분산형 음향 검지 데이터의 이미지 변환과 합성곱 신경망을 통한 레일절손 검지 방안

초록

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.

키워드