Detection of electrical faults in induction motor fed by inverter using Support Vector Machine and Receiver Operating Characteristic

Fault in induction motor is crucial problem in industrial processes. This paper presents the system for electrical fault detection in induction motor fed by inverter. Current spectrum with different frequency is used to fault monitoring. Faults observed includes variation of frequency, unbalance voltage, and inter turn short circuits. Through an experiment, the fault was fired and the current spectrum recorded at steady state condition. Preprocessing is performed before the identification process. It includes noise reduction using wavelet analysis and feature extraction with Principal Component Analysis (PCA). Both processes are intended to eliminate the noise, reducing the dimension of feature, and retrieve components of the optimal features for classification. Strength of identification capability using Support Vector Machine (SVM) is 83.51%. Based on the ROC (Receiver Operating Characteristic) analysis, the SVM classifier has a good enough performance. This is indicated by the sensitivity is 74.31%, specificity is 47.30% and G-Mean is 1.1028.