Graph neural nets predict key flow behaviors with coarser data
Predicting Sparse Observables in Parameterized Unsteady Flows Using Coarse-Mesh Graph Neural Networks
Machine Learning
Summary
Predicting certain important parts of complex fluid flows often requires heavy computations of the entire flow field. The authors present a way to predict just the important details directly on simplified, coarser models of the flow using graph neural networks. Their approach works well across different flow conditions and uses less computing power than detailed methods. They also improve accuracy by focusing training data on challenging cases and by making the model resilient to noise over longer predictions.
What this means in practice
- •For computational fluid dynamics engineers: Predict key flow metrics efficiently under varied conditions without reconstructing full high-resolution flow fields.
- •For process control teams: Estimate flow states from a few sensor measurements to aid monitoring and control of fluid systems.
Authors
Brener Ramos, William R. Wolf, Tarcísio C. D. Oliveira, Scott T. M. Dawson
Abstract
Full-order surrogates for high-dimensional nonlinear flows are often unnecessary when applications require only selected degrees of freedom. We propose reduced-order models that bypass full-state reconstruction by deploying graph neural networks (GNNs) directly on coarse meshes. Applied to parameterized 2D/3D confined cylinder flows and an unevenly heated cavity over Reynolds/Rayleigh-number ranges, the models generalize to unseen conditions and accurately predict sparse probe locations and integrated quantities. Coarse-mesh GNNs match or outperform finer-mesh surrogates while reducing training time and memory, even for 3D chaotic dynamics. A dynamics-informed adaptive sampling strategy guided by frequency-spectrum variations reduces maximum errors by an order of magnitude versus uniform sampling at negligible cost. Coarse-mesh training also makes noise-augmented multi-step unrolling practical, improving long-horizon autoregressive predictions. The approach outperforms parametric DMD, sparse MLPs, U-Nets, and Fourier neural operators at lower cost, and enables system identification and state estimation from sparse observations. These results show that coarse-mesh representations, adaptive sampling, and noise-augmented unrolled training form an efficient framework for reduced-order modeling of parameterized unsteady flows without solver-resolution reconstruction.