DesignXplorer Optimization Tutorials

ANSYS, Inc. Release 2024 R1

Southpointe January 2024
2600 Ansys Drive
Canonsburg, PA 15317 ANSYS, Inc. and
ansysinfo@ansys.com ANSYS Europe,
Ltd. are UL
http://www.ansys.com registered ISO
(T) 724-746-3304 9001: 2015
(F) 724-514-9494 companies.
Copyright and Trademark Information

© 2024 ANSYS, Inc. Unauthorized use, distribution or duplication is prohibited.

ANSYS, Ansys Workbench, AUTODYN, CFX, FLUENT and any and all ANSYS, Inc. brand, product, service and feature
names, logos and slogans are registered trademarks or trademarks of ANSYS, Inc. or its subsidiaries located in the
United States or other countries. ICEM CFD is a trademark used by ANSYS, Inc. under license. CFX is a trademark
of Sony Corporation in Japan. All other brand, product, service and feature names or trademarks are the property
of their respective owners. FLEXlm and FLEXnet are trademarks of Flexera Software LLC.

Disclaimer Notice

THIS ANSYS SOFTWARE PRODUCT AND PROGRAM DOCUMENTATION INCLUDE TRADE SECRETS AND ARE CONFID-
ENTIAL AND PROPRIETARY PRODUCTS OF ANSYS, INC., ITS SUBSIDIARIES, OR LICENSORS. The software products
and documentation are furnished by ANSYS, Inc., its subsidiaries, or affiliates under a software license agreement
that contains provisions concerning non-disclosure, copying, length and nature of use, compliance with exporting
laws, warranties, disclaimers, limitations of liability, and remedies, and other provisions. The software products
and documentation may be used, disclosed, transferred, or copied only in accordance with the terms and conditions
of that software license agreement.

ANSYS, Inc. and ANSYS Europe, Ltd. are UL registered ISO 9001: 2015 companies.

U.S. Government Rights

For U.S. Government users, except as specifically granted by the ANSYS, Inc. software license agreement, the use,
duplication, or disclosure by the United States Government is subject to restrictions stated in the ANSYS, Inc.
software license agreement and FAR 12.212 (for non-DOD licenses).

Third-Party Software

See the legal information in the product help files for the complete Legal Notice for ANSYS proprietary software
and third-party software. If you are unable to access the Legal Notice, contact ANSYS, Inc.

Published in the U.S.A.

Table of Contents
Introduction ............................................................................................................................................... 1
What is DesignXplorer? ........................................................................................................................... 1
DesignXplorer Features .......................................................................................................................... 2
“What If?” Studies ............................................................................................................................. 2
Parameters Correlation ..................................................................................................................... 3
Design of Experiments ...................................................................................................................... 5
Response Surface ............................................................................................................................. 5
Features that Use Response Surface Data .................................................................................... 6
Getting Started with DesignXplorer ........................................................................................................ 7
DesignXplorer Workflow ................................................................................................................... 7
Parametrization Basics ...................................................................................................................... 9
DesignXplorer Workspaces ................................................................................................................... 10
Parameter Set Workspace ................................................................................................................ 11
DesignXplorer Workspaces ............................................................................................................. 14
Parameters Correlation Workspace ............................................................................................ 15
Design of Experiments Workspace ............................................................................................ 16
Response Surface Workspace .................................................................................................... 17
Optimization Workspace ........................................................................................................... 18
Downloading Input Files for Tutorials .................................................................................................... 18
What Have You Learned? ....................................................................................................................... 19
Performing a Parametric “What If” Study ................................................................................................. 21
Getting Started ..................................................................................................................................... 21
Promoting Dimensions and Solution Results to Input and Output Parameters ........................................ 22
Promoting Dimensions in DesignModeler to Input Parameters ........................................................ 22
Promoting a Property and Solution Results in Mechanical to Output Parameters .............................. 24
Working with Design Points in Workbench ............................................................................................ 26
Adding Parameter Charts ...................................................................................................................... 27
Adding a Parameter Chart ............................................................................................................... 27
Adding a Parameters Parallel Chart ................................................................................................. 28
Viewing the Solution ............................................................................................................................ 29
Performing a Parameters Correlation ...................................................................................................... 33
What is a Parameters Correlation? ......................................................................................................... 33
Getting Started ..................................................................................................................................... 36
Defining Parameters ............................................................................................................................. 37
Generating Design Points ..................................................................................................................... 37
Performing the Parameters Correlation ................................................................................................. 37
Viewing the Correlation Matrix and Charts ............................................................................................ 39
Using a Design of Experiments ................................................................................................................. 47
What is a Design of Experiments? .......................................................................................................... 47
DOE Types ............................................................................................................................................ 48
Getting Started ..................................................................................................................................... 56
Working with the DOE .......................................................................................................................... 59
Working with the Response Surface ...................................................................................................... 63
Solving for the Desired Design Point ..................................................................................................... 68
Performing a Goal-Driven Optimization .................................................................................................. 71
What is Goal-Driven Optimization? ........................................................................................................ 71
Getting Started ..................................................................................................................................... 72
Promoting Properties and Results in Mechanical to Parameters ............................................................. 73
Performing the Optimization ................................................................................................................ 75

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. iii
DesignXplorer Optimization Tutorials

Adding the Goal-Driven Optimization System ................................................................................. 75

Configuring and Updating the Design of Experiments Cell ............................................................... 75
Configuring and Updating the Response Surface Cell ...................................................................... 78
Configuring and Updating the Optimization Cell ............................................................................. 80
Viewing Goal-Driven Optimization Charts ............................................................................................. 82
Viewing Candidate Points ..................................................................................................................... 82
Viewing the Solution ............................................................................................................................ 84
Viewing Results for the Most Viable Candidate Point ........................................................................ 84
Viewing Results for the Response Point ........................................................................................... 87
Viewing Results in Mechanical ........................................................................................................ 89
Using ACT to Expose MATLAB Optimizers in DesignXplorer .................................................................... 91
Using Adaptive Single-Objective Optimization ....................................................................................... 93
What is Adaptive Single-Objective Optimization? .................................................................................. 93
Problem Definition ............................................................................................................................... 94
Problem Setup ...................................................................................................................................... 94
Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimization ...................... 95
Run the Response Surface Optimization .......................................................................................... 96
Run the Direct Optimization .......................................................................................................... 103
How Effective was the Approach Used for Scenario 1? ................................................................... 105
Scenario 2: NLPQL Direct Optimization to NLPQL Direct Optimization .................................................. 106
Run the First NLPQL Direct Optimization ....................................................................................... 106
Run the Second NLPQL Direct Optimization .................................................................................. 108
How Effective was the Approach Used for Scenario 2? ................................................................... 111
Scenario 3: Screening Direct Optimization to NLPQL Direct Optimization ............................................. 111
Run the Screening Direct Optimization .......................................................................................... 111
Run the NLPQL Direct Optimization .............................................................................................. 113
How Effective was the Approach Used for Scenario 3? ................................................................... 115
Scenario 4: Adaptive Single-Objective Direct Optimization ................................................................... 116
Run the ASO Direct Optimization System ...................................................................................... 116
How Effective was the Approach Used for Scenario 4? ................................................................... 120
Time to Spare? .................................................................................................................................... 121
What Have You Learned? ..................................................................................................................... 121
Different Types of Optimizations ................................................................................................... 121
Different Optimization Methods .................................................................................................... 122
Using Adaptive Multiple-Objective Optimization .................................................................................. 123
What is Adaptive Multiple-Objective Optimization? ............................................................................. 123
Problem 1 ........................................................................................................................................... 123
Problem 1 Definition ..................................................................................................................... 123
Problem 1 Project Setup ............................................................................................................... 124
Run the MOGA Direct Optimization ............................................................................................... 125
Run the AMO Direct Optimization ................................................................................................. 128
Problem 2 ........................................................................................................................................... 131
Problem 2 Definition ..................................................................................................................... 131
Problem 2 Project Setup ............................................................................................................... 132
Run the MOGA Direct Optimization ............................................................................................... 134
Run the AMO Direct Optimization ................................................................................................. 137
What Have You Learned? ..................................................................................................................... 141
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization ............................... 143
Getting Started ................................................................................................................................... 143
What is MISQP? ............................................................................................................................. 143
Problem Definition ....................................................................................................................... 144

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
iv of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Optimization Tutorials

Project Setup ................................................................................................................................ 144

Direct Optimization Using Screening ................................................................................................... 145
Response Surface Optimization to Direct Optimization Using Screening and MISQP ............................. 147
Run Screening in a Response Surface Optimization ........................................................................ 148
Run MISQP in a Response Surface Optimization ............................................................................. 151
Run MISQP in a Direct Optimization .............................................................................................. 154
Direct Optimization Using MISQP ........................................................................................................ 156
Run MISQP in a Direct Optimization with a Maximum of 20 Iterations ............................................ 157
Run MISQP in a Direct Optimization with a Maximum of 50 Iterations ............................................ 159
What Have You Learned? ..................................................................................................................... 161

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. v
 Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
vi of ANSYS, Inc. and its subsidiaries and affiliates.
Introduction
This guide provides several tutorials for using Ansys DesignXplorer to analyze and optimize design
spaces. The following topics introduce DesignXplorer and explain how to download the input files that
you will need for the tutorials:
What is DesignXplorer?
DesignXplorer Features
Getting Started with DesignXplorer
DesignXplorer Workspaces
Downloading Input Files for Tutorials
What Have You Learned?

Note:

While this guide does not include tutorials for producing ROMs (reduced order models),
you can find comprehensive ROM production and consumption information in "Using
ROMs" in the DesignXplorer User's Guide.

What is DesignXplorer?
DesignXplorer is a tool that uses response surfaces to efficiently analyze a design space. With
DesignXplorer, you can:

• Examine and understand the performance at other design or operating conditions

• Find the conditions which give the best performance

• Determine the key parameters influencing your design

• Evaluate the robustness of your design

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 1
Introduction

DesignXplorer Features
DesignXplorer offers a variety of features to help you to examine and better understand your designs.

This section provides a brief introduction to these features:

“What If?” Studies
Parameters Correlation
Design of Experiments
Response Surface

“What If?” Studies

Ansys Workbench offers the ability to perform a “What if?” study, which runs through a list of manually
specified design points. You don’t need a DesignXplorer license to perform such a study. A “What
if?” study provides you with preliminary information that you can then use in DesignXplorer for more
in-depth analysis.

When you have a Workbench project open, in the Project Schematic, double-clicking the Parameter
Set bar opens it. In the Table pane, you can see the design points that have been generated. Right-
clicking a design point allows you to select an option from the context menu.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
2 of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Features

For more information, see the tutorial Performing a Parametric “What If” Study (p. 21).

Parameters Correlation
A parameters correlation performs simulations on a random sampling of the design space to identify
the correlations between all parameters in the project. A linear association between parameters is
evaluated using the Spearman’s or Pearson’s product-moment coefficient. A correlation or sensitivity
matrix is generated to demonstrate the correlation between input and output parameters and the
sensitivity of output parameters to input parameters.

A parameters correlation allows you to:

• Determine which input parameters have the most (and the least) impact on your design.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 3
Introduction

• Identify the degree to which the relationship is linear or quadratic.

For more information, see the tutorial Performing a Parameters Correlation (p. 33).

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
4 of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Features

Design of Experiments
DesignXplorer uses a technique called a Design of Experiments (DOE) to identify design points with
parameter combinations for exploring the solution space most efficiently. The most efficient solution
uses the fewest number of design points. A DOE works best with fewer than 20 parameters.

DesignXplorer makes several DOE types available so that you can select the type best suited to your
project and purposes. Available DOE types include:

• Central Composite Design (CCD)

• Optimal Space-Filling Design

• Box-Behnken Design

• Custom

• Custom + Sampling

• Sparse Grid Initialization

• Latin Hypercube Sampling Design

These images show examples of design points generated with various DOE types:

For more information, see the tutorial Using a Design of Experiments (p. 47).

Response Surface
From the design points identified in a DOE, DesignXplorer builds a continuous response surface. Just
as there are several DOE types, there are several response surface types:

• Genetic Aggregation

• Full 2nd Order Polynomials

• Kriging

• Non-Parametric Regression

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 5
Introduction

• Neural Network

• Sparse Grid

The following images are response surfaces generated from the design points shown in the previous
topic:

Features that Use Response Surface Data

Because the response surface provides approximated values for output parameters, other
DesignXplorer features can quickly explore and use the data that it provides. Examples of features
that make use of response surface data include:

• Min/Max Search

• Response surface charts

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
6 of ANSYS, Inc. and its subsidiaries and affiliates.
 Getting Started with DesignXplorer

• Sensitivity charts

• Optimization Tradeoff charts

• Robustness of design analyses

Getting Started with DesignXplorer

This section describes the DesignXplorer workflow and parametrization basics:
DesignXplorer Workflow
Parametrization Basics

DesignXplorer Workflow
The workflow for using DesignXplorer basically consists of three steps:

1. Create parameters.

2. Add a DesignXplorer system to the Workbench project.

3. Work through the DesignXplorer system.

Step 1: Create Parameters

To work with DesignXplorer, you must have parameters in your project. Thus, your first step is to
parameterize dimensions, boundary conditions, and such.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 7
Introduction

Once parameters are created, the Parameter Set bar is added to the Workbench Project Schematic.
The following figure shows two values being parameterized in Mechanical and the subsequent inclusion
of the Parameter Set bar in the Project Schematic.

Step 2: Add a DesignXplorer System to the Project

Following parameter creation, you can add one or more DesignXplorer systems to the project. To do
this, in the Workbench Toolbox under Design Exploration, locate the desired system and then drag
it on the Project Schematic, dropping it under the Parameter Set bar. In the following figure, a
Response Surface Optimization system has been added to the Project Schematic.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
8 of ANSYS, Inc. and its subsidiaries and affiliates.
 Getting Started with DesignXplorer

Step 3: Work Through the DesignXplorer System

After adding a DesignXplorer system to the Project Schematic, you must work through its components,
which are the individual cells in the system, in sequence from top to bottom. Each type of design
exploration system has a different workflow:

The Direct Optimization and Parameters Correlation systems are the only ones that do not have
Design of Experiments cells. For all other DesignXplorer systems, the Design of Experiments cell
is first in the sequence. In the following systems, the Response Surface cell is second: Response
Surface and Response Surface Optimization. DesignXplorer systems can share and reuse Design
of Experiments and Response Surface cells.

Parametrization Basics
To work with DesignXplorer, you must have parameters in your project. Parameters can be defined
not only in Workbench applications but also in numerous external CAD packages.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 9
Introduction

For more details on defining parameters, see Parametrization Basics in the Workbench User's Guide.

