NX Nastran 10

WHITEPAPER
NX Nastran Model Reduction
and Superelements:
Theory and
Implementation

www.ata-plmsoftware.com
844-756-7638 (844-PLM-SOFT)
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

Overview
When modeling complex systems in NX
Nastran, detailed models of components
are often available. In many cases it is not
Software: practical to include the full model of every
NX Nastran 10
component in the system. NX Nastran
superelements automate the process
of creating reduced representations of
components and efficiently assembling
them into a system model. The purpose of
this whitepaper is to provide background
on superelements and describe the
three primary ways of creating and using
superelements in NX Nastran: main bulk
data, part, and external superelements.

This whitepaper is part of a series of free Siemens

PLM Software training resources provided by ATA.
For more whitepapers, tutorials, videos, and
macros, visit ATA’s PLM Software website:
http://www.ata-plmsoftware.com/resources.

Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

Reasons to Consider Using Superelements

Below are a number of situations in which to consider using a superelement
approach to develop a system model.

1. An assembly of all component models representing a system is too large

2. Some component models are proprietary
3. Design studies will be performed where most of the system remains
unchanged
4. Nonlinearities in a system can be isolated to a small region
5. Models for a system do not follow unique numbering schemes

Superelement Reduction
The concept behind superelement reduction is to take a finite element model (FEM)
of a component and partition it into a set of interior and exterior nodes as illustrated
in Figure 1. The FEM of the component is then mathematically reduced to just the
exterior nodes, before assembling it into a model of the system. We will refer here to
interior and exterior degrees of freedom (DOF), which are the three translations and
three rotations at each node.

Figure 1: ▶
A superelement is partitioned into exterior
and interior nodes, with the reduced
model including only the exterior nodes.

1 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

The mathematical process used by NX Nastran to reduce a FEM to a small

number of exterior nodes (or DOF) is referred to as static or Guyan reduction.
The reason that it is called static reduction is that it depends only on the static
solution (i.e., the stiffness matrix). The partitioning of the static solution is as
follows:
(1)

The interior deflections {xO} can be calculated from the exterior deflections {xT}
and the interior forces {fO} as follows:

(2)

The first term, multiplying the exterior deflections, is often referred to as the
constraint shapes since they are the shapes that would be calculated by
deflecting each exterior DOF while holding all others fixed. These shapes can
then be used to reduce the equations down to just the exterior DOF as follows:

(3)

The full equations for the component, therefore, have been replaced with
equations on only the exterior DOF. This means that for a component that might
potentially be represented by millions of DOF, but with only a few dozen exterior
DOF, the equations can be reduced to just the few dozen DOF.

While the static reduction is developed entirely using the stiffness matrix, it
can be applied to the mass matrix as well. This captures the rigid body mass
but does not capture internal dynamics, meaning modes of vibration that
would be experienced if all exterior DOF were held fixed. To capture the interior
dynamics, NX Nastran superelements introduce a concept called component
mode synthesis (CMS). CMS appends additional generalized DOF describing
the internal modal dynamics of the component. Using NX Nastran terminology,
these are referred to as the Q-set or {xQ }. This adds an additional term to Eq. 2:

(4)

Note that {xQ} is not a subset of the DOF of the originally unreduced component
FEM but rather additional generalized DOF that are associated with component
mode shapes [ΦO ]. If the component modes are calculated with the exterior
DOF held fixed, the resulting reduced model is referred to as either a Craig-
Bampton (CB) or Hurty/Craig-Bampton (HCB) model. NX Nastran provides a
minor generalization of HCB models since selected DOF can be left free, but for

2 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

the purposes of this whitepaper we will consider only the HCB case. The reduced
equations for an HCB model are

(5)

While CMS is an approximate solution to the dynamics of a component, it is

typically very accurate as long as the component modes are calculated to 1.5–2
times the frequency range of the system modes.

Both static and dynamic (CMS) reductions are supported by NX Nastran

superelements. Static reductions are appropriate for static solutions and
also for dynamic solutions where the internal dynamics of a component are
unimportant or well above the frequency range of the system analysis, or
where the internal dynamics are deliberately neglected. Dynamic reduction is
applicable to a very wide range of dynamic problems, only requiring that the
component remain linear.

General Rules for Superelement Reduction

The reduction of a component model to a superelement can dramatically
reduce the number of DOF (or equations) required to represent that component.
However, the reduction step itself can be computationally expensive, and the
resulting matrices are typically much denser (many fewer zero terms) than the
original equations. Because of this, superelements only provide an advantage
in practice when the number of interface DOF, {xT}, is much less than the number
interior DOF, {xO}. This requires that the interfaces for superelements be chosen
carefully to constitute as small a number of DOF as possible. Typically, having
more than about 100 nodes on the exterior of a superelement results in poor
numerical behavior, though there are exceptions.

