A new intelligent fault classification method using time series data mining and support vector machines

Abstract

Induction motors are the most used machines in industrial applications. Although these motors are generally reliably, they can be exposed many faults due to environmental and wear reasons. In this study, broken rotor bar faults occurred in an induction motor are classified using support vector machines and time series data mining methods. Time series to be used for fault diagnosis is obtained by using two components of park's vector approach. The new time series which is obtained for each fault condition is transformed to a phase space. Support vector machines are used to separate healthy and faulty phase spaces. Fuzzy cluster centers are taken as training data to increase training speed of support vector machines. Healthy motor condition and one, two and three broken rotor bar faults are successfully classified at four different operation speeds with developing method.