Advanced Analysis Techniques Guide Page 1 of 207

Advanced Analysis Techniques Guide

Prev Next

Advanced Analysis Techniques Guide

Table of Contents

1 Design Optimization
1.1 What Is Design Optimization ?
1.2 Basic Definitions
1.3 About GUI Paths and Command Syntax
1.4 How to Do Design Optimization
1.4.1 Create the Analysis File
1.4.2 Establish Parameters for Optimization
1.4.3 Enter OPT and Specify the Analysis File
1.4.4 Declare Optimization Variables
1.4.5 Choose Optimization Tool or Method
1.4.6 Specify Optimization Looping Controls
1.4.7 Initiate Optimization Analysis
1.4.8 Review Design Sets Data
1.5 Multiple Optimization Executions
1.5.1 Restarting an Optimization Analysis
1.6 Optimization Techniques
1.6.1 Subproblem Approximation Method
1.6.2 First Order Method
1.6.3 Random Design Generation
1.6.4 Using the Sweep Tool
1.6.5 Using the Factorial Tool
1.6.6 Using the Gradient Evaluation Tool
1.7 Guidelines on Choosing Optimization Variables
1.7.1 Choosing Design Variables
1.7.2 Choosing State Variables
1.7.3 Choosing the Objective Function
1.8 General Guidelines
1.8.1 Generating the Analysis File
1.8.2 Fixing Design Variable Values After Execution
1.8.3 Modifying the Optimization Variables After Execution
1.8.4 Local Versus Global Minimum
1.8.5 Minimum Weight Versus Minimum Volume
1.8.6 Mesh Density
1.8.7 Using Substructures
1.9 Sample Optimization Analysis
1.9.1 Problem Description
1.9.2 Problem Specifications
1.9.3 Using a Batch File for the Analysis
1.9.4 Using the GUI for the Analysis
1.9.5 Where to Find Other Examples
2 Topological Optimization
2.1 What is Topological Optimization?
2.2 How to Do Topological Optimization

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 2 of 207

2.2.1 Define the Structural Problem

2.2.2 Select the Element Types
2.2.3 Specify Optimized and Non-Optimized Regions
2.2.4 Define and Control Your Load Cases or Frequency Extraction
2.2.5 Define and Control the Optimization Process
2.2.6 Review the Results
2.3 A 2-D Multiple-Load Case Optimization Example
2.3.1 Problem Description - First Scenario
2.3.2 Problem Results - First Scenario
2.3.3 Problem Description - Second Scenario
2.3.4 Problem Results - Second Scenario
2.4 A 2-D Natural Frequency Maximization Example
2.4.1 Problem Description
2.4.2 Problem Results
2.5 Hints and Comments
3 Probabilistic Design
3.1 What Is Probabilistic Design?
3.1.1 Traditional (Deterministic) vs. Probabilistic Design Analysis Methods
3.1.2 When Should Probabilistic Design be Used?
3.2 Basic Definitions
3.3 How to Do Probabilistic Design
3.3.1 Create the Analysis File
3.3.2 Establish Parameters for Probabilistic Design Analysis
3.3.3 Enter the PDS and Specify the Analysis File
3.3.4 Declare Random Input Variables
3.3.5 Visualize Random Input Variables
3.3.6 Specify Correlations Between Random Variables
3.3.7 Specify Random Output Parameters
3.3.8 Choose a Probabilistic Design Method
3.3.9 Executing Probabilistic Analysis Loops
3.3.10 Fitting and Using Response Surfaces
3.3.11 Reviewing Results Data
3.4 Guidelines for Choosing Probabilistic Design Variables
3.4.1 Choosing and Defining Random Input Variables
3.4.2 Choosing Random Output Parameters
3.5 Probabilistic Design Techniques
3.5.1 Monte Carlo Simulations
3.5.2 Response Surface Analysis Methods
3.6 Post-Processing Probabilistic Analysis Results
3.6.1 Statistical Post-Processing
3.6.2 Trend Post-Processing
3.6.3 Generating an HTML Report
3.7 Multiple Probabilistic Design Executions
3.7.1 Saving the Probabilistic Design Database
3.7.2 Restarting a Probabilistic Design Analysis
3.7.3 Clearing the Probabilistic Design Database
3.8 Sample Probabilistic Design Analysis
3.8.1 Problem Description
3.8.2 Problem Specifications
3.8.3 Using a Batch File for the Analysis
3.8.4 Using the GUI for the PDS Analysis
4 Adaptive Meshing
4.1 What Is Adaptive Meshing?
4.2 Prerequisites for Adaptive Meshing
4.3 How to Use Adaptive Meshing: Basic Procedure

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 3 of 207

4.4 Modifying the Basic Procedure

4.4.1 Selective Adaptivity
4.4.2 Customizing the ADAPT Macro with User Subroutines
4.4.3 Customizing the ADAPT Macro (UADAPT.MAC)
4.5 Guidelines for Adaptive Meshing
4.6 Where to Find Examples
5 Submodeling
5.1 What Is Submodeling?
5.2 How to Do Submodeling
5.2.1 Create and Analyze the Coarse Model
5.2.2 Create the Submodel
5.2.3 Perform Cut Boundary Interpolation
5.2.4 Analyze the Submodel
5.2.5 Verify the Distance Between the Cut Boundaries and the Stress Concentration
5.3 Sample Analysis Input
5.4 Shell-to-Solid Submodels
5.5 Where to Find Examples
6 Substructuring
6.1 What Is Substructuring?
6.2 How to Do Substructuring
6.2.1 Generation Pass: Creating the Superelement
6.2.2 Use Pass: Using the Superelement
6.2.3 Expansion Pass: Expanding Results within the Superelement
6.3 Sample Analysis Input
6.4 Top-down Substructuring
6.5 Nested Superelements
6.6 Prestressed Substructures
6.6.1 Static Analysis Prestress
6.6.2 Substructuring Analysis Prestress
6.7 Where to Find Examples
7 Element Birth and Death
7.1 What Is Element Birth and Death?
7.1.1 How Does the Birth and Death Feature Work?
7.2 How to Use the Birth and Death Feature
7.2.1 Build the Model
7.2.2 Apply Loads and Obtain the Solution
7.2.3 Review the Results
7.2.4 Using ANSYS Results to Control Birth and Death
7.3 Further Guidelines
7.4 Where to Find Examples
8 User Routines and Non-Standard Uses
8.1 User Programmable Features (UPFs)
8.1.1 What Are UPFs?
8.1.2 How to Use UPFs
8.1.3 A Brief Description of the Types of UPFs Currently Available
8.2 Non-standard Uses of the ANSYS Program
8.2.1 What Are Non-standard Uses?
8.2.2 Some Guidelines
9 Improving ANSYS Performance and Parallel Performance for ANSYS
9.1 Activating Parallel Processing
9.1.1 System Notes
9.2 Overview of Parallel Performance for ANSYS
9.2.1 What Is Parallel Performance?
9.2.2 About the Algebraic Multigrid (AMG) Solver
9.2.3 About the Distributed Domain Solver (DDS)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 4 of 207

9.3 Running Parallel Performance Solvers

9.3.1 Using the Algebraic Multigrid (AMG) Solver
9.3.2 Using the Distributed Domain Solver (DDS)
9.3.3 Files Written During Distributed Processing
9.4 Troubleshooting Parallel Performance
9.4.1 General Problems
9.4.2 Resolving Errors
9.4.3 Platform-Specific Problems

Prev Next
Bibliography Design Optimization
Design Optimization
Prev Advanced Analysis Techniques Guide Next

Chapter 1. Design Optimization

1.1. What Is Design Optimization ?
Design optimization is a technique that seeks to determine an optimum design. By "optimum
design," we mean one that meets all specified requirements but with a minimum expense of certain
factors such as weight, surface area, volume, stress, cost, etc. In other words, the optimum design is
usually one that is as effective as possible.

Virtually any aspect of your design can be optimized: dimensions (such as thickness), shape (such as
fillet radii), placement of supports, cost of fabrication, natural frequency, material property, and so
on. Actually, any ANSYS item that can be expressed in terms of parameters can be subjected to
design optimization. (See Use ANSYS Parameters in the ANSYS Modeling and Meshing Guide for a
description of ANSYS parameters.)

The ANSYS program offers two optimization methods to accommodate a wide range of optimization
problems. The subproblem approximation method is an advanced zero-order method that can be
efficiently applied to most engineering problems. The first order method is based on design
sensitivities and is more suitable for problems that require high accuracy.

For both the subproblem approximation and first order methods, the program performs a series of
analysis-evaluation-modification cycles. That is, an analysis of the initial design is performed, the
results are evaluated against specified design criteria, and the design is modified as necessary. This
process is repeated until all specified criteria are met.

In addition to the two optimization techniques available, the ANSYS program offers a set of strategic
tools that can be used to enhance the efficiency of the design process. For example, a number of
random design iterations can be performed. The initial data points from the random design
calculations can serve as starting points to feed the optimization methods mentioned above.

Prev Next
Advanced Analysis Techniques Guide Basic Definitions
Basic Definitions
Prev Design Optimization Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 5 of 207

1.2. Basic Definitions

Before describing the procedure for design optimization, we will define some of the terminology:
design variable, state variable, objective function, feasible and infeasible designs, the analysis file,
iterations, loops, design sets, etc. We will start with a typical optimization problem statement:

Find the minimum-weight design of a beam of rectangular cross section subject to the following
constraints:

? Total stress should not exceed max [ < max]

? Beam deflection should not exceed max [ < max]

? Beam height h should not exceed h max [h < hmax]

Example of a Beam for Design Optimization

Design Variables (DVs) are independent quantities that are varied in order to achieve the optimum
design. Upper and lower limits are specified to serve as "constraints" on the design variables. These
limits define the range of variation for the DV. In the above beam example, width b and height h are
obvious candidates for DVs. Both b and h cannot be zero or negative, so their lower limit would be
b,h > 0.0. Also, h has an upper limit of hmax. Up to 60 DVs may be defined in an ANSYS design
optimization problem.

State Variables (SVs) are quantities that constrain the design. They are also known as "dependent
variables," and are typically response quantities that are functions of the design variables. A state
variable may have a maximum and minimum limit, or it may be "single sided," having only one
limit . Our beam example has two SVs: (the total stress) and (the beam deflection). You may
define up to 100 SVs in an ANSYS design optimization problem.

The Objective Function is the dependent variable that you are attempting to minimize. It should be a
function of the DVs, that is, changing the values of the DVs should change the value of the objective
function. In our beam example, the total weight of the beam could be the objective function (to be
minimized). You may define only one objective function in an ANSYS design optimization problem.

The design variables, state variables, and objective function are collectively referred to as the
optimization variables. In an ANSYS optimization, these variables are represented by user-named
variables called parameters. You must identify which parameters in the model are DVs, which are
SVs, and which is the objective function.

A design set (or design) is simply a unique set of parameter values that represents a particular model
configuration. Typically, a design set is characterized by the optimization variable values; however,
all model parameters (including those not identified as optimization variables) are included in the set.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 6 of 207

A feasible design is one that satisfies all specified constraints - constraints on the SVs as well as
constraints on the DVs. If any one of the constraints is not satisfied, the design is considered
infeasible. The best design is the one which satisfies all constraints and produces the minimum
objective function value. (If all design sets are infeasible, the best design set is the one closest to
being feasible, irrespective of its objective function value.)

The analysis file is an ANSYS input file (which can be created in a variety of ways) that contains a
complete analysis sequence (preprocessing, solution, postprocessing). It must contain a
parametrically defined model, using parameters to represent all inputs and outputs which will be
used as DVs, SVs, and the objective function. From this file, an optimization loop file
(Jobname.LOOP) is automatically created and used by the optimizer to perform analysis loops.

A loop is one pass through the analysis cycle. (Think of this as one pass through the analysis file.)
Output for the last loop performed is saved on file Jobname.OPO . An optimization iteration (or
simply iteration) is one or more analysis loops which result in a new design set. Typically, an
iteration equates to one loop. However, for the first order method, one iteration represents more than
one loop.

The optimization database contains the current optimization environment, which includes
optimization variable definitions, parameters, all optimization specifications, and accumulated design
sets. This database can be saved (to Jobname.OPT) or resumed at any time in the optimizer.

Some of the above concepts can be better understood through an illustration. Optimization Data
Flow shows the flow of information during an optimization analysis. The analysis file must exist as a
separate entity, and note that the optimization database is not part of the ANSYS model database.

Optimization Data Flow

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 7 of 207

Prev Next
About GUI Paths and Command
Design Optimization Syntax
About GUI Paths and Command Syntax
Prev Design Optimization Next

1.3. About GUI Paths and Command Syntax

Throughout this document, you will see references to ANSYS commands and their equivalent GUI
paths. Such references use only the command name, because you do not always need to specify all of
a command's arguments, and specific combinations of command arguments perform different
functions. For complete syntax descriptions of ANSYS commands, consult the ANSYS Commands
Reference.

The GUI paths shown are as complete as possible. In many cases, choosing the GUI path as shown
will perform the function you want. In other cases, choosing the GUI path given in this document
takes you to a menu or dialog box; from there, you must choose additional options that are
appropriate for the specific task being performed.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 8 of 207

For all types of analyses described in this guide, specify the material you will be simulating using an
intuitive material model interface. This interface uses a hierarchical tree structure of material
categories, which is intended to assist you in choosing the appropriate model for your analysis. See
Material Model Interface in the ANSYS Basic Analysis Guide for details on the material model
interface.

Prev Next
Basic Definitions How to Do Design Optimization
How to Do Design Optimization
Prev Design Optimization Next

1.4. How to Do Design Optimization

You can approach an ANSYS optimization in two ways: as a batch run or interactively through the
Graphical User Interface (GUI). The approach you take will depend on your ANSYS expertise and
your preference for interacting with the ANSYS program.

If you are very familiar with ANSYS commands, you may choose to perform the entire optimization
by creating an ANSYS command input file and submitting it as a batch job. This may be a more
efficient method for complex analyses (for example, nonlinear) that require extensive run-time.

On the other hand, the interactive features of optimization offer greater flexibility and immediate
feedback for review of loop results. When performing optimization through the GUI, it is important
to first establish the analysis file for your model. Then all operations within the optimizer can be
performed interactively, allowing the freedom to probe your design space before the actual
optimization is done. The insights you gain from your initial investigations can help to narrow your
design space and achieve greater efficiency during the optimization process. (The interactive features
can also be used to process batch optimization results.)

The usual procedure for design optimization consists of the following main steps. These steps may
vary slightly, depending on whether you are performing optimization interactively (through the
GUI), in batch mode, or across multiple machines.

1. Create an analysis file to be used during looping. This file should represent a complete analysis
sequence and must do the following:
? Build the model parametrically (PREP7).

? Obtain the solution(s) (SOLUTION).

? Retrieve and assign to parameters the response quantities that will be used as state
variables and objective functions (POST1/POST26).

2. Establish parameters in the ANSYS database which correspond to those used in the analysis
file; this step is typical, but not required (Begin or OPT).

3. Enter OPT and specify the analysis file (OPT).

4. Declare optimization variables (OPT).

5. Choose optimization tool or method (OPT).

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 9 of 207

6. Specify optimization looping controls (OPT).

7. Initiate optimization analysis (OPT).

8. Review the resulting design sets data (OPT) and postprocess results (POST1/POST26).

Details of the optimization procedure are presented below. Differences in the procedure for a "batch"
versus "interactive" approach are indicated, where appropriate.

1.4.1. Create the Analysis File

The analysis file is a key component which is crucial to ANSYS optimization. The program uses the
analysis file to form the loop file, which is used to perform analysis loops. Any type of ANSYS
analysis (structural, thermal, magnetic, etc.; linear or nonlinear) may be incorporated in the analysis
file.

Note

An explicit dynamics analysis using ANSYS/LS -DYNA cannot be optimized.

In this file, the model must be defined in terms of parameters (which are usually the DVs), and
results data must be retrieved in terms of parameters (for SVs and the objective function). Only
numerical scalar parameters are used by the design optimizer. See Use ANSYS Parameters in the
ANSYS Modeling and Meshing Guide for a discussion of parameters. See the ANSYS APDL
Programmer's Guide for a discussion of the ANSYS Parametric Design Language (APDL).

It is your responsibility to create the analysis file and make sure that it is correct and complete. It
must represent a clean analysis that will run from start to finish. Most nonessential commands (such
as those that perform graphic displays, listings, status requests, etc.) should be stripped off or
commented out of this file. Maintain only those display commands which you would like to see
during an interactive session [EPLOT, etc.], or direct desired displays to a graphics file [/SHOW].
Keep in mind that the analysis file will be used over and over again during optimization looping.
Any commands that do not have direct bearing on the analysis will produce wasted action and
therefore decrease looping efficiency.

There are two ways to create an analysis file:

? Input commands line by line with a system editor.

? Create the analysis interactively through ANSYS and use the ANSYS command log as the
basis for the analysis file.

Both methods have advantages and disadvantages. Creating the file with a system editor is the same
as creating a batch input file for the analysis. (If you are performing the entire optimization in batch
mode, the analysis file will usually be the first portion of the complete batch input stream.) This
method allows you full control of parametric definitions through exact command inputs. It also
eliminates the need to clean out unnecessary commands later. However, if you are not moderately
familiar with ANSYS commands, this method may be inconvenient.

You may find it easier to perform the initial analysis interactively, and then use the resulting
command log as the basis for the analysis file. In this case, final editing of the log file may be
required in order to make it suitable for optimization looping. (See Preparing the Analysis File.)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 10 of 207

No matter how you intend to create the analysis file, the basic information that it must contain is the
same. The steps it must include are explained next.

1.4.1.1. Build the Model Parametrically

PREP7 is used to build the model in terms of the DV parameters. For our beam example, the DV
parameters are B (width) and H (height), so the element real constants are expressed in terms of B and
H:

...
/PREP7
! Initialize DV parameters:
B=2.0 ! Initialize width
H=3.0 ! Initialize height
!
ET,1,BEAM3 ! 2-D beam element
AREA=B*H ! Beam cross-sectional area
IZZ=(B*(H**3))/12 ! Moment of inertia about Z
R,1,AREA,IZZ,H ! Real constants in terms of DV parameters
!
! Rest of the model:
MP,EX,1,30E6 ! Young's modulus
N,1 ! Nodes
N,11,120
FILL
E,1,2 ! Elements
EGEN,10,1,-1
FINISH ! Leave PREP7
...

As mentioned earlier, you can vary virtually any aspect of the design: dimensions, shape, material
property, support placement, applied loads, etc. The only requirement is that the design must be
defined in terms of parameters.

The DV parameters ( B and H in this example) may be initialized anywhere, but are typically defined
in PREP7. The initial values assigned to these parameters represent a starting design, which is later
modified by the optimizer.

Caution

If you build your model interactively (through the GUI), you will encounter many
situations where data can be input through graphical picking (such as when defining
geometric entities). However, some picking operations do not allow parametric input.
Therefore, you should avoid these picking operations when defining items that will be
used as DVs, SVs, or an objective function. Instead, use menu options that allow
direct input of parameters.

1.4.1.2. Obtain the Solution

The SOLUTION processor is used to define the analysis type and analysis options, apply loads,
specify load step options, and initiate the finite element solution. All data required for the analysis
should be specified: master degrees of freedom in a reduced analysis, appropriate convergence
criteria for nonlinear analyses, frequency range for harmonic response analysis, and so on. Loads and
boundary conditions may also be DVs.

The SOLUTION input for our beam example could look like this:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 11 of 207

...
/SOLU
ANTYPE,STATIC ! Static analysis (default)
D,1,UX,0,,11,10,UY ! UX=UY=0 at the two ends of the beam
SFBEAM,ALL,1,PRES,100 ! Transverse pressure (load per unit
! length) = 100
SOLVE
FINISH ! Leave SOLUTION

This step is not limited to just one analysis. You can, for instance, obtain a thermal solution and then
obtain a stress solution (for thermal stress calculations).

If your solution uses the multiframe restart feature, all changes to the parameter set that are made
after the first load step will be lost in a multiframe restart. To ensure that the correct parameters are
used in a multiframe restart, you must explicitly save (PARSAV) and resume (PARESU) the
parameters for use in the restart. See ANSYS Basic Analysis Guide for more information on
multiframe restarts.

1.4.1.3. Retrieve Results Parametrically

This is where you retrieve results data and assign them to parameters. These parameters usually
represent SVs and the objective function. The *GET command (Utility Menu>Parameters>Get
Scalar Data), which assigns ANSYS calculated values to parameters, is used to retrieve the data.
POST1 is typically used for this step, especially if the data are to be stored, summed, or otherwise
manipulated.

In our beam example, the weight of the beam is the objective function (to be minimized). Because
weight is directly proportional to volume, and assuming uniform density, minimizing the total
volume of the beam is the same as minimizing its weight. Therefore, we can use volume as the
objective function. The SVs for this example are the total stress and deflection. The parameters for
these data may be defined as follows:

...
/POST1
SET,...
NSORT,U,Y ! Sorts nodes based on UY deflection
*GET,DMAX,SORT,,MAX ! Parameter DMAX = maximum deflection
!
! Derived data for line elements are accessed through ETABLE:
ETABLE,VOLU,VOLU ! VOLU = volume of each element
ETABLE,SMAX_I,NMISC,1 ! SMAX_I = max. stress at end I of each
! element
ETABLE,SMAX_J,NMISC,3 ! SMAX_J = max. stress at end J of each
! element
!
SSUM ! Sums the data in each column of the element
! table*GET,VOLUME,SSUM,,ITEM,VOLU! Parameter
VOLUME = total volume
ESORT,ETAB,SMAX_I,,1 ! Sorts elements based on absolute value
! of SMAX_I
*GET,SMAXI,SORT,,MAX ! Parameter SMAXI = max. value of SMAX_I
ESORT,ETAB,SMAX_J,,1 ! Sorts elements based on absolute value
! of SMAX_J
*GET,SMAXJ,SORT,,MAX ! Parameter SMAXJ = max. value of SMAX_J
SMAX=SMAXI>SMAXJ ! Parameter SMAX = greater of SMAXI and
! SMAXJ, that is, SMAX is the max. stress
FINISH
...

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 12 of 207

Please see the *GET and ETABLE command descriptions for more information.

1.4.1.4. Preparing the Analysis File

So far, we have described what needs to go into the analysis file. If you create this file as you would
a batch input file (entering commands with a system editor), then you simply save the file and begin
the optimization procedure (see Establish Parameters for Optimization). However, if you choose to
create your model interactively in ANSYS, you must derive the analysis file from the interactive
session. This can be done one of two ways, using the database command log or the session log file.

Database Command Log-You can create a command log file that represents the model database by
using the LGWRITE command (Utility Menu>File>Write DB Log File). LGWRITE writes the
internal database command log to a file ( Jobname.LGW). The internal log contains all commands that
were used to create the current database.

Session Log File - Jobname.LOG contains all commands that are issued during an interactive session.
To use Jobname.LOG as the optimization analysis file, you should edit out all nonessential commands
with a system editor. Because all commands issued are saved to this file, extensive editing may be
needed. Also, if your analysis was performed in several ANSYS sessions, you should piece together
the log file segments for each session to create a complete analysis file. (See Using the ANSYS
Command Log in the ANSYS Operations Guide for more information on the session log file and
database command log.)

Note

With either method, you may have to exit ANSYS or use the /SYS command in order
to edit the analysis file. For more details on creating this file, see Generating the
Analysis File.

1.4.2. Establish Parameters for Optimization

At this point, having completed the analysis file, you are ready to begin optimization. (You may have
to reenter ANSYS if you edited the analysis file at the system level.) When performing optimization
interactively, it is advantageous (but optional) to first establish the parameters from the analysis file
in the ANSYS database. (This step is not necessary for optimization performed in batch mode.)

There are two possible reasons for performing this step. The initial parameter values may be required
as a starting point for a first order optimization. Also, for any type of optimization run, having the
parameters in the database makes them easy to access through the GUI when defining optimization
variables. To establish the parameters in the database do one of the following:

? Resume the database file ( Jobname.DB) associated with the analysis file. This establishes your
entire model database in ANSYS, including the parameters. To resume the database file, use
one of these methods:
Command(s):

RESUME

GUI:
Utility Menu>File>Resume Jobname.db
Utility Menu>File>Resume from

? Read the analysis file into ANSYS to perform the complete analysis. This establishes your

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 13 of 207

entire model database in ANSYS, but may be time-consuming for a large model. To read the
analysis file, use one of these methods:
Command(s):

/INPUT

GUI:

Utility Menu>File>Read Input from

? Resume only the parameters into ANSYS from a previously saved parameter file; that is, a
parameter file that you saved using either the PARSAV command or the Utility
Menu>Parameters>Save Parameters menu path. To resume the parameters, use either of these
methods:
Command(s):

PARRES

GUI:

Utility Menu>Parameters>Restore Parameters

? Recreate the parameter definitions as they exist in the analysis file. Doing this requires that
you know which parameters were defined in the analysis file. Use one of these methods:
Command(s):

*SET or =

GUI:

Utility Menu>Parameters>Scalar Parameters

You may choose to do none of the above, and instead depend on the OPVAR command (Main
Menu>Design Opt>Design Variables) to define the parameters which you declare as optimization
variables (see Declare Optimization Variables).

Note

The ANSYS database does not need to contain model information corresponding to
the analysis file to perform optimization. The model input will be read from the
analysis file automatically during optimization looping.

1.4.3. Enter OPT and Specify the Analysis File

The remaining steps are performed within the OPT processor. When you first enter the optimizer,
any parameters that exist in the ANSYS database are automatically established as design set number
1. It is assumed that these parameter values represent a potential design solution. To enter the
optimizer, use one of these methods:

Command(s):

/OPT

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 14 of 207

GUI:

Main Menu>Design Opt

In interactive mode, you must specify the analysis file name. This file is used to derive the
optimization loop file Jobname.LOOP. There is no default for the analysis file name, therefore it must
be input. To specify the analysis file name, use one of these methods:

Command(s):

OPANL

GUI:

Main Menu>Design Opt>Assign

For an optimization run in batch mode, the analysis file is usually the first portion of the batch input
stream, from the first line down to the first occurrence of /OPT. In batch mode the analysis file name
defaults to Jobname.BAT (a temporary file containing input copied from the batch input file).
Therefore, you normally do not need to specify an analysis file name in batch mode. However, if for
some reason you have separated the batch optimization input into two files (one representing the
analysis and the other containing all optimization operations), then you will need to specify the
analysis file [OPANL] after entering the optimizer [/OPT].

Note

In the analysis file, the /PREP7 and /OPT commands must occur as the first
nonblank characters on a line (that is, do not use the $ delimiter on a line containing
either of these commands). This is required for proper loop file construction.

1.4.4. Declare Optimization Variables

The next step is to declare optimization variables, that is, specify which parameters are DVs, which
ones are SVs, and which one is the objective function. As mentioned earlier, up to 60 DVs and up to
100 SVs are allowed, but only one objective function is allowed. To declare optimization variables,
use one of these methods:

Command(s):

OPVAR

GUI:
Main Menu>Design Opt>Design Variables
Main Menu>Design Opt>State Variables
Main Menu>Design Opt>Objective

Minimum and maximum constraints can be specified for SVs and DVs. No constraints are needed
for the objective function. Each variable has a tolerance value associated with it, which you may
input or let default to a program calculated value.

If the parameter name which you specify on the OPVAR command is not an existing parameter, the
parameter is automatically defined in the ANSYS database with an initial value of zero.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 15 of 207

You may change a previously declared optimization variable at any time by simply redefining it.
You may also delete an optimization variable [OPVAR,Name,DEL]. The delete option does not
delete the parameter; it simply deactivates the parameter as an optimization variable (see Modifying
the Optimization Variables After Execution).

1.4.5. Choose Optimization Tool or Method

In the ANSYS program, several different optimization tools and methods are available. Single loop
is the default. To specify a tool or method to be used for subsequent optimization looping, use one of
these methods:

Command(s):

OPTYPE

GUI:

Main Menu>Design Opt>Method/Tool

Optimization methods are traditional techniques that strive for minimization of a single function (the
objective function) subject to constraints. Two methods are available: the subproblem approximation
method and the first order method. In addition, you can supply an external optimization algorithm, in
which case the ANSYS algorithm will be bypassed. To use one of these methods, you must have an
objective function defined.

? Subproblem Approximation Method: This is an advanced zero-order method which uses

approximations (curve fitting) to all dependent variables (SVs and the objective function). It is
a general method that can be applied efficiently to a wide range of engineering problems.

? First Order Method: This method uses derivative information, that is, gradients of the
dependent variables with respect to the design variables. It is highly accurate and works well
for problems having dependent variables that vary widely over a large range of design space.
However, this method can be computationally intense.

? User-supplied Method: An external optimization routine (USEROP) can be used instead of the
ANSYS optimizer logic.

Optimization tools are techniques used to measure and understand the design space of your problem.
Since minimization may or may not be a goal, an objective function is not required for use of the
tools. However, design variables must be defined. The following tools are available.

? Single Loop Run: This tool performs one loop and produces one FEA solution at a time. You
can do "what if" studies by using a series of single loop runs, setting different design variable
values before each loop.

? Random Design Generation: Multiple loops are performed, with random design variable
values at each loop. A maximum number of loops and a desired number of feasible loops can
be specified. This tool is useful for studying the overall design space, and for establishing
feasible design sets for subsequent optimization analysis.

? Sweep Generation: Starting from a reference design set, this tool generates several sequences
of design sets. Specifically, it varies one design variable at a time over its full range using
uniform design variable increments. This tool makes global variational evaluations of the
objective function and of the state variables possible.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 16 of 207

? Factorial Evaluation: This is a statistical tool that is used to generate design sets at all extreme
combinations of design variable values. This technique is related to the technology known as
design of experiment that uses a 2-level, full and fractional factorial analysis. The primary aim
is to compute main and interaction effects for the objective function and the state variables.

? Gradient Evaluation: At a user-specified reference design set, this tool calculates the gradients
of the objective function and the state variables with respect to the design variables. Using this
tool, you can investigate local design sensitivities.

? User-supplied Design Tool: An external routine (USEROP) can be used to bypass the ANSYS
logic.

As noted above, you can implement your own method or tool by invoking the USEROP routine. The
ANSYS Guide to User Programmable Features contains more details on this user-supplied routine.

1.4.6. Specify Optimization Looping Controls

Each method and tool has certain looping controls associated with it, such as maximum number of
iterations, etc. All of the commands that you use to set these controls are accessed by the menu path
Main Menu>Design Opt>Method/Tool

The commands for setting controls are as follows:

? To set controls for the subproblem approximation method, use OPSUBP and OPEQN.

? To set controls for the first order method, use OPFRST.

? To set controls for the random design generation tool, use OPRAND.

? To set controls for the sweep generation tool, use OPSWEEP.

? To set controls for the factorial evaluation tool, use OPFACT.

? To set controls for the gradient evaluation tool, use OPGRAD.

? To set controls for the user optimization tool, use OPUSER.

There are also a number of general controls which affect how data is saved during optimization.
They are as follows:

? To specify the file where optimization data is to be saved (defaults to Jobname.OPT):

Command(s):

OPDATA

GUI:

Main Menu>Design Opt>Controls

? To activate a detailed summary printout:

Command(s):

OPPRNT

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 17 of 207

GUI:

Main Menu>Design Opt>Controls

? To determine whether information from the best design set is saved (by default, the database
and results files are saved only for the last design set):
Command(s):

OPKEEP

GUI:

Main Menu>Design Opt>Controls

You can also control various looping characteristics, including how the analysis file is to be read for
looping. The file can be read starting from the first line (default) or from the first occurrence
of /PREP7, and parameters assigned as DVs can be ignored (default) or processed during looping. In
addition, you can specify which type of parameters are to be saved during looping: scalar parameters
only (default), or scalar and array parameters. This capability allows for control of parameter values
(DV and non-DV) during looping. To control these looping characteristics, use one of these methods:

Command(s):

OPLOOP

GUI:

Main Menu>Design Opt>Controls

Note

The Parms argument on the OPLOOP command controls which parameters are saved
during looping. The option to save both scalar and array parameters ( Parms = ALL)
should typically not be used, except for the case when array parameters defined
outside of the analysis file [*DIM] need to be preserved during looping.

1.4.7. Initiate Optimization Analysis

After all appropriate controls have been specified, you can initiate looping:

Command(s):

OPEXE

GUI:

Main Menu>Design Opt>Run

If you want to run the optimization loops across multiple machines simultaneously, issue
OPEXE,,,,MULTIRUN. You can specify a time-out value for the slave machines, as well as a
polling frequency with which ANSYS checks for returned tasks. See the OPEXE command
description for a detailed explanation of the correct use of these options.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 18 of 207

To use the multirun feature, you must first create a file called ansys.sys that lists the machines on
which you want to run. You can list up to 256 systems. If a machine has multiple processors, you can
list the machine multiple times, and each processor will be fully utilized. For example, if a machine
has four processors, list the machine four times for maximum efficiency.

The ansys.sys file should reside in your working directory. The format is as follows:

? Line1: The name for ansys execution (default ansys60).

? Line 2+: Machine names

? Columns 1-32 - remote system name

? Columns 33-64 - ANSYS execution name (default name on line 1)

? Columns 65-96 - Directory under which to create the execution directory (default is your
home directory)

? Any line except line 1 that starts with # or ! in column 1 is ignored.

Initially, the samples are distributed equally across the machines. The first machine that returns a
result is then sent the next sample. This process continues until all samples are finished.

You must have a login on each machine listed, and you must be able to execute ANSYS on each
system. You must also have enough available ANSYS tasks to run all systems simultaneously.

Upon execution of OPEXE, an optimization loop file ( Jobname.LOOP) is written from the analysis
file. This loop file, which is transparent to the user, is used by the optimizer to perform analysis
loops. Looping will continue until convergence, termination (not converged, but maximum loop limit
or maximum sequential infeasible solutions encountered), or completion (for example, requested
number of loops for random design generation) has been reached.

If a loop is interrupted due to a problem within the model (for example, a meshing failure, a non-
converged nonlinear solution, conflicting design variable values, etc.), the optimizer aborts that loop,
but can continue looping. In interactive mode, a warning will be issued, and you may choose to
continue or terminate looping. In batch mode, looping will automatically continue. (The NCNV
command (menu path Main Menu>Solution>Sol'n Control:Advanced NL Tab, Main
Menu>Solution>Unabridged Menu>Nonlinear>Criteria to Stop, or Main
Menu>Solution>Nonlinear>Criteria to Stop), which specifies program termination controls for
nonlinear analyses, is ignored during optimization looping.) The design set for the aborted loop will
be saved, but the response quantity parameters for that set will have very large, irrelevant values.

The values of all optimization variables and other parameters at the end of each iteration are stored
on the optimization data file ( Jobname.OPT). Up to 130 such sets are stored. When the 130th set is
encountered, the data associated with the "worst" design are discarded.

Continuing with our beam example, the optimization portion of the input would look like this:

/OPT ! Enter OPT

OPANL,... ! Analysis file name (not needed for batch)
!
! Declare optimization variables:
OPVAR,B,DV,.5,16.5 ! Parameters B and H are DVs
OPVAR,H,DV,.5,8
OPVAR,DMAX,SV,-0.1,0 ! Parameters DMAX and SMAX are SVs

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 19 of 207

OPVAR,SMAX,SV,0,20000
OPVAR,VOLUME,OBJ ! Parameter VOLUME is the obj. function
!
! Specify optimization type and controls
OPTYPE,SUBP ! Subproblem approximation method
OPSUBP,30 ! Maximum number of iterations
OPEXE ! Initiate optimization looping

Several optimization executions may occur in series. For example, we could perform a sweep
generation after the subproblem approximation execution is completed. The following series of
commands executes a sweep with respect to the best design set:

OPTYPE,SWEEP ! Sweep evaluation tool

OPSWEEP,BEST,5 ! 5 evaluations per DV at best design set
OPEXE ! Initiate optimization looping

See the /OPT, OPANL, OPVAR, OPTYPE, OPSUBP, OPSWEEP, and OPEXE command
descriptions for more information.

1.4.8. Review Design Sets Data

After optimization looping is complete, you can review the resulting design sets in a variety of ways
using the commands described in this section. These commands can be applied to the results from
any optimization method or tool.

To list the values of parameters for specified set numbers:

Command(s):

OPLIST

GUI:

Main Menu>Design Opt>List

You can choose to list all scalar parameters, or only those used as optimization variables.

To graph specified parameters versus set number so you can track how a variable changed from
iteration to iteration:

Command(s):

PLVAROPT

GUI:

Main Menu>Design Opt>Graphs/Tables

To change the abscissa of the graph from set number to any other parameter:

Command(s):

XVAROPT

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 20 of 207

Main Menu>Design Opt>Graphs/Tables

To print the values of specified parameters in tabular form (versus the XVAROPT parameter, which
defaults to set number):

Command(s):

PRVAROPT

GUI:

Main Menu>Design Opt>Graphs/Tables

For the PLVAROPT and PRVAROPT operations, the design sets are automatically sorted in a
sequence corresponding to an ascending order of the XVAROPT parameter.

There are several specialized ways to review results from the sweep, factorial, and gradient tools. For
sweep tools, use the OPRSW command to list results and the OPLSW command to plot results. For
factorial tools, use the OPRFA command to list results and the OPLFA command to plot results.
For gradient tools, use the OPRGR command to list results and the OPLGR command to plot
results. (Menu paths appear in detailed discussions of these commands later in this chapter.)

Another way to access optimization data is with the STAT command (Main Menu>Design
Opt>Status). When issued within the optimizer, this command lists other current optimization
information such as the analysis file name, specified optimization technique, number of existing
design sets, optimization variables, etc. Using the STAT command is a good way to check the
current optimization environment (at any point in the optimizer) and to verify that all desired settings
have been input.

In addition to reviewing the optimization data, you may wish to postprocess the analysis results using
POST1 or POST26. By default, results are saved for the last design set in file Jobname.RST (or .RTH,
etc., depending on the type of analysis). The results and the database for the best design set will also
be available if OPKEEP,ON was issued before looping. The "best results" will be in file
Jobname.BRST (.BRTH, etc.), and the "best database" will be in Jobname.BDB.

1.4.8.1. Manipulating Designs Sets

After reviewing the design sets, it may be desirable to manipulate them in some way. For example,
after performing a random design execution, you may wish to discard all non-feasible designs,
keeping the feasible sets as data points for subsequent optimization. There are several ways in which
you can change the design sets.

Two commands are available for discarding unwanted sets.

? To select a number of best design sets, or all feasible sets:

Command(s):

OPSEL

GUI:

Main Menu>Design Opt>Select/Delete

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 21 of 207

(All design sets not selected with OPSEL are permanently removed from the optimization database.)

? To delete the design sets in a specified range, use one of these methods:
Command(s):

OPDEL

GUI:

Main Menu>Design Opt>Select/Delete

For both of these commands, the original set numbers will be retained for the remaining design sets.
(Up to 130 design sets can be stored in the optimization database.)

There are other commands that can affect design sets.

? To form a new design set by adding two existing sets (with scaling factors if desired):
Command(s):

OPADD

GUI:

Main Menu>Design Opt>Combine

? To create a new design set using the active scalar parameter values (without running an
analysis loop):
Command(s):

OPMAKE

GUI:

Main Menu>Design Opt>Create

Prev Next
About GUI Paths and Command
Multiple Optimization Executions
Syntax
Multiple Optimization Executions
Prev Design Optimization Next

1.5. Multiple Optimization Executions

There are various reasons why you might wish to do more than one optimization execution. For
example, your initial optimization run may not find the desired optimum. Or, you may start by using
a design tool and then perform a subsequent optimization (for example, random design generation
followed by a subproblem approximation run). The knowledge you gain from the first few runs may
prompt you to change your design space and optimize yet again.

If you perform all executions within the same ANSYS session (or within the same batch input

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 22 of 207

stream), the procedure is very straightforward. After an execution, simply redefine all optimization
input as desired and initiate the next execution. To initiate the next execution:

Command(s):

OPEXE

GUI:

Main Menu>Design Opt>Run

If you have left the ANSYS program after performing optimization, and would like to continue that
optimization analysis at some later time, you can do a restart as described next.

1.5.1. Restarting an Optimization Analysis

To restart an optimization analysis, simply resume the optimization database file ( Jobname.OPT):

Command(s):

OPRESU

GUI:

Main Menu>Design Opt>Resume

Once the data is read in, you can respecify optimization type, controls, etc., and initiate looping. (The
analysis file corresponding to the resumed database must be available in order to perform
optimization.) To initiate looping:

Command(s):

OPEXE

GUI:

Main Menu>Design Opt>Run

A typical restart might look like this:

....
/OPT
OPRESU,.... ! Read named file (defaults to Jobname.OPT)
OPSEL,10 ! Select 10 best designs
OPTYPE,.... ! Specify optimization tool or method
.... ! Specify other optimization input
....
OPEXE ! Initiate optimization looping

See the /OPT, OPRESU, OPSEL, OPTYPE, and OPEXE command descriptions for more details.

Note

In addition to optimization data, the ANSYS jobname is saved to the optimization

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 23 of 207

database file ( Jobname.OPT). Therefore, when an optimization data file is resumed

[OPRESU], the jobname saved in that file overwrites the current jobname
[/FILNAME].

You can use the OPRESU command (Main Menu>Design Opt>Resume) in an interactive session to
resume optimization data that was created through a batch run, thus allowing convenient interactive
viewing of batch optimization results.

If there is data in the optimization database at the time you want to resume, you should first clear the
optimization database. When you do this, all settings are reset to their default values, and all design
sets are deleted. To clear the optimization database:

Command(s):

OPCLR

GUI:

Main Menu>Design Opt>Clear & Reset

Because the ANSYS database is not affected by the OPCLR command, it may also be necessary to
clear the ANSYS database if the resumed optimization problem is totally independent of the previous
one. To clear the ANSYS database:

Command(s):

/CLEAR

GUI:

Utility Menu>File>Clear & Start New

A counterpart to the OPRESU command is the OPSAVE command (Main Menu>Design

Opt>Save), which writes optimization data to a named file (defaults to Jobname.OPT). Although
optimization data is automatically saved at the end of each optimization loop (see the OPDATA
command), you can save the optimization data at any time by using the OPSAVE command.

Prev Next
How to Do Design Optimization Optimization Techniques
Optimization Techniques
Prev Design Optimization Next

1.6. Optimization Techniques

Understanding the algorithm used by a computer program is always helpful, and is particularly true
in the case of design optimization. In this section, details on the following techniques are presented:
subproblem approximation, first order, random design generation, sweep generation, factorial
evaluation, and gradient evaluation. See the ANSYS, Inc. Theory Reference for more information.

1.6.1. Subproblem Approximation Method

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 24 of 207

The subproblem approximation method can be described as an advanced zero-order method in that it
requires only the values of the dependent variables, and not their derivatives. There are two concepts
that play a key role in the subproblem approximation method: the use of approximations for the
objective function and state variables, and the conversion of the constrained optimization problem to
an unconstrained problem.

1.6.1.1. Approximations

For this method, the program establishes the relationship between the objective function and the DVs
by curve fitting. This is done by calculating the objective function for several sets of DV values (that
is, for several designs) and performing a least squares fit between the data points. The resulting curve
(or surface) is called an approximation. Each optimization loop generates a new data point, and the
objective function approximation is updated. It is this approximation that is minimized instead of the
actual objective function.

State variables are handled in the same manner. An approximation is generated for each state
variable and updated at the end of each loop.

You can control curve fitting for the optimization approximations. You can request a linear fit,
quadratic fit, or quadratic plus cross terms fit. By default, a quadratic plus cross terms fit is used for
the objective function, and a quadratic fit is used for the SVs. To control curve fitting:

Command(s):

OPEQN

GUI:

Main Menu>Design Opt>Method/Tool

OPEQN also gives you control over how the available design data points are weighted in forming
the approximations. See the ANSYS, Inc. Theory Reference for details.

1.6.1.2. Conversion to an Unconstrained Problem

State variables and limits on design variables are used to constrain the design and make the
optimization problem a constrained one. The ANSYS program converts this problem to an
unconstrained optimization problem because minimization techniques for the latter are more
efficient. The conversion is done by adding penalties to the objective function approximation to
account for the imposed constraints.

The search for a minimum of the unconstrained objective function approximation is then carried out
by applying a Sequential Unconstrained Minimization Technique (SUMT) at each iteration.

1.6.1.3. Convergence Checking

At the end of each loop, a check for convergence (or termination) is made. The problem is said to be
converged if the current, previous, or best design is feasible and any of the following conditions are
satisfied:

? The change in objective function from the best feasible design to the current design is less than
the objective function tolerance.

? The change in objective function between the last two designs is less than the objective

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 25 of 207

function tolerance.

? The changes in all design variables from the current design to the best feasible design are less
then their respective tolerances.

? The changes in all design variables between the last two designs are less than their respective
tolerances.

You specify the objective function and design variable tolerances using one of these methods:

Command(s):

OPVAR

GUI:
Main Menu>Design Opt>Design Variables
Main Menu>Design Opt>Objective

Convergence does not necessarily indicate that a true global minimum has been obtained. It only
means that one of the four criteria mentioned above has been satisfied. Therefore, it is your
responsibility to determine if the design has been sufficiently optimized. If not, you can perform
additional optimization analyses.