DesignXplorer Workspaces
When you open the Parameter Set bar or a DesignXplorer system cell, you will see that the layouts of
their workspaces are very similar.

Workspace Access
To open the Parameter Set bar or a DesignXplorer cell, double-click it.

Workspace Layout
Each workspace for a DesignXplorer cell includes the Toolbox with chart options and the following
panes:

• Outline

• Properties

• Table

• Chart

Workspaces are connected between the various Workbench native applications. For example, if you
close the chart on one workspace, it is also closed on the other workspaces. To reset the workspace to
the default layout, select View → Reset Workspace.

The following topics show the various workspaces you’ll use when working in DesignXplorer:
Parameter Set Workspace
DesignXplorer Workspaces

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
10 of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Workspaces

Parameter Set Workspace

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 11
Introduction

Parameter Set Table of Design Points

Parameter Set Charts

The charts available from the Parameter Set bar allow you to do the following things:

• Configure and plot an input parameter versus an output parameter in XY plots

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
12 of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Workspaces

• See how output parameters vary for design point combinations

• See the chart output versus the input parameter

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 13
Introduction

DesignXplorer Workspaces
This section includes images of the following DesignXplorer workspaces:
Parameters Correlation Workspace
Design of Experiments Workspace
Response Surface Workspace
Optimization Workspace

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
14 of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Workspaces

Parameters Correlation Workspace

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 15
Introduction

Design of Experiments Workspace

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
16 of ANSYS, Inc. and its subsidiaries and affiliates.
 DesignXplorer Workspaces

Response Surface Workspace

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 17
Introduction

Optimization Workspace

Downloading Input Files for Tutorials

For the optimization tutorials, download and extract the files that are used as inputs:

1. Click this link to download the ZIP file that contains the input files.

2. Extract these files to a directory that you can access.

Each tutorial will instruct you to navigate to and select one or more files in this directory.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
18 of ANSYS, Inc. and its subsidiaries and affiliates.
 What Have You Learned?

What Have You Learned?

From this tutorial, you've learned the following:

• DesignXplorer uses response surfaces to help you to investigate your design space, letting you find
the combinations of input parameters that provide the best performance, determine how parameters
influence your design, and evaluate the robustness of your design.

• DesignXplorer offers a variety of features to help you to better understand your design, including:

– “What if?” studies

– Parameters correlations

– DOEs

– Response surfaces

– Goal-driven optimizations

• The general workflow for using DesignXplorer consists of the following steps:

– Define parameters.

– Add a DesignXplorer system to the Project Schematic.

– Work through the sequence indicated by the cells in the DesignXplorer system.

• Parameters can be added in most Ansys products and numerous external CAD packages via the user
interface, Mechanical APDL input files, and user-defined expressions.

• The workspaces for the Parameter Set bar and DesignXplorer system cells all have similar layouts.
Each workspace includes the Toolbox with chart options and the following panes:

– Outline

– Properties

– Table

– Chart

• Assuming that you have downloaded and extracted the input files (p. 18) for the subsequent tutorials,
you are now ready to learn more by stepping through example projects.

Note:

As you step through these example projects, your results may not exactly match the results
that are shown in this guide. However, your results should be very similar.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 19
 Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
20 of ANSYS, Inc. and its subsidiaries and affiliates.
Performing a Parametric “What If” Study
This tutorial gives step-by-step instructions for performing a parametric “What If?” study of a hitch re-
ceiver. It investigates the behavior of stress, mass, and deformation in the receiver as geometry para-
meters are changed during vertical loading.

Note:

“What If?” studies use functionality inherent to Ansys Workbench. A DesignXplorer license
is not required to complete this tutorial.

This tutorial covers the following topics:

Getting Started
Promoting Dimensions and Solution Results to Input and Output Parameters
Working with Design Points in Workbench
Adding Parameter Charts
Viewing the Solution

Getting Started
The hitch receiver is composed of a multibody part (three bodies) created in Ansys DesignModeler. The
following figure shows constraints and loads.

In this tutorial, you'll promote the dimensions labeled 1, 2, and 3 in the following figure to input para-
meters.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 21
Performing a Parametric “What If” Study

Opening the Archived Workbench Project

1. Start Workbench and select File → Open.

2. In the file browser, locate and open the archived Workbench project file receiver.wbpz, which
is in the directory to which you extracted the input files (p. 18).

A dialog box appears for saving this archived file to a standard Workbench project file.

3. Save the project as reciever.wbpj in either this same directory or another directory.

Promoting Dimensions and Solution Results to Input and Output Para-

meters
The following topics explain how to promote dimensions and solution results in Ansys DesignModeler
and Mechanical to input parameters and output parameters:
Promoting Dimensions in DesignModeler to Input Parameters
Promoting a Property and Solution Results in Mechanical to Output Parameters

Promoting Dimensions in DesignModeler to Input Parameters

To promote dimensions in DesignModeler to input parameters, do the following:

1. On the Project Schematic, double-click the Geometry cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
22 of ANSYS, Inc. and its subsidiaries and affiliates.
 Promoting Dimensions and Solution Results to Input and Output Parameters

DesignModeler starts.

2. In the Tree Outline, select Gusset.

3. In the Details View, select the check box for FD1, Depth (>0) to promote it to an input parameter.

A dialog box opens for specifying the parameter name.

4. Type thick_ds and click OK.

5. In the Tree Outline under YZPlane, select Sketch4.

6. In the Details View under Dimensions, select the check box for H6 to promote it to an input
parameter.

A dialog box opens for specifying the parameter name.

7. Type horiz_ds and click OK.

8. In the same Details View under Dimensions, select the check box for V7 to promote it to an input
parameter.

A dialog box opens for specifying the parameter name.

9. Type vert_ds and click OK.

10. Exit DesignModeler.

The Project Schematic now includes the Parameter Set bar. An upward arrow leaves the Para-
meter Set bar and connects to the Parameters cell in the Static Structural system, indicating
that input parameters have been defined.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 23
Performing a Parametric “What If” Study

Promoting a Property and Solution Results in Mechanical to Output Para-

meters
To promote a property and two solution results in Mechanical to output parameters, do the following:

1. On the Project Schematic, double-click the Setup cell.

A dialog box opens, indicating that upstream data must be re-read.

2. Click Yes. Mechanical starts.

3. In the Outline view, select Geometry.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
24 of ANSYS, Inc. and its subsidiaries and affiliates.
 Promoting Dimensions and Solution Results to Input and Output Parameters

4. In the Details view under Properties, select the check box for Mass to promote the mass to an
output parameter.

5. In the Outline view under Solution, select Equivalent Stress.

6. In the Details view under Results, select the check box for Maximum to promote the maximum
equivalent stress to an input parameter.

7. In the Outline view under Solution, select Total Deformation.

8. In the Details view underResults, select the check box for Maximum to promote the maximum
total deformation to an output parameter.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 25
Performing a Parametric “What If” Study

9. Exit Mechanical.

On the Project Schematic, a downward arrow now leaves the Parameters cell in the Static
Structural system and connects to the Parameter Set bar, indicating that output parameters
have been defined.

10. Update the project.

Working with Design Points in Workbench

After the project update completes, the Parameter Set bar contains a design point. You can look at
the results for this design point and manually insert additional design points.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
26 of ANSYS, Inc. and its subsidiaries and affiliates.
 Adding Parameter Charts

1. On the Project Schematic, double-click the Parameter Set bar to open it.

The Table pane displays the design point that was generated.

2. Insert three additional design points as shown:

3. In the toolbar, click Update All Design Points.

When the design point update is complete, the Table pane displays results for the added design
points.

Adding Parameter Charts

In the Parameters Set bar, you can add charts. The follow topics explain how to add a Parameter chart
and a Parameters Parallel chart:
Adding a Parameter Chart
Adding a Parameters Parallel Chart

Adding a Parameter Chart

First, you’ll add a Parameter chart, which illustrates the relationship between a selected parameter
and each of the design points.

Note:

If you do not see the Chart pane in the bottom right of the workspace, select View →
Chart.

1. In the Outline pane under Output Parameters, select P6 — Total Deformation Maximum.
Currently, lower in the Outline pane under Charts, no chart instances exist.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 27
Performing a Parametric “What If” Study

2. If the Toolbox is not open, select View → Toolbox to open it. The Toolbox displays the Parameter
charts that can be inserted.

3. In the Toolbox, double-click Design Points vs P6 to add this chart.

4. In the Outline pane under Charts, select Parameter Chart 0.

In the Properties pane, X-Axis (Bottom) is set to Design Points and Y-Axis (Left) is set to P6
— Total Deformation Maximum:

The Graph pane displays this chart. The design points are arranged on the X axis on the bottom
of the chart. The P6 values are arranged on the Y axis on the left side of the chart.

Adding a Parameters Parallel Chart

Next, you’ll add a Parameters Parallel chart, which illustrates the relationship between all parameters
and each of the design points.

1. In the Outline pane, select any input or output parameter.

Because the Parameters Parallel chart plots parallel Y axes for all input and out parameters,
any parameter can be selected.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
28 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Solution

2. In the Toolbox, double-click Parameters Parallel Chart (all) to add this chart.

3. In the Outline pane under Charts, select Parameter Parallel Chart 0.

In the Graph pane, the Parameter Parallel chart displays all input and output parameters
along the X axis on the bottom of the chart. The top and bottom values indicate the range
relative to each parameter. Each color-coded line on the plot represents a design point.

Viewing the Solution

A solution is generated for the design point designated as (Current) in the Table pane. If, based on
your design exploration, you decide that another design point is preferable, you can define it as the
current design point. In this topic, you'll copy the inputs for DP 1 to the current design point.

1. In the Table pane, right-click DP 1 and select Copy inputs to Current.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 29
Performing a Parametric “What If” Study

The values for DP 1 are copied to DP 0, and DP 0 becomes outdated.

2. Right-click DP 0 and select Update Selected Design Points.

When the update is complete, in the Table pane, the results for DP 0 now match those for DP 1.

3. Close the Parameter Set bar.

4. On the Project Schematic, double-click the Model cell. Mechanical starts.

5. In the Outline view under Solution, select Equivalent Stress to view the solution for the current
design point.

In the following figure, the hitch receiver is rotated so that you can clearly see the locations for the
maximum and minimum equivalent stress.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
30 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Solution

6. Exit Mechanical.

7. Exit Workbench, saving project changes.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 31
 Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
32 of ANSYS, Inc. and its subsidiaries and affiliates.
Performing a Parameters Correlation
This tutorial gives step-by-step instructions for performing a parameters correlation to calculate the
deformation of a simple cantilever beam under applied force. It uses a DesignXplorer system that is
linked to a Mechanical APDL file to demonstrate parameter parsing and selection and how to show
results.

This tutorial covers the following topics:

What is a Parameters Correlation?
Getting Started
Defining Parameters
Generating Design Points
Performing the Parameters Correlation
Viewing the Correlation Matrix and Charts

What is a Parameters Correlation?

As you add more input parameters to a DOE, the number of design points increases dramatically, de-
creasing the efficiency of the analysis process. Because DesignXplorer works best with fewer than 20
parameters, you might want to exclude inputs that are not actively contributing toward your intended
design. Removing less important parameters from a DOE reduces the generation of unnecessary sampling
points.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 33
Performing a Parameters Correlation

Benefits
A parameters correlation allows you to:

• Determine which input parameters have the most (and the least) impact on your design.

• Identify the degree to which the relationship is linear or quadratic.

Visual Assessment
For a parameters correlation, DesignXplorer provides the following charts so that you can visually assess
parametric impacts:

• Correlation matrix and chart

• Determination matrix and chart

• Correlation scatter chart

• Sensitivity chart

Sampling
A parameters correlation performs simulations on a random sampling of the design space to identify
correlations between all parameters.

DesignXplorer uses the Latin Hypercube Sampling (LHS) method to generate samples. This means that
the points are randomly generated but no two points share input parameters of the same value. The
following figure shows how samples generated by LHS vary in placement from those generated by
Monte Carlo sampling.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
34 of ANSYS, Inc. and its subsidiaries and affiliates.
 What is a Parameters Correlation?

Correlation Methods
For a parameters correlation, DesignXplorer can use either the Pearson’s linear correlation method or
the Spearman’s rank correlation method.

Pearson’s Linear Correlation Method

• Uses actual data for correlation evaluation.

• Correlation coefficients are based on the sample values.

• Used to correlate linear relationships.

Spearman’s Rank Correlation Method

• Uses ranks of data.

• Correlation coefficients are based on the rank of samples.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 35
Performing a Parameters Correlation

• Recognizes non-linear monotonic relationships, which are less restrictive than linear ones.
In a monotonic relationship, one of the following occurs:

– As the value of one variable increases, the value of the other variable also increases.

– As the value of one variable increases, the value of the other variable decreases.

• Deemed the more accurate method.

Getting Started
This tutorial uses a textbook example of a simple model of a cantilever beam to calculate deformation
under an applied force. You will add this model to the Project Schematic by importing a Mechanical
APDL input file, which the following figure explains:

1. Start Workbench.

2. In the Toolbox under Component Systems, double-click Mechanical APDL to add this system to
the Project Schematic.

3. On the Project Schematic, right-click the Analysis cell and select Add Input File → Browse.

4. In the file browser, locate and open the file BeamEquations.inp, which is in the directory to
which you extracted the input files (p. 18).

5. Update the project.

6. Save the project as PCBeam.wbpj in either this same directory or another directory.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
36 of ANSYS, Inc. and its subsidiaries and affiliates.
 Performing the Parameters Correlation

Defining Parameters
1. On the Project Schematic, double-click the Analysis cell to open it.

2. In the Outline pane, select the Process “BeamEquations.inp” cell.

3. In the Properties pane, identify input parameters and then output parameters as shown:

4. Close the Analysis cell.

Generating Design Points

The Project Schematic now includes the Parameter Set bar. Next, you'll generate a design point.

1. Double-click the Parameter Set bar to open it.

The Table pane displays one design point for which results have not yet been calculated.

2. In the toolbar, click Update All Design Points.

When the update is complete, the Table pane displays results for DP 0, which is the current
design point.

3. Close the Parameter Set bar.

Performing the Parameters Correlation

You'll now add a Parameters Correlation system and then set properties and the lower and upper
bounds for the input parameters.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 37
Performing a Parameters Correlation

1. In the Toolbox under Design Exploration, double-click Parameters Correlation to add this
system to the Project Schematic.

2. In the new Parameters Correlation system, double-click the Parameters Correlation cell to
open it.

3. In the Outline pane, select Parameters Correlation.

4. In the Properties pane, set the properties as shown:

5. In the Outline pane, select the input parameter P1–WIDTH.

6. In the Properties pane under Values, set the lower and upper bounds as shown:

7. Following the process outlined in steps 5 through 6, set the lower and upper bounds for the
other input parameters:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
38 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Correlation Matrix and Charts

• For P2—HEIGHT, set Lower Bound to 4.5 and Upper Bound to 5.5. (The bounds for this
input parameter are already set.)

