Transformer model improves flow field predictions with physics constraints
Physics and Data Driven Transformer-Mamba Framework for Flow Field
Machine Learning
Summary
Simulating fluid flow accurately can be tricky and slow, especially with noisy data or changing conditions. The authors developed a new AI system called Transformer-Mamba that combines smart math tools and physics rules to better predict how fluids move. Their system cleans up noisy input, focuses on important features, and makes sure predictions follow the physics of fluid flow. Tests show it works well across different situations, making fluid simulations faster and more reliable.
What this means in practice
- •For engineering simulation teams: Improve the reliability and speed of fluid dynamics simulations under noisy or varying flow conditions by integrating physics-based AI predictions.
- •For weather modeling centers: Incorporate robust AI models for better airflow predictions in atmospheric simulations where data noise and variability hinder accuracy.
Authors
Zhuo Zhang, Shun Zou, Canqun Yang, Xi Yang
Abstract
While deep learning accelerates expensive partial differential equation solving in computational fluid dynamics (CFD), existing methods like PINNs and FNOs often struggle with generalization, noise robustness, and physical consistency. We introduce the Transformer-Mamba for Flow Field (TM4FF) framework, a physics-constrained operator learning model with three key innovations: a Residual Wavelet Mamba (RWM) layer for feature denoising, a Transformer-based attention mechanism for enhanced feature fusion, and a physics-informed loss using Fourier derivatives to enforce the Navier-Stokes equations. Experiments on four CFD datasets show TM4FF achieves high accuracy and robust generalization across varying flow conditions.