AIAA 2015-2070

AIAA SciTech
5-9 January 2015, Kissimmee, Florida
56th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference

Some Observations on the Current Status

of Performing Finite Element Analyses

Ivatury S. Raju *
NASA Langley Research Center, Hampton, Virginia, 23681, USA

Norman F. Knight, Jr.†

Scitor Corporation, Reston, Virginia, 20190, USA

and
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Kunigal N. Shivakumar ‡
North Carolina A&T State University, Greensboro, North Carolina, 27411, USA

Aerospace structures are complex, high-performance structures. Advances in reliable and

efficient computing and modeling tools are enabling analysts to consider complex
configurations, build complex finite element models, and perform analysis rapidly. Many of
the early-career engineers of today are very proficient in the usage of modern computers,
computing engines, complex software systems, and visualization tools. These young engineers
are becoming increasingly efficient in building complex 3D models of complicated aerospace
components. However, the current trends demonstrate blind acceptance of the results of the
finite element analyses. This paper is aimed at raising an awareness of this situation. Examples
of the common encounters are presented. To overcome the current trends, some guidelines
and suggestions for analysts, senior engineers, and educators are offered.

I. Introduction

F INITE Element Analysis (FEA) systems have made astounding progress in advancing the modeling, analysis, and
visualization of complex structures and structural components. A few years ago, a sophisticated three-dimensional
(3D) model of structural components that would have taken an engineer several months to model, analyze, and
interpret the results of, can now be performed in a few days thanks to advancements in computer hardware, software,
and integration of tools by commercial software developers. Figures 1 and 2 show recent examples of complex models
of aerospace structural components (see also Refs. 1-3).
Many of the early-career engineers of today are very proficient in the use of modern computers, computing engines,
and complex software packages. These young engineers are becoming increasingly efficient in building complex 3D
models of complicated aerospace components. The current day general-purpose, commercially available software
packages such as NASTRAN, ABAQUS, ANSYS, ADINA§, etc., enable sophisticated analyses with 3D complex
models very efficiently and rapidly. Post-processing and visualization of the results can now be performed with
increasingly efficient tools.
The advancements in computing engines and software are allowing users who have little or no background in
engineering mechanics to perform analyses and post-process the results of these analyses. As such, the blind
acceptance and quality of the interpretation of the results are two significant problems today. This trend is alarming
and appears to be worsening. The objective of this paper is to show some of the current trends, identify some of the

*
NASA Technical Fellow for Structures, NASA Engineering and Safety Center. Fellow AIAA, Member ASME,
Member ASCE.
†
Principal Subject Matter Expert, Fellow AIAA, Fellow ASME.
‡
Professor & Director, Center For Composite Materials Research. Associate Fellow AIAA, Member ASME.
§
NASTRAN® is a registered trademark of National Aeronautics and Space Administration,
ABAQUS/Standard is a registered trademark of Dassault Systèmes Simulia Corp., ANSYS is trademark
of ANSYS, Inc., ADINA is a trademark of ADINA R & D, Inc.
1
American Institute of Aeronautics and Astronautics

This material is declared a work of the U.S. Government and is not subject to copyright protection in the United States.
 steps that need to be taken to ensure that this trend is reversed and eliminated, and ensure the use of good quality FE
analysis and results are restored. Guidelines and suggestions for senior engineers and educators are offered. As the
practice of FEA is very broad, the scope of this paper is limited to the performance and use of linear stress analysis
tools. Other challenges outside the scope of the current paper include: composite structures requiring careful definition
of lamination patterns and local coordinate system definition; thermal stress and strain simulations that often produce
non-intuitive solutions; and contact modeling including sliding friction to account for assembly preload that may
require the use of unsymmetric equation solvers.
For convenience in presentation, the phrases analysts and engineers are used interchangeably throughout the paper.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

a) b)

c)
Figure 1. Complex finite element models. a) Aircraft empennage-Global model.
b) Aircraft empennage-Local model. c) Crew module.

Interface flange
Hand Rail
Access platform

Figure 2. Complex finite element models - rocket structures.

2
American Institute of Aeronautics and Astronautics
 II. Current Trends
“The model exactly looks like the part. The analysis ran to completion without any
errors; the results are displayed as contour plots in color - how could the analysis and
results be wrong?”- An Analyst
Many commercial software developers have tried to build in a number of checks (i.e., convergence checks, element
size, shape, type compatibility, etc.) and offer advice to avoid major pitfalls. However, the use of the tools provided
by the software developers and the interpretation of the results are still the purview of the code users. The software
developers implicitly assume that the user is familiar with and well versed in engineering and computational
mechanics. The current and alarming trends suggest that this assumption may be incorrect and at least some of the
users are not well versed in engineering mechanics and the mathematical theories that underlay the finite element
method. The following are some of the questions posted on ResearchGate¶ during the past few months:
1. “Validating Curved Composite Panel Results from FEA: I am currently looking at a large, curved, composite
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

