基于卷积神经网络和自编码器的结构损伤预警方法

骆勇鹏; 何敖力; 余焜明; 张锴; 廖飞宇

doi:10.13206/j.gjgS25092901

基于卷积神经网络和自编码器的结构损伤预警方法

doi: 10.13206/j.gjgS25092901

1. 福建农林大学交通与土木工程学院, 福州 350108;
2. 南平市公路事业发展中心, 福建南平 353000;
3. 福建省高速公路达通检测有限公司, 福州 350108

基金项目:

国家自然科学基金项目（51808122）；福建省自然科学基金面上项目（2024J01424）；福建农林大学科技创新专项基金项目（KFB23168）；中央引导地方科技发展项目（2022L3007）；福建省交通运输科技项目（YB202412）；福建省住房与城乡建设科学技术计划项目（2025-K-23）。

详细信息

作者简介:
骆勇鹏，副教授，主要从事基础设施智能检测和监测研究。

通讯作者:
廖飞宇，教授，主要从事组合结构、基础设施智慧运维等方面的研究工作，feiyu.liao@fafu.edu.cn。

计量
- 文章访问数: 1
- HTML全文浏览量: 0
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2025-09-29
- 网络出版日期: 2026-07-06

A Structural Damage Early Warning Method Based on Convolutional Neural Networks and Autoencoders

1. School of Transportation and Civil Engineering, Fujian Agriculture and Forestry University, Fuzhou 350108, China;
2. Nanping Highway Business Development Center, Nanping 353000, China;
3. Fujian Provincial Expressway Datong Inspection Co., Ltd., Fuzhou 350108, China

摘要

摘要: 足够丰富且多样的样本数据是进行结构损伤智能诊断的前提。由于实际工程应用中缺乏损伤状态数据，限制了上述方法的应用。无监督深度学习算法可在一定程度上解决上述问题。为此，将自编码器引入结构损伤预警中，提出了一种基于卷积神经网络与自编码器结合的结构损伤预警方法，包括数据预处理、特征提取和损伤预警等几个部分。首先构建了基于卷积神经网络的初级特征提取模块对采集的加速度时程信号进行数据预处理和浅层特征提取。其次，基于自编码器模型构建了损伤预警模型。该模型采用重构误差作为损伤预警指标，并基于健康样本重构误差的均值和标准差确定了动态阈值。当重构误差超过预设阈值时，表明结构可能出现损伤。采用卡塔尔看台数据集和室内损伤试验验证所提模型的可行性及可靠性。结果表明，在20 dB信噪比条件下，所提模型的预警准确率均超过98%；在5 dB信噪比条件下，损伤识别准确率均保持在93%以上，具备较好的抗噪性能。
- 结构健康监测 /
- 损伤识别 /
- 深度学习 /
- 卷积神经网络 /
- 自编码器
Abstract: Sufficiently rich and diverse sample data are essential prerequisites for intelligent structural damage diagnosis. However， the scarcity of damage-state data in practical engineering applications limits the implementation of such methods. Unsupervised deep learning algorithms can partially address this challenge， and an autoencoder-based framework for structural damage early warning combining convolutional neural networks （CNNs） and autoencoders （AEs） was therefore proposed. The proposed framework comprises three modules： data preprocessing， feature extraction， and damage early warning. The feature extraction module uses a CNN to derive discriminative features from healthy-state samples， which serve as input to the damage early warning module. This module adopts an autoencoder model trained exclusively on healthy-state data. Structural damage is detected by monitoring the reconstruction error of input data. Damage is identified when this error exceeds a predefined threshold， which is determined through statistical analysis of the reconstruction error distribution from the healthy-state dataset. The feasibility and reliability of the method were validated using the Qatar Stadium benchmark dataset and controlled indoor damage tests. Experimental results demonstrated that the framework achieved a recognition accuracy exceeding 98% at a 20 dB signal-to-noise ratio， a damage recognition accuracy exceeding 93% even at a 5 dB signal-to-noise ratio， demonstrating its excellent noise immunity.
- structural health monitoring /
- damage identification /
- deep learning /
- convolutional neural network /
- autoencoder

HTML全文

参考文献(28)

