Stator Inter-Turn Short Circuit Fault Diagnosis using Wavelet Scattering Network Feature Extraction

Stator inter-turn short circuit (ITSC) fault is among the very serious electrical faults in an induction machine drive system. Incipient detection of the fault increases reliability of the drive system. Data driven models have gained popularity in the diagnosis of induction machine faults at early stage. This enables either taking preventive actions or employ fault tolerant control of the drive system. However, performance of fault diagnosis is highly dependent on the fault feature extraction techniques employed. This paper proposes a wavelet scattering feature extraction, a time-frequency technique, for machine learning based stator ITSC fault diagnosis of three-phase induction machine. Axial leakage flux signature is used to diagnosis fault severity. To evaluate the non-stationary operating condition of the actual drive system, a dataset with different operating conditions is recorded. Three incipient fault severity levels (1.41%, 4.81%, and 9.26%) are considered for high-impedance and low-impedance type ITSC to simulate six classes of fault. Artificial neural network (ANN), support vector machine (SVM), and K-nearest neighbor (KNN) models are trained using features extracted from time-domain, frequency-domain, and Time-frequency methods. Their performances are compared in terms of 10-fold crossvalidation accuracy. The results show that the models trained by the proposed feature extraction method classify fault severity levels with better accuracy.