• For P3—LENGTH, set Lower Bound to 80 and Upper Bound to 120.

• For P4—FORCE, set Lower Bound to 800 and Upper Bound to 1200.

• For P5—YOUNG, set Lower Bound to 1.8E+05 and Upper Bound to 2E+05. (The bounds
for this input parameter are already set.)

8. Update the Parameters Correlation cell. The update may take a few minutes.

Viewing the Correlation Matrix and Charts

Once the update is complete, in the Outline pane under Charts, you can see that default charts have
been generated:

When you select a chart in the Outline pane, the Table pane displays the chart data. The Chart pane
displays the chart itself.

Tip:

To add another instance of a chart, select View → Toolbox to open the Toolbox and then
double-click the type of chart to add. You will then see this other instance in the Outline
pane under Charts.

Descriptions follow for these Parameters Correlation charts: Correlation Matrix, Sensitivities, Correlation
Scatter, and Determination Histogram. For comprehensive information on all Parameters Correlation
charts, see Working with Parameters Correlation Charts in the DesignXplorer User's Guide

Correlation Matrix
When Correlation Matrix is the chart selected in the Outline pane, the Table pane displays correlation
coefficients. A correlation coefficient indicates if there is a relationship between two variables and indic-
ates whether the relationship is positive or negative. The Graph pane displays the Correlation Matrix.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 39
Performing a Parameters Correlation

In this Correlation Matrix, you can see that input parameter P3–LENGTH is a major input because it
drives all the outputs, particularly P8–DISPLACEMENT and P6–VOLUME.

On the other hand, input parameter P5–YOUNG is not important because it has little impact on the
outputs. In this case, you might choose to disable P5–YOUNG by deselecting its check box in the
Properties pane. If you do this, a dialog box will open, indicating that this parameter change will clear
all generated data in this system. If you click Yes and update the Parameters Correlation cell, all charts
are updated accordingly. However, the results shown for this parameters correlation assumes that you
click No so that results are based on the originally generated data.

Sensitivities Chart
When Sensitivities is the chart selected in the Outline pane, the Graph pane shows global sensitivities
of the output parameters with respect to the input parameters. Positive sensitivity occurs when increasing
the input increases the output. Negative sensitivity occurs when increasing the input decreases the
output.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
40 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Correlation Matrix and Charts

In the Properties pane for a Sensitivities chart, you can set Mode to either Bar (default) or Pie. The
following figure shows the same Sensitivities chart in Pie mode.

Generally, the impact of an input parameter on an output parameter is driven by two things:

• Amount by which the output parameter varies across the variation range of an input parameter.

• Variation range of an input parameter. Typically, the wider the variation range is, the larger the
impact of the input parameter.

Statistical sensitivities are based on the Spearman-Rank Order correlation coefficients that simultaneously
take both aspects into account.

Correlation Scatter Chart

When Correlation Scatter is the chart selected in the Outline pane, the Table pane displays the linear
and quadratic coefficient of determination (R2) values. The Graph pane displays either or both linear
and quadratic trend lines.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 41
Performing a Parameters Correlation

For this particular Correlation Scatter chart, in the Properties pane under Axes, X Axis is set to
P1–WIDTH and Y Axis is set to P6–VOLUME.

Under Trend Lines, you can see the Linear and Quadratic values for R2. Because both the linear and
quadratic tread lines are enabled, the chart displays both lines. The closer the samples lie to the curve,
the closer the R2 value is to the optimum value of 1.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
42 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Correlation Matrix and Charts

Determination Histogram
When Determination Histogram is the chart selected in the Outline pane, the Graph pane shows you
what input parameters drive a selected output parameter.

For this particular Determination Histogram chart, in the Properties pane under Axes, Y Axis is set to
P8–DISPLACMENT.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 43
Performing a Parameters Correlation

In the Determination Histogram chart, you can see that input parameters P3–LENGTH, P2–HEIGHT,
and P4 – FORCE all affect output parameter P8–DISPLACEMENT. You can also see that of these three
input parameters, P3–LENGTH has the greatest impact by far.

In either the Chart pane or Properties pane, you can check the Full Model R2 (%) value to see how
well output variations are explained by input variations. The closer this value is to 100, the more certain
it is that output variations result from the input parameters. The lower the value, the more likely that
other factors such as noise, mesh error, or an insufficient number of points might be causing the output
variations.

• In the Chart pane, the value for linear determination is 95%.

• In the Properties pane, the value for linear determination is more precise at 95.31%.

To view the Determination Histogram chart for the quadratic determination, in the Properties pane,
set Determination Type to Quadratic. In the following figure, you can see that with a quadratic de-
termination type, input parameter P3– LENGTH still has the highest impact on P8–DISPLACEMENT.
The Full Model R2 (%) value for the quadratic determination is 96%.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
44 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Correlation Matrix and Charts

In the Properties pane, the more precise Full Model R2 (%) value for the quadratic determination is
96.68%.

Note:

In some cases, the relationship between parameters can be more complex and so cannot
be explained completely with a linear or quadratic correlation. If you pursue the study with
a response surface or goal-driven optimization, it will be difficult to build a Full 2nd Order
Polynomials response surface. In this case, try using another response surface type, such as
Kriging or Non-Parametric Regression.

You can filter your input parameters to keep only the most important by enabling or disabling their
check boxes in the Outline pane. As indicated earlier, when you do this, a dialog box will open, indic-
ating that parameter changes will clear all generated data in the system. If you click Yes and update
the Parameters Correlation cell, all charts are updated accordingly.

When you are finished looking at charts, close the Parameters Correlation cell. Then, exit Workbench,
saving project changes.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 45
 Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
46 of ANSYS, Inc. and its subsidiaries and affiliates.
Using a Design of Experiments
This tutorial introduces a Design of Experiments (DOE) and shows how to use the Design of Experiments
cell in a DesignXplorer system to investigate the impact of input parameters on the mass, equivalent
stress, and total deformation of a geometry model:
What is a Design of Experiments?
DOE Types
Getting Started
Working with the DOE
Working with the Response Surface
Solving for the Desired Design Point

What is a Design of Experiments?

Overview
A Design of Experiments (DOE) is a technique used to determine the location of sampling points in
such a way that the space of random input parameters is most efficiently explored and required inform-
ation is obtained with a minimum number of sampling points. A Design of Experiments cell is available
in these DesignXplorer systems: Response Surface, Response Surface Optimization, and 3D ROM.

Note:

A 3D ROM system is used to produce a ROM. For more information, see "Using ROMs"
in the DesignXplorer User's Guide.

Benefits
A DOE allows you to:

• Define input parameter limits.

• Select the DOE type best suited to your project and purposes.

• Specify DOE properties.

• Preview the DOE.

• Generate the DOE and view associated charts.

• Generate the response surface and view associated charts.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 47
Using a Design of Experiments

• Solve for the desired design point and view results.

Design Points and the DOE Matrix

Once you’ve defined input parameters, you can either preview or update the DOE to generate the DOE
Matrix. The DOE Matrix is constructed of design points that can be submitted to the parent analysis
system for solution.

• When you preview the DOE, the DOE Matrix is generated for viewing, but the design points are
not solved. No output parameter values are available.

• When you update the DOE, the DOE Matrix is generated, and the design points are solved.
Output parameter values are available.

Once the design points in the DOE Matrix are solved, you can update the Response Surface cell of the
DesignXplorer system. Based on the design point information contained in the DOE Matrix, response
surfaces are generated for each of the output parameters.

DOE Types
When setting up your DOE, you can select any of the DOE types available in DesignXplorer:

• Central Composite Design

• Optimal Space-Filling Design

• Box-Behnken Design

• Custom

• Custom + Sampling

• Sparse Grid Initialization

• Latin Hypercube Sampling Design

DOE Type Terminology

To understand the differences between DOE types, you need to know the following terms:

Rotatable

The degree to which the experimental design matrix is biased in any direction. A rotatable design
can be rotated around its center point without changing the prediction variance of a given point.
The predicted response is dependent only on its distance from the center point of the design.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
48 of ANSYS, Inc. and its subsidiaries and affiliates.
 DOE Types

Orthogonality

The degree to which the main effect and interaction estimates of interest are dependent on each
other. For example, two vectors are orthogonal if the sum of the products of their corresponding
elements is zero. A design is orthogonal if the effects of any variable have a sum of zero across the
effects of the other variables. With an non-orthogonal design, some variables are interdependent.
Because you cannot estimate the interactions between those variables, you cannot remove the
unwanted effects that may result from them. With an orthogonal design, however, you can estimate
the impact of each variable independently of the impacts of the other variables, which means that
you can isolate any unwanted effects and remove them from your design.

Leverage

The opportunity of sample points to have an abnormal influence on the outcome.

Central Composite Design (CCD)

A Central Composite Design (CCD) is a five-level fractional factorial design that is suitable for calibrating
the quadratic response model. A CCD consists of:

• 1 center point

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 49
Using a Design of Experiments

• 2*N axis points located at the –a and +a positions on each axis of the selected input parameter

• 2^(N-f ) factorial points located at the -1 and +1 positions along the diagonals of the input
parameter space

There are five design types for CCD that you can use to improve the response surface fit for the DOE:

Face-Centered

Includes three levels and is not rotatable.

Rotatable

Includes five levels and is rotatable. The drawback to this CCD type is that it does not get
sampling points at all extremes. Rotatable is preferred, however, because prediction variance
is the same for any two locations that are the same distance from the design center.

Variance Inflation Factor (VIF) Optimality

Maximizes orthogonalilty.

G-Optimality

Minimizes leverage.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
50 of ANSYS, Inc. and its subsidiaries and affiliates.
 DOE Types

Auto-Defined

This option is the default. It automatically switches between G-Optimality for five input
parameters or VIF Optimality for any other number of input parameters.

Optimal Space-Filling Design (OSF)

To offset the noise associated with physical experimentation, classical DOE types such as CCD focus on
parameter settings near the perimeter of the design region. Because computer simulation is not quite
as subject to noise, though, an Optimal Space-Filling (OSF) design is able to distribute the design
parameters equally throughout the design space with the objective of gaining the maximum insight
into the design with the fewest number of points.

One advantage of an OSF design is that you can specify the number of points to be generated. This is
especially useful when limited computation time is available. Another advantage is its ability to fill the
design space, which makes it appropriate when a more complex meta-modeling technique such as
Kriging, Non-Parametric Regression, or Neural Network is used.

Possible disadvantages of an OSF design are that extremes (such as the corners of the design space)
are not necessarily covered, that the selection of too few design points can result in a low quality of
response prediction, and that some randomness is included in the choice of a starting point.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 51
Using a Design of Experiments

Box-Behnken Design
A Box-Behnken design is a three-level quadratic design that does not contain fractional factorial design.
The sample combinations are treated in such a way that they are located at midpoints of edges formed
by any two factors. The design is rotatable (or in cases, nearly rotatable).

One advantage of a Box-Behnken design is that it requires fewer design points than a full factorial CCD
and generally requires fewer design points than a fractional factorial CCD. Additionally, a Box-Behnken
design avoids extremes (such as the corners of the design space), allowing you to work around extreme
factor combinations.

Possible disadvantages of a Box-Behnken design are that prediction at the corners of the design space
is poor and that there are only three levels per parameter.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
52 of ANSYS, Inc. and its subsidiaries and affiliates.
 DOE Types

Custom and Custom + Sampling

A Custom design and a Custom + Sampling design allow you to create your own DOE types, offering
a variety of ways to customize your DOE.

• A Custom design enables you to create a table of input parameters instead of using the default
DOE, import design point files from an external CSV (Comma-Separated Values) file, and define
design points to enrich an existing DOE.

• A Custom + Sampling design offers the same options as a Custom design. In addition, it allows
you to add sampling to automatically fill up the design space more efficiently.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 53
Using a Design of Experiments

Sparse Grid Initialization

A Sparse Grid Initialization must be used when creating a Sparse Grid response surface. A Sparse Grid
Initialization is an adaptive meta-model driven by the accuracy that you request. It increases the accuracy
of the response surface by automatically refining the matrix of design points in locations where the
relative error of the output parameter is higher. A Sparse Grid Initialization consists of one center point
and 2*n axis points. A Sparse Grid Initialization allows as many linear basis functions as there are points
of discretization.

One advantage of a Sparse Grid Initialization is that it refines only in the directions necessary, which
means fewer design points are needed for the same quality response surface. Another is that Sparse
Grid Initialization is effective at handling discontinuities.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
54 of ANSYS, Inc. and its subsidiaries and affiliates.
 DOE Types

Latin Hypercube Sampling Design

A Latin Hypercube Sampling Design uses an advanced form of the Monte Carlo sampling method to
avoid clustering samples. In Latin Hypercube Sampling (LHS), points are randomly generated in a square
grid across the design space, but no two points share the same value. This means that no point shares
a row or a column of the grid with any other point.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 55
Using a Design of Experiments

One disadvantage of LHS is that extremes, such as the corners of the design space, are not necessarily
covered. Additionally, the selection of too few design points can result in a lower quality of response
prediction. When Sample Types is set to CCD Samples (default), a maximum of 20 input parameters
is supported. This selection generates the same number of samples that a CCD DOE would generate
for the same number of inputs, resulting in an LHS design that has the same cost as the corresponding
CCD design.

Getting Started
For this tutorial, a model for a link with three cutouts is created in Ansys DesignModeler. Boundary
conditions are applied.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
56 of ANSYS, Inc. and its subsidiaries and affiliates.
 Getting Started

Input Parameters

• ds_cutout

• Bearing Load X Component

Output Parameters

• Solid Mass

• Equivalent Stress Maximum

• Total Deformation Maximum

• Equivalent Stress 2 Maximum

Opening the Archived Workbench Project

1. Start Workbench.

2. Select File → Open.

3. In the file browser, locate and open the archived Workbench project file link1.wbpz, which is in
the directory to which you extracted the input files (p. 18).

A dialog box appears for saving this archived file to a standard Workbench project file.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 57
Using a Design of Experiments

4. Save the project as link1.wbpj in either this same directory or another directory.

Verifying Input and Output Parameters

1. On the Project Schematic, double-click the Parameter Set bar to open it.

2. In the Outline pane, confirm that your input and output parameters are defined as shown:

3. Close the Parameter Set bar.

Subsequent topics provide step-by-step instructions for setting up and generating the DOE, generating
a response surface and its associated charts, and viewing results.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
58 of ANSYS, Inc. and its subsidiaries and affiliates.
 Working with the DOE

Working with the DOE

You'll now add a Response Surface system, configure and update the Design of Experiments cell,
and view associated charts.

