ExpertAI

Optimization using unsupervised machine learning, group exploration results according to user-specified criteria.

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


    The Results Explorer Browser opens.
  2. In the Results Explorer Browser, click ExpertAI.
    Figure 2.


    The ExpertAI panel is divided into three areas: Dataset and Clustering, Classifiers, and Constraints.
    Dataset
    Create datasets based on response type, like displacement. Review and visualize clustered results.
    Classifiers
    Generate classifiers from labeled cluster data. A trained classifier can be leveraged as a constraint in a subsequent optimization.
    Constraints
    Create and evaluate an optimization using a constraint created based on selected clusters.
  3. Create a new dataset.
  4. Click to create a new dataset.
  5. In the Dataset dialog, select a Subcase.
  6. Select one or more Result Types, a Step, and Part IDs of parts to be included in the clustering.
    Note:

    If no Steps are selected, then the final timestep will be used in the clustering.

    If no PartIDs are selected, all parts will be used in the clustering.
  7. Click Request.
  8. Click Submit.

    The clustering report is generated. Depending on the size of the model and results, the report may take some time to generate.

    Once generated, the Dataset dialog will close, and the newly created cluster information will appear.

    Figure 3.


  9. Use the Visualize Clusters tools to interrogate clusters.
  10. Click Train Classifier to create a classifier based on the current clustering.
    A trained classifier is listed, along with the following information shown in the table:
    Accuracy with Solver
    Indication of the expected accuracy when running a solver-based optimization.
    Accuracy with Fit
    Indication of the expected accuracy when running a fit-based optimization.
    R-Squared Fit
    Representation of the quality of the fit.
  11. Right-click the classifier and select Create Constraint.
  12. Select a cluster to use as the constraint and enter a lower bound for the constraint.
    The subsequent optimization will attempt to satisfy the constraint such that the optimal result will conform to the selected cluster with probability greater than the constraint value.
  13. Click Optimize and choose the desired optimization type to run.
    Note: The optimization will use design variables and responses, including goals, from the original exploration the clusters were generated from. An objective is required to run the optimization.

    When the originating exploration type is a DOE, both fit-based and solver-based optimizations can be run. When the originating type is an optimization, only solver-based optimizations can be run.

Visualize Clusters

There are a number of tools and features to help with interrogation and visualization of cluster information.

  • Click to display the dendrogram for a given clustering.
    A dendrogram is a visual tool which shows the hierarchical relationship between clusters.
    Tip: Click Save to save an image of the dendrogram.
  • Click to display a silhouette plot for a given clustering.
    The silhouette plot provides an indication of the validity and consistency within the clustering and how well each run has been classified.
  • Select the Show Clusters check box to color the runs on an active scatter plot by cluster.
  • Right-click on a cluster and select Animate Selection.
    This loads an animation in the active HyperView client (if the active client is not a HyperView client, a new one is opened), overlaying the final deformed shape of each run in the selected cluster. This can be useful for visualizing the deformed shape of the given cluster as a whole.
  • Right-click on a cluster or in the white space of the Clusters column and select Animate all Clusters.
    This creates a new page and loads an animation for each cluster in a new HyperView client, overlaying the final deformed shape of each run in each cluster. This can be useful for visualizing the deformed shape of each cluster as a whole and comparing one cluster to another.
  • Select a cluster in the Clusters column, and the runs comprising that cluster are displayed in the Runs in Cluster column. Select one or more runs in the Runs in Cluster column, right-click, and select Overlay Selection.
    This will load an animation in the active HyperView client (if the active client is not a HyperView client, a new one is opened), overlaying the results, including all timesteps, of each of the selected runs.
  • Click Generate Classifier to create a classifier based on the current clustering.
    Classification will allow the clustering information to be used, where the classification assignment can serve as a constraint in an optimization. This feature will be enhanced and leveraged in a coming release.