our elevation on the hill but keeps us inside of the
fence. Here we seek a usable-feasible direction, in which a usable direction is one
that moves us downhill and a feasible direction is one that keeps us inside of the fence.
This situation is shown in Figure D-1. The mathematical definition of a usable search
direction is
Eq. D-7
Eq. D-7 is the scalar product (dot product) of the gradient of the objective function
with the search direction. The dot product is the magnitude of times the
magnitude of times the cosine of the angle between the two vectors. Thus, the cosine
of the angle determines the sign of the product since the magnitudes are positive
numbers. For the cosine to be zero or negative, the angle between the vectors must be
between 90 and 270 degrees. If the cosine is zero, the search direction is at an angle of
90 or 270 degrees from the gradient of the objective function. A move in this direction
follows a contour on the hill and (for a small move) does not increase or decrease the
function. If the cosine is -1.0, the direction is opposite to the direction of and is
the direction of steepest descent. Thus, we wish to find a search direction that makes
the left-hand side of Eq. D-7 as negative as possible. However, this direction must
remain within a critical constraint (fence). This is the feasibility requirement similar to
the usability requirement, but stated with respect to the constraint
Eq. D-8
S
2
F X ( )
T
S 0
F X ( )
S
F X ( )
g
j
X ( )
T
S 0
Main Index Main Index
647 APPENDIX D
Numerical Optimization
Figure D-1 Usable-Feasible Search Direction
Just as for the objective function, the angle between the search direction and the
gradient of the constraint must be between 90 and 270 degrees. If the angle is exactly
90 or 270 degrees, the search direction is tangent to the constraint boundary (fence).
To find the search direction that makes the greatest possible improvement in the
objective function but still follows or moves inside of the fences, we combine the
usability and feasibility requirements. This combination creates the following
suboptimization task: Find the components of the search direction in order to
minimize Eq. D-9
Usable-Feasible
Sector
Usable
Sector
Feasible
Sector
x
0
g
1
X
0
( )
F X
0
( )
S
F X ( ) constant =
g
2
X ( ) 0 =
g
1
X ( ) 0 =
x
1
x
2
0
S
S
F X ( )
T
S ( )
Main Index Main Index


648
subject to: Eq. D-10
Eq. D-11
where J is the set of constraints whose values are zero within some numerical
tolerance. This is the set of active constraints. The purpose of Eq. D-11 is simply to
prevent an unbounded solution to the problem defined by Eq. D-9 and Eq. D-10. In
the case of a simple two-variable problem, finding the appropriate search direction is
quite easy and may be done graphically. In the more general case where there are
numerous design variables as well as several active constraints, this becomes a
subproblem that is solved as part of the optimization. This problem is linear in
except for the quadratic constraint of Eq. D-11.
Assuming we can find a usable-feasible search direction, we can now search in this
direction until we can make no further improvement. The subproblem of finding a
new usable-feasible search direction is repeated and continues until no search
direction can be found that improves the design without violating one or more
constraints. We call this point the optimum. In the present example, we assumed the
initial design was feasible. In practice, we often start outside of one or more fences, or
from an initially infeasible design.
If this is the case, the optimizers first task is to search for a feasible design. Frequently,
the objective function must be increased in this process. For example, it may be
necessary to add structural mass in order to overcome element stress constraints.
However, the optimizer in MSC.Nastran will search for a feasible region, allowing for
the smallest possible increases in the objective. Having found the feasible domain (if
one exists), it can then proceed with minimizing the objective.
The question now arises: How do we know that we have reached the optimum? The
answer can be found in what are known as the Kuhn-Tucker conditions (see
References 4 and 5). In the case of an unconstrained problem (omitting Eq. D-2
through Eq. D-4), this is simply the condition where the gradient objective function
vanishes (i.e., equals zero). In the case of the constrained optimization problem
considered here, the conditions of optimality are more complex. In this case, the
governing equation is the stationary condition of the Lagrangian function:
Eq. D-12
g
j
X ( )
T
S 0
S
T
S 1
S
L X , ( ) F X ( ) =
j
g
j
X ( )
j 1 =
M

+
Main Index Main Index
649 APPENDIX D
Numerical Optimization
Eq. D-13
The Kuhn-Tucker conditions dictate that the Lagrangian function must have a
vanishing gradient at the optimum design, . However, we must also remember the
original optimization problem and the inequality condition of Eq. D-7. In addition, we
note that Eq. D-8 for equality constraints is restated in MSC.Nastran as a set of
inequality constraints. When all of these conditions are considered, they lead to the
statement of the Kuhn-Tucker necessary conditions for optimality:
Condition 1: is feasible; i.e., for all j. Eq. D-14
Condition 2: (the product of and equals zero) Eq. D-15
Condition 3: Eq. D-16
Eq. D-17
The physical interpretation of these conditions is that the sum of the gradient of the
objective and the scalars times the associated gradients of all active constraints must
vectorally add to zero. This is much like the statement for equilibrium where the sum
of all internal and external forces at any given point must vectorally add to zero.
Figure D-2 shows this situation for a simple two-variable function space with two
active constraints at the optimum. Eq. D-15 states that the corresponding Lagrange
multiplier is zero for all non-active constraints.

j
0
L X , ( )
X
X* j g
j
X* ( ) , 0

j
g
j
X* ( ) 0 =
j
g
j
X* ( )
F X* ( )
j
g
j
X* ( )
j 1 =
M

+ 0 =

j
0 j 1 2 M , , , =

j
Main Index Main Index


650
Figure D-2 Kuhn-Tucker Conditions at a Constrained Optimum
These are only necessary conditions and the definition here is actually a bit more
restrictive than required in some cases. However, it does provide the essential idea. In
practice, it is difficult to numerically find a design that precisely satisfies the Kuhn-
Tucker conditions. Also, numerous designs might actually satisfy these conditions
since there may be more than one constrained minimum. The importance of the Kuhn-
Tucker conditions is that an understanding of the necessary conditions for optimality
gives us some knowledge of what is needed to achieve an optimum.
The simple example of descending a hill by the approach outlined above is the
physical interpretation of a class of optimization methods known as feasible direction
techniques. A multitude of other numerical search algorithms is available to solve the
general optimization problem. Most of the more powerful methods update the design
using the aforementioned relation:
Eq. D-18
where:
= iteration number
= vector search direction
= scalar move parameter
x
2
g
2
X ( )
F X ( ) const =
F X* ( )
g
1
X ( )
x
1
X* g
2
X* ( )

2
g
2
X* ( )

1
g
1
X* ( )
g
1
X* ( )

1
g
1
X* ( )
2
g
2
X* ( ) +
X
q
X
q 1
= *S
q
+
q
S
*
Main Index Main Index
651 APPENDIX D
Numerical Optimization
Eq. D-18 is the same as Eq. D-6 except that we have replaced the superscripts with the
iteration counter q. Just as in the example of searching down the hill, the product
is the design modification at the current step. Note that this is very similar to the
traditional engineering approach of modifying a current best design instead of using
a completely new design. An initial design must be provided, but it need not
satisfy all of the constraints. (Indeed, one of the most powerful uses of optimization is
to find a feasible solution to a complicated problem).
In order to determine a search direction which improves the design, gradients of the
objective and critical constraints must be supplied. Ideally, these are computed
analytically or semi-analytically as is done in MSC.Nastran. This process dramatically
increases the size of the problem that can be efficiently solved. Finally, a one-
dimensional search is performed by trying several values of and interpolating for
the one that gives a minimum objective while satisfying the constraints. During this
process, the objective and the constraints must be repeatedly evaluated. Here the use
of approximation methods plays a major role because the evaluation of these
constraints would otherwise require a full finite element analysis. In MSC.Nastran, we
only evaluate the approximate functions, which is a relatively inexpensive process.
In the following sections we will outline the overall optimization process as
implemented in the DOT algorithm. The presentation here will be necessarily brief,
and the reader is referred to sources listed in Reference 15. for more details.
*S
X*
S

