Development of an Artificial Intelligence-Based Predictive Model for the Derailment Coefficient of Railway Tracks

Derailment, one of the most critical accidents in railway systems, is directly associated with running safety. The derailment coefficient, which is widely used as a representative quantitative indicator for derailment evaluation, is determined by the interaction between the vehicle and track and is closely related to track conditions. Previous studies have investigated the effects of operational conditions, such as curve radius, cant, and speed, as well as track conditions, on the derailment coefficient and running safety. However, most of these studies are limited to analytical approaches involving specific conditions or analyses based on limited field-measurement data. This study proposes an artificial intelligence (AI)-based model to predict the derailment coefficient using field-track inspection data. The input variables comprise track geometry parameters, including the alignment, gauge, cross-level, twist, and longitudinal level, all obtained from track inspection vehicles, which effectively reflect actual track conditions. The results demonstrated that the proposed AI-based model effectively captures the complex nonlinear relationships between track inspection data and the derailment coefficient, thereby achieving high prediction accuracy.