Multi-fault diagnosis of rolling-element bearings in electric machines

This paper deals with the diagnosis of faults in roller-element bearings as the core of a dedicated Condition based Maintenance (CBM) system. Vibration signals recorded by accelerometers feed into a classification model in charge of monitoring and evaluating bearing wear. The chosen feature extraction technique is based on the computation of the Discrete Fourier Transform (DFT) and on the estimation of the normalized frequency content in each of the considered spectrum sub-bands as an indicative measure of the state of health of the roller-element bearing. Three different damage attributes have been investigated. For each attribute, both Support Vector Machines (SVM) and neurofuzzy Min-Max classifiers have been employed as the core of the diagnostic system. Test results show that it is possible to achieve high accuracy in all diagnostic problems considered. The pre-processing procedure and the classification stage, especially in the case of Min-Max fuzzy networks, do not require demanding computational hardware resources, and as a result, a simple and effective diagnostic system can be designed by feeding the synthesized Min-Max classifiers with the spectral features computed from vibration sensor outputs.