Add the Response Surface System

In the Toolbox under Design Exploration, double-click Response Surface to add this system to the
Project Schematic.

Configure and Update the Design of Experiments Cell

1. In the Response Surface system, double-click the Design of Experiments cell to open it.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 59
Using a Design of Experiments

You can see your input and output parameters in the Outline pane:

2. In the Outline pane, select the input parameter P1 - ds_cutout.

In the Properties pane, confirm that the upper and lower bounds are set as shown:

3. In the Outline pane, select the input parameter P3 - Bearing Load X Component.

In the Properties pane, confirm that the upper and lower bounds are set as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
60 of ANSYS, Inc. and its subsidiaries and affiliates.
 Working with the DOE

4. In the Outline pane, select Design of Experiments.

5. In the Properties pane, confirm that Design of Experiments Type is set to Central Composite
Design.

6. In the toolbar, click Preview to see what design points are generated as part of the DOE. The Table
pane displays nine design points, none of which are solved.

7. Update the Design of Experiments cell.

When the update is complete, the Table pane displays results for these nine design points.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 61
Using a Design of Experiments

View DOE Charts

The update also generates default charts, which you can see in the Outline pane under Charts:

When you select a chart in the Outline pane, the Table pane displays the chart data. The Chart pane
displays the chart itself.

This topic describes only the Design Points vs Parameter chart.

1. In the Outline pane under Charts, select Design Points vs Parameter.

2. In the Properties pane under Axes, set properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
62 of ANSYS, Inc. and its subsidiaries and affiliates.
 Working with the Response Surface

The Chart pane displays the Design Points vs Parameters chart. The nine solved design points are
shown across the bottom of the chart. The output parameter values for Total Deformation Maximum
are shown on the left side of the chart.

3. If you want, in the Properties pane under Axes, change the selections to p additional charts.

4. When finished, close the Design of Experiments cell.

Working with the Response Surface

You'll now configure and update the Response Surface cell and view associated charts.

Configure and Update the Response Surface Cell

1. In the Response Surface system, double-click the Response Surface cell to open it.

2. Update the Response Surface cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 63
Using a Design of Experiments

3. In the Outline pane under Response Points, you can see that a single response point is generated,
along with charts for this response point.

When you select a chart for the response point, the Table pane displays the chart data. The Chart pane
displays the chart itself.

View Response Charts

When Response is the chart selected, the Graph pane displays the Response chart. A Response chart
allows you to see how changes to an input parameter affect a selected output parameter.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
64 of ANSYS, Inc. and its subsidiaries and affiliates.
 Working with the Response Surface

In the Properties pane for a Response chart, Mode provides three viewing options: 2D, 3D, and 2D
Slices. For this particular Response chart, the properties are set as shown:

The following images show a 3D Response chart and a 2D Slices Response chart.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 65
Using a Design of Experiments

To generate the 3D Response chart, the properties are set as shown:

To generate the 2D Slice Response chart, Mode is changed to 2D Slices. All other properties remain
unchanged.

View Additional Response Point Charts

To further explore your design space, you can view these additional charts for a response point: Local
Sensitivity, Local Sensitivity Curves, and Spider.

Local Sensitivity Chart

The Local Sensitivity chart allows you to see the impact of continuous input parameters (both
with and without manufacturable values) on output parameters. This chart calculates the change
of the output based on the change of each input independently, allowing you to see the weight
of each input.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
66 of ANSYS, Inc. and its subsidiaries and affiliates.
 Working with the Response Surface

Local Sensitivity Curves Chart

The Local Sensitivity Curves chart helps you to focus your analysis by allowing you to view in-
dependent parameter variations within the standard Local Sensitivity chart. This multi-curve
chart provides a means of viewing the impact of each input parameter on a specific output
parameter.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 67
Using a Design of Experiments

Spider Chart

The Spider chart allows you to visualize the impact that changing the input parameter or
parameters has on all of the output parameters simultaneously.

Solving for the Desired Design Point

In the Response Surface cell, you can now specify a design point, solve for it, and view the solution.

1. In the Outline pane under the single response point, select Response to view the Response chart.
This topic assumes that you are viewing the 3D response chart shown in the previous topic. As you
move the mouse cursor over the chart, the tooltip displays data for the current point.

2. Right-click a desired point on the Response chart and select Explore Response Surface at Point.

This new response point is added to the Table pane as Response Point 1.

3. In the Table pane, right-click this new response point and select Insert as Design Point.

4. Close the Response Surface cell.

5. Double-click the Parameters Set bar to open it.

The Table pane displays the newly added design point, DP 1. The input parameter values that you
see will differ from those shown as they depend on the response point that you inserted as a design
point.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
68 of ANSYS, Inc. and its subsidiaries and affiliates.
 Solving for the Desired Design Point

6. Right-click DP 1 and select Copy inputs to Current.

The values for the new design point are copied to the current design point, which is designated by
(Current). The current design point is the one that will be solved and for which the solution will be
displayed.

7. Close the Parameter Set bar.

8. Update the project.

9. When the update is complete, in the Static Structural system, double-click the Solution cell to
open it. Mechanical starts.

Again, the results that you see will differ slightly from those shown.

10. In the Outline view under Solution, select Equivalent Stress to view this result.

11. In the Outline view under Solution, select Total Deformation to view this result.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 69
Using a Design of Experiments

12. Exit Mechanical.

13. Exit Workbench, saving project changes.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
70 of ANSYS, Inc. and its subsidiaries and affiliates.
Performing a Goal-Driven Optimization
This tutorial provides an introduction to goal-driven optimizations in DesignXplorer. The end goal is to
minimize the mass of a support while exceeding a safety factor, with safety as the most important
consideration. This tutorial gives step-by-step instructions for importing an Ansys Mechanical input file,
generating design points and a response surface, defining optimization goals and objectives, and
viewing results for selected points.

This tutorial covers the following goal-driven optimization topics:

What is Goal-Driven Optimization?
Getting Started
Promoting Properties and Results in Mechanical to Parameters
Performing the Optimization
Viewing Goal-Driven Optimization Charts
Viewing Candidate Points
Viewing the Solution
Using ACT to Expose MATLAB Optimizers in DesignXplorer

What is Goal-Driven Optimization?

Overpane
Goal-driven optimization is a set of constrained, multi-objective optimization techniques in which the
"best" possible designs are obtained from a sample set given goals that are set for parameters.

Benefits
A goal-driven optimization allows you to determine the effect on input parameters with certain objectives
applied on output parameters. To do this, you specify a series of design goals or objectives to use to
generate an optimized design. You can define the optimization domain, specify values for input para-
meters, and weight goals in terms of their importance. Based on your specifications, DesignXplorer then
generates a set of sample designs from which you can select the most promising candidate designs.

Response Surface Optimization

This tutorial uses a Response Surface Optimization system that obtains its information from its own
Response Surface cell. Consequently, it is dependent on the quality of the response surface. The
available optimization methods (Screening, MOGA, NLPQL, and MISQP) use response surface evaluations
rather than real solves. Response surfaces are built from the DOE and quickly provide approximated
values of output parameters throughout the design space.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 71
Performing a Goal-Driven Optimization

Getting Started
The model for a support is created in DesignModeler. Constraints and loads are shown.

In this tutorial, you’ll explore the between mass, which is the size of the web on the support, and the
defined safety factor.

Open the Archived Workbench Project

1. Start Workbench.

2. Select File → Open.

3. In the file browser, locate and open the archived Workbench project file support.wbpz, which is
in the directory to which you extracted the input files (p. 18).

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
72 of ANSYS, Inc. and its subsidiaries and affiliates.
 Promoting Properties and Results in Mechanical to Parameters

A dialog box appears for saving this archived file to a standard Workbench project file.

4. Save the project as support.wbpj in either this same directory or another directory.

Promoting Properties and Results in Mechanical to Parameters

1. On the Project Schematic, double-click the Model cell.

Mechanical starts. You'll now promote dimensions in Mechanical to input parameters.

2. In Outline view, select Geometry.

3. In the Details view under Properties, select the check box for Mass to promote it to an output
parameter.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 73
Performing a Goal-Driven Optimization

4. In Outline view under Environment, select Force.

5. In the Details view under Definition, select the check box for Y Component to promote it to an
input parameter.

6. In the Outline view, under Solution > Stress Tool, select Safety Factor.

7. In the Details view under Results, select the check box for Minimum to promote it to an output
parameter.

8. Exit Mechanical.

9. On the Project Schematic, double-click the Parameter Set bar to open it.

In the Outline pane, you can see that the Mechanical values that you promoted to parameters display
as input and output parameters.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
74 of ANSYS, Inc. and its subsidiaries and affiliates.
 Performing the Optimization

10. Close the Parameter Set bar.

Performing the Optimization

The following topics explain how to perform the optimization:
Adding the Goal-Driven Optimization System
Configuring and Updating the Design of Experiments Cell
Configuring and Updating the Response Surface Cell
Configuring and Updating the Optimization Cell

Adding the Goal-Driven Optimization System

Your first step is to add the goal-driven optimization system. In the Toolbox under Design Exploration,
double-click Response Surface Optimization.

This system is added below the Parameter Set bar. You now need to work through each cell.

Configuring and Updating the Design of Experiments Cell

Your first step is to configure and update the Design of Experiments cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 75
Performing a Goal-Driven Optimization

1. In the Response Surface Optimization system, double-click the Design of Experiments cell to
open it.

2. In the Outline pane, select input parameter P2 - ds_web.

3. In the Properties pane, set the lower and upper bounds as shown:

4. In the Outline pane, select input parameter P4 - Force Y Component.

5. In the Properties pane, confirm that the lower and upper bounds are set as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
76 of ANSYS, Inc. and its subsidiaries and affiliates.
 Performing the Optimization

6. In the Outline pane, select Design of Experiments.

7. In the Properties pane under Design of Experiments, set the DOE type and design type as shown:

8. Update the Design of Experiments cell. This will take a few minutes.

When the update is complete, the Table pane displays a total of 9 design points, each of which
is solved.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 77
Performing a Goal-Driven Optimization

9. Close the Design of Experiments cell.

Configuring and Updating the Response Surface Cell

Your next step is to configure and update the Response Surface cell.

1. In the Response Surface Optimization system, double-click the Response Surface cell to open
it.

2. In the Outline pane, select Response Surface.

3. In the Properties pane, ensure that Response Surface Type is set to the default value, Genetic
Aggregation.

4. In the Outline pane, ensure that the check box for Min-Max Search is selected.

5. Update the Response Surface cell.

The Min-Max Search performs a series of NLPQL optimizations to find the minimum and maximums.

6. In the Outline pane, select Min-Max Search.

The Table pane displays the Min-Max Search results.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
78 of ANSYS, Inc. and its subsidiaries and affiliates.
 Performing the Optimization

7. In the Outline pane under Quality, select Goodness Of Fit.

In the Table pane, you can see goodness-of-fit metrics for all output parameters. In the Graph
pane, you can see the Goodness of Fit chart. The response surface fits both output parameters
fairly well.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 79
Performing a Goal-Driven Optimization

8. Close the Response Surface cell.

Configuring and Updating the Optimization Cell

Your next step is to configure and update the Optimization cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
80 of ANSYS, Inc. and its subsidiaries and affiliates.
 Performing the Optimization

1. In the Response Surface Optimization system, double-click the Optimization cell to open it.

2. In the Outline pane, select Optimization.

3. In the Properties pane, set optimization properties as shown:

4. In the Outline pane, select Objectives and Constraints.

5. In the Table pane, for parameters P3 - Geometry Mass and P5 - Safety Factor Minimum, specify
objectives and constraints as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 81
Performing a Goal-Driven Optimization

6. Now, for the two parameters with objectives, ensure that the relative importance of the objectives
are set as indicated:

a. In the Outline pane under Objectives and Constraints, select Minimize P3. In the Properties
pane, ensure that Target is set to 2.6.

b. In the Properties pane under Decision Support Process, ensure that Objective Importance
is set to Default.

c. In the Outline pane under Objectives and Constraints, select Seek P5 = 1.1. In the Properties
pane, ensure that Target is set to 1.1 and Tolerance is set to 0.1.

d. In the Properties pane under Decision Support Process, set Objective Importance to
Higher.

7. Update the Optimization cell.

Viewing Goal-Driven Optimization Charts

Once the update is complete, the following optimization charts are available: Candidate Points, Tradeoff,
Samples, Sensitivities, and History.

• To view all but the History chart, in the Outline pane under Results, select the chart to display it in
the Graph pane.

• To view the History chart for a parameter, in the Outline pane under either Objectives and Con-
straints or Domain, select the parameter to display the chart in the Graph pane.

For comprehensive information about optimization charts, see Goal-Driven Optimization Charts and
Results in the ANSYS DesignXplorer User's Guide.

Viewing Candidate Points

To view the candidate points found by the optimization, in the Outline pane under Results, select
Candidate Points.

The Table pane displays candidate point data. In the following figure, Candidate Point 1 is selected.
The Graph pane plots the data for this candidate point. Placing the mouse cursor over the line for a
candidate point highlights this line.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
82 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing Candidate Points

In the Table pane, you can see the input and output parameter values for the candidate points. The
output parameter values calculated from simulations (design point updates) are displayed in black text.
The output parameter values calculated from a response surface are displayed in the user-specified
custom color, which is blue in the example. The number of gold stars or red crosses shown next to each
goal-driven parameter indicate how well the parameter meets the stated goal, with three red crosses
being the worst and three gold stars being the best.

For each parameter with an objective or constraint defined, the percentage of variation for all parameters
with regard to an initial reference point is calculated. You can set any candidate point as the initial
reference point by selecting the radio button in the Reference column.

The Parameter Value column displays the parameter value and stars indicating the quality of the
candidate. In the Variation from Reference column, green text indicates variation in the expected
direction. Red text indicates variation that is not in the expected direction. When there is no obvious
direction (as for a constraint), the percentage value displays in black text.

Verifying the Most Viable Candidate Point

To verify that Candidate Point 1 is the most viable update, right-click it and select Verify by Design
Point Update. When the update is completed, the Table pane displays Candidate Point 1 (verified),
with solved values, next to the approximated values for Candidate Point 1 so that you can easily
compare them:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 83
Performing a Goal-Driven Optimization

Viewing the Solution

A solution is generated for the design point designated as (Current) in the Table pane. If, based on
your design exploration, you decide that another design point is preferable, you can define it as the
current design point.

The following topics explain how to copy the inputs for DP 1 to the current design point and then view
results:
Viewing Results for the Most Viable Candidate Point
Viewing Results for the Response Point
Viewing Results in Mechanical

Viewing Results for the Most Viable Candidate Point

You'll now insert the most viable candidate point as a design point. Then, you'll copy its input values
to the current design point.

