Multi-fault diagnosis of rotating machinery by using feature ranking methods and SVM-based classifiers

Rotating machinery plays an important role in industries for motion transmission in machines; the breakdowns of gearboxes are mostly produced by gear and bearings failures. Thus, some strategies are sought to avoid unscheduled stops, or catastrophic damages, in order to reduce maintenance costs and increase reliability. This paper describes a methodological framework to detect eleven rotating machinery faults by using feature ranking methods and support vector machine, based on information that comes from the measured vibration signal. Thirty features are calculated from the vibration signal in time domain, for each faulty condition. Feature ranking methods such as ReliefF, Chi square, and Information Gain are used to select the most informative features, and subsequently to reduce the size of the feature vector. The feature ranking methods are compared in order to obtain improved diagnosis results with a reduced feature set. Results show good fault identification accuracy with the first four features of ReliefF ranking method as input to support vector machine classifier.