Acoustic Tunnel Lining Detection with Optimized Support Vector Machine

Tunnels are an essential part of modern transportation infrastructure and their structural health is of significant importance for traffic safety. The cavity of tunnel lining has a serious impact on traffic safety. In this paper, an acoustic-based detection method for assessing the integrity of tunnel lining is studied. The acoustic signal is sampled by tapping on the surface of the tunnel lining. A Particle Swarm Optimized Support Vector Machine (PSO-SVM) classification model is built based on Mel-scale Frequency Cepstral Coefficient (MFCC) feature to classify the cavity and the dense acoustic signals of tunnel linings. The two parameters of the SVM are optimized, and the convergence curves are presented. Experimental results show that the recognition accuracy of PSO-SVM achieves up to 94.7%, which is a 39.1% reduction in the error rate compared to the result of SVM with a recognition accuracy of 91.3%. The missing alarm rate is also improved.