Minimum entropy morphological deconvolution and its application in bearing fault diagnosis

Abstract

The bearing fault signal can be seen as convolution of periodical impulses and interference components. The minimum entropy deconvolution (MED) is effective approach for the deconvolution of signal. However, the MED is vulnerable to random impulse and interference components. To solve the problem, an improved method, named minimum entropy morphological deconvolution (MEMD), is proposed in this paper. Firstly, the amplitude frequency response of two typical morphological operators (MOs) are discussed. These operators are then introduced into MED to filter the sample matrix. The optimal MO is selected based on the amplitude ratio of diagonal slice spectrum (DSS). Eventually, the filtered result is analyzed by DSS to identify the fault type. In MEMD, the influence of random shocks is eliminated and the scale of SE can be determined adaptively. The MEMD is verified by simulation and experimental signals. Comparison study is implemented and the analysis results verify its effectiveness and feasibility.