HS-6000: Approximation on Arm Model

In this tutorial, you will learn how to create approximations for the output responses using Sampling Fit.

Before you begin, complete HS-2000: DOE Method Comparison: Arm Model Study or import the HS-2000.hstx archive file, available at <hst.zip>/HS-6000/.

Sampling Fit is a combination of space-filling DOE and Fit in a single approach.

HS-2000: DOE Method Comparison: Arm Model Study demonstrated studies can be conducted effectively with six input variables rather than nine because the other input variables did not have a great influence on the output responses. This will save computational effort.

Perform Sampling Fit Approach

  1. From the Explorer, right-click and select Add from the context menu.
    The Add - Altair HyperStudy dialog opens.
  2. In the Add - Altair HyperStudy dialog, select Sampling Fit.
  3. Select any previous approach for Definition From and click OK.
  4. Modify input variables.
    1. Go to the Setup > Definition > Define Input Variables step.
    2. In the work area, Active column, clear the radius_1, radius_2, and radius_3 checkboxes.
      Figure 1.


  5. Go to the Specifications step.
  6. In the Specifications tab, verify Mels is the selected method.
  7. In the Fit Specifications tab, verify the Fit Type is set to FAST for all responses.
    Figure 2.


  8. Click Apply.
  9. Go to the Evaluate step and click Evaluate Tasks.
  10. Optional: Monitor the trend of cross-validation and perform what-if scenarios while tasks are being evaluated.
    1. Go to the R2Plot tab to monitor cross-validation value per iteration.
      Figure 3.


    2. Go to the Trade-off tab to review what-if scenarios.
      Note: If there is a data point of interest, add it to the queue for evaluation.
      Figure 4.


  11. Go to the Post Processing step.
    1. Go to the Diagnostics tab and click on each label to review the detailed diagnostics of the fit model for each response.
      Figure 5.


    2. Go to the Residuals tab to review the error between the original output response and the fit for each run of the input and the cross-validation matrix.
      Figure 6.