Superelements and Residual Vectors

Equations (2) and (4) not only provide means for reducing the equations of a
component but also for recovery of internal deflections, {xO}. For a dynamically
reduced component, these equations are typically referred to as mode
displacement data recovery, since the internal displacements are recovered
using the modal displacements rather than the modal accelerations. Mode
displacement data recovery suffers from a well-known modal truncation issue.
Even though the modes may accurately represent the dynamic response of the
component in a particular frequency range, they do not necessarily capture
the internal quasi-static response. That is the response when the component

3 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

experiences a steady-state component in an applied load or acceleration. To

correct for this, NX Nastran adds residual vectors that capture the quasi-static
contribution of all the truncated modes to the applied loads. By default, NX
Nastran calculates residual vectors for internally applied loads and for six rigid
body acceleration components. Because of this, the number of modes used
to represent a component with residual vectors is equal to the number of fixed
interface modes in the frequency range of interest, plus the number of internally
applied loads, plus six for rigid body acceleration. Residual vectors are crucial for
the accurate recovery of all internal responses including displacements, element
forces, stresses and strains, grid point forces, and multipoint constraint forces,
and while it is possible to deactivate them, this is not recommended.

Main Bulk Data Superelements

NX Nastran offers three different methods for generating superelement models;

main bulk data superelements, external superelements, and part superelements.
While the mathematics of superelement reduction are identical across all three
methods, the user interface is very different, and each one may be the best
choice depending on the problem at hand. The most mature of these methods
is referred to as “main bulk data superelements,” because all the superelements
reside in a single NX Nastran model (the main bulk data). In this case, the user
enters a small number of additional inputs that specify how the model is
partitioned. NX Nastran automatically partitions the model, reduces each of
the components, assembles and solves the system, and then propagates the
solution to the interior of all the superelements, as illustrated in Figure 2.

Figure 2: ▼
Main bulk data superelements take a top-down approach. The full model is partitioned, each component solved, and the system is
then reassembled and solved and the results propagated into the individual superelements, potentially in a single run.

4 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

A main bulk data superelement model of a system is identical to a non-

superelement model, with the further restriction that all element numbers
must be unique across the entire model. The partitioning into superelements
is specified by identifying the interior nodes on SESET cards. By using
compact numbering ranges a single SESET card can be used to identify each
superelement. That is all that is required for statically reduced superelements.
Dynamically reduced superelements must also have DOF defined for
component modes (typically SPOINTs). The structure of a typical main bulk data
superelement input file is illustrated in Figure 3.

Figure 3: ▶
Structure of a main bulk data
superelement input deck.

This model contains a single superelement (SE 100) whose internal nodes are
1–18, 40, and 51–54. All other nodes are in the residual structure, which consists
of all parts of the model that are not in a superelement.11 The first SUBCASE
refers to SE 100 and specifies a component mode cutoff of 150 Hz, while the
second SUBCASE refers to the residual and specifies a system mode cutoff of
70 Hz. The SPOINT and SEQSET1 cards specify ten generalized DOF that are used
to store the modal DOF (q-set). Removal of the SESET, SPOINT, and SEQSET1
cards would revert this model back to a FEM of the full system without any
superelement reductions.

One of the advantages of a main bulk data superelement model is that the entire
process can be executed in a single run with minimal alterations to the non-
superelement input file. However, it is also possible to solve each superelement
separately and save results in individual databases by using the SExx Case
Control inputs to control the steps taken in each run. This is described in a later
section.

1 The term “residual structure” in the context of superelements refers to the partition of the FEM that re-
mains after all superelements are reduced. It may or may not contain any elements, but it has nothing to do
with “residual vectors,” which are used to address modal truncation issues.

5 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

The primary disadvantage of the main bulk superelement method is that it

requires that the entire model be made available to NX Nastran with unique
grid and element numbering. Often the reason for using superelements is that
the component FEMs are developed by independent organizations that do not
necessarily adhere to unique numbering schemes and who may want to hide the
component internal details. For this case, an external superelement approach
may be preferred.

Generating External Superelements

External superelements address the case where component FEMs are developed
by different organizations who may only want to share the reduced models
without making the internal details visible. In this case a bottom-up approach is
taken, as illustrated in Figure 4.

Figure 4: ▶
External superelements take a bottom-
up approach to assembling a system.
Each external superelement is solved
independently.

Rather than presenting Nastran with an input file describing the entire system
and then providing SESET cards to partition that system, each superelement is
instead treated as a separate FEM and reduced to a set of matrices in a separate
run. Only the reduced representation, consisting of the interface nodes and the
reduced matrices, is passed on to the system model. The organization doing
the reduction, therefore, only needs to know the location of the exterior nodes
where their structure will interface with the system, and the system analysis
organization only needs to have the exterior nodes and a matrix representation
of each external superelement. This is ideal for situations such as coupled loads
analysis (CLA), where entirely different organizations are responsible for different
components of the system, but it can also provide advantages where a library of
pre-reduced models of various components are desired for system-level analysis.

