Acceleration data-based robust on-board monitoring of hunting proximity in railway vehicles under uncertainty: A feasibility study

Abstract

The feasibility of on-board, real-time proximity assessment to railway hunting initiation under uncertainty arising from vehicle/track and other operational conditions is assessed. In this context, a novel method that classifies hunting proximity into one of several prespecified speed proximity intervals is introduced, based on a grey nonlinear vehicle model with parameters estimated from on-board acceleration measurements and model-based estimation of the Maximum Lyapunov Exponent (MLE) as a hunting proximity measure. The method's performance and robustness are assessed via Monte Carlo simulations using a MATLAB-based multi-body vehicle model and four operating scenarios. The method is shown to achieve excellent performance achieving 97.08 % classification accuracy, which marks a significant improvement over a potential alternative. Its robustness to faulty suspensions and hollow-worn wheels is also demonstrated, although certain performance compromise is observed under worn-track conditions.