A Comparative Classification Study for Broken Rotor Bar Fault and its Severity Analysis using Machine Learning Approaches in Rotating Machines

Early fault detection in rotating machines reduces the amount of time, money, and labor needed to repair or replace the machine in the event of an abrupt breakdown that halts production. Due to an increase in the faults and the loss of valuable time and money, industries are now moving towards accurate, and automatic detection and diagnosis of faults in induction machines. Since broken rotor bar faults (BRBFs) occur inside a motor, it goes unidentified till the motor completely fails. This study presents a comparative approach to detecting broken rotor bar faults and assessing their corresponding severity levels. After acquisition of the experimental data (3-phase currents), the Extended Park’s Vector Approach (EVPA) has been utilized to transform the 3-phase currents, after which, the statistical time domain features have been used to develop the feature set. Thereafter, Principal Component Analysis (PCA), has been used to study the geometry of the data and also used as a dimensionality reduction tool to transform and reduce the feature set. A comparative study has been presented where a variety of neural and non-neural classification models have been trained and tested using PCA and non-PCA transformed feature-sets to determine the best model. Using the developed feature sets, best results are presented by the Ensemble classifier which stems from the Bagged Tree family.