A sneak peek at physicsAI from Altair
Altair EmployeeToday I’d like to give a sneak peek at Altair physicsAI: a new tool for making fast physics predictions. physicsAI learns from your historical simulation data, extracting the relationships between shape and engineering performance. Once trained, physicsAI models output fully animated contours at speeds 10x-100x faster than a solver.
Unlike other machine learning tools which require a design of experiments (DOE) and design variables, physicsAI leverages state-of-the-art geometric deep learning to operate directly on meshes, or even CAD. The result is an accelerated design cycle and better design decisions.
In this example, a physicsAI model is used to predict the surface pressures of an HVAC design. Simply import your parasolid, press “predict”, and the pressure prediction is displayed in 3 seconds.
To train the physicsAI model, we collected some historical HVAC designs from previous makes and models. Using these previously simulated designs, we trained a physicsAI model to predict the performance of new ones. Once the model has been trained, it can be used for near real-time performance predictions. Simply import meshes or CAD and click "predict" to get fast design feedback. When you're happy with the design, validate with the solver.
physicsAI is driven by data and can thus work on a variety of physics. Here’s an example of predicting the crash behavior of a head impacting the hood of a car. Note that physicsAI predicts the full animated contour.
If your design application could benefit from fast physics predictions, stay tuned; physicsAI will be released for the first time later this year. We hope you will give it a try and share your feedback.