Main Index Main Index

652
D.2 The Modified Feasible Direction Algorithm
Now we turn to the actual task of solving the approximate problem. The first method
described here is referred to as the Modified Method of Feasible Directions. At this
point it is assumed that we are provided with an objective function and
constraints , as well as lower and upper bounds on the design
variables. Also, the gradients of the objective and constraints are available. Thus, we
are solving the general problem defined by Eq. D-6, Eq. D-7, and Eq. D-9, which are
repeated here for convenience:
Find the set of design variables that
minimize Eq. D-19
subject to Eq. D-20
Eq. D-21
At this point the optimizer does not know what kind of a problem it is solving. It is
simply minimizing a function subject to inequality constraints.
Given an initial x-vector , the design will be updated according to Eq. D-18, which
is also repeated here
Eq. D-22
The overall optimization process now proceeds in the following steps:
1. Start, .
2. .
3. Evaluate and where .
4. Identify the set of critical and near critical constraints J.
5. Calculate and for all .
6. Determine a usable-feasible search direction .
7. Perform a one-dimensional search to find .
8. Set .
9. Check for convergence to the optimum. If satisfied, exit. Otherwise, go to
Step 2.
F X ( )
g
j
X ( ) 0 j 1 2 n
g
, , , =
x
i
i 1 2 n , , , = ,
F X ( )
g
j
X ( ) 0 j 1 2 n
g
, , , =
x
i
L
x
i
x
i
U
i 1 2 n , , , =
X
0
X
q
X
q 1
=
*
S
q
+
q 0 X X = , =
q q = 1 +
F X ( ) g
j
X ( ) J 1 2 n
g
, , , =
F X ( ) g
j
X ( ) j J
S
q

*
X
q
X
q 1
=
*
S
q
+
Main Index Main Index
653 APPENDIX D
Numerical Optimization
The critical parts of the optimization task consist of:
1. finding a usable-feasible search direction, .
2. finding the scalar parameter that will minimize subject to
the constraints.
3. testing for convergence to the optimum , and terminating if convergence
is achieved.
We will discuss each of these in turn.
Finding the Search Direction
The first step in finding the search direction is to determine which constraints, if any,
are active or violated. Here an active constraint is defined as one with a value between
CT and CTMlN, where CT is a small negative number and CTMlN is a small positive
number. Remember that constraints must be negative to be feasible; therefore, if
is less (e.g., more negative) than CT, it is not considered active. Also, when we
approach a constraint boundary, it makes sense to begin pushing away from that
fence. Therefore, we initially choose a relatively large value for CT, such as -0.03
(i.e., within three percent of being mathematically active).
On the other hand, any constraint with a positive value is mathematically violated.
However, trying to achieve a precise zero on the computer is not meaningful. Also,
loads, material properties, etc., are not really known precisely. Furthermore, the
responses calculated by the finite element analysis are only approximate because of
the nature of the method. Therefore, we allow a small positive constraint value before
identifying a constraint as violated. This is the value of CTMIN typically taken as 0.003
(three-tenths of one percent violation). Thus, the governing definitions are
Inactive Eq. D-23
Active Eq. D-24
Violated Eq. D-25
These tolerances are shown in Figure D-3 for a single constraint in a two design
variable space.
S
*

*
F X
q 1
S
q
+ ( )
X
*
g
j
X ( )
g
j
X ( ) CT <
CT g
j
X ( ) CTMIN
g
j
X ( ) CTMIN >
Main Index Main Index


654
Figure D-3 CT and CTMIN
The use of Eq. D-23 through Eq. D-25 underscores the importance of normalizing
constraints. In MSC.Nastran, this is done automatically using the constraint bounds
as normalizing factors and the caution of avoiding constraint boundaries of 0.0 that
are discussed in Normalized Constraints on page 55 is emphasized once again.
Using the active constraint criteria, the algorithm first sorts all the constraints and
identifies those that are active or violated. Then, the gradients of the objective function
and all the active and violated constraints are calculated. Thereafter, a usable-feasible
search direction is found (if one exists). There are three possibilities:
1. There are no active or violated constraints.
2. There are active constraints but no violated constraints.
3. There are one or more violated constraints.
Each of these possibilities is handled differently.
x
2
x
1
Feasible region
Infeasible
region
g
j
X ( ) 0 >
g
j
X ( ) 0 <
g
j
X ( ) CT =
g
j
X ( ) 0 =
g
j
X ( ) CTMN =
Main Index Main Index
655 APPENDIX D
Numerical Optimization
No Active or Violated Constraints (Unconstrained Minimization). Frequently at
the beginning of the optimization process there are no active or violated constraints.
In this case, the feasibility requirement is automatically met since we can move in any
direction for at least a short distance without violating any constraint. Thus, we only
need to find a usable direction, that is, one that points downhill. It does not have to be
the steepest descent direction, but to start the process, this is the preferred choice.
Therefore, the initial search direction is simply:
Eq. D-26
The steepest descent direction is only used if this is the beginning of the optimization
or if the results of the last iteration had active or violated constraints and the
current one does not.
Now assume the last search direction is in the steepest descent direction and there are
still no active constraints. (If there were no violated constraints before, there will not
be any now.) We could again search in the steepest descent direction. Such a direction
would be perpendicular to the previous direction. However, there are very solid
theoretical reasons not to do this, and experience attests to the accuracy of the theory.
Instead, we move in what is called a conjugate search direction, or more precisely,
an A-conjugate direction, where A is the matrix of second partial derivatives of the
objective function. The A matrix is not actually computed, but there are methods for
approximating A that offer a guaranteed convergence rate for problems where the
objective is a quadratic function. (In fact, the successive use of steepest descent
directions may never converge.)
While there are methods that are considered more powerful than the one used here,
the simple conjugate direction method has proved to be quite reliable. The Fletcher-
Reeves conjugate direction method, used here, defines a search direction as:
Eq. D-27
where Eq. D-28
S
q
F X
q 1
( ) =
q 1 = ( )
S
q
F X
q 1
( ) = S
q 1
+

F X
q 1
( )
2
F X
q 2
( )
2
------------------------------------ =
Main Index Main Index


