Abstract
The dynamical performance of vehicles on railway tracks is significantly influenced by wheel-rail interactions, which makes the wheel-rail force be a safety indicator of vehicle systems. Owing to the high cost of direct measurement, inversion models are widely used to measure the wheel-rail force. This study proposes a model called the Lightweight Wheel-Rail Force Inversion Model (LFIM) to calculate the wheel-rail force by using vibration signals collected from the vehicle system. The model can extract the relevant features, and outperforms prevalent models. It is also resource efficient and small in size, which makes it suitable for edge computing environment. The basic LFIM yielded a correlation coefficient of 0.99 in experiments, significantly higher than the value recorded by the traditional dynamic methods. The compressed LFIM yielded a 91% reduction in floating-point operations (FLOPs) and had an R-2 score of 90%, which satisfies the demands for vehicle monitoring. By balancing accuracy and size, LFIM is a feasible model to deploy on edge devices to monitor the wheel-rail force in real-time and assess the safety of railway vehicles.