panel which is being sized by FEA. Before undertaking a physical test, is there anyone who has experience
validating the FEA results for curved panels? Or know of a hand calc method that can accurately predict
stresses?” A Professor of Aeronautics and Mechanics replied with a suggestion to look at the fundamentals
of shell structures and pointed this person to Timoshenko and Woinowsky-Krieger’s book on Plates and
Shells4.
2. “Why is it important to have a Weak Formulation for FEM and why it does not give accurate results? What
type of method and techniques are available to get accurate results using weak formulation?”
3. “What is the difference between essential boundary conditions and natural boundary conditions?”
4. “What is the shear locking phenomenon and how can we avoid it?”
5. “Why do we always use and plot von Mises stresses in finite element analysis?”
6. “I am writing FEM code for linear buckling analysis of a cylinder shell, but using 3D-solid elements (8-
noded-brick element). I am looking for the FEM formula of geometric stiffness matrix (also called stress
stiffness matrix or stability coefficient matrix). If everyone has or can derive this formula for 3D-solid
element please let me know?”
Clearly, these questions point to the fact the individuals asking these questions have inadequate formal training in
engineering mechanics and FEA theory. Admittedly, this is not a sample that points to the current trends. However,
some senior professors and the authors have noted similar and alarming trends over the past few years.
In the remainder of this section some of the current trends are discussed.
General Observations:
 General lack of understanding of basic assumptions in engineering mechanics is observed.
 Black box software packages are being used without engineering knowledge about finite element theory and
hence, garbage in – garbage out (GIGO) is noticed often.
 The use of building block5 approaches in finite element modeling and analysis is very rare.
 A well-thought-out plan to modeling the analysis region is rarely evident.
 Finer meshes where they are not needed and coarse meshes where large gradients exist are frequently observed,
suggesting lack of knowledge of structural and engineering mechanics.
Meshing and Modeling:
One of the authors observed recently, “Meshing is not modeling – Meshing and modeling are two different things.”
Meshing requires expertise in using a software package, while modeling involves expertise in idealization of the
structure and in understanding the structural response to loads and restraints. Current day users are very adept in
performing quick, rapid, and accurate meshing of complex aerospace structural components. They are adept at
meshing but not always at modeling.
Users need to understand the implications associated with selecting one element type over another. Choices
associated with element shape function order (linear, quadratic, p-versions), element continuity requirements,
placement of mid-side nodes, and facetted modeling of curved surfaces do influence the results.
Boundary and interface conditions:
Often, incorrect boundary and interface conditions are prescribed and applied to the models. Also, boundary and
interface conditions are chosen by what is convenient to prescribe and apply. There are also major issues with

¶
ResearchGate is a social networking site for scientists and researchers to share papers, ask and answer questions,
and find collaborators, Launched in May 2008.
3
American Institute of Aeronautics and Astronautics
 interpreting the results across the interface regions. Consider a bi-material plate subjected to remote tensile loading in
the y-coordinate directrion as shown in Fig. 3. Analysts do not appear to recognize that the interface conditions along
y = 0 required to be satisfied are:

u I = u II ,( y ) I = ( y ) II , and ( xy ) I = ( xy ) II Eq. (1)

where u are the displacement fields and y and xy are the
normal and shear stresses, and the superscripts I and II
denote material I and material II, respectively. As required
by the interface conditions in Eq. (1), only the normal and
shear stresses are continuous along the interface. However,
usually one finds that this fact is ignored and along the bi-
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

material interface all the stresses are averaged and plots of

stress contours are presented.
Three-dimensional (3D) Modeling:
3D mesh models are currently more common than
plate/shell models. The theme appears to be “when in doubt
use 3D meshes”. Often solid models from ProE# and/or
Solidworks** are used to develop shape/configuration.
These packages treat all geometry as 3D. Considerable
amount of effort and judgment is required on the part of the Figure 3. Bi-material plate.
analyst to extract representative planes out of the 3D solid
model so that accurate shell model representations can be built. Many untrained users are not skilled enough to perform
these operations, and hence, rely on the use of an “Automesh” button even though it may generate a very large number
of solid elements to fill a thin plate or a thin shell. The results for bending, transverse shear, and other perhaps
important quantities become much harder to interpret from these types of model representations.
Also, combined plate (or shell) and 3D models are very uncommon. The reason for this is that the analysts are not
sure what they need to do at the plate/shell and 3D interfaces. Most of the time all the nodes on the interface plane of
the 3D model are connected to the plate/shell nodes using a set of kinematic constraints or a rigid body element (RBE)
– see text box below – at that interface.

Rigid Body Element (RBE)

This is the widely used terminology in NASTRAN. In NASTRAN, multipoint constraints

are created internally with the rigid elements RBAR, RBE1, RBE2, RBE3. RBE1 defines a
rigid body connected to an arbitrary number of GRID points; RBE2 defines a rigid body
whose independent degrees of freedom are specified at a single GRID point and whose
dependent degrees of freedom are specified at an arbitrary number of GRID points; RBE3
defines the motion at a reference GRID point as the weighted average of the motions at a set
of other GRID points (UAI NASTRAN Reference Manual, 1999). Note that all RBEs are
rigid. The RBAR and RBE2 do constrain parts to move in total unison as if connected by
infinitely stiff structural member. RBE3 and RSPLINE use interpolation relations to
calculate the movement of some dependent nodes based on either a weighted average or
some polynomial function (spline) of another set of independent points. In these cases the
nodes are not locked together and do not move in unison but rather follow along in some
average sense. RBAR and RBE2 elements add artificial stiffness to a model. RBE3 does not.
Hence, modeling with one versus the other can affect load distribution to the connected
nodes, modes and dynamic behavior, etc.

#
PTC Creo, formerly known as Pro/ENGINEER is a parametric, integrated 3D CAD/CAM/CAE solution created by
Parametric Technology Corporation (PTC).
**
SolidWorks is a solid modeling CAD (computer-aided design) software that runs on Microsoft Windows and is
since 1997 produced by Dassault Systèmes SolidWorks Corp., a subsidiary of Dassault Systèmes.
4
American Institute of Aeronautics and Astronautics
 Global/Local (G/L) approaches:
Global/Local approaches are used in two
situations. The first situation is when after an
initial analysis is performed, a region (or regions)
of large gradients is identified. The analyst would
then like to refine the region of large gradients
(local region) alone without changing the
modeling in the remainder of the model (global)
to obtain a high-fidelity solution in the local
region. Figure 4 shows two examples of model
refinement: global modeling with mesh
transition (upper figure) and G/L modeling
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

(lower figure).
The second situation that may require the G/L
approach is when two regions of a structure are
modeled by two different teams and these two
regions need to be connected and analyzed
together. An example of this situation is shown
in Fig. 5, where a global model of the vertical tail
and a local model of the lug region were modeled Figure 4. G/L modeling.
by two different teams and analyzed using two
different software packages (see Ref. 6).
Global FE
Model

