SuperMeshing: A New Deep Learning Architecture for Increasing the Mesh Density of Physical Fields in Metal Forming Numerical Simulation

Abstract In metal forming physical field analysis, finite element method (FEM) is a crucial tool, in which the mesh-density has significant impact on results. High mesh density usually contributes authentic to an increase accuracy of simulation results but costs more computing resources. To eliminate this drawback, we propose data-driven boosting model named SuperMeshingNet that uses low as inputs, acquire high-density with 2D structured grids instantaneously, shortening time and cost automatically. Moreover, Res-UNet architecture attention mechanism are utilized, enhancing performance SuperMeshingNet. Compared baseline applied linear interpolation method, achieves prominent reduction mean squared error (MSE) absolute (MAE) test data. The well-trained can successfully show improved than models multiple scaled mesh-density, including 2 ×, 4 8 ×. Enhanced by broaden scaling high precision output, FEM be accelerated seldom computational little sacrificed.