Using ParaView Catalyst as a Bridge Between Simulation and AI in CFD

Catalyst is being used to improve the underlying turbulence models with AI/ML. This visualization shows two data extracts, wall distance, and Q-criterion contours colored by velocity magnitude were captured from the PHASTA CFD simulation code running on 256k MPI processes using Catalyst.

Traditional simulation frameworks—honed over decades—are powerful, precise, and deeply specialized. Their complexity and domain specificity, however, make them difficult to modify, and many researchers remain cautious about introducing AI/ML into these workflows. Concerns about the opacity of “black box” models and the lack of consistent, quantifiable benefits have limited the adoption of AI/ML in computational science, where transparency and reproducibility are critical.

To bridge this gap, a promising approach is emerging: targeted, in situ AI/ML integration that complements rather than replaces established simulation codes. Using ParaView Catalyst, we’re developing a modular machine learning toolbox that operates directly within simulation workflows. This approach avoids intrusive code modifications and eliminates the need for costly data movement, enabling researchers to explore tailored AI/ML solutions that enhance performance and insight—while maintaining full control over the scientific process.

You can find more information about this effort here.