Local FE
Model

a)

b)
Figure 5. G/L models. a) NASTRAN shell model. b) Local ABAQUS models.
In both G/L approaches, the results of the global model
G/L
are used to prescribe the conditions across the G/L Interface
interfaces, see Fig. 6. Clearly, across the G/L boundaries
u1 u1
two conditions need to be satisfied1-3, 6-11: (a) continuity of
displacements and (b) reciprocity of tractions. In most
Global Local
implementations, only the continuity of displacements is u2 u2
Region Region
prescribed. During the past five years, several types of G/L
modeling and analysis variations were observed by the u3 u3
authors: Interpolated using
element shape functions
a) Displacements from the global model on the G/L or nodal averaging
interface are prescribed at the matching nodes in from global region

the local model. Displacements of non-matching Figure 6. G/L boundary interface techniques.
nodes are interpolated using some ‘average’

5
American Institute of Aeronautics and Astronautics
 techniques (see Fig. 6). This process is repeated for all the faces of the local model in 3D situations or for all
edges in 2D simulations.
b) The entire face of the local model is constrained using RBEs to a single node and equivalent forces from the
global model are prescribed to this node (see Fig. 7(a)). This process is repeated for all the faces of the local
model in 3D situations (see Fig. 7(b)).
c) The local model is chosen as a convenient rectangular block whose dimensions do not match the local region.
The interface process outlined in (b) is followed to prescribe boundary and interface conditions on all the
surfaces of the block (see Fig. 8).
Obviously, the three strategies outlined here are incorrect and lead to incorrect results and decision making based
on these results is not warranted.
(Fy)G (Fz)G
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Global Local
Region Region Global Local
(Fx)G
Region Region
(Fx)G (Fy)G

a) b)
Interpolated using
RBEs element shape functions

Figure 7. G/L boundary interface techniques using RBEs. a) 2D version. b) 3D version.

Local Region

(Fz)G

(Fx)G
(Fy)G
RBEs
Global Mesh

Local Mesh

Figure 8 G/L boundary interface techniques for non-matching 3D G/L regions.

Curved free edge

Figure 9. Stress concentrations and singularities in composites.

Singularities:
Singularities occur in structural components due to cracks, delaminations, free edges, load discontinuities, material
interfaces, etc. Figure 9 shows typical situations that occur in laminated composites. Stresses at these singular point
locations have no meaning as the FE solutions do not show convergence. Not recognizing singularities and yet
proceeding to perform the analysis is becoming common and structural margins are often reported from analysis using

6
American Institute of Aeronautics and Astronautics
 only a single mesh. Clearly, the stress results would change if the mesh is refined near these singularity locations and
margins would be different or even negative.
Primary, Secondary and other higher order variables:
In displacement-based FE analysis, the primary variables are displacements. These variables occur in the total
potential energy (or virtual work) function and are the ones analysts obtain from the FE solution. The secondary
variables (stresses and strains) are those that are derived by differentiating the primary variables (displacements). As
such, in a finite element solution, the secondary variables are less accurate than the primary variables 1,3. In plate
bending problems, transverse shear resultants Qx and Qy – see text box below – derived from the moment equations
of equilibrium4 are inaccurate1-3. Very often analysts appear to be ignoring this issue of accuracy of the secondary and
other higher-order variables from the FE analysis and are reporting margins based on these results, which are known
to be inaccurate.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Transverse Shear Resultants

If Kirchhoff plate theory is used, there are no transverse shear strains (plane sections remain
plane and are normal to the midplane), and hence transverse shear resultants Q x = Qy = 0.
However, if one integrates the in-plane equations of equilibrium, one could evaluate the
transverse shears and they are accurate. If Mindlin plate theory is used, then the transverse
shear strains can be derived from the displacement field approximations. The shear stresses
then can be evaluated using the constitutive relationships; however, these shear stresses are
inaccurate. To improve the accuracy of the transverse shear stresses, one needs to integrate
the in-plane equations of equilibrium. As such, the accuracy of Q x and Qy from Mindlin
formulations can be poor unless one takes extra precautions and determines them by
integrating the equations of equilibrium12.

Bolts and Bolt Modeling:

Bolts and joints are complex structural configurations. There is no universally accepted process for modeling bolts
in the joints. Some typical modeling strategies13 for bolted structures include:
1. Simply smearing the thickness of the fastened parts together and ignoring the discrete bolts.
2. Introducing discrete constraints or beam element at every fastener location and ignoring features of the bolt.
3. Including the fastener as one or more beams to simulate the bolt shank with sets of constraints to simulate
the bolt head and nut (i.e., beam and spider approach) usually coupled with an explicit model of the fastener
hole.
4. Detailed 3D modeling of the bolt head, shank, washer(s), and nut including contact and friction 14.
In each strategy, the analyst needs to consider the modeling of the parts being fastened together as well as the
fastener modeling details. Some of the major bolting issues that are commonly ignored are bolt pretension, friction,
and contact. For example, bolt pretension (preload) is not usually included and most analysts lack knowledge of how
a preloaded bolted joint works. The preload, can be accounted for by inducing a local strain field in a bar element that
represents the bolt (often with an imposed temperature on just those bar elements). This can be difficult to model
accurately especially since usually what is specified is bolt torque and not the actual tension value.
As mentioned previously, most analysts prefer 3D modeling throughout the analysis region. While the 3D
modeling approach looks easy, complexities associated with contact, friction, interference fits, and preload application
make this approach very unreliable. The unreliability stems from the assumptions made for and implementation of
friction, contact, and interference. Friction of mating parts is also usually ignored at least in initial analyses. Contact
solutions can account for some magnitude of friction, but getting the correct distribution is usually a complex and
nonlinear behavior that is well beyond most everyday user capabilities. Most analysts appear to be unaware of all
these pit falls in bolt modeling and lack knowledge of how a preloaded bolted joint works.
Linear and Nonlinear Analyses:
Linear analysis is based on small deflections, small rotations, infinitesimal strains, and linear elastic stress-strain
law assumptions. The advantages of linear analyses are: simple and direct solutions can be obtained with minimal
computational time and effort; solutions can usually be linearly scaled to obtain solutions for a different magnitude of
loading; various linear solutions can be superposed (superposition); material constants required to described the
constitutive behavior can be kept to a minimum1-3,15-18. One can still use advanced complex materials (fully
anisotropic) which would have 21 elastic constants but the nonlinearity part of those constants (usually defined in the
form of tabulated nonlinear behavior) would not be needed.
Consider a beam undergoing bending as shown in Fig. 10. Linear small deflection analysis implies that ds ~ dx
and slopes (dw/dx) are small; tan ~ sin  ~ When the calculated displacements are higher than half the thickness,
7
American Institute of Aeronautics and Astronautics
 the small deflection assumption is no longer valid and nonlinear analysis is needed. Note that from Fig. 10, the change
in length of the initial differential beam segment dx changes to ds. The change in length is then calculated as:

