Bearing and gear failure detection for brushless DC motors with adaptive feature extraction and classification

This paper presents a novel approach to adaptively select features for early fault detection on bearings and gears connected to brushless DC motors (BLDCM). Multisensor data are collected using a state-of-the-art testing platform to induce faults on BLDCMs with time varying conditions. Due to the high number of features and sensor channels available, determining the right data to identify faults can be a daunting task to achieve. A series of gears and bearings are tested. An algorithm using adaptive selection of features is proposed to improve fault detection. A benchmark data set is built containing multi-sensing data for different fault scenarios for BLDCMs with time varying conditions. The algorithm presented in this work is applied on measurements data to be included in the data set.