基于深度子领域适应卷积神经网络的结构损伤识别

PDF(2821 KB)

振动与冲击 ›› 2025, Vol. 44 ›› Issue (3) : 251-260.

故障诊断分析

基于深度子领域适应卷积神经网络的结构损伤识别

张健飞*，曹雨

作者信息 +

Structural damage recognition based on deep subdomain adaptation CNN

ZHANG Jianfei*, CAO Yu

Author information +

文章历史 +

摘要

针对卷积神经网络（convolutional neural networks, CNN）结构损伤识别模型泛化能力差的问题，提出了一种基于深度子领域适应卷积神经网络（deep subdomain adaptation convolutional neural networks,DSACNN）的结构损伤识别方法。以实际结构为目标域，以有限元模型为源域，根据损伤类别将源域和目标域划分成一系列子领域。在CNN中嵌入子领域适应模块，构建DSACNN模型，通过最小化源域上的损伤分类误差和领域之间的局部最大均值差异，对齐两个领域对应子领域的特征、建立特征与损伤类别之间的映射，从而将源域上的损伤识别能力迁移到目标域之上。模型的训练无需已知目标域样本的损伤标签，采用预训练全局领域适应提高其伪标签的准确率。实验结果表明：与全局领域适应模型相比，基于预训练全局领域适应的DSACNN模型在模拟目标域上准确率最大提高幅度达到21.8%，在实测目标域上提高了9.6%，具有更强的泛化能力。

Abstract

Aiming at the problem of poor generalization ability of the convolutional neural network (CNN) damage identification model, a structural damage identification method based on deep subdomain adaptation CNN (DSACNN) is proposed. The actual structure is taken as the target domain, the finite element model is taken as the source domain, and the source and target domains are divided into subdomains according to the damage classes.The DSACNN model is trained in an unsupervised manner to align the features of the subdomains and establish a mapping between the features and damages by minimizing the damage classification error on the source domain and the local maximum mean discrepancy between the domains. Thereby, the damage identification capability on the source domain is migrated over the target domain. The experimental results show that compared with the global domain adaptation model, the maximum improvement in accuracy of the DSACNN model based on pre-trained global domain adaptation reaches 21.8% on the simulated target domain and 9.6% on the measured target domain. It indicates that the model has stronger generalization ability.

导出引用

张健飞, 曹雨. 基于深度子领域适应卷积神经网络的结构损伤识别[J]. 振动与冲击, 2025, 44(3): 251-260

ZHANG Jianfei, CAO Yu. Structural damage recognition based on deep subdomain adaptation CNN[J]. Journal of Vibration and Shock, 2025, 44(3): 251-260