Sometimes the solution may terminate before convergence is reached. This happens if one of the
following conditions is true:

? The number of loops specified (NITR on the OPSUBP command) has been performed.

? The number of consecutive infeasible designs has reached the specified limit ( NINFS on the
OPSUBP command). The default number is 7.

1.6.1.4. Special Considerations for Subproblem Approximation

Because approximations are used for the objective function and SVs, the optimum design will be
only as good as the approximations. Guidelines to help establish "good" approximations are
presented below.

For subproblem approximation, the optimizer initially generates random designs to establish the state
variable and objective function approximations. Because these are random designs, convergence may
be slow. You can sometimes speed up convergence by providing more than one feasible starting
design. Simply run a number of random designs and discard all infeasible designs. To run a number
of random designs:

Command(s):

OPTYPE, RAND

GUI:

Main Menu>Design Opt>Method/Tool

To discard all infeasible designs, use one of these methods:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 26 of 207

Command(s):

OPSEL

GUI:

Main Menu>Design Opt>Select/Delete

Alternatively, you could create the initial design sets by using multiple single loop runs, specifying a
new set of acceptable design variables before each run:

Command(s):

OPTYPE, RUN

GUI:

Main Menu>Design Opt>Method/Tool

(This latter method works best if you have some insight into the nature of your problem.)

Note

Generating many trial designs may be good for the rate of convergence, but if the
designs are very similar to each other, that is, if the design data points are "clustered"
together, you may be forcing the optimizer to follow a specific path, thereby missing
potentially good designs.

If many infeasible designs are being generated by the subproblem approximation method, it may
mean that the state variable approximation does not adequately represent the actual state variable
function. In that case, you can do the following:

? Increase the allowable number of consecutive infeasible designs and perform an additional
subproblem approximation execution (if it appears likely that a feasible design will be
obtained):
Command(s):

OPSUBP,,NINFS

GUI:

Main Menu>Design Opt>Method/Tool

? Periodically select only the best designs between sequential subproblem approximation runs to
force better curve fit approximations:
Command(s):

OPSEL

GUI:

Main Menu>Design Opt>Select/Delete

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 27 of 207

? Choose cross terms for the state variable approximations:

Command(s):

OPEQN,,KFSV

GUI:

Main Menu>Design Opt>Method/Tool

1.6.2. First Order Method

Like the subproblem approximation method, the first order method converts the problem to an
unconstrained one by adding penalty functions to the objective function. However, unlike the
subproblem approximation method, the actual finite element representation is minimized and not an
approximation.

The first order method uses gradients of the dependent variables with respect to the design variables.
For each iteration, gradient calculations (which may employ a steepest descent or conjugate direction
method) are performed in order to determine a search direction, and a line search strategy is adopted
to minimize the unconstrained problem.

Thus, each iteration is composed of a number of subiterations that include search direction and
gradient computations. That is why one optimization iteration for the first order method performs
several analysis loops.

The OPFRST command (Main Menu>Design Opt>Method/Tool) has two input fields which may be
used to enhance convergence of the first order solution. You can specify the forward difference
applied to the design variable range used to compute the gradient ( DELTA), and also the limit on the
line search step size ( SIZE). Typically, the default values for these two inputs are sufficient. See the
ANSYS, Inc. Theory Reference for details.

1.6.2.1. Convergence Checking

First order iterations continue until either convergence is achieved or termination occurs. The
problem is said to be converged if, when comparing the current iteration design set to the previous
and best sets, one of the following conditions is satisfied:

? The change in objective function from the best design to the current design is less than the
objective function tolerance.

? The change in objective function from the previous design to the current design is less than the
objective function tolerance.

It is also a requirement that the final iteration used a steepest descent search, otherwise additional
iterations are performed.

To specify the objective function tolerance:

Command(s):

OPVAR

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 28 of 207

Main Menu>Design Opt>Objective

The problem may terminate before convergence is reached. This occurs if the number of iterations
specified by NITR on the OPFRST command has been performed.

1.6.2.2. Special Considerations for the First Order Method

Compared to the subproblem approximation method, the first order method is seen to be more
computationally demanding and more accurate. However, high accuracy does not always guarantee
the best solution. Here are some situations to watch out for:

? It is possible for the first order method to converge with an infeasible design. In this case, it
has probably found a local minimum, or there is no feasible design space. If this occurs, it may
be useful to run a subproblem approximation analysis [OPTYPE,SUBP], which is a better
measure of full design space. Also, you may try generating random designs
[OPTYPE,RAND] to locate feasible design space (if any exists), then rerun the first order
method using a feasible design set as a starting point.

? The first order method is more likely to hit a local minimum (see Local Versus Global
Minimum). This is because first order starts from one existing point in design space and works
its way to the minimum. If the starting point is too near a local minimum, it may find that point
instead of the global minimum. If you suspect that a local minimum has been found, you may
try using the subproblem approximation method or random design generation, as described
above.

? An objective function tolerance that is too tight may cause a high number of iterations to be
performed. Because this method solves the actual finite element representation (not an
approximation), it will strive to find an exact solution based on the given tolerance.

1.6.3. Random Design Generation

For random design generation [OPTYPE,RAND], the program performs a specified number of
analysis loops using random design variable values for each loop. You can use the OPRAND
command (Main Menu>Design Opt>Method/Tool) to specify a maximum number of iterations and
(if desired) a maximum number of feasible designs. If a number of feasible design sets is specified,
looping will terminate when that number is reached, even if the maximum number of iterations has
not been reached.

Random design generation is often used as a precursor to a subproblem approximation optimization

(as explained earlier). It can also be used to develop trial designs for a variety of purposes. For
example, a number of random designs can be generated, then reviewed in order to judge the validity
of the current design space.

1.6.4. Using the Sweep Tool

The sweep tool [OPTYPE,SWEEP] is used to perform a global sweep of design space. Exactly
n*NSPS design sets are generated, where n is the number of design variables and NSPS is the number
of evaluation points per sweep (specified on the OPSWEEP command). For each design variable,
the range of the variable is divided into NSPS-1 equal increments, and NSPS loops are performed. The
DV in question is incremented for each loop, while the remaining design variables are held fixed at
their reference values. The DV reference values correspond to the design set specified by Dset on
the OPSWEEP command (Main Menu>Design Opt>Method/Tool).

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 29 of 207

To graph response variable values versus design variable values, use one of these methods:

Command(s):

OPLSW

GUI:

Main Menu>Design Opt>Tool Results>Sweeps

The vertical axis shows actual values for the objective function or state variable. The horizontal axis
shows normalized values (0 to 1) for the design variables, where the normalization is with respect to
the DV maximum and minimum values [OPVAR].

To generate tabular results, use one of these methods:

Command(s):

OPRSW

GUI:

Main Menu>Design Opt>Tool Results>Print

Normalized response values are tabulated against normalized (0 to 1) design variables. The objective
function and state variable results are normalized to the values associated with the reference design
set [OPSWEEP,Dset]. For the design variables, 0 corresponds to its minimum value and 1 to its
maximum.

1.6.5. Using the Factorial Tool

The factorial tool [OPTYPE,FACT] employs a 2-level technique to generate design set results at all
extreme points (or corners) of design space. (The 2-level technique samples the 2 extreme points of
each DV.) Either a full or fractional evaluation will be performed, as specified by the OPFACT
command (Main Menu>Design Opt>Method/Tool). For a full evaluation, the program performs 2 n
loops, where n is the number of design variables. A 1/2 fractional factorial evaluation will perform
2n /2 loops; a 1/4 fractional factorial evaluation will perform 2 n/4 loops; etc.

You can display graphics in the form of bar charts and generate tables that show certain effects for
either the objective function or any state variable. For example, you may request a graph of the main
effect that each design variable has on the objective function. You can also see the effects associated
with 2- and 3-variable interactions.

To display graphics in the form of bar charts, use one of these methods:

Command(s):

OPLFA

GUI:

Main Menu>Design Opt>Tool Results>Factorial

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 30 of 207

To generate tables that show effects for the objective function or any state variable, use one of these
methods:

Command(s):

OPRFA

GUI:

Main Menu>Design Opt>Tool Results>Print

See the ANSYS, Inc. Theory Reference for more information.

1.6.6. Using the Gradient Evaluation Tool

The gradient tool [OPTYPE,GRAD] computes a gradient at a point in design space. Gradient results
are useful for studying the sensitivities of the objective function or state variables. To identify the
design set where the gradient is to be computed:

Command(s):

OPGRAD

GUI:

Main Menu>Design Opt>Method/Tool

The number of loops performed for this tool equals the number of design variables.

You can graph response variables with respect to design variable values. The vertical axis shows
actual values for the objective function or state variable graphed. The horizontal axis shows a plus or
minus 1% change in the DVs. To graph response variables:

Command(s):

OPLGR

GUI:

Main Menu>Design Opt>Tool Results>Gradient

You can also generate tabular results for the objective function and the state variables. Changes in
objective function and state variable values are tabulated against plus or minus 1% changes in the
design variables. To generate these tabular results:

Command(s):

OPRGR

GUI:

Main Menu>Design Opt>Tool Results>Print

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 31 of 207

Note

The 1% change in the DVs is with respect to the range of the DV (MAX - MIN value
from the OPVAR command) and, therefore, is not based on the current DV values.

Prev Next
Guidelines on Choosing Optimization
Multiple Optimization Executions
Variables
Guidelines on Choosing Optimization Variables
Prev Design Optimization Next

1.7. Guidelines on Choosing Optimization Variables

There are many useful guidelines you can follow in defining your DVs, SVs, and objective function.
Some of these are presented below.

1.7.1. Choosing Design Variables

DVs are usually geometric parameters such as length, thickness, diameter, or model coordinates.
They are restricted to positive values. Some points to remember about DVs are:

? Use as few DVs as possible. Having too many design variables increases the chance of
converging to a local minimum rather than the true global minimum, or even diverging if your
problem is highly nonlinear. Obviously, more DVs demand more iterations and, therefore,
more computer time. One way to reduce the number of design variables is to eliminate some
DVs by expressing them in terms of others. This is commonly referred to as design variable
linking.

DV linking may not be practical if the DVs are truly independent. However, it may be possible to
make judgements about your structure's behavior which allow a logical link between some DVs. For
example, if it is thought that an optimum shape will be symmetric, use one DV for symmetric
members.

? Specify a reasonable range of values for the design variables ( MIN and MAX on the OPVAR
command). Too wide a range may result in poor representation of design space, whereas too
narrow a range may exclude "good" designs. Remember that only positive values are allowed,
and that an upper limit must be specified.

? Choose DVs such that they permit practical optimum designs. For example, you can perform
weight optimization of a cantilever beam with just one design variable, X1, as shown in
Choosing DVs for a Tapered Cantilever Beam (a). However, this excludes a tapered or curved
design that may offer a lower weight. To allow for such a design, you may choose four design
variables X1 to X4 (Figure b), but that may result in an undesirable local minimum (Figure c).
A different scheme for such a situation would be to relate the DVs as shown in Figure d. Also,
avoid choosing DVs that may result in unrealistic or undesirable designs.

Choosing DVs for a Tapered Cantilever Beam

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 32 of 207

1.7.2. Choosing State Variables

SVs are usually response quantities that constrain the design. Examples of SVs are stresses,
temperatures, heat flow rates, frequencies, deflections, absorbed energy, elapsed time, and so on. A
state variable need not be an ANSYS-calculated quantity; virtually any parameter can be defined as a
state variable. Some points to keep in mind while choosing state variables are:

? When defining SVs [OPVAR command], a blank input in the MIN field is interpreted as "no
lower limit." Similarly, a blank in the MAX field is interpreted as "no upper limit." A zero input
in either of these fields is interpreted as a zero limit. Example:

OPVAR,SIG,SV,,1000 ! SIG<=1000
OPVAR,SIG,SV,0,1000 ! 0<=SIG<=1000

? Choose enough SVs to sufficiently constrain the design. In a stress analysis, for example,
choosing the maximum stress as the only SV may not be a good idea because the location of
the maximum stress may change from loop to loop. Also avoid the other extreme which would
be to choose the stress in every element as a state variable. The preferred method is to define
the stresses at a few key locations as state variables.

? For the subproblem approximation method, if possible, choose SVs that have a linear or
quadratic relationship with the DVs. For example, the state variable G = Z1/Z2 subject to G <
C (where Z1 and Z2 are design variables and C is a constant) may not lead to a good
approximation for G because of its inverse relationship with Z2. By restating the state variable
to be G = Z1 - (C x Z2) subject to G < 0, the state variable approximation will be exact.

? If a state variable has both an upper and lower limit, specify a reasonable range of limit values
[MIN and MAX on the OPVAR command]. Avoid very small ranges, because feasible designs
may not exist. A stress range of 500 to 1000 psi, for example, is better than 900 to 1000 psi.

? If an equality constraint is to be specified (such as frequency = 386.4 Hz), define two state
variables for the same quantity and bracket the desired value, as illustrated below:

...
*GET,FREQ,ACTIVE,,SET,FREQ ! Parameter FREQ = calculated frequency
FREQ1=FREQ
FREQ2=FREQ
...
/OPT
OPVAR,FREQ1,SV,,387 ! Upper limit on FREQ1 = 387
OPVAR,FREQ2,SV,386 ! Lower limit on FREQ2 = 386

The effective feasible region is now between 386 and 387, but each state variable has a wide enough

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 33 of 207

range for smooth approximations. (Please see the OPVAR command description for more
information.)

? Avoid choosing SVs near singularities (such as concentrated loads) by using selecting before
defining the parameters.

1.7.3. Choosing the Objective Function

The objective function is the quantity that you are trying to minimize or maximize. Some points to
remember about choosing the objective function are:

? The ANSYS program always tries to minimize the objective function. If you need to maximize
a quantity x, restate the problem and minimize the quantity x1 = C-x or x1 = 1/x, where C is a
number much larger than the expected value of x. C-x is generally a better way to define the
objective function than 1/x because the latter, being an inverse relationship, cannot be as
accurately represented by the approximations used in the subproblem approximation method.

? The objective function should remain positive throughout the optimization, because negative
values may cause numerical problems. To prevent negative values from occurring, simply add
a sufficiently large positive number to the objective function (larger than the highest expected
objective function value).

Prev Next
Optimization Techniques General Guidelines
General Guidelines
Prev Design Optimization Next

1.8. General Guidelines

This section outlines some key points you should keep in mind when doing design optimization. The
most important point to remember is that optimization involves a series of analyses, that is, several
loops of the preprocessing-solution-postprocessing-optimization cycle. We recommend, therefore,
that you start with a simple problem first and understand fully the optimization procedure. Once you
understand the various steps involved, you will find it easier to solve the real problem at hand.

1.8.1. Generating the Analysis File

As discussed in Preparing the Analysis File there are two ways to derive this file if the model was
built interactively: from the internal database command log [LGWRITE] (Utility Menu>File>Write
DB Log File), or from the session log file ( Jobname.LOG). Using the internal database log instead of
Jobname.LOG has several advantages.

The LGWRITE command has an option ( Kedit field) to exclude nonessential commands, or to
write them to the file as comment lines. This option does some automatic cleanup of the log file for
you, although you should still do a final review to ensure that the file is appropriate for optimization.
Also, the internal database log will represent the complete database so no piecing together of files is
necessary. Because the database log is saved on the database file ( Jobname.DB), a resumed database
will contain its own complete database log. (See Using the Database Command Log in the ANSYS
Operations Guide for more information on using the database command log.)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 34 of 207

Caution

It is recommended that you use Kedit = COMMENT (not Kedit = REMOVE) on the
LGWRITE command. Some commands that are filtered out by the Kedit option may
be required for subsequent *GET operations (for example, EXTREM and
PLNSOL). These commands should be uncommented during the final edit of
Jobname.LGW.

Note

The /CLEAR command clears the database from memory, and therefore also clears
the database log. A /CLEAR is issued automatically at the beginning of each
optimization loop. If LGWRITE is issued after optimization looping, the resulting
file will not contain a complete command history. Typically, the database log should
be written before performing optimization looping.

As stated earlier, you should avoid picking operations when defining items which will be used as
optimization variables. If you did use picking because it is more convenient, be aware that picking
operations cause special GUI-generated commands (such as FLST and FITEM) to be written to the
command log. These commands are documented in the ANSYS Commands Reference. However, it
may be tedious to convert them to parametric input during the final cleanup of the command log file.
You should avoid editing such commands on the log file. Any data change within the FITEM
command, for example, could render the data to be invalid, and could cause unpredictable results.

1.8.2. Fixing Design Variable Values After Execution

After performing one or more optimization executions [OPEXE], you may decide to eliminate
certain design variables [OPVAR,Name,DEL] for subsequent optimization. Typically, you would
want these parameters to maintain current values (the values assigned during the last loop, or new
values you assign to them), and not to revert back to the original values assigned in the analysis file.
Assuming that no reassignment of the parameter value occurs within the loop file, the value of a
"deleted" design variable can be fixed as follows:

1. In the analysis file, initialize the design variable parameters before the /PREP7 command.
(Only the parameters which you wish to later fix in value must occur before /PREP7.)

2. Before the next optimization execution, issue OPLOOP,PREP (Main Menu>Design

Opt>Controls) to begin reading the analysis file from the first occurrence of /PREP7.

If you do not perform both steps, each parameter that was previously a design variable will be reset
to its initial value during subsequent optimization analyses.

In the following example, we start with two design variables, AREA1 and AREA2, and perform
optimization. Then AREA2 is "deleted" (no longer a design variable) and held at its current value
(fixed) by following the steps above.

AREA1=5.00 ! AREA1 is first area

AREA2=5.00 ! AREA2 is second area
/PREP7 ! Enter PREP7 preprocessor
! Use AREA1 and AREA2 to build a parametric model
....
FINISH
/SOLVE
! Apply loads, etc. and solve
....

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 35 of 207

FINISH
/POST1
SET,...
....
*GET,SIG1,.... ! Define parameters which will be SVs and OBJ
*GET,SIG2,....
*GET,TVOL,....
....
FINISH
/OPT ! Enter optimization module
OPVAR,AREA1,DV,.... ! Assign parameters AREA1 and AREA2 as DVs
OPVAR,AREA2,DV,....
OPVAR,SIG1,SV,.... ! Assign state variables
OPVAR,SIG2,SV,....
OPVAR,TVOL,OBJ ! Assign objective function
OPTYPE,SUBP ! Use subproblem approximation method
OPEXE ! Execute
OPVAR,AREA2,DEL ! Delete design variable AREA2
STATUS ! Verify current optimization variables
OPLOOP,PREP ! Read analysis file from first /PREP7
OPTYPE,.... ! Specify desired optimization type
.... ! Specify other OPT controls
OPEXE ! Execute optimization
FINISH

Please see the OPVAR, OPTYPE, OPEXE, and OPLOOP command descriptions for more
information.

1.8.3. Modifying the Optimization Variables After Execution

Optimization variables can be modified between executions [OPEXE] by use of the OPVAR
command (Main Menu>Design Opt>Design Variables). For example, you may wish to change the
tolerance of the objective function, the upper or lower limit of a state variable, or you may decide to
delete a design variable or define a new one. Whatever the reason, if optimization variables are
modified (or new ones are added) after an optimization analysis, a partial clearing of the optimization
database is automatically performed. This does not affect the existing design sets and optimization
settings you have defined. Only information relating to the optimization calculations (transparent to
the user) is cleared. This is done to eliminate data which might be invalid for the modified
optimization variable set.

1.8.4. Local Versus Global Minimum

Sometimes the solution might terminate at a local minimum instead of at the global minimum (see
Local and Global Minima). To verify that this has not happened, you could rerun the problem with a
different starting design (that is, different initial DV values). Additional strategies for avoiding local
minima are found in Special Considerations for the First Order Method.

Local and Global Minima

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 36 of 207

1.8.5. Minimum Weight Versus Minimum Volume

Avoid specifying material density if it is not necessary for the analysis. This will save some
computer time because the mass matrix is not calculated. You can still calculate weight
parametrically as weight = density x volume, where volume is the item to be minimized (assuming
density is uniform throughout the model).

1.8.6. Mesh Density

In shape optimization problems, where the finite element mesh may change from loop to loop, it is
important to verify the adequacy of the mesh. By specifying the mesh divisions in terms of
parameters or absolute size, you can vary them appropriately for each loop.

Also, for a linear stress or thermal analysis, you can list the percent error in energy norm (see
Estimating Solution Error in the ANSYS Basic Analysis Guide) in each loop. An interesting extension
to this is to run an adaptive meshing loop within a design optimization loop to ensure that the
meshing error does not exceed a certain value. Details of adaptive meshing are described in
Guidelines for Adaptive Meshing. To list the percent error, use one of these methods:

Command(s):

PRERR

GUI:
Main Menu>General Postproc>List Results>Percent Error
Utility Menu>List>Results>Percent Error

1.8.7. Using Substructures

If only portions of a model change during design optimization, consider substructuring the
unchanging portions. The optimization run would then loop through just the use pass (and expansion
pass if necessary), resulting in a significant savings in computer time. Substructuring is described in
Substructuring.

Prev Next
Guidelines on Choosing Optimization
Sample Optimization Analysis
Variables
Sample Optimization Analysis

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 37 of 207

Prev Design Optimization Next

1.9. Sample Optimization Analysis

In the following example, you will perform an optimization analysis of a hexagonal steel plate.

1.9.1. Problem Description

You will build a parametric model of a hexagonal steel plate, using thickness t1 and fillet radius fil
as the parameters. All other dimensions are fixed.

This example uses a 2-D model and takes advantage of symmetry.

1.9.2. Problem Specifications

The loading for this example is tensile pressure (traction) of 100 MPa at the three flat faces.

The following material properties are used for this analysis:

Thickness = 10 mm
Young's modulus (E) = 2.07e5 MPa
Poisson's ratio ( ) = 0.3

1.9.3. Using a Batch File for the Analysis

You can perform the example optimization analysis of a hexagonal steel plate using the ANSYS
commands shown below. Items prefaced with an exclamation point (!) are comments.

The analysis happens in essentially two passes. In the first pass, you create the analysis file. In the
second pass, you create the optimization input. If you prefer, you can perform the second pass of the
example analysis using the GUI method rather than ANSYS commands. See Using the GUI for the
Analysis for details.

! *******************************************************
! First Pass: Create analysis file.
! *******************************************************

! GEOMETRY (in mm)

!-----------------
*afun,deg ! Degree units for trig. functions
inrad=200*cos(30)-20 ! Inner radius
t1=30 ! Thickness
fil=10 ! Fillet radius

/prep7
! Create the three bounding annuli
cyl4,-200,,inrad,-30,inrad+t1,30
cyl4,200*cos(60),200*sin(60),inrad,-90,inrad+t1,-150
cyl4,200*cos(60),200*sin(-60),inrad,90,inrad+t1,150
aplot
aadd,all
adele,all ! Delete area, keep lines
lplot
! Fillets on inner slot

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 38 of 207

lsel,,radius,,inrad+t1 ! Select inner arcs

l1 = lsnext(0) ! Get their line numbers
l2 = lsnext(l1)
l3 = lsnext(l2)
lfillet,l1,l2,fil ! Fillets
lfillet,l2,l3,fil
lfillet,l3,l1,fil
lsel,all
lplot
! Keep only symmetric portion
wprot,,90
lsbw,all
wprot,,,60
lsbw,all
csys,1
lsel,u,loc,y,0,60
ldele,all,,,1
lsel,all
ksll
ksel,inve
kdele,all ! Delete unnecessary keypoints
ksel,all
lplot
! Create missing lines and combine right edge lines
csys,0
ksel,,loc,y,0
lstr,kpnext(0),kpnext(kpnext(0)) ! Bottom symmetry edge
ksel,all
csys,1
ksel,,loc,y,60
lstr,kpnext(0),kpnext(kpnext(0)) ! 60-deg. symm. edge
ksel,all
csys,0
lsel,,loc,x,100
lcomb,all ! Add lines at the right edge
lsel,all
! Create the area
al,all
aplot

! MESHING
! -------
et,1,82,,,3 ! Plane stress with thickness
r,1,10 ! Thickness
mp,ex,1,2.07e5 ! Young's modulus of steel, MPa
mp,nuxy,1,0.3 ! Poisson's ratio
smrt,3
amesh,all
eplot
finish

! LOADING
! -------
/solu
csys,1
lsel,u,loc,y,1,59
dl,all,,symm ! Symmetry b.c.
csys,0
lsel,,loc,x,100
sfl,all,pres,-50 ! Pressure load (MPa)
lsel,all
lplot

! SOLUTION

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 39 of 207

! --------
eqslv,pcg
solve

! POSTPROCESSING
! --------------
/post1
plnsol,s,eqv ! Equivalent stress contours
/dscale,,off ! Displacement scaling off
/expand,6,polar,half,,60 ! Symmetry expansion
/replot
/expand
! Retrieve maximum equivalent stress and volume
nsort,s,eqv
*get,smax,sort,,max ! smax = max. equivalent stress
etable,evol,volu
ssum
*get,vtot,ssum,,item,evol ! vtot = total volume
finish

! *********************************************************
! Second Pass: Create optimization input.
! *********************************************************

anfile='hexplate'
anext='inp'

! ENTER OPT AND IDENTIFY ANALYSIS FILE

/opt
opanl,anfile,anext

! IDENTIFY OPTIMIZATION VARIABLES

opvar,t1,dv,20.5,40 ! DVs: Thickness
opvar,fil,dv,5,15 ! Fillet radius
opvar,smax,sv,,150 ! SV: Maximum equivalent stress
opvar,vtot,obj,,,1 ! OBJ: Total volume, tolerance = 1.0

! RUN THE OPTIMIZATION

opkeep,on ! Save best design
optype,subp ! Subproblem approximation method
opsave,anfile,opt0 ! Save the current opt database
opexe

! REVIEW RESULTS
oplist,all,,,1 ! List all design sets
plvaropt,t1,fil ! DVs t1 & fil vs. set number
plvaropt,smax ! SV smax vs. set number
plvaropt,vtot ! OBJ vtot vs. set number
finish

1.9.4. Using the GUI for the Analysis

Using a Batch File for the Analysis describes this example optimization analysis as consisting of two
passes. In the first you create an analysis file, and in the second you create the optimization input. As
discussed earlier in this chapter, you should avoid graphical picking operations when defining a
parametric model. Thus, the GUI method is not recommended for performing the first pass of the
example analysis and will not be presented here. However, it is acceptable to perform the
optimization pass of the hexagonal steel plate example using the GUI method instead of the ANSYS
commands shown earlier. The GUI procedure for performing the optimization pass follows.

Step 1: Test Analysis File

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 40 of 207

To test the analysis file, you clear the database and then read input from the hexplate.lgw file.

1. Choose menu path Utility Menu>File>Clear & Start New. Click on OK.

2. When the Verify dialog box appears, click Yes.

3. Change the jobname. Choose menu path Utility Menu>File>Change Jobname. The Change
Jobname dialog box appears.

4. Change the jobname to hexplate and click on OK.

5. Choose menu path Utility Menu>File>Read Input from. In the Files list, click on hexplate.lgw.
Then click on OK. You see a replay of the entire analysis. Click on Close when the “Solution
is done!” message appears.

Step 2: Enter the Optimizer and Identify Analysis File

In the next several steps of this problem, you optimize the design. The overdesigned steel plate under
tension loading of 100 MPa needs to be optimized for minimum weight subject to a maximum von
Mises stress limit of 150 MPa. You are allowed to vary the thickness t1 and fillet radius fil.

First, enter the optimizer and identify the analysis file.

1. Choose menu path Main Menu>Design Opt>-Analysis File-Assign. The Assign Analysis File
dialog box appears.

2. In the Files list, click once on hexplate.lgw and then click on OK.

Step 3: Identify the Optimization Variables

1. Choose menu path Main Menu>Design Opt>Design Variables. The Design Variables dialog
box appears.

2. Click on Add. The Define a Design Variable dialog box appears.

3. In the list of parameter names, click on T1. Type 20.5 in the MIN field and 40 in the MAX
field. Click on Apply.

4. In the list of parameter names, click on FIL. Type 5 in the MIN field and 15 in the MAX field.
Click on OK.

5. Click on Close to close the Design Variables dialog box.

6. Choose menu path Main Menu>Design Opt>State Variables. The State Variables dialog box
appears.

7. Click on Add. The Define a State Variable dialog box appears.

8. In the list of parameter names, click on SMAX. Type 150 in the MAX field. Click on OK.

9. Click on Close to close the State Variables dialog box.

10. Choose menu path Main Menu>Design Opt>Objective. The Define Objective Function dialog

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 41 of 207

box appears.

11. In the list of parameter names, click on VTOT. Set the TOLER field to 1.0. Click on OK.

Step 4: Run the Optimization

This step involves specifying run-time controls and the optimization method, saving the optimization
database, and executing the run.

1. Choose menu path Main Menu>Design Opt>Controls. The Specify Run-Time Controls dialog
box appears.

2. Change the OPKEEP setting from “Do not save” to “Save.” Click on OK.

3. Specify an optimization method. Choose menu path Main Menu>Design Opt>Method/Tool.

The Specify Optimization Method dialog box appears.

4. Choose Sub-Problem. Click on OK. Click OK again.

5. Save the optimization database. Choose menu path Main Menu>Design Opt>-Opt Database-
Save. In the Selection field, type hexplate.opt0. Click on OK.

6. Execute the run. Choose menu path Main Menu>Design Opt>Run. Review the settings and
click on OK. (If you receive any warning messages during the run, close them.)

7. Notes will appear to let you know which design set ANSYS is currently running. When the run
converges, review the Execution Summary. Click on OK.

Step 5: Review the Results

In this step, you start by listing design sets, then graph the objective function and state variables
versus set number.

1. Choose menu path Main Menu>Design Opt>-Design Sets-List. The List Design Sets dialog
box appears.

2. Verify that the ALL Sets option is selected. Click on OK.

3. Review the information that appears in the window. Click on Close.

4. Choose menu path Main Menu>Design Opt>Graphs/Tables. The Graph/List Tables of Design
Set Parameters dialog box appears.

5. For X-variable parameter, select Set number. For Y-variable parameters, select VTOT. Click
on OK. Review the graph.

6. Choose menu path Main Menu>Design Opt>Graphs/Tables. The Graph/List Tables of Design
Set Parameters dialog box appears.

7. For X-variable parameter, select Set number. For Y-variable parameters, select SMAX.
Unselect VTOT by clicking on it. Click on OK. Review the graph.

Step 6: Restore the Best Design

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 42 of 207

In this step, you restore the best design. First, however, save the optimization database to a file.

1. Choose menu path Main Menu>Design Opt>-Opt Database-Save. The Save Optimization Data
dialog box appears.

2. In the Selection field, type hexplate.opt1. Then click on OK.

3. Choose menu path Main Menu>Finish

4. Issue the following commands in the ANSYS Input window. After you type each command in
the window, press ENTER.
resume,hexplate,bdb
/post1
file,hexplate,brst
lplot

5. Choose menu path Main Menu>General Postproc>-Read Results-First Set.

6. Choose menu path Main Menu>General Postproc>Plot Results>-Contour Plot -Nodal Solu.
The Contour Nodal Solution Data dialog box appears.

7. Choose Stress from the list on the left. Choose von Mises SEQV from the list on the right.
Click on OK. Review the plot.

8. Choose menu path Utility Menu>PlotCtrls>Style>Displacement Scaling. For DMULT, select

0.0 (off). Click on OK.

9. Choose menu path Utility Menu>PlotCtrls>Style>Symmetry Expansion>User-Specified

Expansion. The Expansion by Values dialog box appears.

10. Fill in the 1st Expansion of Symmetry section. For NREPEAT, type 6. For TYPE, choose
Polar. For PATTERN, choose Alternate Symm. Type 0, 60, and 0 in the DX, DY, and DZ
fields, respectively. Click on OK.

Step 7: Exit ANSYS

Click on Quit in the ANSYS Toolbar. Select an option to save, then click on OK.

1.9.5. Where to Find Other Examples

Several ANSYS publications, particularly the ANSYS Verification Manual and the Design
Optimization Seminar course notes, describe additional optimization analyses.

The ANSYS Verification Manual consists of test case analyses demonstrating the analysis capabilities
of the ANSYS program. While these test cases demonstrate solutions to realistic analysis problems,
the ANSYS Verification Manual does not present them as step-by-step examples with lengthy data
input instructions and printouts. However, most ANSYS users who have at least limited finite
element experience should be able to fill in the missing details by reviewing each test case's finite
element model and input data with accompanying comments.

The ANSYS Verification Manual contains the following optimization analysis test cases:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 43 of 207

VM155 - Shape Optimization of a Cantilever Beam

VM157 - Optimization of a Frame Structure

Prev Next
General Guidelines Topological Optimization
Topological Optimization
Prev Advanced Analysis Techniques Guide Next

Chapter 2. Topological Optimization

2.1. What is Topological Optimization?
Topological optimization is a form of "shape" optimization, sometimes referred to as "layout"
optimization. The goal of topological optimization is to find the best use of material for a body such
that an objective criteria (that is, global stiffness, natural frequency, etc.) takes on a
maximum/minimum value subject to given constraints (such as volume reduction).

Unlike traditional optimization (see Design Optimization), topological optimization does not require
you to explicitly define optimization parameters (that is, independent variables to be optimized). In
topological optimization, the material distribution function over a body serves as the optimization
parameter. You define the structural problem (material properties, FE model, loads, etc.) and the
objective function (the function to be minimized/maximized), then select the state variables (the
constrained dependent variables) from among a set of predefined criteria.

The goal of topological optimization - the objective function - is to minimize/maximize the criteria
selected (minimize the energy of structural compliance, maximize the fundamental natural
frequency, etc.) while satisfying the constraints specified (volume reduction, etc.). This technique
uses design variables ( i) that are internal pseudo-densities assigned to each finite element. These
densities are plotted with the PLNSOL,TOPO and PLESOL,TOPO commands (see Review the
Results).

The standard formulation of topological optimization defines the problem as minimizing the the
structural compliance while satisfying a constraint on the volume (V) of the structure. Minimizing
the compliance is equivalent to maximizing the global structural stiffness. For example, specifying V
= 60 means that 60% of the material is to be removed in a manner that maximizes the stiffness, with
the given load configuration. An Optimization Sample With 60% Volume Reduction shows a
constrained and loaded rectangular area that is to be subjected to topological optimization. Figure (a)
shows the loads and boundary conditions and Figure (b) shows the "shape" results in terms of density
contours.

An Optimization Sample With 60% Volume Reduction

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 44 of 207

Prev Next
Sample Optimization Analysis How to Do Topological Optimization
How to Do Topological Optimization
Prev Topological Optimization Next

2.2. How to Do Topological Optimization

The procedure for topological optimization consists of the following main steps.

1. Define the structural problem.

2. Select the element types.

3. Specify optimized and non-optimized regions.

4. Define and control the load cases or frequency extraction.

5. Define and control the optimization process.

6. Review the results.

Details of the optimization procedure are presented below. Where appropriate, differences in the
procedure for a "batch" versus "interactive" approach are indicated.

2.2.1. Define the Structural Problem

Define the problem as you would for any linear elastic analysis. You need to define material
properties (Young's modulus, Poisson's ratio, and possibly the material density). Poisson's ratio must
be between 0.1 and 0.4. You then select the proper element types for topological optimization,
generate a finite element model, and depending on the criteria you need for your particular
topological optimization problem, you will need to apply either:

? Load and boundary conditions for a single or multiple load case linear structural static
analysis, or

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 45 of 207

? Boundary conditions for a modal frequency analysis.

See Getting Started with ANSYS and Loading in the ANSYS Basic Analysis Guide for more
information on defining the problem.

2.2.2. Select the Element Types

Topological optimization supports 2-D planar, 3-D solid, and shell elements. To use this technique,
your model must contain only the following element types:

2-D Solids: PLANE2 or PLANE82

3-D Solids: SOLID92 or SOLID95
Shells: SHELL93

The 2-D elements should be used for plane stress or axisymmetric applications.

2.2.3. Specify Optimized and Non-Optimized Regions

Only those elements identified as type 1 (TYPE) will be subjected to topological optimization. Use
this rule to control which regions of your model to optimize or not. For example, if you want to keep
material near a hole or a support, you should identify elements in those areas as type 2 or higher:

...
ET,1,SOLID92
ET,2,SOLID92
...
TYPE,1
VSEL,S,NUM,,1,2 ! The volume modeled by these elements will be
VMESH,ALL ! optimized
TYPE,2
VSEL,S,NUM,,3 ! The volume modeled by these elements will not
VMESH,ALL ! be optimized
...

You can use any appropriate ANSYS select and modification command to control the type
definitions for various elements.

2.2.4. Define and Control Your Load Cases or Frequency Extraction

You can perform topological optimization based on either linear structural static or modal analysis.

2.2.4.1. Linear Structural Static Analysis

When defining the structural compliance as either the objective or constraint for your topological
optimization (TOCOMP, TOVAR), you must perform a linear structural static analysis during
optimization looping. You can perform topological optimization for a single load case or collectively
for several load cases. The single load case situation is the simplest and is explained in Solve the
Problem.

However, to obtain a single optimization solution from several independent load cases, you must use
load case write and solve capabilities. After each load case is defined, you must write that data to a
file (use the LSWRITE command). Finally, you need to solve the collection of load cases by using
the LSSOLVE command. The TOLOOP command performs this last step for you.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 46 of 207

For example, the following input fragment shows how three load cases can be combined for a single
topological optimization analysis.

...
...
D,10,ALL,0,,20,1 ! Define Loads and constraints for 1st load case
NSEL,S,LOC,Y,0
SF,
ALLSEL
LSWRITE,1 ! Write 1st load case
DDEL, ! Clear and then define 2nd load case
SFDEL,
NSEL,S,LOC,X,0,1
D,ALL,ALL,0
F,212,FX
LSWRITE,2 ! Write 2nd load case
... ! Etc.
LSWRITE,3 ! Write 3rd load case
... ! Etc.
FINISH
TOCOMP,MCOMP,MULTIPLE,3 ! Define weighted multiple compliance function
"MCOMP"
! considering all three load cases
TOVAR,MCOMP,OBJ ! Define "MCOMP" as topological objective
TOVAR,VOLUME,CON,,10 ! Define "VOLUME" as topological constraint
TODEF ! Initialize topo opt.
TOLOOP,20 ! Solve and perform topological optimization

2.2.4.2. Modal Analysis

When defining a natural frequency formulation as an objective for topological optimization

(TOFREQ, TOVAR) a modal analysis must be performed during optimization looping. Depending
on the frequency model you need to specify the number of frequencies to solve for during modal
analysis by using the MODOPT command. You also must specify the number of modes to expand
and write by means of MXPAND. Note that the element calculation key must be set to 'YES'. The
following input fragment shows the modal analysis set-up for topological optimization. When using
the TOLOOP command macro, you need only the ANTYPE command because the MODOPT and
MXPAND commands are done by the TOLOOP command. Only the Block Lanczos eigensolver is
available with the TOLOOP command.

/SOLUTION
ANTYPE,MODAL ! Switch to modal analysis
FINISH

TOFREQ,MFREQ,RECIPROCAL,3 ! Define reciprocal frequency function "MFREQ"

! for topological optimization
TOVAR,MFREQ,OBJ ! Define function "MFREQ" as objective for
topological
! optimization
TOVAR,VOLUME,CON,,50 ! Define "VOLUME" as constraint for topological
! optimization
TODEF,1.0D-4 ! Initialize topological optimization
(accuracy=1
TOLOOP,20 ! Solve for first 3 natural frequencies and
corresponding
! mode shapes and then perform topological
optimization
...

2.2.5. Define and Control the Optimization Process

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 47 of 207

The topological optimization process consists of four parts: defining optimization functions, defining
objective and constraints, initializing optimization, and executing topological optimization. You can
run the fourth part, executing topological optimization, in two ways; carefully controlling and
executing each iteration, or automatically performing many iterations. ANSYS recommends the
latter approach.

Seven ANSYS commands define and execute topological optimization: TOFREQ, TOCOMP,
TOVAR, TOTYPE, TODEF, TOEXE, and TOLOOP. The commands TOCOMP and TOFREQ
are used to define topological optimization functions. Command TOVAR defines the objective and
constraint functions for the optimization problem. The TOTYPE command defines the solution
approach employed to solve the optimization problem. The TODEF command defines a tolerance
for convergence and initializes the topological optimization process. TOEXE executes a single
iteration of optimization. TOLOOP executes several iterations.

Note

To correctly perform topological optimization, after you have defined the

optimization parameters, you must solve the problem (SOLVE) before executing one
or more optimization iterations (TOEXE). If you do not solve the problem, you will
receive an error message when you try to execute an optimization step. The
TOLOOP command macro performs the solve steps for you.

2.2.5.1. Defining Optimization Functions

First, you define the optimization functions involved in your topological optimization problem. Use
TOCOMP to define a compliance function for single or multiple load case conditions in a linear
static structural analysis. In contrast, TOFREQ is intended to define natural frequency type
functions. For every topological function, a unique reference name must be specified (note that the
reference name "VOLUME" is a predefined name for the total volume function). You may also
delete a topological optimization function using TOCOMP,RefName or TOFREQ,RefName (with
remaining fields left blank). The delete option also deactivates the function as a topological objective
or constraint if the TOVAR command has already been used.

To define a function for topological optimization, use one of these methods:

Command(s):

TOCOMP, TOFREQ

GUI:

Main Menu>Topological Opt>Advanced Opt>Topo Function

Main Menu>Topological Opt>Basic Opt

To list all topological optimization functions currently defined use TOLIST .

Command(s):

TOLIST

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 48 of 207

Main Menu>Topological Opt>Advanced Opt>List Functions

2.2.5.2. Defining Objective and Constraints

The next step is to assign the objective and the constraint(s) for the topological optimization
problem, that is, specify which functions, previously defined by means of TOCOMP and TOFREQ,
are constraints, and which one is the objective function. A pre-defined topological function
"VOLUME" is provided (representing the total volume function), which can be used for either
objective or constraint. Note that only the following combinations of objective and constraints are
allowed:

Objective Valid Constraints

Single Compliance (TOCOMP) VOLUME
Multiple Compliance (TOCOMP) VOLUME
Single Frequency (TOFREQ) VOLUME
Weighted mean frequency
VOLUME
(TOFREQ)
Reciprocal mean frequency
VOLUME
(TOFREQ)
Euclidean norm frequency VOLUME
(TOFREQ)
Single Compliance (TOCOMP), multiple constraint
VOLUME
definition allowed
VOLUME Multiple Compliance (TOCOMP)

To assign the objective and constraint(s) for topological optimization, use one of these methods:

Command(s):

TOVAR

GUI:

Main Menu>Topological Opt>Advanced Opt>Topo Objective

Main Menu>Topological Opt>Basic Opt

The objective function must be defined before you define constraints. Minimum and maximum
bounds can be specified for constraint definition. No constraints are needed for the objective
function.

You may change a previously declared objective or constraint at any time by simply redefining it.
You may also delete an objective or constraint (TOVAR,RefName,DEL). The delete option does not
delete the topological function; it simply deactivates the function as a topological objective or
constraint.

For example, the following input fragment shows how a single compliance minimization subject to a
volume reduction of 25% is defined

...
TOCOMP,COMP,SINGLE,1 ! Define single compliance (load case 1) as

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 49 of 207

topological
! optimization function
! "COMP"
TOVAR,COMP,OBJ ! Define the compliance function "COMP" as
objective for
! topological optimization
TOVAR,VOLUME,CON,,25 ! Define pre-defined total volume function
"VOLUME" as constraint
! for topological optimization; Specify a volume
reduction of 25%
TODEF,1.0d-4 ! Initialize topological optimization
TOLOOP,10,1 ! Do 10 topological optimization iterations
automatically
...

At any time you can query the current active status from TOVAR, TODEF, and TOTYPE by using
the command TOSTAT.

Command(s):

TOSTAT

GUI:

Main Menu>Solution>-Solve- Topological Opt>Advanced Opt>Status [Topological

Opt>Status]

2.2.5.3. Solving and Initializing Optimization

After defining your optimization parameters, solve the problem (SOLVE). You must solve the
problem before you perform topological optimizations.

Command(s):

SOLVE

GUI:

Main Menu>Solution>-Solve - Current LS

After specifying the optimization problem (see Defining Optimization Functions and Defining
Objective and Constraints) you may select the solution approach you want employed to solve the
optimization problem; TOTYPE allows you to choose Optimality Criteria (OC) or Sequential
Convex Programming (SCP). The OC approach is applicable to problems with volume as constraint
only. The SCP approach is applicable to all valid combinations of objective and constraints (see
Defining Objective and Constraints for a list of valid combinations of objective and constraints).

Command(s):

TOTYPE

GUI:

Main Menu>Topological Opt>Run

As a last preparation step you must initialize the topological optimization process. Here you also

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 50 of 207

define the termination/convergence accuracy.

Command(s):

TODEF

GUI:

Main Menu>Topological Opt>Run

The specification details generated at this point are not saved in the ANSYS database. Therefore, if
you want to perform another topological optimization after a RESUME, you need to reissue all of
the commands you used to set up the topological optimization problem (TOCOMP, TOFREQ,
TOVAR, TOTYPE, and TODEF).

