A topology description function‐enhanced neural network for topology optimization

• DOI: 10.1111/mice.12933
• Publication Date: 2022-10-25T04:54:36Z
• Authors: Wenliang Qian; Yang Xu; Hui Li
• Keywords: Topology optimization

Abstract Topology optimization aims to find an economic and efficient structure with a lighter overall weight. description functions (TDFs), which are explicit level‐set approach for topology optimization, can obtain the geometry function of topology. However, due original hard thresholding in conventional TDF, TDF is derivative‐free method that requires significant computational resources, creates barriers its widespread adoption among structural engineers. To fix this problem, novel TDF‐enhanced neural network (TDF‐NN) proposed, introduces sigmoid activation as soft allow gradient‐based approaches. Meanwhile, TDF‐NN uses single‐layer model weights converted TDF‐NN, be obtained directly by instead solving linear equations. The performance proposed validated through compliance minimization problems heat conduction it definitely also used other problems. cost method, effect number knots on diverse designs, application real‐life structures discussed. These results indicate universal method. Numerical examples show has higher efficiency than solid isotropic material penalization It speedup factor approximately two 10 times. Moreover, multiple competitive solutions generated changing knots, will architects select suitable design task.