Search

It is generally observed that the condition of rail tracks degrades rapidly over time until and unless effective maintenance is carried out. In the rail industry, rail maintenance actions are usually reactive, which means that maintenance is carried out after a defect has been identified. Unfortunately, this approach can lead to general safety concerns and may result in costly maintenance. Predictive maintenance, which aims to predict the future behaviour of track degradation based on the analysis of already recorded data, can be used to identify defects in advance, thus providing a solution for the above safety and cost concerns. Two important questions for which answers are sought in predictive maintenance of rail track are: where does the fault occur and when. The aim of the research presented in this thesis is to develop a novel predictive rail track degradation model that answers the above questions. The proposed model consists of an alignment component for effective alignment of data and a degradation component for understanding rail track degradation based on rail profile and track geometry parametric analysis. The thesis takes an incremental approach to data alignment proposing three different algorithms namely, distance alignment, fixed window based alignment and parameter based alignment. It is proven that the latter approach provides the most accurate data alignment algorithm. The degradation component of the proposed model is based on a comprehensive multivariate and univariate analysis. In multivariate analysis, parameters of a base file i.e. a file consisting of parameters belonging to the same segment of the rail track at a given time of measurement are predicted using all other parameters of the same file. In univariate analysis, every parameter of a given base file is predicted, temporally, from the corresponding parameters in the previous base files. Such contribution analysis manifests the level to which each parameter contributes in predicting other parameters and over time. Subsequent to univariate and iii multivariate analysis the predictive errors are thresholded into either exceedences i.e. they exceed the threshold line, needing immediate maintenance, or normal i.e. they are below the threshold line, needing no immediate maintenance. The research presented in this thesis shows that in multivariate analysis, rail profile parameters were predicted with 97% prediction accuracy below threshold, whereas track geometry parameters were predicted with 99% prediction accuracy below threshold. Both univariate and multivariate analysis will serve as the basis in monitoring track conditions and thus finding track degradation problems. This will greatly aid in planning predictive track degradation by providing an objective means of evaluating track conditions and hence the over all life of the rail track will increase.