1. In the Outline pane under Results, select Candidate Points.

2. Right-click Candidate Point 1 and select Insert as Design Point.

In the Table pane, Candidate Point 1 (verified) now has the label DP 2 to its right.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
84 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Solution

3. Close the Optimization cell.

4. On the Project Schematic, double-click the Parameter Set bar to open it.

5. In the Table pane, right-click DP 1 and select Copy inputs to Current.

The current design point now has the same input parameter values as your best candidate, DP
1. The Note column indicates that this design point is created from Candidate Point 1 in the
Optimization cell.

6. Right-click the current design point and select Update Selected Design Points.

Once the design point is updated, input values for the current design point are now the same as
for DP 1. Updating only the current design point saves time and resources.

7. For this tutorial, also update DP 2 to check that the input values are the same:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 85
Performing a Goal-Driven Optimization

8. Close the Parameter Set bar.

9. In the Static Structural system, double-click the Solution cell to open it.

Mechanical starts.

10. In the Outline view, under Solution > Stress Tool, select Safety Factor to view the results.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
86 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Solution

11. Exit Mechanical.

Viewing Results for the Response Point

Your next step is to view results and copy values for a selected point in the response surface to the
response point in the Table pane. You'll then insert this response point as a design point, update it,
and copy its values to the current design point. Lastly, you'll update the project.

1. In the Response Surface Optimization system, double-click the Response Surface cell to open
it.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 87
Performing a Goal-Driven Optimization

2. In the Outline pane under Response Points > Response Point, select Response, which displays
the Response chart in the Graph pane.

3. In the Properties pane for this chart, set Mode to 3D.

4. In the Graph pane, move the mouse cursor over the chart to view data for various points.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
88 of ANSYS, Inc. and its subsidiaries and affiliates.
 Viewing the Solution

5. Right-click a point and select Copy Values to Response Point.

The values for this point are copied to the response point in the Table pane.

6. In the Table pane, right-click the response point and select Insert as Design Point.

7. Close the Response Surface cell.

8. On the Project Schematic, double-click the Parameter Set bar to open it.

The Table pane now includes a design point named DP 3. The Note column indicates that this
design point is created from the response surface.

9. In the Table pane:

a. For DP 3, select the Retain check box.

b. Right-click DP 3 and select Update Selected Design Points.

c. Right-click DP 3 and select Set as Current.

Note:

It is also possible to select Export Selected Design Points to view these

results in a separate Workbench project.

10. Close the Parameter Set bar.

11. On the Project Schematic, update the project.

Viewing Results in Mechanical

To view results in Mechanical:

1. In the Static Structural system, double-click the Model cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 89
Performing a Goal-Driven Optimization

Mechanical starts.

2. In the Outline view under Solution, select Stress Tool > Safety Factor to view the results.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
90 of ANSYS, Inc. and its subsidiaries and affiliates.
 Using ACT to Expose MATLAB Optimizers in DesignXplorer

3. Exit Mechanical.

4. Exit Workbench, saving changes to the project.

Using ACT to Expose MATLAB Optimizers in DesignXplorer

DesignXplorer can delegate the resolution of an optimization to third-party optimizers. The optimizers
are installed as ACT extensions and hosted in the DesignXplorer environment. When invoked, the installed
optimizer becomes responsible for the resolution of the optimization. The complete definition of the
optimization, along with utilities to trigger design point updates and provide progress information, is
passed to the installed optimizer. This architecture enables the optimizer to implement any optimization
strategy or workflow, from the simplest to the most complicated.

From the Ansys App Store, you can download the free app MATLab Optmizers for DX for your Ansys
version. In the downloaded folder, the child folder doc contains a PDF file that explains how to install
the extension and use the MATLAB optimizers that this ACT extension exposes in DesignXplorer.

You need no knowledge of ACT to use the MATLAB optimizers in this extension. You can use them exactly
as they are delivered, without having to change or recompile the ACT extension. However, if these op-

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 91
Performing a Goal-Driven Optimization

timizers do not meet your need, the supplied PDF file explains how you can either customize the gen-
eric optimizer or modify source code to create your own ACT extension with custom MATLAB optimizers.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
92 of ANSYS, Inc. and its subsidiaries and affiliates.
Using Adaptive Single-Objective Optimization
To choose the correct optimization method for a given problem, you must understand the problem.
To understand the problem, you must first explore it, which requires the selection of an optimization
method. DesignXplorer’s Adaptive Single-Objective (ASO) optimization method is a robust, adaptive al-
gorithm that simplifies this process, allowing you to explore your design space during an actual optim-
ization run.

In this advanced tutorial, you’ll use four different optimization scenarios, including one that uses ASO,
to explore the design space and find the global optimum for the same problem. You’ll examine the
results and benefits of each method for solving this particular problem, learning how the performance
of different algorithms in combination compare with the performance of ASO.

Note:

This advanced tutorial assumes that you are familiar with Ansys Workbench and with
DesignXplorer’s goal-driven optimization functionality. For an introduction to such optimiza-
tions, see the tutorial Performing a Goal-Driven Optimization (p. 71).

This tutorial is divided into the following sections:

What is Adaptive Single-Objective Optimization?
Problem Definition
Problem Setup
Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimization
Scenario 2: NLPQL Direct Optimization to NLPQL Direct Optimization
Scenario 3: Screening Direct Optimization to NLPQL Direct Optimization
Scenario 4: Adaptive Single-Objective Direct Optimization
Time to Spare?
What Have You Learned?

What is Adaptive Single-Objective Optimization?

Adaptive Single-Objective (ASO) is a gradient-based mathematical optimization method that is available
only for direct optimizations. It combines an Optimal Space-Filling (OSF) DOE, a Kriging response surface,
and the Mixed-Integer Sequential Quadratic Programming (MISQP) optimization algorithm with domain
reduction to locate the global optima.

When an optimization method is adaptive, it is internally powered by response surface technology.

When the level of accuracy is not acceptable, ASO performs design point updates and refines the surface.
When the level of accuracy is good enough, it uses approximation instead.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 93
Using Adaptive Single-Objective Optimization

Problem Definition
For this tutorial, the problem is a non-convex analytic function with two input parameters. The definition
of the problem is as follows:

Minimize

Where

And

This analytic function has three local maxima, one local minimum, and one global minimum point at
(0.2282;-1.6256), with a corresponding objective function value of -6.5511.

Problem Setup
To create this project, you’ll add an analysis system to the Project Schematic and then link the Analysis
cell to an input file.

1. Start Workbench.

2. From under Component Systems in the Toolbox, add the system to the Project Schematic:

• For Windows, add a Microsoft Office Excel system.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
94 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimiz-
ation

• For Linux, add a Mechanical APDL system.

3. In this system, double-click the Analysis cell to open it.

4. In the Outline pane, right-click in the Input column for Setup and select Add File → Browse.

5. In the file browser, navigate to the directory to which you extracted the input files (p. 18) and
open the appropriate file:

• For Windows, open AnalyticFunction2D.xlsx.

• For Linux, open AnalyticFunction2D.inp.

6. Define input and output parameters as shown.

Note:

The images in the remainder of this tutorial are for the XLSX file. The order in which
you define parameters determines the system-generated names for the parameters.
For input names to match those shown, you must select the Input check boxes for
wb_x1 first and wb_x2 second. You then select the Output check box for wb_y
last.

7. Close the Analysis cell. The Project Schematic now includes the Parameter Set bar.

8. Update the project.

9. Save the project as DX_ASO.wbpj in either this same directory or another directory.

In this project, you’ll run four different optimization scenarios, compare their results, and determine
which optimization method was best for this particular problem.

Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL

Direct Optimization
For the first scenario, you’ll run a response surface optimization. You’ll then use its results to run a direct
optimization.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 95
Using Adaptive Single-Objective Optimization

Run the Response Surface Optimization

You'll begin by adding a Response Surface Optimization system to the Project Schematic. You'll
then configure each of its cells, run the optimization, and view results.

Add the Goal-Driven Optimization System

On the Project Schematic, in the Toolbox under Design Exploration, double-click Response Surface
Optimization.

This system is added below the Parameter Set bar.

Configure and Update the Design of Experiments Cell

1. Double-click the Design of Experiments cell to open it.

2. In the Outline pane, select Design of Experiments.

3. In the Properties pane, set DOE properties as shown:

4. In the Properties pane for each input parameter, set lower and upper bounds as shown here for
P1 - wb_x1:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
96 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimiz-
ation

5. Update the Design of Experiments cell. Once the update completes, the Table pane displays
the solved design points to be used as inputs for the response surface.

6. Close the Design of Experiments cell.

Configure and Update the Response Surface Cell

Because the problem is a type of function that cannot be approximated with a quadratic response
surface, you must select an alternate type of response surface. Kriging is a good choice because it
can approximate the function by using automatic refinement to enrich the response surface and obtain
the required accuracy.

1. On the Project Schematic, double-click the Response Surface cell to open it.

2. In the Outline pane, select Response Surface.

3. In the Properties pane, set properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 97
Using Adaptive Single-Objective Optimization

4. In the Outline pane under Output Parameters, ensure that the check box for Min-Max
Search is selected.

5. Update the Response Surface cell. When the update is complete, the Table pane displays
a response point. The value for output P3 - wb_y is 3.0293.

View Results
1. In the Outline pane, select Response Surface.

In the Properties pane under Refinement, Number of Refinement Points displays 54. This indic-
ates that Kriging with automatic refinement converged after 54 additional refinement points were
created.

2. In the Outline pane under Output Parameters, select Min-Max Search.

The Table pane displays the approximate value of the objective function (-5.8017) and the para-
meter minimums (P1 = 0.35346 and P2 = –1.5925). You will use these minimum values to initialize
the Optimization cell of the Response Surface Optimization system.

3. In the Outline pane, select Refinement. The Graph pane displays the Convergence Curves chart.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
98 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimiz-
ation

4. To view the three-dimensional Kriging response surface:

a. In the Outline pane under Response Points → Response Point, select Response to display
this chart.

b. In the Properties pane:

• Set Mode to 3D.

• Set Chart Resolution Along X and Chart Resolution Along Y to 50.

The chart displays the three-dimensional response surface.

c. In the Chart pane, rotate the chart so that only the Y axis and Z axis are visible.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 99
Using Adaptive Single-Objective Optimization

5. To view the design point distribution of the Kriging response surface:

a. In the Properties pane, select the Show Design Points check box.

b. Right-click the response surface and select Edit Properties.

c. In the Properties pane for the chart, set Display As to None.

d. In the Chart pane, rotate the chart so that only the X axis and Y axis are visible.

The chart displays the distribution of design points in the response surface as shown:

6. Close the Response Surface cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
100 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimiz-
ation

Configure and Update the Optimization Cell

1. On the Project Schematic, double-click the Optimization cell to open it.

2. In the Outline pane, select Optimization.

3. In the Properties pane, set optimization properties as shown:

4. In the Outline pane, select Objectives and Constraints.

5. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Minimize
and Target to -6:

6. In the Outline pane, select Domain.

7. In the Table pane, set Starting Value for each input parameter to the minimum found earlier by
the Min-Max Search:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 101
Using Adaptive Single-Objective Optimization

8. Update the Optimization cell.

View Results
1. When the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, Number of Evaluations shows that the op-
timization has converged after 14 evaluations.

2. In the Outline pane under Results, select Candidate Points. The Table pane shows that the best
candidate is the original NLPQL Starting Point. This is expected because the Min-Max Search is
based on NLPQL.

3. Right-click Candidate Point 1 and select Verify by Design Point Update.

The objective function value for the verified point is -6.4009.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
102 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimiz-
ation

4. Close the Optimization cell.

Run the Direct Optimization

Next, you'll add, configure, and run a Direct Optimization system that uses NLPQL. Although NLPQL
is dependent on the Starting Point, you can get reasonable starting points for the inputs by using
the results of the response surface optimization. Also, you can use response surface exploration to
reduce the domain of the direct optimization +/-0.3 in each direction.

1. From under Design Exploration Systems in the Toolbox, double-click Direct Optimization to
add this system to the Project Schematic.

2. In the Direct Optimization system, double-click the Optimization cell to open it.

3. In the Outline pane, select Optimization.

4. In the Properties pane, set optimization properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 103
Using Adaptive Single-Objective Optimization

5. In the Outline pane, select Objectives and Constraints.

6. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Minimize
and Target to -6:

7. In the Outline pane, select Domain.

8. In the Table pane, assign values for the input parameters as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
104 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 1: Kriging-NLPQL Response Surface Optimization to NLPQL Direct Optimiz-
ation

Note:

The starting values are set to the minimums found earlier by the response surface Min-
Max Search. The lower and upper bounds are set to +0.3 and –0.3 of the starting values.

9. Update the Optimization cell.

View Results
1. When the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that the optimization has con-
verged. Four iterations and 15 design points were needed to find the minimum.

2. In the Outline pane under Results, select Candidate Points. The Table pane shows that Candidate
Point 1 now has an objective function value of -6.5511.

3. Close the Optimization cell.

How Effective was the Approach Used for Scenario 1?

A response surface optimization is a good way to explore the design space. However, for this problem,
it is expensive in terms of the number of design points required (80):

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 105
Using Adaptive Single-Objective Optimization

• 10 to build the DOE

• 54 to enrich the Kriging response surface

• 1 to verify the candidate

• 15 to run the NLPQL direct optimization

Once built, the Kriging response surface does allow you to find the area containing the global minimum,
but the response surface optimization alone cannot obtain an accurate candidate point unless more
design points are generated to further enrich the Kriging response surface.

Running an NLPQL direct optimization afterward, with the candidate from the response surface op-
timization as the starting point and with a reduced domain, is a good way to get more accuracy from
the response surface-based approach.

Scenario 2: NLPQL Direct Optimization to NLPQL Direct Optimization

In Scenario 2, you'll begin by adding and running a Direct Optimization system that uses NLPQL. You’ll
then run a second direct optimization that is the same as the first, except with a different starting point.

Run the First NLPQL Direct Optimization

You'll begin by adding, configuring, and running the first NLPQL direct optimization:

1. From under Design Exploration in the Toolbox, double-click Direct Optimization to add this
system to the Project Schematic.

2. Rename the system to Scenario 2: First NLPQL Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

5. In the Properties pane, specify optimization properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
106 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 2: NLPQL Direct Optimization to NLPQL Direct Optimization

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Minimize
and Target to -6:

8. In the Outline pane, select Domain.

9. In the Table pane, set input parameter properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 107
Using Adaptive Single-Objective Optimization

10. Update the Optimization cell.

When the update is complete, the Table pane displays 14 design points and their results.

