SVM-based stator fault diagnosis for induction motors using DQ0 voltage components

Abstract

This paper presents a new technique for diagnosis of stator winding shorted turns in three phase induction motors using support vector machine (SVM) as a classifier in conjunction with a voltage signature analysis as feature extraction. The proposed dqO voltage components related to three phase voltages are obtained by using Park's transformation. Unlike general dqO quantity, the required dq components are considered from only harmonics in a certain range resulted from the stator shorted turn fault whilst the required zero component is determined from the fundamental voltage excitation only. The procedure for finding the harmonic spectrum associated with the stator fault employs a FFT technique together with digital signal processing in order to enlarge such harmonics to be easily detected. The data of d, q and zero components will be used as the input features for a SVM classification and diagnosis. In order to confirm that the dqO voltage components could be used as quantity for identifying the stator fault, patterns of dqO voltage components in three-dimension are plotted. The obtained trajectories are obviously different depending on the severity of the shorted turn fault. In training and testing with a multiclass-SVM classification, the good accuracy is obtained under various load conditions for both healthy and stator fault conditions. The SVM classification performance of the proposed method has high accuracy greater than 98% comparable to other feature extraction processes such as a motor current signature method (MCSA) but the number of training data and features for the proposed technique are less. © 2010 ICIC International.