Induction motor bearing fault classification using WPT, PCA and DSVM

In industry loading on bearing and its fault severity in an induction motor is unpredictable. Hence, in the present article wide range of vibration data from induction motor bearing surface has been taken for extraction of fault features and classified for the detection of mechanical faults presents in the bearing so that condition based monitoring possible. The vibration data which is selected in this paper includes four different kinds of loading and three different types of fault with three different fault sizes. Firstly, Wavelet Packet Transform (WPT) is applied to decompose the vibration signal and develop Bearing Damage Index (BDI) from the decomposed signal to select the useful signal from the original recorded signal. This BDI based useful signal is further applied for extraction of statistical features and fed to the classifier. Total eleven time domain features has been calculated and Principal Component Analysis (PCA) is applied for the selection of significant features. The selected features further used as input to the Dendogram Support Vector Machine (DSVM) classifier to identify the faults. This proposed method shows significant improvement in classification rate as compared to conventional method, which is quite promising and encouraging.