View Results
1. In the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that 4 iterations and 14 design
points were needed to find the minimum.

In the Table pane, you can see that the objective function of Candidate Point 1 is 0.012741.

2. Close the Optimization cell.

Run the Second NLPQL Direct Optimization

Next, you’ll add, configure, and run the second NLPQL direct optimization. While you are going to
change the starting point, you will not be using the results of the last optimization because the ob-
jective function obtained was not close enough to the expected value to be usable.

1. From under Design Exploration in the Toolbox, double-click Direct Optimization to add this
system to the Project Schematic.

2. Rename the system to Scenario 2: Second NLPQL Direct Optimization.

3. In the system, double-click the Optimization cell to open it. The next four steps are the same as
for the first NLPQL direct optimization.

4. In the Outline pane, select Optimization.

5. In the Properties pane, specify optimization properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
108 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 2: NLPQL Direct Optimization to NLPQL Direct Optimization

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Minimize
and set Target to -6:

8. In the Outline pane, select Domain.

9. In the Table pane, set input parameter parameters as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 109
Using Adaptive Single-Objective Optimization

Note:

The starting values are randomly selected. They are not based on the results of the
previous direct optimization.

10. Update the Optimization cell.

When the update is complete, the Table pane displays 88 design points and their results.

View Results
1. In the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that 13 iterations and 88 design
points were needed to find the global minimum.

In the Table pane, you can see that the objective function of Candidate Point 1 is -6.5511, which
is the expected value.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
110 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 3: Screening Direct Optimization to NLPQL Direct Optimization

2. Close the Optimization cell.

How Effective was the Approach Used for Scenario 2?

In this scenario, the problem required 102 design points, 14 for the first NLPQL direct optimization
and 88 for the second NLPQL direct optimization. If you didn't know the global minimum ahead of
time, you might think that the results of the first NLPQL direct optimization are good—until you run
the second one and achieve a better result.

The two optimizations in this scenario illustrate the importance of the Starting Value in a gradient-
based method optimization such as NLPQL, especially when the objective function is not convex and
contains several local optima. Because this problem has one local minimum and one global minimum,
NLPQL alone cannot find the global optimum without a good starting point. This is also true for the
Mixed-Integer Sequential Quadratic Programming (MISQP) optimization method.

Scenario 3: Screening Direct Optimization to NLPQL Direct Optimization

In Scenario 3, you'll use what you learned from Scenario 2: NLPQL needs to have a good starting point.
You’ll begin by running a Screening direct optimization to explore the design space. You will then use
results from this optimization as the starting point for running an NLPQL direct optimization.

Run the Screening Direct Optimization

You'll begin by adding, configuring, and updating the Screening direct optimization:

1. From under Design Exploration in the Toolbox, double-click Direct Optimization to add
this system to the Project Schematic.

2. Rename the system to Scenario 3: Screening Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

5. In the Properties pane, specify optimization properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 111
Using Adaptive Single-Objective Optimization

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Min-
imize and Target to -6:

8. In the Outline pane, select Domain.

9. In the Properties pane, set upper and lower bounds as shown:

10. Update the Optimization cell.

When the update is complete, the Table pane displays 20 design points and their results.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
112 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 3: Screening Direct Optimization to NLPQL Direct Optimization

View Results
1. In the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that Screening generated a
sample set of 20 design points and identified 3 candidate points.

In the Table pane, you can see Candidate Point 1 is the best candidate. For this candidate point,
parameter P1 - wb_x1 has a value of 0.75, parameter P2 - wb_x2 has a value of –1.725, and the
objective value of the function (output P3 - wb_y) is -4.2983.

2. Close the Optimization cell.

Run the NLPQL Direct Optimization

Next, you'll add, configure, and update the NLPQL direct optimization:

1. From under Design Exploration in the Toolbox, double-click Direct Optimization to add
this system to the Project Schematic.

2. Rename the system to Scenario 3: NLPQL Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

5. In the Properties pane, set optimization properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 113
Using Adaptive Single-Objective Optimization

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Min-
imize and Target to -6:

8. In the Outline pane, select Domain.

9. In the Properties pane, set input parameter properties as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
114 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 3: Screening Direct Optimization to NLPQL Direct Optimization

10. Update the Optimization cell.

When the update is complete, the Table pane displays 38 design points and their results.

View Results
1. In the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that 5 iterations and 38 new
design points were needed to find the global minimum.

From the Convergence Criteria chart and the Raw Optimization Data table, you can see that the
candidate was already found at the third iteration with 13 design points.

In the Table pane, you can see Candidate Point 1 is the best candidate. For this candidate point,
the objective value of the function is -6.5511, which again matches the expected value.

2. Close the Optimization cell.

How Effective was the Approach Used for Scenario 3?

Scenario 3 is quite reliable because you began by running a Screening direct optimization to explore
the design space and find the best global candidate point, which you then used as the starting point
for the NLPQL direct optimization. The best candidate was actually found after 13 design points in
the NLPQL, so this approach is more effective than the approaches in the two previous scenarios. It
requires only 33 design points, 20 to run the Screening and 13 more for NLPQL.

However, remember that the candidate point found by Screening must be good enough to guarantee
convergence of NLPQL. Convergence depends on the space-filling ability of Screening to create
enough samples to adequately explore the parameter space. Also, a Screening direct optimization
can be expensive when you have a large number of input parameters.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 115
Using Adaptive Single-Objective Optimization

Scenario 4: Adaptive Single-Objective Direct Optimization

In Scenario 4, you will run only a single Adaptive Single-Objective (ASO) direct optimization and then
view results.

Run the ASO Direct Optimization System

Add, configure, and update the ASO direct optimization:

1. From under Design Exploration in the Toolbox, double Direct Optimization to add this
system to the Project Schematic.

2. Rename the system to Scenario 4: ASO Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

5. In the Properties pane, set optimization properties as shown:

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, for Parameter, select P3 - wb_y and then set the objective type to Min-
imize and Target to -6:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
116 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 4: Adaptive Single-Objective Direct Optimization

8. In the Outline pane, select Domain.

9. In the Properties pane, set the lower and upper bounds as shown:

10. Update the Optimization cell.

View Results
1. In the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that the optimization converged
after 56 evaluations and 3 candidates were identified.

In the Table pane, you can see Candidate Point 1 is the best candidate. For this candidate point,
parameter P1 has a value of 0.23301, parameter P2 has a value of -1.6264, and the corresponding
objective function value for P3 is –6.5509. These values show that the optimization has reached
the expected global minimum.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 117
Using Adaptive Single-Objective Optimization

2. In the Outline pane under Results, select Tradeoff. In the Chart pane, you can see that the re-
finement is targeted to a small area of the surface.

For the following Tradeoff chart, in the Properties pane under Chart, Mode is set to 2D. Under
Axes, X Axis is set to P1 - wb_x1 and Y Axis is set to P2 - wb_x2.

For the next Tradeoff chart, in the Properties pane under Chart, Mode is set to 3D. Under Axes,
the X Axis and Y Axissettings remain the same. The Z Axis is set to P3 - wb_y. This chart displays
the refinement in the minimum area.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
118 of ANSYS, Inc. and its subsidiaries and affiliates.
 Scenario 4: Adaptive Single-Objective Direct Optimization

The History charts for the input parameters show the successive steps of the domain reduction
performed by ASO.

3. In the Outline pane under Domain, select P1 - wb_x1 to view the History chart of this input
parameter in the Graph pane:

4. In the Outline pane under Domain, select P2 - wb_x2 to view the History chart of this input
parameter in the Graph pane:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 119
Using Adaptive Single-Objective Optimization

5. In the Outline pane under Objectives and Constraints, select Minimize P3 to view the History
chart of this output parameter in the Graph pane:

6. Close the Optimization cell.

7. Exit Workbench, saving project changes.

How Effective was the Approach Used for Scenario 4?

This scenario required 56 design points. Although ASO needed more points than the methods used
in Scenarios 2 and 3, it has the added benefit of ease of use. ASO automatically zooms in on a solution
by adaptive methods.

ASO has these advantages:

• Offers a fully automated method of finding the global optimum using sampling, a response
surface, and the NLPQL algorithm.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
120 of ANSYS, Inc. and its subsidiaries and affiliates.
 What Have You Learned?

• Employs targeted refinement (the refinement of the internal response surface is driven by the
optimization objective), which means that it does not expend time or resources on refining
the surface in areas not relevant to the optimization.

• Finds the optimal point without requiring results from a prior optimization.

• Reaches a high level of accuracy early in the optimization process, accelerating the optimization
process by enabling you to accept intermediate results.

Time to Spare?
Open the saved DX_ASO project and try to find the global maximum of the objective function by using
ASO.

What Have You Learned?

During this tutorial, you learned that DesignXplorer offers multiple ways to find the global optimum
for a given function.

Different Types of Optimizations

You looked at different types of optimizations: response surface optimizations, direct optimizations,
and an approach that combined both of these types.

Response Surface Optimization

• Allows you to select the type of DOE and response surface best suited to your problem.

• Quickly finds interesting areas of the design space.

• Offers candidate verification and is excellent for exploring sensitivities, determination, and responses.

• Excellent for design exploration and finding an approximated optimum quickly.

• Prevents targeted refinement.

The response surface is built before optimization objectives are defined. This means that time and
resources are spent on parts of the design space that are not relevant to the optimization, making
it a less effective approach for optimization. Because the entire design space is refined, the optim-
ization can be very expensive, requiring a large number of design point updates to obtain a response
surface that is accurate all over.

• Sacrifices some accuracy to achieve greater efficiency.

For very complex response surfaces, the number of trials to generate a high enough quality response
surface can exceed the number that would have been required for a direct solve.

Direct Optimization
• Gives you the ability to reuse data.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 121
Using Adaptive Single-Objective Optimization

Although it does not have a single associated response surface, data reuse is possible because it
can retrieve information via data links from other DesignXplorer system cells that contain design
point data.

• Good choice when a large number of parameters or problems with building a good response surface
make response surface optimization infeasible.

• Provides targeted refinement.

Refinement is driven by the objective, with the creation of a new response surface on a smaller
domain with each iteration. The smaller the domain, the easier the surface construction and the
more accurate the approximation.

• Although it uses real solves, each design point update is worth the expense.

Each update is targeted on the area most relevant to the optimization, allowing the refinement
process to progressively zoom in on the optimum.

Different Optimization Methods

Screening
• Good option for the initial exploration of a design space because its space-filling abilities allow it
to locate a viable candidate point, which can possibly be used as a starting point for an NLPQL
optimization.

• Can be expensive when there are many input parameters.

NLPQL
• Can add accuracy to the response surface-based approach.

• Highly dependent on the quality of the starting point.

ASO
• Adaptive method that uses actual solves to refine the response surface, but it uses approximation
when accuracy is good enough.

• Combines the best of DesignXplorer technologies: a DOE, an internal response surface, domain
reduction, and error prediction.

• Provides both accuracy and speed without needing prior results to initialize the optimization.

• Allows you to balance your available time and resources with your desired level of accuracy.

• While a response surface optimization or NLPQL might be sufficient for exploring problems that
are convex or smooth, the ASO is a better optimization choice when you are not already very famil-
iar with your problem.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
122 of ANSYS, Inc. and its subsidiaries and affiliates.
Using Adaptive Multiple-Objective Optimization
In this advanced tutorial, you’ll use DesignXplorer's Adaptive Multiple-Objective (AMO) optimization
method as part of a direct optimization to locate a Pareto front. First, you'll create two separate projects
to examine different functions and apply both the Multiple-Objective Genetic Algorithm (MOGA) and
AMO to the problem. Then, you'll review the results and examine why AMO is better suited to finding
the Pareto front for the given problem.

Note:

This advanced tutorial assumes that you are familiar with Ansys Workbench and with
DesignXplorer’s goal-driven optimization functionality. For an introduction to such op-
timizations, see the tutorial Performing a Goal-Driven Optimization (p. 71).

This tutorial is divided into the following sections:

What is Adaptive Multiple-Objective Optimization?
Problem 1
Problem 2
What Have You Learned?

What is Adaptive Multiple-Objective Optimization?

Adaptive Multiple-Objective (AMO) is a hybrid optimization method that combines a Kriging response
surface optimization and MOGA direct optimization. It uses the same general approach as MOGA but
extends it by using the Kriging error predictor to reduce the number of evaluations needed to locate
the global optimum.

Problem 1
This tutorial covers the following topics for Problem 1:
Problem 1 Definition
Problem 1 Project Setup
Run the MOGA Direct Optimization
Run the AMO Direct Optimization

Problem 1 Definition
The analytic function you’ll examine for Problem 1 has two input parameters and four output para-
meters (with two objectives and two constraints). The definition of the problem is as follows:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 123
Using Adaptive Multiple-Objective Optimization

Minimize

With

Where and .

Problem 1 Project Setup

1. Start Workbench.

2. From under Component Systems in the Toolbox, double-click Microsoft Office Excel to add
this system to the Project Schematic.

3. Right-click the Analysis cell and select Add File → Browse

4. In the file browser, locate and open the file DX_AMO_input1.xlsx, which is in the directory to
which you extracted the input files (p. 18).

5. On the Project Schematic, double-click the Analysis cell to open it.

Note:

The order in which you define parameters determines the system-generated names for
the parameters. For this tutorial, you are using a specific order so that the parameter
names in your project will match the parameter names in the tutorial.

6. In the Outline pane, define parameters in the following order:

• In the Input column, select the check boxes for x1_ and x2_, which appear at the bottom
of the list.

• In the Output column, select the check boxes for f1_, f2_, g1_, and g2_.

The Outline pane will look like this.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
124 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 1

7. Close the Analysis cell. The Project Schematic now includes the Parameter Set bar.

8. Double-click the Parameter Set bar to open it. Given that you added the parameters in the order
indicated, the Outline pane looks like this:

9. Close the Parameter Set bar.

10. Update the project.

11. Save the project as DX_AMO_Problem1.wbpj in either this same directory or another directory.

In this project, you’ll add, configure, and update two Direct Optimization systems. One will use
MOGA, and the other will use AMO.

Run the MOGA Direct Optimization

First, add, configure, and update the MOGA Direct Optimization system.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 125
Using Adaptive Multiple-Objective Optimization

1. From under Design Exploration in the Toolbox, double Direct Optimization to add this system
to the Project Schematic.

2. Rename the system to MOGA Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

5. In the Properties pane, set optimization properties as shown:

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, specify objectives and constraints as shown:

8. In the Outline pane, select Domain.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
126 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 1

9. In the Table pane, set the lower and upper bounds as shown:

10. Update the Optimization cell.

The update may take a few minutes. While the update is in process, you can look at the History
charts to monitor the progress of the input parameters and the objectives and constraints. In the
Outline pane under Objectives and Constraints and Domain, a sparkline image of the History
chart is displayed for each object in the Monitoring column. To view the full-sized History chart
in the Graph pane, select the object in the Outline pane.

