PCA based feature extraction for classification of stator-winding faults in induction motors

Nowadays,induction motors are widely used for many industrial processes. the shorted-turn fault of the stator-winding is the initial point of stator winding faults. this paper proposes using the principal component analysis (pca) to reduce the dimension of the feature set which is obtained from the motor current signature analysis (mcsa). the six original features consist of the signal power of the three-phase filtered current signal at 20 hz to 80 hz and 120 hz to 180 hz of the phases a,b and c. after using the pca,the dimension of the feature set decreases to two new features. these two new features are then used to classify the shorted-turn phases of the stator-winding by applying the artificial neural network (ann) classifier. the experimental results demonstrate that the new feature set can decrease the complexity of the system. additionally,the accuracy rate using the new feature set is higher than using the original feature set. therefore,the new feature set can properly improve the efficiency of the classification. © 2017 universiti putra malaysia press.