Subway tunnel damage detection based on in-service train dynamic response, variational mode decomposition, convolutional neural networks and long short-term memory

This study describes a method for detecting tunnel damage by using vertical acceleration of an in-service subway train and a multi-step strategy based on variational mode decomposition, convolutional neural networks and long short-term memory. In contrast to conventional methods, the strategy is based on multiple classifiers and can extract the damage information using a step-by-step approach at low cost and high efficiency. Laboratory tests were conducted to verify the performance of the proposed method on tunnel damages such as lining concrete spalling, surface overload, and voids behind the tunnel segment. Results show that the proposed strategy can accurately identify the location, type, and degree of the damage with an accuracy of 95%, 95%, and 91% and Kappa coefficients of 0.94, 0.93, and 0.88, respectively. Compared to CNN, CNN-LSTM, and WPD used in the identification of tunnel damages, the proposed method exhibited higher performance in terms of accurate classification.

Automated summary

This study presents a novel method for detecting tunnel damage by combining multiple classifiers with variational mode decomposition, convolutional neural networks, and long short-term memory. The proposed method can accurately identify the location, type, and degree of the damage at low cost and high efficiency.