2.2.5.4. Executing a Single Iteration

After defining your optimization parameters, you can launch a single iteration. After execution, you
can check convergence and display and/or list your current topological results. You may continue to
solve and execute additional iterations until you achieve the desired result. The TOEXE command is
not available in the GUI.

Command(s):

TOEXE

The following example demonstrates how you would execute topological optimization one iteration
at a time:

...
TOCOMP,COMP,SINGLE,1 ! Define single compliance (load case 1) as topological
! optimization function
! "COMP"
TOVAR,COMP,OBJ ! Define the compliance function "COMP" as objective for
! topological optimization
TOVAR,VOLUME,CON,,25 ! Define pre-defined total volume function "VOLUME" as
constraint
! for topological optimization; Specify a volume
reduction of 25%
TOTYPE,OC ! Use OC approach for optimization problem
TODEF,1.0d-4 ! Initialize topological optimization
/SOLUTION
SOLVE ! Perform 1st stress analysis
TOEXE ! Perform 1st topological iteration
FINISH
/POST1 ! Enter post processing
PLNSOL,TOPO ! Plot topological results
*GET,TOPSTAT,TOPO,,CONV ! Get the topological convergence status
*STAT,TOPSTAT ! List convergence status
/SOLUTION
SOLVE ! Perform 2nd stress analysis
TOEXE ! Perform 2nd topological optimization
FINISH
/POST1 ! Etc.
...

The following is an input fragment showing how you could perform a frequency maximization
formulation one iteration at a time.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 51 of 207

...
TOFREQ,FREQ1,SINGLE,1 ! Define single frequency as topological optimization
! function "FREQ1"
TOVAR,FREQ1,OBJ ! Define the frequency function "FREQ1" as objective for
! topological optimization
TOVAR,VOLUME,CON,,25 ! Define pre-defined total volume function "VOLUME" as
! constraint for topological optimization; Specify a
volume
! reduction of 25%
TOTYPE,SCP ! Use SCP approach for optimization problem
TODEF,1.0d-4 ! Initialize topological optimization
/SOLUTION
ANTYPE,MODAL ! Switch to modal analysis
MODOPT,LANB,1 ! Specify modal analysis options: Choose Block Lanczos
! solver (default); extract 1 eigenmode.
MXPAND,1,,,YES ! Expand fundamental mode, and set element
calculation key to YES.
SOLVE ! Perform modal analysis
TOEXE ! Perform 1st topological iteration
FINISH
TOPLOT,0 ! Plot topological results
*GET,TOPSTAT,TOPO,,CONV ! Get the topological convergence status
*STAT,TOPSTAT ! List convergence status
/SOLUTION
SOLVE ! Perform 2nd modal analysis
TOEXE ! Perform 2nd topological optimization
FINISH
TOPLOT,0
... ! Etc.

One of the main advantages of TOEXE is that you can use it to devise your own iterative macros for
automatic optimization looping and plotting. In Executing Several Iterations Automatically, you will
see that the TOLOOP command is actually an ANSYS macro that can be used to perform several
optimization iterations.

2.2.5.5. Executing Several Iterations Automatically

After defining your optimization parameters, you can launch several iterations to be executed
automatically. After all of the iterations have run, you can check convergence and display and/or list
your current topology. You may continue to solve and execute additional iterations if you want. The
TOLOOP command is actually an ANSYS macro and, as such, can be copied and customized (see
the ANSYS APDL Programmer's Guide).

Command(s):

TOLOOP

GUI:

Main Menu>Solution>-Solve-Topological opt>Run

The following example demonstrates how you would use the TOLOOP macro to execute multiple
iterations (in this case, 20) automatically:

... ! Setup, define and

LSWRITE ! write 1st load case
... ! Setup, define and
LSWRITE ! write 2nd load case
... ! Setup, define and

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 52 of 207

LSWRITE ! write 3rd load case

...
TOCOMP,MCOMP,MULTIPLE,3 ! Define multiple compliance (3 load cases) as
topological
! optimization function "MCOMP"
TOVAR,MCOMP,OBJ ! Define the compliance function "MCOMP" as objective
for
! topological optimization
TOVAR,VOLUME,CON,,80 ! Define pre-defined total volume function "VOLUME" as
constraint for
! topological optimization; Specify a volume reduction
of 80%
TODEF,0.001 ! Initialize topological optimization with .001
convergence tolerance
...
/DSCALE,,OFF ! Remove distortion
/CONTOUR,,3 ! Request 3 contours for each display
TOLOOP,20,1 ! Perform 20 (max.) iterations. Each iteration solves and
! plots results
...

Each topological iteration executes an LSSOLVE command, a TOEXE command, and a

PLNSOL,TOPO display (optional) command. The optimization iteration process terminates once
convergence is attained (defined with TODEF) or when the maximum iteration number is achieved
(defined with TOLOOP)

2.2.6. Review the Results

After your topological optimization solutions are complete, pertinent results are stored on the
ANSYS results file ( Jobname.RST) and are available for additional POST1 processing. You can use
the following postprocessing options. For more information on any of these options, see the ANSYS
Commands Reference for the particular command description, or see The General Postprocessor
(POST1) in the ANSYS Basic Analysis Guide.

For a nodal listing or plot of the pseudo densities, use the TOPO argument of the PRNSOL and
PLNSOL commands. Or you can use the command TOPLOT,0.

For an element-based listing or plot of pseudo densities, use the TOPO argument of the PLESOL or
PRESOL commands. Or you can use the command TOPLOT,1.

Additionally you may graph or print the history of the topological optimization course over iterations
by using the commands TOGRAPH or TOPRINT.

...
/POST1 ! Enter post processor
TOPLOT,1 ! Plot non-averaged element pseudo densities
PLNS,TOPO ! Plot averaged nodal pseudo densities
TOGRAPH,OBJ ! Plot iteration history of topological objective
TOGRAPH,CON,VOLUME ! Plot iteration history of topological constraint
"VOLUME"
TOPRINT,OBJ ! Print iteration history of topological objective
TOPRINT,CON ! Plot iteration history of topological constraint
"VOLUME"
...

You can also view the results via ANSYS' tabular capabilities:

...
ETABLE,EDENS,TOPO

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 53 of 207

PLETAB,EDENS
PRETAB,EDENS
ESEL,S,ETAB,EDENS,0.9,1.0
EPLOT
...

To check the most recent convergence status (that is, the last iteration) and the objective or constraint
values, use *GET:

...
*GET,TOPCV,TOPO,,CONV ! If TOPCV = 1 (converged)
*GET,TITER,TOPO,,ITER ! TITER = Iteration counter
*GET,TOBJ,TOPO,ITER-1,TOHO ! TOBJ = objective function value of last
iteration
*GET,TCON,TOPO,ITER-1,TOHC,1 ! TCON = constraint function value of last
iteration
*STAT

For a frequency solution, particularly when using the weighted, reciprocal, or Euclidean
formulations, you should look at the true frequencies of the structure at the converged solution by
issuing the *GET command:

...
*GET,FREQ1,MODE,1,FREQ ! First fundamental frequency
*GET,FREQ2,MODE,2,FREQ ! Second fundamental frequency

You should also look at the mode shapes in POST1 by using the PLDISP command, and you should
review the ifreq.out file. For more information on reviewing results of a modal analysis, see
Modal Analysis in the ANSYS Structural Analysis Guide.

Prev Next
A 2-D Multiple-Load Case
Topological Optimization Optimization Example
A 2-D Multiple-Load Case Optimization Example
Prev Topological Optimization Next

2.3. A 2-D Multiple-Load Case Optimization Example

In this sample analysis, you will run topological optimization on a beam subjected to two load cases.

2.3.1. Problem Description - First Scenario

A loaded, elastic beam is shown in Beam With Two Load Cases. The beam is fixed along both ends
and subjected to two load cases. Notice that one area of the beam is modeled by type 1 (TYPE) finite
elements and can therefore be subjected to topological optimization. The other area, identified and
modeled by type 2 elements, is not optimized. The goal is to minimize the energy of structural
compliances (for each load case independently) subject to a 50% reduction in volume of type 1
material.

Beam With Two Load Cases

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 54 of 207

This problem is solved using the ANSYS commands below. Notice that the two load cases are
defined and written by the LSWRITE command. Using ANSYS selection commands and logic, type
1 and type 2 PLANE82 elements are used to represent optimized and non-optimized areas,
respectively. The TOCOMP command defines a 2 load case compliance function with the reference
name MCOMP. TOVAR defines MCOMP as the objective and calls for a 50% volume reduction
(TOVAR,VOLUME,CON,,50). The TOEXE command is not explicitly used in this example.
Instead, up to 12 iterations of topological optimization are specified via the TOLOOP command
macro. After the optimization execution, the final objective (compliance) and constraint (volume)
histories are graphed and printed, and the optimum weighted compliance value is retrieved from the
ANSYS database (*GET).

/TITLE, A 2-d, multiple compliane minimization problem subjected

to volume constraint
/PREP7
BLC4,0,0,3,1 ! Create solid model (3 x 1 rectangle)
ET,1,82 ! Use 2-d solids. Type 1 is optimized
ET,2,82 ! Type 2 is not optimized.
MP,EX,1,118E9 ! Linear isotropic, material
MP,NUXY,1,0.3
ESIZE,0.05 ! Use a relatively fine mesh density
TYPE,1
AMESH,ALL ! Free, rectangular-element meshing
NSEL,S,LOC,X,0,0.4 ! Select region not to be optimized
ESLN
TYPE,2
EMODIF,ALL ! Define type 2 elements
ALLSEL
NSEL,S,LOC,X,0
D,ALL,ALL,0 ! Fixed at X = 0
NSEL,S,LOC,X,3
D,ALL,ALL,0 ! Fixed at X = 3
FORCE = 1000 ! Value for applied load
NSEL,S,LOC,X,1
NSEL,R,LOC,Y,1
F,ALL,FY,FORCE ! Define first load case

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 55 of 207

ALLSEL
LSWRITE,1 ! Write first load case
FDEL,ALL
NSEL,S,LOC,X,2
NSEL,R,LOC,Y,0
F,ALL,FY,-FORCE ! Define second load case
ALLSEL
LSWRITE,2 ! Write second load case
FDEL,ALL
TOCOMP,MCOMP,MULTIPLE,2 ! Define multiple compliance function
! "MCOMP" for topological optimization
TOVAR,MCOMP,OBJ ! Define "MCOMP" as topological objective
TOVAR,VOLUME,CON,,50 ! Define "VOLUME" as topological constraint; 50%
volume reduction
TOTYPE,OC ! Specify solution approach
TODEF ! Initialize topological opt.
/SHOW,topo,grph ! Put graphics in a file (remove if interactive)
/DSCALE,,OFF
/CONTOUR,,2
TOLOOP,12,1 ! Perform no more than 12 iterations
FINISH
TOGRAPH,OBJ ! Graph final objective (compliance) history
TOGRAPH,CON ! Graph final constraint (volume) history
TOPRINT,OBJ ! Print final objective (compliance) history
TOPRINT,CON ! Print final constraint (volume) history
*GET,TITER,TOPO,,ITER ! Get iteration counter
*GET,OCMP,TOPO,TITER-1,TOHO ! Get final compliance value

2.3.2. Problem Results - First Scenario

The final optimal shape obtained from the above input stream is shown in Final Topological Shape -
50% Volume Reduction. Notice that these results were diverted to the topo.grph file for subsequent
display. If running ANSYS in an interactive mode, you should remove the /SHOW command so you
can view graphical results every iteration.

Final Topological Shape - 50% Volume Reduction

A graph of the objective (compliance) and the constraint (volume) history is shown in History of
Objective and Constraint Functions. The final optimal weighted compliance value is 0.6E-04.

History of Objective and Constraint Functions

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 56 of 207

2.3.3. Problem Description - Second Scenario

In contrast to the first scenario, where we have designed for minimum 2-load-case compliance
subject to 50% volume reduction, in this second scenario we will optimize for minimum volume
design subject to 2 compliance constraints. The analysis is based on the same finite element model,
boundary and load conditions as in the first scenario (see Final Topological Shape - 50% Volume
Reduction and the input fragment in Problem Description). Here, the topological optimization
problem is modeled as follows. We first define the volume of type 1 material as objective function
by using command TOVAR. We then specify two single compliance functions, "COMP1" for the
first load case and "COMP2" for the second load case by means of command TOCOMP and define
them as topological constraints (TOVAR) with an upper bound value of 0.6E-4 (the optimum

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 57 of 207

compliance value for the first scenario). Notice that SCP was selected as the solution approach for
the topological optimization problem because this problem cannot be solved with the OC approach.

/TITLE, A 2-d, volume minimization problem subjected to 2 compliance

constraints
... ! Same modeling, boundary conditions and load cases as in
previous listing

TOCOMP,COMP1,SINGLE,1 ! Define single compliance function "COMP1"

(1st load case)
TOCOMP,COMP2,SINGLE,2 ! Define single compliance function "COMP2"
(2nd load case)
TOVAR,VOLUME,OBJ ! Define "VOLUME" of type 1 material as
topological objective
TOVAR,COMP1,CON,,0.6E-4,ACTUAL ! Define first constraint "COMP1" with an
actual upper bound of
! 0.6E-4
TOVAR,COMP2,CON,,0.6E-4,ACTUAL ! Define second constraint "COMP2" with an
actual upper bound of
! 0.6E-4

TOTYPE,SCP ! Specify SCP solution approach

TODEF,1.0d-5 ! Initialize topological opt; set accuracy to
1.0d-5
/SHOW,topo2,grph ! Put graphics in a file (remove if interactive)
TOLOOP,25,1 ! Perform 25 iterations
FINISH
TOPLOT,1 ! Plot final densities un-averaged
TOGRAPH,OBJ ! Graph final objective (compliance) history
TOGRAPH,CON,COMP1 ! Graph final constraint (volume) history
TOGRAPH,CON,COMP2 ! Graph final constraint (volume) history

2.3.4. Problem Results - Second Scenario

The final optimal shape obtained from the above commands are shown in Final Topological Shape
for Second Scenario. Notice that these results were diverted to the topo2.grph file for subsequent
display. If running ANSYS in an interactive mode, you should remove the /SHOW command in
order to view graphical results every iteration.

Final Topological Shape for Second Scenario

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 58 of 207

A graph of the objective (volume) and both constraint (compliances) histories is shown in History of
Objective and Constraint Functions for Second Scenario.

History of Objective and Constraint Functions for Second Scenario

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 59 of 207

Prev Next
A 2-D Natural Frequency
How to Do Topological Optimization Maximization Example
A 2-D Natural Frequency Maximization Example
Prev Topological Optimization Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 60 of 207

2.4. A 2-D Natural Frequency Maximization Example

In this example analysis, you will run an optimal reinforcement problem of a two -story planar frame
as shown in Two-Story Planar Frame.

2.4.1. Problem Description

This is an initial core-structure (type 2 material) with four concentrated masses specified and
assumed to be unchanged during the optimization process. As shown in Two-Story Planar Frame, the
design domain is specified as a rectangle, 5.0 in horizontal length and 8.0 in vertical height with two
fixed supported boundaries at the bottom of the domain. Within the optimization problem, material is
added (type 1 area) to reinforce the core-structure subjected to dynamic stiffness. The design domain
is filled by a material with a Young's Modulus of E = 100, Poisson's Ratio = 0.3, and density = 1.0D-
06. Each concentrated mass is 5.0D-06.

Two-Story Planar Frame

In this scenario, we maximize the fundamental frequency, whereas the constraint of the total type 1

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 61 of 207

volume was specified as V = 14. Thus, we define a single frequency function "FREQ1" using
TOFREQ and specify this function to be the objective for topological optimization (TOVAR). We
also define the volume constraint with an actual upper bound of 14
(TOVAR,VOLUME,,14,ACTUAL). Again, the TOEXE command is not explicitly used in this
example. Instead, a maximum of 40 iterations of topological optimization are selected via the
TOLOOP command macro.

/TITLE, 2-D Two-Story reinforcement problem - Maximize fundamental frequency

A=0.25 ! Prepare Model
B=5
C=0.375
D=8
E=3.75
/PREP7
K,1
K,2,C
K,3,C+A
K,4,B/2
K,5,,E
K,6,C,E
K,7,C+A,E
K,8,B/2,E
KSEL,S,,,5,8
KGEN,2,ALL,,,,A
KSEL,S,,,9,12
KGEN,2,ALL,,,,E
KSEL,S,,,13,16
KGEN,2,ALL,,,,A
ALLSEL
A,1,2,6,5
A,5,6,10,9
A,9,10,14,13
A,13,14,18,17
A,2,3,7,6
A,6,7,11,10
A,10,11,15,14
A,14,15,19,18
A,3,4,8,7
A,7,8,12,11
A,11,12,16,15
A,15,16,20,19

ET,1,82 ! Define two element type regions

ET,2,82 ! 1 - optimized region
ASEL,S,,,4,8 ! 2 - non-optimized region
ASEL,A,,,10,12,2
TYPE,2
ESIZE,0.1
AMESH,ALL
ASEL,INVE
TYPE,1
AMESH,ALL
ALLSEL
MP,EX,1,100 ! Material of structure
MP,NUXY,1,0.3
MP,DENS,1,1.0D-6
MP,EX,2,100 ! Material of concentrated masses
MP,NUXY,2,0.3
MP,DENS,2,5.0D-6

ASEL,S,,,6,8,2
ESLA,S,1
EMODIF,ALL,MAT,2 ! Define concentrated masses

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 62 of 207

ALLSEL
LOCAL,11,0,2.5
ARSYM,X,ALL ! Full model
NUMM,KP
NUMM,ELEM
NUMM,NODE
LSEL,S,,,14
LSEL,A,,,45
NSLL,S,1
D,ALL,ALL
ALLSEL
FINISH
TOFREQ,FREQ1,SING,1 ! Define single frequency function (1st)
TOVAR,FREQ1,OBJ ! Define objective for topological optimization
TOVAR,VOLUME,CON, ,14,ACTUAL ! Define volume constraint (upper bound = 14)
TOTYPE,SCP ! Select SCP solution approach
TODEF,0.00001, ! Initialize topological optimization process,
accuracy = 0.00001
TOLOOP,40,1 ! Perform up to 40 iterations
FINISH
TOPLOT,1 ! Plot final pseudo densities
TOGRAPH,OBJ ! Graph final objective (1st frequency) history
TOGRAPH,CON ! Graph final constraint (volume) history

2.4.2. Problem Results

Final Topological Shape for Maximum Fundamental Frequency shows the optimal shape of the
reinforcement optimization problem and History of Fundamental Frequency the history of the first
frequency and volume, respectively (TOGRAPH) over optimization iteration.

Final Topological Shape for Maximum Fundamental Frequency

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 63 of 207

History of Fundamental Frequency

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 64 of 207

Prev Next
A 2-D Multiple-Load Case
Hints and Comments
Optimization Example
Hints and Comments
Prev Topological Optimization Next

2.5. Hints and Comments

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 65 of 207

? Results are sensitive to your load configuration. Small changes to your applied loads or load
distributions can lead to significant differences in results.

? Results are sensitive to the density of the finite element mesh. In general, a very fine mesh will
produce "clear" topological results. A coarse mesh will lead to "fuzzier" results. However, a
model with a large number of elements will take more time to reach convergence.

? Under certain circumstance, a truss-like solution will result. This will happen when you
request a high volume reduction and a very fine finite element mesh. For example, a large
reduction could be 80% or more volume removed (set in TOVAR).

? If you have several load cases, there are many ways to combine your loads and to generate
topological results. Consider, for example, what you could do with five different load cases.
You may choose to perform five separate topological optimization analyses, or you could
perform a single optimization analysis for all five load independent cases. Also, you could
combine all five loads into one load case and perform a single topological analysis. In
summary, you could produce seven different topological optimization solutions:
? 5 independent topological solutions (1 for each load case)

? 1 topological solution for 5 independent load cases

? 1 topological solution for the combined load cases

Additional results and combination of results are also possible.

? Results are sensitive to Poisson's ratio but not Young's modulus. However, the effects of the
dependence to Poisson's ratio are usually not significant.

? Maximizing a chosen natural frequency is usually used as the objective in a frequency

topological optimization problem. However, in the frequency optimization problem, when one
maximizes a lower frequency, higher eigenvalues may fall down to the lower values. This
means that if the optimization process is to follow a specified mode of the structure, then the
order number of this mode may be changed during the optimization process. For example, at
the beginning we may wish to optimize the kth eigenfrequency, but the optimal solution
obtained may correspond to the k+p'th mode, where p > 0. Thus the problem can have an
unexpected solution. In contrast, if you follow the number of modal order (for example, to
optimize the kth eigenfrequency), then the mode being optimized may change to another one.
In this case, the sensitivities of the objective function become discontinuous, and may cause
oscillation and divergence in the iterative optimization process. To overcome this problem,
several mean-frequency functions (see TOFREQ) can be used to smooth out the frequency
objective.

? The specifications set using TOCOMP, TOFREQ, TOVAR, TODEF, TOTYPE, and
TOLOOP are not saved in the ANSYS database; therefore, you will need to specify your
optimization goals and definitions each time you use topological optimization.

Prev Next
A 2-D Natural Frequency
Probabilistic Design
Maximization Example
Probabilistic Design
Prev Advanced Analysis Techniques Guide Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 66 of 207

Chapter 3. Probabilistic Design

3.1. What Is Probabilistic Design?
Probabilistic Design is a technique you can use to assess the effect of uncertain input parameters and
assumptions on your analysis model. Using a probabilistic analysis you can find out how much the
results of a finite element analysis are affected by uncertainties in the model. An uncertainty or
random quantity is a parameter where it is impossible to tell the exact value at a given point in time
(if it is time dependent) or at a given location (if it is location dependent). One example of this is
ambient temperature, where you never know exactly what the temperature will be one week from
now in a certain city. In a probabilistic analysis these uncertain parameters are described by
statistical distribution functions such as the Gaussian or normal distribution, the uniform distribution,
etc. (See the ANSYS Modeling and Meshing Guide for a description of ANSYS parameters.)

Computer models are expressed and described with specific numerical and deterministic values;
material properties are entered using certain values, the geometry of the component is assigned a
certain length or width, etc. An analysis based on a given set of specific numbers and values is called
a deterministic analysis. Naturally, the results of a deterministic analysis are only as good as the
assumptions and input values used for the analysis. The validity of those results depend on how
correct the values were for the component under real life conditions.

In reality, literally every aspect of an analysis model is subjected to scatter (in other words, is
uncertain in some way). Material property values are different if one specimen is compared to the
next. This kind of scatter is inherent for materials and varies among different material types and
material properties. For example, the scatter of the Young's modulus for many materials can often be
described as a Gaussian distribution with standard deviation of ±3 - 5%. Likewise, the geometric
properties of components can only be reproduced within certain manufacturing tolerances. The same
variation holds true for the loads that are applied to a finite element model. However, in this case the
uncertainty is often due to a lack of engineering knowledge. For example, at elevated temperatures
the heat transfer coefficients are very important in a thermal analysis, yet it is almost impossible to
measure the heat transfer coefficients. This means that almost all input parameters used in a finite
element analysis are inexact, each associated with some degree of uncertainty.

It is neither physically possible nor financially acceptable to eliminate the scatter of input parameters
completely. The reduction of scatter is typically associated with higher costs either through better
and more precise manufacturing methods and processes or increased efforts in quality control.
Hence, accepting the existence of scatter and dealing with it rather then trying to eliminate it makes
products more affordable and production of those products more cost effective.

To deal with uncertainties and scatter, you can use the ANSYS Probabilistic Design System (PDS) to
answer the following questions:

? If the input variables of a finite element model are subjected to scatter, how large is the scatter
of the output parameters? How robust are the output parameters? Here, output parameters can
be any parameter that ANSYS can calculate. Examples are the temperature, stress, strain, or
deflection at a node, the maximum temperature, stress, strain, or deflection of the model, etc.

? If the output is subjected to scatter due to the variation of the input variables, then what is the
probability that a design criterion given for the output parameters is no longer met? How large
is the probability that an unexpected and unwanted event takes place (what is the failure
probability)?

? Which input variables contribute the most to the scatter of an output parameter and to the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 67 of 207

failure probability? What are the sensitivities of the output parameter with respect to the input
variables?

Probabilistic design can be used to determine the effect of one or more variables on the outcome of
the analysis. In addition to the probabilistic design techniques available, the ANSYS program offers
a set of strategic tools that can be used to enhance the efficiency of the probabilistic design process.
For example, you can graph the effects of one input variable versus an output parameter. And you
can easily add more samples and additional analysis loops to refine your analysis.

3.1.1. Traditional (Deterministic) vs. Probabilistic Design Analysis Methods

In traditional deterministic analyses, uncertainties are either ignored or accounted for by applying
conservative assumptions.

Uncertainties are typically ignored if the analyst knows for certain that the input parameter has no
effect on the behavior of the component under investigation. In this case, only the mean values or
some nominal values are used in the analysis. However, in some situations the influence of
uncertainties exists but is still neglected; for example, the Young's modulus mentioned above or the
thermal expansion coefficient, for which the scatter is usually ignored. Let's assume you are
performing a thermal analysis and you want to evaluate the thermal stresses (thermal stresses are
directly proportional to the Young's modulus as well as to the thermal expansion coefficient of the
material). The equation is

therm =E T

If the Young's modulus alone has a Gaussian distribution with a 5% standard deviation, then there is
almost a 16% chance that the stresses are more than 5% higher than what you would think they are in
a deterministic case. This figure increases if you also take into account that, typically, the thermal
expansion coefficient also follows a Gaussian distribution.

Probability that the

Probability that the thermal
Random Input Variables taken thermal stresses are more
stresses are more than 10%
into account than 5% higher than
higher than expected
expected
Young's modulus (Gaussian
distribution with 5% standard ~16% ~2.3%
deviation)
Young's modulus and thermal
expansion coefficient (each with
~22% ~8%
Gaussian distribution with 5%
standard deviation)

When a conservative assumption is used, this actually tells you that uncertainty or randomness is
involved. Conservative assumptions are usually expressed in terms of safety factors. Sometimes
regulatory bodies demand safety factors in certain procedural codes. If you are not faced with such
restrictions or demands, then using conservative assumptions and safety factors can lead to
inefficient and costly over-design. You can avoid over-design by using probabilistic methods while
still ensuring the safety of the component.

Probabilistic methods even enable you to quantify the safety of the component by providing a
probability that the component will survive operating conditions. Quantifying a goal is the necessary
first step toward achieving it. Probabilistic methods can tell you how to achieve your goal.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 68 of 207

3.1.2. When Should Probabilistic Design be Used?

Probabilistic Design should be used when the issues are:

? The reliability of products or its components

? Customer satisfaction in terms of quality and robustness

Reliability is usually an issue because product or component failures have significant financial
consequences (costs of repair, replacement, warranty, or penalties), or even impact peoples lives and
health. As infinite safety and perfection is neither physically possible nor financially feasible,
probabilistic design helps you to design safe and reliable products while avoiding costly over-design
and conserve manufacturing resources (machining accuracy, efforts in quality control, etc).

Quality is the perception by a customer that the product performs as expected or better. In a quality
product, the customer rarely receives unexpected and unpleasant events where the product or one of
its components fails to perform as expected. By nature, those rare "failure" events are driven by
uncertainties in the design. Here, probabilistic design methods help you to assess how often "failure"
events may happen. This in turn enables you to improve the design for those cases where the
"failure" event rate is above your customers' tolerance limit.

Prev Next
Hints and Comments Basic Definitions
Basic Definitions
Prev Probabilistic Design Next

3.2. Basic Definitions

Before describing the procedure for probabilistic design, we will define some of the terminology.
First, let's start with a typical probabilistic design problem statement:

Find the scatter of the maximum deflection of a beam under a random load of snow. The snow
should have a linear distribution along the length of the beam with a height of H1 at one end and a
height of H2 at the other. The beam material is described by various parameters including the
Young's modulus, for which a mean value and standard deviation have been derived.

A Beam Under a Snow Load

Random Input Variables (RVs) are quantities that influence the result of an analysis. In probabilistic
literature, these random input variables are also called the "drivers" because they drive the result of
an analysis. For random input variables you must specify the type of the statistical distribution they
follow and the parameter values of their distribution functions. For the beam example the heights H1

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 69 of 207

and H2 and the Young's modulus E are clearly the random input variables. Naturally, the heights H1
and H2 cannot be negative and more often there will be little snow and only a few times there will be
a lot of snow. Therefore, it might be appropriate to model the height of the snow as an exponential or
a lognormal distribution, both of which have the bulk of the data at lower values.

If two (or more) random input variables are statistically dependent on each other, then you have a
correlation between those variables. In our beam example, it is very unlikely that there is a lot of
snow on one side of the roof (beam) while there is almost no snow on the other. It is not necessary
that H1 and H2 are exactly identical, but with a lot of snow then H1 and H2 probably both have larger
values and with little snowfall then both would be lower. This means that H1 and H2 are correlated,
although this correlation must not be mistaken with a direct mathematical dependency.
(Mathematical dependencies have some numerical dependence on each other; for example, when one
parameter doubles then another parameter also doubles.) In the example above, there is no numerical
dependency, but rather a certain trend between the two heights; with this particular H1 and H2 it is
simply very unlikely that their values will be drastically different at any given point in time.

Random Output Parameters (RPs) are the results of a finite element analysis. The RPs are typically a
function of the RVs; that is, changing the values of the random input variables should change the
value of the random output parameters. In our beam example, the maximum beam deflection is a
random output parameter.

The random input variables and random output parameters are collectively referred to as
probabilistic design variables. In the ANSYS Probabilistic Design System (PDS), you must identify
which parameters in the model are RVs and which are RPs. You can define up to a total of 5,000
RVs and RPs.

A sample is a unique set of parameter values that represents a particular model configuration. A
sample is characterized by random input variable values. In the example above, if you measure the
Young's modulus of a particular beam, and the snow height on a particular day at each end of that
beam, then those three values for E, H1 , and H2 would constitute one sample. You can think of a
sample as one virtual prototype. In this sense every component that is manufactured represents one
sample, because you can measure its particular properties (material, geometry, etc.) and you would
get specific values for each. However, sample also has a wider and more general use in discussing
statistics. For example, any single measured value of any physical property is considered one sample.
Because a probabilistic analysis is based on a statistical evaluation of the random output parameters,
the values of the random output parameters are also called samples.

The collection of all samples that are required or that you request for a certain probabilistic analysis
is called a simulation. The simulation contains the information used to determine how the component
would behave under real-life conditions (with all the existing uncertainties and scatter), and all
samples therefore represent the simulation of this behavior.

The analysis file is an ANSYS input file (which you can create in a variety of ways) that contains a
complete analysis sequence (preprocessing, solution, postprocessing). It must contain a
parametrically defined model using parameters to represent all inputs and outputs which will be used
as RVs and RPs. From this file, a probabilistic design loop file ( Jobname.LOOP) is automatically
created and used by the probabilistic design system to perform analysis loops.

A loop (or simulation loop) is one pass through the analysis file. In each loop the probabilistic design
system uses the values of the RVs from one sample and executes the user-specified analysis. The
PDS collects the values for the RPs following each loop.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 70 of 207

The probabilistic model is a term describing the combination of definitions and specifications for the
deterministic model (in form of the analysis file), the random input variables, correlations, random
output parameters, and the selected probabilistic method and its parameters. If you change any part
of the probabilistic model, then you will generate different results for the probabilistic analysis
(different results values and/or a different number of results). For example, if you modify the
analysis file, then the results may be affected. If you add or take away a random input variable or
change its distribution function, then you are solving a different probabilistic problem, which again
leads to different results. If you add a random output parameter, you will still solve the same
probabilistic problem, but more results are generated.

The probabilistic design database contains the current probabilistic design environment, which
includes

? Random input variables

? Correlations between random input variables

? Random output parameters

? Settings for probabilistic methods

? Which probabilistic analyses have been performed and in which files the results are stored

? Which output parameters of which probabilistic analyses have been used for a response
surface fit, the regression model that has been used for the fitting procedure, and the results of
that fitting procedure

This database can be saved (to Jobname.PDS) or resumed at any time. The results of a probabilistic
analysis are not stored in the database but in separate files. The samples generated for a fitted
response surface are neither in the database nor in files, because they can be re-generated very
quickly (files consume a lot of disk space, and reading these files would take as much time as simply
re-generating the sample data).

The mean value is a measure of location that is often used to describe the general location of the bulk
of the scattering data of a random output parameter or of a statistical distribution function.
Mathematically, the mean value is the arithmetic average of the data, while in literature the mean
value is often also called the expected value. The mean value also represents the center of gravity of
the data points.

The median value is the statistical point where 50% of the data is below the median value and the
50% is above. For symmetrical distribution functions (Gaussian, uniform, etc.) the median value and
the mean value are identical, while for non-symmetrical distributions they are different.

The standard deviation is a measure of variability (dispersion or spread) about the arithmetic mean
value; this is often used to describe the width of the scatter of a random output parameter or of a
statistical distribution function. The larger the standard deviation the wider the scatter and the more
likely it is that there are data values further apart from the mean value.

A solution set is the collection of results derived from the simulation loops performed for a particular
probabilistic analysis. It includes the values of all random input variables and all random output
parameters for all simulation loops of a probabilistic analysis. A unique label identifies each solution
set.

A response surface set is the collection of response surfaces derived from a fitting procedure

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 71 of 207

(regression analysis) plus the results derived from using the response surfaces for a probabilistic
analysis. A response surface set is identified by a unique response surface label.

A remote host is a computer in your local area network that is used to execute a probabilistic analysis
in parallel mode. A remote host can have more than one CPU; in parallel processing you can use
multiple CPUs on remote hosts.

Some of the above concepts can be better understood through an illustration. The following figure
(Probabilistic Design Data Flow) shows the flow of information during a probabilistic design
analysis. Note that the analysis file must exist as a separate entity, and that the probabilistic design
database is not part of the ANSYS model database.

Probabilistic Design Data Flow

Prev Next
Probabilistic Design How to Do Probabilistic Design
How to Do Probabilistic Design
Prev Probabilistic Design Next

3.3. How to Do Probabilistic Design

You can approach an ANSYS probabilistic design two ways: as a batch run or interactively through
the Graphical User Interface (GUI). The approach you take will depend on your ANSYS expertise
and your preference for interacting with the ANSYS program.

If you are very familiar with ANSYS commands, you may choose to perform the entire probabilistic
design analysis by creating an ANSYS command input file and submitting it as a batch job. This may
be a more efficient method for complex analyses (for example, nonlinear) that require extensive run-
time.

On the other hand, the interactive features of probabilistic design offer greater flexibility and

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 72 of 207

immediate feedback for review of loop results. When performing probabilistic design through the
GUI, it is important to first establish the analysis file for your model. Then all operations can be
performed interactively.

The usual procedure for probabilistic design consists of the following main steps. These steps may
vary slightly, depending on whether you are performing probabilistic design interactively (through
the GUI) or in batch mode. The items in parentheses indicate which ANSYS processor you need to
be in to perform the given task.

1. Create an analysis file to be used during looping. This file should represent a complete analysis
sequence and must do the following:

? Build the model parametrically (PREP7).

? Obtain the solution(s) (SOLUTION).

? Retrieve and assign to parameters the quantities that will be used as random input
variables and random output parameters (POST1/POST26).

2. Establish parameters in the ANSYS database which correspond to those used in the analysis
file. This step is typical, but not required (Begin or PDS); however, if you skip this step, then
the parameter names are not available for selection in interactive mode.

3. Enter PDS and specify the analysis file (PDS).

4. Declare random input variables (PDS).

5. Visualize random input variables (PDS). Optional.

6. Specify any correlations between the RVs (PDS).

7. Specify random output parameters (PDS).

8. Choose the probabilistic design tool or method (PDS).

9. Execute the loops required for the probabilistic design analysis (PDS).

10. Fit the Response Surfaces (if you did not use a Monte Carlo Simulation method) (PDS).

11. Review the results of the probabilistic analysis (PDS).

Details of the probabilistic design procedure are presented below. Differences in the procedure for a
"batch" versus "interactive" approach are indicated, where appropriate.

Because analyzing complex problems can be time consuming, ANSYS offers you the choice of
running a probabilistic analysis on a single machine, or distributing the analyses across multiple
machines. By using the ANSYS PDS parallel run capabilities, you can run many analysis loops
simultaneously and reduce the overall run time for a probabilistic analysis.

3.3.1. Create the Analysis File

The analysis file is crucial to ANSYS probabilistic design. The PDS uses the analysis file to form the
loop file, which is used to perform analysis loops. Any type of ANSYS analysis (structural, thermal,

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 73 of 207

magnetic, etc.; linear or nonlinear) may be incorporated in the analysis file.

In this file, the model must be defined in terms of parameters (RVs), and results data must be
retrieved in terms of parameters (RPs). Only numerical scalar parameters are used by the
probabilistic design system. You can use up to a total of 5000 parameters (RVs and RPs together).
See Use ANSYS Parameters in the ANSYS Modeling and Meshing Guide for a discussion of
parameters. See the ANSYS APDL Programmer's Guide for a discussion of the ANSYS Parametric
Design Language (APDL).

It is your responsibility to create the analysis file and make sure that it is correct and complete. It
must represent a clean analysis that will run from start to finish. Most nonessential commands (such
as those that perform graphic displays, listings, status requests, etc.) should be stripped off or
commented out of this file. Maintain only those display commands that you want to see during an
interactive session [EPLOT, etc.], or direct desired displays to a graphics file [/SHOW]. Keep in
mind that the analysis file will be used over and over again during probabilistic design looping. Any
commands that do not have direct bearing on the analysis will produce wasted action and therefore
decrease efficiency.

You can create an analysis file by inputting commands line by line with a system editor, or you can
create the analysis interactively through ANSYS and use the ANSYS command log as the basis for
the analysis file.

Both methods have advantages and disadvantages. Creating the file with a system editor is the same
as creating a batch input file for the analysis. (If you are performing the entire probabilistic design in
batch mode, the analysis file is usually the first portion of the complete batch input stream.) This
method allows you full control of parametric definitions through exact command inputs. It also
eliminates the need to clean out unnecessary commands later. However, if you are not moderately
familiar with ANSYS commands, this method might be inconvenient.

You might find it easier to perform the initial analysis interactively, and then use the resulting
command log as the basis for the analysis file. In this case, you will need to edit the log file to make
it suitable for probabilistic design looping. For more information on using the log files, see Using the
ANSYS Command Log in the ANSYS Operations Guide.

3.3.1.1. Sample Problem Description

To illustrate probabilistic design analysis, we will use the simple beam problem introduced earlier in
this chapter.

A Beam Under a Snow Load

We know the following about this problem:

? Young's modulus is 20E4

3.3.1.2. Build the Model Parametrically

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 74 of 207

PREP7 is used to build the model in terms of the RV parameters. For our beam example, the RV
parameters are H1 (snow height at left end), H2 (snow height at right end), and the Young's modulus
E.

...
! Initialize ANSYS parameters:
H1=100 ! Initialize snow height H1 @ left end (in mm)
H2=100 ! Initialize snow height H2 @ right end(in mm)
YOUNG=200.0e3 ! Initialize the Young's modulus (in N/mm**2)
ROOFWDT=1000.0 ! Initialize roof width left and right of beam (in mm)
BWDT=10.0 ! Initialize beam width (in mm)
BHGT=40.0 ! Initialize beam height (in mm)
BLEN=3000.0 ! Initialize beam length (in mm)
SNOWDENS = 200e-9 ! Density of snow (200 kg/m**3)
GRAVACC = 9.81 ! Earth gravity (in N/kg)
LOAD1 = H1*GRAVACC*ROOFWDT*SNOWDENS ! Pressure load due to snow @ left end
LOAD2 = H2*GRAVACC*ROOFWDT*SNOWDENS ! Pressure load due to snow @ right end
DELLOAD = LOAD2-LOAD1
!
! Material definitions:
MP,EX,1,YOUNG ! Young's modulus
MP,PRXY,1,0.3 ! Poisson's ratio
!
! Create the geometry (a line)
K,1,0,0,0 ! keypoint at left end
K,2,BLEN,0,0 ! keypoint at right end
L,1,2,100 ! line between keypoints
!
! Mesh definitions
ET,1,BEAM3 ! 2-D beam element
AREA=BWDT*BHGT ! Beam cross-sectional area
IZZ=(BWDT*(BHGT**3))/12 ! Moment of inertia about Z
R,1,AREA,IZZ,BHGT ! Real constants in terms of RV parameters
LATT,1,1,1
LMESH,1 ! mesh the line
FINISH ! Leave PREP7
...

As mentioned earlier, you can vary virtually any aspect of the design: dimensions, shape, material
property, support placement, applied loads, etc. The only requirement is that the design be defined in
terms of parameters. The RV parameters (H1, H2, and E in this example) may be initialized
anywhere, but are typically defined in PREP7.

Caution

If you build your model interactively (through the GUI), you will encounter many
situations where data can be input through graphical picking (such as when defining
geometric entities). Because some picking operations do not allow parametric input
(and PDS requires parametric input), you should avoid these picking operations.
Instead, use menu options that allow direct input of parameters.

3.3.1.3. Obtain the Solution

The SOLUTION processor is used to define the analysis type and analysis options, apply loads,
specify load step options, and initiate the finite element solution. All data required for the analysis
should be specified: master degrees of freedom in a reduced analysis, appropriate convergence
criteria for nonlinear analyses, frequency range for harmonic response analysis, and so on. Loads and
boundary conditions may also be RVs as illustrated for the beam example here.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 75 of 207

The SOLUTION input for our beam example could look like this:

...
/SOLU
ANTYPE,STATIC ! Static analysis (default)
D,1,UX,0,,,,UY ! UX=UY=0 at left end of the beam
D,2,UY,0,,,, ! UY=0 at rigth end of the beam
!D,2,UX,0,,,,UY ! UX=UY=0 at rigth end of the beam
elem=0
*get,numele,ELEM,,COUNT
*DO,i,1,numele
elem=elnext(elem) ! get number of next selected element
node1=NELEM(elem,1) ! get the node number at left end
node2=NELEM(elem,2) ! get the node number at right end
x1 = NX(node1) ! get the x-location of left node
x2 = NX(node2) ! get the x-location of rigth node
ratio1 = x1/BLEN
ratio2 = x2/BLEN
p1 = LOAD1 + ratio1*DELLOAD ! evaluate pressure at left node
p2 = LOAD1 + ratio2*DELLOAD ! evaluate pressure at left node
SFBEAM,elem,1,PRES,p1,p2 ! Transverse pressure varying linearly
! as load per unit length in negative Y
*ENDDO
SOLVE
FINISH ! Leave SOLUTION
...

This step is not limited to just one analysis. You can, for instance, obtain a thermal solution and then
obtain a stress solution (for thermal stress calculations).

If your solution uses the multiframe restart feature, all changes to the parameter set that are made
after the first load step will be lost in a multiframe restart. To ensure that the correct parameters are
used in a multiframe restart, you must explicitly save (PARSAV) and resume (PARESU) the
parameters for use in the restart. See the ANSYS Basic Analysis Guide for more information on
multiframe restarts.

3.3.1.4. Retrieve Results and Assign as Output Parameters

This is where you retrieve results data and assign them to random output parameters to be used for
the probabilistic portion of the analysis. Use the *GET command (Utility Menu>Parameters>Get
Scalar Data), which assigns ANSYS calculated values to parameters, to retrieve the data. POST1 is
typically used for this step, especially if the data are to be stored, summed, or otherwise manipulated.

In our beam example, the maximum deflection and the maximum stress of the beam are random
output parameters (RPs). The parameters for these data may be defined as follows:

...
/POST1
SET,FIRST
NSORT,U,Y ! Sorts nodes based on UY deflection
*GET,DMAX,SORT,,MIN ! Parameter DMAX = maximum deflection
!
! Derived data for line elements are accessed through ETABLE:
ETABLE,VOLU,VOLU ! VOLU = volume of each element
ETABLE,SMAX_I,NMISC,1 ! SMAX_I = max. stress at end I of each
! element
ETABLE,SMAX_J,NMISC,3 ! SMAX_J = max. stress at end J of each
! element
!
ESORT,ETAB,SMAX_I,,1 ! Sorts elements based on absolute value

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 76 of 207

! of SMAX_I
*GET,SMAXI,SORT,,MAX ! Parameter SMAXI = max. value of SMAX_I
ESORT,ETAB,SMAX_J,,1 ! Sorts elements based on absolute value
! of SMAX_J
*GET,SMAXJ,SORT,,MAX ! Parameter SMAXJ = max. value of SMAX_J
SMAX=SMAXI>SMAXJ ! Parameter SMAX = greater of SMAXI and
! SMAXJ, that is, SMAX is the max. stress
FINISH
...

See the *GET and ETABLE commands for more information.

3.3.1.5. Prepare the Analysis File

If you choose to create your model interactively in ANSYS, you must now derive the analysis file
from the interactive session. Use the command log or the session log file to do so. For more
information on using these log files, see Using the ANSYS Command Log in the ANSYS Operations
Guide.

