[Demo] Industrial AI: CFD PhysicsAI

From my experience, training industrial AI model is challenging but not the hardest part. Generating high-quality physics data (i.e. morphing, meshing, solver runs…) and trustworthy, explainability are. We’ve been exploring how advanced AI engineering can tackle those burdens. Please see our recent 3D ROM AI research and development (R&D) demo on CFD aerodynamic dataset. In particular,

• Explainable AI: which vehicle regions drive AI predictions.
• Uncertainty quantification: how much trust to place in each prediction
• Active learning: guide which physical parameters or 3D regions actually require new simulations

Hope you enjoy it 😊 . It is not just AI prediction, but a (simple) example of trustworthy AI and industrial engineering workflows.

Thank you the teams behind the DrivAerNet & DrivAerStar dataset (especially Mohamed Elrefaie for the exchange).

Note that this week Siemens has just also introduced an important industrial AI tool in Simcenter Star-CCM+, bringing AI directly into CFD workflow, from data generation to model validation and deployment inside design tools.