Feature extraction enhancement based on parameterless empirical wavelet transform: Application to bearing fault diagnosis

Rolling-element bearings are usually subject to faults that need prompt
detection in order to prevent sudden failures. Many time-frequency analysis
techniques have been used for the purpose of bearing fault detection and
diagnosis. From these techniques, wavelets and empirical mode
decomposition (EMD) stand out as the most widely applied methods in
bearing fault diagnosis. Recently, a novel method named the parameterless
empirical wavelet transform (PEWT) has been proposed to combine the
wavelet formulation with the adaptability of the empirical mode
decomposition. In this paper, the parameterless empirical wavelet transform
(PEWT) is combined with envelope detection (ED) to present a new scheme
named PEWT-ED for non-stationary signal analysis. The capabilities and
limitations of the new method in bearing fault diagnosis are investigated
using simulation and experiment. The results show that the new approach
can effectively extract the bearing fault characteristics. The PEWT-ED is
found to be a powerful tool in signal de-noising and enhancement for fault
diagnosis purposes.