WITHDRAWN: Condition Monitoring of Metro Wheelsets Using Wayside Vibration and Acoustic Sensors

Abstract

This paper presents a suitable framework that can diagnose wheelset related faults on metro trains using acoustic and vibration based techniques on the wayside. Proposed condition monitoring system includes four main stages; data acquisition, signal segmentation via one period analysis, feature extraction; WPE – Wavelet Packet Energy, TDF – Time-Domain Features and LCP-K – Linear Configuration Pattern Kurtograms and classification with state-of-art classifiers; FLDA – Fisher’s Linear Discriminant Analysis, SVM-I – Support Vector Machine, Dec. Tree – Decision Tree and PERLC – Linear Perceptron in addition to classifier combination techniques. Throughout the study, two databases (A1-A2), each of which is consistent of measured vibration and acoustic signals belong to healthy and faulty cases of the wheelsets of Prague metros respectively, are created. Since the number observations of faulty cases are limited, features are generated with ADASYN – Adaptive Synthetic Sampling and larger vibration and acoustic databases (SA1-SA2) are created. Acquired results show that TDF with Dec. Tree algorithm is capable of detecting wheelset faults with a 100\% accuracy in A1 and novel LCP-K method outperforms both in A2 and SA2 acoustic databases up to 93\% success rate while WPE outperforms via combined classifier approach by reaching 100\% detection performance. The proposed framework is considered to be helpful to the maintenance specialists since it can provide cost-effective maintenance of wheelsets not only for metros but also for other types of railway vehicles with its outstanding performance.

A Survey on Fault Diagnosis Approaches for Rolling Bearings of Railway Vehicles

“…Zhang et al proposed a Doppler feature matching search algorithm (DFMS) based on the fusion time-frequency distribution (TFD) of a raw signal to solve the Doppler distortion problem [117]. Kilinc and Vagner combined the support vector machine (SVM) and fisher linear discriminant analysis (FLDA) for the defect classification [118]. The test results have verified the availability of the model in multi-source acoustic diagnosis for defective train bearings.…”

Wheel Defect Detection Using a Hybrid Deep Learning Approach

“…For instance, in the Swedish railway network, the wayside equipment for monitoring rolling stock consists of almost 200 wayside inspection devices [ 16 ]. Moreover, this method also requires detailed information about a target vehicle (e.g., number of axles and type of wagon) for accurate condition monitoring [ 17 ]. The high cost and maintenance issues associated with this method limit its widespread use.…”