Note

Do not use the /CLEAR command in your analysis file as this will delete the
probablilistic design database during looping. If this happens, the random input
variables are no longer recognized during looping and you will get the same
(deterministic) results for all simulation loops. However resume the ANSYS database
using the RESUME command as part of your analysis file. For example, this is
helpful if the variations of the random input variables do not require that meshing is
done in every loop (because the mesh is not effected). In this case you can mesh your
model, save the ANSYS database, and resume the database at the beginning of the
analysis file.

3.3.2. Establish Parameters for Probabilistic Design Analysis

At this point, having completed the analysis file, you are ready to begin probabilistic design analysis.
(You may have to reenter ANSYS if you edited the analysis file at the system level.) When
performing probabilistic design interactively, it is advantageous (but optional) to first establish the
parameters from the analysis file in the ANSYS database. (This step is not necessary for probabilistic
design performed in batch mode.)

To establish the parameters in the database:

? Resume the database file ( Jobname.DB) associated with the analysis file. This establishes your
entire model database in ANSYS, including the parameters. To resume the database file:
Command(s):

RESUME

GUI:

Utility Menu>File>Resume Jobname.db

Utility Menu>File>Resume from

? Read the analysis file into ANSYS to perform the complete analysis. This establishes your
entire model database in ANSYS, but might be time-consuming for a large model. To read the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 77 of 207

analysis file:
Command(s):

/INPUT

GUI:

Utility Menu>File>Read Input from

? Restore only the parameters from a previously saved parameter file; that is, read in a parameter
file that you saved using either the PARSAV command or the Utility Menu>Parameters>Save
Parameters menu path. To resume the parameters:
Command(s):

PARRES

GUI:

Utility Menu>Parameters>Restore Parameters

? Recreate the parameter definitions as they exist in the analysis file. Doing this requires that
you know which parameters were defined in the analysis file.
Command(s):

*SET or =

GUI:

Utility Menu>Parameters>Scalar Parameters

You may choose to do none of the above, and instead use the PDVAR command to define the
parameters that you declare as probabilistic design variables. See Declare Random Input Variables
for information on using PDVAR.

Note

The ANSYS database does not need to contain model information corresponding to
the analysis file to perform probabilistic design. The model input is automatically read
from the analysis file during probabilistic design looping.

3.3.3. Enter the PDS and Specify the Analysis File

The remaining steps are performed within the PDS processor.

Command(s):

/PDS

GUI:

Main Menu>Prob Design

In interactive mode, you must specify the analysis file name. This file is used to derive the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 78 of 207

probabilistic design loop file Jobname.LOOP. The default for the analysis file name is Jobname.pdan.
You can also specify a name for the analysis file:

Command(s):

PDANL

GUI:

Main Menu>Prob Design>Assign

For a probabilistic design run in batch mode, the analysis file is usually the first portion of the batch
input stream, from the first line down to the first occurrence of /PDS. In batch mode, the analysis file
name defaults to Jobname.BAT (a temporary file containing input copied from the batch input file).
Therefore, you normally do not need to specify an analysis filename in batch mode. However, if for
some reason you have separated the batch probabilistic design input into two files (one representing
the analysis and the other containing all probabilistic design operations), then you will need to
specify the analysis file using PDANL after entering the PDS [/PDS].

Note

In the analysis file, the /PREP7 and /PDS commands must occur as the first nonblank
characters on a line (that is, do not use the $ delimiter on a line containing either of
these commands). This is required for proper loop file construction.

Note

You cannot assign a different analysis file using the PDANL command after a
probabilistic analysis has been performed. This ensures the integrity of the previously
generated results with the specified probabilistic model.

Of course, ANSYS cannot restrain you from editing the analysis file or exchanging it with system
level commands. If you do so, then it is your responsibility to ensure the integrity of the generated
results with the definitions in the analysis file. If you are not sure that this integrity is maintained or
if you know that it is not, then we recommend that you save the current PDS database using the
PDSAVE command and then clear the probabilistic analysis results from the probabilistic design
database using the PDCLR, POST command. The PDCLR command does not delete the result files
that have been generated; it erases the link to the result files from the database.

In the example of a beam supporting a roof with a snow load you could store the analysis file in a
macro called "beam.mac". Here, the analysis is specified with the commands:

...
/PDS
PDANL,beam,mac
...

3.3.4. Declare Random Input Variables

The next step is to declare random input variables, that is, specify which parameters are RVs. As
mentioned earlier, the PDS allows for a combined total of 5000 random input variables and random
output parameters. To declare random input variables:

Command(s):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 79 of 207

PDVAR

GUI:

Main Menu>Prob Design>Random Input

If the parameter name that you specify with the PDVAR command is not an existing parameter, the
parameter is automatically defined in the ANSYS database with an initial value of zero.

For random input variables you must specify the type of statistical distribution function used to
describe its randomness as well as the parameters of the distribution function. For the distribution
type, you can select one of the following:

? Gaussian (Normal) (GAUS):

You provide values for the mean value and the standard deviation of the random variable
x.

? Truncated Gaussian (TGAU):

You provide the mean value and the standard deviation of the non-truncated Gaussian

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 80 of 207

distribution and the truncation limits xmin and xmax. F(…) denotes the cumulative distribution
function of the non-truncated Gaussian distribution.

? Lognormal option 1 (LOG1):

You provide values for the mean value and the standard deviation of the random variable
x. The PDS calculates the logarithmic mean and the logarithmic deviation :

? Lognormal option 2 (LOG2):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 81 of 207

You provide values for the logarithmic mean value and the logarithmic deviation . The
parameters and are the mean value and standard deviation of ln(x):

? Triangular (TRIA):

You provide the minimum value xmin, the most likely value limit xmlv and the maximum value
xmax.

? Uniform (UNIF):

You provide the lower and the upper limit xmin and xmax of the random variable x.

? Exponential (EXPO):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 82 of 207

You provide the decay parameter and the shift (or lower limit) xmin of the random variable
x.

? Beta (BETA):

You provide the shape parameters r and t and the lower and the upper limit xmin and xmax of
the random variable x. B(…) is the Betafunction.

? Gamma (GAMM):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 83 of 207

You provide the decay parameter and the power parameter k. (…) is the Gammafunction.

? Weibull (Type III smallest) (WEIB):

You provide the Weibull characteristic value xchr , the Weibull exponent m and the minimum
value xmin. Special cases: For xmin = 0 the distribution coincides with a two-parameter Weibull
distribution. The Rayleigh distribution is a special case of the Weibull distribution with =
xchr - xmin and m = 2.

You may change the specification of a previously declared random input variable by redefining it.
You may also delete a probabilistic design variable [PDVAR,Name,DEL]. The delete option does not
delete the parameter from the database; it simply deactivates the parameter as a probabilistic design
variable.

Note

Changing the probabilistic model by changing a random input variable is not allowed
after a probabilistic analysis has been performed. This ensures the integrity of the
previously generated results with the specified probabilistic model. If you need to
change one or more random input variables (for example, because you learned that

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 84 of 207

some specifications were incorrect after running an analysis), then we recommend

that you save the current PDS database (using the PDSAVE command) and then clear
the probabilistic analysis results from the probabilistic design database (using the
PDCLR,POST command). The PDCLR command does not delete the result files that
have been generated, it simply removes the link to the results file from the database.

In the example of a beam supporting a roof with a snow load, you could measure the snow height on
both ends of the beam 30 different times. Suppose the histograms from these measurements look like
the figures given below.

Histograms for the Snow Height H1 and H2

From these histograms you can conclude that an exponential distribution is suitable to describe the
scatter of the snow height data for H1 and H2. Suppose from the measured data we can evaluate that
the average snow height of H1 is 100mm and the average snow height of H2 is 200mm. The
parameter can be directly derived by 1.0 divided by the mean value which leads to 1 = 1/100 =
0.01 for H1, and 1 = 1/200 = 0.005 for H2. From measurements of the Young's modulus you see
that the Young's modulus follows a Gaussian distribution with a standard deviation of 5%. Given a
mean value of 200,000N/mm2 for the Young's modulus this gives a standard deviation of
10,000N/mm 2 . These definitions can be specified using the following commands:

...
PDVAR,H1,EXPO,0.01
PDVAR,H2,EXPO,0.005

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 85 of 207

PDVAR,YOUNG,GAUS,200000,10000
...

3.3.5. Visualize Random Input Variables

After you define your random input variables, you should use ANSYS' visual tools to verify them.
You can plot individual RVs, and you can obtain specific information about them through an inquiry.

Command(s):

PDPLOT

PDINQR

GUI:

Main Menu>Prob Design>Plot

Main Menu>Prob Design>Inquire

The PDPLOT command plots the probability density function as well as the cumulative distribution
function of a defined random input variable. This allows you to visually check that your definitions
are correct.

Use the PDINQR command to inquire about specific information for a defined random input
variable by retrieving statistical properties or probing the two function curves that are plotted with
the PDPLOT command. For example you can inquire about the mean value of a random input
variable or evaluate at which value the cumulative distribution function reaches a certain probability.
The result of this inquiry is stored in a scalar ANSYS parameter.

3.3.6. Specify Correlations Between Random Variables

In a probabilistic analysis, random input variables can have specific relationships to each other,
called correlations. If two (or more) random input variables are statistically dependent on each other,
then there is a correlation between those variables.

To define these correlations:

Command(s):

PDCORR

GUI:

Main Menu>Prob Design>Correlations

You specify the two random input variables for which you want to specify a correlation, and the
correlation coefficient (between -1 and 1). To remove a correlation, enter DEL in the correlation
coefficient field (PDCORR ,Name1,Name2,DEL)

In the example of a beam supporting a roof with a snow load, the data for the snow height indicates
that there is a correlation between the snow height at one end versus the snow height at the other end.
This is due to the fact that it is unlikely that one end of the beam has very little snow (or no snow at

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 86 of 207

all) at the same time that the other end carries a huge amount of snow. In the average the two snow
heights tend to be similar. This correlation is obvious if you plot the measured data values for H2
versus the data value for H1. This scatter plot looks like this:

A Scatter Plot of Snow Height H1 vs. H2

Performing a statistical evaluation of the data, we can conclude that the linear correlation coefficient
between the values of H1 and H2 is about 0.8. You can define this correlation using the commands:

...
PDVAR,H1,EXPO,0.01
PDVAR,H2,EXPO,0.005
PDCORR,H1,H2,0.8
...

You may have a more complex correlation where you have a spatial dependency. If so, you can use
the PDCFLD command to calculate a correlation field and store it into an ANSYS array.

Random fields are random effects with a spatial distribution; the value of a random field not only
varies from simulation to simulation at any given location, but also from location to location. The
correlation field describes the correlation coefficient between two different spatial locations. Random
fields can be either based on element properties (typically material) or nodal properties (typically
surface shape defined by nodal coordinates). Hence, random fields are either associated with the
selected nodes or the selected elements. If a random field is associated with elements, then the
correlation coefficients of the random field are calculated based on the distance of the element
centroids.

Note that for correlation fields, the “domain distance” D(xi , xj) is not the spatial distance |xi - xj|, but
the length of a path between xi and xj that always remains inside the finite element domain. However,
exceptions are possible in extreme meshing cases. For elements that share at least one node, the
PDCFLD evaluates the distance by directly connecting the element centroids with a straight line. If
these neighboring elements form a sharp inward corner then it is possible that the “domain distance”
path lies partly outside the finite element domain, as illustrated below.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 87 of 207

After the correlation coefficients have been calculated and stored in an ANSYS parameter
(PDCFLD,ParR), then you can use the PDCORR command to define the correlations between the
elements of the random field.

Example

The structure illustrated below is modeled with 12 elements. We will evaluate the domain distances
of the element centroids.

First, build the finite element structure:

…
/PREP7
et,1,shell63
! create the nodes
N,1,0,4
N,2,1,4
N,3,2,4
N,4,3,4
N,5,4,4
N,6,0,3
N,7,1,3
N,8,2,3
N,9,3,3
N,10,4,3
N,11,1,2
N,12,2,2
N,13,3,2
N,14,4,2
N,15,0,1
N,16,1,1
N,17,2,1
N,18,3,1
N,19,4,1
N,20,0,0
N,21,1,0
N,22,2,0
N,23,3,0
N,24,4,0
! create the elements
E,1,2,7,6
E,2,3,8,7
E,3,4,9,8
E,4,5,10,9
E,7,8,12,11
E,9,10,14,13
E,11,12,17,16
E,13,14,19,18

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 88 of 207

E,15,16,21,20
E,16,17,22,21
E,17,18,23,22
E,18,19,24,23
…

Next, calculate the domain distances and store the results in the array “elemdist”:

…
/PDS
PDCFLD,elemdist,ELEM,DIST
…

Finally, get all the element domain distances and print them:

…
*GET,numsel,ELEM,0,COUNT ! Get the number of selected elements
!
! Outer loop through all selected elements from first to last
index=0
elem1=0
! Pipe output to file
/OUT,elements,dat
*DO,i,1,numsel
elem1=ELNEXT(elem1) ! get number of next selected element
*IF,elem1,EQ,0,CYCLE ! Leave do loop if no more elements
!
! Inner loop through selected elements from "elem1+1" to last
elem2=elem1
*DO,j,i+1,numsel
elem2=ELNEXT(elem2) ! get number of next selected element
*IF,elem2,EQ,0,CYCLE ! Leave do loop if no more elements
index=index+1
!
! Print out the element distance
*MSG,INFO,elem1,elem2,elemdist(index)
Distance between element %i and %i is %g
*ENDDO ! goto next element for inner loop
*ENDDO ! goto next element for outer loop
…

The print out will show that for the structure illustrated above the "domain distance" between the
element centroids of elements 1 and 9 is 3.8284 and between the element centroids of elements 1 and
12 it is 4.8284. The paths related to these distances are sketched in the illustration with a solid line
and a dashed line respectively. In this example there are 12 elements, thus the array "elemdist" has a
length of 12*(12-1) = 66.

3.3.7. Specify Random Output Parameters

After declaring your input variables and correlations among them you must define the random output
parameters. The random output parameters (RPs) are results parameters that you are interested in. To
define random output parameters:

Command(s):

PDVAR,Name,RESP

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 89 of 207

Main Menu>Prob Design>Random Output

3.3.8. Choose a Probabilistic Design Method

In the probabilistic design system, several different probabilistic design methods are available. You
can choose one of two primary methods, the Monte Carlo Simulation (default) or the Response
Surface Method. Options under the Monte Carlo Simulation method include the Latin Hypercube
Sampling method (default) and the Direct Monte Carlo Sampling method. Options under the
Response Surface Method include the Central Composite Design and the Box-Behnken Matrix
Design method. Both the Monte Carlo Simulation and the Response Surface Methods allow a user-
defined option. See the PDDMCS, PDLHS, PDCCD, and the PDBBM commands or Probabilistic
Design Techniques for further details about these methods.

To specify a method to be used for probabilistic design looping:

Command(s):

PDMETH

GUI:

Main Menu>Prob Design>Monte Carlo Sims

Main Menu>Prob Design>Response Surface

Note

To use Response Surface Methods, the random output parameters must be smooth and
continuous functions of the involved random input variables. Do not use Response
Surface Methods if this condition is not satisfied.

3.3.8.1. Probabilistic Method Determination Wizard

You can use the Probabilistic Method Determination Wizard to find the fastest method for solving
your probabilistic analysis. You should have completed one analysis prior to starting the wizard.

Use Main Menu>Prob Design>-Prob Method>Methods Wizard to access the wizard. Answer the
questions on the screens as they are presented. Before you start the wizard, you should know:

? How long did it take to run the analysis file (hours, minutes, seconds)? Or, you should give an
estimation of the time it will take if you haven't run the analysis yet.

? How many CPUs are available to run the probabilistic analysis (if running parallel
processing)?

? Is your output parameter a smooth/continuous function of the input parameters?

? What results are you interested in evaluating? Options are mean values, standard deviation,
sensitivities, or acceptable part failures.

Below is one of the wizard screens, as an example.

The PDS Methods Wizard

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 90 of 207

Based on the information you provide, the wizard will tell you the fastest method for solving your
probabilistic design problem. The wizard will issue the PDMETH command and one of the PDLHS ,
PDCCD, or PDBBM commands. You will still need to run the analysis, then fit the results to a
response surface, etc. to evaluate your results.

3.3.9. Executing Probabilistic Analysis Loops

You can execute your analysis on your machine alone (serial execution), or using other computers in
your network to save running time and speed processing (parallel execution). If you want to use
serial processing only, choose the serial option from the Run menu. If you want to run parallel
analyses on multiple CPUs, you must first set the parallel options before performing the analysis (see
PDS Parallel Analysis Runs for more details).

Command(s):

PDEXE

GUI:

Main Menu>Prob Design>Run>-Exec Serial- Run Serial

Main Menu>Prob Design>Run>-Exec Parallel- Run Parallel

Caution

For security reasons ANSYS strongly recommends that you use parallel processing
only within the firewall of your local area network.

If you choose serial processing to perform a probabilistic analysis then you will utilize only the CPU
of the computer you are working on. If you have access to only one license of ANSYS or if you have
access to only one computer, then this is the only way in which you can run a probabilistic analysis.
While the simulation loops are running in serial mode, your ANSYS session is locked (you cannot
perform other tasks in the same ANSYS session). If you are running your ANSYS session in
interactive mode then the simulation loops are also performed in interactive mode. If you are running
your ANSYS session in batch mode then the simulation loops are performed in batch mode.

If you choose the PDS parallel processing option, you can use other CPUs that you have access to for
running the probabilistic analysis. PDS parallel processing can distribute the necessary jobs in a local
area network. With this option, the simulation loops are sent to CPUs that you can specify, where
they are executed in "server mode". This looks the same as a batch run (in other words, there is no
interactive visualization during the execution of a simulation loop). While the simulation loops are
running in parallel mode, your ANSYS session is locked; however, you can instruct the ANSYS
session to start post-processing the probabilistic results as they are calculated so you can review and
visualize the results before all simultaion loops are finished.

In parallel processing, you can monitor the running jobs and the completed results.

3.3.9.1. Probabilistic Design Looping

Regardless of whether you choose serial or parallel processing, ANSYS controls the analysis
looping; you cannot modify this process. It does the following:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 91 of 207

? Always reads the analysis file from the beginning.

? Always ignores the RV settings and replaces their value with the derived value that the
probabilistic method has assigned to an RV for a particular loop.

? Always reads the PDS database after each loop.

For the execution of the simulation loops you must specify a solution label ( Slab on the PDEXE
command). The solution label is used for several purposes:

? The results are stored in an ASCII readable file under the name "jobname_Slab.pdrs". Here,
Slab is the user specified solution label.

? If you want to fit response surfaces to the results of the random output parameters then you
need to specify the solution label to indicate which set of results you want to use for the fitting
procedure.

? If you want to post-process the results generated with the PDEXE command then you must
specify the solution label to indicate which results you want to post-process.

When you execute the probabilistic analysis (PDEXE), ANSYS creates a probabilistic design loop
file ( Jobname.LOOP) from the analysis file. This loop file is used to perform analysis loops. Looping
continues until all parameters have been evaluated.

If a loop is interrupted due to an error in the execution run (for example, a meshing failure, a non-
converged nonlinear solution, etc.), ANSYS PDS aborts that loop. Further processing depends if you
are in serial or parallel processing mode. If you are using:

? Serial interactive processing: you can choose to terminate when you receive the error, or
continue processing.

? Serial batch processing: processing terminates at the first error.

? Parallel processing: processing terminates if the allowed number of failed loops is exceeded
(set in PDEXE), otherwise it continues.

Note that for all failed loops (loops with errors), the results for that loop are discarded, no data from
that loop is used for post processing.

After the PDEXE command is issued, the PDS generates a file containing the input sample values.
The file is called jobname.samp. An example of the content of the file is given below:

TEST1
ITER CYCL LOOP X1 X2 X3
1 1 1 1.619379209e+000 2.364528435e-001 1.470789050e+000
1 1 2 2.237676559e-001 5.788049712e-001 1.821263115e+000
1 1 3 7.931615474e+000 8.278689033e-001 2.170793522e+000
.. .. .. ... ... ...
.. .. .. ... ... ...

The first line contains the solution label (the parameter Slab is set via the PDEXE command); the
second line contains the headers of the data columns - the iteration number, cycle number, loop
number, and the random variable names. The iteration number and cycle number tell the PDS to
which group (with specific PDS method and settings) the loops belong. Subsequent lines provide

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 92 of 207

specific iteration, cycle, loop, and input sample values for the defined input variables.

The PDS also creates a file where the results are stored. The name of the results file is
jobname_Slab.pdrs. Before the job is executed, the file looks like this:

TEST1
ITER CYCL LOOP ERR X1 X2 X3 RESULT

In the first line, the PDS enters the solution label. In the second line are the headers for the data
columns: the first four columns are the iteration, cycle number, loop number, and an error flag. The
fifth and subsequent columns are for the random input variable and random output parameter values.
If you run a subsequent analysis (same type of analysis with the same solution label), the cycle is
incremented, the loop count is reset to 1, and the result file is appended.

For example, the content of the result file could look like this:

TEST1
ITER CYCL LOOP ERR X1 X2 X3 RES
1 1 1 0 1.619379209e+000 2.364528435e-001 1.470789050e+000 4.162928057e+
1 1 2 0 2.237676559e-001 5.788049712e-001 1.821263115e+000 4.744249212e+
1 1 3 0 7.931615474e+000 8.278689033e-001 2.170793522e+000 1.149997825e+
.. .. .. .. ... ... ...
.. .. .. .. ... ... ...

3.3.9.2. Serial Analysis Runs

Serial execution of the probabilistic analysis is useful for smaller models, or if you have access to
only one machine.

Command(s):

PDEXE

GUI:

Main Menu>Prob Design>Run>-Exec Serial- Run Serial

3.3.9.3. PDS Parallel Analysis Runs

To save time when analyzing large models with PDS, use the parallel processing option available
under the PDEXE command, if you have multiple CPUs available.

To successfully execute a probabilistic analysis in parallel mode you need to follow three steps:

1. Define the remote hosts that you want to use for parallel processing. See Hosts and Product Set
Up for details. You must have account privileges on all remote hosts.

2. Specify the files that must be copied to the remote hosts. See Files Needed for Parallel Run for
details.

3. Start the parallel execution.

Caution

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 93 of 207

For security reasons we strongly recommend that you use parallel processing only
within the firewall of your local area network.

Each of these steps is explained in detail in the following sections.

3.3.9.3.1. Hosts and Product Set Up

For parallel processing you need to specify the remote hosts you want to use. This information is
placed in a file called hosts60.ans. You can create this file using a text editor or you can use the
ANS_ADMIN utility (see the discussion on the ANS_ADMIN utility in the ANSYS Installation and
Configuration Guide for your platform for more information). This file contains host information for
all remote machines on which you may want to run. This file is global and does not contain job-
specific information. If you have multiple users running PDS at your site, you should have one
hosts60.ans file for all users. ANSYS searches for this file first in the local directory, followed by
the home directory, and finally the docu directory.

A sample hosts60.ans file looks like this:

# This file is used to specify those hosts that the ANSYS Nanny may
# run children on.
#
# Each host entry is to be on its own line. The host entry consists of
# several fields which are space delimited.
#
# Field 1 - host IP address or name
# Field 2 - host machine type
# Field 3 - execution key:
# 0-Use a remote shell to start the child process;
# this requires a remote shell server to be
# running on the host machine.
# >1024-Use a running ANSYS thin server on the host
# which is listening on this port number.
# Field 4 - The default maximum number of jobs to run on this host
# Field 5 - The time in minutes to check again if the host is available.
# If this is zero then the host will not be checked again.
# Field 6 - The local port number to start the communication with the
# Ansys Thin Server on. This is tied to authentication on the
# Ansys Thin Server.
# Field 7 - The directory to create the children subdirectories in
# Field 8 - The cluster type. Only valid entry is MPI.
# Field 9 - The speed factor (relative speed to other machines listed).
Only valid entry is 1.
# Field 10 - Number of OpenMP threads. Only valid entry is 1.
# Example:
#
# UNIX box that has five processers
# zeus sgi64 0 5 30 2000 /scratch/wjc
# Microsoft box using the ANSYS Thin Server
# wjcpc NT 2010 1 0 2000 C:\TEMP
alpha1 alpha 0 1 15 2000 /scratch/epc MPI 1 1
athena sgi64 0 1 15 2000 /scratch/epc MPI 1 1
rs43p rs6000 0 1 15 2000 /home/pdstest MPI 1 1
rs260 rs64 0 1 15 2000 /home/pdstest MPI 1 1
snoopy hppa8000 0 1 15 2000 /home/pdstest MPI 1 1
alpha24 alpha 0 1 15 2000 /home/pdstest MPI 1 1
hp770 hppa8000 0 1 15 2000 /home/pdstest MPI 1 1
us60 usparc 0 1 15 2000 /home/pdstest MPI 1 1
ss60 sun64 0 1 15 2000 /home/pdstest MPI 1 1

You can designate any machine that has a licensed copy of ANSYS installed as a remote host

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 94 of 207

(slave); for NT machines, you will need to install your own rsh server or invoke the ANSYS Thin
Server manually (ansysts). ANSYS suggests specifying port number 50057 when you use ansysts.

It is good practice to follow these guidelines when selecting remote hosts:

? Use idling computer power. During regular work hours, computers usually have designated
"owners" using the machines for interactive work. Using these computers for parallel
processing can slow down the machine performance and possibly disturb your colleagues in
their work. However, computers are usually idle at night and can then be used without
disturbing others.

? Use machines that are dedicated as computer servers. There might be some computers in your
work environment which are not used for interactive work at all. It is fairly safe to use these
computers as remote hosts for parallel-processing at all times.

? Check with your System Administrator for additional guidelines.

When this file is in place, you can proceed with the parallel execution either interactively or via
command or batch mode:

Command(s):

PDEXE

GUI:

Main Menu>Prob Design>Run>-Exec Parallel- Host Select

If you are working interactively, this selection defines information that ANSYS uses to create a
Jobname.hosts file. If you are working in batch mode, you must create this file using a text editor.
This file must reside in the directory where you are running ANSYS PDS. This file must include the
following information:

? Remote hosts to be used for this particular parallel run. You must have a user account on each
remote host that you specify, and ANSYS must be installed and accessible on (in the path for)
that machine.

? Number of processes that can run in parallel on each host. If a remote host has more than one
CPU, you can use all of the CPUs on the remote host. For performance reasons, we
recommend leaving one CPU for system tasks and using only N-1 CPUs for parallel
processing (if N is the number of CPUs on the remote host).

? Directories in which you want the remote jobs to be executed. We recommend that you use
temporary directories like "/scratch" or "/tmp" on UNIX or "C:\TEMP" on PCs. These remote
directories are cleaned up automatically after the parallel jobs finish running.

A sample jobname.hosts file looks like this:

# This file is used to specify those hosts that the ANSYS Parent may
# run children processes on.
#
# Each host entry is to be on its own line. The host entry consists of
# several fields which are space delimited.
#
# Field 1 - host IP address or name

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 95 of 207

# Field 2 - username to use for a remote shell to the host

# Field 3 - execution key:
# 0 - Use a remote shell to start the child process
# >1024 - Use a running ANSYS thin server on the host
# which is communicating on this port number
# Field 4 - the product code sequence to attempt to start ansys jobs
# on the host with. This is a colon delimited list.
# Field 5 - the number of jobs to run on this host
# Field 6 - The time in minutes to check again if the host is available.
# If this is zero then the host will not be checked again.
# Field 7 - If field 3 is non-zero then this is the local port number
# or range to start the communication with the Ansys Thin
# Server on. This is tied to the "authentication" on the
# Ansys Thin Server.
# If field 3 is zero then this should be set to zero.
# Field 8 - directory to create the children subdirectories in
#
# Example:
#
# UNIX box that has five processers and will first attempt to
# run with ANSYS/Mechanical Batch Child and then ANSYS/Mechanical
# zeus wjc 0 MEBACH:ANSYS 5 30 2000 /scratch/wjc
# NT box running the thin server on port 2010
# wjcpc wjc 2010 ANSYS 1 0 2000 C:\TEMP
alpha1 epc 0 MEBA:MEBACH 1 15 2000 /scratch/epc
athena epc 0 MEBA:MEBACH 1 15 2000 /scratch/epc
rs43p pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest
rs260 pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest
snoopy pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest
hp160 pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest
alpha24 pdstest 0 MEBA:MEBACH 2 15 2000 /home/pdstest
hp770 pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest
us60 pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest
ss60 pdstest 0 MEBA:MEBACH 1 15 2000 /home/pdstest

3.3.9.3.2. Files Needed for Parallel Run

You also need to specify all other files that will be used in the distributed processing run. These files
are listed in a separate file called Jobname.ifiles. At a minimum, this file must include the analysis
file. Additionally, you may need other files, such as data files or ANSYS macros. You must include
all files that the analysis files calls in order for the PDS run to execute successfully. These files,
which are typically found only on your local machine, will then be copied to all remote hosts during
the PDS run. The Jobname.ifiles must reside in your working directory. If you are working in
batch mode, you need to create this file manually. The first file listed must be the analysis file. Other
files listed must include the full path.

Command(s):

PDANL

GUI:

Main Menu>Prob Design>Assign

A sample jobname.ifiles file looks like this:

# Files to copy to the ansys server for job Jobname

# created on Fri Oct 13 10:33:44 EDT 2000.
pdsrun1.inp
/home/staff/epc/ddts/pds/tags

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 96 of 207

3.3.9.3.3. Controlling Server Processes

If you are working interactively, you need to complete one additional step. When you select Run
Parallel, you see a dialog box listing all remote hosts specified in the Jobname.hosts file, as well as
their current status. You can use this dialog box to monitor the processes and remote machines, and
stop or start processes on those machines as necessary. The status codes you could see include:

? DOWN - The remote host is not available (reason unknown).

? DN-TS - The Thin Server is down.

? DN-DRF - ANSYS was unable to create the necesary directories on that host.

? DN-ANS - ANSYS would not start on that host.

? DN-ATH - Authentication failure (the batch child product was unable to authenticate with the
parent product).

? UP - The Thin Server is up.

? RUN - The ANSYS job is running on that host.

If you want to check diagnositcs for a staus code in parallel processing mode, choose the Diagnostics
button in the Server Process Control dialog. The diagnostics system will show you the status of the
slave machine(s), and details to help you address any issues that arise. Suggested actions are
presented.

Note

In batch mode, ANSYS will always attempt to start remote processing on all
machines listed in the Jobnames.hosts file.

3.3.9.3.4. Initiate Parallel Run

You can now activate the parallel execution:

Command(s):

PDEXE

GUI:

Main Menu>Prob Design>Run>-Exec Parallel- Run Parallel

Several things happen when you initiate a parallel processing run.

1. The networked machines are initialized (ANSYS checks for machine permissions, licenses,
and directory permissions), and any necessary directories are created.

2. The relevant files are copied to the networked machines.

3. ANSYS is launched in server mode on each networked machine.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 97 of 207

Jobs are sent to each machine as that machine finishes a previous job; faster machines will naturally
process more jobs. If a slave machine (for whatever reason) does not complete a job, that job is
automatically sent to another machine to be processed.

When the analyses are finished (either completed or stopped manually), then the PDS removes any
files or directories it created and stops any processes it started.

3.3.10. Fitting and Using Response Surfaces

After you have executed a probabilistic analysis you can use the results that are stored in the result
files to fit responses surfaces. If the probabilistic analysis was based on the Monte Carlo Simulation
method this step is optional and you can go directly to the results post-processing. If the probabilistic
analysis was based on the Response Surface Method, this step is mandatory. The random output
parameter values generated using the Response Surface Method are meant to be fitted with a
response surface. Therefore, the Response Surface Method determines the values of the random
input variables such that fitting of a response surface is most efficient (that is, so that it uses the
fewest sampling points). If you use Monte Carlo Simulation results to derive response surfaces, then
the sampling points are not as efficiently located for the fitting process; you should accomodate this
by using more sample points.

? You should use at least 20% more Monte Carlo Simulation points than what would be required
for a Response Surface Method for the same problem. For a list of the number of sampling
points required for a Response Surface Method please see Probabilistic Design Techniques.

? If you cannot determine how many sampling points a Response Surface Method needs (for
example, because there are too many random input parameters), then you should have at least
two times more sampling points than the number of coefficients of the response surfaces.

3.3.10.1. About Response Surface Sets

The results generated by the response surface fitting procedure as well as the results generated by
applying the approximation equation (Monte Carlo Simulation) are combined in a response surface
set. Each response surface set includes the following information:

? A unique name that you provide. This name is used to identify the response surface set during
probabilistic post-processing.

? The name of the solution set you used for the fitting procedure (containing the data points).

? The number and the names of the random output parameters for which a response surface has
been evaluated. If you have many random output parameters you might not be interested in
fitting a response surface for every one, but only for those that are most important.

? For each random output parameter that was fitted with a response surface, the response surface
set includes information about the regression model that was used to fit the response surface
(linear, quadratic, or quadratic with cross-terms), as well as the terms and coefficients that
were derived as result of the regression analysis.

? The Monte Carlo Simulation samples created using the response surface equations.

There is a one-to-one relationship between solution sets and response surface sets. For each solution
set containing sample points you can have only one response surface set containing the response
surfaces fitting these sample points. The reverse is also true, that each response surface set can only
contain the response surfaces that are based on the sample points of one solution set.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 98 of 207

3.3.10.2. Fitting a Response Surface

To fit a response surface you must specify a name for the response surface set where you want the
results to be stored, the solution set you want to use, one random output parameter, the regression
model, an output transformation technique (if any), and whether to filter terms.

The regression model describes which regression terms are used for the approximation function of
the response surface. In the ANSYS PDS, the following regression models are implemented:

? Linear approximation function

? Quadratic approximation function without cross-terms

? Quadratic approximation function including cross-terms

While you can use all terms included in the regression model, the ANSYS PDS also offers an option
that automatically filters out insignificant terms. This technique is called the forward-stepwise
regression analysis. For example, where the Young's modulus E and the thermal expansion
coefficient are random input variables, a full quadratic regression model reads:

A full regression model uses the available sampling points to determine values for all regression
coefficients c0 to c3 . Of course the values for c0 to c2 will be zero or very close to zero; taking more
coefficients into account than really necessary reduces the degrees of freedom of the algebraic
equation to be solved to evaluate the coefficients. This in turn reduces the accuracy of the
coefficients that are important for the regression fit. The forward-stepwise regression analysis takes
this into account and automatically eliminates terms that are not needed.

The ANSYS PDS offers a variety of transformation functions that can be used to make the random
response parameter to be more appropriately described by a quadratic function after the
transformation has been applied. These transformation functions include:

? Exponential transformation = exp(Y i)

? Logarithmic transformations = ln(Y i); = log 10 (Y i); = (Yi)

? Square root transformation =

? Power transformation =

? Box-Cox transformation

= for 0

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 99 of 207

= ln(Y i) for =0

Here, Yi is the value of a random output parameter obtained in the i-th sampling loop and is the
corresponding transformed value. The physical nature of the problem should indicate which
transformation to use; for example, lifetime parameters (such as the number of cycles until low cycle
fatigue occurs) are usually transformed with a logarithmic transformation. If you do not have this
kind of information, then you should start with the Box-Cox transformation. The PDS automatically
searches for an optimum value for the Box-Cox parameter within the interval [-2,2]. As
guidelines:

? If is close to -1.0 then the data is best transformed by a reciprocal transformation, which is a
power transformation with an exponent of -1.0.

? If is close to zero then the data is best transformed by a logarithmic transformation.

? If is close to 0.5 then use the square root transformation.

? If is close to 1.0, then no transformation should be used.

? If is not close to any of these specific values then the Box-Cox transformation is
appropriate.

To fit a response surface:

Command(s):

RSFIT

GUI:

Main Menu>Prob Design>Fit Resp Surf

3.3.10.3. Plotting a Response Surface

Whether a response surface is a good representation of the sampling point that it is supposed to fit
can be best illustrated by plotting the response surface. The ANSYS PDS plots the sampling points
as symbols and the response surface as a contour plot so you can visually compare them. However,
you can only plot one random output parameter as a function of two random input variables at a
time.

To plot a response surface:

Command(s):

RSPLOT

GUI:

Main Menu>Prob Design>Plot Resp Surf

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 100 of 207

3.3.10.4. Printing a Response Surface

After using a regression analysis to fit a response surface, ANSYS automatically prints all necessary
results in the output window:

? The transformation function that has been used or has been determined automatically (in case
of Box-Cox transformation)

? Regression terms

? Regression coefficients

? Goodness-of-fit measures

The goodness-of-fit measures provide a means to verify the quality of the response surface and
whether it is a good representation of the underlying data (in other words, the sample points).

You can request a print out of this data at any time.

Command(s):

RSPRNT

GUI:

Main Menu>Prob Design>Prnt Resp Surf

3.3.10.5. Generating Monte Carlo Simulation Samples on the Response Surfaces

After you have generated a response surface set that includes one or more response surfaces for one
or more random output parameters then you also need to perform Monte Carlo Simulations using
these response surfaces to generate probabilistic results. This is where the PDS generates sampling
values for the random input variables in the same way it did for the simulation looping performed
using your analysis file. But instead of using the random input variable values in the analysis file and
running through the analysis file, it uses the approximation function derived for the response surfaces
to calculate approximated response values. The process of calculating an explicitly known equation
is much faster than running through the analysis file and performing a finite element analysis, so you
can run a large number of simulation loops in a relatively short time. Usually, several thousand
simulation loops are performed if you utilize the response surfaces.

After you have generated the Monte Carlo Simulation loops on the response surfaces, you can begin
probabilistic post-processing and review the probabilistic results the same way as you would for
Monte Carlo Simulations. However, there is one difference for post-processing between Monte Carlo
results and Monte Carlo results derived from response surface approximations. For Monte Carlo
simulation results, the accuracy of the results is determined by the number of simulation loops that
are performed. The PDS can visualize the accuracy of Monte Carlo results by means of confidence
limits or confidence bounds. For Monte Carlo results derived from response surface approximations,
the confidence bounds are suppressed. This is necessary because the accuracy is not determined by
the number of simulation loops (as mentioned above, you typically perform a large number of these)
but by the goodness-of-fit or the response surface model itself. With increasing numbers of
simulation loops the confidence bounds tend to merge with the result curve you are plotting (the
width of the confidence band shrinks to zero). This could lead you to conclude that the results are
very, very accurate. However, the underlying response surface approximation could have been
completely inadequate (for example, using a linear approximation function for a highly non-linear

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 101 of 207

problem).

Command(s):

RSSIMS

GUI:

Main Menu>Prob Design>RS Simulation

3.3.11. Reviewing Results Data

After probabilistic design looping is complete, you can review the results sets in a variety of ways
using the commands described in this section. These commands can be applied to the results from
any probabilistic design method or tool.

? Statistics
? Sample History

? Histogram

? Cumulative Distribution Function

? Probabilities

? Inverse Probabilities

? Trends
? Scatter Plot

? Sensitivities

? Correlation Matrix

? Report
? Print HTML Report

3.3.11.1. Viewing Statistics

To post-process one particular design variable, choose this option. The statistics options are
described below.

Plot Sample History

Use the PDSHIS command to request a sample history plot.

Command(s):

PDSHIS

GUI:

Main Menu>Prob Design>-Prob Results-Statistics>Sampl History

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 102 of 207

You must choose the results set you want to use, the design variable you want to review, the plot
type to use, and the confidence level.

Plot Histogram

Use the PDHIST command to request a histogram plot of a design variable.

Command(s):

PDHIST

GUI:

Main Menu>Prob Design>-Prob Results-Statistics>Histogram

You must choose the results set you want to use, the design variable you want to review, the number
of classes/points to use, and the type of histogram.

CumulativeDF

Use the PDCDF command to request a histogram plot of a design variable.

Command(s):

PDCDF

GUI:

Main Menu>Prob Design>-Prob Results-Statistics>CumulativeDF

You must choose the results set you want to use, the design variable you want to review, and the
confidence level.

The confidence level is a probability expressing the confidence that the values for the cumulative
distribution function are in fact between the confidence bounds. The larger the confidence level, the
wider the confidence bounds. Plotting of the confidence bounds only makes sense for the post-
processing of Monte Carlo simulation results. Here, the confidence bounds represent the accuracy of
the results and with increasing sample size the width of the confidence bounds gets smaller for the
same confidence level. For response surface methods the number of simulations done on the
response surface is usually very large. Therefore, the accuracy of the results is determined by the
goodness of the response surface fit and not by the confidence level.

Probabilities

Use the PDPROB command to request the value of a design variable at an specific point on the
cumulative distribution curve.

Command(s):

PDPROB

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 103 of 207

Main Menu>Prob Design>-Prob Results-Statistics>Probabilities...

You must choose the results set you want to use, the design variable you want to review, the relation
(greater than, less than), the limit value, and the confidence level.

Inverse Probabilities

Use the PDPINV command to request the value of a design variable at an specific point on the
cumulative distribution curve.

Command(s):

PDPINV

GUI:

Main Menu>Prob Design>-Prob Results-Statistics>Inverse Prob...

You must choose the results set you want to use, the design variable you want to review, the relation
(greater than, less than), the limit value, and the confidence level.

3.3.11.2. Viewing Trends

To post-process one particular design variable as it relates to another, choose this option. The trend
options are described below.

Scatter Plot

Use the PDSCAT command to request a scatter plot showing the correlation between two design
variables.

Command(s):

PDSCAT

GUI:

Main Menu>Prob Design>-Prob Results-Trends>Scatter Plot...

You must choose the results set you want to use, the design variables you want to review, the type of
trendline curve to use (and if plotted, the polynomial order), and the maximum number of point to
include in the scatter plot.

Sensitivities

Use the PDSENS command to request the sensitivities of an output parameter to the input variables.

Command(s):

PDSENS

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 104 of 207

Main Menu>Prob Design>-Prob Results-Trends>Sensitivities

You must choose the results set and output parameter you want to use, the type of chart to plot, the
type of correlation coefficient, and the sensitivity level.

Correlation Matrix

Use the PDCMAT command to calculate the correlation coefficient matrix.

Command(s):

PDCMAT

GUI:

Main Menu>Prob Design>-Prob Results-Trends>Probabilities...

You must choose the results set you want to use, which type of design variables you are looking at,
the specific design variable names, the type of correlation, the significance level, and whether you
want to see the probabilities with the correlation coefficients.

3.3.11.3. Creating Reports

To visualize and summarize all results of a probabilistic analysis, choose this option. Details on
specifying a report are described below.

Report Options

Use the PDROPT command to request an HTML report.

Command(s):

PDROPT

GUI:

Main Menu>Prob Design>-Prob Results-Report>Report Options...

You must choose which statistics and trends to show in the report and in what form you want to see
them. See the PDROPT command for details, and see the other probabilistic results options for
further details.

Generate Report

Use the PDWRITE command to request the sensitivities of an output parameter to the input
variables.

Command(s):

PDWRITE

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 105 of 207

Main Menu>Prob Design>-Prob Results-Report>Generate Report

You must enter a name for the report file, your first and last name, and whether links should be
generated between your report and the analysis file, each analysis loop, and the response surface
output parameter details (if the response surface method was used).

Prev Next
Guidelines for Choosing Probabilistic
Basic Definitions
Design Variables
Guidelines for Choosing Probabilistic Design Variables
Prev Probabilistic Design Next

3.4. Guidelines for Choosing Probabilistic Design Variables

There are many useful guidelines you can follow in defining your probabilistic design variables.
Some of these are presented below.

3.4.1. Choosing and Defining Random Input Variables

Here are a few tips you can use to determine which random input variable in your finite element
model follows which distribution function and what the parameters of the distribution function are.

First, you should know to

? Specify a reasonable range of values for each random input variable.

? Set reasonable limits on the variability for each RV.

Note

The values and hints given below are simply guidelines; none are absolute. Always
double check this information with an expert in your organization and adapt it as
needed to suit your analysis.

3.4.1.1. Random Input Variables for Monte Carlo Simulations

The number of simulation loops that are required for a Monte Carlo Simulation does not depend on
the number of random input variables. The required number of simulation loops only depends on the
amount of the scatter of the output parameters and the type of results you expect from the analysis. In
a Monte Carlo Simulation, it is a good practice to include all of the random input variables you can
think of even if you are not sure about their influence on the random output parameters. Exclude only
those random input variables where you are very certain that they have no influence. The
probabilistic design system will then automatically tell you which random input variables have
turned out to be significant and which one are not. The number of simulations that are necessary in a
Monte Carlo analysis to provide that kind of information is usually about 50 to 200. However, the
more simulation loops you perform, the more accurate the results will be.

3.4.1.2. Random Input Variables for Response Surface Analyses

The number of simulation loops that are required for a Response Surface analysis depends on the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 106 of 207

number of random input variables. Therefore, you want to select the most important input variable
(s), the ones you know have a significant impact on the random output parameters. If you are unsure
which random input variables are important, it is usually a good idea to include all of the random
variables you can think of and then perform a Monte Carlo Simulation. After you learn which
random input variables are important and therefore should be included in your Response Surface
Analysis, you can eliminate those that are unnecessary.

3.4.1.3. Choosing a Distribution for a Random Variable

The type and source of the data you have determines which distribution functions can be used or are
best suited to your needs.

