Accurate Detection of Bearing Faults Using Difference Visibility Graph and Bi-directional Long Short-term Memory Network Classifier

This paper proposes a novel bearing fault detection framework for the real-time condition monitoring of induction motors based on difference visibility graph (DVG) theory. In this regard, the vibration signals of healthy as well as different rolling bearing defects were acquired from both fan-end and drive-end accelerometers. These data were recorded for three different bearing defects and under four loading conditions. The acquired vibration time series were converted to a topological network using DVG. From the transformed vibration data in the graph domain, degree distribution (DD) was selected as features to discriminate different fault networks. Using analysis of variance (ANOVA) test and false discovery rate (FDR) correction, most discriminative DD features were selected. These features were subsequently fed as inputs to a deep learning model, i.e., a bi-directional long short-term network (bi-LSTM) classifier for fault classification. In this study, 112 classification problems have been addressed, and for all of them, the proposed approach delivered very high fault detection accuracy. Finally, the classification performance of the proposed framework is compared with other well-known machine-learning classifiers all of which delivered satisfactory results.