𝑑𝑤 2
𝑑𝑠 2 = 𝑑𝑥 2 + 𝑑𝑤 2 = 𝑑𝑥 2 [1 + ( ) ] Eq. (2)
𝑑𝑥

(1/2)
𝑑𝑤 2 1 𝑑𝑤 2
𝑑𝑠 = 𝑑𝑥 [1 + ( ) ] ~ 𝑑𝑥 [1 + ( ) ] Eq. (3)
𝑑𝑥 2 𝑑𝑥

In writing Eq. (3), the cubic and

z, w
higher-order terms are neglected4. When Undeformed beam
the slope (dw/dx) is not small, the second dx
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

term in the last equation above indicates x

that the line segment is being stretched
and needs to be accounted for in a large dx
Deformed beam dw
deflection analysis, which makes the ds
analysis geometrically nonlinear. Figure 10. Beam undergoing deformations.
As mentioned previously, the
majority of the present-day analyses are linear analyses. Nonlinear analyses are becoming more and more popular for
several reasons15-18: higher performance and light-weight structures are pushing the limits, as such every analyses that
takes advantages of all the material and structural efficiencies are being pursued; nonlinear analyses are used to assess
continued use of and to expand the capabilities of the structures already in service and extend service life; and to
identify efficiencies and simulate manufacturing processes of aerospace structural components.
Nonlinearities can be due to material constitutive behavior, geometric deformations, and due to boundary
interactions (e.g., contact and friction). Some characteristics of nonlinear analyses are: principle of superposition
cannot be applied, only one load case can be applied at a time, solutions are path dependent (i.e., loading sequence is
important). In other words, one cannot scale and obtain the solution of a particular load case from a different load
case. Also, note that the nonlinear analysis solutions can be non-unique and dependent on solution procedure
parameters used to solve the nonlinear algebraic equations.
Nonlinear analyses offer many more opportunities for analysts to misinterpret results and as such one needs to be
more vigilant when performing nonlinear analyses. Also, note that many analysts are unaware that different stress and
strain measures may crop up in nonlinear geometric analyses. Depending on the element/material combinations 2 nd
Piola-Kirchhoff stresses and Green-Lagrange strains may need to be used3,15-18. Often, the analysis tool reports
engineering stresses and strains as the default option and analysts need to know and understand what output is
provided.
Inertia Relief 19,20:
Software manufacturers have incorporated techniques, such as inertia relief, to facilitate easy analyses of aerospace
components subjected to inertial forces, such as accelerations and gravitational loads. In the conventional inertia relief
calculations, reference points or special points are usually chosen where rigid body motions are prescribed. For
example, center of gravity of traditional aircrafts is usually chosen as reference support point for inertia relief
calculation; for rockets, center of gravity or center of buoyancy are usually used for these support location. Support
conditions up to 6 degrees of freedom (DOF) are needed to remove all rigid body motions and normalize the
displacement field. For conventional inertia relief, resultant forces at the reference support node need to be zero
(machine zero in the actual computations) for an acceptable solution. Automatic inertia relief is available in some of
the software packages. Very often an inertia relief analysis is performed without understanding the theory behind the
analysis and the resultant contour plots of displacements and stresses are incorrect. Very frequently, the analysts have
failed to perform simple checks to ensure that the presented results did not violate common sense.

III. Verification and Validation

Two issues that are very often overlooked today are Verification and Validation (V&V) of finite element models.
Analysis of any physical problem needs to be subjected to V&V. The definitions for V&V are taken from the NASA
Standard for Models and Simulations, NASA-STD-7009 (Ref. 21). Similar definitions can also be found in references
21-25.

8
American Institute of Aeronautics and Astronautics
 Verification is the process of determining that a computational model accurately represents the underlying
mathematical model and its solution, i.e. verification is ensuring the computational model is correct in terms of the
governing equations (stress, strain, motion) and input data. Verification generally involves two different aspects. One
aspect relates to code verification and is primarily the responsibility of the code developer. The other aspect relates to
calculation or model verification and is primarily the responsibility of the user.
The code verification activity focuses on identifying and removing errors in the software implementation of the
finite element model. Code verification and calculation verification are the two parts of this process. Code verification
is often performed by the code and software developers and focuses on making the computer software error free.
Calculation or model verification focuses on the removal of errors introduced during operation and use of the
software and is the responsibility of the analysts. Model verification may involve comparisons of highly accurate or
exact solutions of a mathematical model and includes an assessment of discretization errors, input data errors, and the
overall numerical solution of the finite element representation. The analysts need to capture errors due to missing
elements, disconnected nodes, improper material assignments, consistency of various coordinate systems, boundary
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

