Intelligent monitoring of concrete-rock interface debonding via ultrasonic measurement integrated with convolutional neural network

This paper presents an intelligent methodology for monitoring and assessing the interfacial debonding damage process of concrete-rock composites through the piezoelectric-based ultrasonic testing method coordinated with convolutional neural network (CNN). The main novelties of this research lie in the following aspects. Firstly, the ultrasonic sensing method is successfully extended to monitoring the debonding of concrete-rock composites with different interfacial roughness for the first time, and changes in recorded time-domain signals, continuous wavelet packet (CWT) time-frequency spectra and wavelet packet energy are thoroughly analyzed. The results confirm the good consistency of wavelet packet energy with the interfacial debonding damage. The key inno-vation of this study is that a CNN model is constructed and well trained to identify different debonding phases from image-based datasets of CWT spectra. The results well demonstrate the effectiveness and high accuracy of the CNN model. Moreover, comparative analysis with three machine learning-based models is also performed, which demonstrates the superiority of the CNN model. The proposed methodology could facilitate intelligent monitoring and automatic classification of concrete-rock interfacial debonding without manual feature extraction.

Automated summary

This paper presents an intelligent methodology for monitoring and assessing the interfacial debonding damage process of concrete-rock composites through the piezoelectric-based ultrasonic testing method coordinated with convolutional neural network (CNN). The ultrasonic sensing method is successfully extended to monitoring the debonding of concrete-rock composites with different interfacial roughness for the first time. A CNN model is constructed and well trained to identify different debonding phases from image-based datasets of CWT spectra. Comparative analysis with three machine learning-based models demonstrates the superiority of the CNN model.