Crack Detection of Concrete Structures Using Acoustic Emission Sensors and Convolutional Neural Networks

Structure deterioration is regarded as one of the crucial problems in the construction industry. One method of detecting cracks in concrete structures is using acoustic emission sensors. The conventional acoustic emission sensor-based approach concentrates on measuring the time of arrival, time difference of arrival, and received signal strength indicator. However, these conventional methods are susceptible to a high degree of error due to the presence of inhomogeneous materials. In this study, we propose a new, deep learning-based method for detecting cracks using AE sensors. The objective of this method is to automate the process of detecting cracks and improve accuracy. The proposed method involves the following steps: collecting acoustic emission sensor signals and transforming them into a time-frequency representation using continuous wavelet transform. Next, these representations are inputted into a convolutional neural network that has been designed to localize the crack. Lastly, the trained convolutional neural network is utilized to estimate the coordinates of the crack. The effectiveness and progressiveness of the proposed method were validated through tests on a concrete block with an artificially created crack caused by pencil-lead breaks.