and interface conditions, mechanical, thermal, and inertia loadings, etc. in their models and that the model is
appropriate for its intended use. Ensuring that the finite element solution will yield accurate results by successively
refining the mesh is a necessary part of this verification activity. In addition, analysts should ensure that the software
correctly yields accurate solutions for known problems instead of blindly accepting the general-purpose software
without any independent assessment. This can be achieved by running verification problems provided by a software
developer and problems for which exact solutions4,26 are available. Such a process verifies the tool’s function within
the analyst’s computing environment, verifies the mathematical modeling and idealization process, and provides a
check on the analyst’s modeling process. This process is referred to as model verification and involves building the
model right.
Validation, on the other hand, is the process of determining the degree to which a model is an accurate
representation of the real world from the perspective of the intended uses of the model, i.e. validation is ensuring the
modeling effort captures the physics of the intended application. Validation is the process for testing the model’s
predictive capabilities of the physical reality being modeled. It has been recommended that validation experiments be
designed and performed in concert with the analysis effort to ensure consistent objectives and intended uses of the
results extracted from the analysis.
The goal of validation is to increase confidence in the predictive capability of the models. This is achieved by
quantifying uncertainties and errors by comparing the results from the computational model to experimental
measurements. Validation is application specific, as a model is validated and not the code. The use of building block
approaches for testing and analysis is a key aspect of model validation involving the development of higher-level or
system models using test-validated models of sub-elements of the structure. Model validation is concerned with
building the right model for the intended use of the model.
While most analysts pay little attention to V&V, it is very important to perform verification of all the FE models
to the extent possible and validate the model with test data. Figure 2 of Ref. 24 discusses in-depth the model V&V
process.
The NASA Standard 700921 is developed so that the decision makers can rank analyses performed by various
analysts and make rational decisions. The standard is new and experience with the use of this standard is limited. It
will take a few years for analysts and decision makers to gain experience in actually implementing and using this
standard effectively. Various professional societies are working on V&V (see Refs. 22,23).

IV. Guidelines and Suggestions

In this section, some guidelines and suggestions are offered to the analysts, senior engineers, and educators. These
guidelines and suggestions should be useful in avoiding pitfalls discussed in Section II.
Most finite element packages offer suggestions in their user manuals to ensure proper use of their software. Some
of the general checks are discussed next for analysts, senior engineers, and educators.

A. Analysts:
General Checks:
 Most FEA packages have element distortion checks, and will warn the user if elements exceed those limits. These
limits are subjective, and a ‘bad’ element might not give erroneous results and a ‘good’ element might not give
accurate results.

9
American Institute of Aeronautics and Astronautics
  Check element normals – all normals need to point in the same direction to ensure proper nodal numbering and
to insure the proper application of distributed pressure type loading whose direction is referenced to element
normals.
 Use shrink plot to check connectivity and missing elements
 Apply a dummy load which will exercise all components of the model, such as unit inertial loads, solve, and then
view displacements to check for mechanisms, spurious constraints and the like.
 Add density and perform a free-free modal analysis. In 3D analyses, six rigid body (zero frequency) modes should
result. If more or less than of this number occur, the model most likely has mechanisms, excessive internal
constraints or is a planar model. Mode shapes may reveal places to look for these mechanisms and strain energy
summaries may identify areas where internal constraints have developed.
 Large differences in stiffnesses in the same model can cause problems.
 A simple, necessary but not sufficient, test for geometric nonlinearity is to impose the full load and solve the
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

problem as a nonlinear system. If the solution converges in one or two iterations, the structural response is
probably linear. If the maximum stresses are lower than the yield stress, then the problem is probably elastic.
Tricks of the trade:
Obtain as much advice as possible from the mentors and educators and learn the tricks of the trade. For example,
most experienced analysts know that visualization can be an invaluable tool. However, to capture many trends, one
must first ensure that the contour plotting and averaging, which are default settings in most tools, are turned off. With
these plots, one can decide on the main features of the solution and decide the regions that need careful scrutiny or
remeshing.
Element Selection:
Numerous elements are available in the
w Rigid Diaphragm
general-purpose programs. Many elements v u u=w=0
have issues and do not perform as expected
(Refs. 1-3, 27-29). Some of the older elements
even converge to the incorrect results or
convergence to the correct solution is very L=
slow. For example, the 3-node triangular 50
bending element with 9 DOF fails the patch f
test; converges to the incorrect result for an
union jack mesh and a load that twists the Rigid Diaphragm 400
R=25 ft.
plate (see Ref. 1, page 321). u=w=0
Another example can be illustrated using
the widely studied problem of Scordelis-Lo Figure 11. Scordelis-Lo Shell Roof:
shell roof problem (see Fig. 11) under its own Loading: Self Weight = 90 lbs/sq.ft (E = 3x106 psi,  = 0, t = 3 in.).
weight. This problem was studied by many
researchers who were developing shell finite elements27,30-34. The vertical displacement at the center of the free edge
of the shell roof was used as the figure of merit for comparison. Some results are presented in Fig. 12 as a function of
the number of DOF in the model (see Ref. 27, pages 12 and 13, and page 446 for numerical data). Note that the number
of DOF in the model is directly proportional to the computing effort and hence used to compare element performance.
As seen from this plot, the accuracy of the elements varies widely and the more recent elements, shown with filled
symbols, give accurate displacement results compared to the older elements. Comparison of stress results are not
available, but convergence is expected to be slower than for displacements.
There are two paths available for the analysts to ensure that the elements that they wish to use do perform as
expected. The first path is to study thoroughly the documentation supplied by the software provider for the elements
in the software library. The analysts should ensure the modeling is performed with elements that perform well within
reasonable ranges of distortions and element aspect ratios.

10
American Institute of Aeronautics and Astronautics
 The second path is instead of 1.6
z
depending on the documentation y x
supplied by the software tool L
1.4
developers, the analysts should =
A5
develop their own experience database. 1.2 0

Normalized Vertical Displacement at A

They should experiment with various
f
element types, mesh configurations,
1 ∂
and material combinations, apply them
to well known classical problems1-4,
26,27
, and study the performance of the 0.8
Strickland & Loden (Ref. 30)
elements. Examples of some such Bonnes et al. (Ref. 31)
problems, MacNeal-Harder problems, 0.6
Clough & Johnson (Ref. 32)
can be found in Ref. 28. Fortunately, it
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Cowper et al. (Ref. 33)

is not a difficult job to select few well- 0.4 Raju et al. (Ref. 34)
performing elements for use and stick
MacNeal-MSC-Quad 4 RI (Ref. 27)
with those elements until one finds 0.2
other good performers to add to one’s MacNeal-MSC-Quad 8R (Ref. 27)

