Fault classification for induction motor using Hilbert-based bispectral analysis and probabilistic neural networks

This paper addresses the development of a new signal processing approach based on the fusion of Hilbert transform and bispectral analysis to extract features of defects in a number of induction motor conditions. The motor conditions considered are a normal motor and motors with outer, inner race and rotor faults. The signal processing techniques based on Hilbert transform have been used to extract the modulating components which are able to characterize the motor fault patterns. The use of bispectral analysis provides great capabilities for detection and characterization of nonlinearity in the motor vibration systems. Feature selection based on principal component analysis are used to extract from the vibration signatures so obtained and these features are used as inputs to probabilistic neural networks trained to identify the motor conditions. The results obtained show that the diagnostic system using a supervised radial basis type neural network is capable of classifying motor conditions with high accuracy recognition rate.