Backpropagation Algorithms of Neural Networks to construct the Railway Track Deterioration Model

Abstract

Rail transport is the fastest growing transport sector with rapid increase in rail passengers and freight movement resulting in accelerated rate of railway track deterioration and maintenance costs in the United Kingdom. The Network Rail requires the reliable prediction model for railway track deterioration rate to support the decision support system of railway track management. This paper applies the Backpropagation-Artificial Neural Network (BP-ANN) with Generalised Delta Rule (GDR) learning algorithm to construct the deterioration model for railway track of 200-m sections at London-Wolverhampton and Wolverhampton-London routes. The track geometry, Ballast Fouling Index, train speed, catch pits, ballast age and sleeper age are the parameters of railway track deterioration rate. The BP-ANN models estimate that standard deviation of 35m vertical profile (mm) is the most significant factor of railway track deterioration followed by train speed, Ballast Fouling Index, rail sleeper age and ballast age. The findings of this paper can support the Network Rail and other railway agencies by predicting reliable models for railway track deterioration rate.