Use ExpertAI for Expert Emulation

Tutorial Level: Advanced In this tutorial, you will learn how to use the expertAI tools in Design Explorer by running two optimizations. the first a standard solver-based optimization, and the subsequent one through expertAI to allow for subjective criteria to be considered in the optimization.

Before you begin, copy the file(s) used in this tutorial to your working directory:

Open the Model

  1. Start HyperMesh.
  2. From the menu bar, click File > Open > HyperMesh Model.
  3. Browse to your working directory, select clip.hm, and click Open.
    If prompted, click Change to change the solver interface to OptiStruct while opening the model.
    A finite element model appears in the modeling window.
Figure 1.


Create an Exploration

  1. From the Design Explorer ribbon, Exploration tool group, click the Create Explorations tool.
    Figure 2.


    The Explorations dialog opens.
  2. Click then select Optimization.
  3. In the Study Path field, browse to and select the folder to store your exploration files.
    Figure 3.


  4. Close the Explorations dialog.
  5. From the Exploration tool group, click the Design Explorer tool.
    Figure 4.


    The Design Explorer browser opens. You can see the newly created optimization exploration. Additional exploration entities will appear here as well.

Create the Exploration Inputs

  1. From the Design Explorer ribbon, click the Gauge tool.
    Figure 5.


  2. In the modeling window, select the Fuse property of the main clip.
    Figure 6.


  3. In the microdialog, change the lower and upper bound percentages to 50.
    Figure 7.


  4. Click Create.
One gauge design variable is created.
Figure 8.


Create the Exploration Responses

  1. From the Design Explorer ribbon, click the Mass/Volume tool.
    Figure 9.


  2. On the guide bar, click to create the response.
  3. In the microdialog, change the drop-down option from Response to Objective.
    An objective is created.
  4. On the guide bar, click .
  5. From the Design Explorer ribbon, click the Disps. tool.
    Figure 10.


  6. On the guide bar, click to open the Advanced Selection dialog.
  7. In the Advanced Selection dialog, select By ID, enter 904, and click OK.
    A displacement response is created.
  8. On the guide bar, click .
The optimization now consists of one design variable, two response variables, and one objective.
Figure 11.


Evaluate the Exploration

  1. From the Design Explorer ribbon, Evaluate tool group, click the Evaluate tool.
    Figure 12.


    The Evaluate dialog opens.
  2. Change the No. of Runs to 10.
  3. Optional: Increase the number of concurrent runs by increasing the Multi Execution number.
  4. Click Run.
    The optimization is evaluated. In this case, there will be a nominal run plus ten optimization runs. This may take a few minutes depending on your computer.

    When the evaluation is complete, the Evaluation Status dialog should look like Figure 13.

    Figure 13.


  5. Close the Evaluation Status dialog.

Review the Evaluation

  1. From the Design Explorer ribbon, Evaluate tool group, click the Results Explorer tool.
    Figure 14.


    The Results Explorer Browser opens.
  2. Review the Summary table, which shows the input and response values for each run of the optimization, including the optimal run, in green.
    Figure 15.


  3. Right-click on the optimal run, run 5, and select Load Results from the context menu.
The results are loaded into a HyperView window, shown in Figure 16 when set to the final timestep.
Figure 16.


We have achieved our optimization objective to minimize the mass of the clip. However, the resulting breakage shown in Figure 16 is undesirable. We would like to meet our objective while trying to ensure that the clip does not break.

Expert Emulation

  1. If it's not the active client, make the original HyperMesh window active, and return to the Results Explorer Browser.
  2. In the Results Explorer Browser, select the Expert AI tab.
  3. Next to the Dataset drop-down, click .
    A clustering report is generated from the original optimization.
  4. In the Dataset dialog, make the following selections, as shown in Figure 17.
    1. For Result Type, select Displacement.
    2. For Steps, select 1.0000e-02.
    3. For Part ID, select Base and Fuse.
    Figure 17.


  5. Click Request.
    The dataset is created.
  6. Click Submit.
    Using unsupervised machine learning, the results from the original optimization have been clustered by their displacements.

    You can select either cluster to see which optimization runs it contains.

  7. In the Results Explorer Browser, right-click in the area beneath the two clusters and select Animate all Clusters from the context menu.
    A new page is created containing animations, one for each cluster. Each animation cycles through the displacements for each run in a given cluster.
    Figure 18.


  8. Click to animate the windows and cycle through the runs.
    Cluster 1 contains all the runs resulting in a broken clip. Cluster 2 contains all the runs where the clip remains intact.
  9. Click to stop the animations.
  10. Return to Page 1, make the HyperMesh window active, and return to the Results Explorer tab.
  11. Beneath the cluster area, click Train Classifier.
    A classifier is trained where we can supervised machine learning, based on our previous observations of the clusters.
  12. Right-click on the classifier, and select Create Constraint from the context menu.
  13. In the Constraint dialog, select Cluster2 for Cluster.
  14. For Lower Bound, enter 0.7 and click OK.
    Figure 19.


    A constraint is created.
  15. Click Optimize, and select Solver-Based.
    The Evaluate dialog opens.
  16. Change the No. of Runs to 10.
  17. Optional: Increase the number of concurrent runs by increasing the Multi Execution number.
  18. Click Run.
    The optimization is evaluated. In this case, there will be a nominal run plus ten optimization runs. This may take a few minutes depending on your computer.
    When the evaluation is complete, the Evaluation Status dialog should look like Figure 20.
    Figure 20.


  19. Close the Evaluation Status dialog.

Review the Evaluation

  1. From the Design Explorer ribbon, Evaluate tool group, click the Results Explorer tool.
    Figure 21.


    The Results Explorer Browser opens.
  2. Ensure that Optimzation_2 is the selected exploration, and review the Summary table, Figure 22, which shows the optimal run, minimizing the mass of the click while also conforming to the constraint that the clip does not break.
    Figure 22.


    Figure 22 shows the loaded results for the optimal run, confirming that the clip is not broken.
    Figure 23.