Data-Driven Structural Topology Optimization Method Using Conditional Wasserstein Generative Adversarial Networks with Gradient Penalty  

作　　者：Qingrong Zeng Xiaochen Liu Xuefeng Zhu Xiangkui Zhang Ping Hu 

机构地区：[1]School of Automotive Engineering,Dalian University of Technology,Dalian,116024,China

出　　处：《Computer Modeling in Engineering & Sciences》2024年第12期2065-2085,共21页工程与科学中的计算机建模（英文）

基　　金：supported by the National Key Research and Development Projects (Grant Nos.2021YFB3300601,2021YFB3300603,2021YFB3300604);Fundamental Research Funds for the Central Universities (No.DUT22QN241).

摘　　要：Traditional topology optimization methods often suffer from the“dimension curse”problem,wherein the com-putation time increases exponentially with the degrees of freedom in the background grid.Overcoming this challenge,we introduce a real-time topology optimization approach leveraging Conditional Generative Adversarial Networks with Gradient Penalty(CGAN-GP).This innovative method allows for nearly instantaneous prediction of optimized structures.Given a specific boundary condition,the network can produce a unique optimized structure in a one-to-one manner.The process begins by establishing a dataset using simulation data generated through the Solid Isotropic Material with Penalization(SIMP)method.Subsequently,we design a conditional generative adversarial network and train it to generate optimized structures.To further enhance the quality of the optimized structures produced by CGAN-GP,we incorporate Pix2pixGAN.This augmentation results in sharper topologies,yielding structures with enhanced clarity,de-blurring,and edge smoothing.Our proposed method yields a significant reduction in computational time when compared to traditional topology optimization algorithms,all while maintaining an impressive accuracy rate of up to 85%,as demonstrated through numerical examples.

关 键 词：Real-time topology optimization conditional generative adversarial networks dimension curse CMES 2024 vol.141 no.3 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]