Data-Driven In-Train Force Monitoring Under Changing Consist Configurations and Loads

Abstract

The reliable monitoring of in-train forces is a critical component for ensuring the safety and efficiency of heavy-haul railway operations. Excessive longitudinal forces can lead to severe operational failures, including coupler breakage, train separation, and catastrophic derailments. Traditionally, these forces have been estimated using computationally expensive physics-based longitudinal train dynamics models, which struggle to adapt to real-time operations due to their reliance on static parameters and deterministic assumptions. This paper presents a comprehensive data-driven approach to monitor and predict in-train forces under dynamically changing consist configurations and varying payload distributions. By leveraging advanced machine learning algorithms, the proposed framework integrates multidimensional operational data, including locomotive control inputs, track topography, and historical sensor readings. We explore the transition from classical physical modeling to modern artificial intelligence techniques, evaluating models ranging from ensemble tree-based methods to recurrent neural networks designed for sequential time-series forecasting. Special attention is given to the challenge of varying train configurations, where the number, type, and weight of wagons change frequently, rendering static models obsolete. Through extensive analysis of simulated and real-world operational datasets, this research demonstrates that data-driven methodologies can achieve high accuracy in predicting draft gear forces and slack action occurrences. The findings provide a scalable, real-time solution for intelligent train control systems, ultimately contributing to enhanced railway safety, reduced maintenance costs, and optimized network capacity.