Performance of a load-immune classifier for robust identification of minor faults in induction motor stator winding

Reliable detection of induction motor stator winding insulation failure at its early stages is a challenging issue in modern industry. Insulation failure between small number of turns, involving less than 5% turns of phase winding are often indiscernible and detection becomes even more complicated when motor operates at varying load levels. In line-fed motors, supply voltage unbalance is another inadvertent issue which may tend to exhibit current signature similar to stator winding inter-turn insulation failure case. The proposed work presents a robust system, to identify severity of stator winding insulation faults when an induction motor with random wound stator winding works under such operating conditions. In the present work, various features obtained from time, frequency, timefrequency, and non-linear analysis of stator currents at various stator winding short circuit faults and supply voltage unbalance conditions for different load levels have been studied. A Support Vector Machine based Recursive Feature Elimination (SVM-RFE) algorithm is used to identify the features which can provide discrimination information related to severity of fault level, independent of supply voltage unbalance and immune to load level variations. Among the extracted features, features obtained through Detrended Fluctuation Analysis (DFA) are found to be most robust for this purpose. Finally a Support Vector Machine in Regression mode (SVR) has been formed to identify winding failures employing the optimum number of features selected through SVM-RFE technique.