3.4.1.3.1. Measured Data

If you have measured data then you first have to know how reliable that data is. Data scatter is not
just an inherent physical effect, but also includes inaccuracy in the measurement itself. You must
consider that the person taking the measurement might have applied a "tuning" to the data. For
example, if the data measured represents a load, the person measuring the load may have rounded the
measurement values; this means that the data you receive are not truly the measured values.
Depending on the amount of this "tuning," this could provide a deterministic bias in the data that you
need to address separately. If possible, you should discuss any bias that might have been built into
the data with the person who provided that data to you.

If you are confident about the quality of the data, then how to proceed depends on how much data
you have. In a single production field, the amount of data is typically sparse. If you have only few
data then it is reasonable to use it only to evaluate a rough figure for the mean value and the standard
deviation. In these cases, you could model the random input variable as a Gaussian distribution if the
physical effect you model has no lower and upper limit, or use the data and estimate the minimum
and maximum limit for a uniform distribution. In a mass production field, you probably have a lot of
data, in which case you could use a commercial statistical package that will allow you to actually fit
a statistical distribution function that best describes the scatter of the data.

3.4.1.3.2. Mean Values, Standard Deviation, Exceedence Values

The mean value and the standard deviation are most commonly used to describe the scatter of data.
Frequently, information about a physical quantity is given in the form that its value is; for example,
"100±5.5". Often, but not always, this form means that the value "100" is the mean value and "5.5" is
the standard deviation. To specify data in this form implies a Gaussian distribution, but you must
verify this (a mean value and standard deviation can be provided for any collection of data regardless
of the true distribution type). If you have more information (for example, you know that the data
must be lognormal distributed), then the PDS allows you to use the mean value and standard
deviation for a definition of a lognormal distribution.

Sometimes the scatter of data is also specified by a mean value and an exceedence confidence limit.
The yield strength of a material is sometimes given in this way; for example, a 99% exceedence limit
based on a 95% confidence level is provided. This means that derived from the measured data we
can be sure by 95% that in 99% of all cases the property values will exceed the specified limit and
only in 1% of all cases they will drop below the specified limit. The supplier of this information is
using mean value, the standard deviation, and the number of samples of the measured data to derive
this kind of information. If the scatter of the data is provided in this way, the best way to pursue this
further is to ask for more details from the data supplier. Because the given exceedence limit is based
on the measured data and its statistical assessment, the supplier might be able to provide you with the
details that were used.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 107 of 207

If the data supplier does not give you any further information, then you could consider assuming that

the number of measured samples was large. If the given exceedence limit is denoted with

and the given mean value is denoted with then the standard deviation can be derived from the
equation:

where the values for the coefficient C are:

Exceedence Probability C
99.5% 2.5758
99.0% 2.3263
97.5% 1.9600
95.0% 1.6449
90.0% 1.2816

3.4.1.3.3. No Data

In situations where no information is available, there is never just one right answer. Below are hints
about which physical quantities are usually described in terms of which distribution functions. This
might help you with the particular physical quantity you have in mind. Also below is a list of which
distribution functions are usually used for which kind of phenomena. Keep in mind that you might
need to choose from multiple options.

Geometric Tolerances

? If you are designing a prototype, you could assume that the actual dimensions of the
manufactured parts would be somewhere within the manufacturing tolerances. In this case it is
reasonable to use a uniform distribution, where the tolerance bounds provide the lower and
upper limits of the distribution function.

? Sometimes the manufacturing process generates a skewed distribution; for example, one half
of the tolerance band is more likely to be hit than the other half. This is often the case if
missing half of the tolerance band means that rework is necessary, while falling outside the
tolerance band on the other side would lead to the part being scrapped. In this case a Beta
distribution is more appropriate.

? Often a Gaussian distribution is used. The fact that the normal distribution has no bounds (it
spans minus infinity to infinity), is theoretically a severe violation of the fact that geometrical
extensions are described by finite positive numbers only. However, in practice this is irrelevant
if the standard deviation is very small compared to the value of the geometric extension, as is
typically true for geometric tolerances.

Material Data

? Very often the scatter of material data is described by a Gaussian distribution.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 108 of 207

? In some cases the material strength of a part is governed by the "weakest-link-theory". The
"weakest-link-theory" assumes that the entire part would fail whenever its weakest spot would
fail. for material properties where the "weakest-link" assumptions are valid, then the Weibull
distribution might be applicable.

? For some cases, it is acceptable to use the scatter information from a similar material type.
Let's assume that you know that a material type very similar to the one you are using has a
certain material property with a Gaussian distribution and a standard deviation of ±5% around
the measured mean value; then let's assume that for the material type you are using, you only
know its mean value. In this case, you could consider using a Gaussian distribution with a
standard deviation of ±5% around the given mean value.

? For temperature-dependent materials it is prudent to describe the randomness by separating the

temperature dependency from the scatter effect. In this case you need the mean values of your
material property as a function of temperature in the same way that you need this information
to perform a deterministic analysis. If M(T) denotes an arbitrary temperature dependent
material property then the following approaches are commonly used:

? Multiplication equation — M(T)rand = Crand (T)

? Additive equation — M(T)rand = (T) + Mrand

? Linear equation — M(T)rand = Crand (T) + Mrand

Here, (T) denotes the mean value of the material property as a function of temperature. In
the "multiplication equation" the mean value function is scaled with a coefficient Crand and
this coefficient is a random variable describing the scatter of the material property. In the
"additive equation" a random variable Mrand is added on top of the mean value function
(T). The "linear equation" combines both approaches and here both Crand and M rand are
random variables. However, you should take into account that in general for the "linear
equation" approach Crand and Mrand are, correlated.

Deciding which of these approaches is most suitable to describing the scatter of the temperature
dependent material property requires that you have some raw data about this material property. Only
by reviewing the raw data and plotting it versus temperature you can tell which approach is the better
one.

Load Data

? For loads, you usually only have a nominal or average value. You could ask the person who
provided the nominal value the following questions: If we have 1000 components that are
operated under real life conditions, what would the lowest load value be that only one of these
1000 components is subjected to and all others have a higher load? What would the most likely
load value be, i.e. the value that most of these 1000 components have (or are very close to)?
What would the highest load value be that only one of the 1000 components is subjected to
and all others have a lower load? To be safe you should ask these questions not only of the
person who provided the nominal value, but also to one or more experts who are familiar with
how your products are operated under real-life conditions. From all the answers you get, you
can then consolidate what the minimum, the most likely, and the maximum value probably is.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 109 of 207

As verification you can compare this picture with the nominal value that you would use for a
deterministic analysis. If the nominal value does not have a conservative bias to it then it
should be close to the most likely value. If the nominal value includes a conservative
assumption (is biased), then its value is probably close to the maximum value. Finally, you can
use a triangular distribution using the minimum, most likely, and maximum values obtained.

? If the load parameter is generated by a computer program then the more accurate procedure is
to consider a probabilistic analysis using this computer program as the solver mechanism. Use
a probabilistic design technique on that computer program to assess what the scatter of the
output parameters are, and apply that data as input to a subsequent analysis. In other words,
first run a probabilistic analysis to generate an output range, and then use that output range as
input for a subsequent probabilistic analysis.

Here, you have to distinguish if the program that generates the loads is ANSYS itself or your
own in-house program. If you have used ANSYS to generate the loads (for example,
FLOTRAN analysis calculating fluid loads on a structure or a thermal analysis calculating the
thermal loads of a structure) then we highly recommend that you include these load calculation
steps in the analysis file (and therefore in the probabilistic analysis). In this case you also need
to model the input parameters of these load calculation steps as random input variables. If you
have used your own in-house program to generate the loads, you can still integrate the load
calculation program in the analysis file (see the /SYS command for details), but you must have
an interface between that program and ANSYS that allows the programs to communicate with
each other and thus automatically transfer data.

You also have to distinguish if the load values are random fields or single random variables. If the
load is different from node to node (element to element) then it is most appropriate to include the
program calculating the load in the analysis file. If the load is described by one or very few constant
values then you can also consider performing a probabilistic analysis with the program calculating
these load values. Again you need to provide an interface to transfer input data to this program and
get output data (the loads) back to ANSYS. If there is more than just one single load value generated
by the program then you should also check for potential correlations.

3.4.1.4. Distribution Functions

Beta Distribution

The Beta distribution is very useful for random variables that are bounded at both sides. If linear
operations are performed on random variables that are all subjected to a uniform distribution then the
results can usually be described by a Beta distribution. An example is if you are dealing with
tolerances and assemblies, where the components are assembled and the individual tolerances of the
components follow a uniform distribution. In this case the overall tolerances of the assembly are a
function of adding or subtracting the geometrical extension of the individual components (a linear
operation). Hence, the overall tolerances of the assembly can be described by a Beta distribution.
Also, as previously mentioned, the Beta distribution can be useful for describing the scatter of
individual geometrical extensions of components as well. The uniform distribution is a special case
of the Beta distribution.

Exponential Distribution

The exponential distribution is useful in cases where there is a physical reason that the probability
density function is strictly decreasing as the random input variable value increases. The distribution
is mostly used to describe time-related effects; for example, it describes the time between
independent events occurring at a constant rate. It is therefore very popular in the area of systems
reliability and lifetime-related systems reliability, and it can be used for the life distribution of non-

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 110 of 207

redundant systems. Typically, it is used if the lifetime is not subjected to wear-out and the failure rate
is constant with time. Wear-out is usually a dominant life-limiting factor for mechanical components,
which would preclude the use of the exponential distribution for mechanical parts. However in cases
where preventive maintenance exchanges parts before wear-out can occur, then the exponential
distribution is still useful to describe the distribution of the time until exchanging the part is
necessary.

Gamma Distribution

The Gamma distribution is again a more time-related distribution function. For example it describes
the distribution of the time required for exactly k events to occur under the assumption that the
events take place at a constant rate. It is also used to describe the time to failure for a system with
standby components.

Gaussian (Normal) Distribution

The Gaussian or normal distribution is a very fundamental and commonly used distribution for
statistical matters. It is typically used to describe the scatter of the measurement data of many
physical phenomena. Strictly speaking, every random variable follows a normal distribution if it is
generated by a linear combination of a very large number of other random effects, regardless which
distribution these random effects originally follow. The Gaussian distribution is also valid if the
random variable is a linear combination of two or more other effects if those effects also follow a
Gaussian distribution.

Lognormal Distribution

The lognormal distribution is a basic and commonly used distribution. It is typically used to describe
the scatter of the measurement data of physical phenomena, where the logarithm of the data would
follow a normal distribution. The lognormal distribution is very suitable for phenomena that arise
from the multiplication of a large number of error effects. It is also correct to use the lognormal
distribution for a random variable that is the result of multiplying two or more random effects (if the
effects that get multiplied are also lognormally distributed). If is often used for lifetime distributions;
for example, the scatter of the strain amplitude of a cyclic loading that a material can endure until
low-cycle-fatigue occurs is very often described by a lognormal distribution.

Uniform Distribution

The uniform distribution is a very fundamental distribution for cases where no other information
apart from a lower and an upper limit exists. If is very useful to describe geometric tolerances. It can
also be used in cases where there is no evidence that any value of the random variable is more likely
than any other within a certain interval. In this sense it can be used for cases where "lack of
engineering knowledge" plays a role.

Triangular Distribution

The triangular distribution is most helpful to model a random variable when actual data is not
available. It is very often used to cast the results of expert-opinion into a mathematical form, and is
often used to describe the scatter of load parameters. However, regardless of the physical nature of
the random variable you want to model, you can always ask some experts questions like "What is the
one-in-a-thousand minimum and maximum case for this random variable? and other similar
questions. You should also include an estimate for the random variable value derived from a
computer program, as described earlier. This is also described in more detail above for load
parameters in Choosing a Distribution for a Random Variable.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 111 of 207

Truncated Gaussian Distribution

The truncated Gaussian distribution typically appears where the physical phenomenon follows a
Gaussian distribution, but the extreme ends are cut off or are eliminated from the sample population
by quality control measures. As such, it is useful to describe the material properties or geometric
tolerances.

Weibull Distribution

In engineering, the Weibull distribution is most often used for strength or strength-related lifetime
parameters, and it is the standard distribution for material strength and lifetime parameters for very
brittle materials (for these very brittle material the "weakest-link-theory" is applicable). For more
details see Choosing a Distribution for a Random Variable.

3.4.2. Choosing Random Output Parameters

Output parameters are usually parameters such as length, thickness, diameter, or model coordinates.

The ANSYS PDS does not restrict you in the number of random output parameters as long as the
total number of probabilistic design variable (that is random input variables and random output
parameters together) does not exceed 5000. We therefore recommend that you include all output
parameters you can think of and that might be useful to you. The additional computing time that is
required to handle more random output parameters is marginal compared to time it takes to solve the
problem. It is better to define random output parameters that you might not consider important before
you start the analysis. If you forgot to include a random output parameter that later turns out to be
important, you must redo the entire analysis.

Prev Next
How to Do Probabilistic Design Probabilistic Design Techniques
Probabilistic Design Techniques
Prev Probabilistic Design Next

3.5. Probabilistic Design Techniques

Understanding the algorithm used by a computer program is always helpful; this is particularly true
in the case of probabilistic design. This section presents details on the method types and the sampling
options associated with each. See the ANSYS, Inc. Theory Reference for more information.

3.5.1. Monte Carlo Simulations

The Monte Carlo Simulation method is the most common and traditional method for a probabilistic
analysis. This method lets you simulate how virtual components behave the way they are built. One
simulation loop represents one manufactured component that is subjected to a particular set of loads
and boundary conditions.

For Monte Carlo simulations, you can choose either the Direct Sampling method or the Latin
Hypercube Sampling method. Both are described in later sections.

When you manufacture a component, you can measure its geometry and all of its material properties
(although typically, the latter is not done because this can destroy the component). In the same sense,

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 112 of 207

if you started operating the component then you could measure the loads it is subjected to. Again, to
actually measure the loads is very often impractical. But the bottom line is that once you have a
component in your hand and start using it then all the input parameters have very specific values that
you could actually measure. With the next component you manufacture you can do the same; if you
compared the parameters of that part with the previous part, you would find that they vary slightly.
This comparison of one component to the next illustrates the scatter of the input parameters. The
Monte Carlo Simulation techniques mimic this process. With this method you “virtually”
manufacture and operate components or parts one after the other.

The advantages of the Monte Carlo Simulation method are:

? The method is always applicable regardless of the physical effect modeled in a finite element
analysis. It not based on assumptions related to the random output parameters that if satisfied
would speed things up and if violated would invalidate the results of the probabilistic analysis.
Assuming the deterministic model is correct and a very large number of simulation loops are
performed, then Monte Carlo techniques always provide correct probabilistic results. Of
course, it is not feasible to run an infinite number of simulation loops; therefore, the only
assumption here is that the limited number of simulation loops is statistically representative
and sufficient for the probabilistic results that are evaluated. This assumption can be verified
using the confidence limits, which the PDS also provides.

? Because of the reason mentioned above, Monte Carlo Simulations are the only probabilistic
methods suitable for benchmarking and validation purposes.

? The individual simulation loops are inherently independent; the individual simulation loops do
not depend on the results of any other simulation loops. This makes Monte Carlo Simulation
techniques ideal candidates for parallel processing.

The Direct Sampling Monte Carlo technique has one drawback: it is not very efficient in terms of
required number of simulation loops.

3.5.1.1. Direct Sampling

Direct Monte Carlo Sampling is the most common and traditional form of a Monte Carlo analysis. It
is popular because it mimics natural processes that everybody can observe or imagine and is
therefore easy to understand. For this method, you simulate how your components behave based on
the way they are built. One simulation loop represents one component that is subjected to a particular
set of loads and boundary conditions.

The Direct Monte Carlo Sampling technique is not the most efficient technique, but it is still widely
used and accepted, especially for benchmarking and validating probabilistic results. However,
benchmarking and validating requires many simulation loops, which is not always feasible. This
sampling method is also inefficient due to the fact that the sampling process has no "memory".

For example, if we have two random input variables X1 and X2 both having a uniform distribution
ranging from 0.0 to 1.0, and we generate 15 samples, we could get a cluster of two (or even more)
sampling points that occur close to each other if we graphed the two variables (see figure below).
While in the space of all random input variables, it can happen that one sample has input values close
to another sample, this does not provide new information and insight into the behavior of a
component in a computer simulation if the same (or almost the same) samples are repeated.

Graph of X1 and X2 Showing Two Samples with Close Values

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 113 of 207

To use Direct Monte Carlo Sampling, do the following

Command(s):

PDMETH,MCS,DIR

PDDMCS

GUI:

Main Menu>Prob Design>Monte Carlo Sims

In this sampling method, you set the number of simulations, whether to stop simulations loops when
certain criteria are met (accuracy for mean values and standard deviations), and the seed value for
randomizing input variable sample data.

3.5.1.2. Latin Hypercube Sampling

The Latin Hypercube Sampling (LHS) technique is a more advanced and efficient form for Monte
Carlo Simulation methods. The only difference between LHS and the Direct Monte Carlo Sampling
technique is that LHS has a sample "memory," meaning it avoids repeating samples that have been
evaluated before (it avoids clustering samples). It also forces the tails of a distribution to participate
in the sampling process. Generally, the Latin Hypercube Sampling technique requires 20% to 40%
fewer simulations loops than the Direct Monte Carlo Simulation technique to deliver the same results
with the same accuracy. However, that number is largely problem dependent.

Graph of X1 and X2 Showing Good Sample Distribution

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 114 of 207

To use the Latin Hypercube Sampling technique:

Command(s):

PDMETH,MCS,LHS

PDLHS

GUI:

Main Menu>Prob Design>Monte Carlo Sims

In this sampling method, you set the number of simulations and repetitions, the location in the
interval for the sample, whether the simulations stop when certain criteria are met (accuracy of mean
values and standard deviations), and random number seed for variability in the sample input variable
data.

3.5.1.3. User-Defined Sampling

For this method, you provide the file containing the samples.

Command(s):

PDMETH,MCS,USER

PDUSER

GUI:

Main Menu>Prob Design>-Prob Method- Method/Tool>Monte Carlo Sims

By using this option you have complete control over the sampling data. You are required to give the
file name and path.

If user-specified sampling methods are requested with the PDMETH,MCS,USER command or the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 115 of 207

PDMETH,RSM,USER command, then you need to specify which file contains the sample data. The
sample data is a matrix, where the number of columns is equal to the number of defined random
variables and the number of lines is equal to the number of simulation loops requested. This data
must be contained in an ASCII file and the content must obey the following notations and format
requirements:

? Column separators allowed: blank spaces, commas, semi-colons, and tabs.

? Multiple blank spaces and multiple tabs placed directly one after the other are allowed and are
considered as one single column separator.

? Multiple commas or semi-colons placed directly one after the other are not allowed; for
example, two commas with no data between them (just blanks) are read as an empty column,
which leads to an error message.

? The first line of the file must contain a solution label. No additional data is allowed on the first
line, and if found, will lead to an error message. An error message is also issued if the solution
label is missing.

? The solution label is just a placeholder. For consistency, you should use the same solution
label you specify in the PDEXE command, but if they are different, you will always use the
solution label specified in the PDEXE command for post-processing. The PDS system does
not check if the solution label in the user-specified file and the one given in the PDEXE
command match.

? The second line of the file must contain the headers of the data columns. The first three
column headers must be “ITER ”, “CYCL”, and “LOOP”, respectively; then subsequent
columns should contain the names of the random variables. You must use one of the allowed
separators as described above between the column headers. No additional data is allowed on
this line, and if found, will prompt an error message. An error message is also issued if any of
the required column headers are missing.

? The random variable names in the file must match the names of the defined random variables.

? Columns four to n can be in arbitrary order. The ANSYS PDS tool determines the order for the
random variable data based on the order of the random variable names in the second line.

? The third and subsequent lines must contain the order number for the iteration, cycle, and
simulation loop, then the random variable values for that loop. The iteration, cycle, and
simulation loop numbers must be in the first, second, and third columns respectively, followed
by the random variable values. The iteration and cycle numbers are used by the ANSYS PDS
(internally) and for a user-defined sampling method you will typically use a value of "1" for all
simulation loops. The loop number is an ascending number from 1 to the total number of loops
requested. Additional data is not allowed, and if found, will lead to an error message. An error
message is also issued if any of the data columns are missing.

? You must be sure that the order of the random variable values in each line is identical to the
order of the random variable names in the second line.

? The user-specified sampling file must contain a minimum of one data line for the random
variable values.

When the PDUSER command is issued, the PDS checks that the specified file exists, then verifies it
for completeness and consistency. Consistency is checked according to the possible minimum and

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 116 of 207

maximum boundaries of the distribution type of the individual random variables. An error message is
issued if a random variable value is found in the file that is below the minimum boundary or above
the maximum boundary of the distribution of that random variable. This also means that any value
will be accepted for a random variable if its distribution has no minimum or maximum boundary; for
example, this is the case for the gaussian (normal) distribution. Apart from this check, it is your
responsibility to provide values for the random variables that are consistent with their random
distribution.

Note

It is your responsibility to ensure that parameters defined as random input variables

are actually input parameters for the analysis defined with the PDANL command.
Likewise, you must ensure that parameters defined as random output parameter are in
fact results generated in the analysis file.

Example

An excerpt of the content of a user-specified sampling file is given below. This example is based on
three random variables named X1, X2, and X3. A total of 100 simulation loops are requested.

USERSAMP
ITER CYCL LOOP X1 X2 X3
1 1 1 1.619379209e+000 2.364528435e-001 1.470789050e+000
1 1 2 2.237676559e-001 5.788049712e-001 1.821263115e+000
1 1 3 7.931615474e+000 8.278689033e-001 2.170793522e+000
.. .. .. ... ... ...
.. .. .. ... ... ...
1 1 98 1.797221666e+000 3.029471373e-001 1.877701081e+000
1 1 99 1.290815540e+001 9.271606216e-001 2.091047328e+000
1 1 100 4.699281922e+000 6.526505821e-001 1.901013985e+000

3.5.2. Response Surface Analysis Methods

For Response Surface Analysis, you can choose from three sampling methods: Central Composite
Design, Box-Behnken Matrix, and User-defined.

Response Surface Methods are based on the fundamental assumption that the influence of the
random input variables on the random output parameters can be approximated by mathematical
function. Hence, Response Surface Methods locate the sample points in the space of random input
variables such that an appropriate approximation function can be found most efficiently; typically,
this is a quadratic polynomial. In this case the approximation function is described by

where c0 is the coefficient of the constant term, ci, i = 1,...NRV are the coefficients of the linear terms
and cij , i = 1,...NRV and j = i,...,NRV are the coefficients of the quadratic terms. To evaluate these
coefficients a regression analysis is used and the coefficients are usually evaluated such that the sum
of squared differences between the true simulation results and the values of the approximation
function is minimized.

Hence, a response surface analysis consists of two steps:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 117 of 207

1. Performing the simulation loops to calculate the values of the random output parameters that
correspond to the sample points in the space of random input variables.

2. Performing a regression analysis to derive the terms and the coefficients of the approximation
function.

The fundamental idea of Response Surface Methods is that once the coefficients of a suitable
approximation function are found, then we can directly use the approximation function instead of
looping through the finite element model. To perform a finite element analysis might require minutes
to hours of computation time; in contrast, evaluating a quadratic function requires only a fraction of a
second. Hence, if using the approximation function, we can afford to evaluate the approximated
response parameter thousands of times.

A quadratic polynomial is sufficient in many cases of engineering analysis (for example, the
evaluation of the thermal stress mentioned above). For that evaluation, the Young's modulus and the
thermal expansion coefficient both have a linear effect on the thermal stresses, which is taken into
account in a quadratic approximation by the mixed quadratic terms. However, there are cases where
a quadratic approximation is not sufficient; for example, if the finite element results are used to
calculate the lifetime of a component. For this evaluation, the lifetime typically shows an exponential
behavior with respect to the input parameters; thus the lifetime results cannot be directly or
sufficiently described by a quadratic polynomial. But often, if you apply a logarithmic
transformation to the lifetime results, then these transformed values can be approximated by a
quadratic polynomial. The ANSYS PDS offers a variety of transformation functions that you can
apply to the response parameters, and the logarithmic transformation function is one of them.

Assuming the approximation function is suitable for your problem, the advantages of the Response
Surface Method are:

? It often requires fewer simulation loops than the Monte Carlo Simulation method.

? It can evaluate very low probability levels. This is something the Monte Carlo Simulation
method cannot do unless you perform a great number of simulation loops.

? The goodness-of-fit parameters tell you how good the approximation function is (in other
words, how accurate the approximation function is that describes your "true" response
parameter values). The goodness-of-fit parameters can warn you if the approximation function
is insufficient.

? The individual simulation loops are inherently independent (the individual simulation loops do
not depend on the results of any other simulation loops). This makes Response Surface
Method an ideal candidate for parallel processing.

The disadvantages of the Response Surface Method are:

? The number of required simulation loops depends on the number of random input variables. If
you have a very large number of random input variables (hundreds or even thousands), then a
probabilistic analysis using Response Surface Methods would be impractical.

? This method is not usually suitable for cases where a random output parameter is a non-
smooth function of the random input variables. For example, a non-smooth behavior is given if
you observe a sudden jump of the output parameter value even if the values for the random
input variables vary only slightly. This typically occurs if you have instability in your model
(such as bulking). The same might happen if the model includes a sharp non-linearity such as a
linear-elastic-ideal-plastic material behavior. Or, if you are analyzing a contact problem, where

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 118 of 207

only a slight variation in your random input variables can change the contact situation from
contact to non-contact or vice versa, then you also might have problems using the Response
Surface Method.

Note

To use Response Surface Methods, the random output parameters must be smooth and
continuous functions of the involved random input variables. Do not use Response
Surface Methods if this condition is not satisfied.

3.5.2.1. Central Composite Design Sampling

A central composite design consists of a central point, the N axis point plus 2 N-f factorial points
located at the corners of an N-dimensional hypercube. Here, N is the number of random input
variables and f is the fraction of the factorial part of the central composite design. A fraction f = 0 is a
called a full factorial design, f = 1 gives a half-factorial design, and so on. The PDS gradually
increases the fraction f as you increase the number of random input variables. This keeps the number
of simulation loops reasonable. The fraction f is automatically evaluated such that a resolution V
design is always maintained. A resolution V design is a design where none of the second order terms
of the approximation function are confined with each other. This ensures a reasonable accuracy for
the evaluation of the coefficients of the second order terms.

The locations of the sampling points for a problem with three random input variables is illustrated
below.

Locations of Sampling Points for Problem with Three Input Variables for CCD

The number of sample points (simulation loops) required for a central composite design as a function
of the number of random input variables is given in the table below:

Number of random Number of coefficients in a Factorial Number sample points

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 119 of 207

input variables quadratic function (with cross- number f (simulation loops)

terms)
1 3 N/A N/A
2 6 0 9
3 10 0 15
4 15 0 25
5 21 1 27
6 28 1 45
7 36 1 79
8 45 2 81
9 55 2 147
10 66 3 149
11 78 4 151
12 91 4 281
13 105 5 283
14 120 6 285
15 136 7 287
16 153 8 289
17 171 9 291
18 190 9 549
19 210 10 551
20 231 11 553

To use the Response Surface Method with a Central Composite Design, do the following:

Command(s):

PDMETH,RSM,CCD

PDCCD

GUI:

Main Menu>Prob Design>-Prob Method- Response Surface

PDCCD also allows you to specify the factorial fraction (Fract) and five probability levels. See the
ANSYS, Inc. Theory Reference for more details.

3.5.2.2. Box-Behnken Matrix Sampling

A Box-Behnken Design consists of a central point plus the midpoints of each edge of an N-
dimensional hypercube.

The location of the sampling points for a problem with three random input variables is illustrated
below.

Location of Sampling Points for Problem with Three Input Variables for BBM

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 120 of 207

The number of sample points (simulation loops) required for a Box-Behnken design as a function of
the number of random input variables is given in the table below:

Number of random Number of coefficients in a quadratic Number sample points

input variables function (with cross-terms) (simulation loops)
1 N/A
2 6 N/A
3 10 12
4 15 25
5 21 41
6 28 49
7 36 57
8 45 65
9 55 121
10 66 161
11 78 177
12 91 193

To use Response Surface Analysis with the Box-Behnken design option, do the following:

Command(s):

PDMETH,RSM,BBM

PDBBM

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 121 of 207

Main Menu>Prob Design>-Prob Method- Response Surface

For the Box-Behnken method, you can modify the default probability levels.

3.5.2.3. User-defined Sampling

For this method, you provide the file containing the samples.

Command(s):

PDMETH,RSM,USER

PDUSER

GUI:

Main Menu>Prob Design>-Prob Method- Response Surface

By using this option, you have complete control over the sampling data. You are required to give the
file name and path.

Prev Next
Guidelines for Choosing Probabilistic Post-Processing Probabilistic Analysis
Design Variables Results
Post-Processing Probabilistic Analysis Results
Prev Probabilistic Design Next

3.6. Post-Processing Probabilistic Analysis Results

There are two groups of post-processing functions in the probabilistic design system - statistical and
trend analyses.

A statistical analysis is an evaluation function performed on a single probabilistic design variable; for
example, a histogram plot of a random output parameter. The ANSYS PDS allows statistical
evaluation of either random output parameters or random input variables.

A trend analysis typically involves two or more probabilistic design variables; for example, a scatter
plot of one probabilistic design variable versus another.

Probabilistic post-processing is generally based on result sets. A result set is either a solution set
generated by a probabilistic analysis run (see the PDEXE command), or a response surface set (see
the RSFIT command). With a few exceptions, you can perform the same type of probabilistic post-
processing operations regardless if you post-process the results of a solution set or the results of a
response surface set.

3.6.1. Statistical Post-Processing

Statistical post-processing allows you several options for reviewing your results.

3.6.1.1. Sample History

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 122 of 207

The most fundamental form of post-processing is directly reviewing the simulation loop results as a
function for the number of simulation loops. Here, you can review the simulation values (for either
response surface or Monte Carlo methods), or for Monte Carlo Simulations only, the mean,
minimum, or maximum values, or the standard deviations.

It is most helpful to review the mean values and standard deviation history for Monte Carlo
Simulation results if you want to decide if the number of simulation loops was sufficient. If the
number of simulation loops was sufficient, the mean values and standard deviations for all random
output parameters should have converged. Convergence is achieved if the curve shown in the
respective plots approaches a plateau. If the curve shown in the diagram still has a significant and
visible trend with increasing number of simulation loops then you should perform more simulation
loops. In addition, the confidence bounds plotted for the requested history curves can be interpreted
as the accuracy of the requested curve. With more simulation loops, the width of the confidence
bounds is reduced.

Typically, post-processing Monte Carlo results based on response surfaces is based on a very large
number of simulation loops using the response surface equations. Therefore, the accuracy of the
results is not a function of the number of simulation loops, but rather the goodness-of-fit measures of
the individual response surfaces. As one example, suppose the goodness-of-fit measures indicate a
very poor quality of a response surface fit, but the mean value and standard deviation history indicate
that the results have converged (because of the large number of loops) and the accuracy is excellent
(again because confidence bounds shrink with increasing number of loops). This could lead you to an
incorrect conclusion. This is why the PDS only allows you to visualize the sample value history
directly, but not the mean value history and so on.

To review the simulation loop results:

Command(s):

PDSHIS

GUI:

Main Menu>Prob Design>Statistics>Sampl History

You need to select the results set label and the design variable, choose a plot type, and set the
confidence level.

3.6.1.2. Histogram

A histogram plot is most commonly used to visualize the scatter of a probabilistic design variable. A
histogram is derived by dividing the range between the minimum value and the maximum value into
intervals of equal size. Then the PDS determines how many samples fall within each interval, that is,
how many "hits" landed in the intervals.

Most likely, you will use histograms to visualize the scatter of your random output parameters. The
ANSYS PDS also allows you to plot histograms of your random input variables so you can double
check that the sampling process generated the samples according to the distribution function you
specified. For random input variables, the PDS not only plots the histogram bars, but also a curve for
values derived from the distribution function you specified. Visualizing histograms of the random
input variables is another way to make sure that enough simulation loops have been performed. If the
number of simulation loops is sufficient, the histogram bars will:

? Be close to the curve that is derived from the distribution function

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 123 of 207

? Be "smooth" (without large “steps”)

? Not have major gaps

A major gap is given if you have no hits in an interval where neighboring intervals have many hits.
However, if the probability density function is flattening out at the far ends of a distribution (for
example, the exponential distribution flattens out for large values of the random input variable) then
there might logically be gaps. Hits are counted only as positive integer numbers and as these
numbers gradually get smaller, a zero hit can happen in an interval.

To plot histograms:

Command(s):

PDHIST

GUI:

Main Menu>Prob Design>Statistics>Histogram

3.6.1.3. Cumulative Distribution Function

The cumulative distribution function is a primary review tool if you want to assess the reliability or
the failure probability of your component or product. Reliability is defined as the probability that no
failure occurs. Hence, in a mathematical sense reliability and failure probability are two sides of the
same coin and numerically they complement each other (are additive to 1.0). The cumulative
distribution function value at any given point expresses the probability that the respective parameter
value will remain below that point. The figure below shows the cumulative distribution function of
the random property X:

Cumulative Distribution Function of X

The value of the cumulative distribution function at the location x0 is the probability that the values
of X stay below x0 . Whether this probability represents the failure probability or the reliability of
your component depends on how you define failure; for example, if you design a component such
that a certain deflection should not exceed a certain admissible limit then a failure event occurs if the
critical deflection exceeds this limit. Thus for this example, the cumulative distribution function is

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 124 of 207

interpreted as the reliability curve of the component. On the other hand, if you design a component
such that the eigenfrequencies are beyond a certain admissible limit then a failure event occurs if an
eigenfrequency drops below this limit. Thus for this example, the cumulative distribution function is
interpreted as the failure probability curve of the component.

The cumulative distribution function also lets you visualize what the reliability or failure probability
would be if you chose to change the admissible limits of your design.

Often you are interested in visualizing low probabilities and you want to assess the more extreme
ends of the distribution curve. In this case plotting the cumulative distribution function in one of the
following ways is more appropriate:

? As a Gauss plot (also called a "normal plot"). If the probabilistic design variable follows a
Gaussian distribution then the cumulative distribution function is displayed as a straight line in
this type of plot.

? As a lognormal plot. If the probabilistic design variable follows a lognormal distribution then
the cumulative distribution function is displayed as a straight line in this type of plot

? As a Weibull plot. If the probabilistic design variable follows a Weibull distribution then the
cumulative distribution function is displayed as a straight line in this type of plot.

The advantage of these plots is that the probability axis is scaled in a non-linear fashion such that the
extreme ends of the distribution function are emphasized and more visible.

To plot the cumulative distribution function:

Command(s):

PDCDF

GUI:

Main Menu>Prob Design>Statistics>CumulativeDF

3.6.1.4. Print Probabilities

The PDS offers a function where you can determine the cumulative distribution function at any point
along the axis of the probabilistic design variable, including an interpolation function so you can
evaluate the probabilities between sampling points. This feature is most helpful if you want to
evaluate the failure probability or reliability of your component for a very specific and given limit
value.

To print probabilities:

Command(s):

PDPROB

GUI:

Main Menu>Prob Design>Statistics>Probabilities

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 125 of 207

3.6.1.5. Print Inverse Probabilities

The PDS offers a function where you can probe the cumulative distribution function by specifying a
certain probability level; the PDS tells you at which value of the probabilistic design variable this
probability will occur. This is helpful if you want to evaluate what limit you should specify to not
exceed a certain failure probability, or to specifically achieve a certain reliability for your
component.

To print inverse probabilities:

Command(s):

PDPINV

GUI:

Main Menu>Prob Design>Statistics>Inverse Prob

3.6.2. Trend Post-Processing

Trend post-processing allows you several options for reviewing your results.

3.6.2.1. Sensitivities

Probabilistic sensitivities are important in allowing you to improve your design toward a more
reliable and better quality product, or to save money while maintaining the reliability or quality of
your product. You can request a sensitivity plot for any random output parameter in your model.

There is a difference between probabilistic sensitivities and deterministic sensitivities. Deterministic

sensitivities are mostly only local gradient information. For example, to evaluate deterministic
sensitivities you can vary each input parameters by ±10% (one at a time) while keeping all other
input parameters constant, then see how the output parameters react to these variations. As illustrated
in the figure below, an output parameter would be considered very sensitive with respect to a certain
input parameter if you observe a large change of the output parameter value.

Sensitivities

These purely deterministic considerations have various disadvantages that are taken into
consideration for probabilistic sensitivities, namely:

? A deterministic variation of an input parameter that is used to determine the gradient usually
does not take the physical range of variability into account. An input parameter varied by
±10% is not meaningful for the analysis if ±10% is too much or too little compared with the
actual range of physical variability and randomness. In a probabilistic approach the physical

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 126 of 207

range of variability is inherently considered because of the distribution functions for input
parameters. Probabilistic sensitivities measure how much the range of scatter of an output
parameter is influenced by the scatter of the random input variables. Hence, two effects have
an influence on probabilistic sensitivities: the slope of the gradient, plus the width of the
scatter range of the random input variables. This is illustrated in the figures below. If a random
input variable has a certain given range of scatter, then the scatter of the corresponding random
output parameter is larger, and the larger the slope of the output parameter curve is (first
illustration). But remember that an output parameter with a moderate slope can have a
significant scatter if the random input variables have a wider range of scatter (second
illustration).

Range of Scatter

? Gradient information is local information only. It does not take into account that the output
parameter may react more or less with respect to variation of input parameters at other
locations in the input parameter space. However, the probabilistic approach not only takes the
slope at a particular location into account, but also all the values the random output parameter
can have within the space of the random input variables.

? Deterministic sensitivities are typically evaluated using a finite-differencing scheme (varying

one parameter at a time while keeping all others fixed). This neglects the effect of interactions
between input parameters. An interaction between input parameters exists if the variation of a
certain parameter has a greater or lesser effect if at the same time one or more other input
parameters change their values as well. In some cases interactions play an important or even
dominant role. This is the case if an input parameter is not significant on its own but only in
connection with at least one other input parameter. Generally, interactions play an important
role in 10% to 15% of typical engineering analysis cases (this figure is problem dependent). If
interactions are important, then a deterministic sensitivity analysis can give you completely
incorrect results. However, in a probabilistic approach, the results are always based on Monte
Carlo simulations, either directly performed using you analysis model or using response
surface equations. Inherently, Monte Carlo simulations always vary all random input variables
at the same time; thus if interactions exist then they will always be correctly reflected in the
probabilistic sensitivities.

To display sensitivities, the PDS first groups the random input variables into two groups: those
having a significant influence on a particular random output parameter and those that are rather
insignificant, based on a statistical significance test. This tests the hypothesis that the sensitivity of a
particular random input variable is identical to zero and then calculates the probability that this
hypothesis is true. If the probability exceeds a certain significance level (determining that the
hypothesis is likely to be true), then the sensitivity of that random input variable is negligible. The
PDS will plot only the sensitivities of the random input variables that are found to be significant.
However, insignificant sensitivities are printed in the output window. You can also review the
significance probabilities used by the hypothesis test to decide which group a particular random input
variable belonged to.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 127 of 207

The PDS allows you to visualize sensitivities either as a bar chart, a pie chart, or both. Sensitivities
are ranked so the random input variable having the highest sensitivity appears first.

In a bar chart the most important random input variable (with the highest sensitivity) appears in the
leftmost position and the others follow to the right in the order of their importance. A bar chart
describes the sensitivities in an absolute fashion (taking the signs into account); a positive sensitivity
indicates that increasing the value of the random input variable increases the value of the random
output parameter for which the sensitivities are plotted. Likewise, a negative sensitivity indicates that
increasing the random input variable value reduces the random output parameter value. In a pie
chart, sensitivities are relative to each other.

In a pie chart the most important random input variable (with the highest sensitivity) will appear first
after the 12 o'clock position, and the others follow in clockwise direction in the order of their
importance.

Using a sensitivity plot, you can answer the following important questions.

How can I make the component more reliable or improve its quality?

If the results for the reliability or failure probability of the component do not reach the expected
levels, or if the scatter of an output parameter is too wide and therefore not robust enough for a
quality product, then you should make changes to the important input variables first. Modifying an
input variable that is insignificant would be waste of time.

Of course you are not in control of all random input parameters. A typical example where you have
very limited means of control are material properties. For example, if it turns out that the
environmental temperature (outdoor) is the most important input parameter then there is probably
nothing you can do. However, even if you find out that the reliability or quality of your product is
driven by parameters that you cannot control, this has importance — it is likely that you have a
fundamental flaw in your product design! You should watch for influencial parameters like these.

If the input variable you want to tackle is a geometry-related parameter or a geometric tolerance,
then improving the reliability and quality of your product means that it might be necessary to change
to a more accurate manufacturing process or use a more accurate manufacturing machine. If it is a
material property, then there is might be nothing you can do about it. However, if you only had a few
measurements for a material property and consequently used only a rough guess about its scatter and
the material property turns out to be an important driver of product reliability and quality, then it
makes sense to collect more raw data. In this way the results of a probabilistic analysis can help you
spend your money where it makes the most sense — in areas that affect the reliability and quality of
your products the most.

How can I save money without sacrificing the reliability or the quality of the
product?

If the results for the reliability or failure probability of the component are acceptable or if the scatter
of an output parameter is small and therefore robust enough for a quality product then there is
usually the question of how to save money without reducing the reliability or quality. In this case,
you should first make changes to the input variables that turned out to be insignificant, because they
do not effect the reliability or quality of your product. If it is the geometrical properties or tolerances
that are insignificant, you can consider applying a less expensive manufacturing process. If a
material property turns out to be insignificant, then this is not typically a good way to save money,
because you are usually not in control of individual material properties. However, the loads or
boundary conditions can be a potential for saving money, but in which sense this can be exploited is
highly problem dependent.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 128 of 207

Command(s):

PDSENS

GUI:

Main Menu>Prob Design>Trends>Sensitivities

3.6.2.2. Scatter Plots

While the sensitivities point indicate which probabilistic design parameters you need to modify to
have an impact on the reliability or failure probability, scatter plots give you a better understanding
of how and how far you should modify the input variables. Improving the reliability and quality of a
product typically means that the scatter of the relevant random output parameters must be reduced.

The PDS allows you to request a scatter plot of any probabilistic design variable versus any other
one, so you can visualize the relationship between two design variables (input variables or output
parameters). This allows you to verify that the sample points really show the pattern of correlation
that you specified (if you did so). Typically, random output parameters are correlated as because they
are generated by the same set of random input variables. To support the process of improving the
reliability or quality of your product, a scatter plot showing a random output parameter as a function
of the most important random input variable can be very helpful.

When you display a scatter plot, the PDS plots the sampling points and a trendline. For this trendline,
the PDS uses a polynomial function and lets you chose the order of the polynomial function. If you
plot a random output parameter as a function of a random input variable, then this trendline
expresses how much of the scatter on the random output parameter (Y-axis) is controlled by the
random input variable (X-axis). The deviations of the sample points from the trendline are caused
and controlled by all the other random input variables. If you want to reduce the scatter of the
random output parameter to improve reliability and quality, you have two options:

? Reduce the width of the scatter of the most important random input variable(s) (that you have
control over).

? Shift the range of the scatter of the most important random input variable(s) (that you have
control over).

The effect of reducing and shifting the scatter of a random input variable is illustrated in the figures
below. "Input range before" denotes the scatter range of the random input variable before the
reduction or shifting, and "input range after" illustrates how the scatter range of the random input
variable has been modified. In both cases, the trendline tells how much the scatter of the output
parameter is affected and in which way if the range of scatter of the random input variable is
modified.

Effects of Reducing and Shifting Range of Scatter

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 129 of 207

It depends on your particular problem if either reducing or shifting the range of scatter of a random
input variable is preferable. In general, reducing the range of scatter of a random input variable leads
to higher costs. A reduction of the scatter range requires a more accurate process in manufacturing or
operating the product — the more accurate, the more expensive it is. This might lead you to conclude
that shifting the scatter range is a better idea, because it preserves the width of the scatter (which
means you can still use the manufacturing or operation process that you have). Below are some
considerations if you want to do that:

? Shifting the scatter range of a random input variable can only lead to a reduction of the scatter
of a random output parameter if the trendline shows a clear nonlinearity. If the trendline
indicates a linear trend (if it is a straight line), then shifting the range of the input variables
anywhere along this straight line doesn't make any difference. For this, reducing the scatter
range of the random input variable remains your only option.

? It is obvious from the second illustration above that shifting the range of scatter of the random
input variable involves an extrapolation beyond the range where you have data. Extrapolation
is always difficult and even dangerous if done without care. The more sampling points the
trendline is based on the better you can extrapolate. Generally, you should not go more than
30-40% outside of the range of your data. But the advantage of focusing on the important
random input variables is that a slight and careful modification can make a difference.

Note

ANSYS strongly recommends that you redo the entire probabilistic analysis using the
new and modified random input variables if you have reduced of shifted the scatter
range of any input variables using the procedure and recommendations above. To
redo the probabilistic analysis, save the probabilistic model using the PDSAVE
command and clear the current probabilistic analysis results using the PDCLR,POST
command. Then you can modify the random input variable definitions and redo the
probabilistic analysis.

