Deep learning based intelligent rail track health monitoring system

Abstract

This paper describes the possible way of monitoring the health of the rail track to increase the comfort and ride quality of rail transportation. The abnormalities present in the track are identified and rectified at the initial stage. In this paper, Convolutional Neural Network and Extreme Learning Machine Algorithm (CNN-ELMA) based rail track monitoring is proposed to estimate the exact abnormality. The micro-electro-mechanical sensor (MEMS) accelerometers are fixed in the axle box for measuring the acceleration signal. The location of abnormality is measured by a new sensing method even if the signal of the global positioning system (GPS) is absent. To pre-process the raw signal received from the accelerometer is done by using a Continuous Wavelet Transform (CWT). Then the high-level features are extracted using CNN with a square pooling architecture. To evaluate the performance of the proposed CNN-ELMA, it is simulated and compared with four other methods. The comparison results show that the proposed CNN-ELMA is an effective and accurate method useful for the maintenance department of railways. An experiment has been conducted for the four different abnormal locations and the performance of the proposed method is studied.