.png){kind=icon}
.png){kind=icon}
Return
Steel structural damage identification based on multisensor same-class graph and GraphSAGE network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.istruc.2025.108388.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Structural damage identification often faces challenges in extracting appropriate damage features when there are limited damage samples and noise interference, which can lead to a decrease in identification accuracy. This paper presents a steel structural damage identification approach based on a multisensor same-class graph and GraphSAGE network (MSC-GraphSAGE). Firstly, the spatial-temporal graphs of the data are established by Continuous Wavelet Transform (CWT). Then, the Laplacian matrix in graph theory extracts the features in each spatial-temporal graph. Multiple sensor spatial-temporal graph features are merged to construct a multisensor same-class graph to address the problems of poor model noise immunity and degradation of recognition performance for small sample datasets. Ultimately, the GraphSAGE network is utilized to classify the nodes in the multisensor same-class graph to identify the damage to the steel structure. The method in this paper is evaluated using the Steel Truss Bridge (STB) dataset and the Qatar University Grandstand Simulator (QUGS) dataset, and experiment findings indicate that the identification accuracy of MSC-GraphSAGE is superior and robust, and the recognition accuracy can reach 97.7% for the STB dataset and 99% for the QUGS dataset. With limited damage samples and noise interference, the identification results remain stable without significant fluctuation.
Keywords:
Structural health monitoring
Structural damage identification
Continuous wavelet transform
Laplacian matrix
Multisensor same-class graph
GraphSAGE network