Command(s):

PDSCAT

GUI:

Main Menu>Prob Design>Trends>Scatter

3.6.2.3. Correlation Matrix

Probabilistic sensitivities are based on a statistical correlation analysis between the individual

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 130 of 207

probabilistic design variables. The PDS lets you review the correlation data that has been used to
derive sensitivities and decide if individual sensitivity values are significant or not. This information
is collected in the correlation matrix of the random output parameters versus the random input
variables. The PDS also lets you review the correlations that have been sampled between random
input variables, which is stored in the random input variables correlation matrix. The correlations
between random output parameters are important if you want to use the probabilistic results of your
probabilistic analysis as input for another probabilistic analysis.

To print out a correlation matrix:

Command(s):

PDCMAT

GUI:

Main Menu>Prob Design>Trends>Correl Matrix

3.6.3. Generating an HTML Report

The ANSYS probabilistic design system automatically generates an HTML report for all
probabilistic analyses that you performed with your probabilistic model. The report explains the
problem you analyzed, which probabilistic methods were used, and the results. The report uses
pictures as well as verbal description and explanations to document all this. The report is a pre-
defined HTML document that you can modify and add to using an HTML editor. However, you can
also influence the contents of the report to some extent.

To generate an HTML report for probabilistic design:

Command(s):

PDROPT

PDWRITE

GUI:

Main Menu>Prob Design>Report>Report Options

Main Menu>Prob Design>Report>Generate Report

Use the PDROPT command to choose the items you want to include in your report, then use the
PDWRITE command to generate it.

Prev Next
Multiple Probabilistic Design
Probabilistic Design Techniques
Executions
Multiple Probabilistic Design Executions
Prev Probabilistic Design Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 131 of 207

3.7. Multiple Probabilistic Design Executions

There are various reasons why you may wish to do more than one probabilistic design execution. For
example, your initial probabilistic design run may not contain enough samples or loops to find the
desired probability limits. Or, you may start by using one probabilistic design analysis method, then
try another method to compare or validate results. The knowledge you gain from the first few loops
may prompt you to change the method you use or other analysis settings.

If you perform all executions within the same ANSYS session (or within the same batch input
stream), the procedure is very straightforward. After an execution, redefine the probabilistic design
input as desired and initiate the next execution. To initiate the execution:

Command(s):

PDEXE

GUI:

Main Menu>Prob Design>Run>-Exec Serial- Run Serial

Main Menu>Prob Design>Run>-Exec Parallel- Run Parallel

For more information on these execution options, see Executing Probabilistic Analysis Loops. If you
left the ANSYS program after performing an analysis and would like to continue your probabilistic
design analysis at some later time, you can do a save and restart as described next.

3.7.1. Saving the Probabilistic Design Database

Because the probabilistic database is completely separate from the ANSYS database, if you save the
ANSYS database using the command SAVE it does not save the probabilistic design database. An
exception is when you leave the ANSYS session and choose the "Save Everything" button in the
"EXIT from ANSYS" menu, in which case the PDS database is also saved. If you already saved the
probabilistic database to a specific file, then the "Save Everything" command will cause the current
PDS database to overwrite the data in that file; otherwise, the data is written to jobname.pds in your
current working directory.

The probabilistic design data is automatically saved at the end of each probabilistic design analysis
loop. You can save the probabilistic design data at any time by using the PDSAVE command.

To save a probabilistic design analysis:

Command(s):

PDSAVE

GUI:

Main Menu>Prob Design>Save

Caution

Currently, you must use the Save and Resume commands on the Prob Design menu to
save your work if interrupted or if you simply want to resume the project later.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 132 of 207

3.7.2. Restarting a Probabilistic Design Analysis

The probabilistic database is maintained completely separate from the ANSYS database. This means
that if you resume the ANSYS database then the probabilistic design database is not also resumed.

To restart a probabilistic design analysis, resume the probabilistic design database file
(Jobname.PDS):

Command(s):

PDRESU

GUI:

Main Menu>Prob Design>Resume

Once the data is read in, you can respecify probabilistic design type, controls, etc., and initiate
looping. (The analysis file corresponding to the resumed database must be available in order to
perform probabilistic design.) Note that the previous results are not loaded into the database until a
probabilistic post-processing command is issued.

See the /PDS, PDRESU, PDSAVE, and PDEXE command descriptions for more details.

Note

In addition to probabilistic design data, the ANSYS jobname is saved to the

probabilistic design database file ( Jobname.PDS). Therefore, when a probabilistic
design data file is resumed [PDRESU], the jobname saved in that file will overwrite
the current jobname [/FILNAME].

You can use the PDRESU command (Main Menu>Prob Design>Resume) in an interactive session
to resume probabilistic design data that was created through a batch run, thus allowing convenient
interactive viewing of batch probabilistic design results.

3.7.3. Clearing the Probabilistic Design Database

The probabilistic database is completely separate from the ANSYS database. This usually means that
saving and resuming the ANSYS database has no effect on saving and resuming the PDS database.
However, if you issue the /CLEAR command to clear the ANSYS database, then the PDS database
is also cleared. For this reason you should never use the /CLEAR command as part of your PDS
analysis file. Clearing the PDS database has no effect on the ANSYS database.

For clearing the probabilistic database you have two options. You can clear the entire PDS database
or just the results and solution portion. The latter maintains the probabilistic model definition (the
random input variable definitions, correlations, and random output parameter definitions). Clearing
only the solution and results part of the PDS database is helpful if you want to modify the
probabilistic model (change RVs, set correlations between RVs or RPs, etc.). After you have
performed a probabilistic analysis, the PDS will not allow you to change any of the probabilistic
model details. This ensures consistency of the probabilistic model with the results.

To clear the probabilistic design database:

Command(s):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 133 of 207

PDCLR

GUI:

Main Menu>Prob Design>Clear & Reset

Because the ANSYS database is not affected by the PDCLR command, it may also be necessary to
clear the ANSYS database if the resumed probabilistic design problem is totally independent of the
previous one. See the /CLEAR command for details.

Prev Next
Post-Processing Probabilistic Analysis
Sample Probabilistic Design Analysis
Results
Sample Probabilistic Design Analysis
Prev Probabilistic Design Next

3.8. Sample Probabilistic Design Analysis

In the following example, you will perform a probabilistic design analysis of a simple indeterminate
three-bar truss. This analysis uses the Direct Monte Carlo method.

3.8.1. Problem Description

You will analyze the effect of variation in the 3-bar truss cross-section that is subjected to a force
with both vertical and horizontal components.

3.8.2. Problem Specifications

The loading for this example is a force of 20000 lbs. located at the vertex of the three truss members.
The force acts at a 45-degree angle from the vertical direction.

The following material property is used for this analysis:

Young's modulus (E) = 30E6 psi

The following geometric properties are used for this analysis. These properties are the initial cross-
sectional areas for each truss member:

ARE1 = 5.0
ARE2 = 5.0
ARE3 = 5.0

3.8.2.1. Problem Sketch

The Simple Indeterminate 3-Bar Truss for the Sample Problem

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 134 of 207

3.8.3. Using a Batch File for the Analysis

You can perform the example probabilistic design analysis of this 3-bar truss using the ANSYS
commands shown below.

The analysis file is created for use during the probabilistic analysis. It is a parametric model of the
problem geometry, materials, and loads. Within the analysis file, input variables are initialized and
output variables are retrieved. If you prefer, you can perform the second pass of the example analysis
using the GUI method rather than ANSYS commands. See Using the GUI for the PDS Analysis for
details.

The following input listing sets up the analysis file for the 3-bar truss.

/com
/com, Create an analysis file to be used during looping
/com
*create,pds3bar,pdan
*SET,ARE1,5.00 !INITIALIZE CROSS SECTIONAL AREAS
*SET,ARE2,5.00
*SET,ARE3,5.00
/PREP7
/title, PROBABILISTIC ANALYSIS OF A SIMPLE INDETERMINATE 3-BAR TRUSS
ET,1,1
EX,1,30E6
R,1,ARE1
R,2,ARE2
R,3,ARE3
N,1
N,2,10
N,3,20
N,4,10,-10,,-45 ! ROTATE TIPNODE 45 DEGREES
REAL,1

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 135 of 207

E,1,4
REAL,2
E,2,4
REAL,3
E,3,4
D,1,ALL,,,3
F,4,FX,20000
FINISH
/SOLU
SOLVE
FINISH
/POST1
SET,1
ETABLE,VOLU,VOLU ! STORE MEMBER VOLUMES
ETABLE,AXST,LS,1 ! STORE MEMBER AXIAL STRESSES
*GET,SIG1,ELEM,1,ETAB,AXST ! SIG1 IS DEFINED TO BE AXIAL STRESS IN ELEMENT 1
*GET,SIG2,ELEM,2,ETAB,AXST ! SIG2 IS DEFINED TO BE AXIAL STRESS IN ELEMENT 2
*GET,SIG3,ELEM,3,ETAB,AXST ! SIG3 IS DEFINED TO BE AXIAL STRESS IN ELEMENT 3
SSUM
*GET,TVOL,SSUM,,ITEM,VOLU
FINI
*end

After the analysis file has been created, you can proceed with the probabilistic design analysis. You
can do this though ANSYS commands or though the GUI. If you prefer using commands, the
following input sets up the probabilistic analysis for the 3-bar truss example.

/inp,pds3bar,pdan
/com
/com, Enter PDS and specify the analysis file
/com
/PDS ! enter probabilistic design system
pdanl,pds3bar,pdan
/com
/com, Declare random input variables
/com
PDVAR,ARE1,GAUS,5,0.5 ! define area1 with Gaussian distribution
! having mean of 5 and std. dev of 0.5
PDVAR,ARE2,tria,10,11,12 ! define area2 with triangular distribution
! having low bound of 10, most likely point of 11
! and high bound of 12
PDVAR,ARE3,unif,5,6 ! define area3 with uniform distribution
! with low bound of 5 and high bound of 6
/com
/com, Specify any correlations between the random variables
/com
PDCOR,ARE1,ARE3,0.25 ! define a correlation coef of 0.25 between ARE1 and ARE3
/com
/com, Declare random output variables
/com
PDVAR,SIG1,resp ! define SIG1 a response parameter
PDVAR,SIG2,resp ! define SIG2 a response parameter
PDVAR,SIG3,resp ! define SIG3 a response parameter
PDVAR,TVOL,resp ! define TVOL a response parameter
/com
/com, Choose the probabilistic design tool or method
/com
PDMETH,MCS,DIR ! specify direct Monte Carlo simulation
PDDMCS,100,NONE,ALL,,,,123457 ! use all 100 samples, initial seed of 123457
/com
/com, Execute the loops required for the probabilistic design analysis
/com
PDEXE,mcs3bar ! run analysis and define solution label 3bar_mcs

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 136 of 207

/com
/com, Review the results of the probabilistic analysis
/com
PDSENS,MCS3BAR,TVOL,BOTH,RANK,0.025 !Create Sensitivity plot
fini
/exit

3.8.4. Using the GUI for the PDS Analysis

Because of the parametric definition of some variables in the analysis file, it is recommended that it
be created through ANSYS. Thus, the GUI method is not recommended for the analysis file creation
and will not be presented here. However, it is acceptable to perform the probabilistic design analysis
of the 3-bar truss example using the GUI method instead of the ANSYS commands shown earlier.
The GUI procedure for performing the probabilistic design analysis follows. Using a Batch File for
the Analysis describes how to create an analysis file. The GUI procedure for performing the
probabilistic design analysis pass follows.

Step 1: Test Analysis File

To test the analysis file, you clear the database and then read input from the pdbeam.lgw file.

1. Choose menu path Utility Menu>File>Clear & Start New. Click on OK.

2. In the Verify dialog box, click Yes.

3. Change the jobname. Choose menu path Utility Menu>File>Change Jobname. The Change
Jobname dialog box appears.

4. Change the jobname to pds3bar.pdan and click on OK.

5. Choose menu path Utility Menu>File>Read Input from. In the Files list, click on
pds3bar.pdan. Then click on OK. You see a replay of the entire analysis. Click on Close
when the “Solution is done!” message appears.

In the next several steps of this problem, you explore the effects of variation in the parameters.

Step 2: Enter the Probabilistic Design Module and Identify Analysis File

First, enter the optimizer and identify the analysis file.

1. Choose menu path Main Menu>Prob Design>-Looping File-Assign. The Assign Analysis File
dialog box appears.

2. In the Files list, click once on pds3bar.pdan and then click on OK.

Step 3: Identify the Probabilistic Design Variables

1. Choose menu path Main Menu>Design Opt>Design Variables. The Design Variables dialog
box appears.

2. Click on Add. The Define a Random Variable dialog box appears.

3. In the list of parameter names, click on ARE1. Select GAUSS in the distribution type list.
Click on OK.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 137 of 207

4. Type 5 in the MEAN VALUE field and 0.5 in the STANDARD DEVIATION field. Click on
OK.

5. Click on Add. The Define a Random Variable dialog box appears.

6. In the list of parameter names, click on ARE2. Select TRIANGULAR in the distribution type
list. Click on OK.

7. Type 10 in the LOWER BOUNDARY field, 11 in the MOST LIKELY VALUE field, and 12
in the UPPER BOUNDARY field. Click on OK.

8. Click on Add. The Define a Random Variable dialog box appears.

9. In the list of parameter names, click on ARE3. Select UNIFORM in the distribution type list.
Click on OK.

10. Type 5 in the LOW BOUND field, and 6 in the HIGH BOUND field. Click on OK.

11. Click on Close to close the Define a Random Variable dialog box.

12. Choose menu path Main Menu>Prob Design>Prob Definitions>Correlation. The Add/Edit or
Delete Correlation dialog box appears.

13. Select ARE1 and ARE2 from the list of variable names. Click on OK. Type 0.25 in the
Correlation Coeff field.

14. Choose menu path Main Menu>Prob Design>Prob Definitions>Random Output. The Random
Output Parameters dialog box appears.

15. Click on Add. The Define a Random Output Parameter dialog box appears. In the list of
parameter names, click on SIG1. Click on OK.

16. Click on Add. The Define a Random Output Parameter dialog box appears. In the list of
parameter names, click on SIG2. Click on OK.

17. Click on Add. The Define a Random Output Parameter dialog box appears. In the list of
parameter names, click on SIG3. Click on OK.

18. Click on Close to close the Random Output Parameters dialog box.

Step 4: Run the Probabilistic Design Analysis

This step involves specifying the probabilistic design method, executing the run, and saving the data.

1. Choose menu path Main Menu>Prob Design>Prob Method>Monte Carlo Sims. The Monte
Carlo Simulation dialog box appears.

2. Select Direct Sampling from the Sampling Methods.

3. Type 100 in the Number of Simulations field. Choose Use 123457 from the Random Seed
Option. Click on OK.

4. Choose menu path Main Menu>Prob Design>Run>Exec Serial>Run Serial. Type mcs3bar in

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 138 of 207

the Solution Set Label field. Click on OK.

5. The Run Monte Carlo Simulations dialog box appears. Click on OK.

6. Choose menu path Main Menu>Prob Design>Prob Database>Save. Type mcs3bar in the
selection field. Click on OK.

When the data is saved, you are ready to review the results.

Step 5: Review the Probabilistic Results

In this step, you visualize the probabilistic results.

1. Choose menu path Main Menu>Prob Design>Prob Results>Trends>Sensitivities. The

Sensitivity of a Response Parameter dialog box appears.

2. Select MCS3BAR in the Select Results Set field. Select TVOL in the Select Response Parameter
field. Click on OK.

3. A Rank-Order Correlation Sensitivity bar and pie chart is shown in the ANSYS Graphics
window.

Step 7: Exit ANSYS

1. Click on Quit in the ANSYS Toolbar.

2. Select an option to save, then click on OK.

Prev Next
Multiple Probabilistic Design
Adaptive Meshing
Executions
Adaptive Meshing
Prev Advanced Analysis Techniques Guide Next

Chapter 4. Adaptive Meshing

4.1. What Is Adaptive Meshing?
The ANSYS program provides approximate techniques for automatically estimating mesh
discretization error for certain types of analyses. (See The General Postprocessor (POST1) in the
ANSYS Basic Analysis Guide for more information about error-value approximation.) Using this
measure of mesh discretization error, the program can then determine if a particular mesh is fine
enough. If it is not, the program will automatically refine the mesh so that the measured error will
decrease. This process of automatically evaluating mesh discretization error and refining the mesh,
called adaptive meshing, continues through a series of solutions until the measured error drops below
some user-defined value (or until a user-defined limit on the number of solutions is reached).

Prev Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 139 of 207

Sample Probabilistic Design Analysis Prerequisites for Adaptive Meshing

Prerequisites for Adaptive Meshing
Prev Adaptive Meshing Next

4.2. Prerequisites for Adaptive Meshing

The ANSYS program includes a prewritten macro, ADAPT.MAC, to perform adaptive meshing. Your
model must satisfy certain preconditions before you can successfully activate the ADAPT macro. (In
some cases, models that do not conform to these criteria can be adaptively meshed using a modified
procedure, as discussed below.) These requirements include the following:

? The standard ADAPT procedure is valid only for single-solution linear static structural and
linear steady-state thermal analyses.

? Your model should preferably use only one material type, as the error calculations are based in
part on average nodal stresses, and would thus often be invalid at the material interfaces. Also,
an element's error energy is affected by its elastic modulus. Therefore, even if the stress
discontinuity is the same in two adjoining elements, their error energy will be different if they
have different material properties. You should also avoid abrupt changes in shell thickness, as
such discontinuities will cause similar stress-averaging problems.

? Your model must use element types that support error calculations. (See Element Types That
can be Used in Adaptive Meshing for a list of valid element types.)

? You must build your model using meshable solid model entities: that is, characteristics that
will cause meshing failure must not be built into your model.

Element Types That Can Be Used in Adaptive Meshing

Type Element Description

PLANE2 2-D 6-Node Triangular Solid
PLANE25 Axisymmetric Harmonic Solid
2-D Structural Solids PLANE42 2-D 4-Node Isoparametric Solid
PLANE82 2-D 8-Node Solid
PLANE83 Axisymmetric Harmonic 8-Node Solid
SOLID45 3-D 8-Node Isoparametric Solid
SOLID64 3-D Anisotropic Solid
3-D Structural Solids
SOLID92 3-D 10-Node Tetrahedral Solid
SOLID95 3-D 20-Node Isoparametric Solid
SHELL43 Plastic Quadrilateral Shell
3-D Structural Shells SHELL63 Elastic Quadrilateral Shell
SHELL93 8-Node Isoparametric Shell

PLANE35 2-D 6-Node Triangular Solid

PLANE75 Axisymmetric Harmonic Solid
2-D Thermal Solids
PLANE55 2-D 4-Node Isoparametric Solid
PLANE77 2-D 8-Node Solid

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 140 of 207

PLANE78 Axisymmetric Harmonic 8-Node Solid

SOLID70 3-D 8-Node Isoparametric Solid
3-D Thermal Solids SOLID87 3-D 10-Node Tetrahedral Solid
SOLID90 3-D 20-Node Isoparametric Solid
3-D Thermal Shells SHELL57 Plastic Quadrilateral Shell

Prev Next
How to Use Adaptive Meshing: Basic
Adaptive Meshing
Procedure
How to Use Adaptive Meshing: Basic Procedure
Prev Adaptive Meshing Next

4.3. How to Use Adaptive Meshing: Basic Procedure

The basic procedure for running the adaptive meshing macro follows these steps:

1. As in any linear static structural or steady state thermal analysis, first enter the preprocessor
(/PREP7 command or menu path Main Menu>Preprocessor). Then, specify the element type,
real constants, and material properties, in accordance with the prerequisites listed above.

2. Model your system using solid modeling procedures, creating meshable areas or volumes that
describe the geometry of your system. You do not need to specify element sizes, nor do you
need to mesh these areas and volumes - the ADAPT macro will automatically initiate meshing
for you. (If you need to mesh your model with both area and volume elements, create an
ADAPTMSH.MAC user subroutine - see below.)

3. You can either proceed to SOLUTION (/SOLU or menu path Main Menu>Solution) or remain
in PREP7 to specify analysis type, analysis options, loads, and load step options. Apply only
solid model loads and inertia loads (accelerations, rotational accelerations, and rotational
velocities) in a single load step. (Finite element loads, coupling, and constraint equations can
be introduced through the ADAPTBC.MAC user subroutine. Multiple load steps can be introduced
through the ADAPTSOL.MAC subroutine. The subroutines are discussed later in this chapter.)

4. If in PREP7, exit the preprocessor [FINISH]. (You can invoke the ADAPT macro from either
SOLUTION or the Begin level.)

5. Invoke the adaptive solution. To do so, use one of these methods:

Command(s):

ADAPT

GUI:

Main Menu>Solution>Adaptive Mesh

Notice that you can use the ADAPT macro in either a thermal or a structural analysis, but that
you cannot mix the two disciplines in one adaptive solution. As the adaptive meshing
iterations proceed, element sizes will be adjusted (within the limits set by FACMN and FACMX ) to
decrease and increase the elemental error energies until the error in energy norm matches the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 141 of 207

target value (or until the specified maximum number of solutions has been used).

After you have invoked the adaptive solution, this macro controls all program operations until
the solution is completed. The ADAPT macro will define element sizes, generate the mesh,
solve, evaluate errors, and iterate as necessary till the target value of error in energy norm is
met. All these steps are performed automatically, with no further input required from you.

6. Once adaptive meshing has converged, the program automatically turns element shape
checking on [SHPP,ON]. It then returns to the SOLUTION phase or to the Begin level,
depending on which phase you were in when you invoked ADAPT. You may then enter
POST1 and postprocess as desired, using standard techniques.

Prev Next
Prerequisites for Adaptive Meshing Modifying the Basic Procedure
Modifying the Basic Procedure
Prev Adaptive Meshing Next

4.4. Modifying the Basic Procedure

4.4.1. Selective Adaptivity

If you know that mesh discretization error (measured as a percentage) is relatively unimportant in
some regions of your model (for instance, in a region of low, slowly-changing stress), you can speed
up your analysis by excluding such regions from the adaptive meshing operations. Also, you might
want to exclude regions near singularities caused by concentrated loads. Selecting logic provides a
way of handling such situations.

Selective Adaptivity

Selective adaptivity can improve the performance of models having concentrated loads.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 142 of 207

If you select a set of keypoints, the ADAPT macro will still include all your keypoints (that is, the
ADAPT macro will modify the mesh at both selected and non-selected keypoints), unless you also
set KYKPS = 1 in the ADAPT command (Main Menu>Solution>Adaptive Mesh).

If you select a set of areas or volumes, the ADAPT macro will adjust element sizes only in those
regions that are in the selected set. You will have to mesh your entire model in PREP7 before
executing ADAPT.

4.4.2. Customizing the ADAPT Macro with User Subroutines

The standard ADAPT macro might not always be applicable to your particular analysis needs. For
instance, you might need to mesh both areas and volumes, which is not possible with the standard
macro. For this and other such situations, you can modify the ADAPT macro to suit your analysis
needs. By using a macro to perform the adaptive meshing task, we have intentionally given you
access to the logic involved, and have thereby furnished you with the capability for modifying the
technique as you might desire.

Fortunately, you do not always need to change the coding within the ADAPT macro to customize it.
Three specific portions of the macro can be readily modified by means of user subroutines, which
are separate user files that you can create and that will be called by the ADAPT macro. The three
features that can be modified by user subroutines are:

? the meshing command sequence

? the boundary condition command sequence,

? the solution command sequence

The corresponding user subroutine files must be named ADAPTMSH.MAC, ADAPTBC.MAC, and
ADAPTSOL.MAC, respectively.

4.4.2.1. Creating a Custom Meshing Subroutine (ADAPTMSH.MAC)

By default, if your model contains one or more selected volumes, the ADAPT macro will mesh only
volumes (no area meshing will be done). If you have no selected volumes, then the ADAPT macro
will mesh only areas. If you desire to mesh both volumes and areas, you can create a user subroutine,
ADAPTMSH.MAC, to perform all the desired operations. You will need to clear any specially-meshed
entities before remeshing. Such a subroutine might look like this:

C*** Subroutine ADAPTMSH.MAC - Your name - Job Name - Date Created

TYPE,1 ! Set element TYPE attribute for area meshing
ACLEAR,3,5,2 ! Clear areas and volumes to be meshed by this subroutine
VCLEAR,ALL
AMESH,3,5,2 ! Mesh areas 3 and 5 (no other areas will be meshed by ADAPT)
TYPE,2 ! Change element type for volume mesh
VMESH,ALL ! Mesh all volumes

Please see the TYPE, ACLEAR, VCLEAR, AMESH, and VMESH command descriptions for
more information.

We strongly recommend that you include a comment line (C***) to identify your macro uniquely.
This comment line will be echoed in the job printout, and will provide assurance that the ADAPT
macro has used the correct user subroutine.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 143 of 207

4.4.2.2. Creating a Custom Subroutine for Boundary Conditions (ADAPTBC.MAC)

The ADAPT macro clears and remeshes with every new solution loop. As a result, your model's
nodes and elements will be repeatedly changing. This situation generally precludes the use of finite
element loads, DOF coupling, and constraint equations, all of which must be defined in terms of
specific nodes and elements. If you need to include any of these finite-element-supported items, you
can do so by writing a user subroutine, ADAPTBC.MAC. In this subroutine, you can select nodes by
their location, and can then define finite element loads, DOF coupling, and constraint equations for
the selected nodes. A sample ADAPTBC.MAC subroutine follows:

C*** Subroutine ADAPTBC.MAC - Your name - Job Name - Date Created

NSEL,S,LOC,X,0 ! Select nodes @ X=0.0
D,ALL,UX,0 ! Specify UX=0.0 for all selected nodes
NSEL,S,LOC,Y,0 ! Select nodes @ Y=0.0
D,ALL,UY,0 ! Specify UY=0.0 for all selected nodes
NSEL,ALL ! Select all nodes

4.4.2.3. Creating a Custom Solution Subroutine (ADAPTSOL.MAC)

The default solution command sequence included in the ADAPT macro is simply:

/SOLU
SOLVE
FINISH

This default sequence will solve only a single load step. You might be able to implement other
solution sequences by incorporating them into the user subroutine ADAPTSOL.MAC.

4.4.2.4. Some Further Comments on Custom Subroutines

You can create these subroutines as you would any user file. In other words, you can use either the
APDL command *CREATE (Utility Menu>Macro>Create Macro), the APDL command *END, or
an external text editor. When the ADAPT macro calls these subroutines, the program will search
first through the ANSYS root directory, then through your root directory, and last through the current
directory. Thus, you should take care to ensure that no identically-named files from another directory
will be read in place of your intended files. The comment lines (C***) shown in the example
subroutines above would be echoed in your printout, and would provide one means of checking that
the proper files were used. Additionally, executing /PSEARCH,OFF (Utility Menu>Macro>Macro
Search Path) before running the ADAPT macro would restrict the file search to the ANSYS root
directory and your current directory, reducing somewhat the possibility that the wrong file will be
used. No matter where they reside, these subroutines will be accessed only if you set KYMAC = 1 on
the ADAPT command (Main Menu>Solution>Adaptive Mesh). See the ANSYS Guide to User
Programmable Features for more information on user subroutines and the ANSYS APDL
Programmer's Guide for more information on APDL.

4.4.3. Customizing the ADAPT Macro (UADAPT.MAC)

For those instances when you need to modify the ADAPT macro but are unable to do so through
separate user subroutines, you can modify the main body of the macro. However, for a variety of
reasons, we do not recommend that modifications be made directly to the macro file ADAPT.MAC.
(For one thing, if other users access the same copy of the ANSYS program as you, these other users
might become upset if they invoked ADAPT and found out that you had changed its logic!)
Consequently, an identical copy of this macro file, UADAPT.MAC, is supplied on the ANSYS
installation media for the specific purpose of user modification to suit a particular need.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 144 of 207

If modifications are made to UADAPT.MAC, we suggest that you rename the modified file to denote the
particular version you have created. Then, instead of issuing the command ADAPT, you will call the
modified adaptive meshing procedure by issuing the new filename. Be aware that if you use the new
filename as an "unknown command," the program will first search through the high-level directory,
then through the login directory, and finally through the working directory, until the macro is found.
If a modified adaptive procedure is to be accessed by only a single user, then it makes sense to store
this file in a directory no higher than the user's login directory. If the macro file is stored in such a
low-level directory, the file search can be streamlined by calling the macro using the *USE
command (Utility Menu>Macro>Execute Data Block) in place of the unknown command format.

Prev Next
How to Use Adaptive Meshing: Basic
Guidelines for Adaptive Meshing
Procedure
Guidelines for Adaptive Meshing
Prev Adaptive Meshing Next

4.5. Guidelines for Adaptive Meshing

The following suggestions might improve your implementation of adaptive meshing:

? No initial element sizes are required, but they may be specified to aid the adaptive
convergence when desired. If you specify initial element sizes at keypoints, the ADAPT
macro will use these sizes in its first loop, and will adjust these sizes as appropriate in
subsequent loops. To specify initial element sizes, use one of these methods:
Command(s):

KESIZE

GUI:
Main Menu>Preprocessor>-Meshing-Size Cntrls>-Keypoints-All KPs
Main Menu>Preprocessor>-Meshing-Size Cntrls>-Keypoints-Picked KPs

If you specify line divisions or spacing ratios, the ADAPT macro will use these values in every loop.
(That is, line divisions or spacing ratios that you specify will not be changed by the ADAPT macro.)
If you do not specify mesh divisions of any kind, default element sizing [SMRTSIZE and DESIZE]
will be used for the initial mesh. To specify line divisions or spacing ratios, use one of these
methods:

Command(s):

LESIZE

GUI:
Main Menu>Preprocessor>-Meshing-Size Cntrls>-Lines-All Lines
Main Menu>Preprocessor>-Meshing-Size Cntrls>-Lines-Picked Lines

? Mapped meshing (all quadrilaterals) is available for 2-D solid and 3-D shell analyses. The
benefits of using mapped area meshing are usually minimal, however.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 145 of 207

? Mapped meshing (all hexahedral bricks) is available for 3-D solid analyses. Mapped volume
meshing, if possible for a given model, will probably give superior performance, compared to
tetrahedral meshing.

? In general, midside-noded elements will perform better than linear elements in an adaptive
meshing operation.

? Do not use concentrated loads or other singularities (such as sharp re-entrant corners), because
the ADAPT procedure cannot show energy value convergence with mesh refinement in the
vicinity of these singularities. If a concentrated loading condition is present in your model,
replace it with an equivalent pressure load, applied over a small area. (Or, exclude the region
of the concentrated load from adaptive modification using the select options discussed
previously.)

? For many problems, it could be preferable to use a number of relatively small, simple regions
in place of fewer large, complicated regions, for best meshing performance.

? If the location of maximum response is known or suspected beforehand, then a keypoint

should be placed near that location.

? If the ADAPT procedure is used in an interactive run and the ANSYS program aborts abruptly
without issuing the proper error message, then the output file for the adaptive meshing portion
of your run (Jobname.ADPT) should be reviewed to determine the cause.

? Similarly, if the ADAPT procedure is used in a batch run, then Jobname.ADPT should be saved
and examined in case of an abrupt abort.

? If a model has an excessive curvature in some region, then your model might experience mesh
failure. In this case, use the SIZE field on the KESIZE command (Main Menu>Preprocessor>-
Meshing-Size Cntrls>-Keypoints-Picked KPs) to define the maximum element edge length
settings at keypoints near the excessive curvature region. FACMX should also be set equal to 1
(in the ADAPT command) so that element size will not be permitted to grow in the vicinity of
the excessive curvature.

? You should save the results file ( Jobname.RST or Jobname.RTH). In case of a program abort in
the middle of the ADAPT procedure, the results file will still have the entire analysis data
from the previous solution completed by the ADAPT procedure.

? You should issue a SAVE (Utility Menu>File>Save as Jobname.db) before starting the
adaptive meshing operation. In case of system abort, Jobname.DB can then be used to restart
the analysis.

Prev Next
Modifying the Basic Procedure Where to Find Examples
Where to Find Examples
Prev Adaptive Meshing Next

4.6. Where to Find Examples

The ANSYS Verification Manual describes several analyses that demonstrate adaptive meshing.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 146 of 207

The ANSYS Verification Manual consists of test case analyses demonstrating the analysis capabilities
of the ANSYS program. While these test cases demonstrate solutions to realistic analysis problems,
the ANSYS Verification Manual does not present them as step-by-step examples with lengthy data
input instructions and printouts. However, most ANSYS users who have at least limited finite
element experience should be able to fill in the missing details by reviewing each test case's finite
element model and input data with accompanying comments.

The ANSYS Verification Manual contains the following adaptive meshing test cases:

VM193 - Adaptive Analysis of Two-dimensional Heat Transfer with Convection

VM205 - Adaptive Analysis of an Elliptic Membrane Under a Uniformly Distributed Load

Prev Next
Guidelines for Adaptive Meshing Submodeling
Submodeling
Prev Advanced Analysis Techniques Guide Next

Chapter 5. Submodeling
5.1. What Is Submodeling?
Submodeling is a finite element technique used to get more accurate results in a region of your
model. Often in finite element analysis, the finite element mesh may be too coarse to produce
satisfactory results in a region of interest, such as a stress concentration region in a stress analysis
(see Submodeling of a Pulley (a). The results away from this region, however, may be adequate.

To obtain more accurate results in such a region, you have two options: (a) reanalyze the entire
model with greater mesh refinement, or (b) generate an independent, more finely meshed model of
only the region of interest and analyze it. Obviously, option (a) can be time-consuming and costly
(depending on the size of the overall model). Option (b) is the submodeling technique (see
Submodeling of a Pulley (b).

Submodeling of a Pulley

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 147 of 207

Submodeling of a pulley hub and spokes: (a) coarse model, (b) submodel shown
superimposed over coarse model

Submodeling is also known as the cut-boundary displacement method or the specified boundary
displacement method. The cut boundary is the boundary of the submodel which represents a cut
through the coarse model. Displacements calculated on the cut boundary of the coarse model are
specified as boundary conditions for the submodel.

Submodeling is based on St. Venant's principle, which states that if an actual distribution of forces is
replaced by a statically equivalent system, the distribution of stress and strain is altered only near the
regions of load application. This implies that stress concentration effects are localized around the
concentration; therefore, if the boundaries of the submodel are far enough away from the stress
concentration, reasonably accurate results can be calculated in the submodel.

The ANSYS program does not restrict submodeling to structural (stress) analyses only. Submodeling
can be used effectively in other disciplines as well. For example, in a magnetic field analysis, you
can use submodeling to calculate more accurate magnetic forces in a region of interest.

Aside from the obvious benefit of giving you more accurate results in a region of your model, the
submodeling technique has other advantages:

? It reduces, or even eliminates, the need for complicated transition regions in solid finite
element models.

? It enables you to experiment with different designs for the region of interest (different fillet
radii, for example).

? It helps you in demonstrating the adequacy of mesh refinements.

Some restrictions for the use of submodeling are:

? It is valid only for solid elements and shell elements.

? The principle behind submodeling assumes that the cut boundaries are far enough away from
the stress concentration region. It is up to you to verify that this assumption is adequately
satisfied.

Prev Next
Where to Find Examples How to Do Submodeling
How to Do Submodeling
Prev Submodeling Next

5.2. How to Do Submodeling

The procedure for submodeling consists of the following steps:

1. Create and analyze the coarse model.

2. Create the submodel.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 148 of 207

3. Perform cut boundary interpolation.

4. Analyze the submodel.

5. Verify that the distance between the cut boundaries and the stress concentration is adequate.

5.2.1. Create and Analyze the Coarse Model

The first step is to model the entire structure and analyze it.

Note

To easily identify this initial model, we will refer to it as the coarse model. This does
not mean that the mesh refinement has to be coarse, only that it is relatively coarse
compared to the submodel.

The analysis type may be static (steady-state) or transient and follows the same procedure as
described in the individual analysis guides. Some additional points to keep in mind are listed below.

Jobname-You should use different jobnames for the coarse model and the submodel. This way, you
can keep files from being overwritten. Also, you can easily refer to files from the coarse model
during cut boundary interpolation. To specify a jobname, use one of these methods:

Command(s):

/FILNAME

GUI:

Utility Menu>File>Change Jobname

Element Types - Only solid and shell elements support the submodeling technique. Your analysis
may include other element types (such as beams added as stiffeners), but the cut boundary should
only pass through the solids or shells.

A special submodeling technique, called shell-to-solid submodeling, allows you to build your coarse
model with shell elements and your submodel with 3-D solid elements. This technique is discussed
in Shell-to-Solid Submodels.

Modeling - In many cases, the coarse model need not include local details such as fillet radii, as
shown in the following figure. However, the finite element mesh must be fine enough to produce a
reasonably accurate degree-of-freedom solution. This is important because the results of the
submodel are almost entirely based on interpolated degree-of-freedom results at the cut boundary.

Coarse Model

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 149 of 207

Initial, coarse model may not need to include many details

Files - Both the results file ( Jobname.RST, Jobname.RMG , etc.) and the database file (Jobname.DB,
containing the model geometry) are required from the coarse-model analysis. Be sure to save the
database before going on to create the submodel. To save the database, use one of these methods:

Command(s):

SAVE

GUI:
Utility Menu>File>Save as
Utility Menu>File>Save as Jobname.db

5.2.2. Create the Submodel

The submodel is completely independent of the coarse model. Therefore, the first step after the initial
analysis is to clear the database at the Begin level. (Another way is to leave and re-enter the ANSYS
program.) To clear the database at the Begin level, use one of these methods:

Command(s):

/CLEAR

GUI:

Utility Menu>File>Clear & Start New

Also, be sure to use a different jobname for the submodel so that the coarse-model files are not
overwritten. To specify a jobname, use one of these methods:

Command(s):

/FILNAME

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 150 of 207

GUI:

Utility Menu>File>Change Jobname

Then enter PREP7 and build the submodel. Some points to remember are:

? Use the same element type (solid or shell) that was used in the coarse model. Also, specify the
same element real constants (such as shell thickness) and material properties. (Another type of
submodeling - shell-to-solid submodeling - allows you to switch from shell elements in the
coarse model to 3-D solid elements in the submodel; see 3D Solid Submodel Superimposed on
Coarse Shell Model.)

? The location of the submodel (with respect to the global origin) must be the same as the
corresponding portion of the coarse model, as shown in Submodel Superimposed Over Coarse
Model.

Submodel Superimposed Over Coarse Model

? Specify appropriate node rotations. Node rotation angles on cut boundary nodes should not be
changed after they have been written to the node file in interpolation step 1 (see Perform Cut
Boundary Interpolation). To specify node rotations, use one of these methods:
Command(s):

NROTAT

GUI:
Main Menu>Preprocessor>Create>Nodes>-Rotate Node CS-To Active CS
Main Menu>Preprocessor>Move / Modify>-Rotate Node CS-To Active CS

Be aware that node rotation angles might be changed by application of nodal constraints [DSYM],
by transfer of line constraints [SFL], or by transfer of area constraints [SFA], as well as by more
obvious methods [NROTAT and NMODIF].

The presence or absence of node rotation angles in the coarse model has no effect upon the
submodel.

Loads and boundary conditions for the submodel will be covered in the next two steps.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 151 of 207

5.2.3. Perform Cut Boundary Interpolation

This is the key step in submodeling. You identify the nodes along the cut boundaries, and the
ANSYS program calculates the DOF values (displacements, potentials, etc.) at those nodes by
interpolating results from the full (coarse) model. For each node of the submodel along the cut
boundary, the ANSYS program uses the appropriate element from the coarse mesh to determine the
DOF values. These values are then interpolated onto the cut boundary nodes using the element shape
functions.

The following tasks are involved in performing the cut boundary interpolation:

1. Identify and write the cut boundary nodes of the submodel to a file ( Jobname.NODE by
default). You can do this in PREP7 by selecting nodes along the cut boundaries and then using
one of these methods to write the nodes to a file:

Command(s):

NWRITE

GUI:

Main Menu>Preprocessor>Create>Nodes>Write Node File

Here is an example of issuing the NWRITE command:

NSEL,... ! Select nodes along cut boundaries

NWRITE ! Writes all selected nodes to Jobname.NODE

Cut Boundaries on the Submodel

At this point, it may be worthwhile to discuss temperature interpolation. In an analysis with

temperature-dependent material properties, or in a thermal-stress analysis, the temperature
distribution must be the same between the coarse model and the submodel. For such cases, you
must also interpolate the temperatures from the coarse model to all nodes in the submodel. To
do this, select all submodel nodes and write them to a different file using
NWRITE,Filename,Ext. Be sure to specify a file name; otherwise, your file of cut boundary
nodes will be overwritten! Step 7 shows the command to do temperature interpolation.

2. Restore the full set of nodes, write the database to Jobname.DB, and leave PREP7. You must
write the database to Jobname.DB because you need to continue with the submodel later.

To restore the full set of nodes, use one of these methods:

Command(s):

ALLSEL

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 152 of 207

GUI:

Utility Menu>Select>Everything

To write the database to Jobname.DB, use one of these methods:

Command(s):

SAVE

GUI:

Utility Menu>File>Save as Jobname.db

3. To do the cut boundary interpolation (and the temperature interpolation), the database must
contain the geometry for the coarse model. Therefore, you must resume the database using one
of the methods shown below, making sure to identify the name of the coarse-model database
file:

Command(s):

RESUME

GUI:

Utility Menu>File>Resume from

For example, if the jobname for the coarse-model analysis was COARSE, issue the command
RESUME,COARSE,DB.

4. Enter POST1, which is the general postprocessor (/POST1 or menu path Main Menu>General
Postproc). Interpolation can only be performed in POST1.

5. Point to the coarse results file (FILE or menu path Main Menu>General Postproc>Data & File
Opts).

6. Read in the desired set of data from the results file (SET or menu path Main Menu>General
Postproc>-Read Results-option).

7. Initiate cut boundary interpolation. To do so, use one of these methods:

Command(s):

CBDOF

GUI:

Main Menu>General Postproc>Submodeling>Interpolate DOF

By default, the CBDOF command assumes that the cut boundary nodes are on file
Jobname.NODE. The ANSYS program will then calculate the cut boundary DOF values and
write them in the form of D commands to the file Jobname.CBDO.

To do temperature interpolation, use one of these methods, being sure to identify the name of
the file containing all submodel nodes:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 153 of 207

Command(s):

BFINT

GUI:

Main Menu>General Postproc>Submodeling>Interp Body Forc

Interpolated temperatures are written in the form of BF commands to the file

Jobname.BFIN.

Note

If real and imaginary data are involved, steps 6 and 7 will need to be done
twice. First issue the SET command to get the real data, followed by the
interpolation step [CBDOF and/or BFINT]. Then issue the SET command
with the field set to 1 to get the imaginary data, and repeat the interpolation
step, this time writing the interpolated imaginary data to a different file.

8. All interpolation work is now done, so leave POST1 [FINISH] and restore the submodel
database (RESUME or menu path Utility Menu>File>Resume from). (Be sure to use the
submodel database jobname.)

5.2.4. Analyze the Submodel

In this step, you define the analysis type and analysis options, apply the interpolated DOF values
(and temperatures), define other loads and boundary conditions, specify load step options, and obtain
the submodel solution.

The first step is to enter SOLUTION (/SOLU or menu path Main Menu>Solution).

Then define the appropriate analysis type (usually static) and analysis options.

To apply the cut boundary DOF constraints, simply read in the file of D commands (created by
CBDOF) using one of these methods (for example, /INPUT,,CBDO):

Command(s):

/INPUT

GUI:

Utility Menu>File>Read Input from

Similarly, to apply the interpolated temperatures, read in the file of BF commands (created by
BFINT) using one of these methods (for example, /INPUT,,BFIN):

Command(s):

/INPUT

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 154 of 207

Utility Menu>File>Read Input from

If real and imaginary data are involved, first read in the file(s) containing the real data, specify
whether DOF constraint values and/or nodal body force loads are to be accumulated, and then read in
the file containing the imaginary data.

Specify that DOF constraint values are to be accumulated:

Command(s):

DCUM,ADD

GUI:
Main Menu>Preprocessor>Loads>-Loads-Settings>Constraints
Main Menu>Solution>-Loads-Settings>Constraints

Specify that nodal body force loads are to be accumulated:

Command(s):

BFCUM,,ADD

GUI:
Main Menu>Preprocessor>Loads>-Loads-Settings>Nodal Body Ld
Main Menu>Solution>-Loads-Settings>Nodal Body Ld

Be sure to reset the DCUM and BFCUM commands to their default status before proceeding.

It is important that you duplicate on the submodel any other loads and boundary conditions that
existed on the coarse model. Examples are symmetry boundary conditions, surface loads, inertia
forces (such as gravity), concentrated force loads, etc. (see Loads on the Submodel).

Then specify load step options (such as output controls) and initiate solution calculations using one
of these methods:

Command(s):

SOLVE

GUI:
Main Menu>Solution>Current LS
Main Menu>Solution>Run FLOTRAN

After the solution is obtained, leave SOLUTION [FINISH].

The overall data flow for submodeling (without temperature interpolation) is shown in Data Flow
Diagram for Submodeling.