6 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

To define an external superelement, the user simply identifies the interface DOF
on ASET cards2 and optionally provides generalized DOF for dynamic reduction
using SPOINT and QSET cards. Every DOF that is not identified on an ASET is
treated as an interior DOF. A sample external superelement generation input file
is illustrated in Figure 5.

Figure 5: ▶
Example of an external superelement
generation.

This generates an external superelement with 24 exterior DOF (six at each of four
nodes) and up to 150 modal DOF (including residual vectors). The EXTSEOUT
card in Case Control specifies that the matrices be written in OUTPUT4 format to
Unit 11 and that the superelement be labeled as SE 100. Data recovery matrices
will be generated for the displacement of nodes 18, 40, and 42 and forces in
elements 87–90.

External superelements do support internal data recovery as illustrated in

Figure 5, though all requests must be made at the time that the superelement
is reduced in order to support system results. The output requests can include
displacement, velocities and accelerations (include a DISPLACEMENT request
for those nodes in the reduction), element forces, stresses and strains, grid point
forces, and MPC and SPC forces. These are converted to data recovery matrices,
which are used after the system solution to recover the internal results. As long as
residual vectors are included, the accuracy of the internal responses is typically
excellent.

Internal loads can also be applied to external superelements, though they need
to be specified at the time of reduction and an SELOAD card needs to be used
to associate external superelement loads to system-level loads. Because of this,
the use of internal loads is significantly more complicated when using external
superelements, and it is usually preferable to avoid this as much as possible.

2 Exterior DOF can also be specified on BSET, CSET, BNDFIX, and BNDFREE cards for dynamic reduction. This
affects the boundary conditions used for the component mode calculation.

7 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

NX Nastran offers a large number of formats for storing the reduced matrices.
These include two database formats (DMIGDB and MATDB), an OUTPUT2 format
(DMIGOP2), an OUTPUT4 format (MATOP4), and a DMIG format (DMIGPCH). The
most popular formats are DMIGOP2, MATOP4, and DMIGPCH. DMIGOP2 is a
binary format that stores both the matrices and the exterior geometry. MATOP4
(Figure 5) stores the matrices in OUTPUT4 format (either binary or formatted)
and the exterior geometry in standard bulk data. DMIGPCH stores the matrices in
a DMIG format and the exterior geometry in standard bulk data. The formatted
versions of MATOP4 and DMIG are probably the most commonly used because
the formatted matrices are easy to interpret and read into other programs such
as MATLAB if necessary. The MATOP4 format, in particular, forms the basis of
most CLA, while DMIG3 can be very easily passed to another model without
requiring a superelement license and is therefore the most portable format.

Part Superelements

The previous section describes the generation of an external superelement

reduced model, but to use an external superelement, a part superelement model
is required. An NX Nastran part superelement model is one where the input file is
divided into sections (or parts) using BEGIN SUPER nnn cards, where nnn is the
superelement number. All bulk data between BEGIN BULK and the first BEGIN
SUPER card represents the residual structure, all bulk data between BEGIN
SUPER mmm and BEGIN SUPER nnn represents superelement ID mmm, etc. This
is a bottom-up approach as illustrated in Figure 4. Each part superelement can
be a previously reduced external superelement, or it can be a FEM that will be
reduced by Nastran during the execution of the part superelement run. A typical
part superelement input file is illustrated in Figure 6.

Figure 6: ▶
Structure of a part superelement input
deck.

3 The primary disadvantage of the DMIG option is that the matrices are entered with only 10 digits of
precision, while the other options all use 16 digits of precision. The missing digits often result in artificial
grounding that is picked up by the GROUNDCHECK command in NX Nastran.

8 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

This model consists of a residual structure (gpbus.blk) and two part

superelements (gpsecant.blk and se200.pch). The residual and SE 100 are each
valid models in their own right. Each contains all required grids, elements,
properties, coordinate systems, parameters, etc. In fact, in this case, the residual
uses a PARAM WTMASS of 0.00259, while SUPER 100 uses the default value of
1.0. Each superelement also has its own solution cards such as an EIGRL card to
specify a frequency cutoff. In this case METHOD=70 is called out in Case Control,
but this refers to a 70 Hz cutoff for the system results and a 150 Hz cutoff for the
superelement. The SENQSET card automatically assigns twenty generalized
(q-set) DOF to SE 100 for component modes. In this case, SE 200 is an external
superelement in MATOP4 format. The OUTPUT4 file is assigned to UNIT 11 at the
top of the deck, the superelement is identified in the bulk data using an SEBULK
card, and the BEGIN SUPER 200 and all other required input are in the se200.pch
file, which was generated by NX Nastran when reducing SE 200 as an external
superelement.