[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.
[2]	Domaneschi M, Martinelli L, Cucuzza R, et al. Structural control and health monitoring contributions to service-life extension of Bridges[J]. Proceedings in Civil Engineering, 2023, 6（5）:741-745.
[3]	李宏男, 李东升. 土木工程结构安全性评估、健康监测及诊断述评[J]. 地震工程与工程振动, 2002, 22（3）:82-90.
[4]	Alkayem N F, Cao M, Zhang Y, et al. Structural damage detection using finite element model updating with evolutionary algorithms:a survey[J]. Neural Computing and Applications, 2018, 30:389-411.
[5]	Chellini G, De Roeck G, Nardini L, et al. Damage analysis of a steel-concrete composite frame by finite element model updating[J]. Journal of Constructional Steel Research, 2010, 66（3）:398-411.
[6]	Lee J J, Lee J W, Yi J H, et al. Neural networks-based damage detection for bridges considering errors in baseline finite element models[J]. Journal of Sound and Vibration, 2005, 280（3/4/5）:555-578.
[7]	Salehi M, Ziaei-Rad S, Ghayour M, et al. A frequency response based structural damage localization method using independent component analysis[J]. Journal of Mechanical Science and Technology, 2013, 27（3）:609-619.
[8]	Fallahian M, Ahmadi E, Khoshnoudian F. A structural damage detection algorithm based on discrete wavelet transform and ensemble pattern recognition models[J]. Journal of Civil Structural Health Monitoring, 2022, 280:323-338.
[9]	Okafor A C, Dutta A. Structural damage detection in beams by wavelet transforms[J]. Smart Materials and Structures, 2000, 9（6）:906.
[10]	Han J, Ren W, Sun Z. Wavelet packet based damage identification of beam structures[J]. International Journal of Solids and Structures, 2005, 42（26）:6610-6627.
[11]	Esfandiari A, Nabiyan M S, Rofooei F R. Structural damage detection using principal component analysis of frequency response function data[J]. Structural Control and Health Monitoring, 2020, 27（7）:e2550.
[12]	Gui G, Pan H, Lin Z, et al. Data-driven support vector machine with optimization techniques for structural health monitoring and damage detection[J]. KSCE Journal of Civil Engineering, 2017, 21（2）:525-534.
[13]	Lecun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521:436-444.
[14]	Cheng F, Zhang H, Fan W, et al. Image Recognition Technology Based on Deep Learning[J]. Wireless Personal Communications, 2018, 102:1917-1933.
[15]	Otter D W, Medina J R, Kalita J K. A survey of the usages of deep learning for natural language processing[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 32（2）:604-624.
[16]	Mei Q, Gül M. A fixed-order time series model for damage detection and localization[J]. Journal of Civil Structural Health Monitoring, 2016, 6（5）:763-777.
[17]	杨铄, 许清风, 王卓琳. 基于卷积神经网络的结构损伤识别研究进展[J]. 建筑科学与工程学报, 2022, 39（4）:38-57.
[18]	郭旭, 骆勇鹏, 王林堃, 等. 基于CNN与DCGAN的结构振动监测传感器故障诊断及监测数据恢复[J]. 铁道科学与工程学报, 2022, 19（11）:3383-3395.
[19]	Sharma S, Sen S. One-dimensional convolutional neural network-based damage detection in structural joints[J]. Journal of Civil Structural Health Monitoring, 2020, 10（5）:1057-1072.
[20]	李雪松, 马宏伟, 林逸洲. 基于卷积神经网络的结构损伤识别[J]. 振动与冲击, 2019, 38（1）:159-167.
[21]	骆勇鹏, 王林堃, 廖飞宇, 等. 基于一维卷积神经网络的结构损伤识别[J]. 地震工程与工程振动, 2021, 41（4）:145-156.
[22]	崔广新, 李殿奎. 基于自编码算法的深度学习综述[J]. 计算机系统应用, 2018, 27（9）:47-51.
[23]	Pathirage C S N, Li J, Li L, et al. Structural damage identification based on autoencoder neural networks and deep learning[J]. Engineering Structures, 2018, 172:13-28.
[24]	Rastin Z, Ghodrati A G, Darvishan E. Unsupervised structural damage detection technique based on a deep convolutional autoencoder[J]. Shock and Vibration, 2021, 2021（1）:6658575.
[25]	Shang Z, Sun L, Xia Y, et al. Vibration-based damage detection for bridges by deep convolutional denoising autoencoder[J]. Structural Health Monitoring, 2020, 20（2）:1880-1903.
[26]	Rumelhart D E, Hinton G E, Williams R J. Learning representations by back-propagating errors[J]. Nature, 1986, 323:533-536.
[27]	Avci O, Abdeljaber O, Kiranyaz S, et al. Wireless and real-time structural damage detection:a novel decentralized method for wireless sensor networks[J]. Journal of Sound and Vibration, 2018, 424:158-172.
[28]	李行, 骆勇鹏, 郭旭, 等. 强噪声小样本条件下基于图卷积神经网络的结构损伤识别[J]. 地震工程与工程振动, 2024, 44（3）:52-60.