Adaptive wavelet filtering for bearing monitoring based on block bootstrapping and white noise test

Abstract

This study describes a novel scheme of adaptive wavelet filtering for bearing monitoring based on block bootstrapping and white noise test. The scheme consists of three main steps. First, the vibration signal is decomposed into wavelet domain, and the correlations between the wavelet coefficients are measured by lag autocorrelations. Second, according to the intensity of correlation at each level, either the block bootstrapping or general bootstrapping procedure is adopted to produce new pseudo-samples from the original wavelet coefficient series. Finally, as actual signal and noise have different translating characters along the levels in wavelet domain, the optimal decomposition level is achieved through whitening test on the wavelet coefficients, and the accuracy of the test is also obtained by the pseudo-samples. The simulation and experimental results show that the proposed procedure can be used to adaptively determine the optimal decomposition level and obtain superior filtering capability.