Physics foundation models improve control of fluid flows efficiently
Physics Foundation Models for Efficient Control of Fluid Flows
Machine Learning
Summary
Controlling how fluids move can be very complicated and needs smart methods. The authors study if a special type of computer model, trained to predict fluid patterns, can also be used to control fluids in real time. They show their model works better and uses less effort than a typical approach. This suggests that these physics-based models can learn useful ways to control fluids without extra training from scratch.
What this means in practice
- •For fluid dynamics engineers: Implement controllers for fluid systems that perform better and reduce control effort using pretrained physics models.
- •For simulation software developers: Integrate pretrained physics models to enable efficient fluid flow control directly within simulation environments.
Authors
Mahindra Rautela, Shaifalee Saxena, Alexander Scheinker
Abstract
We investigate whether a physics foundation model (PFM) pretrained for PDE forecasting can be repurposed for closed-loop fluid-flow control. We construct a PFM-based controller by attaching lightweight Gaussian actor and critic heads to frozen pretrained representations and train the policy using proximal policy optimization (PPO). We evaluate the approach on HydroGym's two-dimensional Kolmogorov-flow environment using vorticity observations at \(64\times64\) and \(128\times128\) resolutions and Reynolds numbers \(\mathrm{Re}=100\) and \(200\). Across these settings, we compare the frozen PFM controller against a baseline MLP policy trained with the same PPO framework. The PFM controller achieves consistently higher reward while requiring smaller executed action magnitudes over training, indicating improved control performance with reduced control effort. These results provide initial evidence that representations learned through self-supervised next-frame prediction can be transferred directly to closed-loop fluid flow control. Project page: https://mahindrautela.github.io/morph.