11. When the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that Converged is set to Yes.
Eight iterations and 374 design points were needed for convergence.

Note:

You can confirm this in the Table pane under Optimization Method, where Status
displays Converged after 374 evaluations.

12. In the Outline pane under Results, select Tradeoff.

In the Graph pane, you can see the Tradeoff chart, which shows the Pareto front found by the
optimization in blue by default.

For the following Tradeoff chart, in the Properties pane under Chart, Mode is set to 2D. Under
Axes, X Axis is set to P3 - f1_ and Y Axis is set to P4 - f2_ .

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 127
Using Adaptive Multiple-Objective Optimization

For the next Tradeoff chart, in the Properties pane under Chart, Mode is set to 3D. Under Axes,
Z Axis is set to P5 - g1_ .

13. Close the Optimization cell.

Run the AMO Direct Optimization

Next, add, configure, and update the AMO Direct Optimization system.

1. From under Design Exploration in the Toolbox, double-click Direct Optimization to add this
system to the Project Schematic.

2. Rename the system to AMO Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
128 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 1

5. In the Properties pane, set optimization properties as shown:

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, specify the same objectives and constraints as you did for the MOGA optimiz-
ation:

8. In the Outline pane, select Domain.

9. In the Table pane, set the lower and upper bounds the same as you did for the MOGA optimization:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 129
Using Adaptive Multiple-Objective Optimization

10. Update the Optimization cell.

While the update is in process, you can look at the History chart to monitor the progress of the
input parameters and the objectives and constraints as before. The AMO optimization runs much
more quickly than the previous MOGA optimization.

11. When the update is complete, in the Outline pane, select Optimization.

12. In the Properties pane under Optimization Status, Converged is set to Yes. You can see that 8
iterations and 131 design points were needed for convergence.

Note:

You can confirm this in the Table pane under Optimization Method, where Status
displays Converged after 131 evaluations.

13. In the Outline pane under Results, select Tradeoff.

As before, in the Graph pane you can see the Tradeoff chart, which shows the Pareto front found
by the optimization in blue by default.

For the following Tradeoff chart, in the Properties pane under Chart, Mode is set to 2D. Under
Axes, X Axis is set to P3 - f1_ and Y Axis is set to P4 - f2_ .

The Pareto front found by the AMO optimization is similar to the one found by the MOGA optim-
ization. While the results for the MOGA and AMO methods are similar, the number of design points

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
130 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

needed to reach these results differ greatly. While MOGA had to run 374 design points, AMO had
to run only 131 design points. Because AMO required 243 fewer points, the optimization process
was much faster.

Note:

In this particular example, the Kriging response surface had excellent error prediction,
so the optimization results are similar to those found by MOGA. This is not always the
case. The quality of the Kriging error prediction can vary according to the type of
function.

14. Close the Optimization cell.

15. Exit Workbench, saving your changes to this project.

Problem 2
This tutorial covers the following topics for Problem 2:
Problem 2 Definition
Problem 2 Project Setup
Run the MOGA Direct Optimization
Run the AMO Direct Optimization

Problem 2 Definition
The analytic function you’ll examine for this problem has six input parameters, two objectives, and
six constraints. The definition of the problem is as follows:

Minimize

With

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 131
Using Adaptive Multiple-Objective Optimization

Where , , and

Problem 2 Project Setup

1. Start Workbench.

2. From under Component Systems in the Toolbox, double-click Microsoft Office Excel to add
the system to the Project Schematic.

3. Right-click the Analysis cell and select Add File → Browse

4. In the file browser that opens, locate and open the file DX_AMO_input2.xlsx, which is in the
directory to which you extracted the input files (p. 18).

5. On the Project Schematic, double-click the Analysis cell to open it.

Note:

The order in which you define parameters determines the system-generated names for
the parameters. For this tutorial, you are using a specific order so that the parameter
names in your project will match the parameter names in the tutorial.

6. In the Analysis cell, define inputs and outputs in the following order:

• In the Input column, select the check boxes for x1_ through x6_, which appear at the
bottom of the list.

• In the Output column, select the check boxes for f1_ through f8_.

The Outline pane will look like this.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
132 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

7. Close the Analysis cell. The Project Schematic now includes the Parameter Set bar.

8. Double-click the Parameter Set bar to open it. Given that you added the parameters in the order
indicated, the Outline pane looks like this:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 133
Using Adaptive Multiple-Objective Optimization

9. Close the Parameter Set bar.

10. Update the project.

11. Save the project as DX_AMO_Problem2.wbpj in either this same directory or another directory.

In this project, you'll add, configure, and update two Direct Optimization systems. Again, one will
use MOGA, and the other will use AMO.

Run the MOGA Direct Optimization

First, you'll add, configure, and update the MOGA Direct Optimization system.

1. From under Design Exploration in the Toolbox, double-click Direct Optimization to add this
system to the Project Schematic.

2. Rename the system to MOGA Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
134 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

5. In the Properties pane, set optimization properties as shown:

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, set objectives and constraints as shown:

8. In the Outline pane, select Domain.

9. In the Table pane, set lower and upper bounds as shown:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 135
Using Adaptive Multiple-Objective Optimization

10. Update the Optimization cell.

The update takes several minutes. As before, you can watch the optimization process, either on
the sparkline image in the Outline pane or on the full History chart in the Graph pane.

11. When the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, Converged is set to Yes. You can see that
10 iterations and 535 design points were needed for convergence.

Note:

You can confirm this in the Table pane under Optimization Method, where Status
displays Converged after 535 evaluations.

12. In the Outline pane under Results, select Tradeoff.

As before, in the Graph pane, you can see the Tradeoff chart, which shows the Pareto front found
by the optimization in blue by default.

For the following Tradeoff chart, in the Properties pane under Chart, Mode is set to 2D. Under
Axes, X Axis is set to P7 - f1_ and Y Axis is set to P8 - f2_ .

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
136 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

For the next Tradeoff chart, in the Properties pane under Chart, Mode is set to 3D. Under Axes,
the X Axis and Y Axis settings remain the same. Z Axis is set to P9 - f3_.

13. Close the Optimization cell.

Run the AMO Direct Optimization

Next, you'll add, configure, and update the AMO Direct Optimization system.

1. Under Design Exploration in the Toolbox, double-click Direct Optimization to add the second
system to the Project Schematic.

2. Rename the system to AMO Direct Optimization.

3. In the system, double-click the Optimization cell to open it.

4. In the Outline pane, select Optimization.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 137
Using Adaptive Multiple-Objective Optimization

5. In the Properties pane, set optimization properties as shown:

6. In the Outline pane, select Objectives and Constraints.

7. In the Table pane, set the same objectives and constraints as you did for the MOGA optimization:

8. In the Outline pane, select Domain.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
138 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

9. In the Table pane, set the lower and upper bounds as you did for the MOGA optimization:

10. Update the Optimization cell.

As before, you can watch the optimization process, either on the sparkline image in the Outline
pane or on the full History chart in the Graph pane. The AMO optimization runs much more
quickly than the previous MOGA optimization.

11. When the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, Converged is set to Yes. You can see that
10 iterations and 144 design points were needed for convergence.

Note:

You can confirm this in the Table pane under Optimization Method, where Status
displays Converged after 144 evaluations..

12. In the Outline pane under Results, select Tradeoff.

As before, in the Graph pane, you can see the Tradeoff chart, which shows the Pareto front found
by the optimization in blue by default.

For the following Tradeoff chart, in the Properties pane under Chart, Mode is set to 2D. Under
Axes, X Axis is set to P7 - f1_ and Y Axis is set to P8 - f2_.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 139
Using Adaptive Multiple-Objective Optimization

For the next Tradeoff chart, in the Properties pane under Chart, Mode is set to 3D. Under Axes,
the X Axis and Y Axis settings remain the same. Z Axis is set to P9 - f3_.

The AMO optimization found a Pareto front that is approximately the same as the one found by
the MOGA optimization.

The number of design points needed to reach this result is especially significant. While MOGA
had to run 535 design points, AMO had to run only 144. Because AMO required 391 fewer points,
the optimization process was much faster.

13. Close the Optimization cell.

14. Exit Workbench, saving project changes.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
140 of ANSYS, Inc. and its subsidiaries and affiliates.
 What Have You Learned?

What Have You Learned?

Both MOGA and AMO are designed to perform an optimization with multiple objectives and constraints.
You used two separate problems to compare the ability of each method to find a Pareto front. You
then reviewed Tradeoff charts to compare the Pareto fronts found by the two methods. As you worked
through both problems, you observed that:

• MOGA obtains a high degree of accuracy, but it generally requires more design points to find
the Pareto front, which increases the time.

• AMO obtains a high degree of accuracy with fewer design points. However, the accuracy depends
on how well the Kriging error predictor is able to simulate the function being examined.

– With some functions (as in Problems 1 and 2), results are extremely accurate. The AMO
optimization finds the exact same Pareto front as the MOGA optimization, but the AMO
optimization uses far fewer design points.

– With other functions, the AMO optimization might find a Pareto front that approximates
the one found by the MOGA optimization, but the AMO optimization uses far fewer design
points.

During this tutorial, you have learned that AMO offers the accuracy of MOGA. AMO also applies the
Kriging error predictor to reduce the number of evaluations. In general, when time and resources are
considerations, AMO is the better choice for finding an accurate Pareto front.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 141
 Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
142 of ANSYS, Inc. and its subsidiaries and affiliates.
Using Mixed-Integer Sequential Quadratic
Programming (MISQP) Optimization
This advanced tutorial introduces you to the Mixed-Integer Sequential Quadratic Programming (MISQP)
optimization algorithm and shows how it supports all three types of input parameters: continuous,
continuous with manufacturable values, and discrete. To illustrate the effectiveness of MISQP, this tu-
torial has you apply several different optimization methods to the same problem.

Note:

This tutorial assumes that you are familiar with Ansys Workbench and DesignXplorer's goal-
driven optimization functionality. For more information on goal-driven optimization, see the
tutorial Performing a Goal-Driven Optimization (p. 71).

This tutorial is divided into the following sections:

Getting Started
Direct Optimization Using Screening
Response Surface Optimization to Direct Optimization Using Screening and MISQP
Direct Optimization Using MISQP
What Have You Learned?

Getting Started
Before beginning the tutorial, you should review the following topics:
What is MISQP?
Problem Definition
Project Setup

What is MISQP?
Adaptive Multiple-Objective (AMO) is a hybrid optimization method that combines a Kriging response
surface and the MOGA algorithm in a Direct Optimization system. It uses the same general approach
as MOGA but extends it by using the Kriging error predictor to reduce the number of evaluations
needed to locate the global optimum.

MISQP is a multiple-objective optimization algorithm that is available for both Direct Optimization
and Response Surface Optimization systems. It is used to solve mixed-integer nonlinear programming
by modified Sequential Quadratic programming (SQP), with the assumption that integer variables have
a smooth influence on the model functions. MISQP has the same settings as NLPQL, but unlike NLPQL,
it supports discrete parameters and continuous parameters with manufacturable values.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 143
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

Problem Definition
MISQP can be used to drive any Workbench project containing any combination of solvers. Because
almost everyone has access to Microsoft Office Excel, this tutorial uses an analytic function in Excel.
The optimization problem has five input parameters: two continuous and three continuous with
manufacturable values:

The global minimum is at point (1.33;100;1;-100;-1) with a corresponding objective function value
of -10099.69 and a constraint function value of 6.1564.

Below, the Local Sensitivity Curves chart for the function shows the impact of the five input parameters
on the output parameter P6 - f.

Project Setup
1. Start Workbench.

2. Select File → Open.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
144 of ANSYS, Inc. and its subsidiaries and affiliates.
 Direct Optimization Using Screening

3. In the file browser, locate and open the archived Workbench project file DX_MISQP.wbpz, which
is in the directory to which you extracted the input files (p. 18).

A dialog box appears for saving this archived file to a standard Workbench project file.

4. Save the project as DX_MISQP.wbpj in either this same directory or another directory.

Direct Optimization Using Screening

In this project, the Project Schematic already includes two systems: Microsoft Office Excel and Response
Surface Optimization.

You will use the Response Surface Optimization system to explore the parameters that have already
been defined and are used in both systems.

1. In the Response Surface Optimization system, double-click the Design of Experiments cell
to open it.

In the Outline pane under Input Parameters, you can see that five inputs are defined.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 145
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

2. Select any one of these input parameters: P3 - x_3, P4 - x_4, or P5 - x_5.

In the Properties pane, you can see that Classification is set to Continuous and Allowed
Values is set to Manufacturable Values.

In the Table pane for any of these three parameters, you can see levels that define the manu-
facturable values. Each of them has the same 201 levels.

The following figure shows properties and manufacturable values for P3 - x_3.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
146 of ANSYS, Inc. and its subsidiaries and affiliates.
 Response Surface Optimization to Direct Optimization Using Screening and MISQP

The number of manufacturable values in the problem means that for full-factorial optimization
of just these three parameters, the combination of values is equal to 8120601 (201^3). Unless
each solve is very fast and you are running a Screening optimization on a cluster of machines
via Ansys Remote Solve Manager, running all of the combinations is difficult. Consequently,
you can conclude that for this problem, a Direct Optimization system using Screening is not
likely to discover the optimum.

The only way to study all of these combinations is to assume that they have a smooth influence
on the model functions and then either interpolate a response surface or use an optimization
method that includes this assumption, such as MISQP.

3. Close the Design of Experiments cell.

Response Surface Optimization to Direct Optimization Using Screening

and MISQP
For this problem, you’ll use Screening and MISQP in two Response Surface Optimization systems to
find a good starting point for MISQP in a Direct Optimization system. These are basically the steps
that you’ll take:

1. Run Screening in a Response Surface Optimization system to find a good candidate point.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 147
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

2. Run MISQP in a Response Surface Optimization system, using the candidate point found by the
Screening as its starting point.

3. Run MISQP in a Direct Optimization system, using the candidate point found by the previous
MISQP as a starting point.

Run Screening in a Response Surface Optimization

You can try this approach because the problem doesn’t have a large number of input parameters
and because you can evaluate the analytic functions at any point: both on manufacturable values
and between manufacturable values—for example, at point (1.33;100;1;-100;-1) This means that it
may be feasible to build a response surface.

First, you will run a Screening optimization to find a feasible candidate point to serve as a starting
point for an MISQP optimization.

The Project Schematic already has a Response Surface Optimization system named Screening
= 10000 samples.

For each cell, you need to configure properties and then update it and view results. For this first
system, properties are already set. However, follow the instructions for setting properties and then
update the cells as described.