656
Figure D-4 Geometric Interpretation of the Steepest Descent Method
Figure D-5 Geometric Interpretation of the Fletcher-Reeves Method
x
2
x
1
-2
-4 -2 0 2 4 6 8 10 12
2
8
6
4
-40
-30
-20
-10
8
10
S
2
S
1
30
40
60
80
100
+
x
5
x
4
x
2
x
3
20
X
1
x
o
x
2
x
1
-2
-4 -2 0 2 4 6 8 10 12
2
8
6
4
-40
-30
-20
-10
8
10 S
2
S
1
30
40
60
80
100
+
x
2
x
4
x
2
x
3
20
x
1
x
o
S
3
Main Index Main Index
657 APPENDIX D
Numerical Optimization
The advantage of using the conjugate search direction is seen from Figure D-4 and
Figure D-5, which show a simple two-variable design space. In Figure D-4 the search
directions are the steepest descent directions, whereas in Figure D-5 the conjugate
directions are used. Note that in Figure D-4 the search direction is always
perpendicular to the previous direction. On the other hand, in Figure D-5 each search
direction uses the steepest descent direction plus some fraction of the previous
direction. This makes physical sense. If progress is made in direction , it is
reasonable to bias direction a bit in that direction rather than simply using a
steepest descent direction. This is exactly what Eq. D-27 and Eq. D-28 accomplish.
This algorithm is extremely simple but, as seen from Figure D-4 and Figure D-5, it
dramatically improves the rate of convergence to the optimum. In the rare case of
there being no constraints specified at all in the optimization task ( ), a special
purpose BFGS (Broydon, Fletcher, Goldfarb and Shannon) algorithm is used.
MSC.Nastran imposes constraints that are in addition to those imposed by the user so
that it is indeed extremely rare that (see Objective and Constraint Value
Evaluation on page 116).
The BFGS method is called a quasi-Newton method because it creates an
approximation to the inverse of the Hessian matrix (matrix of second derivatives of
the objective function). The matrix contains the (inverse of) of the second
derivatives. Initially, is set to the identity matrix, . The search direction is defined
as:
Eq. D-29
so that the first search direction is the direction of steepest descent.
After the first iteration, is updated using the following formula:
Eq. D-30
where
The change vectors and are defined as:
S
1
S
2
n
g
0 =
n
g
0 =
H
H I
S
q
H F X
q 1
( ) =
H
H
q 1 +
H
q
D
q
+ =
D
q +

2
--------------
\ .
| |
pp
T 1

--- H
q
yp p H
q
y ( )
T
+ [ ] =
p
T
y =
y
T
H
q
y =
p y
Main Index Main Index


658
This method is considered to be theoretically best, but requires significant memory to
store the matrix (actually just the upper half, since is symmetric). This method is
considered to be less sensitive to the accuracy of the one-dimensional search than the
Fletcher-Reeves method, through experience shows that the two methods are roughly
comparable in efficiency and reliability as they are coded in DOT.
This method can be proven to converge in N or few iterations for strictly quadratic
functions. However, since most optimization problems of interest are not quadratic,
the method may have to be restarted by setting to the identity matrix as the
optimization progresses. This restarting is automatic in DOT.
Active but no Violated Constraints. The more common search direction problem,
in which the design is feasible and there is at least one active constraint, is to find a
search direction that improves the design but moves parallel to or away from the
constraint. To solve this problem, a search direction must first be found that reduces
the objective function without violating any currently active constraints. The
following equations state the problem in mathematical terms:
Find the search direction that
minimize Eq. D-31
subject to Feasibility Condition Eq. D-32
Bounds on Eq. D-33
This is the same problem as shown in Eq. D-9 through Eq. D-11 for finding a usable-
feasible search direction in the physical example of searching for the lowest point on
the hill. Note that the scalar product is the magnitude of the two vectors times the
cosine of the angle between them. Thus, the objective of this suboptimization problem
is
Eq. D-34
p X
q
X
q 1
=
y F X
q
( ) F X
q 1
( ) =
H H
H
S
S
F X
q 1
( )
T
S
q
g
j
X
q 1
( )
T
S
q
S
q
( )
T
S
q
1 S
F X
q 1
( ) S ( ) cos
Main Index Main Index
659 APPENDIX D
Numerical Optimization
Since we are minimizing this function, we want the cosine of the angle between the
two vectors to be as large a negative number as possible but within the restriction of
Eq. D-32. Alternatively, for any angle between 90 and 270 degrees, Eq. D-34 can be
made more negative if the magnitude of is increased. Also, if satisfies the
requirements of Eq. D-32, then any increase in the magnitude of also satisfies this
equation. Therefore, it is necessary to bound , which is accomplished using Eq. D-33.
Assuming the resulting objective function from this subproblem is negative, the
usability requirement of Eq. D-17 is satisfied. If the objective function (defined by
Eq. D-31) cannot be forced to be negative, then it follows that no direction exists that
reduces the objective function while staying inside of the constraints. If this is the case,
the Kuhn-Tucker conditions are satisfied, and the optimization process may be
terminated.
In practice, the decision is not quite this simple because Eq. D-32 may include
constraints that are within a tolerance of CT but are not really critical. Therefore, we
reduce CT and delete any constraints from the active set J that are not within the new
tolerance. We then solve this direction finding problem again to determine if a more
precise satisfaction of the constraints can be achieved. If the value of CT is reduced in
magnitude to CTMlN (CT is negative) and no negative objective is found for the
direction finding problem, then this design is judged to have satisfied the Kuhn-
Tucker conditions, and the optimization process is stopped.
An unanswered question is: How do we solve the sub-problem of finding the usable-
feasible search direction ? The details are beyond the scope of this discussion. For
our purposes, it is sufficient to note that the problem can be converted to an iterative
form of solving a set of simultaneous equations with some conditions on the variables.
The actual algorithm is similar to that used in traditional linear programming. The
details of this suboptimization problem can be found in Vanderplaats, G. N. (see
Reference 13.).
One or More Violated Constraints. Now consider the case where one or more
constraints are violated. Such a case is shown in Figure D-6, where constraint is
violated and is active. Now we must find a search direction back toward the
feasible region even if it is necessary to increase the objective function to do so. To
achieve this, we augment our direction-finding problem of Eq. D-31 through Eq. D-
33 with a new variable W. This process has no direct physical significance to the
problem except as a measure of the constraint violation.

S S
S
S
S
g
1
x ( )
g
2
x ( )
Main Index Main Index


660
Figure D-6 Violation of Constraint(s)
The new direction finding problem is now:
Find the search direction S and artificial variable W that will
minimize Eq. D-35
subject to Eq. D-36
Eq. D-37
For discussion, assume that the parameter in Eq. D-36 is equal to 1.0 and the
parameter is a very large positive number. Then the second term dominates the
minimizing of the function defined by Eq. D-35 so any increase in the variable W will
drive the objective more and more negative (i.e., reduces the objective of the direction
finding problem). However, for W to increase, the first term in Eq. D-36 must become
more and more negative. Since is bounded by Eq. D-37, the cosine of the angle
between
and
must be moved closer and closer to -1.0. For this to happen, must point in a
direction opposite to
S
0
Feasible
Region
g
1
0 =
g
2
0 =
x
2
x
1
F X
q 1
( )
T
S
q
W
g
j
X
q 1
( )
T
S
q

j
W + 0 j J
S
q
( )
T
S
q
W + 1

j
S
q
g
j
X
q 1
( ) S
q
S
q
Main Index Main Index
661 APPENDIX D
Numerical Optimization
That is, must point straight back toward the feasible region. The first term in Eq. D-
35 is included simply as a means to reduce the true objective function if possible while
seeking a feasible design.
Now, consider the value of in Eq. D-36. This is referred to as a push-off factor since
its value determines how hard to push away from this violated constraint. If ,
even increasing W will not require a move anywhere except tangent to the constraint,
and this move will probably not bring the design back to the feasible region.
Therefore, some positive value is needed. In the optimizer, the more the constraint is
violated, the greater the push-off factor value should be.
This can be accomplished by expressing the push-off factor as a quadratic function
of the j-th constraint such that at
Thus,
Eq. D-38
Eq. D-39
Therefore, when a constraint becomes active, it is included in the direction finding
process. However, as the constraint becomes more critical, particularly as it becomes
violated, the push-off factor increases. The value of has been limited to 50.0 as an
arbitrary numerical limit. This limit is based on experience that shows this value to be
large enough to force the design back to the feasible region without causing too much
numerical ill-conditioning in the algorithm.
Finally, consider the value of . This parameter is initially chosen as a small number,
such as 5.0. Also, we always normalize and so that the first term in
Eq. D-35 and Eq. D-36 is near unity. Thus, the second term in Eq. D-35 dominates, but
not too strongly, allowing for the possibility of reducing the objective function while
overcoming the constraint violation. If this iteration does not overcome the violation
is increased by a factor of 10, although limited to an upper bound of about 1000, to
avoid numerical ill-conditioning. Usually, this has the result of bringing the design
g
j
X
q 1
( )
S
q

j
0.0 =

j
0 =
g
j
X
q 1
( ) CT =

j

0
1.0
g
j
X
q 1
CT
--------------------- =

j
50.0

F X
q 1
( ) g
j
X
q 1
( )

