Predictive modeling of track degradation rates considering drainage conditions for maintenance scheduling

High Track Degradation Rates (TDR) can lead to increased maintenance costs and a decline in ride quality. As a result, it is crucial to examine the factors contributing to high TDR and develop methods for predicting its value. In this study, geometry data of 280 miles of passenger revenue track from 2011 to 2021 are investigated. Ground Penetrating Radar (GPR) and LiDAR data are utilized alongside geometry data to probe the subsurface and drainage conditions of the tracks, respectively. Key properties of ditches, including depth, distance, and longitudinal condition, are extracted from LiDAR data. The findings emphasize the significant impact of lateral external drainage on TDR values; specifically, blocks with poor lateral drainage (marked by shallow ditches located far from the track) exhibit a TDR approximately 55% higher than blocks with favorable lateral drainage conditions. Furthermore, 12 features are extracted for TDR prediction, encompassing parameters such as track subsurface condition, ditch properties, and geometry history. Four Machine Learning (ML) models are employed to predict TDR, with XGBoost marginally outperforming the others, demonstrating successful TDR prediction with a MAE of 0.019 and RMSE of 0.029.