ON THE PROBLEM OF ON-BOARD VIBRATION-BASED FAULT DETECTION IN RAILWAY SUSPENSIONS UNDER VARYING OPERATING CONDITIONS: A FEASIBILITY STUDY

摘要

The feasibility of on-board random vibration-based fault detection in railway suspension systems based on a single, car body mounted, vibration sensor and variable nominal traveling speeds (Operating Conditions, OCs) is explored via robust Statistical Time Series (STS) type detection methods. The study is based on field tests performed on an Athens Metro vehicle traveling at 60, 70, or 80 km/h, while two fault scenarios are considered: Air Spring (AS) pressure reduction by 25% and Lateral Damper (LD) subject to lubricant loss. The problem is shown to be challenging due to the use of a single sensor, the considerable influence of the OCs (varying nominal speed) on the dynamics, and the minor nature of the LD fault. The study focuses on two Multiple Model type methods: A non-parametric Power Spectral Density based method and a parametric AutoRegressive model based method. The results of the study, based on thousands of test cases, demonstrate perfect detection performance of both methods for the AS fault, but, expectedly, degraded performance (especially by the non-parametric method) for the LD fault. Preliminary comparisons with an alternative method, employing the Multiscale Permutation Entropy (MPE) and Linear Local Tangent Space Alignment (LLTSA), have indicated significant degradation in performance for both fault scenarios.