基于数据驱动的声振耦合系统微结构拓扑优化方法研究  

Research on data-driven topology optimization method for microstructures of acoustic-structure interaction system

作　　者：徐炅阳 余秋子 张佳龙 操小龙[2] 陈海波[1] XU Jiongyang;YU Qiuzi;ZHANG Jialong;CAO Xiaolong;CHEN Haibo(CAS Key Laboratory of Mechanical Behavior and Design of Materials,Department of Modern Mechanics,University of Science and Technology of China,Hefei 230027,China;Beijing Research Institute of Mechanical and Electrical Engineering,Beijing 100074,China)

机构地区：[1]中国科学技术大学近代力学系中国科学院材料力学行为与设计重点实验室,合肥230027 [2]北京机电工程研究所,北京100074

出　　处：《振动与冲击》2025年第8期133-142,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金项目(12172350)。

摘　　要：传统声振耦合系统微结构拓扑优化依赖于有限元、边界元等数值方法,存在计算成本高、耗时长的问题。为此,提出一种基于数据驱动的声振耦合系统微结构拓扑优化方法。该方法的核心是以微结构密度分布为特征,以系统响应和灵敏度值为标签构建数据集分别训练人工神经网络,建立微结构材料分布与响应及灵敏度之间的非线性映射关系。数值测试表明,所提方法通过神经网络预测的方式替代传统的响应分析和灵敏度计算,在保证计算精度的同时减少计算成本,最终显著提升声振耦合系统微结构拓扑优化计算效率。同时该方法具有较好的泛化能力,可以针对不同的初始结构快速给出收敛的优化构型,这对拓扑优化设计中的全局优化解的搜寻具有重要意义。Traditional microstructural topology optimization for acoustic-structure interaction system relies on numerical methods such as the finite element method and boundary element method,which has the problems of high computational cost and long time consumption.In this paper,a data-driven topology optimization method for the microstructures of acoustic-structure interaction system was proposed to overcome the problems.In this method,the density distribution of microstructures was used as the feature,and the system response and sensitivity values were used as labels to construct a dataset to train the artificial neural network respectively,and the nonlinear mapping relationships between the distribution of microstructures and the response/sensitivity values were established.The response and sensitivity calculation part were replaced by neural network prediction in the optimization process,so as to reduce the computational cost.Numerical tests indicate that the proposed method can significantly improve the computational efficiency while ensuring computational accuracy.At the same time,the proposed method has good generalization ability and can quickly converge to optimization configurations for different initial structures.This work is significant in searching for the global topology optimization design of the microstructures in acoustic-structure interaction system.

关 键 词：声振耦合 拓扑优化 微结构 数据驱动 神经网络 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3