深度学习赋能结构拓扑优化设计方法研究  

Research on structure topology optimization design empowered by deep learning method

作　　者：陈小前 张泽雨 李昱 姚雯 周炜恩 CHEN Xiaoqian;ZHANG Zeyu;LI Yu;YAO Wen;ZHOU Weien(Chinese Academy of Military Science,Beijing 100195,China;Defense Innovation Institute,Chinese Academy of Military Science,Beijing 100071,China;College of Aerospace Science and Engineering,National University of Defense Technology,Changsha 410073,China;Intelligent Game and Decision Laboratory,Beijing 100071,China)

机构地区：[1]军事科学院,北京100195 [2]军事科学院国防科技创新研究院,北京100071 [3]国防科技大学空天科学学院,长沙410073 [4]智能博弈与决策实验室,北京100071

出　　处：《力学进展》2024年第2期213-258,共46页Advances in Mechanics

基　　金：国家自然科学基金重大项目92371206;湖南省研究生创新项目CX20220059.

摘　　要：本文综合论述了近年来结构拓扑优化领域与深度学习技术交叉融合发展的相关研究进展.围绕结构拓扑优化设计的核心方法与关键环节,从深度学习赋能的角度系统性梳理了两大类赋能方法.研究指出,基于深度学习技术的结构优化设计全局代理模型构建方法作为一种直接映射式结构设计方法,因其简单而典型的设计思想目前已被广泛研究,然而全局代理模型在计算性和泛化性上的局限与不足也尤为明显;融合深度学习技术的结构优化设计局部子环节加速与替代方法是一种更加灵活与多样的局部赋能形式,具有较好的普适性和独特的优越性.文章对智能赋能结构优化未来的发展进行了展望,研究重点在于深度学习与结构设计的有机结合方式,以及数据和知识的混合驱动设计范式.This article comprehensively discusses the relevant research progress in the field of structural topology optimization and the cross-integration development of deep learning technology in recent years.Focusing on the core methods and key modules of structural topology optimization design,two major types of empowerment methods are systematically sorted out from the perspective of deep learning empowerment.The study points out that the global surrogate model construction method for structural optimization design based on deep learning technology,as a direct mapping structural design method,has been widely studied because of its simple and typical design ideas.However,the global surrogate model has limitations in computation and generalization.The limitations and deficiencies in performance are also particularly obvious.The structural optimization design method with local sub-link acceleration and replacement integrated with deep learning technology is a more flexible and diverse form of local empowerment,with good universality and unique advantages.The article looks forward to the future development of intelligently empowered structural optimization.Further research work would focus on the organic combination of deep learning and structural design,as well as the co-driven design paradigm of data and knowledge.

关 键 词：拓扑优化 深度学习 人工神经网络 代理模型 

分 类 号：O342[理学—固体力学]