Classification and diagnosis of mechanical faults using the RBF network based on the local bispectra

Abstract

This paper introduces an efficient technique to design a classifier for classifying and diagnosing mechanical faults. The bispectra features of vibration signals resulting from mechanical faults are extracted and then evaluated using the Fisher's class-separability discriminant measure. The local bispectra with the most discriminant power and sensitivity are selected as the classification feature vector that can effectively represent the fault class of concern over a broad range of sample data. A RBF neural network is implemented to realize identification and diagnosis for different mechanical faults. The suggested technique is demonstrated to design a classifier for fault signals of rolling bearings that is verified to be highly accurate and robust even in the presence of excessive noise. © Springer-Verlag Berlin Heidelberg 2007.