Efficient, high-resolution topology optimization method based on convolutional neural networks

Abstract Topology optimization is a pioneer design method that can provide various candidates with high mechanical properties. However, resolution desired for optimum structures, but it normally leads to computationally intractable puzzle, especially the solid isotropic material penalization (SIMP) method. In this study, an efficient, high-resolution topology developed based on superresolution convolutional neural network (SRCNN) technique in framework of SIMP. SRCNN involves four processes, namely, refinement, path extraction and representation, nonlinear mapping, image reconstruction. High computational efficiency achieved pooling strategy balance number finite element analyses output mesh process. A combined treatment uses 2D built as another speed-up reduce cost memory requirements 3D problems. Typical examples show using demonstrates excellent applicability when used problems arbitrary boundary conditions, any domain shape, varied load.