Data-driven fault diagnosis of bogie suspension components with on- board acoustic sensors

Abstract

This paper proposes a data-driven approach for fault-detection and isolation of bogie suspension components with on-board acoustic sensors. The fault detection technique is based on the acoustic emissions variation due to structural modal coupling changes in the presence of faulty components. A suspensions component failure introduces an imbalance into the system, resulting in dynamics interferences between the motions. These interferences modify the energy introduced into the system as well as its acoustic emissions. The unknown arbitrary track irregularities generate together with a variable train speed a random nonstationary vehicle excitation. Speech recognition techniques were used to generate features that consider this phenomenon. Frequency spectrums were analysed in different operating conditions to design efficient features. The robustness of the methodology was verified with data from two different test measurement campaigns on a test ring, where the influence of the sensor locations for the fault classification process was studied. The proposed methodology achieved good fault classification performance on the investigated use cases, removed dampers and 50% damper degradation on primary and secondary vertical suspension.