基于生成对抗网络的跨分辨率拓扑优化方法的比较研究

摘要
图/表
参考文献(22)
相关文章 (15)

全文: PDF (2243 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要拓扑优化的迭代过程涉及大量有限元分析与灵敏度更新步骤，且随着划分网格数目的增加，其优化过程将耗费大量的计算成本。利用深度学习方法，通过建立低分辨率中间构型与高分辨率拓扑结构之间的映射关系，可实现结构的跨分辨率拓扑优化设计，从而大幅提高其计算效率。本文基于两种不同的生成对抗网络，构建了可实现跨分辨率预测的拓扑优化方法，并将其扩展至三维结构的优化预测。首先,以柔顺度最小化为目标函数，利用SIMP法生成不同载荷条件、初始位移和体积分数下的优化结构数据集；然后，分别利用Pix2pix和Esrgan网络解决其跨分辨率预测问题，其中对于Pix2pix网络，使用了残差模块代替其生成器内部的卷积模块，以增强网络对低层信息的复用；最后，通过二维和三维算例验证了所提方法的有效性，并与现有基于CGAN网络的方法进行了比较研究。结果表明：综合考虑计算精度和计算效率，基于Esrgan网络的方法表现更优，最适合应用于跨分辨率拓扑优化设计。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	朱赫鹏1
	2
	3
	丁喆1
	2
	3
	张严1
	2
	3
	李小白1
	2
	3

关键词 ：拓扑优化, 柔顺度, 跨分辨率预测, Pix2pix, Esrgan

Abstract：The iterative process of topology optimization involves a large number of finite element analyses and sensitivity update steps. As the number of mesh divisions increases, the optimization process consumes a significant amount of computational cost. By using deep learning methods and establishing a mapping relationship between low-resolution intermediate configurations and high-resolution topological structures, cross-resolution topology optimization design of structures can be achieved, thus greatly improving computational efficiency. This article constructs a topology optimization method capable of cross-resolution prediction based on two different generative adversarial networks, and extends it to the optimization prediction of three-dimensional structures. Firstly, with the minimization of compliance as the objective function, a dataset of optimized structures under different loading conditions, initial displacements, and volume fractions is generated using the Solid Isotropic Material with Penalization method. Then, the Pix2pix and Esrgan networks are used to solve their cross-resolution prediction problems, where for the Pix2pix network, residual modules are used to replace the convolution modules inside the generator to enhance the reuse of low-level information. Finally, the effectiveness of the proposed method is verified through two-dimensional and three-dimensional examples, and comparative studies are conducted with existing methods based on the CGAN network. The results show that considering both calculation accuracy and computational efficiency, the method based on the Esrgan network performs better and is most suitable for cross-resolution topology optimization design.

Key words： topology optimization compliance cross-resolution prediction Pix2pix Esrgan

收稿日期: 2023-06-01 出版日期: 2024-03-28

引用本文:

朱赫鹏1,2,3，丁喆1,2,3，张严1,2,3，李小白1,2,3. 基于生成对抗网络的跨分辨率拓扑优化方法的比较研究 [J]. 振动与冲击, 2024, 43(6): 132-140.
ZHU Hepeng1,2,3，DING Zhe1,2,3，ZHANG Yan1,2,3，LI Xiaobai1,2,3. Comparative study on cross-resolution topology optimization methods based on generative adversarial networks. JOURNAL OF VIBRATION AND SHOCK, 2024, 43(6): 132-140.

链接本文:

http://jvs.sjtu.edu.cn/CN/ 或 http://jvs.sjtu.edu.cn/CN/Y2024/V43/I6/132

[1]Chuang C H, Chen S, Yang R J, et al. Topology optimization with additive manufacturing consideration for vehicle load path development [J]. International Journal for Numerical Methods in Engineering, 2018, 113(8): 1434-1445. [2]黄垲轩, 丁喆, 张严, 等.高承载梯度分层点阵结构的拓扑优化设计方法[J].力学学报,2023,55(2):433-444. Huang Kai-xuan, Ding Zhe, Zhang Yan, Li Xiao-bai. Topological optimization design method of layer-wise graded lattice structures with high load-bearing. Chinese Journal of Theoretical and Applied Mechanics, 2023, 55(2): 433-444. [3]Tcherniak D. Topology optimization of resonating structures using SIMP method [J]. International Journal for Numerical Methods in Engineering, 2002, 54(11): 1605-1622. [4]Xie Y M, Steven G P. A simple evolutionary procedure for structural optimization [J]. Computers and Structures, 1993, 49(5): 885-896. [5]Allaire G, Jouve F, Toader AM. Structural optimization using sensitivity analysis and a level-set method [J]. Journal of Computational Physics, 2004, 194(1): 363-393. [6]Ye H L, Wang W W, Chen N, et al. Plate/shell structure topology optimization of orthotropic material for buckling problem based on independent continuous topological variables[J]. Acta Mechanica Sinica, 2017, 33(5): 899-911. [7]Li B, Huang C J, Li X, et al. Non-iterative structural topology optimization using deep learning [J]. Computer-Aided Design, 2019, 115(10):172-180. [8]Xue L, Liu J, Wen G, et al. Efficient,high-resolution topology optimization method based on convolutional neural networks [J]. Frontiers of Mechanical Engineering, 2021, 16(1): 80-96. [9]Woldseth R V, Aage N, Bærentzen J A. On the use of artificial neural networks in topology optimisation [J]. Structural and Multidisciplinary Optimization, 2022, 65(10): 294. [10]韩坤林, 汤宝平, 刘大洋, 等.基于注意力改进卷积网络的隧道风机预埋基础损伤识别[J].振动与冲击,2023,42(10):307-313. Han Kun-lin, Tang Bao-ping, Liu Da-yang, et al. Damage identification of embedded foundation of tunnel fan based on attention improved convolution network[J]. Journal of vibration and shock, 2023, 42(10): 307-313. [11]付君健, 张跃, 杜义贤, 等.周期性多孔结构特征值拓扑优化[J].振动与冲击,2022,41(03):73-81. Fu Jun-jian, Zhang Yue, Du Yi-xian, et al. Eigenvalue topology optimization of periodic cellular structures[J]. Journal of vibration and shock, 2022,41(03):73-81. [12]项程, 陈艾荣.基于深度学习的多材料结构拓扑优化方法[J].同济大学学报(自然科学版),2022,50(07):975-982. Xiang Cheng, Chen Ai-rong. Topology Optimization of Multi-material Structures Based on Deep Learning [J]. Journal of Tongji University (Natural Science), 2022,50(07):975-982. [13]阎军, 许琦, 张起, 等.人工智能在结构拓扑优化领域的现状与未来趋势[J].计算力学学报,2021,38(04):412-422. Yan Jun, Xu Qi, Zhang Qi, et al. Current and future trends of artificial intelligence in the field of structural topology optimization [J]. Chinese Journal of Computational Mechanics, 2021,38(04):412-422. [14]Sosnovik I, Oseledets I. Neural networks for topology optimization [J]. Russian Journal of Numerical Analysis and Mathematical Modelling, 2019, 34(4): 215-223. [15]Deng H, To A C. Topology optimization based on deep representation learning (DRL) for compliance and stress-constrained design [J]. Computational Mechanics, 2020, 66(2): 449-469. [16]Zhang Y, Peng B, Zhou X, et al. A deep Convolutional Neural Network for topology optimization with strong generalization ability [J]. arXiv preprint arXiv, 2019, 1901.07761 [17]Yu Y, Hur T, Jung J, et al. Deep learning for determining a near-optimal topological design without any iteration [J]. Structural and Multidisciplinary Optimization, 2019, 59(3): 787-799. [18]邢晓松, 郭伟.基于改进半监督生成对抗网络的少量标签轴承智能诊断方法[J].振动与冲击,2022,41(22):184-192. Xing Xiaosong, Guo Wei. Intelligent diagnosis method for bearings with few labelled samples based on an improved semi-supervised learning-based generative adversarial network[J]. Journal of vibration and shock, 2022, 41(22): 184-192. [19]Isola P, Zhu J Y, Zhou T, et al. Image-to-image translation with conditional adversarial networks [J]. Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, 1125-1134. [20]Wang X, Yu K, Wu S, et al. Esrgan: Enhanced super-resolution generative adversarial networks [J]. Proceedings of the European Conference on Computer Vision Workshop, 2018, 1-23. [21]叶红玲, 李继承, 魏南, 等.基于深度学习的跨分辨率结构拓扑优化设计方法 [J].计算力学学报,2021,38(4):430-436. Ye Hong-ling, Li Ji-cheng, Wei Nan, et al. Cross-resolution acceleration design for structural topology optimization based on deep learning [J]. Chinese Journal of Computational Mechanics, 2021,38(4):430-436. [22]Kang X, Liu L, Ma H. ESR-GAN: Environmental Signal Reconstruction Learning With Generative Adversarial Network [J]. IEEE Internet of Things Journal, 2020, 8(1): 636-646.