A Machine Learning approach to enhance the SUPG stabilization method for advection-dominated differential problems

TL;DR

This paper introduces a machine learning approach using neural networks to optimize the stabilization parameter in the SUPG method, improving accuracy for advection-dominated differential problems in finite element analysis.

Contribution

The paper presents a novel machine learning framework to predict optimal stabilization parameters for SUPG, enhancing solution accuracy over traditional fixed-parameter methods.

Findings

ANN-based parameter prediction improves solution accuracy

Method effective in 1D and 2D advection problems

Outperforms conventional stabilization parameter choices

Abstract

We propose using machine learning and artificial neural networks (ANNs) to enhance residual-based stabilization methods for advection-dominated differential problems. Specifically, in the context of the finite element method, we consider the streamline upwind Petrov-Galerkin (SUPG) stabilization method and we employ ANNs to optimally choose the stabilization parameter on which the method relies. We generate our dataset by solving optimization problems to find the optimal stabilization parameters that minimize the distances among the numerical and the exact solutions for different data of differential problem and the numerical settings of the finite element method, e.g., mesh size and polynomial degree. The dataset generated is used to train the ANN, and we used the latter ``online'' to predict the optimal stabilization parameter to be used in the SUPG method for any given numerical setting and problem data. We show, by means of 1D and 2D numerical tests for the advection-dominated differential problem, that our ANN approach yields more accurate solution than using the conventional stabilization parameter for the SUPG method.