Using Adaptive Multiple-Objective Optimization

In this advanced tutorial, we’ll use the Adaptive Multiple-Objective (AMO) optimization method as
part of a Direct Optimization system to locate a Pareto front. To illustrate how AMO optimization works,
we will create two separate projects to examine different functions and apply both the Multi-Objective
Genetic Algorithm (MOGA) and the Adaptive Multiple-Objective (AMO) optimization methods to
the problem. Then, we will review the results and examine why Adaptive Multiple-Objective method 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
DesignXplorer’s Goal Driven Optimization functionality. For details on Goal Driven Op-
timization in version 14.5, see the tutorial Performing a Goal Driven Optimization
Study.

This tutorial is divided into the following sections:

1. What is Adaptive Multiple-Objective Optimization?
2. Problem 1
3. Problem 2
4. What Have We Learned?

1. What is Adaptive Multiple-Objective Optimization?

Adaptive Multiple-Objective is a hybrid optimization method which 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 local the global optimum.

Download the Project Input Files

1. To access tutorials and their input files on the ANSYS Customer Portal, go to http://support.ansys.com/
training.

2. Download the ANSYS_DX_AMO.zip file to your working folder.

3. Extract the files DX_AMO_input1.xlsx and DX_AMO_input2.xlsx.

2. Problem 1
This tutorial covers the following topics for Problem 1:
2.1. Problem 1: Definition
2.2. Problem 1: Project Setup
2.3. Problem 1: Optimization Setup

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 1
Using Adaptive Multiple-Objective Optimization

2.1. Problem 1: Definition

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

Minimize
2 2
f1 ( x1 x2 ) = x1 + x2 − + x2 − x1 +

 (

 ) =
 − +

With
3
(
g    )=  − +  − ≤

 ( 
 ) = 
+  − (  − 
+ ) − ≤

≤  ≤ ≤  ≤
Where and .

2.2. Problem 1: Project Setup

1. Open ANSYS Workbench 14.5.

2. Create a new project.

3. Add a Microsoft Office Excel to the Project Schematic.

4. Attach the input file DX_AMO_input1.xlsx.

5. In the Analysis workspace, define parameters in the following order:

• Define x1_ and x2_ as inputs.

• Define f1_, f2_, g1_, and g2_ as outputs.

Note

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

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
2 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 1

6. Return to the Project Schematic. Note that the Parameter Set has been added.

Note

You can verify the parameters in the Parameter Set workspace.

7. Update the project and save it as DX_AMO_Problem1.wbpj.

2.3. Problem 1: Optimization Setup

For Problem 1, we’ll add, configure, and update two Direct Optimization systems. One will use the
MOGA optimization method, and the other will use the Adaptive Multiple-Objective optimization
method.

2.3.1. Run the MOGA Optimization

1. Add the first Direct Optimization system to the Project Schematic, renaming it MOGA.

2. On the Project Schematic, right-click the Optimization component and select Edit to open the Optimiz-
ation workspace.

3. In the Outline view:

• In the Optimization node Properties view, set Optimization Method to MOGA

• In the Objectives and Constraints node Table view, specify objectives and constraints as follows:

4. Under the Domain node, select each input parameter in turn. In the Properties view, set the Lower
Bound, and Upper Bound properties as follows:

• For P1 —X1_:

– Set Lower Bound to 0.

– Set Upper Bound to 5.

• For P2 —X2_:

– Set Lower Bound to 0.

– Set Upper Bound to 3.

5. Update the project.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 3
Using Adaptive Multiple-Objective Optimization

6. The update may take a few minutes. 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. In the Outline view,
a sparkline image of the History chart is displayed next to each object. To view the full-sized History
chart in the Charts view, select the object in the Outline view.

7. When the update is complete, you can review final optimization data by selecting the Optimization
node of the Outline view. In the Properties view, you can see that the Converged status is Yes and that
11 iterations and 474 evaluations were needed for convergence.

Note

You can confirm this in the Table view; the optimization Status says “Converged after
474 evaluations.”

8. Under the Outline view Results node, click Tradeoff.

In the Charts view, the 2D version of the Tradeoff chart shows the Pareto front found by the optim-
ization. You can also set the chart Mode property to 3D to see the three-dimensional version. The
Pareto front is displayed in blue for both versions.

