Analytical differences between seven prediction models and the description of the rail track deterioration process through these methods

Abstract

The quality of the rail track is one of the most important indicators to manage a reliable maintenance system so that plan the costs and the technical interventions. In this study, the track state will be defined by the qualifying numbers of the track, which shows the geometrical condition of a given rail section. More than one million measured data with the car (FMK-004) were processed than analysed and defined by configuring and programming a new regression method. The aim was the perfection of an analytic examination, which describes the differences between models of the track deterioration process, through characterized the correspondences more precisely and better to use in practice. Seven models were built and tested for prediction of the track qualifying numbers. One conventional non-linear regression model based on Vaszary’s model using VBA, four new model using basic predictable equations, one new model using linear regression in VBA and one new model using an artificial neural network in MATLAB. When compare the predictions of models, the result shows that exist some basic models with more accurate predictions than a complex model.