Intelligent on-line railway maintenance based on distributed acoustic sensing and machine learning

With the increasing density of railway operations, infrastructure issues such as wheelset flats and loosening of sound barriers occur more frequently, while open railway environments are more susceptible to external disturbances, including intrusions, construction activities, and rockfalls, creating an urgent demand for efficient monitoring methods. Traditional railway safety monitoring methods face notable limitations, including restricted monitoring distances, vulnerability to environmental influences, and difficulty in maintenance. Distributed acoustic sensing (DAS) technology, with its long-distance monitoring capability, high sensitivity, and strong resistance to electromagnetic interference, has substantial potential to meet these pressing demands. To address the three typical monitoring targets in railway operation and maintenance, this paper proposes an integrated monitoring method based on DAS technology and machine learning methods. Experiments demonstrate that the proposed method possesses multi-track monitoring capability and can simultaneously detect and identify wheelset fault conditions. Meanwhile, it achieves identification accuracies of 97.03% for abnormal disturbance detection and 99.6% for sound barrier fault recognition. This method provides a reliable and promising technical solution for railway monitoring. It can even allow for the implementation of sensing functions using existing track-side communication cables.