repository. In addition, analysts should

0
develop their own suite of test cases
0 500 1000 1500
that are representative of the problem Degrees of Freedom in the Model
classes they typically model and Figure 12. Comparison of various solutions
analyze. (Vertical displacement at the center of the free edge, A = 3.7031 in.).
Stress (or Strain) Recovery:
Frequently, analysts generate contour plots of stress and strain results where the post processing tool averages
element values across inter-element boundaries. Most FEA tools do not guarantee continuity of secondary variables
such as stresses and strains across inter-element boundaries; however, options are available that result in ‘smooth’
contour plots. Plotting element values often reveals jumps in these secondary variables indicating that the spatial
resolution of the mesh has not converged. In addition, it may be that while some stress component exhibit ‘smooth’
contours, a different component may not.
To reduce post-processing effort and to ensure that all stress components are considered, ‘stress bond’ plots35 based
on un-averaged element stress values have been used to compare different FE solutions and assess mesh convergence.
G/L approaches:
Global/local approaches are very useful in obtaining accurate solutions. As mentioned previously, continuity of
displacements and reciprocity of tractions need to be maintained across the G/L boundaries. Any of the following
strategies can be used to implement the procedure correctly:
a) Remodel the analysis region with finer mesh over the entire analysis region.
b) Remodel the region where large gradients exist (local region) with a much finer mesh, and have an
interface region that transitions from a finer mesh in the local region to a coarser mesh in the global
region (see Fig. 4)
c) Work with the G/L model (as in (b) without a transition region – for example as in Fig. 4 lower part)
and use the procedures outlined in Refs. 7-11 to perform the analysis with interface methodologies.
Most FE software packages provide 2D and 3D interface elements in their element libraries.
d) Use an iterative procedure outlined in Refs. 6 and 36. These procedures are especially helpful in
performing G/L analyses when different parts of the models are being analyzed by two different
software packages (see Ref. 6).
Bolts and Bolt Modeling:
Simply smearing the thickness of the fastened parts together and ignoring the discrete bolts or introducing discrete
constraints or beam element at every fastener location and ignoring features of the bolt are often adequate to model
the overall response of the bolted connection.
The next most common approach used to model mechanically connected parts is to use the beam and spider
approach in which the bolt shank is represented as one-dimensional linear elastic beams, while the bolt head,
washer(s), and nut are represented as sets of kinematic coupling constraints (RBEs) – see Fig. 13. If the fastener is
installed with a specified preload, then analysis procedures to impose the preload need to be developed (e.g., use of
thermal loading to obtain the preload, certain RBE formulations using Lagrange multipliers that can account for
11
American Institute of Aeronautics and Astronautics
 thermal distortion in the element, or some special feature of the analysis tool). The washer-bearing-surface modeling
assumptions are discussed in Ref. 14. Most experienced analysts use ring-type modeling around the bolt hole and
discard the stresses in the two rings that are closest to the hole, as the kinematic coupling spider constraints develop
spurious stress results in those rings.
In critical bolted connections, the bolt head, bolt shank, washer(s), and the nut are explicitly modeled by 3D
elements along with the explicit modeling of the bolt holes. While the 3D modeling approach looks easy, complexities
associated with contact, friction, interference fits, and preload application make this approach very unreliable. The
unreliability stems from the assumptions made for and implementation of friction, contact, and interference.
Extracting forces from the FE model at the bolt location and then performing hand calculations using classical
methods to assess the adequacy of the joint is recommended. The stresses from the FE analysis may be artificially
high due to concentrated or point loading. Bolt forces can be extracted from the FE results and hand calculations of
the local response performed. Analysts need to determine the number of structural members bolted together (e.g.,
single vs. double lap joint) and assess any load sharing simulated by the FE model.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Different size
washers
Figure 13. Beam and spider approach to bolt modeling.
Shell-3D Regions: 3D Region
As mentioned previously, there 
j
are many instances where it is z i
convenient to model most of the uj=ui-z,
analysis region with plate and shell etc. 
elements, while regions requiring
high-fidelity results are modeled
with 3D elements. However, the a)
difficulty then would be to insure b)
the transitions between these 3D Region
regions are performed correctly.
There are many methods used to
perform the connections.
a) Use multipoint constraints to
ensure the continuity of
displacements and reciprocity
of tractions. Figure 14 c) Offset

demonstrates a plate-to-3D Figure 14. Shell-3D strategies. a) Multi-point constraints.

connection between nodes i b) Extension of the shell region.
and j. c) Plate nodes offset to the top of plate.

12
American Institute of Aeronautics and Astronautics
 b) The second most popular method is to use RBE elements.
c) The third method is to use tied constraints that are provided by the software developers. Study the theory manuals
of software developers. The developers also explain restrictions and model recommendations. Try out their test
cases.
d) Figure 14(b) demonstrates another method that is simple and easy to implement. In this method, one continues
the plate/shell region into the 3D region for about two elements deep. While locally the stresses are inaccurate,
one or two thicknesses away the stresses are unchanged compared to those by other methods. As such, this
procedure can be used in regions where higher-fidelity stresses are not needed.
e) Another popular method is to use offsets. Figure 14(c) shows all the plate element nodes offset to the top of the
3D region, as in many applications this surface represents the outer mold line.
Stiffener Modeling:
Aerospace structures are made up of skin-stiffener construction. As such, stiffener modeling is very important.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Many variations are available for stiffener modeling as branched plates by defining appropriate offsets for webs and
flanges etc. Figure 15 shows some of these modeling strategies. Each of these modeling stragegies needs to be checked
out thoroughly and the check-out modeling strategies need to be used consistently.
Mid-planes
T-stiffener Skin T-stiffener
Skin

a) d)

b) e)

c) f)
Figure 15. Various offsets used in stiffener models. a) T-stiffened skin. b) Nodes offset to the top of skin.
c) Nodes at mid planes of skin and stiffener flange and connected by link elements.
d) Skin, flange, and web modeled by their mid-planes. e) Skin, skin-flange, and web modeled.
f) Skin, flange, and skin-flange-web modeled.

B. Senior Engineers and Educators:

“There is nothing so wrong with analysis as believing the answer! Uncertainties appear
everywhere ……….. When using a mathematical model careful attention must be given
to uncertainties in the model …..” Richard Feynman, Appendix F, Vol. II, Rogers
Commission Report on Space Shuttle Challenger Accident, Washington, D. C., June 6, 1986.

