Sensitive sub-band selection criteria for empirical wavelet transform to detect bearing fault based on vibration signals

Under varying speed, a bearing exhibits a severely modulated signal, to which the classical techniques of time and frequency domain fails to detect. An experimental investigation exhibiting detection of bearing faults at both constant and varying speed using empirical wavelet transform (EWT) is presented. EWT employs a filter bank to decompose a non-stationary signal into different sub-bands. The selection of sub-band influences fault detection because residual noises remain unfiltered. Therefore, identification of sensitive sub-band is crucial for a detailed investigation, which is done using correlated correlation and kurtosis. Upon selection of sensitive sub-band the envelope spectrum was used to detect the presence of fault based on the frequencies noticed. The results were validated by permutation entropy. Two case studies: in-house bearing vibration signal at a constant speed and, online available bearing data set at varying speed, were performed to exhibit the promising result of the proposed approach to detect the bearing fault.