SuperMeshing: A Deep Learning Method for Boosting Mesh Density in Numerical Computation Within 2D Domain
Abstract Due to the limit of mesh density, the improvement of the spatial resolution of numerical computation always leads to a decrease in computing efficiency. Aiming at this inability of numerical computation, we propose a novel method for boosting the mesh density in finite element method (FEM) within 2D domain. Based on the von Mises stress fields of 2D plane-strain problems computed by the FEM, this method utilizes a deep neural network named SuperMeshingNet to learn a non-linear mapping from low mesh-density to high mesh-density in stress fields, and realizes the improvement of numerical computation accuracy and efficiency simultaneously. We adopt residual dense blocks into our mesh-density boost model – SuperMeshingNet to extract abundant local features and enhance the prediction capacity. The results indicate that SuperMeshingNet is able to effectively increase the spatial resolution of the von Mises stress fields under the multiple scaling factors: 2X,4X,and8X. Compared with the targets, the relative error of SuperMeshingNet is 2.44%, which shows better performance than the interpolation methods. Besides, SuperMeshingNet reveals an astonishing strength in predicting the maximum stress value. We publicly share our work with full detail of implementation at https://github.com/zhenguonie/2021_SuperMeshing_2D_Plane_Strain.