基于多尺度自适应融合图卷积网络的结构损伤识别

倪艳春; 靳启源; 胡睿

doi:10.3724/j.gyjzG25041503

基于多尺度自适应融合图卷积网络的结构损伤识别

doi: 10.3724/j.gyjzG25041503

同济大学土木工程学院, 上海 200092

基金项目:

国家自然科学基金项目（52378312）。

详细信息

作者简介:
倪艳春，副教授，主要从事桥梁结构健康监测，yanchunni@tongji.edu.cn。

通讯作者:
胡睿,博士研究生,主要从事桥梁结构健康监测,2310099@tongji.edu.cn。

计量
- 文章访问数: 7
- HTML全文浏览量: 1
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2025-04-15
- 网络出版日期: 2026-01-06

Structural Damage Identification Based on a Multi-Scale Adaptive Fusion Graph Convolutional Neural Network

School of Civil Engineering，Tongji University，Shanghai 200092，China

摘要

摘要: 目前，基于图卷积网络（Graph Convolutional Network， GCN）的损伤识别方法多依赖于静态图或单一尺度动态图结构对不同变量进行建模，难以同时提取局部与全局的特征。鉴于此，提出一种基于多尺度自适应融合图卷积网络（Multi-scale Adaptive Fusion Graph Convolutional Network，MAFGCN）的结构损伤识别方法。首先，基于本征正交分解（Proper Orthogonal Decomposition， POD）构建物理信息驱动的动态图结构，并基于传感器空间拓扑关系构建静态图结构。其次，使用简化图卷积提取多尺度动态特征图的特征，与静态图特征进行融合。最后，设计多尺度异构图融合框架，通过自适应权重机制实现多尺度邻接矩阵融合，使用图卷积在不同的尺度上提取高阶特征。通过简支梁数值模型和卡塔尔看台的实际检测对该方法进行了验证，并通过向原始振动数据中加入不同水平的噪声验证了方法的抗噪性。结果表明：该方法能够有效地捕捉振动数据的多尺度特征，在15％噪声水平下仍然能够达到90％以上的损伤识别准确率。
- 图卷积 /
- 神经网络 /
- 多尺度特征 /
- 本征正交分解 /
- 损伤识别
Abstract: Graph Convolutional Networks （GCNs） have attracted significant attention in structural damage identification due to their capacity to explicitly model relations among variables from different sensors. However， existing GCN-based methods predominantly rely on static graphs or single-scale dynamic graph structures， often failing to effectively extract both local and global features. To address this limitation， this study proposes a structural damage identification method using a Multi-scale Adaptive Fusion Graph Convolutional Neural Network （MAFGCN）. Firstly， a physics-informed dynamic graph structure was constructed based on Proper Orthogonal Decomposition （POD）， while a static graph structure was established using the spatial topology of the sensors. Secondly， simplified graph convolution was employed to extract multi-scale features from the dynamic graphs， which were then fused with the static graph features. Finally， a multi-scale heterogeneous graph fusion framework was designed， incorporating an adaptive weighting mechanism to fuse the multi-scale adjacency matrices， thereby enabling the extraction of high-order spatial features across different scales. The proposed method was validated through a numerical simulation of a simply-supported beam and an experiment on a Qatar Stadium stand. Additionally， its noise robustness was verified by introducing varying levels of noise into the original vibration data. The results demonstrated that the proposed method effectively extracted multi-scale features of the vibration data， achieving a damage identification accuracy exceeding 90% even under a 15% noise level.
- graph convolution /
- neural network /
- multi-scale features /
- Proper Orthogonal Decomposition /
- damage identification

HTML全文

参考文献(21)

