SVM Based Bearing Fault Diagnosis in Induction Motors Using Frequency Spectrum Features of Stator Current

Abstract

Induction motors are the most popular motors in the world. Unscheduled breakdowns often lead to financial losses. The most common failure of induction motors is bearing related. Typically, vibration measuring methods are used to diagnose this type of faults. This study relies on stator current based diagnosis of bearing faults. Compared to the measurement of vibration, the stator's current-based method is less invasive and physically do not require to reach the motor housing. In this study, the most informative features are selected from stator current spectrum amplitudes. Feature weight vector is created by the application of Neighbourhood Component Feature Selection method. Support Vector Machine is used as supervised machine learning method for classification. In order to investigate feature selection and classifier performance an experiment with three artificially caused bearing faults were performed. The most informative spectrum points are discussed.