Eagle: Large-Scale Learning of Turbulent Fluid Dynamics with Mesh Transformers

TL;DR

EAGLE introduces a large-scale 2D fluid dynamics dataset and a novel mesh transformer model that efficiently predicts unsteady fluid behavior, outperforming existing methods and capturing complex airflow interactions in a single step.

Contribution

The paper presents a new large-scale fluid dynamics dataset and a mesh transformer model that improves prediction accuracy and efficiency over existing graph neural network approaches.

Findings

Transformer outperforms state-of-the-art GNNs on synthetic and real datasets.

Model learns to attend to airflow, capturing complex interactions.

Single-iteration learning of complex fluid dynamics.

Abstract

Estimating fluid dynamics is classically done through the simulation and integration of numerical models solving the Navier-Stokes equations, which is computationally complex and time-consuming even on high-end hardware. This is a notoriously hard problem to solve, which has recently been addressed with machine learning, in particular graph neural networks (GNN) and variants trained and evaluated on datasets of static objects in static scenes with fixed geometry. We attempt to go beyond existing work in complexity and introduce a new model, method and benchmark. We propose EAGLE, a large-scale dataset of 1.1 million 2D meshes resulting from simulations of unsteady fluid dynamics caused by a moving flow source interacting with nonlinear scene structure, comprised of 600 different scenes of three different types. To perform future forecasting of pressure and velocity on the challenging EAGLE dataset, we introduce a new mesh transformer. It leverages node clustering, graph pooling and global attention to learn long-range dependencies between spatially distant data points without needing a large number of iterations, as existing GNN methods do. We show that our transformer outperforms state-of-the-art performance on, both, existing synthetic and real datasets and on EAGLE. Finally, we highlight that our approach learns to attend to airflow, integrating complex information in a single iteration.