Cross domain pretraining improves neural fluid simulations accuracy
Cross-Domain Pretraining for Steady-State Neural CFD Surrogates
Machine Learning
Summary
Simulating fluid flow is important but slow, and AI models called neural surrogates can make it faster. However, these models often don't work well on new shapes or conditions they weren't trained on. The authors show that training a model across many different fluid flow scenarios first helps it learn better. This makes the model much more accurate on new cases even with less retraining data. Their approach works regardless of the model type and benefits from bigger models and more varied training data.
What this means in practice
- •For engineering simulation teams: Train fluid simulation models that quickly adapt to new geometries and conditions with less data and higher accuracy.
- •For automotive design teams: Reduce computation costs when testing new car shapes by reusing pretrained fluid flow models across diverse aerodynamic scenarios.
Authors
Anthony Zhou, Amir Barati Farimani, Shirley Ho, Rudy Morel
Abstract
Neural surrogates for computational fluid dynamics (CFD) have the potential to greatly enhance engineering innovation through accelerating simulation. However, the primary limitation for neural surrogates is the lack of generalization to geometries and applications beyond the training set, which is significant given the diversity of engineering scenarios. Currently, this is addressed by generating a new dataset for a specific application; however, this requires running costly numerical solvers. In this work, we take a step toward addressing this by studying neural surrogates trained across different geometries, boundary conditions, and fidelities. We find that cross-domain pretraining improves zero- and few-shot performance on held-out datasets relative to both training from scratch and transferring from domain-specific experts. In particular, finetuning a pretrained, cross-domain model can achieve 2-3x lower errors at the same sample size and use 8x fewer samples to achieve the same error, compared to training from scratch. This benefit is architecture agnostic and improves with model size and pretraining dataset diversity. Furthermore, we study how and why cross-domain pretraining works in CFD surrogates, and find that simply pooling steady-state datasets is both sufficient and effective. Given the high cost of generating CFD data, leveraging existing datasets through cross-domain pretraining will likely be a valuable strategy as future surrogates expand to tackle new problems and use cases.