Period-oriented multi-hierarchy deconvolution and its application for bearing fault diagnosis

Deconvolution methods have been proven to be effective tools to extract excitation sources from the noisy measured signal. However, its application is confined by the extraction of incomplete information. To tackle this problem, a new deconvolution method, named period-oriented multi-hierarchy deconvolution (POMHD) is proposed in this paper. Various filters are designed adaptively by the iterative algorithm to update the filter coefficient using the harmonic-to-noise ratio as the deconvolution orientation. Additionally, a novel index, called normalized proportion of harmonics, is proposed as the evaluation criteria for the fault feature. Based on upon, a harmonics proportion diagram is constructed for the diagnostic decisions. The new deconvolution method overcomes the disadvantages of the traditional methods. More importantly, without an accurate fault period as the prior knowledge, the proposed POMHD can simultaneously extract multiple latent fault components by using the adaptive filter and intuitively present different fault information in one diagram. Finally, the simulated and experimental data which includes the signals collected from bearings with both single faults and compound faults is used to evaluate the new method. The results validate the feasibility and robustness of the proposed POMHD.