Main Index Main Index

662
back to the feasible region in a few iterations. If 20 iterations pass without overcoming
the constraint violations, the optimization process is terminated on the assumption
that no feasible solution exists.
The One-Dimensional Search. Having determined a usable-feasible search
direction, the problem now becomes one of determining how far the design can be
moved in that direction. Again, a variety of possibilities exist depending on the
starting point . However, in each case, polynomial approximations are used for
the objective and constraint functions in the one-dimensional search direction, .
The basic concept is to try some initial value for in Eq. D-22 and evaluate the
corresponding objective and constraint functions. Therefore, the first question to ask
is: What is a good first estimate for ? At the beginning of the optimization process,
very little information is available except the function values and their derivatives
with respect to . In fact, at first glance, it is not obvious that the derivatives of the
objective and constraints will respect to are available. However, consider the
objective function (the same algebra applies to constraints) and create a first order
MacLaurin series approximation to in terms of . Remember that
Eq. D-40
Thus, a first-order approximation to is
Eq. D-41
or
Eq. D-42
But
is just the j-th entry of . Also, from Eq. D-22,
X
q 1
S

F ( )
F X
q
( ) F X
q 1

*
S
q
+ ( ) =
F X
q
( )
F X
q
( ) F X
q 1
( ) =
FX
q 1
( )
x
i
----------------------------
x
i

-------
)
`

i 1 =
N

+
*
F X
q
( ) F X
q 1
( ) =
dF X
q 1
( )
d
---------------------------- +
*
F X
q 1
( )
x
i
----------------------------
F X
q 1
( )
x
i

------- s
i
=
Main Index Main Index
663 APPENDIX D
Numerical Optimization
Therefore,
Eq. D-43
Since both terms in Eq. D-41 are available, we have the slope of the function at
for any function (objective or constraint) for which the gradient is available.
Now consider how this information might be used. Since this is the first step in the
optimization process, we may optimistically expect to reduce the objective function by
some fraction, for example, 10%, which can be stated as
Eq. D-44
from which a proposed is:
Eq. D-45
This results in an estimate of which reduces the objective function by 10%.
However, since the gradients of some constraints are probably available, other
possible moves can be calculated as well. Remember that Eq. D-42 applies to a
constraint by simply substituting the constraint gradient for the objective gradient.
Now assume that some constraint gradients exist for constraints that are not critical,
and it is desired to estimate how far to move to make one of them critical. That is,
instead of reducing the value of the function by 10% as was done for the objective, it
is desired to estimate a move in that places it near the constraint boundary. Thus, a
linear approximation to find
is
Eq. D-46
and an estimate for is
dF X
q 1
( )
d
---------------------------- F X
q 1
( )
T
= S
q
0 =
F X
q 1
( ) F X
q 1
( ) =
dF X
q 1
( )
d
---------------------------- +
*
F X
q 1
( ) 0.1 = F X
q 1
( )

est
*
0.1 F X
q 1
( )
dF X
q 1
( )
d
*
----------------------------
------------------------------------- =

g
j
X
q 1
( ) 0.0 =
g
j
X
q
( ) g
j
X
q 1
( ) =
dg
j
X
q 1
( )
d
----------------------------- +
*
0.0 =

*
Main Index Main Index


664
Eq. D-47
Therefore, even at the beginning of the one-dimensional search, a considerable
amount of information is available to direct the process. Using the estimates for
given by Eq. D-45 and Eq. D-47 for each noncritical constraint, the smallest positive
proposed is taken as the first estimate of how far to move.
If constraints are currently violated, Eq. D-47 can still be used. Applying this
approximation to all violated constraints, the largest proposed value of is selected
as an estimate of how far to move to overcome all constraint violations. If the projected
move to a constraint that is not currently active is smaller than this, then a move to that
constraint is made as a first estimate since we do not want to add new constraints to
the set of violated constraints.
Using similar approximations, we can also estimate an upper bound on that forces
all design variables to their lower or upper bounds. This then provides a maximum
allowable value for .
While this brief discussion is not detailed enough to cover all of the intricacies of
estimating the first step in the one-dimensional search, it does provide a sense of the
complexity of the decision processes. Also, it should be noted that on subsequent
search directions, this process is modified to account for knowledge gained on
previous iterations. For example, if an average of 5% improvement is obtained on the
last two one-dimensional searches, then another 5% improvement is attempted on this
one.
The one-dimensional search process now proceeds to find the bounds on the that
contain the solution. Once the bounds on are known, the constrained minimum is
found by interpolation. Since has been defined as a direction of improving design,
the search can be limited to positive values of .
Finding Bounds On
At the beginning of this process, the values of the objective and constraints are known
at and where is the initial estimate for . Following are three cases
that must be considered when deciding if is the upper bound .

est
*
g
j
X
q 1
( )
dg
j
X
q 1
( )
d
-----------------------------
----------------------------------- =

*
S
q

*
0 =
1
=
1

*

1

u
Main Index Main Index
665 APPENDIX D
Numerical Optimization
Case 1 - All constraints are initially satisfied (all ). In this case, the
search direction is the one that reduces the objective. Therefore, if at , the objective
is greater than at , then is an upper bound on . Also, since the initial
constraints are all satisfied, if any , then is an upper bound because
the search has moved into the infeasible design space. If either the objective function
has begun to increase or some constraint has become violated, then is chosen as the
upper bound .
Case 2 - One or more constraints are violated (greater than CTMlN) at , and the
objective function is increasing . Here, a search direction has been
chosen that should reduce the constraint violation. Thus, at , if the maximum
constraint value is greater than the maximum constraint value at , then is the
upper bound . This usually occurs when some new constraint becomes violated at
. If this happens, it is probably not possible to overcome the constraint violations in
this direction, but the maximum constraint violation may at least be reduced.
Alternatively, a value of might be found where all constraints are satisfied. If this
happens, a feasible design is found and so, again, is an upper bound on . Here,
there is little concern that the objective function is increasing since the first priority is
to overcome the constraint violations. If the constraint violations can be overcome, it
is desirable to do so with a minimum increase in the objective function.
Case 3 - One or more constraints are violated (greater than CTMlN) at , and the
objective function is decreasing . This is the same as Case 2 except
that the objective function can be reduced. Therefore, everything is the same except
that the search is not stopped if the constraint violation(s) are overcome. If a feasible
design can be found that allows further reduction of the objective function, movement
continues in this direction until some new constraint becomes active or the objective
begins to increase. The corresponding is the upper bound .
During the process of finding the upper bound on , it may be required to move
further than the initial step . If does not yield an upper bound, another step
is taken. In practice, it may take many steps before a bound is found on the
solution. If this is the case, the best four solutions are retained and called , , ,
and . If more than three steps are required, the lower bound will not equal zero.
Instead, a lower bound is carried forward to retain the nearest four proposed values
of . If fewer than four values are available (including ), then all are retained.
For each retained value of , the value of the objective and all constraint functions are
also retained. These are then used to interpolate for an improved solution to the value
of .
g
j
0 = ( ) CTMIN <