“One can now make mistakes with more confidence than ever before.”…….... “Although
the finite element method can make a good engineer better, it can make a poor engineer
more dangerous” – R. D. Cook, et al. (Ref. 1, p. 24)

13
American Institute of Aeronautics and Astronautics
 Senior engineers and educators consciously need to educate and mentor junior and freshman engineers as a key
aspect of their legacy to the engineering profession. Some strategies †† are suggested here:
Courses: Recognize that students need to have strong prerequisite courses such as Strength of Materials, Advanced
Strength of Materials, Theory of Elasticity, and Plates and Shells. They also need two FEA theory courses - one basic
and another advanced. These courses represent minimum requirements for those performing FEA.
Classical Methods: Teach them and insist that students learn the classical methods. Recognize that it may take
many years to become proficient in these methods. Also, understanding the problems and methods used to analyze
and solve these problems will enhance the analysts’ ability to ask the right questions at the right time.
Hand calculations: There are many problems that can be solved using classical methods utilizing beam, plate, and
shell theories. If there are classical solutions available for a reasonable model of the structure, those classical solutions
should be used. Aerospace companies have analysis manuals that capture application of these methods to aerospace
structures. The use of classical hand analyses to verify the finite element results is critical. Development of hand
analyses, bounding analyses, skills, and tools need to be taught.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Demonstrate Evaluation of FE results: Develop skills needed to evaluate the results of a FE analysis. Engage them
in group discussions of their analyses and results. Teach by example. Recognize that these skills cannot be learned
overnight and it will probably take a number of years to become proficient in this endeavor. Modeling is a life-long
learning process as the analyst matures and tools advance.
Best Practices: Many textbooks offer many best practices (see Ref. 1, Chapter 19 and Ref. 27). Journal papers,
conferences, and workshops are other sources for self learning.

V. Concluding Remarks
Aerospace structures are complex and are high-performance structures. As such, there is an increased reliance on
modeling and analysis and considerable emphasis on accurate analyses of these structures. Advances in reliable and
efficient computing and modeling tools are enabling analysts to model complex three dimensional (3D) configurations
and perform analysis rapidly and efficiently. Many of the early-career engineers of today are very proficient in the
usage of modern computers, computing engines, and complex software and visualization tools. However, the current
trends also suggest that there is blind acceptance, at many levels, of the results of the finite element analysis. There
appears to be a general lack of understanding of basic assumptions in engineering mechanics. The current-day analysts
are adept at meshing but not developing a well-thought-out plan for modeling and analysis. Most of the pitfalls are
observed in arriving at proper boundary and interface conditions, in regions of plate/shell interfaces, singularities, and
bolts and bolt modeling. Improper use of primary and secondary variable results, linear and nonlinear analyses and
inertia relief methods are also frequently observed.
To overcome these trends, guidelines and suggestions are offered to analysts, senior engineers, and educators.
Analysts need to study software developers’ manuals, actively pursue verification and validation of their finite element
models, experiment with various elements to develop their own library of elements with good performance
characteristics. Senior engineers and educators need to ensure that the young engineers receive proper grounding in
classical methods, finite element theory, simple yet bounding models, hand calculation techniques, and methods for
evaluating finite element results.

VI. References
1
Cook, R. D., Malkus, D. S., and Plesha, M. E., Concepts and Applications of Finite Element Analysis, Third
Edition, John Wiley & Sons, 1989.
2
Zienkiewicz, O. C. and Taylor, R. L., The Finite Element Method, Vols. 1 and 2, Fourth Edition, McGraw Hill
Publishers, 1989.
3
Bathe, K. J., Finite Element Procedures, Prentice-Hall, Englewood Cliffs, NJ, 1996.
4
Timoshenko, S. P. and Woinowsky-Krieger, S., Theory of Plates and Shells, 2nd Edition, McGraw-Hill Book
Company, New York, 1959.
5
Raju, I. S., Knight, N. F., Jr., Song, K., and Phillips, D. R., “Fracture Mechanics Analyses of the Slip-Side Joggle
Regions of Wing-Leading-Edge Panels”, Frattura ed Integrità Strutturale, Vol. 15, 2010, pp. 35-49.

††
See also the references quoted in this footnote. The authors came across three excellent presentations/documents
that offer invaluable advice: T. Rose, “Your Answers are Wrong!!!”, Presentation to the combined Loads & Dynamics
and Structures NESC Technical Discipline Teams, April 14,2014; L. Procter, “Modal checkout in MSC. NASTRAN”,
MSC Lunch-n-Learn Series, April 2008; ANSYS: FEA-Best Practices, Document can be downloaded from:
http://innomet.ttu.ee/martin/MER0070/Loengud/FEA_Best_Practices.pdf
14
American Institute of Aeronautics and Astronautics
 6
Raju, I. S., Glaessgen, E. H., Mason, B. H., Krishnamurthy, T., and Davila, C. G, “Structural Analysis of the
Right Rear Lug of American Airlines Flight 587”, Computer Modeling in Engineering and Sciences, Vol. 22, pp.1-
30, 2007.
7
Aminpour, M. A., Ransom, J. B., and McCleary, S. L., “A Coupled Analysis for Structures with Independently
Modeled Finite Element Subdomains,” International Journal for Numerical Methods in Engineering, Vol. 38, 1995,
pp. 3695-3718
8
Ransom, J. B., McCleary, S. L., and Aminpour, M. A., “A New Interface Element for Connecting Independently
Modeled Substructures,” AIAA Paper No. 93-1503, Proceedings of the AIAA/ASME/ASCE/AHS/ASC 34 th Structures,
Structural Dynamics, and Materials Conference, 1993, pp. 1693-1703.
9
Ransom, J. B., “Interface Technology for Geometrically Nonlinear Analysis of Multiply Connected Subdomains,”
AIAA Paper No. 97-1298, Proceedings of the AIAA/ASME/ASCE/AHS/ASC 38th Structures, Structural Dynamics, and
Materials Conference, 1997, pp. 1862-1872 .
10
Ransom, J. B. and Knight, N. F., Jr., “Global/Local Stress Analysis of Composite Panels,” Computers and
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

Structures, Vol. 37, No. 4, 1990, pp. 375-395.

