WAYSIDE DIAGNOSIS OF METRO WHEELSETS USING ACOUSTIC SENSOR DATA AND ONE-PERIOD ANALYSIS

Abstract eng:
This research promises a wheelset fault diagnosis methodology for metro train sets using wayside acoustic sensor information. Throughout the research, two different feature extraction techniques; Wavelet Packet Energy (WPE) and Time-domain Features (TDF) are employed in association with two state-of-art classifiers Fisher Linear Discriminant Analysis (FLDA) and Support Vector Machines (SVMs). The database is prepared by the acquisition of wayside acoustic sensor data accompanied by optical gates that detect wheelset center position while multiple passing of a single metro train set of type 81-71M is in daily operation with the contribution of a novel approach; one-period analysis. Acquired database is then divided into two classes which represent the healthy and faulty states of the wheelsets referring to the ground truth information of a faulty wheelset. Since the faulty states are insufficient to demonstrate the real classification performance, an adaptive synthetic sampling technique (ADASYN) is utilized to increase the number of faulty states. Promising results are observed up to 93 % in classification of faulty wheelsets of the metro with the proposed techniques on acoustic sensor data. This study may aid to maintenance specialists by providing a cost effective monitoring of faulty condition of metro wheelsets.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket