Towards Long-Term predictions of Turbulence using Neural Operators

TL;DR

This paper investigates neural operator models, especially Fourier Neural Operators and U-NET variants, for long-term turbulence prediction, highlighting the importance of regularization and model configuration for accuracy and stability.

Contribution

It introduces U-NET based neural operator models for turbulence prediction, demonstrating improved accuracy and stability over standard FNO, especially at higher Reynolds numbers.

Findings

U-FNET outperforms FNO in turbulence prediction accuracy.

Regularization terms improve model stability and accuracy.

U-NET based models are promising for complex turbulent flows.

Abstract

This paper explores Neural Operators to predict turbulent flows, focusing on the Fourier Neural Operator (FNO) model. It aims to develop reduced-order/surrogate models for turbulent flow simulations using Machine Learning. Different model configurations are analyzed, with U-NET structures (UNO and U-FNET) performing better than the standard FNO in accuracy and stability. U-FNET excels in predicting turbulence at higher Reynolds numbers. Regularization terms, like gradient and stability losses, are essential for stable and accurate predictions. The study emphasizes the need for improved metrics for deep learning models in fluid flow prediction. Further research should focus on models handling complex flows and practical benchmarking metrics.