Using machine learning techniques as track geometry predictors for railway track

Abstract

This paper investigates utilising machine learning (ML) techniques to predict the five major parameters of track geometry in railway infrastructure. Track geometry and incurred vehicle acceleration measurements were collected on a railway line and matched according to their GPS coordinates. The data were then split into test/train and validation datasets and processed using various ML methods. The predictive results of each ML method were compared for each track geometry parameter and the best methods were highlighted. The quality of the results was mixed with accurate results obtained for cant and alignment but inaccurate results for gauge, profile and twist. Overall, this research paper contributes to the field of railway engineering by demonstrating the potential to utiliseML in the field by predicting track geometry parameters. The findings have practical implications for improving track maintenance and ensuring passenger safety and comfort in railway operations. The promising results of this paper warrant more research being conducted and potential methods for improvement are highlighted.