参考文献

[1] 朱宏平, 余璟, 张俊兵. 结构损伤动力检测与健康监测研究现状与展望[J]. 工程力学, 2011, 28(2):1-11, 17.
ZHU Hong-ping, YU Jing, ZHANG Jun-bing. A summary review and advantages of vibration-based damage identification methods in structural heath monitoring [J]. Engineering Mechanics, 2011, 28(2):1-11, 17.
[2] 闫桂荣, 段忠东, 欧进萍. 基于结构振动信息的损伤识别研究综述[J]. 地震工程与工程振动, 2007, 27(3): 95-103.
YAN Gui-rong, DUAN Zhong-dong, OU Jin-ping. Review on structural damage detection based on vibration data[J]. Journal of Earthquake Engineering and Engineering Vibration, 2007, 27(3): 95-103.
[3] 袁旭东, 周晶, 黄梅. 基于静力位移和频率的结构损伤识别神经网络方法[J]. 哈尔滨工业大学学报, 2005, 37(4): 488-490.
YUANG Xu-dong, ZHOU Jing, HUANG Mei. A method of structural damage identification using neural networks based on static displacements and natural frequencies[J]. Journal of Harbin Institute of Technology, 2005, 37(4): 488-490.
[4] 李兆,唐雪松,陈星烨. 基于曲率模态和神经网络的分步损伤识别法及其在桥梁结构中的应用[J]. 长沙理工大学学报(自然科学版)，2008, 5(2): 32-37
LI Zhao, TANG Xue-song, CHEN Xing-ye. Two-step damage identification method and its application based on curvature mode and neural network[J], Journal of Changsha University of Science and Technology (Natural Science), 2008, 5(2): 32-37.
[5] 周邵萍, 郝占峰, 韩红飞, 等. 基于应变模态差和神经网络的管道损伤识别[J]. 振动、测试与诊断, 2015, 35(2): 334-338.
ZHOU SHao-ping, HAO Zhan-feng, HAN Hong-fei, et al. Damage identification in straight pipeline using strain modal difference and neural network[J]. Journal of Vibration, measurement & Diagnosis, 2015, 35(2): 334-338.
[6] Avci O, Abdeljaber O, Kiranyaz S, et al. A review of vibration-based damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning applications[J]. Mechanical Systems and Signal Processing, 2021, 147: 107077.
[7] Khodabandehlou H, Pekcan G, Sami Fadali M. Vibration-based structural condition assessment using convolution neural networks [J].Structural Control & Health Monitoring, 2019, 26, e2308.
[8] Lin Y, Nie Z. Structural damage detection with automatic feature-extraction through deep learning [J]. Computer-Aided Civil and Infrastructure Engineering, 2017, 32(12): 1025-1046.
[9] 马亚飞, 李诚, 何羽, 等. 基于小波散射卷积神经网络的结构损伤识别[J]. 振动与冲击, 2023, 42(14): 138-146.
MA Ya-fei, LI Cheng, HE yu, et al. Structural damage identification based on the wavelet scattering convolution neural network[J]. Journal of Vibration and Shock, 2023, 42(14): 138-146.
[10] 李延强, 韩家浩. 基于格拉姆角场和卷积神经网络的斜拉索损伤识别研究[J]. 石家庄铁道大学学报(自然科学版), 2023, 36(4): 1-7.
LI Yan-qiang, HAN Jia-hao. Reaearch on cable damage identification based on Gramian angular field and convolutional neural network[J]. Journal of Shijiazhuang Tiedao University (Natural Science Edition), 2023, 36(4): 1-7.
[11] 李晶晶, 孟利超, 张可, 等. 领域自适应研究综述[J]. 计算机工程, 2021, 47(6): 1-13.
LI Jing-jing, MENG Li-chao, ZHANG Ke, et al. Review of studies on domain adaptation [J]. Computer Engineering, 2021, 47(6): 1-13.
[12] Wang M, Deng W. Deep Visual Domain Adaptation: A Survey[J]. Neurocomputing, 2018, 312: 135-153.
[13] Zhu Y, Zhuang F, Wang J, et al. Deep Subdomain Adaptation Network for Image Classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(4): 1713-1722.
[14] Hoffman J，Guadarrama S，Tzeng E S，et al. LSDA：large scale detection through adaptation [J]. Advances in Neural Information Processing Systems, 2014, 27: 3536-3544.
[15] Ghifary M, Kleijn W B, Zhang M. Domain adaptive neural networks for object recognition[[C]//Proceddings of Pacific Rim International Conference on Artificial Intelligence. Queensland: Springer, 2014.
[16] 康守强, 邹佳悦, 王玉静, 等. 基于无监督特征对齐的变负载下滚动轴承故障诊断方法[J]. 中国电机工程学报, 2020, 40(1): 274-281.
KANG Shou-qiang, ZOU Jia-yue, WANG Yu-jing, et al. Fault Diagnosis Method of a Rolling Bearing Under Varying Loads Based on Unsupervised Feature Alignment[J]. Proceedings of the CSEE, 2020, 40(1): 274-281.
[17] 董绍江, 朱朋, 朱孙科, 等. 基于仿真数据驱动和领域自适应的滚动轴承故障诊断方法[J]. 中国机械工程, 2023, 34(6): 694-702.
DONG Shao-jiang, ZHU Peng, ZHU Sun-ke, et al. Fault Diagnosis Method of Rolling Bearing Based on Simulation Data Drive and Domain Adaptation[J]. China Mechanical Engineering, 2023, 34(6): 694-702.
[18] Lin Y, Nie Z, Ma H. Dynamics-based cross-domain structural damage detection through deep transfer learning[J]. Computer-Aided Civil and Infrastructure Engineering, 2022, 37: 24-54.
[19] 王育鹏, 吕帅帅, 杨宇, 等. 基于域自适应的复合材料结构损伤识别方法[J]. 航空学报, 2022, 43(6): 526752.
WANG Yu-peng, LV Shuai-shai, YANG Yu, et al. Damage recognition of composite structures based on domain adaptive model[J]. Acta Aeronautica et Astronautica Sinica, 2022, 43(6): 526752.
[20] Li Z, He W, Ren W, et al. Damage detection of bridges subjected to moving load based on domain-adversarial neural network considering measurement and model error[J]. Engineering Structures，2023，293: 116601.
[21] Zhu Y, Zhuang F, Wang J, et al. Deep Subdomain Adaptation Network for Image Classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(4): 1713 - 1722.
[22] 宋向金, 孙文举, 刘国海, 等. 深度子领域自适应网络电机滚动轴承跨工况故障诊断[J].电工技术学报, 2024,39(1):182-193.
Song Xiang-jin, Sun Wen-ju, Liu Guo-hai, el al. Across Working Conditions Fault Diagnosis for MotorRolling Bearing Based on Deep Subdomain Adaption Network[J]. Transactions of China Electrotechnical Society, 2024,39(1):182-193.
[23] Pan S J, Tsang I W, Kwok J T, et al. Domain adaptation via transfer component analysis[J]. IEEE Transactions on Neural Networks, 2011, 22(2): 199–210.
[24] Der Maaten L V, Hinton G E. Visualizing Data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9: 2579-2605.