Rail Breakage Detection Method Using Convolutional Neural Network through Image Conversion of Distributed Acoustic Sensing Data

Recently, research on implementing broken rail detection functions using distributed acoustic sensing (DAS) technology has gained attention worldwide. DAS is a nondestructive testing technology that detects the location and size of a vibration by sending and receiving a signal of a specific frequency to an optical cable and analyzing the signal distortion caused by the vibration. To apply DAS to broken rail detection, the specificity of the noise and vibration caused by rail breakage during train operation must be used. However, the specificity of the noise and vibration not only includes the size or frequency characteristics at a specific time but also the changes in the noise and vibration over time. One method for recognizing these changes cumulatively is through a spectrogram. In this study, a method to determine rail breakage by converting DAS detection data into spectrograms and performing image classification using a convolutional neural network was developed.