11
Farhat, C. and Roux, F. X., “A method of finite element tearing and interconnecting and its parallel solution
algorithm”, International Journal for Numerical Methods in Engineering, Vol. 32, 1991, pp. 1205-1228.
12
Hartman, T. B., Hyer, M. W., Case, S. W., “ Stress Recovery in Composite Laminates”, Paper presented at the
52nd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, Denver, Colorado,
AIAA Paper No. 2011-1856, April 2011.
13
Kim, J., Yoon, J.-C., and Kang, B.-S., “Finite element analysis and modeling of structures with bolted joints,”
Applied Mathematical Modeling, Vol. 31, pp. 895-911, 2007.
14
Knight, N. F., Jr., Phillips, D. R., and Raju, I. S., “Simulating Structural Response of a Preloaded Bolted Joint,”
Paper presented at the 49th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference,
Schaumburg, Illinois, AIAA Paper No. 2008-1842, April 2008.
15
Hinton, E. (Ed.), NAFEMS: Introduction to Nonlinear Finite Element Analysis, NAFEMS, Glasgow, UK, 1992.
16
Crisfield, M. A., Nonlinear Finite Element Analysis of Solids And Structures, - Volume 1, John Wiley & Sons,
Chichester, 1991.
17
Reddy, J. N., An Introduction to Nonlinear Finite Element Analysis, Oxford University Press. New York, 2004.
18
Belytschko, T., Liu, W. K., Moran, B., and Elkhodary, K. I., Nonlinear Finite Elements for Continua and
Structures, Second Edition, John Wiley & Sons, 2013.
19
Barnett, A. R., Wildrick, T. W., and Ludwiczak, D. R., “Closed-form static analysis with inertia relief and
displacement-dependent loads using a MSC/NASTRAN DMAP Alter”, 1995 World User’s Conference, MacNeal-
Schwendler Corporation, Los Angeles, CA, May 8-12, 1995, Also available as NASA TM 106836, Jan. 1995.
20
Liao, L., “A study of inertia relief analysis”, Paper presented at the 52nd AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics and Materials Conference, Denver, Colorado, AIAA Paper No. 2011-2002, April
2011.
21
Anonymous, NASA Standard for Models and Simulations, NASA-STD-7009, National Aeronautics and Space
Administration, Washington, D.C., July 2009.
22
Anonymous, AIAA, Guide for the Verification and Validation of Computational Fluid Dynamics Simulations,
American Institute of Aeronautics and Astronautics, AIAA-G-077-1998, Reston, VA, 1998.
23
Anonymous, ASME, Guide for the Verification and Validation of Computational Solid Mechanics, American
Society of Mechanical Engineers, ASME-V&V-10-2006, New York, 2006.
24
Thacker, B. H., “The Role of Nondeterminism in Verification and Validation of Computational Solid Mechanics
Models”, Society of Automotive Engineers, Paper 2003-01-1353, 2003. (See also: Thacker, B.H., Anderson, M.C.,
Senseny, P.E., Rodriguez, E.A , “The Role of Nondeterminism in Computational Model Verification and Validation,”
Int. J. Materials and Product Technology, Vol. 25, 2006, , pp 144-163.)
25
Roache, P. J., Verification and Validation in Computational Science and Engineering, Hermosa Publishers,
Albuquerque, NM, 1998.
26
Timoshenko, S. P. and Goodier, J. N., Theory of Elasticity, McGraw-Hill Book Company, 1970
27
MacNeal, R. H., Finite Elements: Their Design and Performance, Marcel Dekker, Inc., 1994.
28
MacNeal, R. H., and Harder, R. L., “A Proposed Standard Set of Problems to Test Finite Element Accuracy,”
Finite Elements in Analysis and Design, Vol. 1, 1985, pp. 3-20.
29
Sun, E. Q., “Shear Locking and Hourglassing in MSC Nastran, ABAQUS, and ANSYS,” Paper 27. Presented at
the MSC Software Virtual Product Development Conference, (VPD 2006), Huntington Beach, CA., July 17-19, 2006.
30
Strickland, G. E., and Loden, W. A., “A Doubly-Curved Triangular Shell Element”, Proc. 2nd Conf. on Matrix
Methods in Structural Mechanics, Wright-Patterson AFB, Ohio, 1968.
15
American Institute of Aeronautics and Astronautics
 31
Bonnes, G., Dhatt, G., Gironn, Y. M., and Robichand, P. A., “Curved Triangular Elements For The Analysis Of
Shells”, Proc. 2nd Conf on Matrix Methods in Structural Mechanics, Wright-Patterson AFB, Ohio, 1968.
32
Clough, R. W., and Johnson, R. L., “A Finite Element Approximation For The Analysis Of Thin Shells”, Int.
Jnl. Solids and Structures, Vol. 4, 1968, pp. 43-60.
33
Cowper, G. R., Lindberg, G. M., Olson, M. D., “A Shallow Shell Finite Element of Triangular Shape”, Int. Jnl.
Solids Structures, Vol. 6, 1970, pp. 1133-1156.
34
Raju, I. S., Venkateswara Rao, G., Prakasa Rao, B., and Venkataramana, J., “A Conical Shell Finite Element”,
Computers and Structures, Vol. 4, 1974, pp. 901-915.
35
Sussman, T., and Bathe, K. J., “Studies of Finite Element Procedures – Stress Band Plots and the Evaluation of
Finite Element Meshes”, Engineering Computations, Vol. 3, Sep 1986, pp. 178-191.
36
Whitcomb, J. D., and Woo, K., “Application of Iterative Global/Local Finite Element Analysis. Part 1: Linear
Analysis, Part 2: Geometrically Non-Linear Analysis”, Communications in Numerical Methods in Engineering, Vol.
9, 745-756 (Part 1), 757-766 (Part 2), 1993.
Downloaded by KUNGLIGA TEKNISKA HOGSKOLEN KTH on March 4, 2016 | http://arc.aiaa.org | DOI: 10.2514/6.2015-2070

16
American Institute of Aeronautics and Astronautics