A Hybrid Signal Processing Technique for Bearing Defect Severity Estimation

This paper presents a hybrid signal processing technique for bearing defect feature extraction and severity estimation. This is achieved by decomposing vibration signals measured on multiple bearings with different defect conditions into multiple sub-bands by means of the wavelet packet transform (WPT). Representative statistical features for each sub-band are then calculated. Subsequently, Principal Component Analysis (PCA) is performed on the statistical features to choose the best-suited representative features as inputs to a diagnostic classifier for bearing health diagnosis.