[1]	孙利民，尚志强，夏烨. 大数据背景下的桥梁结构健康监测研究现状与展望[J]. 中国公路学报，2019，32（11）:1-20.
[2]	陈鑫婷，张军，鲁东明，等. 基于多头卷积自编码器的桥梁结构信号重构与损伤识别方法研究[J]. 振动与冲击，2025，44（6）:298-305.
[3]	NGUYEN Q T，NGUYEN T T，NGUYEN T P. Analysis of vibration characteristics of bridge spans based on the viscoelastic material model:investigating the relationship between material properties and dynamic parameters[J]. Structures，2025，75，108788.
[4]	杨银枪，康帅，王自法，等. 基于卷积自编码和相关函数的钢框架损伤识别研究[J]. 工业建筑，2024，54（11）:78-86.
[5]	ABDELJABER O，AVCI O，KIRANYAZ S，et al. Real-time vibration-based structural damage detection using one-dimensional convolutional neural networks[J]. Journal of Sound and Vibration，2017，388:154-170.
[6]	GIGLIONI V，VENANZI I，POGGIONI V，et al. Autoencoders for unsupervised real-time bridge health assessment[J]. Computer-Aided Civil and Infrastructure Engineering，2023，38（8）:959– 974.
[7]	TU J，LIU H，LI C，et al. Correntropy-induced weighted gan for bridge anomaly detection[J]. IEEE Sensors Journal，2024，24（4）:5335-5346.
[8]	SHANG Z，SUN L，XIA Y，et al. Vibration-based damage detection for bridges by deep convolutional denoising autoencoder[J]. Structural Health Monitoring，2021，20（4）:1880– 1903.
[9]	DENG A，HOOI B. Graph neural network-based anomaly detection in multivariate time series[C]// Proceedings of the AAAI Conference on Artificial Intelligence. 2021，35（5）:4027-4035.
[10]	LIAO J，LI J，CHEN Y，et al. DPDGAD:a dual-process dynamic graph-based anomaly detection for multivariate time series analysis in cyber-physical systems[J]. Advanced Engineering Informatics，2024，61，102547.
[11]	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.
[12]	ZHAN P，QIN X，ZHANG Q，et al. A novel structural damage detection method via multisensor spatial-temporal graph-based features and deep graph convolutional network[J]. IEEE Transactions on Instrumentation and Measurement，2023，72:1-14.
[13]	DANG V H，VU T C，NGUYEN B D，et al. Structural damage detection framework based on graph convolutional network directly using vibration data[J]. Structures，2022，38:40-51.
[14]	DANG V H，LE-NGUYEN K，NGUYEN T T. Semi-supervised vibration-based structural health monitoring via deep graph learning and contrastive learning[J]. Structures，2023，51:158-170.
[15]	史子凡，郭文华，郭柳君，等. 基于图卷积神经网络的结构损伤识别研究[J/OL]. 铁道科学与工程学报，2025[2025-06-14]. https://doi.org/10.19713/j.cnki.43-1423/u.T20250627.
[16]	KIM M，SONG J. Seismic damage identification by graph convolutional autoencoder using adjacency matrix based on structural modes[J]. Earthquake Engineering and Structural Dynamics，2024，53（2）:815-837.
[17]	KIM M，SONG J，KIM C W，et al. Near-real-time damage identification under vehicle loads using dynamic graph neural network based on proper orthogonal decomposition[J]. Mechanical Systems and Signal Processing，2025，224，112175.
[18]	CUSIDO J，ROMERAL L，ORTEGA J A，et al. Fault detection in induction machines using power spectral density in wavelet decomposition[J]. IEEE Transactions on Industrial Electronics，2008，55（2）:633-643.
[19]	ZHANG Z，GENG Z，HAN Y. Graph structure change-based anomaly detection in multivariate time series of industrial processes[J]. IEEE Transactions on Industrial Informatics，2024，20（4）:6457-6466.
[20]	杨少冲，姚远，张凯，等. 基于本征正交分解和卡尔曼滤波的结构损伤识别[J]. 振动与冲击，2023，42（18）:304-312.
[21]	XU W，YANG B，BI F，et al. Dual-graph collaboration:bidirectional fusion graph convolution network for structure multidefect positioning and assessment[J]. IEEE Transactions on Industrial Informatics，2024，20（11）:13284-13295.