Structural Damage Recognition Based on Filtered Feature Selection and Convolutional Neural Network

作者：Zihan Jin, Shuai Teng, Jiqiao Zhang, Gongfa Chen, Fangsen Cui · 发表于：International Journal of Structural Stability and Dynamics · 年份：2022 · DOI：10.1142/s0219455422501346 · 被引用次数：19 · 研究领域：Structural Health Monitoring Techniques、Infrastructure Maintenance and Monitoring、Ultrasonics and Acoustic Wave Propagation

This study proposes a structural damage recognition method based on the filtered feature selection (FFS) and the convolutional neural network (CNN). The FFS usually provides a better sample feature input for a CNN, avoiding the problem that the CNN is prone to over-fitting for the data containing a large amount of invalid feature information. To demonstrate the efficiency and accuracy of the method, a steel frame structure is investigated. The acceleration signals under three different measures (Chi-square test, F test and Mutual information method) of the FFS are studied, along with their influence on the CNN recognition accuracy, network training time and feature dimension. Studies have shown that the Chi-square test has the best effect over the other two measures in terms of efficiency and accuracy. The results of numerical simulations and vibration experiments show that the method has achieved good results in terms of recognition accuracy and training time, and it can significantly reduce the feature dimension while ensuring the accuracy of the CNN recognition.