Damage localization using acoustic emission sensors via convolutional neural network and continuous wavelet transform

Abstract

Due to aging structures, deterioration is becoming an essential issue in the engineering and facility management industry. Especially for nuclear power plants, the deterioration of structures could be directly related to safety issues. One of the popular methods for localizing damage such as cracks in nuclear power plants in the early stage is using acoustic emission sensors. The conventional methods for localizing damage using the acoustic emission sensor include methods such as time of arrival, time difference of arrival, and received signal strength indicator measurements. However, the conventional methods have large errors especially when the material is not homogeneous, or the propagation path of signals is non-straight. In this study, we propose a new deep learning-based damage localization method using acoustic emission sensors to automate the damage localization process and improve accuracy. First, the signals from acoustic emission sensors were collected and transformed into time–frequency domain images using continuous wavelet transform. Next, the convolutional neural networks were designed to localize the damage using the continuous wavelet transform images as the input. Finally, the trained convolutional neural networks were used to estimate the location or coordinates of damages. To validate the performance of the proposed method, experimental tests were conducted in the concrete panel and cube with artificially generated damages. The results express that the proposed method is effective and progressive.