Wavelet-based multi-class support vector machine for stator fault diagnosis in induction motor

This work proposes a novel online detection scheme to diagnose incipient inter-turn short circuit fault and estimate the failure severity in induction motor. Incipient detection of the stator failure during the machine running, as well as identification of its intensity can reduce the risk of additional damage to the phase winding, improve the operational efficiency, and ensure machine availability. Hence, the incipient fault diagnosis provides a safe operating area for the motor. This work aims to specify the percentage of defective turns in the shorted winding by proposing a new mathematical parameter based on wavelet analysis, in addition to employ a multi-class support vector machine to perform the classification task. Discrete Wavelet Transform is used to analyze the stator currents after modeling the motor utilizing Clarke-Concordia transformation. From the detailed coefficients, Max and L2 norms are calculated. The adopted parameter is computed depending on the previous norms, which form the input vector to feed the classifier. The multi-class support vector machine–based one versus one algorithm is used to determine accurately the defect intensity. The acquired outcomes prove that the proposed approach, depending on the novel parameter along with multi-class support vector machine can give a robust and accurate indication about the machine status, which enables the estimation of the fault severity. To verify the competency of the methodology, various hardware experiments are carried out on the motor. The experimental results demonstrate the validity and practicability of the method, with a higher level of correctness, exceeding 96%.