Predicting Transient Scalar Fields

The topic of AI-based transient predictions in the simulation community is a poular one. Transient simulations, atleast in CFD space, are often very expensive (computationally) and time consuming to complete. Ordinarily, the problems we seek to simulate are transient in nature, and thus more accurate to simulate them that way as well. That is one such appeal for the transient AI use case.

However, it is quite a moon shot to learn unsteady PDEs in a general enough manner that can tackle a large amount of flow simulations. Thankfully, historic data for current and past designs usually exists within company archives, and can be utilized for machine learning.

This study exemplifies that machine learning models can accurately predict scalar fields for transient simulations. The gif pictured herein compares Simcenter STAR-CCM+, as the ground truth, to the machine learning prediction. This case was nt included in training in any way.