9. Return to the Project Schematic.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
4 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 1

2.3.2. Run the Adaptive Multiple-Objective Optimization

1. Add the second Direct Optimization system to the Project Schematic, renaming it Adaptive Multiple-
Optimization.

2. Right-click the Optimization component and select Edit to open the Optimization workspace.

3. Select the Outline view Optimization node. In the Properties view, set Optimization Method to Adaptive
Multiple-Optimization.

4. Right-click the Objectives and Constraints node. In the Table view, specify the same objectives and
constraints as you did for the MOGA optimization.

5. Under the Domain node of the Outline view, select each input parameter in turn. In the Properties view,
set Lower Bound, and Upper Bound properties to the same values as you did for the MOGA optimization.

6. Update the project.

7. While the update is in process, you can monitor the optimization via the History chart, as before. (Note
the difference in the time available for reviewing the History chart, compared with the MOGA optimization.)

8. When the update is complete, you can review final optimization data as before, by selecting the Optim-
ization node. In the Properties view, you can see that the Converged status is Yes and that 11 iterations
and 131 evaluations were needed for convergence.

Note

You can confirm this in the Table view; the optimization Status says “Converged after
131 evaluations.”

9. Under the Outline view Results node, click on Tradeoff.

As before, in the Charts view, you can view the Pareto front found by the optimization in both the
2D and 3D versions of the Tradeoff chart.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 5
Using Adaptive Multiple-Objective Optimization

Note that Pareto front found by the Adaptive Multiple-Objective optimization is exactly the same
one that was found by the MOGA optimization; the results for each method are identical. The differ-
ence is in the number of design points needed to reach this result. Whereas MOGA had to run 474
design points, Adaptive Multiple-Objective only had to run 131 design points. Because the Adaptive
Multiple-Objective method required 353 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 were identical to those found by the MOGA algorithm. This may
not always be the case, however; the quality of the Kriging error prediction can vary ac-
cording to the type of function.

3. Problem 2
This tutorial covers the following topics for Problem 2:
3.1. Problem 2: Definition
3.2. Problem 2: Project Setup
3.3. Problem 2: Optimization Setup

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
6 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

3.1. Problem 2: Definition

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

Minimize
f1 (x) = − x1 − 2
+ ( x2 − )2 + x3 − 2
+ ( x4 − )2 + x5 − 2

n

 (
) = ∑
i
i =

With
g  ( )= − (  +  ) ≤

 (  ) = ( 
+  ) − ≤

 () = − (  −  ) − ≤

 (  ) = (  −  )− ≤


 (  ) = (  − ) +   − ≤


6 () = − ( 

− 
+ 6 ≤ )
≤ " " "
!
≤ ≤ %# %
$
≤ ≤ '& ≤
Where , , .

3.2. Problem 2: Project Setup

1. Open ANSYS Workbench 14.5.

2. Create a new project.

3. Add a Microsoft Office Excel system to the Project Schematic.

4. Attach the input file, DX_AMO_input2.xlsx.

5. In the Analysis workspace, define inputs and outputs in the following order:

• Define x1_ throughx6_ as inputs.

• Define f1_, f8_ as outputs.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 7
Using Adaptive Multiple-Objective Optimization

Note

Again, for the purposes of this tutorial, we are defining parameters in specific order so
that the parameter names in your project will match the parameter names in the tutorial
examples.

6. Return to the Project Schematic. The Parameter Set has been added.

Note

You can verify the parameters in the Parameter Set workspace.

7. Update the project.

8. Save the project as DX_AMO_Problem2.wbpj.

3.3. Problem 2: Optimization Setup

For Problem 2, we will add, configure, and update two Direct Optimization systems. Again, one will use
the MOGA optimization method, and the other will use the Adaptive Multiple-Objective optimization
method.

3.3.1. Run the MOGA Optimization

Run the MOGA Optimization
1. Add the first Direct Optimization system to the Project Schematic, renaming it MOGA.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
8 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

2. On the Project Schematic, right-click the Optimization component and select Edit to open the Optimiz-
ation workspace.

3. Select the Outline view Optimization node. In the Properties view, set Optimization Method to MOGA.

4. Select Objectives and Constraints node. In the Table view, specify objectives and constraints as follows:

5. Under the Domain node, select each input parameter in turn. In the Properties view, set the Lower
Bound, and Upper Bound properties as follows:

6. Update the project.

7. The update may take a few minutes. As before, you can watch the optimization process, either on the
full History chart in the Charts view or on the sparkline image in the Outline view.

