A new adaptive finite element analysis using deep learning

Abstract

The adaptive finite element method has been developed as a method for obtaining highly accurate solutions while suppressing the increase of analysis scale. However, it is not always simple and fast because it repeats adaptive remeshing and reanalysis. In this study, we propose a new adaptive method using deep learning, one of remarkably advancing machine learning algorithms. The proposed method only need to remesh once and can estimate accurate solutions by deep learning. In this paper, the fundamental formulation of the proposed method is explained in detail, and its basic performance is tested through sample two-dimensional stress analyses.