1
0 =
1

*
g
j

1
( ) CTMIN >
1

u
0 =
dF X
q 1
( ) d 0 > ( )

1
0 =
1

1

*
0 =
dF X
q 1
( ) d 0 > ( )

u

1

1

2
2
1
=

L

1

2

U

L

*
0 =

*
Main Index Main Index


666
Interpolation for
Once bounds on the solution to the one-dimensional search problem have been
established, it is desirable to refine the solution as much as possible. To achieve this, a
polynomial interpolation of the objective and constraint functions is used. The basic
tool used here is a simple polynomial curve fit, which may be linear, quadratic, or
cubic, depending on the amount of information available. Experience has shown that
a cubic fit is adequate to approximate the functions without introducing too much
numerical error so higher order functions are not attempted.
Whether the objective function or a constraint is to be approximated, the basic
approach is the same. The process is demonstrated using a quadratic fit to three
function values.
A quadratic polynomial is defined as
Eq. D-48
where: are constants, and a is the variable.
In this case, corresponds to the one-dimensional search parameter .
Next, assume that three values of exist for three values of x as listed in the table
below:
Substituting the values of and into Eq. D-48 results in three equations with three
unknowns
Eq. D-49
Eq. D-50
Eq. D-51
Solving for the values of , , and results in
x F
0.5 -1.25
1.0 -3.00
2.0 -2.00

*
F a
0
= a
1
x a
2
x
2
+ +
a
0
a
1
a
2
, , x
x
F
F x
a
0
0.5 + a
1
0.25 + a
2
1.25 =
a
0
1.0 + a
1
1.00 + a
2
3.0 =
a
0
2.0 + a
1
4.00 + a
2
2.0 =
a
0
a
1
a
2
Main Index Main Index
667 APPENDIX D
Numerical Optimization
Eq. D-52
Substituting this into Eq. D-48 results in
Eq. D-53
Eq. D-53 can be used in different ways depending on whether the function is the
objective or a constraint. First, assume that this is the objective function and we wish
to find the value of that minimizes it. The function has a minimum or maximum
value when the derivative of with respect to is zero. Differentiating Eq. D-53 with
respect to yields:
Eq. D-54
Setting this to zero and solving for gives a proposed solution . Substituting
this into Eq. D-53 gives an estimated value of the function of . However, this
is not sufficient to indicate that is a minimum. The second derivative must also be
considered, where
Eq. D-55
Since the result is positive, the proposed solution represents a minimum of instead
of a maximum.
Now, assume that the function to be approximated is a constraint. In this case, the
value of is sought that drives the function to zero (i.e., forces the design to a
constraint boundary). To achieve this, the value of must be found that makes Eq. D-
54 equal to zero. Setting Eq. D-53 to zero and solving for yields
Eq. D-56
For each of these values of , the estimated value of is zero. The question is: Which
one should be chosen? In this case, note that the initial value is positive.
Therefore, is a proposed lower bound on since this overcomes the
constraint violation (to a quadratic approximation). Thus, this is a possible lower
bound on , . However, more information is available. The value of the variable
is the point at which the constraint again becomes violated. Therefore, this
value of is a possible upper bound on , . This can be verified by calculating the
a
0
2.0 a
1
8.0 = a
2
3.0 = , , =
F 2.0 = 8.0x 3.0x
2
+
F
x F
F x
x
dF
dx
------- 8.0 = 6.0 + x
x x 1.33 =
F 3.33 =
F
d
2
F
dx
2
---------- 6.0 =
F
x
x
x
x
4.0 10.0 + [ ]
3.0
----------------------------------- 0.2792 2.3874 , = =
x F
F x 0 = ( )
x 0.2792 = *

L
x 2.3974 =
x *
U
Main Index Main Index


668
gradient of F at and to determine if it is negative at , (i.e., forcing to be more
negative) and positive at (i.e., forcing to be violated). To verify this, simply
substitute the values of into Eq. D-54.
The optimization process leads to an estimate for for the minimum of the objective
and the zero of one constraint. In this case, the constraint is initially violated so we
have both a lower and an upper bound on . However, we must also estimate values
of for all constraints. The maximum lower bound and the minimum upper bound
define the range for the proposed solution to the one-dimensional search. It may
happen that the proposed upper bound is less than the proposed lower bound. If this
happens, it usually means that for each proposed there are one or more violated
constraints. In this case, Eq. D-54 is used to minimize the maximum constraint
violation, and this decision overrides other conditions.
Let us assume that the polynomial approximation for the objective and all constraints
is used and concludes with a lower and upper bound on the constraint values so that
the upper bound is greater than the lower bound. (If no constraints are violated at
, this is the usual case.) Next a choice must be made as to which value of to
select. If the search starts with violated constraints and the lower bound is less than
the upper bound, either the lower bound is selected, or the value is chosen that
minimizes the objective (assuming that it is between the lower and upper bounds). If
no constraints are violated at , the minimum proposed is selected from the
that minimizes the objective and the that represents the upper bound.
As indicated by the difficulty in describing the one-dimensional search process, this is
a particularly complex part of the optimization process. The description given here is
only a simple outline of the process. In practice, decisions must be made based on how
many proposed values of are available using linear, quadratic and cubic curve fits
to estimate the best value of the move parameter. Since it is impossible to predict all
the possible combinations of the situations that may arise, this is necessarily a search
process that is based more on experience than theory. The key is the need to
consistently improve the design. If the absolute best solution does not appear during
this one-dimensional search, further iterations provide other chances until the
convergence criteria are satisfied. This explains why more complex methods are not
used (with their associated number of function evaluations and cost) during the one-
dimensional search. In the optimization algorithm used in the Modified Feasible
Directions algorithm used in DOT, the one-dimensional search process is somewhat
more sophisticated but basically follows the process outlined here.

L

U

L
g
j

U
g
j
x
*
*
*
*
0 = *
0 =
*
*
*
*
Main Index Main Index
669 APPENDIX D
Numerical Optimization
At this point, the best search direction to improve the design has been determined and
a search conducted in that direction. The question now is: How can it be determined
if the optimal design has been achieved? The answer is based on the decisions that
determine when to terminate the optimization process.
Convergence to the Optimum
Since numerical optimization is an iterative process, one of the most critical and
difficult tasks is determining when to stop. The optimization software uses several
criteria to decide when to end the iterative search process, and these are described
here. It is important to remember that the process described in this section relates only
to the solution of the approximate optimization problem. The number of cycles
through the entire design process is controlled by other similar criteria as given in
Tests for Convergence on page 125.
Maximum Iterations
As with any iterative process, a maximum iteration counter is included. The default
for this is 40 iterations (search directions). Usually, an optimum is found sooner than
this; therefore, the maximum is mainly intended to avoid excessive computations.
No Feasible Solution
If the initial design is infeasible (constraints are violated), the first priority is to
overcome these violations and find a feasible solution. However, if there are
conflicting constraints, a feasible solution may not exist. Therefore, if a feasible design
is not achieved in 20 iterations, the optimization process is terminated.
Point of Diminishing Returns
Probably the most common situation is where the optimum is approached
asymptotically. Therefore, while some progress is still being made, continued
iterations are not justified. This is particularly true in MSC.Nastran because this
program is solving an approximate problem that is to be updated on the next design
cycle. Here, two criteria are used. The first criteria requires that the relative change in
the objective between iterations be less than a specified tolerance DELOBJ. Thus, the
criteria is satisfied if:
Eq. D-57
F X
q
( ) F X
q 1
( )
F X
q 1
( )
----------------------------------------------------- DELOBJ
Main Index Main Index


