Create and Manage Datasets

Datasets are collections of one or more simulation results used for training or scoring a model.

Note: Dataset creation times vary because each .h3d file must be processed separately.
The following files are supported for physicsAI:
Table 1. List of Supported Files
Supported Solvers File Format
Abaqus .odb
AcuSolve .ensight

.h3d

ANSYS .rst

.rth

CCM+ .ensight
Custom/user-generated .ensight

.unv

.h3d

Fluent .ensight
LS-DYNA .d3plot

.d3eigv

Marc .t16
Nastran .op2

.h5

.xdb

PAM-CRASH .dsy
OptiStruct .h3d

.op2

Radioss .h3d

.anim

ultraFluidX .ensight
  1. Create dataset.
    1. From the PhysicsAI ribbon, select the Create Dataset tool.
      Figure 1.


      The Create Dataset dialog opens.
    2. Enter a name for the dataset.
      Important: A dataset name should only consist of letters, numbers, and underscores.
    3. For Directory, click Open and navigate to the project folder.
    4. Select and transfer all of the necessary result files.
      These can be results files with field results, such as an .h3d file. For use cases where the only output is vector data, the geometry information can be supplied in mesh file, for example a .fem file.
      Figure 2.


      Note: If the solid elements exist in the CAE data, selecting Extract solid faces means the prediction will be made using faces from the outer surfaces, and results will only be visible on the outer surface. The internal 3D elements do not contribute. Extract solid faces is selected by default because it ensures faster prediction time.

      If Extract solid faces is not selected, predictions will be made for (and using) all nodes/elements throughout the interior of the solid body as well.

      Figure 3. With Extract Solid Faces Selected


      Figure 4. Without Extract Solid Faces Selected


      Select Extract Simulation Properties to automatically extract the thicknesses and material IDs if there are solver decks available along with the result files. These decks should have a matching name with the result files. For example, A.h3d and A.fem, B.h3d and B.fem and so on. It is recommended that Extract Simulation Properties be selected only if the thicknesses and/or material IDs are changing between the files as additional processing time is needed. For more details on the supported file formats for thickness extraction, please refer to the Frequently Asked Questions.

    5. Click OK.
  2. Manage Datasets.
    1. From the PhysicsAI ribbon, select the Manage Datasets tool.
      Figure 5.


      The Datasets dialog opens.
    2. Create, import, or remove datasets.
    3. Optional: Select a data file and click Display File to preview the selected file.
      Figure 6.


      Note: Outliers in the datasets can be detected based on one or more responses. A Z-score distribution is used to highlight the points which fall in 3 sigma tails.
      Figure 7.


    4. Click OK.

CAE Data Compatibility

Learn more about the data compatibility of PhysicsAI.

File Types

PhysicsAI is compatible with any CAE file format which can be read by HyperView, including formats like .h3d, as well as several 3rd party CAE formats.

See the full list of formats here.

Element Types

PhysicsAI supports all readable CAE element types including 0D, 1D, 2D and 3D elements, and both first and second order elements.

Polyhedral elements are not supported, so they are not compatible with physicsAI.

Static/Transient

Both static and transient simulations are supported.

If transient, all samples must have the same number of time steps.

Physics

Nearly all types of physics simulation are supported, including structural, CFD, thermal, manufacturing, and so on.

PhysicsAI expects that results are defined on all nodes/elements. If the result is undefined for some parts (for example, “NA” or gray color in HyperView), they should first be removed using hvtrans before creating a PhysicsAI dataset. One exception is rigid elements, which are not expected to have results.

Erosion in explicit analysis is not currently supported.

Tensor Results

For tensor results like stress and strain, PhysicsAI predicts only the von Mises component. If another component is desired, like Tresca stress, a scalar field can be created from this component as a preprocessing step (outside of PhysicsAI).

Layered Results

When shell elements have multiple layers, PhysicsAI predicts the maximum value across all layers.