MeshMask: Physics-Based Simulations with Masked Graph Neural Networks

TL;DR

MeshMask introduces a masked pre-training approach for graph neural networks in CFD, significantly improving accuracy and efficiency across diverse fluid simulation datasets, including complex 3D aneurysm models.

Contribution

It presents a novel masking strategy combined with an asymmetric encoder-decoder architecture for GNNs applied to CFD, achieving state-of-the-art results and enabling effective multi-dataset pre-training.

Findings

Up to 60% improvement in long-term prediction accuracy.

Effective pre-training on multiple CFD datasets reduces data and time requirements.

State-of-the-art performance on seven CFD datasets, including complex 3D aneurysm simulations.

Abstract

We introduce a novel masked pre-training technique for graph neural networks (GNNs) applied to computational fluid dynamics (CFD) problems. By randomly masking up to 40\% of input mesh nodes during pre-training, we force the model to learn robust representations of complex fluid dynamics. We pair this masking strategy with an asymmetric encoder-decoder architecture and gated multi-layer perceptrons to further enhance performance. The proposed method achieves state-of-the-art results on seven CFD datasets, including a new challenging dataset of 3D intracranial aneurysm simulations with over 250,000 nodes per mesh. Moreover, it significantly improves model performance and training efficiency across such diverse range of fluid simulation tasks. We demonstrate improvements of up to 60\% in long-term prediction accuracy compared to previous best models, while maintaining similar computational costs. Notably, our approach enables effective pre-training on multiple datasets simultaneously, significantly reducing the time and data required to achieve high performance on new tasks. Through extensive ablation studies, we provide insights into the optimal masking ratio, architectural choices, and training strategies.