670
The default value for DELOBJ is 0.001.
The second criteria is that the absolute change in the objective between the iterations
is less than a specified tolerance DABOBJ. This criteria is satisfied if
Eq. D-58
The default value for DABOBJ is the maximum of and 1.0E-20.
The reason for the two criteria is that if the objective function is large, the relative
change between two successive iterations is an indication of convergence. However,
if is a very small number, a relative change is not meaningful, and thus the
absolute change controls the convergence.
Often, no progress is made on one iteration, but significant progress is produced on
the next. Therefore, one of these criteria is required to be satisfied on ITRMOP
consecutive iterations where the default value of ITRMOP is 2. This default may be
changed on the DOPTPRM Bulk Data entry.
Satisfaction of the Kuhn-Tucker Conditions
The Kuhn-Tucker conditions necessary for optimality are stated in Eq. D-14 through
Eq. D-17. In the uncommon case where there are no constraints, these conditions
degenerate to the familiar case of the vanishing gradient of the objective. In practice,
this is assumed to be true if all components of the gradient are less than 0.001.
The usual optimization task in MSC.Nastran is that many constraints are imposed. In
the process of finding a usable-feasible search direction, it is possible to detect if the
Kuhn-Tucker conditions are satisfied. If they are, the optimization process should be
terminated. However, in practice, other considerations may affect the results.
Remember that a constraint is defined as active if it is numerically greater than the
parameter CT. The optimization starts with CT = 0.03 (any constraint within three
percent of being satisfied is called critical). Since a constraint with this value is
considered in the Kuhn-Tucker calculations, these conditions may be satisfied even if
the search is up to three percent away from the constraint boundary. Therefore, if CT
is greater than CTMIN in magnitude, the CT value is reduced by some fraction
(typically 50%) and another attempt can be made to find a usable-feasible direction. If
CT is reduced in magnitude to a value of -CTMIN, and can still satisfy the Kuhn-
Tucker conditions, the optimization process is terminated.
F X
q
( ) F X
q 1
( ) DABOBJ
0.001 F X
0
( )
F x ( )
Main Index Main Index
671 APPENDIX D
Numerical Optimization
D.3 Sequential Linear Programming
The second method to be discussed is Sequential Linear Programming (SLP) (see
Kelly, J. E., in Reference 5.).
This method is regarded as the least effective of those available within MSC.Nastran
but is retained because it may prove of value in particular cases. This section explains
how the SLP method has been implemented in the DOT algorithm.
The basic concept is actually quite simple. First, a Taylor Series approximation for the
objective and constraint functions is created. This approximation is then used for
optimization, instead of the original nonlinear functions. Now when the optimizer
requires the values of the objective and constraint functions, these are very easily and
inexpensively calculated from the linear approximation. Also, since the approximate
problem is linear, the gradients of the objective and constraints are available directly
from the Taylor Series expansion.
Traditionally, the SIMPLEX algorithm is used for solving the linear approximate
problem. However, DOT uses the Modified Method of Feasible Directions, since it
solves linear problems nicely and works well for the size of problems normally
considered.
The general algorithm proceeds in the following steps:
1. For the current values of the design variables, sort the constraints and retain
the most critical for use during this cycle. Typically, the number of retained
constraints is about five times the number of design variables. The reason
that all constraints are not retained is that there may be thousands, and most
may be far from critical. Therefore, it would be wasted effort to calculate
their gradients.
2. Create a first order Taylor Series expansion of the objective and retained
constraints with respect to the design variables.
3. Define move limits on the design variables. Typically, during one cycle, the
design variables will be allowed to change by 20-40%, but this is adjusted
during later cycles.
4. Solve the linear approximate optimization problem.
5. Check for convergence. If satisfied, EXIT. Otherwise, repeat the process from
step 1.
In step 1, Taylor Series expansions are created in the form:
Main Index Main Index


672
Eq. D-59
Eq. D-60
where Eq. D-61
and J is the set of retained constraints.
Note that everything in Eq. D-59 and Eq. D-60 is constant except the values of the
design variables, . Therefore, Eq. D-59 and Eq. D-60 can be rewritten as
Eq. D-62
Eq. D-63
where Eq. D-64
and Eq. D-65
The following linear approximate optimization problem is now solved;
minimize Eq. D-66
subject to Eq. D-67
Eq. D-68
where Eq. D-69
and Eq. D-70
The multiplier in Eq. D-69 and Eq. D-70 is initially set to RMVLMZ. Depending on
the progress of the optimization, this parameter will be sequentially reduced.
Typically, the first approximate optimization produces a design that violates one or
F

X ( ) F X
q 1
( ) = F X
q 1
( )
T
+ X
g

j
X ( ) g
j
X
q 1
( ) = g
j
X
q 1
( )
T
+ X j J
X X
q
= X
q 1

X
q
F

X
q
( ) F

0
= F X
q 1
( )
T
+ X
q
g

j
X
q
( ) g

j
0
= g
j
X
q 1
( )
T
+ X
q
j J
F

0
F X
q 1
( ) = F X
q 1
( )
T
X
q 1
g

1
0
g
j
X
q 1
( ) = g
j
X
q 1
( )
T
X
q 1
F

X
q
( )
g

j
X
q
( ) 0 j J
x

i
L
x
i
q
x

i
U
i 1 N , =
x

i
L
x
i
q
= D X
i
q

i
U
x
i
q
= D X
i
q
+
D
Main Index Main Index
673 APPENDIX D
Numerical Optimization
several constraints. However, as the optimization proceeds, this constraint violation
should be reduced. If it actually increases, then is reduced, as well as the move
limits on the design variables.
The convergence criteria for the SLP method are much the same as for the other
methods, with the addition that the process will not terminate if there are significant
constraint violations. In this case, the process continues until the constraints are
satisfied or the maximum number of iterations is reached.
D
Main Index Main Index


