Wavelet packet transform-based hybrid signal processing for machine health monitoring and diagnosis

This paper presents a hybrid signal processing technique for machine health monitoring and diagnosis. Vibration signals measured from a spindle test bed with different defect conditions are first decomposed into multiple sub-frequency bands by means of the wavelet packet transform. The statistical parameters (energy and Kurtosis) of these sub-frequency bands are then calculated. Subsequently, Principal Feature Analysis, which is an extension of the Principle Component Analysis, is performed on the statistical parameters to choose the most representative features, which can be distinctively separated from each other, as inputs to a diagnostic classifier. Experimental analysis of sensor data measured from the test bed has verified the effectiveness of the developed technique.