Configure and Update the Design of Experiments Cell

1. In the Project Schematic, double-click the Design of Experiments cell to open it.

2. In the Outline pane, select Design of Experiments.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
148 of ANSYS, Inc. and its subsidiaries and affiliates.
 Response Surface Optimization to Direct Optimization Using Screening and MISQP

3. In the Properties pane:

a. Set Design of Experiments Type to Central Composite Design.

b. Set Design Type to Auto Defined.

4. Update the Design of Experiments cell.

When the update is complete, the Table pane displays 27 design points and their results.

5. Close the Design of Experiments cell.

Configure and Update the Response Surface Cell

1. In the Project Schematic, double-click the Response Surface cell to open it.

2. In the Outline pane, select Response Surface.

3. In the Properties pane:

a. Set Response Surface Type to Kriging.

b. Set Maximum Predicted Relative Error % to 5.

4. Update the Response Surface cell.

The Table pane displays a response point.

5. Close the Response Surface cell.

Note:

Kriging with auto-refinement does not add new refinement points because, in this case,
the response surface generated from the DOE is sufficiently accurate.

Configure and Update the Optimization Cell

1. In the Project Schematic, double-click the Optimization cell to open it.

2. In the Outline pane, select Optimization.

3. In the Properties pane:

a. Set Method Selection to Manual.

b. Set Method Name to Screening.

c. Set Number of Samples to 10000.

4. In the Outline pane, select Objectives and Constraints.

The Table pane displays the objectives and constraints that have already been set:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 149
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

5. Update the Optimization cell.

6. When the update is complete, in the Outline pane, select Optimization.

In the Table pane, you can see that Candidate Point 2 is the best candidate (11.51;77.866;3;-
94;7), with a corresponding objective value of -9043.7 and a constraint value of 166.32. This
sample was chosen (screened) as the best of the 10,000 that were generated using the Kriging
response surface meta-model, which was generated from the 27 design points that were actually
solved for the DOE.

The verification process has already been run. It takes the candidate input points and runs that
design point through the actual solver (in this case, Excel). For this problem, verification found
that the candidate was even better than predicted. The difference is due to the accuracy of the
response surface prediction of the sample based on the 27 design points that were actually solved.
However, because the Screening method is good at separating global maxima (or minima) from
local maxima, this would be a good place to start an optimization with a more efficient method
or even a direct solve method such as MISQP.

7. Close the Optimization cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
150 of ANSYS, Inc. and its subsidiaries and affiliates.
 Response Surface Optimization to Direct Optimization Using Screening and MISQP

Run MISQP in a Response Surface Optimization

Now that Screening has found a suitable candidate, you’ll use this candidate as the Starting Value
in an MISQP response surface optimization.

Add Another Response Surface System

1. To add the second Response Surface Optimization system in the Project Schematic, right-click
the header of the existing Response Surface Optimization system and select Duplicate.

2. Rename this new system to MISQP based on Response Surface: Starting Point
= Screening Candidate Point.

3. Drag the Design of Experiments cell in the original system and drop it on the Design of Exper-
iments cell in the new system so that these two cells share data.

4. In the dialog box that opens, click OK to confirm that the two cells are to share data.

5. Drag the Response Surface cell in the original system and drop it on the Response Surface cell
in the new system so that these two cells share data.

6. In the dialog box that opens, click OK to confirm that the two cells are to share data.

The Project Schematic will now look like this:

Configure and Update the Optimization Cell

1. In the second Response Surface Optimization system, double-click the Optimization cell to
open it.

2. In the Outline pane, select Optimization.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 151
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

3. In the Properties pane:

a. Set Method Selection to Manual.

b. Set Method Name to MISQP.

c. Set Allowable Convergence (%) to 1E-06.

d. Set Maximum Number of Iterations to 55.

4. In the Outline pane, select Objectives and Constraints.

5. In the Table pane, set the same objectives and constraints as in the other system:

6. In the Outline pane, select Domain.

When Optimization Type is set to MISQP, the Table pane provides for entering not only lower
and upper bounds but also starting values.

7. Adjust the starting values to match those for the candidate found by Screening (11.51;77.866;3;-
94;7) as shown:

8. Update the Optimization cell.

9. Once the update is complete, in the Outline pane, select Optimization.

In the Table pane, you can see that Candidate Point 1 is the best candidate (1.323;94.688;1;-
100;-1), with an objective value of -10045 and a constraint value of 6.1205. These values have
been evaluated on the Kriging response surface.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
152 of ANSYS, Inc. and its subsidiaries and affiliates.
 Response Surface Optimization to Direct Optimization Using Screening and MISQP

10. Right-click Candidate Point 1 and select Verify By Design Point Update.

In the following figure, these values have been verified by a design point update. For this verific-
ation point, the objective value is -10094 and the constraint value remains the same at 6.1205.
Because this verification point (real solve) is very close to the predicted value, you now that the
Kriging response surface is very accurate in this location.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 153
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

Screening, which is somewhat random, missed this better location on the model because its res-
olution was not high enough. The benefit of Screening is that it spreads its samples across the
entire domain so that it is better able to find the region of the global minimum or maximum. The
region of the global minimum or maximum is an ideal starting point for an algorithm like MISQP,
which is capable of a more active search but might otherwise have fallen into a local minimum.

11. Close the Optimization cell.

Run MISQP in a Direct Optimization

Finally, you’ll use the candidate found by the MISQP response surface optimization as the starting
value in an MISQP direct optimization.

Add a Direct Optimization System

1. Under Design Exploration in the Toolbox, drag the Direct Optimization system template and
drop it on the Project Schematic to the right of the last system.

2. Rename it to MISQP: Starting Point = MISQP Candidate Point from RSO.

The Project Schematic will now look like this:

Configure and Update the Optimization Cell

1. In the Direct Optimization system, double-click the Optimization cell to open it.

2. In the Outline pane, select Optimization.

3. In the Properties pane:

a. Set Method Selection to Manual.

b. Set Method Name to MISQP.

c. Clear the Tolerance Setting check box.

d. Set Allowable Convergence (%) to 1E-06.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
154 of ANSYS, Inc. and its subsidiaries and affiliates.
 Response Surface Optimization to Direct Optimization Using Screening and MISQP

e. Set Maximum Number of Iterations to 20.

4. In the Outline pane, select Objectives and Constraints.

5. In the Table pane, set the same objectives and constraints as in the other two systems:

6. In the Outline pane under Domain, do the following for input parameters P3 - x_3, P4 - x_4, and
P5 - x_5 in their respective Properties panes:

a. Set Allowed Values to Manufacturable Values.

b. Add the 201 manufacturable value levels by copying all manufacturable values from the
Design of Experiments cell in the first Response Surface Optimization system and
pasting them into this cell.

To perform the paste operation, right-click in the Manufacturable Values column for the
first row and select Paste. If a dialog box opens, indicating that an entered value falls
outside of the current variation range and that existing results and design points will be
lost, click OK.

7. In the Outline pane, select Domain.

When Optimization Type is set to MISQP, the Table pane provides for entering not only lower
and upper bounds but also starting values.

8. Adjust the starting values to match those for the candidate found by the previous MISQP optim-
ization (1.323;94.688;1;-100;-1) as shown:

9. Update the Optimization cell.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 155
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

Once the update is complete, in the Outline pane, select Optimization.

In the Table pane, the best candidate point is (0.82299;95.014;1;-100;-1), with an objective value
of –10094 and a constraint value of 3.7973.

Response surface optimization allows you to approximate the objective and constraint values,
explore them in the parameter space with a minimum number of design points, and find a good
candidate point as starting point of the direct optimization. Essentially, this first step in a response
surface-based exploration is an acceleration of a direct optimization. MISQP convergence can be
very dependent on starting values, especially for non-convex problems.

10. Close the Optimization cell.

Direct Optimization Using MISQP

As demonstrated by the previous system, using response surface optimization can be an effective ap-
proach to finding a good starting value for MISQP. However, it’s not always possible to build a response
surface. Possible reasons include the following:

• Too many input parameters

• Too many combinations of discrete parameter levels

• Too difficult to approximate the objective or constraint values

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
156 of ANSYS, Inc. and its subsidiaries and affiliates.
 Direct Optimization Using MISQP

When a response surface cannot be generated, a direct optimization is necessary. In this scenario, you
will use MISQP for a direct optimization.

Run MISQP in a Direct Optimization with a Maximum of 20 Iterations

You'll now add a Direct Optimization system that uses the central point of the domain as the starting
value for the MISQP optimization.

Add a Direct Optimization System

1. To add the system in the Project Schematic, right-click the header of the existing Direct Optim-
ization system and select Duplicate.

2. Rename this new system to MISQP (MaxIter = 20): Starting Point = Central
Point.

The Project Schematic should now look like this:

Configure and Update the Optimization Cell

1. In this second Direct Optimization system, double-click the Optimization cell to open it.

2. In the Outline pane, select Optimization.

3. In the Properties pane:

a. Set Finite Difference Approximation to Central to indicate the method for calculating
the derivative.

b. Set Allowable Convergence (%) to 1E-06.

c. Set Maximum Number of Iterations to 20.

4. In the Outline pane, select Objectives and Constraints.

The Table pane already displays the same objectives and constraints as in the other three systems:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 157
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

5. In the Outline pane, select Domain.

6. Set all starting values to 0 because you are assuming a scenario with no response surface to find
a starting point.

The Table pane should look like this:

7. Update the Optimization cell.

8. Once the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that MISQP reaches the maximum
number of iterations (20) with 222 design points, and it does not converge.

In the Table pane, the best candidate point is (5.9531;100;2;-100;24), with an objective value of
–9416.1 and a constraint value of 34.119.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
158 of ANSYS, Inc. and its subsidiaries and affiliates.
 Direct Optimization Using MISQP

9. Close the Optimization cell.

Run MISQP in a Direct Optimization with a Maximum of 50 Iterations

In the previous MISQP direct optimization, the candidate is very close to the optimum. To try to reach
convergence, you can perform the same optimization but increase the maximum number of iterations.

Add Another Direct Optimization System

1. To add another Direct Optimization system in the Project Schematic, right-click the header of
the previous Direct Optimization system and select Duplicate.

2. Rename this new system to MISQP (MaxIter = 50): Starting Point = Central
Point.

The Project Schematic should now look like this:

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 159
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

Configure and Update the Optimization Cell

1. In this third Direct Optimization system, double-click the Optimization cell to open it.

2. In the Outline pane, select Optimization.

3. In the Properties pane, set Maximum Number of Iterations to 50, leaving the settings for all
other properties the same.

4. Update the Optimization cell.

Because you changed only Maximum Number of Iterations, this MISQP optimization reuses the
design points generated by the previous one.

5. Once the update is complete, in the Outline pane, select Optimization.

In the Properties pane under Optimization Status, you can see that MISQP converges in 40 iter-
ations with 402 design points.

In the Table pane, you can see that MISQP did now find the optimum point (1.33;100;1;-100;-
1), with an objective value of –10100 and a constraint value of 6.1564.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
160 of ANSYS, Inc. and its subsidiaries and affiliates.
 What Have You Learned?

The candidate point found by this second MISQP is more accurate than the candidate point found
by the first MISQP. To achieve the improved accuracy, the second optimization required more it-
erations and more design points (402 versus 222).

Tip:

DesignXplorer’s History charts allow you to view the optimization history of objective
and constraint values. You can stop the optimization at any time if you think that the
current values are good enough. In this case, the candidate points will be extracted
from the existing samples.

6. Close the Optimization cell.

7. Exit Workbench, saving project changes.

What Have You Learned?

In summary, for the analytic function problem, you tried three different optimization approaches:

1. Running a Screening direct optimization

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 161
Using Mixed-Integer Sequential Quadratic Programming (MISQP) Optimization

Screening is an exhaustive method that distributes solutions across all Pareto fronts (across the
entire design space, instead of focusing refinement on a section). A direct optimization uses
real solves rather than estimates. In addition, 3 of the 5 input parameters each have 201 man-
ufacturable value levels. Given these factors, a Screening direct optimization is not likely to find
the optimum. The only way to study all of the combinations is to assume that they have a
smooth influence on the model functions and then run either a DOE and interpolate a response
surface or use an optimization method like MISQP that includes this assumption.

2. Running Screening and MISQP response surface optimizations to find a starting value for
an MISQP direct optimization

MISQP can be very dependent on the quality of its starting point, especially in a non-convex
problem. The closer the starting point is to the optimum point, the better the probability of
finding the optimum and the fewer the design points needed for convergence. You used a
Screening response surface optimization to locate a starting point for an MISQP response surface
optimization, which in turn found a starting point for an MISQP direct optimization. Because
MISQP had a good starting point, the MISQP direct optimization needed only 222 design points
to locate the optimum. In cases where it is possible to build a response surface, this is a good
approach for solving problems of this type.

3. Running an MISQP direct optimization

In cases where it is not possible to build a response surface, you can run MISQP in a direct op-
timization. MISQP still needs a good starting point, so in the absence of a response surface, you
used the central point of the domain to initialize the optimization. In the first attempt, the op-
timization did not converge, but you could see that it was close to convergence. When you
increased the maximum number of iterations, MISQP converged in 402 design points and ob-
tained slightly more accurate results. Although more design points are needed for this approach,
it is still possible to find the global minimum, even without a preliminary response surface.

The following table summarizes the results for each approach.

x1 x2 x3 x4 x5 f g Design
Points
Needed
Global Minimum 1.33 100 1 -100 -1 -10099.69 6.1564 ---
Response Surface 11.51 77.866 3 -94 7 -9043.7 166.32 27+1
Optimization
Screening
Response Surface 1.323 94.688 1 -100 -1 -10045 6.12095 53+1
Optimization MISQP
(Starting Value =
RSO Screening
Candidate)
Direct Optimization 0.82299 95.014 1 -100 -1 -10094 3.7973 102
MISQP (Starting
Value = RSO MISQP
Candidate)
Direct Optimization -5.9531 100 2 -100 24 -9416.1 34.119 222
MISQP

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
162 of ANSYS, Inc. and its subsidiaries and affiliates.
 What Have You Learned?

x1 x2 x3 x4 x5 f g Design
Points
Needed
(Max. Iterations =
20; Staring Value =
Central Point)
Direct Optimization 1.33 100 1 -100 -1 -10100 6.1564 402
MISQP

(Max. Iterations =
50; Staring Value =
Central Point)

Multiple-Objective Optimization Problems

For multiple-objective optimization problems, you can define a derived output parameter that is a
combination of your objective functions.

Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 163
 Release 2024 R1 - © ANSYS, Inc. All rights reserved. - Contains proprietary and confidential information
164 of ANSYS, Inc. and its subsidiaries and affiliates.