Simulation Data Sanity Checks and Requirements

The Ansys SimAI Pro application model predictions can only be as accurate as the data provided. An AI model is generally only as good as the data it is trained on. Inaccurate training data inevitably leads to unreliable or flawed outputs.

Perform the following sanity checks to your simulation data to ensure an optimal usage of the application’s prediction capabilities:

• Data accuracy: Simulations that you import to the Ansys SimAI Pro application must be converged. Monitored variables (for example, physical quantities of interest such as lift, drag, pressure or velocity) are stabilized to consistent values.

• Normal orientation: Normals should be oriented consistently, either all towards the exterior, or all towards the interior. This consistency should be kept for both AI-model creation and prediction.

  Tip

  To check if normals are correctly oriented after the data were uploaded, use the Model configuration page to compute a Global coefficient that integrates surface fields using normals. If its value is way off, there is a chance that normals are not well oriented, and that your model will not compute global coefficients correctly.

• Overlapping surface: Use geometric algorithms (bounding box checks or mesh intersection tests) to identify and correct potential overlapping regions between surfaces to ensure the data do not contain duplicate or conflicting information.

• Underflow and overflow values: Verify that numerical values of the simulation data (dataset) remain within a physically meaningful range.