SVM Based Fault Classification of Three Phase Induction Motor

Early detection of abnormal conditions during induction motor's operation would eliminate consequential damages on motor, so that outage time and costs of repairing can be reduced. Due to unique fingerprints from faults in line currents, it is possible to detect faults by extracting fault information from line currents. From the literature review it is observed that in many model based on ANN based techniques, the stator current spectra is used as input feature space. In this paper, simple thirteen statistical parameters are used as an input feature space. Support Vector Machine (SVM) is proposed as a fault classifier. Since the basic SVM is essentially a 2-class classifier, the synergism of three classifiers is proposed to overcome the limitation. Principal Component Analysis (PCA) is used as data fusion method to reduce the dimension of classifier. To verify the performance various kernel function as (Radial Basis Function (RBF), Quadratic, Linear, Polynomial, Multilayer Perceptron (MLP)) are applied and tested with real experimental datasets. In order to generate the experimental data, specially designed 2 HP, three phase, 4 pole, 415V, 50 Hz induction motor is used.