674
D.4 Sequential Quadratic Programming
The Sequential Quadratic Programming (SQP) algorithm has a firm theoretical basis
and its implementation in DOT has been refined to the extent that it is considered
competitive with the Modified Method of Feasible Directions (MMFD) in many cases.
Although MSC.Nastran has selected the MMFD as its default optimizer, this is a close
call and users may find SQP works better for their applications. SQP is selected by
setting METHOD=3 on the DOPTPRM Bulk Data entry. For further information, see
Powell, M.J.D., and Vanderplaats, G.N. and Sugimoto, H., in Reference 12.
The basic concept is very similar to Sequential Linear Programming. First the objective
and constraint functions are approximated using Taylor Series Approximations.
However, a quadratic, rather than a linear approximation of the objective function is
used. Linearized constraints are used with this to create a direction finding problem
of the form;
minimize Eq. D-71
subject to Eq. D-72
This sub-problem is solved using the Modified Method of Feasible Directions. The
matrix B is a positive definite matrix which is initially the identity matrix. On
subsequent iterations, B is updated to approach the Hessian of the Lagrangian
functions.
Now assume the approximate problem of minimizing Q subject to the linearized
constraints has been solved. At the optimum for this problem, the Lagrange
multipliers can be calculated. With the Lagrange multipliers available, an
approximate Lagrangian function can be constructed and used to search in the
direction defined by . The task is to find in order to:
minimize Eq. D-73
where:
=
Q S ( ) F
0
= F
T
+ S 0.5S
T
BS +
g
j
T
S g
j
0
+ 0 j 1 m , =

j
j 1 m , =
S
F X ( ) = u
j
max 0 g
j
X ( ) , [ ]
j 1 =
m

+
X
X
q 1
S +
Main Index Main Index
675 APPENDIX D
Numerical Optimization
and from the previous iteration.
During the one-dimensional search, approximations are made to the components of
because this function has discontinuous derivatives at the constraint boundaries,
but the components are smooth.
After the one-dimensional search is complete, the B matrix is updated using the BFGS
formula:
Eq. D-74
where:
Now replaces and a new iteration is begun.
= first iteration
= subsequent iterations
=
=
=
=
if
=
if
u
j

j
j 1 m , =
u
j
max
j
1
2
--- u
j
'
j
+ ( ) ,
j 1 m , =
u
j
' u
j
=

B
*
[ ] B [ ] =
B [ ] p { } p { }
T
B [ ]
p { }
T
B [ ] p { }
-------------------------------------------
{ } { }
T
p { }
T
{ }
------------------------ +
p { }
X
q
X
q 1

{ } y { } 1 ( ) B [ ] p { } +
y { }

q
x

q 1
x

F X ( )
j
g
j
X ( )
j 1 =
m

+
p { }
T
y { } 0.2 p { }
T
B [ ] p { }
1.0
0.8 p { }
T
B [ ] p { }
p { }
T
B [ ] p { } p { }
T
y { }
-----------------------------------------------------------
p { }
T
y { } 0.2 p { }
T
B [ ] p { } ( ) <
B
*
[ ] B [ ]
Main Index Main Index


676
D.5 Summary
The purpose of this Appendix has been to provide an overview of the computational
details of the optimization process. The techniques used in the optimizer have been
chosen because they are reasonably robust and because they allow for considerable
flexibility in formulating the design problem.
From the variety of decisions that must be made during the optimization process it is
clear that anything the user can do to improve the numerical conditioning of the
problem will have a strong influence on the efficiency and reliability of the result. It is
hoped that some understanding of the actual optimization process will assist in this.
Main Index Main Index
I N D E X
MSC.Nastran Design Sensitivity and Optimization Users Guide

A
A-conjugancy, 655
alternate reduction, 639
ALTRED, 639
analysis, 208
discipline, 535
available types, 156
model, 535
definition of, 162
properties, 162
field position, 164
analytic boundary shapes, 182
approximate
design task, 97
model, 29, 36, 101, 535
optimization, 535
approximation
concepts, 28, 535
constraint screening, 537
deletion, 28, 537
regionalization, 65, 537
definition of, 25, 28
design variable linking, 28, 539
formal approximations, 29, 540
direct variables, 111
intermediate variables, 113
reciprocal variables, 111
methods
convex linearization, 113
approximations
direct, 112
formal, 29, 101, 540
reciprocal, 112
auxiliary boundary models, 432, 536
auxiliary models, 166, 422, 536
external, 176
in analytic boundary shapes, 52
examples using, 182, 430
in direct input of shapes, 52, 170
examples using, 175, 419
AVG, 225
B
basic optimization
problem statement, 14, 645
design constraints, 14
design objective, 14
design variables, 14, 160
basis vector, 74, 420, 536
basis vectors, basis functions, see also
reduced basis formulations
beam libraries
user defined, 205
beam library, 536
beta method, 536
boundary conditions
multiple, in design, 209
boundary shapes
analytic, 430
Bulk Data Entries
TOPVAR, 300
Bulk Data entries, 216
BNDGRID, 216
DCONADD, 218
DCONSTR, 218
DEQATN, 222
DESVAR, 158, 228
DLINK, 230
DOPTPRM, 231, 232
DRESP1, 188, 244
I N D E X
MSC.Nastran
Design
Sensitivity
and
Optimization
Users Guide
Main Index Main Index
Index
INDEX

678
DRESP2, 267
DSCREEN, 278
DTABLE, 279
DVBSHAP, 281
DVCREL1, 282
DVCREL2, 284
DVGRID, 287
DVMREL1, 288
DVMREL2, 290
DVPREL1, 293
DVPREL2, 296
DVSHAP, 298
FREQ3, 227
FREQ4, 227
FREQ5, 227
MODTRAK, 299
PBAR, 148
PBARL, 149
PBEAM, 149
PBEAML, 150
RANDPS, 493
role in design modeling, 154
C
Case Control, 155, 208
output requests, 156
Case Control commands
ANALYSIS, 156
AUXCASE, 214
AUXMODEL, 214
DESGLB, 201, 210
DESOBJ, 209
DESSUB, 201, 210
CDD
conservative discrete design, 133
composite materials, 470
conservative discrete design
(CDD), 133
constraints, 14, 54, 64, 537
active, 98
deletion, 28, 64, 537
equality, 14
inequality, 14
normalized, 55
properties, 117
regionalization, 65, 537
screening, 36, 64, 537
side, 14
threshold, 64
violated, 98, 544
converged, 37
convergence, 125, 537
hard, 125, 129, 541
soft, 127, 544
convergence, See also design cycle
convergence and optimizer convergence,
125
CSV, 537
cyclic symmetry, 639
D
DB, 225
DBA, 225
design, 162
compromise infeasible, 132
cycle, 538
feasible, 540
infeasible, 24, 130, 541
initial, 35
iteration, 538
model, 26, 538
definition of, 7
integrity of, 8
nonunique, 130
optimization, 2, 538
unique, 130
variables, 14
design constraints, 210, 537
defining using DESGLB, 201, 210
defining using DESSUB, 210
definition of, 7, 12
equality
implementing, 56
in basic optimization problem
statement, 13
violated, 98
Main Index Main Index
679 INDEX
design cycle
convergence, 125
hard convergence, 125, 126, 541
decision logic, 129
soft convergence, 125, 127, 544
decision logic, 127
design modeling
for shape, 47
overview, 153
process overview, 153
design objective, 209
definition of, 7, 12, 542
formulating, 54, 199
identifying with DESOBJ, 156
design of experiments
(DOE), 133
design responses, 53, 188
first-level (type-1), 540
limitations in second-level, 192
second-level (type-2), 543
design sensitivity, 2
formatted, 364
unformatted, 369
design sensitivity analysis, 67, 538
coupled fluid-structure interaction, 85
dynamic responses, 80
force derivatives, 82
modal transient responses, 84
design sensitivity coefficient, 538
definition of, 67
design sensitivity print, 212
design space, 539
definition of, 7, 12
flat, 540
graphical examples of, 21
design variables, 28, 158
continuous, 537
definition of, 7
dependent, 537
discrete, 539
in basic optimization problem statement,
160
in property optimization, 162
in shape optimization, 47
independent, 69, 541
linking, 28, 539
role in design modeling, 154
set, 211
to property relations, 162
designed coordinates, 539
designed grids, 539
designed properties, 538
difference, 70
central, 70
forward, 70
differential stiffness, approximation of, 92
direct input of shapes, 539
examples using, 175, 419
discrete variable optimization, 360
distributed memory parallel, 37, 139
E
energy norm, used in shape sensitivity
analysis, 74
equations
to define properties, 46, 162
to define responses, 53, 188, 192
Euler buckling
defining as a design response, 192
evaluation
property, 111
type-2 response, 115
type-3 response, 116
executive control, 207
external responses, 206
F
failure modes, in design, 24
feasible design, 110, 540
feasible region, 24
file management, 205
finite difference methods, 70
finite differences, 540
central, 75
first-forward, 17
in approximations, 73
finite element analysis, 26
Main Index Main Index
Index
INDEX

