Wayside Diagnosis of Wheelset Faults of Metros using One-period Analysis

Abstract

This research is focused on detection of wheelset faults of Prague metro train set of type 81-71M using vibration sensors on the wayside with the contribution of a novel one period analysis. Vibration sensor activities for each metro passing are recorded by two accelerometer sensors which are mounted on both right and left rail for all day while metros are in routine operation. Signal samples of two known faulty wheels of a wheelset on ID-108 metro are used as ground truth information in comparison to ID-119 healthy data. The database has 16 faulty signal samples against 16 healthy ones. Three different methodologies; Wavelet Packet Energy (WPE), Time-Domain Features (TDF) and Linear Configuration Pattern Kurtograms (LCP-K) are used with Fisher Linear Discriminant Analysis and Support Vector Machines classifiers. Outstanding results are observed among proposed techniques up to 100%. Proposed methods may be used for a cost-effective wayside diagnostic system for railway vehicles.