Loads on the Submodel

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 155 of 207

Data Flow Diagram for Submodeling (Without Temperature Interpolation)

5.2.5. Verify the Distance Between the Cut Boundaries and the Stress Concentration

The final step is to verify that the cut boundaries of the submodel are far enough away from the
concentration. You can do this by comparing results (stresses, magnetic flux density, etc.) along the
cut boundaries with those along the corresponding locations of the coarse model. If the results are in
good agreement, it indicates that proper cut boundaries have been chosen. Otherwise, you will need
to re-create and reanalyze the submodel with different cut boundaries further away from the region
of interest.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 156 of 207

An effective way to compare results is to obtain contour displays and path plots, as shown in Contour
Plots to Compare Results and Path Plots to Compare Results.

Contour Plots to Compare Results

Path Plots to Compare Results

Prev Next
Submodeling Sample Analysis Input
Sample Analysis Input
Prev Submodeling Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 157 of 207

5.3. Sample Analysis Input

A sample input listing for a submodeling analysis is shown below:

! Start with coarse model analysis:

/FILNAME,coarse ! Jobname = coarse
/PREP7 ! Enter PREP7
...
... ! Generate coarse model
...
FINISH

/SOLU ! Enter SOLUTION

ANTYPE,... ! Analysis type and analysis options
...
D,... ! Loads and load step options
DSYMM,...
ACEL,...
...
SAVE ! Coarse model database file coarse.db
SOLVE ! Coarse model solution
! Results are on coarse.rst (or rmg, etc.)
FINISH

! Create submodel:
/CLEAR ! Clear the database (or exit ANSYS and re-enter)
/FILNAME,submod ! New jobname = submod
/PREP7 ! Re-enter PREP7
...
... ! Generate submodel
...

! Perform cut boundary interpolation:

NSEL,... ! Select nodes on cut boundaries
NWRITE ! Write those nodes to submod.node
ALLSEL ! Restore full sets of all entities
NWRITE,temps,node ! Write all nodes to temps.node (for
! temperature interpolation)
SAVE ! Submodel database file submod.db
FINISH

RESUME,coarse,db ! Resume coarse model database (coarse.db)

/POST1 ! Enter POST1
FILE,coarse,rst ! Use coarse model results file
SET,... ! Read in desired results data
CBDOF ! Reads cut boundary nodes from submod.node and
! writes D commands to submod.cbdo
BFINT,temps,node ! Reads all submodel nodes from temps.node and
! writes BF commands to submod.bfin (for
! temperature interpolation)
FINISH ! End of interpolation

RESUME ! Resume submodel database (submod.db)

/SOLU ! Enter SOLUTION
ANTYPE,... ! Analysis type and options
...
/INPUT,submod,cbdo ! Cut boundary DOF specifications
/INPUT,submod,bfin ! Interpolated temperature specifications
DSYMM,... ! Other loads and load step options
ACEL,...
...
SOLVE ! Submodel solution

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 158 of 207

FINISH

/POST1 ! Enter POST1

...
... ! Verify submodel results
...
FINISH

Prev Next
How to Do Submodeling Shell-to-Solid Submodels
Shell-to-Solid Submodels
Prev Submodeling Next

5.4. Shell-to-Solid Submodels

In the shell-to-solid submodeling technique, the coarse model is a shell model, and the submodel is a
3-D solid model. A sample is shown in 3D Solid Submodel Superimposed on Coarse Shell Model.

3-D Solid Submodel Superimposed on Coarse Shell Model

The procedure for shell-to-solid submodeling is essentially the same as the solid-to-solid
submodeling procedure discussed in the previous section. Some points to keep in mind about shell-
to-solid submodeling are listed below.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 159 of 207

? Shell-to-solid submodeling is activated by setting KSHS to 1 on the CBDOF command (Main

Menu>General Postproc>Submodeling>Interpolate DOF) and the BFINT command (Main
Menu>General Postproc>Submodeling>Interp Body Forc). This feature is not applicable to
offsets used with SHELL91 or SHELL99 (KEYOPT(11) 0).

? Cut boundaries on the submodel are the end planes that are normal to the shell plane (see Node
Rotations). Nodes on these cut boundaries are written to the node file [NWRITE] (Main
Menu>Preprocessor>Create>Nodes>Write Node File).

? To determine the DOF values at a cut boundary node [CBDOF], the program first projects the
node onto the nearest element in the shell plane. The DOF values of this projected point are
then calculated by interpolation and assigned to the corresponding node. Interpolated
temperatures [BFINT] are calculated based on the average temperature at the midplane of the
nearest shell element.

Note

The nodes on the cut boundary must lie within a distance of 0.75 times the
average thickness of the nearest shell element, as shown in Node Rotations.
That is, the submodel should be approximately centered on the coarse model.

? In a structural analysis, only translational displacements are calculated for the cut boundary
nodes, but their values are based on both the translations and rotations of the projected point.
Also, the node is rotated such that the nodal UY direction is always perpendicular to the shell
plane, as shown in Node Rotations. A UY constraint will be calculated only for nodes that are
within 10% of the average shell element thickness from the shell plane. This prevents over-
constraint of the submodel in the transverse direction.

Node Rotations

Node rotations: (a) before CBDOF command, (b) after CBDOF command

? The .CBDO file written by the CBDOF command will consist of two blocks:
? a block of NMODIF commands (indicating node rotation angles) and DDELE

commands (to delete UY constraints)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 160 of 207

? a block of D commands (to apply the interpolated DOF values).

The two blocks are separated by a /EOF command and a :CBnn label (where nn is the cumulative
iteration number of the results set used).

? You must read in the .CBDO file in PREP7, because the NMODIF command is only valid in
PREP7. To do so, enter the preprocessor, then use one of these methods:
Command(s):

/INPUT

GUI:

Utility Menu>File>Read Input from

Also, you will have to read in the .CBDO file twice, because the two blocks of commands are
separated by a /EOF command. The second time you read in the file, use the LINE field on /INPUT
("Optional line number or label" in the GUI) to instruct the program to read the file starting with
the :CB nn label, as shown below:

/PREP7! The .CBDO file must be read in PREP7

/INPUT,,cbdo ! Reads Jobname.cbdo up to the /EOF command
/INPUT,,cbdo,,:cb1 ! Reads same file from the label :cb1

Prev Next
Sample Analysis Input Where to Find Examples
Where to Find Examples
Prev Submodeling Next

5.5. Where to Find Examples

The ANSYS Verification Manual consists of test case analyses demonstrating the analysis capabilities
of the ANSYS program. While these test cases demonstrate solutions to realistic analysis problems,
the ANSYS Verification Manual does not present them as step-by-step examples with lengthy data
input instructions and printouts. However, most ANSYS users who have at least limited finite
element experience should be able to fill in the missing details by reviewing each test case's finite
element model and input data with accompanying comments.

The ANSYS Verification Manual contains the following submodeling test case:

VM142 - Stress Concentration at a Hole in a Plate

Prev Next
Shell-to-Solid Submodels Substructuring
Substructuring
Prev Advanced Analysis Techniques Guide Next

Chapter 6. Substructuring

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 161 of 207

6.1. What Is Substructuring?

Substructuring is a procedure that condenses a group of finite elements into one element represented
as a matrix. This single matrix element is called a superelement. You can use a superelement in an
analysis just as you would any other ANSYS element type. The only difference is that you first
create the superelement by performing a substructure generation analysis. Substructuring is available
in the ANSYS/Multiphysics, the ANSYS/Mechanical, and the ANSYS/Structural products.

The reasons for substructuring are (a) to reduce computer time and (b) to allow solution of very large
problems with limited computer resources. Examples of reason (a) are nonlinear analyses and
analyses of structures that contain repeated geometrical patterns. In a nonlinear analysis, you can
substructure the linear portion of the model so that the element matrices for that portion need not be
recalculated every equilibrium iteration. In a structure with repeated patterns (such as the four legs of
a table), you can generate one superelement to represent the pattern and simply make copies of it at
different locations, thereby saving a significant amount of computer time.

You can also use substructuring on models with large rotations. For these models, ANSYS assumes
each substructure to rotate about its mass center. In 3-D cases, the substructures contain three rigid
body rotations and three translational motions. With a large rotation model, you do not constrain the
substructure until the use pass because each substructure is treated as a single finite element that
should allow rigid body motions.

An example of reason (b) is an analysis that is too large for the computer in terms of wavefront size
or disk space requirements. In such a situation, you can analyze the model in pieces, where each
piece is a superelement small enough to fit on the computer.

Prev Next
Where to Find Examples How to Do Substructuring
How to Do Substructuring
Prev Substructuring Next

6.2. How to Do Substructuring

A substructure analysis involves three distinct steps or passes:

? Generation pass

? Use pass

? Expansion pass

The generation pass is where you condense a group of "regular" finite elements into a single
superelement. The condensation is done by identifying a set of master degrees of freedom, which are
mainly used to define the interface between the superelement and other elements and to capture
dynamic characteristics for dynamic analyses. Example of a Substructuring Application shows a
plate-like structure that is to be analyzed with contact (interface) elements. Since the contact
elements require an iterative solution, substructuring the plate portion can result in a significant
savings in computer time. The master DOF required in this case are the degrees of freedom that
connect the plate to the contact elements.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 162 of 207

Example of a Substructuring Application

The use pass is where you use the superelement in an analysis by making it part of the model. The
entire model may be a superelement or, as in the plate example, the superelement may be connected
to other non-superelements (see Example of a Substructuring Application). The solution from the use
pass consists only of the reduced solution for the superelement (that is, the degree of freedom
solution only at the master DOF) and complete solution for non-superelements.

Data Flow in a Typical Substructuring Analysis

The expansion pass is where you start with the reduced solution and calculate the results at all
degrees of freedom in the superelement. If multiple superelements are used in the use pass, a separate
expansion pass will be required for each superelement.

In the matrix generation pass of a substructure analysis (ANTYPE, SUBSTR), the equation solver
selected through the EQSLV command has no impact. However, in the substructure matrix use pass
(SOLVE ), specifying an equation solver (such as SPARSE or JCG) through the EQSLV command
allows the appropriate solver to be used in the solution of the model with the substructure.

Data Flow in a Typical Substructuring Analysis shows the data flow for a complete substructure
analysis and some of the files involved. The three passes are explained in detail next.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 163 of 207

Note

Perform each step while using the same version of ANSYS - do not go from one
version of ANSYS to another while performing these steps.

6.2.1. Generation Pass: Creating the Superelement

The procedure to generate a superelement consists of two main steps:

1. Build the model.

2. Apply loads and create the superelement matrices.

6.2.1.1. Build the Model

In this step, you specify the jobname and analysis title and then use PREP7 to define the element
types, element real constants, material properties, and the model geometry. These tasks are common
to most analyses and are described in the ANSYS Basic Analysis Guide. For the generation pass, you
should keep in mind the following additional points:

Jobname-This takes on special significance in a substructure analysis. By using jobnames

effectively, you can eliminate much of the file handling inherent in a three-pass analysis.

Use one of these methods to specify the jobname:

Command(s):

/FILNAME

GUI:

Utility Menu>File>Change Jobname

For example,

/FILNAME,GEN

gives the jobname GEN to all files produced by the generation pass. The default jobname is FILE (or
file) or whatever name was specified while entering the ANSYS program.

Element Types - Most ANSYS element types can be used to generate a substructure. In general, the
only restriction is that elements within the superelement are assumed to be linear. If you include
bilinear elements, they will be treated as linear elements (in their initial state).

Caution

Coupled-field elements used in a direct method coupled-field analysis with load

vector coupling are not valid in a substructure analysis. Other elements in the same
shape family should be used instead. See the ANSYS Coupled-Field Analysis Guide
for details.

Material Properties - Define all necessary material properties. For example, if the mass matrix is to
be generated, density (DENS) (or mass in some form) must be defined; if the specific heat matrix is

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 164 of 207

to be generated, the specific heat (C) must be defined; and so on. Again, because a superelement is
linear, any nonlinear material properties will be ignored.

Model Generation - In the generation pass, you are mainly concerned with creating the superelement
portion of the model. The non-superelement portion, if any, is defined later in the use pass. However,
you should plan the modeling approach for both portions before you start building the model. In
particular, decide on how you want to connect the superelement to the other elements. To ensure the
connection, use the same node numbers at the interface. (Other methods requiring less effort on your
part are discussed in the use pass section later in this chapter.)

One approach might be to develop the entire model, save it on a named database file, and select only
the portion to be substructured for the generation pass. In the use pass then, you can RESUME
(Utility Menu>File>Resume from) from the named database file, unselect the portion that was
substructured and replace it with the superelement matrix.

6.2.1.2. Apply Loads and Create the Superelement Matrices

The "solution" from a substructure generation pass consists of the superelement matrix (or matrices).
As with any other analysis, you define the analysis type and options, apply loads, specify load step
options, and initiate the solution. Details of how to do these tasks are explained below.

Enter SOLUTION using either of these methods:

Command(s):

/SOLU

GUI:

Main Menu>Solution

Define the analysis type and analysis options. The applicable options are explained below.

Analysis Type - Choose a substructure generation using one of these methods:

Command(s):

ANTYPE

GUI:

Main Menu>Solution>-Analysis Type-New Analysis

New analysis or restart - If you are starting a new analysis, choosing the analysis type (as described
above) is all you need to do. However, if the run is a restart, you must also indicate this by setting
STATUS = REST on the ANTYPE command (Main Menu>Solution>-Analysis Type-Restart). A
restart is applicable if you need to generate additional load vectors. (The files Jobname.EMAT,
Jobname.ESAV, and Jobname.DB from the initial run must be available for the restart.)

Name of the superelement matrix file - Specify the name ( Sename) to be assigned to the superelement
matrix file. The program will automatically assign the extension SUB, so the complete file name will
be Sename.SUB. The default is to use the jobname [/FILNAME]. To specify the name of the
superelement matrix file:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 165 of 207

Command(s):

SEOPT

GUI:

Main Menu>Solution>Analysis Options

Matrices to be generated - You can request generation of just the stiffness matrix (or conductivity
matrix, magnetic coefficient matrix, etc.); stiffness and mass matrices (or specific heat, etc.); or
stiffness, mass, and damping matrices. The mass matrix is required if the use pass is a structural
dynamic analysis or if you need to apply inertia loads in the use pass. For the thermal case, the
specific heat matrix is required only if the use pass is a transient thermal analysis. Similar
considerations apply to other disciplines and to the damping matrix. To make your request, use the
SEOPT command as described above.

Matrices to be printed - This option allows you to print out the superelement matrices. You can
request listing of both matrices and load vectors, or just the load vectors. The default is not to print
any matrices. To print out the matrices, use the SEOPT command:

Mass matrix formulation - Applicable only if you want the mass matrix to be generated. You can
choose between the default formulation (which depends on the element type) and a lumped mass
approximation. We recommend the default formulation for most applications. However, for dynamic
analyses involving "skinny" structures, such as slender beams or very thin shells, the lumped mass
approximation has been shown to give better results. To specify a lumped mass approximation, use
one of these methods:

Command(s):

LUMPM

GUI:

Main Menu>Solution>Analysis Options

Define master degrees of freedom using one of these methods:

Command(s):

GUI:

Main Menu>Solution>Master DOFs>Define

In a substructure, master DOFs serve four purposes:

? They serve as the interface between the superelement and other elements. Be sure to define
master DOFs at all nodes that connect to non-superelements, as shown in Example of a
Substructuring Application. All degrees of freedom at these nodes should be defined as master
DOFs ( Lab = ALL on the M command). Master DOFs must be defined even if you plan to
have no elements in the model other than a superelement.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 166 of 207

? If the superelement is to be used in a dynamic analysis, master DOFs characterize the dynamic
behavior of the structure. See Modal Analysis in the ANSYS Structural Analysis Guide for
guidelines.

? If constraints [D] or force loads [F] are to be applied in the use pass, master DOFs must be
defined at those locations with the M command.

? Master DOFs are required in the use pass for large deflections [NLGEOM,ON] (Main
Menu>Solution>Analysis Options) or when used with the SETRAN command (Main
Menu>Preprocessor> Create>Elements>-Superelements-By CS Transfer). For these cases, all
nodes that have master DOFs must have all six (UX, UY, UZ, ROTX, ROTY, ROTZ) defined.

Apply loads on the model. You can apply all types of loads in a substructure generation pass (see
Loads Applicable in a Substructure Analysis), but keep in mind the following points:

? The program will generate a load vector that includes the effect of all applied loads. One load
vector per load step is written to the superelement matrix file. This load vector is the
equivalent load for the combined loads in the load step. A maximum of 31 load vectors are
allowed.

? Non-zero DOF constraints can be used in the generation pass and will become part of the load
vector. (In the expansion pass, if the load step being expanded contains non-zero DOF
constraints, the database must have matching DOF values. If it does not, the DOF constraints
must be respecified [D] in the expansion pass.)

? Application of constraints [D] or force loads [F] can be postponed until the use pass, but
master DOF must be defined at those locations with the M command or corresponding GUI
path.

? Similarly, application of linear and angular accelerations can be postponed until the use pass,
but only if a mass matrix is generated. This is recommended if you plan to rotate the
superelement in the use pass, because load vector directions are "frozen" and rotated with the
superelement.

? The Maxwell force flag (MXWF label on the SF family of commands) is normally used in a
magnetic analysis to flag element surfaces on which the magnetic force distribution is to be
calculated. This flag has no effect (and therefore should not be used) for a superelement in a
magnetic analysis.

Note

If a mass matrix is generated, it is recommended that degree of freedom constraints be

applied in the use pass at the master DOF (defined in the generation pass). This
ensures that all mass is accounted for in the substructure.

For large rotation analyses - Do not apply constraints to the model in this pass. You will apply
constraints for large rotation analyses in the use pass.

Loads Applicable in a Substructure Analysis

Load Commands[1]
Load Name
Category Solid Model Loads Finite Element Loads

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 167 of 207

Displacement DK, DKLIST, DKDELE , DL,

D, DSYM, DLIST,
Temperature Mag. Constraints DLLIST, DLDELE, DA,
DDELE, DSCALE, DCUM
Potential... DALIST, DADELE, DTRAN
Force Heat Flow FK, FKLIST, FKDELE, F, FLIST, FDELE,
Forces
Rate Mag. Flux... FTRAN FSCALE, FCUM
SF, SFLIST, SFDELE,
SFL, SFLLIST, SFLDELE, SFE, SFELIST, SFEDELE,
Pressure Convection Surface
SFA, SFALIST, SFADELE, SFBEAM, SFGRAD,
Maxwell Surface... Loads
SFGRAD, SFTRAN SFFUN, SFSCALE,
SFCUM
BFK, BFKLIST, BFKDELE ,
BF, BFLIST, BFDELE,
Temperature Heat BFL, BFLLIST , BFLDELE,
BFE, BFELIST,
Generation Rate Body Loads BFA, BFALIST, BFADELE ,
BFEDELE, BFSCALE,
Current Density... BFV, BFVLIST, BFVDELE ,
BFCUM
BFTRAN
Gravity, Linear and
Inertia
Angular ACEL, DOMEGA
Loads
Acceleration

1. The menu path used to access each command in the GUI will vary depending on the
engineering discipline of the analysis (structural, magnetic, etc.). For a list of menu paths, see
the description of individual commands in the ANSYS Commands Reference.

Specify load step options. The only options valid for the substructure generation pass are dynamics
options (damping).

Damping (Dynamics Options) - Applicable only if the damping matrix is to be generated.

To specify damping in the form of alpha (mass) damping:

Command(s):

ALPHAD

GUI:

Main Menu>Solution>Time/Frequenc>Damping

To specify damping in the form of beta (stiffness) damping:

Command(s):

BETAD

GUI:

Main Menu>Solution>Time/Frequenc>Damping

To specify damping in the form of material-dependent beta damping:

Command(s):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 168 of 207

MP,DAMP

GUI:
Main Menu>Preprocessor>Material Props>Material Models>Structural>Damping

Save a back-up copy of the database on a named file. Doing this is required because you need to
work with the same database in the expansion pass. To save a back-up copy, use one of these
methods:

Command(s):

SAVE

GUI:

Utility Menu>File>Save as Jobname.db

Start solution calculations using one of these methods:

Command(s):

SOLVE

GUI:

Main Menu>Solution>Current LS

Output from the solution consists of the superelement matrix file, Sename.SUB, where Sename
is the name assigned as an analysis option [SEOPT ] or the jobname [/FILNAME]. The matrix
file includes a load vector calculated based on the applied loads. (The load vector will be zero
if no loads are defined.)

Repeat for additional load steps (that is, to generate additional load vectors). The load vectors are
numbered sequentially (starting from 1) and appended to the same superelement matrix file. See
Loading in the ANSYS Basic Analysis Guide for other methods for multiple load steps.

Leave SOLUTION using one of these methods:

Command(s):

FINISH

GUI:

Main Menu>Finish

6.2.2. Use Pass: Using the Superelement

The use pass may be any ANSYS analysis type (except a FLOTRAN or explicit dynamics analysis).
The only difference is that one or more of the elements in the model is a superelement that has been
previously generated. The individual analysis guides contain detailed descriptions of how to do
different analyses. In this section, we will concentrate on the steps you need to make the
superelement a part of your model.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 169 of 207

6.2.2.1. Clear the Database and Specify a New Jobname

The use pass consists of a new model and new loads. Therefore, the first step is to clear the existing
database. This has the same effect as leaving and re-entering the ANSYS program. To clear the
database, use one of these methods:

Command(s):

/CLEAR

GUI:

Utility Menu>File>Clear & Start New

By default, clearing the database causes the START.ANS file to be re-read. (You can change this
setting if you so desire.)

Caution

If you are using the command input method to clear the database, additional
commands may not be stacked on the same line (using the $ delimiter) as
the /CLEAR command.

Be sure to define a jobname that is different from the one used for the generation pass. This way, you
can ensure that no generation pass files will be overwritten. To define a jobname, use one of these
methods:

Command(s):

/FILNAME

GUI:

Utility Menu>File>Change Jobname

6.2.2.2. Build the Model

This step is performed in PREP7 and consists of the following tasks:

1. Define MATRIX50 (the superelement) as one of the element types. Use one of these methods:

Command(s):

ET

GUI:

Main Menu>Preprocessor>Element Type>Add/Edit/Delete

2. Define other element types for any non-superelements. Nonlinear behavior may or may not be
allowed, depending on the type of analysis to be performed.

3. Define element real constants and material properties for the non-superelements. Nonlinear

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 170 of 207

properties may or may not be allowed, depending on the type of analysis to be performed.

4. Define the geometry of the non-superelements. Take special care in defining the interfaces
where the non-superelements connect to the superelements. The interface node locations must
exactly match the locations of the corresponding master nodes on the superelements (see Node
Locations).

There are three ways to ensure connectivity at these interfaces:

? Use the same node numbers as the ones in the generation pass.

? Use the same node number increment (or offset) between interface nodes in the
generation pass and interface nodes in the use pass. (Use SETRAN, as described below
in step 5b.)

? Couple the two sets of nodes in all degrees of freedom using the CP family of
commands [CP, CPINTF , etc.]. This method is helpful if you cannot use one of the
above two methods. For example, to define a set of coupled degrees of freedom use one
of the following:

Command(s):

CP

GUI:

Main Menu>Preprocessor>Coupling / Ceqn>Couple DOFs

If the superelement is not connected to any other elements, you do not need to define any
nodes in the use pass.

Node Locations

Interface nodes between superelement and non-superelement must exactly match

the master node locations

5. Define the superelement by pointing to the proper element type reference number and reading
in the superelement matrix. To point to the element type:

Command(s):

TYPE

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 171 of 207

Main Menu>Preprocessor>-Modeling-Create>Elements>Elem Attributes

Now read in the superelement matrix using one of these methods (you may first need to
use other commands to modify the matrix, as explained below):

Command(s):

SE

GUI:

Main Menu>Preprocessor>Create>Elements>-Superelements -From .SUB File

a. If there are no non-superelements in the model, or if there are non-superelements and the
interface nodes have the exact same node numbers as the master nodes on the
superelement, then simply read in the superelement using the SE command:

TYPE,...! Element type reference number

SE,GEN! Reads in superelement from file GEN.SUB

The Sename field on the SE command shown above identifies the name of the
superelement matrix file. The extension .SUB is assumed, so the complete file name is
Sename.SUB (GEN.SUB in the above example). The superelement is given the next
available element number.

b. If there are non-superelements in the model and the interface nodes have a constant node
number offset from the master nodes, you must first create a new superelement matrix
with new node numbers and then read in the new matrix.

To create a new superelement matrix, use one of these methods:

Command(s):

SETRAN

GUI:
Main Menu>Preprocessor>Create>Elements>-Superelements -By CS Transfer
Main Menu>Preprocessor>Create>Elements>-Superelements -By Geom Offset

To read in the new matrix, use one of these methods:

Command(s):

SE

GUI:

Main Menu>Preprocessor>Create>Elements>-Superelements -From .SUB File

For example, given an existing superelement matrix file GEN.SUB and a node number
offset of 2000, the commands would be:

SETRAN,GEN,,2000,GEN2,SUB ! Creates new superelement GEN2.SUB with

! node offset = 2000
TYPE,... ! Element type reference number
SE,GEN2 ! Reads in new superelement from file GEN2.SUB

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 172 of 207

c. If there are non-superelements in the model and the interface nodes have no relationship
with the master nodes (as would be the case with automatically meshed models), first
observe the following caution.

Caution

It is quite likely that the node numbers of the master nodes from the
generation pass overlap with node numbers in the use pass model. In such
cases, reading in the superelement [SE] will cause existing use pass
nodes to be overwritten by the superelement's master nodes. To avoid
overwriting existing nodes, use a node number offset [SETRAN] before
reading in the superelement. In any case, save the database [SAVE]
before issuing the SE command.

Thus you should first save the database [SAVE], use the SETRAN command to create a
new superelement matrix with a node number offset, and then use the SE command to
read in the new matrix. The CPINTF command (Main
Menu>Preprocessor>Coupling/Ceqn>Coincident Nodes) can then be used to connect the
pairs of nodes at the interface. For example, given a superelement matrix file called
GEN.SUB:

*GET,MAXNOD,NODE,,NUM,MAX ! MAXNOD = maximum node number

SETRAN,GEN,,MAXNOD,GEN2,SUB ! Creates new superelement with
! node offset = MAXNOD, name = GEN2.SUB
SE,GEN2 ! Reads in new superelement
NSEL,... ! Select all nodes at the interface
CPINTF,ALL ! Couples each pair of interface nodes in
! all DOF
NSEL,ALL

d. If the superelement is to be transformed - moved or copied to a different position, or

symmetrically reflected - you must first use the SETRAN command or SESYMM
command (Main Menu>Preprocessor>Create> Elements>By Reflection), with the
appropriate node number offsets, to create new superelement matrix files and then use
SE to read in the new matrices. Connecting the superelements to the non-superelements
is done the same way as above - by using common node numbers, a constant node
number offset, or the CPINTF command.

Note

If you use SETRAN to transfer the superelement to a different coordinate

system, the superelement's master nodes are rotated with it by default. This is
typically useful if the original superelement's master nodes are rotated, into a
cylindrical system for example. (In this case, the transfer does not effect the
superelement stiffness matrix.) If the original superelement has no rotated
nodes, it is likely that the transferred superelement will not need rotated nodes
either. You can prevent node rotation in such cases by setting the NOROT field
on SETRAN to 1. (The superelement stiffness matrix and load vector are
modified by the program for this type of transfer.)

6. Verify the location of the superelement using graphics displays and listings. Superelements are
represented by an edge outline display, the data for which are written to the matrix file in the
generation pass. To produce a graphics display:

Command(s):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 173 of 207

EPLOT

GUI:

Utility Menu>Plot>Elements

To produce a listing:

Command(s):

SELIST

GUI:

Utility Menu>List>Other>Superelem Data

7. Save the complete model database:

Command(s):

SAVE

GUI:

Utility Menu>File>Save as Jobname.db

Leave PREP7 using one of these methods:

Command(s):

FINISH

GUI:

Main Menu>Finish

6.2.2.3. Apply Loads and Obtain the Solution

This step is performed in SOLUTION. The procedure to obtain the use pass solution depends on the
type of analysis being performed. As mentioned earlier, you can subject a superelement to any type
of analysis. You should, of course, have the appropriate matrices generated during the generation
pass. For example, if you intend to do a structural dynamic analysis, the mass matrix must be
available. The procedure is as follows:

1. Enter SOLUTION using one of these methods:

Command(s):

/SOLU

GUI:

Main Menu>Solution

2. Define the analysis type and analysis options.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 174 of 207

For large rotation analyses - turn large deformation effects on [NLGEOM,ON], and define
the proper number of substeps for the nonlinear analysis.

3. Apply loads on the non-superelements. These may consist of DOF constraints and symmetry
conditions [D family of commands], force loads [F family], surface loads [SF family], body
loads [BF family], and inertia loads [ACEL, etc.]. Remember that inertia loads will affect the
superelement only if its mass matrix was generated in the generation pass.

Note

For large rotation analyses, be sure to apply the proper constraints in this step.

4. Apply superelement load vectors (if any) using one of these methods:

Command(s):

SFE

GUI:

Main Menu>Solution>-Loads-Apply>Load Vector>For Superelement

One load vector per load step (created during the generation pass) is available on the
superelement matrix file, and is identified by its reference number:

SFE,63,1,SELV,,0.75

applies, on element number 63, load vector number 1, with a scale factor of 0.75. Thus
the ELEM field represents the element number of the superelement, LKEY represents the
load vector number (default = 1), Lab is SELV, and VAL1 represents the scale factor
(default = 0.0). (See the SFE command description for more information.)

Note

The load vector orientation is fixed (frozen) to the superelement, so if the

superelement is used in a rotated position, the load vector rotates with it.
The same applies to the degree of freedom directions (UX, UY, ROTY,
etc.). They too are fixed to the superelement and will rotate with the
superelement if it is rotated (unless NOROT = 1 on the SETRAN
command, in which case the nodal coordinate systems will not be
rotated).

5. Specify load step options that are appropriate for the analysis type.

Note

Do not use the PCG solver with the MATRIX50 superelement.

6. Initiate the solution:

Command(s):

SOLVE

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 175 of 207

GUI:

Main Menu>Solution>Current LS

Results from the solution consist of the complete solution for non-superelements and the
reduced solution - DOF solution at masters - for the superelements. The complete solution for
non-superelements is written to the results file ( Jobname.RST, RTH, or RMG), which you can
postprocess using normal postprocessing procedures.

The reduced solution is written to the file Jobname.DSUB . You can review this file using one of
these methods:
Command(s):

SEDLIST

GUI:

Main Menu>General Postproc>List Results>Superelem DOFUtility

Menu>List>Results>Superelem DOF Solu

To expand the reduced solution to all elements within the superelement, you will need to
perform the expansion pass, explained next.

7. Leave SOLUTION:

Command(s):

FINISH

GUI:

Main Menu>Finish

6.2.3. Expansion Pass: Expanding Results within the Superelement

The expansion pass starts with the reduced solution calculated in the use pass and calculates the
complete solution within the superelement. The procedure for the expansion pass is explained below.
It assumes that the .EMAT, .ESAV, .SUB, .TRI, .DB, and .SELD files from the generation pass and
the .DSUB file from the use pass are available. If an offset of node numbers was used in the use pass
[SETRAN or SESYMM], it will automatically be taken into account in the expansion pass.

1. Clear the database:

Command(s):

/CLEAR

GUI:

Utility Menu>File>Clear & Start New

This has the same effect as leaving and re-entering the ANSYS program.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 176 of 207

2. Change the jobname to what it was during the generation pass. This way, the program can
easily identify the files required for the expansion pass:

Command(s):

/FILNAME

GUI:

Utility Menu>File>Change Jobname

3. Restore the generation pass database:

Command(s):

RESUME

GUI:

Utility Menu>File>Resume Jobname.db

4. Enter SOLUTION using one of these methods:

Command(s):

/SOLU

GUI:

Main Menu>Solution

5. Activate the expansion pass and its options:

Command(s):

EXPASS

GUI:

Main Menu>Solution>ExpansionPass

Expansion pass on or off - Choose "on."

Name of superelement to be expanded - Specify the name ( Sename):

Command(s):

SEEXP

GUI:

Main Menu>Solution>ExpansionPass>Expand Superelem

(The complete name of the file is assumed to be Sename.SUB.)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 177 of 207

Name of the reduced solution file from use pass - Specify the name ( Usefil) using the SEEXP
command (or the menu path shown above). The complete name of the file is assumed to be
Usefil.DSUB.

Real or imaginary component of displacement - Applicable only if the use pass was a
harmonic response analysis. Use the SEEXP command (or the menu path shown above).

Solution to be expanded - Identify the use pass solution to be expanded. You can use either the
load step and substep numbers or the time (or frequency) to identify the solution:
Command(s):

EXPSOL

GUI:

Main Menu>Solution>ExpansionPass>By Load StepMain

Menu>Solution>ExpansionPass>By Time/Freq

Note

If the load step being expanded contains non-zero DOF constraints, the
database must have matching DOF values. If it does not, the DOF constraints
must be respecified [D] in the expansion pass.

6. Specify load step options. The only options valid for a substructure expansion pass are output
controls:

Output Controls - These options control printed output, database and results file output, and
extrapolation of results.

If you want to include any results data on the printed output file (Jobname.OUT):
Command(s):

OUTPR

GUI:

Main Menu>Solution>Output Ctrls>Solu Printout

If you want to control the data on the results file ( Jobname.RST):

Command(s):

OUTRES

GUI:

Main Menu>Solution>Output Ctrls>DB/Results File

If you want to review element integration point results by copying them to the nodes instead of
extrapolating them (default):
Command(s):

ERESX

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 178 of 207

GUI:

Main Menu>Solution>Output Ctrls>Integration Pt

7. Start expansion pass calculations:

Command(s):

SOLVE

GUI:

Main Menu>Solution>Current LS

8. Repeat steps 5 to 7 for additional use pass solutions to be expanded. If you need to expand the
solution for a different superelement, you will need to leave and re-enter SOLUTION.

9. Finally, leave SOLUTION:

Command(s):

FINISH

GUI:

Main Menu>Finish

10. Postprocess results in the superelement using standard techniques.

Note

An expansion pass is not valid if the use pass was a PSD analysis.

Prev Next
Substructuring Sample Analysis Input
Sample Analysis Input
Prev Substructuring Next

6.3. Sample Analysis Input

A sample command input listing for a substructuring analysis is shown below. This example assumes
a single superelement which is possibly combined with non-superelements.

! GENERATION PASS
! Build the model (superelement portion)
/FILNAME,GEN ! Jobname = GEN (for example)
/TITLE,...
/PREP7 ! Enter PREP7
---
--- ! Generate superelement portion of model
FINISH

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 179 of 207

! Apply loads and create the superelement matrices

/SOLU ! Enter SOLUTION
ANTYPE,SUBST ! Substructure analysis
SEOPT,GEN,... ! Superelement name and other substructure analysis options
M,... ! Master DOF
D,... ! Loads. A load vector will be generated and
--- ! written to the superelement matrix file
--- ! Load step options
SAVE ! Save the database for later expansion pass
SOLVE ! Initiate solution -- creates GEN.SUB file
! containing superelement matrix and load vector
--- ! Loads for second load vector (D and M may not changed)
SOLVE ! Add load vector 2
--- ! Repeat loading and SOLVE sequence for additional load vectors
--- ! (Up to 31 total)
FINISH
! USE PASS
! Build the model
/CLEAR ! Clear the database (or exit and re-enter ANSYS)
/FILNAME,USE ! Jobname = USE (for example)
/PREP7 ! Enter PREP7
ET,1,MATRIX50 ! MATRIX50 is the superelement type
ET,2,... ! Element type for non-superelements
--- ! Generate non-superelement model
---
TYPE,1 ! Point to superelement type reference number
SETRAN,... ! May be required for node number offset
SE,... ! Read in the superelement created by SETRAN
EPLOT ! Verify location of superelement
NSEL,... ! Select nodes at interface
CPINTF,ALL ! Couple node pairs at interface (required if
! node numbers at interface don't match)
NSEL,ALL
FINISH

! Apply loads and obtain the solution

/SOLU ! Enter SOLUTION
ANTYPE,... ! Analysis type and analysis options
---
---
D,... ! Loads on non-superelements
---
---
SFE,... ! Apply superelement load vector
--- ! Load step options
---
SAVE ! Save database before solution
SOLVE ! Initiate solution -- calculates complete solution
! for non-superelements (USE.RST, RTH or RMG) and
! reduced solution for superelements (USE.DSUB)
FINISH

! ... Review results in non-superelements

! EXPANSION PASS
/CLEAR ! Clear the database
/FILNAME,GEN ! Change jobname back to generation pass jobname
RESUME ! Restore generation pass database
/SOLU ! Enter SOLUTION
EXPASS,ON ! Activate expansion pass
SEEXP,GEN,USE... ! Superelement name to be expanded (GEN, unless SETRAN used)
--- ! Load step options (mainly output controls)
---

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 180 of 207

SOLVE ! Initiate expansion pass solution. Full

! superelement solution written to GEN.RST (or
! RTH or RMG).
FINISH

! ... Review results in superelements

Please see the ANTYPE, SEOPT, M, ET, SETRAN, SE, CPINTF , EXPASS, and SEEXP
command descriptions for more information.

Prev Next
How to Do Substructuring Top-down Substructuring
Top-down Substructuring
Prev Substructuring Next

6.4. Top-down Substructuring

The substructuring procedure described in the previous section is called bottom-up substructuring,
meaning that each superelement is separately generated in an individual generation pass, and all
superelements are assembled together in the use pass. This method is suitable for very large models
which are divided into smaller superelements so that they can "fit" on the computer.

For substructuring of smaller models or of systems with global project geometry controls and
isolated component analysis, you can use a slightly different technique known as top-down
substructuring. This method is suitable, for instance, for substructuring of the linear portion of
nonlinear models that are small enough to fit on the computer. An advantage of this method is that
the results for multiple superelements can be assembled in postprocessing. The procedure for top-
down substructuring is briefly explained below, and is followed by a sample input.

1. First build the entire model, including both the superelement and non-superelement portions.
Save this model on a named database file (for example, FULL.DB). The full model database is
later required for the expansion pass. It will also be required for the use pass if the model
consists of non-superelements.

2. Perform the generation pass on a selected subset of the entire model. Because the full model
has already been built, all you need to do is to select the elements for the superelement portion,
apply the desired loads (for the load vector), and create the superelement with the SOLVE
command (Main Menu>Solution>-Solve-Current LS).

The use of components may be helpful for this. To group items into a component, use the CM
command (Utility Menu>Select> Comp/Assembly>Create Component).

If multiple superelements are to be generated, you will need to exit and re-enter SOLUTION
each time and repeat the select-load-solve steps. Be sure to use a different jobname for each
superelement.

3. Perform the use pass. Enter PREP7 and start by restoring the full model database and then
selecting only the non-superelement portion of the model. Next, define the superelement type
[ET, TYPE] and read in the appropriate superelement matrices. In most cases, you don't need
to worry about the connecting nodes between the superelements, because they were all
generated from a single model.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 181 of 207

Enter SOLUTION and define the analysis type and analysis options. Apply loads on the non-
superelements, read in load vectors (if any), specify load step options, and initiate the use pass
solution.

4. Perform the expansion pass. Start by restoring the full model database, with all elements and
nodes active. Then expand each superelement separately, using the appropriate jobnames and
exiting and re-entering SOLUTION each time. You can then review the results in each
superelement using normal postprocessing procedures. Use of the full database, FULL.DB,
allows the reading in of multiple superelement results:

RESUME,FULL,DB
/POST1
FILE,GEN1
SET,...
FILE,GEN2
SET,...!Will not clear previous superelement results

A sample input for top-down substructuring follows. This example assumes a model with one
superelement and other non-superelements.

! Sample input for top-down substructuring

!
! BUILD THE FULL MODEL
!
/FILNAME,FULL ! Jobname = FULL (for example)
/TITLE,...
/PREP7 ! Enter PREP7
---
--- ! Generate entire model, including both the
--- ! superelement and non-superelement portions
---
SAVE ! Save the full model database. It is required for
! the (use pass and) expansion pass.
FINISH

! GENERATION PASS

/FILNAME,GEN ! Jobname = GEN (for example)

/SOLU ! Enter SOLUTION
ANTYPE,SUBSTR ! Substructure analysis type
SEOPT,GEN,... ! Analysis options
ESEL,... ! Select elements and
NSLE ! nodes in the superelement portion
M,... ! Master DOF
D,... ! Loads. A load vector will be generated and written to the
--- ! superelement matrix file
--- ! Load step options
---
SOLVE ! Initiate solution -- creates superelement
! matrix file GEN.SUB.
--- ! Loads for second load vector (D and M may not changed)
SOLVE ! Add load vector 2
--- ! Repeat loading and SOLVE sequence for additional load vectors
--- ! (Up to 31 total)
FINISH

! USE PASS
!
/CLEAR ! Clear database for use pass

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 182 of 207

/FILNAME,USE ! Jobname = USE (for example)

RESUME,FULL,DB ! Restore full model database (for non-superelements)
ESEL,... ! Select elements and
NSLE ! nodes in the non-superelement portion

/PREP7
ET,...,MATRIX50 ! Superelement type (type number not used by non-superelements)
TYPE,... ! Point to superelement type reference number
SE,GEN ! Read in superelement matrix (GEN.SUB created above)
EPLOT
FINISH

/SOLU
ANTYPE,... ! Analysis type and analysis options
---
D,... ! Loads on non-superelements
---
---
SFE,... ! Superelement load vector
---
--- ! Load step options
---
SOLVE ! Initiates solution -- calculates complete
! solution for non-superelements (USE.RST, etc.)
! and reduced solution for superelement (USE.DSUB)
FINISH

! EXPANSION PASS
!
/CLEAR ! Clear database for expansion pass
/FILNAME,GEN ! Change jobname back to generation pass jobname
RESUME,FULL,DB ! Restore full model database

/SOLU ! Enter SOLUTION

ANTYPE,SUBSTR
EXPASS,ON ! Activate expansion pass
EXPSOL,... ! Specifies the solution to be expanded
SEEXP,GEN,USE,... ! Superelement name to be expanded
--- ! Load step options (mainly output controls)
---
SOLVE ! Initiate expansion pass solution. Full
! superelement solution written to GEN.RST (or
! RTH or RMG).
FINISH

! ... Review results in superelement

Please see the ANTYPE, SEOPT, M, ET, SE, EXPASS, and SEEXP command descriptions for
more information.

Prev Next
Sample Analysis Input Nested Superelements
Nested Superelements
Prev Substructuring Next

6.5. Nested Superelements

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 183 of 207

A powerful substructuring feature available in the ANSYS program is the ability to use nested
superelements: one superelement containing another. That is, when you generate a superelement, one
of the elements in the generation pass may be a previously generated superelement.

For instance, you may have a superelement named PISTON. You can then generate another
superelement named CYLINDER which contains the superelement PISTON. Now, for a complete
analysis of the cylinder and its piston, you will need to perform one use pass and two expansion
passes. The use pass will calculate the reduced solution for the master DOF in the superelement
CYLINDER. The first expansion pass will calculate the complete solution for CYLINDER and the
reduced solution for PISTON. The second expansion pass will then give you the complete solution in
the piston.

Prev Next
Top-down Substructuring Prestressed Substructures
Prestressed Substructures
Prev Substructuring Next

6.6. Prestressed Substructures

In modeling a system's behavior properly, it may be important to consider its stress state. That stress
state will influence the values of the stiffness matrix terms. The stress state from a previous structural
solution may be included when the stiffness matrix is formed in a superelement generation pass.
Stress stiffening can provide strength in a structure which would normally not have any resistance to
certain loadings. For example, a stretched cable can resist a normal loading while a slack cable
cannot. Stress stiffening can also change the response frequencies of a system which impacts both
modal and transient dynamic problems.

Two different approaches can be used to generate prestressed substructures. These approaches are a
static analysis prestress and a substructuring analysis prestress.

6.6.1. Static Analysis Prestress

1. Build the model, define a static analysis (ANTYPE command or menu path Main
Menu>Solution>New Analysis) and apply the stiffening load.

2. Specify that prestress effects be calculated (PSTRES command or menu path Main
Menu>Solution>Analysis Options or Main Menu>Solution>Unabridged Menu>Analysis
Options).

3. Specify that the File.EMAT be written (EMATWRITE command; not available via the GUI)

4. Initiate the static analysis (SOLVE command or menu path Main Menu> Solution>-Solve-
Current LS).

5. Perform the generation pass. Include the prestress effects from the static analysis by issuing
the PSTRES command. (It is important to have the prestress effects active during the static
analysis and the generation pass.)

6. Perform the use and expansion passes.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 184 of 207

7. Review the results.

6.6.2. Substructuring Analysis Prestress

Note

This method does not require a static analysis on the full DOF model.

1. Build the model and perform the generation pass. Make sure to reserve space for the stress
stiffening matrix by setting SESST = 1 on the SEOPT command (Main
Menu>Solution>Analysis Options).

2. Apply loads and perform the static use pass.

3. Perform the expansion pass using the PSTRES command (Main Menu>Solution>Analysis
Options) to calculate prestress effects.

