Fault Diagnosis in Railway Track Circuits Using Support Vector Machines

This paper presents a fault diagnosis system for a key component called electrical separation joint in railway track circuits using multi-class support vector machines by one-versus-one strategy. Firstly, a track circuit electrical model is developed based on transmission line theory. A signal known as short-circuit current signal is obtained and the influences on it are then investigated for the existence of defective electrical separation joints. The signal is composed of arched curve segments, and each of the segments can be approximated by a quadratic polynomial. The coefficients of the polynomials for the first three arched segments are used as fault features to train the proposed diagnosis mode. Training parameters are selected using cross-validation technique. Polynomial and Gaussian RBF kernel functions are employed. Experiments with simulated data show that the correct diagnosis rates over 96% can be achieved using this approach, which can meet the requirement of practical application.