Train detection and classification using distributed fiber-optic acoustic sensing

Abstract

With the rapid development of the high-speed railway industry, train detection and identification play a vital role in capacity improvement and safe operation in railway systems. Conventional detection methods such as track circuit and axle counting tend to be interfered with by severe weather conditions and irrelevant conductive objects, leading to false detections. Fiber-optic distributed acoustic sensing (DAS) technology is a prevailing sensing method in geophysics research, petroleum exploration, and structure inspection. Compared to traditional detection techniques, DAS is suitable for long-distance detection and is resistant to severe weather conditions and electrical interference. We have developed a train detection and classification system using DAS technology and have explored an effective classification method for train identification. Specifically, we conduct a field experiment by the side of a railroad over viaducts and the data are collected with the DAS detection system. To eliminate the impact of background noise, DC noise, and motor vehicle signals from the original data, we adopt a wavelet denoising method and Chebyshev filter to extract the features of three types of train signals. The vibration signals of these different trains indicate remarkable cyclical variations related to the number of wheelsets in the time domain and have similar narrow-band discrete spectrums with different characteristic peak frequencies. Furthermore, based on the features of the train signals, we select a support vector machine classifier to identify three types of trains, with accuracy greater than 97%.