4. Repeat the generation pass for a new substructure generation, while continuing the prestress
effects using PSTRES.

5. Solve the second generation pass and perform the use pass.

6. Perform the expansion pass and review the results.

Prev Next
Nested Superelements Where to Find Examples
Where to Find Examples
Prev Substructuring Next

6.7. Where to Find Examples

The ANSYS Verification Manualconsists of test case analyses demonstrating the analysis capabilities
of the ANSYS program. While these test cases demonstrate solutions to realistic analysis problems,
the ANSYS Verification Manual does not present them as step-by-step examples with lengthy data
input instructions and printouts. However, most ANSYS users who have at least limited finite
element experience should be able to fill in the missing details by reviewing each test case's finite
element model and input data with accompanying comments.

The ANSYS Verification Manual contains the following substructuring cases:

VM125 - Radiation Heat Transfer Between Concentric Cylinders

VM141 - Diametrical Compression of a Disk
VM147 - Gray-Body Radiation within a Frustrum of a Cone
VM153 - Nonaxisymmetric Vibration of a Stretched Circular Membrane (Using Cyclic Symmetry)

Prev Next
Prestressed Substructures Element Birth and Death
Element Birth and Death

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 185 of 207

Prev Advanced Analysis Techniques Guide Next

Chapter 7. Element Birth and Death

7.1. What Is Element Birth and Death?
If material is added to (or removed from) a system, certain elements in your model may become
"existent" (or "nonexistent"). The element birth and death options can be used to deactivate or
reactivate selected elements in such cases. (Applicable element types are listed in Elements With
Birth and Death Capability.) This feature can be useful in analyzing excavation (as in mining and
tunneling), staged construction (as in shored bridge erection), sequential assembly (as in fabrication
of layered computer chips), and many other applications in which you can easily identify activated or
deactivated elements by their known locations. The birth and death feature is only available in the
ANSYS/Multiphysics, ANSYS/Mechanical, and ANSYS/Structural products.

Elements With Birth and Death Capability

LINK1 MASS21 SOLID45 SOLID70 SOLID98 CONTA173

PLANE2 BEAM23 PLANE53 MASS71 PLANE121 CONTA174
BEAM3 BEAM24 BEAM54 PLANE75 SOLID122 LINK180
BEAM4 PLANE25 PLANE55 PLANE77 SOLID123 SHELL181
SOLID5 MATRIX27 SHELL57 PLANE78 SHELL143 PLANE182
LINK8 LINK31 PIPE59 PLANE82 SURF151 PLANE183
LINK10 LINK32 PIPE60 PLANE83 SURF152 SOLID185
LINK11 LINK33 SOLID62 SOLID87 SURF153 SOLID186
PLANE13 LINK34 SHELL63 SOLID90 SURF154 SOLID187
COMBIN14 PLANE35 SOLID64 SOLID92 SHELL157 BEAM188
PIPE16 SHELL41 SOLID65 SHELL93 TARGE169 BEAM189
PIPE17 PLANE42 PLANE67 SOLID95 TARGE170
PIPE18 SHELL43 LINK68 SOLID96 CONTA171
PIPE20 BEAM44 SOLID69 SOLID97 CONTA172

User elements may also be given the birth and death capability. See the ANSYS Guide to User
Programmable Features for more information on user elements

In some circumstances, an element's birth and death status may be dependent on an ANSYS-
calculated quantity, such as temperature, stress, strain, etc. You can use commands such as
ETABLE (Main Menu>General Postproc>Element Table>Define Table) and ESEL (Utility
Menu>Select>Entities) to determine the value of such quantities in selected elements, and to change
the status (melted, solidified, ruptured, etc.) of those elements accordingly. This procedure can be
useful for modeling effects due to phase changes (as in welding processes, when structurally inactive
molten material solidifies and becomes structurally active), failure-surface propagation, and other
analysis-dependent element changes.

7.1.1. How Does the Birth and Death Feature Work?

To achieve the "element death" effect, the ANSYS program does not actually remove "killed"

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 186 of 207

elements. Instead, it deactivates them by multiplying their stiffness (or conductivity, or other
analogous quantity) by a severe reduction factor [ESTIF]. This factor is set to 1.0E-6 by default, but
can be given other values (see Apply Loads and Obtain the Solution for more information). Element
loads associated with deactivated elements are zeroed out of the load vector, however, they still
appear in element-load lists. Similarly, mass, damping, specific heat, and other such effects are set to
zero for deactivated elements. The mass and energy of deactivated elements are not included in the
summations over the model. An element's strain is also set to zero as soon as that element is killed.

In like manner, when elements are "born," they are not actually added to the model; they are simply
reactivated. You must create all elements, including those to be born in later stages of your analysis,
while in PREP7. You cannot create new elements in SOLUTION. To "add" an element, you first
deactivate it, then reactivate it at the proper load step.

When an element is reactivated, its stiffness, mass, element loads, etc. return to their full original
values. Elements are reactivated having no record of strain history (or heat storage, etc.). However,
initial strain defined as a real constant (for elements such as LINK1) will not be affected by birth and
death operations. Also, unless large-deformation effects are turned on [NLGEOM,ON], some
element types will be reactivated in their originally specified geometric configuration (large-
deformation effects should be included to obtain meaningful results). Thermal strains are computed
for newly-activated elements based on the current load step temperature and the reference
temperature. Thus, newborn elements with thermal loads may not be stress-free as intended. The
material property REFT can be used instead of the global TREF to specify material-dependent
reference temperatures. This allows you to specify the activation temperature as a stress-free
temperature.

Prev Next
How to Use the Birth and Death
Where to Find Examples Feature
How to Use the Birth and Death Feature
Prev Element Birth and Death Next

7.2. How to Use the Birth and Death Feature

You can include element birth and death behavior in most static and nonlinear transient analyses,
using the same basic procedures described in the various analysis guides.

Modify your basic analysis procedure as follows to incorporate the element birth and death feature:

7.2.1. Build the Model

While in PREP7, create all elements - even those that won't be activated until later load steps. You
cannot create new elements outside of PREP7.

7.2.2. Apply Loads and Obtain the Solution

Perform the following actions within SOLUTION:

7.2.2.1. Define the First Load Step

In the first load step, you must choose the analysis type and all appropriate analysis options:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 187 of 207

Command(s):

ANTYPE

GUI:

Main Menu>Solution>-Analysis Type-New Analysis

For a structural analysis, large-deflection effects should be on:

Command(s):

NLGEOM,ON

GUI:
Main Menu>Solution>Sol'n Controls>basic

For all birth and death applications, you must be sure to set the Newton-Raphson option explicitly in
the first load step, as the program cannot predict the presence of an EKILL command in a
subsequent load step:

Command(s):

NROPT

GUI:
Main Menu>Solution>Analysis Options
Main Menu>Solution>Unabridged Menu>Analysis Options

Deactivate [EKILL] all the initially inactive elements that are to be added (reactivated) in later load
steps:

Command(s):

EKILL

GUI:

Main Menu>Solution>-Load Step Opts-Other>Kill Elements

Elements are deactivated (or activated) in the first substep of the load step, and maintain that status
through the rest of the load step. Note that the default reduction factor used as a stiffness multiplier
might not suffice for some problems. Sometimes, you might have to consider using a more severe
reduction factor. To provide a new value for the reduction factor:

Command(s):

ESTIF

GUI:

Main Menu>Solution>Other>StiffnessMult

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 188 of 207

Nodes that are not connected to any active elements may "float," or pick up stray degree of freedom
(DOF) responses. In some cases, you might want to constrain inactive DOFs [D, CP, etc.] to reduce
the number of equations to be solved, and to avoid ill-conditioning. However, constraining inactive
DOFs can become more important for cases in which you want to reactivate elements with a specific
shape (or temperature, etc.). Of course, you must be sure to remove such artificial constraints when
you reactivate elements. You must also remove nodal loads from inactive DOFs (i.e., at nodes that
are not connected to any active elements). Similarly, you must specifically add nodal loads (if any) to
reactivated DOFs.

Part of your input listing could look like this for your first load step:

! First load step

TIME,... ! Sets TIME value (optional for static analyses)
NLGEOM,ON ! Turns large-deflection effects on
NROPT,FULL ! You must explicitly set the Newton-Raphson option
ESTIF,... ! Sets non-default reduction factor (optional)
ESEL,... ! Selects elements to be deactivated in this load step
EKILL,... ! Deactivates selected elements
ESEL,S,LIVE ! Selects all active elements
NSLE,S ! Selects all active nodes
NSEL,INVE ! Selects all inactive nodes (those not attached to any
! active elements)
D,ALL,ALL,0 ! Constrains all inactive DOFs (optional)
NSEL,ALL ! Selects ALL nodes
ESEL,ALL ! Selects ALL elements
D,... ! Adds constraints as appropriate
F,... ! Adds nodal loads to active DOFs as appropriate
SF,... ! Adds element loads as appropriate
BF,... ! Adds body loads as appropriate
SAVE
SOLVE

Please see the TIME, NLGEOM, NROPT, ESTIF, ESEL, EKILL, NSLE, NSEL, D, F, SF, and
BF command descriptions for more information.

7.2.2.2. Subsequent Load Steps

In your remaining load steps, you can deactivate and reactivate elements as desired. As before, be
sure to apply and delete constraints and nodal loads as appropriate.

To deactivate elements, use one of these methods:

Command(s):

EKILL

GUI:

Main Menu>Solution>-Load Step Opts-Other>Kill Elements

To reactivate elements, use one of these methods:

Command(s):

EALIVE

GUI:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 189 of 207

Main Menu>Solution>-Load Step Opts-Other>Activate Elem

! Second (or subsequent) load step:

TIME,...
ESEL,...
EKILL,... ! Deactivates selected elements
ESEL,...
EALIVE,... ! Reactivates selected elements
...
FDELE,... ! Deletes nodal loads at inactive DOFs
D,... ! Constrains inactive DOFs
...
F,... ! Adds nodal loads as appropriate to active DOFs
DDELE,... ! Deletes constraints from reactivated DOFs
SAVE
SOLVE

See the TIME, ESEL, EKILL, EALIVE, FDELE , D, F, and DDELE command descriptions for
more information.

7.2.3. Review the Results

For the most part, you will follow standard procedures when postprocessing an analysis that contains
deactivated or reactivated elements. However, you should realize that "killed" elements are still
present in your model, even though they make an insignificant contribution to the stiffness
(conductivity, etc.) matrix. Therefore, they will be included in element displays, output listings, etc.
For example, deactivated elements will be included in nodal results averaging (PLNSOL command
or menu path Main Menu>General Postproc>Plot Results>Nodal Solu) and will "smear" the results.
The entire element printout for deactivated elements should be ignored, since many items computed
make little physical sense. A suggestion is to use selecting (ESEL command or menu path Utility
Menu>Select>Entities) to remove deactivated elements for element displays and for other
postprocessing operations.

7.2.4. Using ANSYS Results to Control Birth and Death

At times, you will not explicitly know the location of elements that you need to deactivate or
reactivate. For instance, if you want to "kill" melted elements in a thermal analysis (i.e., to model the
removal of melted material), you will not know the location of those elements beforehand; you will
have to identify them on the basis of their ANSYS-calculated temperatures. When the decision to
deactivate or reactivate an element depends on the value of an ANSYS result item (such as
temperature, stress, strain, etc.), you can use commands to identify and select the critical elements.

To identify the critical elements:

Command(s):

ETABLE

GUI:

Main Menu>General Postproc>Element Table>Define Table

To select the critical elements:

Command(s):

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 190 of 207

ESEL

GUI:

Utility Menu>Select>Entities

You could then deactivate or reactivate the selected elements. (You could also use the ANSYS
Parametric Design Language to write a macro to perform such an operation. See the ANSYS APDL
Programmer's Guide for more information.)

To deactivate the selected elements:

Command(s):

EKILL, ALL

GUI:

Main Menu>Solution>-Load Step Opts-Other>Kill Elements

To reactivate the selected elements, use one of these methods:

Command(s):

EALIVE,ALL

GUI:

Main Menu>Solution>-Load Step Opts-Other>Activate Elem

The following simplified input listing demonstrates how you might deactivate elements that rupture
when their total strain has exceeded some critical value:

/SOLU ! Enter SOLUTION

... ! Typical solution procedure
SOLVE
FINISH
!
/POST1 ! Enter POST1
SET,...
ETABLE,STRAIN,EPTO,EQV ! Store total equivalent strain in ETABLE
ESEL,S,ETAB,STRAIN,0.20 ! Select all elements with total equivalent strain
! greater than or equal to 0.20
FINISH
!
/SOLU ! Re-enter SOLUTION
ANTYPE,,REST
EKILL,ALL ! Deactivate selected (overstrained) elements
ESEL,ALL ! Restore full element set
... ! Continue with solution

Please see the ETABLE, ESEL, ANTYPE, and EKILL command descriptions for more
information.

Prev Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 191 of 207

Element Birth and Death Further Guidelines

Further Guidelines
Prev Element Birth and Death Next

7.3. Further Guidelines

? Constraint equations [CE, CEINTF, etc.] cannot be applied to inactive DOFs. (Inactive DOFs
occur when a node has no active ("alive") elements attached to it.)

? You can model stress-relieving operations (such as annealing) by deactivating and then
reactivating elements.

? In nonlinear analyses, be careful not to deactivate or reactivate elements in such a way as to

create singularities (such as sharp re-entrant corners in a structural analysis) or sudden large
changes in stiffness. Such situations are likely to cause convergence difficulties.

? If the model is entirely linear (has no contact or other nonlinear element present, and the
material is linear) except for birth and death elements, ANSYS will treat the analysis as linear,
and will not set SOLCONTROL,ON.

? The FULL Newton-Raphson option with adaptive descent turned on often yields good results
in analyses that incorporate element birth and death. To use this option with adaptive descent
turned on:
Command(s):

NROPT,FULL,,ON

GUI:
Main Menu>Solution>Analysis Options
Main Menu>Solution>Unabridged Menu>Analysis Options

? You can retrieve a parameter whose value will indicate the status (active or inactive) of an
element [*GET,Par,ELEM, n, ATTR, LIVE] (Utility Menu> Parameters>Get Scalar Data).
This parameter could be used in APDL logical branching ([*IF], etc.) or in other applications
in which you need to monitor the birth-death status of an element.

? You might be tempted to try to deactivate or reactivate elements by changing their material
properties [MPCHG ] (Main Menu>Preprocessor> Material Props>Change Mat Num).

However, you must proceed cautiously if you attempt such a procedure. The safeguards and
restrictions that affect "killed" elements will not apply to elements that have their material properties
changed in SOLUTION. (Element forces will not be automatically zeroed out; nor will strains, mass,
specific heat, etc.) Many problems could result from careless use of MPCHG. For instance, if you
reduce an element's stiffness to almost zero, but retain its mass, it could result in a singularity if
subjected to acceleration or inertial effects.

One application of MPCHG would be in modeling construction sequences in which the strain
history of a "born" element is maintained. Using MPCHG in such cases will enable you to capture
the initial strain experienced by elements as they are fitted into the displaced nodal configuration.

? The load step file method [LSWRITE] for solving multiple load steps cannot be used with the

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 192 of 207

birth-death option, because it will not write the status of deactivated or reactivated elements to
the load step file. Birth and death analyses having multiple load steps must therefore be
performed using a series of explicit SOLVE commands (Main Menu>Solution>Current LS).

Prev Next
How to Use the Birth and Death
Where to Find Examples
Feature
Where to Find Examples
Prev Element Birth and Death Next

7.4. Where to Find Examples

The ANSYS Verification Manual consists of test case analyses demonstrating the analysis capabilities
of the ANSYS program. While these test cases demonstrate solutions to realistic analysis problems,
the ANSYS Verification Manual does not present them as step-by-step examples with lengthy data
input instructions and printouts. However, most ANSYS users who have at least limited finite
element experience should be able to fill in the missing details by reviewing each test case's finite
element model and input data with accompanying comments.

The ANSYS Verification Manual contains the following contains the following test cases that feature
birth and death:

VM194 - Element Birth/Death in a Fixed Bar with Thermal Loading

Prev Next
Further Guidelines User Routines and Non-Standard Uses
User Routines and Non-Standard Uses
Prev Advanced Analysis Techniques Guide Next

Chapter 8. User Routines and Non-Standard Uses

8.1. User Programmable Features (UPFs)
The ANSYS program's open architecture allows you to link it to your own FORTRAN routines and
subroutines. In fact, some of the ANSYS features you see today as "standard" offerings originally
started as user-programmed features. This chapter provides a basic introduction to UPFs. For
detailed information, see the ANSYS Guide to User Programmable Features.

8.1.1. What Are UPFs?

User programmable features (UPFs) are ANSYS capabilities for which you can write your own
FORTRAN routines. UPFs are only available in the ANSYS/Multiphysics, ANSYS/Mechanical,
ANSYS/Structural, ANSYS/PrepPost, and ANSYS/University (Research Faculty/Student version)
products.

UPFs allow you to customize the ANSYS program to your needs, which may be a user-defined
material behavior option, a user element, a user-defined failure criterion (for composites), and so on.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 193 of 207

You can even write your own design optimization algorithm that calls the entire ANSYS program as
a subroutine.

Note

Using UPFs requires extreme care and thought on your part. By linking in your own
FORTRAN routines, you are creating a customized, site-dependent version of the
ANSYS program. In addition, when you use UPFs, you are using ANSYS in a non-
standard way, one that ANSYS Inc.'s Quality Assurance verification testing program
does not cover. You are responsible for verifying that the results produced are
accurate and that the routines you link to ANSYS do not adversely affect other,
standard areas of the program.

You should take special care when using UPFs on parallel systems. You should never
use the /CONFIG command or a config.ans file to activate parallelization on a
system with UPFs.

8.1.2. How to Use UPFs

UPFs can range from a simple element output routine for customized output to a much more
complex user element or user optimization algorithm. Therefore, it is difficult to present the
procedures without describing specific programming details. This section presents a general
sequence of steps to follow. The ANSYS Guide to User Programmable Features contains more detail
on UPFs.

A typical UPF involves the following steps:

1. Design and program the desired user routine in FORTRAN 77. The source codes for all user
routines are available on your ANSYS distribution medium. Most of them demonstrate at least
simple functionality, so you should obtain a printout of the appropriate routine before you
begin programming.

2. Compile and link your user routine into the ANSYS program. The ANSYS Installation and
Configuration Guide describes how to do this on your system.

3. You may want to verify that the changes you have made do not affect other, standard ANSYS
features. You can do this, for example, by running a set of ANSYS Verification Manual
problems. Input for these problems is also available on your ANSYS distribution medium.

4. Verify the user routine using whatever procedures you feel are adequate. Remember that this is
your responsibility.

The ANSYS program activates some UPFs, such as user elements, automatically when you use
them. For example, to activate a user element, all you need to do is specify it as one of the element
types in the model (ET command or menu path Main Menu>Preprocessor>Element
Type>Add/Edit/Delete), set the element type attribute pointer (TYPE command or menu path Main
Menu>Preprocessor> -Attributes-Define>Default Attribs), and define elements using the solid
modeling [AMESH, VMESH, etc.] or direct generation [ET, etc.] method.

For some other UPFs, you need to take an additional action to activate them. That is, you need to
issue the USRCAL command or pick one of the following menu paths: Main
Menu>Preprocessor>Loads>-Load Step Opts-Other>User Routines or Main
Menu>Solution>Other>User Routines. If you fail to do this, standard ANSYS logic will be used by
default.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 194 of 207

For example, when you apply a convection load, the default is to use standard ANSYS logic even if
you have a user convection routine linked in. You must activate the appropriate user routine with the
USRCAL command if you want the user convection routine to be used. Refer to the USRCAL
command description for a list of user routines affected by the command. Use the NSVR command
to define the number of extra variables that need to be saved for such user-programmable element
options as user plasticity. (The NSVR command has no equivalent GUI path.)

Another useful command is /UCMD , which allows you to make your own command out of a user
routine. (The /UCMD command has no GUI equivalent.) Suppose you link in a user routine for a
parabolic distribution of pressure. If you name the routine USERnn (where nn = 01 to 10), you can
create your own command to call the routine:

/UCMD,PARAB,1

PARAB now becomes a valid ANSYS command that simply calls the user routine USER01. You
can call up to ten such user routines as commands. By including /UCMD commands in your start-up
file ( START5x.ANS), you can make the user routines available in all your ANSYS sessions.

8.1.3. A Brief Description of the Types of UPFs Currently Available

User element - This is a user-written element type you can add to the ANSYS element library and
use as a "regular" element. You can create up to six independent element types (USER100 -
USER105). Example copies of the routines for MASS21, the structural mass element, and LINK8,
the 3-D spar, are included in ANSYS distribution media for demonstration purposes.

User element coordinate system orientation - This is available for the following element types:
SHELL43, SHELL63, SHELL91, SHELL93, SHELL99 , SHELL181, SOLID46, and SOLID64. For
the layered elements (SOLID46, SHELL91, and SHELL99 ), you can also customize the orientation
of a given layer.

User real constants - Elements COMBIN7 and COMBIN37 allow the input real constants to be
modified based on your own nonlinear function.

User thickness - This is available for SHELL181.

User stresses - This is available for PLANE2, PLANE42, SOLID45, PLANE82, SOLID92,
SOLID95, LINK180, SHELL181, PLANE182, PLANE183, SOLID185, SOLID186, SOLID187,
BEAM188, and BEAM189.

User coefficient of friction - This is available for general contact elements CONTAC48 and
CONTAC49.

User plasticity law - This allows you to calculate plastic strains and form the tangent stress-strain
matrix at an integration point based on your own plasticity law.

User creep equation - This allows you to specify your own creep equation.

User swelling law - If you need to account for swelling in an analysis (such as due to neutron
bombardment), you must write the appropriate swelling law as a user routine. No built-in swelling
laws are available in the ANSYS program.

User hygrothermal growth - This allows you to induce growth caused by moisture content, and is
available for the SHELL91 element.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 195 of 207

User hyperelasticity - This is available for hyperelastic elements HYPER56, HYPER58, HYPER74,
and HYPER158.

User failure criteria - This is available for the layered elements SOLID46 and SHELL99 . Up to six
user-defined failure criteria can be supplied.

User viscosity - You can define viscosity as a function of pressure, temperature, position, time,
velocity, and velocity gradients for FLUID141 and FLUID142.

User loads - Body loads such as temperatures, heat generations, and fluences (such as neutron flux),
as well as surface loads such as pressures, convections, heat fluxes and charge density may be
defined by way of user-written logic.

User load vector - This allows you to create a complex load vector for the frequency domain logic of
the PIPE59 element. You can use it to represent hydrodynamic forces.

ANSYS as a subroutine - You can call the entire ANSYS program as a subroutine in your own
program, such as a user-written design optimization algorithm.

User optimization - You can replace the ANSYS optimization logic with your own algorithm and
termination logic.

User access at the beginning and end of each ANSYS run solution, load step, substep, and
equilibrium iteration - This allows you to evaluate results and perform any desired calculations
during solution.

Prev Next
Non-standard Uses of the ANSYS
Where to Find Examples
Program
Non-standard Uses of the ANSYS Program
Prev User Routines and Non-Standard Uses Next

8.2. Non-standard Uses of the ANSYS Program

The ANSYS program goes through a rigorous verification testing plan before its release. You can be
reasonably assured of obtaining good results when you exercise documented features using
"standard," recommended procedures. However, in some situations you may need to employ non-
standard techniques - techniques that have not been or cannot be fully tested at ANSYS, Inc. because
of their very nature, such as user-programmable features. Be aware that verifying the results in such
cases is your responsibility.

8.2.1. What Are Non-standard Uses?

The results of non-standard uses of the ANSYS program cannot be predicted. Therefore, ANSYS,
Inc.'s Quality Assurance verification program cannot fully cover such uses. ANSYS, Inc. does not
discourage non-standard uses, but you should exercise caution and your engineering judgment in
these situations. For example, if you write your own user element and use it in an ANSYS analysis,
the results depend primarily on how well the element was programmed. You must verify the results
in such cases, and make sure that other, standard areas of the program are not adversely affected.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 196 of 207

The following is a partial list of non-standard ANSYS features and uses:

? User programmable features (UPFs) - writing your own user routines, linking them into the
ANSYS executable, and using them in an analysis. UPFs are described earlier in this chapter.

? Reading into the ANSYS program an ANSYS file which was created or modified external to
the ANSYS program, for example, a results file or a superelement file created by you or by
another program.

? High-risk capabilities such as the following:

? Changing element real constants in SOLUTION, between load steps. Depending on the

element type being used, the element may not properly use the updated real constant
value.

? Deactivating the cross-reference checking of the solid model [MODMSH,NOCHECK]

(Main Menu>Preprocessor>Checking Ctrls>Model Checking).

? Turning off element shape checking [SHPP,OFF] (Main Menu>

Preprocessor>Checking Ctrls>Shape Checking).

? Using undocumented features, such as an element option not documented in the ANSYS
Elements Reference or a command argument not mentioned in the ANSYS Commands
Reference. Undocumented features, by definition, are "unofficial," unverified, and should be
used with caution.

If the ANSYS program can detect the use of a non-standard feature, it will often issue a warning
message to that effect.

8.2.2. Some Guidelines

Follow these guidelines when you need to use the ANSYS program in a non-standard manner:

? Use your engineering judgment and carefully review the results of the analysis.

? Do not assume that other, standard areas of the program are not affected. Run a few test
problems to verify this.

? If you need to obtain customer support on an analysis involving non-standard use of the
program, be sure to mention the nature and extent of the non-standard feature that you used.

For detailed information on UPFs, see the ANSYS Guide to User Programmable Features.

Prev Next
Improving ANSYS Performance and
User Routines and Non-Standard Uses
Parallel Performance for ANSYS
Improving ANSYS Performance and Parallel Performance for ANSYS
Prev Advanced Analysis Techniques Guide Next

Chapter 9. Improving ANSYS Performance and Parallel

Performance for ANSYS

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 197 of 207

Solving a large model with millions of DOFs can require many hours. To decrease processing time,
ANSYS offers several options to spread model-solving over multiple processors:

? Increasing the default number of processors on certain types of parallel systems for all ANSYS
runs by specifying the NUM_PROC configuration parameter in your config60.ans file.

? Increasing the number of processors on certain types of parallel systems during an ANSYS run
via the /CONFIG,NPROC command or the SETNPROC macro.

? Using one of the multiprocessor solvers from the add-on product Parallel Performance for
ANSYS to further extend the use of multiple processors.

Note

If you are running ANSYS/LS-DYNA, you can use the shared memory parallel
processing (SMP) capabilities of LS-DYNA. For more information, see Section 5.3 in
the ANSYS/LS-DYNA User's Guide.

9.1. Activating Parallel Processing

The following options are available to activate the parallel processing capability of ANSYS:

? If you are running ANSYS on certain multiprocessor systems, you can specify multiple
processors via the NUM_PROC configuration parameter. Place this line (which must begin in
column 1) in your config60.ans file:

NUM_PROC=N

The N value represents the number of processors to use. If N exceeds the number of available
processors, ANSYS uses all available processors.

For more information about the config60.ans file, see The Configuration File in the ANSYS
Basic Analysis Guide.

? Issue the ANSYS /CONFIG command, as follows::

/CONFIG,NPROC,N

The N value represents the number of processors to use. If N exceeds the number of available
processors, ANSYS uses all available processors.

? Issue the SETNPROC macro in one of the following ways:

SETNPROC

Issuing SETNPROC sets the number of processors to the maximum allowed on the system.
(The value may be less than the actual number of processors.)

SETNPROC,N

Issuing SETNPROC,N sets the number of processors to N. (If N exceeds the number of
available processors, ANSYS uses all available processors.)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 198 of 207

SETNPROC,-P

Issuing SETNPROC,-P sets the number of processors to the maximum allowed on the system
minus the number specified by P (but never less than one processor).

ANSYS recommends setting N no higher than the number of available processors minus one. For
example, on a four-processor system, set N to 3.

When you invoke parallel processing, the CPU time indicated at the end of the ANSYS execution is
the sum of all CPU times used for the ANSYS process.

For a complete up-to-date list of systems on which ANSYS supports parallel processing, point your
Web browser to the following URL:

http://www.ansys.com/services/hardware_support/parallel/index.htm

9.1.1. System Notes

The number of processors that the ANSYS program uses is limited to the lesser of one of the
following:

? The actual quantity of CPUs

? The value specified via the /CONFIG,NPROC command or the NUM_PROC configuration
parameter (if used).

Note

Issuing the /CONFIG,NPROC command to change the number of processors

overrides the number specified via the NUM_PROC configuration parameter.

You can specify multiple settings for the number of CPUs to use during an ANSYS session.
However, on SGI systems (and possibly others), ANSYS recommends that you issue the
use /CLEAR command before resetting the number of processors for subsequent analyses.

On SGI systems, the number of processors is limited to the lesser of the actual quantity of CPUs and
the value of the environment variable MP_SET_NUMTHREADS (if used).

On Sun Solaris systems, if you set the number of processors to more than one (via
the /CONFIG,NPROC command or the NUM_PROC configuration parameter), the ANSYS program
activates parallel processing and uses all available processors.

On IBM AIX systems, verify that the host.list file is consistent with the number of processors
specified. For more information, see the IBM Parallel Environment for AIX documentation,
Operation and Use, Volume 1, Using the Parallel Operating Environment.

Prev Next
Non-standard Uses of the ANSYS Overview of Parallel Performance for
Program ANSYS
Overview of Parallel Performance for ANSYS
Prev Improving ANSYS Performance and Parallel Next

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 199 of 207

Performance for ANSYS

9.2. Overview of Parallel Performance for ANSYS

The Parallel Performance for ANSYS add-on consists of two solvers which can run large models
over multiple processors:

? Algebraic Multigrid iterative equation solver (AMG) -- Solves static or transient analyses over
multiple processors on the same system.

? Distributed Domain Solver (DDS) -- Solves static or transient analyses over multiple systems.

Parallel Performance for ANSYS lets you solve models of ten million DOFs or more by using
multiple processors so that you can run the models when the systems are available.

The rest of this chapter introduces Parallel Performance for ANSYS, and shows you how ANSYS
uses parallel processing to reduce solution time. It does not document the configuration of the
Parallel Performance for ANSYS solvers. If you want to run Parallel Performance for ANSYS
solvers on Windows systems, see Configuring Parallel Performance Solvers in the ANSYS
Installation and Configuration Guide for Windows. If you want to run Parallel Performance for
ANSYS solvers on UNIX systems, see Configuring Parallel Performance Solvers in the ANSYS
Installation and Configuration Guide for UNIX .

9.2.1. What Is Parallel Performance?

The Parallel Performance for ANSYS add-on is a powerful tool consisting of two solvers which can
solve large models over multiple processors. When ANSYS solves parts of the model in parallel, the
total solution time is reduced.

Occasionally, the memory limitations of ANSYS can make it impossible to solve very large models.
For greater control over running deferred jobs and over recovering from systems failures during
deferred jobs, we recommend using LSF/Batch from Platform Computing or a similar batch
management product. For more information on this product, see Accessing the ANSYS LSF/Batch
Facility in the ANSYS Operations Guide.

The Algebraic Multigrid (AMG) solver is an iterative solver that uses multiple processors on the
same system to speed up processing time.

The Distributed Domain Solver (DDS) solver breaks large problems into smaller domains, transfers
the domains to multiple processors, solves each domain, and creates a complete solution to the
model. Because the solutions are running in parallel, the whole model solution takes much less time
to solve. The memory required is also distributed over multiple systems.

Before you can run the DDS solver, you must first install Message Passing Interface (MPI) software
on every system on which you intend to use the DDS solver. Throughout ANSYS documentation,
the term “MPI software” is used to refer to the MPI software that is right for your platform. See the
appendix on Configuring Parallel Performance Solvers in the ANSYS Installation and Configuration
Guide for UNIX or the ANSYS Installation and Configuration Guide for Windows for more
information on acquiring MPI and on platform-specific MPI issues.

9.2.2. About the Algebraic Multigrid (AMG) Solver

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 200 of 207

The Algebraic Multigrid (AMG) solver, which is based on the multi-level method, is an iterative
solver that you can use in single- and multiprocessor environments. To use this solver, you must
have a license for the Parallel Performance for ANSYS add-on.

The AMG solver is intended for problems in which the PCG and ICCG solvers would have difficulty
converging (for example, large, ill-conditioned problems where the ill-conditioning is due to large
element aspect ratios within a mesh, or cases in which shell or beam elements are attached to solid
elements). In terms of CPU time when used in a single-processor environment, the AMG solver
performs better than the PCG and ICCG solvers for ill-conditioned problems, and it delivers about
the same level of performance for ordinary problems. In a multiprocessor environment, the AMG
solver scales better than the PCG and ICCG solvers on shared memory parallel machines. It also
handles indefinite matrix problems for nonlinear analyses. However, the AMG solver typically uses
50% more memory than the PCG solver.

The AMG solver is available only for static analyses and full transient analyses. (These analyses can
be linear or nonlinear.) In addition, the efficiency of the AMG solver is limited to single-field
structural analyses in which the solution DOFs are combinations of UX, UY, UZ, ROTX, ROTY,
and ROTZ. For analyses such as single-field thermal analyses in which the solution DOF is TEMP,
the AMG solver is less efficient than the PCG or ICCG.

9.2.3. About the Distributed Domain Solver (DDS)

The Distributed Domain Solver (DDS) is a solver which decomposes large models into smaller
domains, then sends those domains to multiple processors. To use this solver, you must have a
license for the Parallel Performance for ANSYS add-on.

The DDS solver is a scalable solver intended for large static or full transient analyses, with
symmetric matrices that do not involve prestress, inertia relief, coupling, constraint equations or
problems using the probabilistic design system (PDS). This solver cannot be used with models
containing LINK elements, CONTAC elements, TARGE elements, p-elements, superelements,
PRETS179 elements, or with pure truss problems. This solver speeds up solving time by using
multiple processors on multiple homogeneous systems.

The DDS solver works by dividing the meshed model into domains without user intervention. The
number of DOFs in any domain is generally under 10,000 and is ideally around 2,000. The number
of domains is almost always greater than the number of processors available to solve the problem.
During the solution phase, the solution is launched on the host system and on all designated systems
simultaneously. Each processor solves its own set of domains, communicating with the other
processors. The host system then builds the total DOF solution, and the analysis continues on the
host system only.

Note

The probabilistic design system can also distribute solutions across processors, but
uses a system independent of DDS. See Probabilistic Design in the ANSYS Advanced
Analysis Techniques Guide for more information.

Prev Next
Improving ANSYS Performance and
Running Parallel Performance Solvers
Parallel Performance for ANSYS
Running Parallel Performance Solvers
Improving ANSYS Performance and Parallel

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 201 of 207

Prev Performance for ANSYS Next

9.3. Running Parallel Performance Solvers

Since both the AMG and DDS solvers are part of the Parallel Performance for ANSYS, a license for
Parallel Performance for ANSYS is required in order to run either solver.

For optimal performance on solving a large model, close down all other applications before
launching ANSYS with a multiprocessor solver. We recommend running the Distributed Domain
Solver when the network is not too busy, such as at night and on weekends.

9.3.1. Using the Algebraic Multigrid (AMG) Solver

The AMG solver calculates solutions over multiple processors on the same system. If you have a
Parallel Performance for ANSYS license, you can use the AMG solver by following these steps:

1. When starting your ANSYS session, specify the Parallel Performance for ANSYS add-on from
the operating system command line:

ansys60 -pp

2. When defining the analysis, choose the AMG solver [EQSLV ,AMG].

3. (Optional.) To use the AMG solver in a multiprocessor environment, activate parallel

processing. You can do this either by issuing the /CONFIG,NPROC, VALUE command, or by
executing the SETNPROC macro. ( VALUE indicates the number of processors to be used, and can
be a maximum of 16. The AMG solver is expected to scale linearly when used with up to 8
processors. When the AMG solver is used with more than 8 processors, any additional
scalability is minimal; the larger the problem, the higher the potential for scalability.) The
SETNPROC macro defines the maximum number of processors available.

9.3.2. Using the Distributed Domain Solver (DDS)

If you have a Parallel Performance for ANSYS license, you can use the DDS solver by following
these steps:

1. Verify that you have access (valid logins, etc.) to all remote systems on which you will be
running. On Windows systems, verify that each machine being used has a working directory
with a path identical to that on the host machine.

2. When starting your ANSYS session, specify the Parallel Performance for ANSYS add-on from
the operating system command line:

ansys60 -pp

3. When defining the analysis, choose the DDS solver [EQSLV,DDS,TOLER].

4. Specify system information for the multiprocessor solution run [DDSOPT,Configopt].

9.3.2.1. Setting Options for the Distributed Domain Solver (DDS)

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 202 of 207

When you invoke DDSOPT, you can enter one of three configuration options:

? local -- Sends domains to processors on the local system only.

? file -- Calls config60.dds to set processor information for the DDS solver. Use of the
config60.dds file is explained in more detail in later sections. See your MPI software
documentation for more information about MPI commands and options.

Note

The file option will not run multiple machines across a network on Sun and
IBM platforms. Use the script option.

? script -- Calls the ansdds60 executable ( ansdds.bat on Windows platforms) to set

processor information for the DDS solver. This option should only be used by people
experienced with writing MPI applications.

Note

The script option will not run multiple machines across a network on the HP
platform. Use the file option.

Using the config60.dds File

The file option specifies that MPI should look in the config60.dds file for configuration options to
define processor information.

In general, the config60.dds file contains the following information:

SystemName

The name of the system to send DDS solver processes to.

NumProcessors

The number of processors on the named system to send DDS solver processes to.

ExecutablePath

The full path name to the ansdds executable on the local system. Do not use relative path
names.

Also on Windows system, if the config60.dds file contains entries for multiple machines, the
master machine must be the first machine referenced in the file.

On a Windows system, the config60.dds file format is:

SystemName NumProcessors ExecutablePath

The following example shows two systems, DEVNT and DEVSERV, with 2 and 1 processor,
respectively:

DEVNT 2 "C:\Program Files\ANSYS Inc\ANSYS60\bin\intel\ansdds.exe"

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 203 of 207

DEVSERV 1 "C:\Program Files\ANSYS Inc\ANSYS60\bin\intel\ansdds.exe"

On UNIX platforms, the format of the config60.dds file varies, depending on the platform. Refer to
the MPI documentation for the specific format of each platform. The following examples show a
config60.dds file on an SGI platform:

<server>,<server2> -np 2 /ansys60/bin/sgi64/ansdds.e60

Or, to use a different number of processors on each server, use this format:

<server> -np 2 /ansys60/bin/sgi64/ansdds.e60 : <server2> -np 4 /ansys60/bin/sgi64/ansdds.e6

Note

The -np immediately before the number of processors is an mpirun command

argument indicating "number of local processors." Other mpirun command
arguments can be used in this position in the file. For information about the mpirun
command arguments, type:

man mpirun

from the UNIX command prompt.

When you invoke the DDSOPT,file command, ANSYS searches your directories in the following
order to look for the config60.dds file:

1. Your current directory

2. Your home directory

3. /ansys60/docu

Using the mpitest.exe Program

The mpitest.exe is a simple ping program to verify that the mpi is set up correctly. Execute is as
follows

mpirun -np 2 ../ansys60/bin/mpitest.exe

You must use an even number of nodes (we used 2 in the above example).

Using the ansdds60 Script

If you are familiar with writing MPI applications, you can create an ansdds60/ansdds.bat script to
further customize the application. When you invoke DDSOPT,script, ANSYS searches your
directories in the following order to look for the ansdds60 script:

1. Your current directory

2. Your home directory

3. The /ansys60/bin directory

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 204 of 207

Note

If the ansdds60 script does not have execution permission, then ANSYS
automatically searches in /ansys60/bin.

If you are running an SC cluster on an Alpha, you must rename ansdds_sc.e60 to ansdds.e60 to
use the ansdds60 script. You must use this script (DDSOPT,script).

Here is a sample ansdds60 script for a UNIX platform. This script defines a DDS job using one
processor on each of three SGI machines:

#!/bin/csh
# mpirun machine1, machine2, machine3 1 \
$ANSYS_DIR/bin/$ANSYS_SYSDIR/ansdds.e60
# exit

For Windows, this script file must be named ansdds.bat . Here is a sample ansdds.bat script for a
Windows platform. This script refers to a configuration file named config60.dds:

mpirun -pg_file config60.dds

This sample script (also for a Windows platform) will run two processes on the first machine and
two on the second machine:

mpirun -np 4 -pg_file host.list "C:\PROGRAM FILES\ANSYS INC

\ANSYS60\bin\intel\ansdds.exe"

and where the host.list file contains:

machine1
machine2

Use the file option to specify exactly how many processes you want sent to certain machines. The
first name in the machines file should be the local hostname.

Considerations

To run a DDS solver process, you must be able to log in to each system listed in the config60.dds
file or ansdds60 script and both ANSYS Release 6.0 and MPI software must be installed on each
system. If you want to see more examples of ansdds60 scripts, you can look in
the ../ansys_inc/ansys60/bin directory for ansdds60.

For both Windows and UNIX environments, the working directory structure must be the set up the
same for all the distributed machines as it is on the host machine.

Windows NT platforms with most MPI software have a limit of eight processors for DDS (that is,
eight CPUs regardless of where they physically exist in your configured network). You can can
contact MPI Software Technology to purchase an MPI version that allows a higher number of
processors.

9.3.3. Files Written During Distributed Processing

The DDS solver writes out a number of files during the solution phase. Here is a list of those files
and a brief description of their contents:

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 205 of 207

Jobname.DCS Contains the status of the ANSDDS run

Jobname.dds
Contains the nodal coordinates, element connectivity and displacement boundary
conditions
Jobname.erot Contains element matrices and load vectors
scratch.dds Contains the name of the Jobname, Jobname.dds, and Jobname.erot files

Prev Next
Overview of Parallel Performance for
Troubleshooting Parallel Performance
ANSYS
Troubleshooting Parallel Performance
Improving ANSYS Performance and Parallel
Prev Next
Performance for ANSYS

9.4. Troubleshooting Parallel Performance

9.4.1. General Problems

Restarting stopped jobs

You must restart a job from the beginning; it is impossible to restart a job from the point where it
ended.

Recovering from a computer or network crash

Be sure to kill any lingering processes (UNIX: type kill -9 from command level, Windows: invoke
the Task Manager) on all processors and start the job again.

9.4.2. Resolving Errors

No permission to system

This error can occur if you do not have log-in access to a remote system where the DDS solver is
supposed to run. If you use UNIX, you can experience this problem if you have not set permissions
on the .rhosts file properly. Before starting a DDS run, be sure you have access to all the remote
systems you have specified. If you run UNIX, be sure you have run the following command:

chmod 600 .rhosts

DDS will not run on a Windows system

Windows requires that each machine being used must have a working directory with a path identical
to that on the host machine.

DDS did not complete successfully. Check your DDS and MPI settings.

For Windows systems: The machines file must be present in either the ANSYS working directory or
in the directory set by the MPI_HOME environment variable. Be certain this file does not have an
extension. This file contains a list of the host machines, and the first machine on the list must be the
local host.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 206 of 207

DDS solver failure, no U file detected

If you receive this error with a Windows error of The instruction at “nnnnnnnnnn” referenced
memory at “mmmmmmmmmm ”. The memory could not be “written”. Click on OK to
terminate the program, verify that all machines are running the same version of MPI/Pro. Also,
you might have provided MPI/Pro with an invalid network password.

Check that the directory structures on subsequent machines are identical to the host machine's
structure. (This can occur while running the SCRIPT option.)

Slave Machine Name:_desktop failed.

User name unable to CreatProcess for \\pathname\ansdds.exe

The directory name is invalid.

Local Machine Name: Process exited with code 1

Local Machine Name: Process exited with code 1

Check that the directory structures are identical in subsequent machines as it is on the host machine.

Machine Name: Access violation

This can be caused when more than one host tries to access the same machine while using an older
version of MPI/Pro. The version of MPI/Pro provided with ANSYS Release 6.0 resolves this
problem.

The Distributed Domain Solver execution script not found.

Looking for: .\ansdds.bat or M:\ansdds.bat

Verify that the ansdds.bat file exists and is in the correct directory.

MPI configuration file config60.dds not found.

Be certain that the config60.dds file exists and is in the correct directory.

9.4.3. Platform-Specific Problems

IBM Problems

Solver aborts

If you try to access more nodes than available the solver will hang.

SGI Problems

ansdds process aborts without warning

This can happen on a single system when at least two processors have been specified. There may be
insufficient memory allocated, if virtual swap is in use. Display the vswap setting for the system
(from command level, type: /sbin/swap -1) and, if necessary, reallocate memory.

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004
Advanced Analysis Techniques Guide Page 207 of 207

SUN Problems

no_mp_jobs: Not enough resources available

This can happen when you request more processors than are available. This happens via ansdds60 if
-np is greater than the number of avialable processors. Use

mpinfo -N

to check the number of available processors.

DDS solver failure

If you are running on a SUN cluster, you may require a shared file system to enable file availability
to ANSYS from nodes within the cluster when running the DDS solver.

Prev Next
Running Parallel Performance Solvers APDL Programmer's Guide

file://D:\Documents%20and%20Settings\Administrator\Local%20Settings\Temp\~hh... 20.05.2004