680
first-level response, 540
FREQ4, 227
fully stressed design, 37, 359, 470, 540
(FSD), 134, 135, 136
function evaluation, 111
G
GDACMS, 143
geometric boundary shapes, 179
GMAX, 129
gradient evaluation, 119
gradient-based numerical optimizer, 541
gradients, 12, 17, 541
beam library constraints, 124
dependent design variables, 123
property constraints, 124
type-3 responses, 123
grid
perturbation, 541
variation, 541
I
inertia relief, 639
infeasible design, 541
INREL, 639
INVDB, 225
INVDBA, 225
K
Kuhn-Tucker conditions, 19, 541, 649, 670
L
Lagrange multipliers, See also Kuhn-Tucker
conditions, 19
Lagrangian function, 649
limits
design variable, 104
move, 104
property, 106
load case deletion, 541
load conditions, multiple, in statics, 155
examples using, 388
lower bound, 105
M
mathematical programming
(MP), 134, 135
MDACMS, 143
mode tracking, 62, 212, 542
Modified Method of Feasible Directions
(MMFD), 23
Modules
AXMDRV, 317
AXMPR1, 317
AXMPR2, 317
DOM10, 318
DOM11, 318
DOM12, 318
DOM6, 317
DOM9, 317
DOPR1, 318
DOPR2, 318
DOPR3, 319
DOPR4, 319
DOPR5, 319
DSABO, 319
DSAD, 320
DSAE, 320
DSAF, 320
DSAH, 320
DSAJ, 321
DSAL, 321
DSAM, 321
DSAN, 321
DSAP, 321
DSAPRT, 212, 365
DSAR, 321
DSAW, 322
DSCM2, 370
DSPRM, 322
DSTAP2, 322
DSVG1, 323
DSVG2, 323
Main Index Main Index
681 INDEX
DSVG4, 323
MDCASE, 323
SDSA, 323
SDSB, 324
SDSC, 324
SHPCAS, 324
SOFTEXIT, 305
WEIGHT, 324
modules, 317
move limits, 542
automatic update of, 108
implemented as equivalent constraints,
109
multidisciplinary analysis, 58
N
NEWBULK, 215, 377
NSTR, 65
O
objective, 14, 54
objective function, 542
one-dimensional search, 18, 542, 645, 650,
664, 666
search parameter, 666
finding bounds on, 664
interpolation, 666
vector, 650
operations research, 11
optima
non-unique, 131
optimization, 14, 37, 97
acoustic, 478
discrete variable, 37, 133, 461
dynamic response, 447
problem statement, 14
property, 542
superelement, 144, 461
transient dynamic, 497
optimize
optimization parameters, 636
subcase create, 638
subcase parameters, 638
subcase select optimize, 639
optimizer, 36
convergence, 669
general limitations, 8
gradient based, 12, 16
optimum
constrained, 19
design, 542
global, 541
nonunique, 542
searching for, 16
unconstrained, 20
output, 329
design punch, 374
design sensitivity, 364
output parameters, 327
P
parallel
DOF domain, 142
frequency domain, 141
parameter, 304
Parameters
Nastran
AUTOADJ, 304
CDIF, 75, 304
DESPCH, 304, 327, 374
DESPCH1, 304, 374
DPEPS, 305
DSNOKD, 305
IUNIT, 305
NASPRT, 305, 327
OPTEXIT, 305, 327
POST, 328
SENSUOO, 305
SOFTEXIT, 127, 305
SOLADJC, 306
STPSCL, 74
UPDTBSH, 306
Optimization
APRCOD, 232
CONV1, 128, 240
CONV2, 128, 240
Main Index Main Index
Index
INDEX

682
CONVDV, 129, 240
CONVPR, 128, 240
CT, 99, 243, 653
CTMIN, 100, 243, 653
DABOBJ, 670
DELB, 73, 241
DELOBJ, 670
DELP, 107, 240
DELX, 105, 240
DELXV, 105
DESMAX, 129, 239
DISBEG, 242
DISCOD, 242
DPMIN, 107, 240
DXMIN, 105, 240
FSDALP, 136, 240
FSDMAX, 136, 240
GMAX, 129, 240
GSCAL, 243
IPRINT, 238, 327
METHOD, 233
P1, 236, 327
P2, 236, 327
P2CALL, 328
P2CBL, 328
P2CC, 328
P2CDDV, 328
P2CM, 328
P2CP, 328
P2CR, 327
P2RSET, 328
PLVIOL, 241
PTOL, 241
STPCSL, 241
perturbation, 73
property, 73
shape basis, 73
primary model, 542
properties, 44, 162
designed, 69
type-1, 44
type-2, 46
pseudo-load vector, 74, 543
push-off factor, 661
R
Rayleigh Quotient Approximation, 114
reduced basis formulations, 543
basis vectors, 536
for properties, 44
for shape, 48
example of, 388
in property optimization, 46
in shape optimization, 48
analytic boundary shapes, 186
direct input of shapes, 179
relative minima, 8
response
RMS, 490
responses, 53, 59, 62
external, 62, 540
PSD, 60
rms, 60
second level, 543
synthetic, 62, 544
third level, 544
type-1, 53, 59
type-2, 53, 61
type-3, 54, 61
restarts, in design sensitivity and
optimization, 521
results
supported entities, 641
RMS, 490
RMSDISP, 493
RSS, 225
S
search direction, 12, 543
determination of, 653
feasible, 646
steepest descent, 17, 22, 655
usable, 646
usable-feasible, 648
search vector, 17
semianalytic sensitivity analysis, 543
sensitivities
central difference, 536
Main Index Main Index
683 INDEX
forward, 540
PSD response, 95
real eigenvalue, 89
real eigenvector, 90
RMS response, 95
volume, 89
weight, 89
sensitivity, 71, 76
adjoint, 71, 88
coefficient, 67, 538
complex eigenvalue, 93
direct, 71
flutter, 94
frequency response, 83
sensitivity analysis, 36, 67, 538
direct, 539
shape basis vector, 51, 168, 214, 543
analytic boundary shapes, 52, 172, 182
direct input of shapes, 52, 170, 175
geometric boundary shapes, 52, 171
manual grid variation, 51, 169
shape design variable, 543
shape optimization, 166, 205, 419, 544
using analytic boundary shapes, 52
shape sensitivity
definition of, 544
side constraints, 544
SOFTCV, 128
SOL 200, 143
Solution 200, 309
flowchart, 309
sparse data recovery, 95
spectral density, 60
steepest descent, 17
structural analysis, 35
structural optimization, 26
stuctural optimization, early attempts at, 27
subDMAPS, 314
superelement design modeling
design responses in, 145
design variables in, 145
DVGRID entries in, 145
external superelements in, 147
image superelements in, 147
limitations, 146
shape optimization and, 145
T
Taylor series, 101
Topology optimization, 137
TRIM Bulk Data entry
specification of, 300
TRS (truncation threshold), 64
U
upper bound, 105
Main Index Main Index
INDEX

684
Main Index Main Index