Bridge deck pavement debonding detection system using impact sound and deep learning approach

The present study involved the design and development of a machine learning model for automatically classifying impact sounds from bridge deck pavement using the impact-sounding technique. This technique is non-destructive, eliminating the need to bore holes in the pavement to detect debonding between the asphalt and concrete layers. The model leverages extracted features such as Mel Frequency Cepstrum Coefficients and Short Time Fourier Transforms from the sound data. These features are then represented using chromagrams and spectrograms, and a convolutional neural network (CNN) is employed for image classification. Remarkably, the CNN achieved 97% accuracy in classifying debonded sounds. These findings demonstrate the efficiency of the impact-sounding technique as a non-destructive system for detecting bridge deck pavement debonding. Furthermore, the use of CNNs proves highly suitable for aiding in debonding detection, with significant implications for real-world applications.