Models for Forecasting Railroad Track Geometry Degradation Using Machine Learning Methods

Abstract

Railroad track maintenance has always been complex, both because of its responsibility in terms of ensuring the safety of train traffic, and because of the high labor intensity of work processes and continuous work planning. Diagnostics and monitoring of all elements of the railroad track is carried out to ensure the safety of train traffic. One of the main parameters affecting the safety and uninterrupted movement of trains is the condition of the track. Deviations and malfunctions in rail track geometry lead to both speed limitation and complete closure of the track for train traffic. Failure to correct faults in a timely manner can often lead to more serious consequences. The main parameters of track geometry have a significant impact on the smoothness of train movement and the risks of derailment of rolling stock. Therefore, monitoring and control of these parameters is a priority task to ensure the stable operation of railroads. To automate this process, data-driven fault detection and diagnosis models can be used. To solve the problem, we used modern methods of solving classification problems for tabular data collected by special track-measuring tools. Automated machine learning model generation systems served as the basis for solving the problem. These systems make it much easier to train and configure machine learning models, as well as to implement them in a production environment. The practical significance of this work is that the solution of the problem of predicting track geometry degradation can be considered as part of the decision-making system for track repair and maintenance.