Because part superelements are placed in different parts of the input file, there
is no requirement for unique numbering from superelement to superelement.
In fact, the connectivity between superelements depends on geometrically
coincident nodes rather than numbering and is fully automated.4 Unlike external
superelements, however, the entire model is available to NX Nastran so all
reductions, the system solution, and data recovery can be performed in a single
run, though that is not necessary. Another advantage of making all the bulk
data available for each superelement is that application and updates of loads
or additional data recovery requests are all handled automatically and do not
require manual reprocessing of external superelements.

Using Restart with Superelements

When using external superelements, each superelement is reduced

independently and then assembled into a system. Because of this, external
superelements are typically only reduced once. For both main bulk data and
part superelements, on the other hand, the entire process can be executed in a
single run. While this may seem like an advantage, the computational advantage
of superelements can typically only be realized by avoiding reducing the
superelements every time. Because of this, it is typically critical to use existing
databases to avoid repeating superelement reductions. There are two ways
to do this. The first is to use a DBLOCATE to attach an existing database with a
superelement solution. This is illustrated in Figure 7.

4 NX Nastran offers a number of options for controlling how part superelements are connected. These vary
from a fully automated approach based on finding nodes within some distance of each other to completely
manual connection using SECONCT cards.

9 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

Figure 7: ▶
Superelement restart using DBLOCATE.

In this case, the database se100.MASTER was created in a previous run and
contains a reduced model of SEID 100. The SEALL = 1000 card indicates that
superelement operations will only be performed on SEID 0 (the residual) and
SEID 200, assuming that SEID 100 has already been reduced. This will cause NX
Nastran to locate all required datablocks for SEID 100 in the existing database
rather than repeating the reduction.

An alternative method is to use NX Nastran’s automated RESTART capability.

RESTART works independently of superelements and works out what parts of
a previous solution need to be repeated based on changes in the input file. It
is particularly powerful in the context of superelements because if nothing has
changed in a superelement, the reduction will not be repeated, but if there are
changes to a superelement such as new loads or data recovery requests, the
minimum steps required to handle these will be automatically applied. There
are two ways of performing an automated RESTART in NX Nastran. The first uses
just an existing database and a RESTART card. This will create a new “version”
in the database with the updated results. The downside of this approach is that
if there are any errors in the restart run, the database can very easily become
corrupted and unusable for further restarts. As such, a better approach is to treat
the original database as “read-only” and save new results in a new database file.
An example of this is illustrated in Figure 8.

Figure 8: ▶
Superelement restart using RESTART
LOGICAL.

10 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 ATA Engineering NX Nastran 10

NX Nastran Model Reduction and Superelements: Theory and Implementation

In this case, NX Nastran will compare the updated and original Case Control
and Bulk Data on a superelement-by-superelement basis and determine what,
if anything, has changed in each superelement. It will automatically decide
which steps on which superelements need to be repeated. The /,1,9999999 card
is required to delete the old bulk data and replace it with the new. Without this
card, NX Nastran will append the new bulk data to the old and generate multiple
errors due to duplicate cards. Alternatively, if no changes are made to Bulk
Data, or cards are only added and not replaced, only the new cards need to be
included and /,1,99999999 is left out.

Conclusion

If used correctly, superelements can provide a very powerful tool for analyzing
complex systems or sharing component models among organizations. NX
Nastran provides three different ways of developing a superelement model, each
with its own advantages and disadvantages: main bulk data superelements,
external superelements, and part superelements. Much more information
can be found in the Superelement User’s Guide found with the NX Nastran
documentation or by taking a two-day superelement course taught by ATA
Engineering.

11 Content subject to change without notice. © 2016 ATA Engineering, Inc. NX Nastran is a trademark of Siemens PLM Software, Inc.
 www.ata-plmsoftware.com

ATA Engineering

Copyright © ATA Engineering, Inc. 2015 www.ata-plmsoftware.com

Corporate Headquarters 13290 Evening Creek Drive South, Suite 250, San Diego, California 92128 Phone 858.480.2000
Los Angeles Regional Office 1960 East Grand Avenue, Suite 560, El Segundo, California 90245 Phone 424.277.5689
Northwestern Regional Office 1026 Lancaster Way SE, Sammamish, Washington 98075 Phone 425.677.2200
Rocky Mountain Regional Office 1597 Cole Boulevard, Suite 375, Lakewood, Colorado 80401 Phone 303.945.2375
Southeastern Regional Office 1500 Perimeter Parkway NW, Suite 285, Huntsville, Alabama 35806 Phone 256.850.3850
Eastern Regional Office 13921 Park Center Road, Suite 340, Herndon, Virginia 20171 Phone 703.225.7434