8. When the update is complete, you can review final optimization data by selecting the Optimization
node. In the Properties view, you can see that the Converged status is Yes and that 19 iterations and
974 evaluations were needed.

9. Under the Results node in the Outline view, click Tradeoff.

In the Charts view, the 2D version of the Tradeoff chart shows the Pareto front found by the optim-
ization. You can also set the chart Mode property to 3D to see the three-dimensional version. The
Pareto front is displayed in blue for both versions.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 9
Using Adaptive Multiple-Objective Optimization

10. Return to the Project Schematic.

3.3.2. Run the Adaptive Multiple-Objective Optimization

1. Add the second Direct Optimization system to the Project Schematic, renaming it Adaptive Multiple-
Optimization.

2. On the Project Schematic, right-click the Optimization component and select Edit to open the Optimiz-
ation workspace.

3. Select the Outline view Optimization node. In the Properties view, set Optimization Method to Adaptive
Multiple-Optimization.

4. Right-click the Objectives and Constraints node. In the Table view, specify the same objectives and
constraints as you did for the MOGA optimization.

5. Under the Domain node, select each input parameter in turn. In the Properties view, set Lower Bound,
and Upper Bound properties to the same values as you did for the MOGA optimization.

6. Update the project.

7. While the update is in process, you can monitor the optimization via the History chart, as before. (Note
the difference in the time available for reviewing the History chart, compared with the MOGA optimization.)

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
10 of ANSYS, Inc. and its subsidiaries and affiliates.
 Problem 2

8. When the update is complete, you can review final optimization data as before, by selecting the Optim-
ization node of the Outline view. In the Properties view, you can see that the optimization ran through
20 iterations (the maximum number defined) and ran 150 design points, but that the optimization has
not converged yet.

Note

You can confirm this in the Table view.

9. Under the Results node in the Outline view, click Tradeoff.

As before, in the Charts view, you can view the Pareto front found by the optimization in both the
2D and 3D versions of the Tradeoff chart.

Although the Adaptive Multiple-Objective optimization did not converge, it still found a Pareto front
that is approximately the same as the one found by the MOGA optimization; when you compare
the two Pareto fronts side-by-side, you can see that they are not identical, but are very similar.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 11
Using Adaptive Multiple-Objective Optimization

The number of design points needed to reach this result is especially significant. Whereas MOGA
had to run 974 design points, Adaptive Multiple-Objective only had to run 150 design points. Because
the Adaptive Multiple-Objective method required 824 fewer points, the optimization process was
much faster.

If you want the Adaptive Multiple-Objective optimization to converge, try raising the maximum number
of evaluations possible. Change Maximum Number of Iterations from 20 (the default) to 30 and update
the optimization again.

Now, when you look in the Properties and Table views, you can see that the optimization has converged.
The optimization required 3 more iterations to converge (for a total of 23 iterations), but because it
was very close to convergence before, did not require any additional design points (the Number of
Evaluations is still 150). When you compare all three of the Problem 2 Tradeoff charts side-by-side,
you can see that finding a Pareto front with the Adaptive Multiple-Objective method is less expensive
than using MOGA.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
12 of ANSYS, Inc. and its subsidiaries and affiliates.
 What Have We Learned?

4. What Have We Learned?

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

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
of ANSYS, Inc. and its subsidiaries and affiliates. 13
Using Adaptive Multiple-Objective Optimization

• MOGA obtains a high degree of accuracy, but in general requires more design points (and so takes
a longer time) to find the Pareto front.

• Adaptive Multiple-Objective obtains a high degree of accuracy with fewer design points, but the
accuracy depends on how well the Kriging error predictor is able to simulate the function being ex-
amined.

– With some functions (as in Problem 1), results are extremely accurate; the optimization finds the
exact same Pareto front as the one found by the MOGA optimization, but uses fewer design points.

– With other functions (as in Problem 2), results are fairly accurate; the optimization finds a Pareto
front that approximates the one found by the MOGA optimization, but uses far fewer design points.

During this tutorial, we have learned that the Adaptive Multiple-Objective optimization method offers
the accuracy of the MOGA algorithm, but also applies the Kriging error predictor to reduce the number
of evaluations. In general, the Adaptive Multiple-Objective method is the better choice for finding an
accurate Pareto front when time and resources are a consideration.

Release 14.5 - © SAS IP, Inc. All rights reserved. - Contains proprietary and confidential information
14 of ANSYS, Inc. and its subsidiaries and affiliates.