Use PhysicsAI for Test Data

Tutorial Level: Advanced In this tutorial, you will train a physicsAI model without simulation results.

As inputs, physicsAI will use meshes (supplied using solver decks) and it can train to predict experimentally measured KPIs (supplied using .json files).

This tutorial uses 50 .fem files and corresponding .json files containing the KPIs. The KPIs can be experimentally derived and can be read from a .csv file. For additional details, refer to the Custom Outputs section. The original CAD models are also provided. These have been meshed to generate the .fem files.

Before you begin, copy the file(s) used in this tutorial to your working directory.
Warning: In this tutorial, you will train the model. If possible, it is recommended that you train the model using a GPU. Without a GPU, expect the training portion of this tutorial to take a long time. Learn more about training a model using a GPU here.

Create Project

  1. Open HyperMesh.
  2. From the menu bar, click View > Ribbons > PhysicsAI to open the PhysicsAI ribbon.
  3. From the PhysicsAI ribbon, select the Create Project tool.


    Figure 1.
    The Create Project dialog opens.
  4. For Project Name, enter Mesh_only.
  5. For Location, click Choose and select a save location for the project.
    Note: The save location for the project contains all files created by PhysicsAI, but the original files used for training do not need to reside in the project folder.
  6. Click OK.

Create a Dataset

  1. From the PhysicsAI ribbon, select the Create Dataset tool.


    Figure 2.
    The Create Dataset dialog opens.
  2. For Dataset Name, enter Ibeam_mesh.
  3. For File System, click and navigate to the Mesh_only_data folder.
  4. Change the filter from *.h3d to *.fem.
  5. Select and transfer all of the .fem files.
  6. Select Enable train test split check box and enter 90 for the Train %.
  7. Clear the Extract solid faces and Extract Simulation Propertiescheck boxes.


    Figure 3.
  8. Click OK.
    The dataset is extracted and the Datasets dialog opens.
  9. Click Close.

Train Model

  1. From the PhysicsAI ribbon, select the Train an ML Model tool.


    Figure 4.
    The Train Model dialog opens.
  2. Define the following details and click Train.
    1. For Model Name, enter Max_displacement.
    2. For Inputs, select cae.coord.
    3. For Vector, select max_displacement
    4. Under Method, select Transformer.
    5. Change the number of Epochs to 500.


    Figure 5.
    The Model Training dialog opens.
    Tip: Once the status changes to Running, you can click Show Log view the training logs.
  3. Once the training is complete, click Loss Curve to see if the model has converged after 500 curves.


    Figure 6.

Test Model

  1. From the PhysicsAI ribbon, select the Test ML Model tool.


    Figure 7.
    The Test Model dialog opens.
  2. In the Models area, select Max_displacement.
  3. In the Datasets area, select IBeam_mesh_test and click OK.
    The Model Testing dialog opens.
  4. In the Model Testing dialog, select a Run ID and click Display File to view the results in the modeling window.
    The closer the points are to the 45-degree diagonal, the better the fit. This models shows a good fit.


    Figure 8.
  5. Close the Model Testing dialog.