Research On Railway Safety Monitoring Based On Micro-structured Fiber Distributed Sensing System

Railway is one of the most important infrastructures in the country.Real-time monitoring of the health status of railways is an effective guarantee for the safe operation of railway systems.The traditional electrical sensors have the problems of poor resistance to electromagnetic interference,limited signal transmission distance,and high cost of forming a sensing network.The micro-structured optical fiber distributed acoustic sensing(MOF-DAS)system combines sensing and transmission into one,and multiple sensing units are integrated on a single fiber.A long-distance,high-sensitivity and distributed acoustic sensing can be realized with a single fiber.This technology has great application prospects in the field of railway health monitoring.This paper applies DAS to the field of railway safety monitoring,designing experiments,collecting and analyzing data.For the two key problems of train positioning and rail defect detection,algorithms for solving problems is proposed.This paper has mainly completed the following research work:(1)The basic principles of DAS were introduced.The advantages of DAS in the field of railway safety monitoring were analyzed.A railway safety monitoring program based on DAS was designed.The microstructure sensing fiber is fixed at the rail waist position,and the data of the rail vibration signal is collected through experiments,and the signal is analyzed.(2)A DAS-based train positioning algorithm is designed.The method takes the absolute value of the signal per second of the sensing node and takes the average value in the time dimension.The mean value is used to characterize the vibration of the rail.By comparing this mean value with the set threshold value,it is judged whether there is a train passing on the sensor node.(3)A DAS-based rail defect detection method is designed.The method uses the experimental data of the train passing a standard damage rail to obtain the signal of the rail vibration of the sensor node at the defect-free and the sensor node at the defect.The timefrequency domain characteristics of these signals are used as input to machine learning algorithms.The LightGBM framework is used to classify the two types of signals and train the model to detect defects.