Determination of Crack Signals Using the Deep Learning Technique Based on a 1D Convolutional Neural Network for Smart Detection of Structural Damage Cracking

Abstract

Structures can be damaged by natural disasters such as earthquakes and typhoons. In particular, any damage to the foundation of a structure can present critical problems. Therefore, a smart monitoring technique such as the acoustic emission method is required to detect internal cracks and other types of structural damage. Many laboratory studies on this method have been conducted to estimate the locations and sizes of cracks as well as the resulting changes in structural durability using collected acoustic signals. However, the method has rarely been applied in the field because identifying damage signals from acquired signals, which can contain ambient noise, is difficult. We developed a deep learning algorithm based on a one-dimensional convolutional neural network method that can identify damage or crack signals generated from concrete failure from randomly synthesized signals. Using the developed algorithm, we were able to distinguish damage signals from random ambient noise signals. This algorithm enables real-time monitoring of concrete structures, thus providing a smart monitoring strategy. Keywords: Structure Health Monitoring, Machine Learning, 1D Convolution Network, Accelerometer