Broken Rail Prediction and Detection Using Wavelets and Artificial Neural Networks

Abstract

Current track health monitoring requires time consuming use of railway monitoring vehicles. This paper presents a rail defect detection and classification algorithm that could potentially be used with bogie side frame vertical acceleration data from a data acquisition system located onboard a train car during daily operation. The algorithm uses wavelets to process the vertical acceleration data and detect irregularities in the signal. Wavelets have proven themselves to be useful in event detection and other applications where localization is needed in both the time and frequency domains. Traditional signal processing methods may use the Fourier transform which is limited to localization only in the frequency domain. Wavelets provide a solution for recognizing rail defects and determining their location. The wavelet-processed data is fed into an artificial neural network for defect classification. Neural networks can be a powerful tool in pattern recognition and classification because of their ability to be taught. The network in this algorithm has been trained to recognize impending breaks and breaks in a rail from the original vertical acceleration signal and the first four scales of the wavelet transformed signal. This paper presents an offline analysis of a set of collected data using the proposed defect detection and classification algorithm.