VSMKD: A new deconvolution method and application to rolling bearing fault diagnosis

Abstract

The vibration signal of rolling bearing fault has significant periodic impulse characteristics, which is a crucial symbol for the fault diagnosis of rolling bearing. Commonly employed deconvolution methods use the invariable fault period as the basis for periodic selection. However, the actual fault period of the bearing vibration signal is not fixed and fluctuates randomly in a certain range, leading to the failure of these deconvolution methods in some condition. To overcome this limitation, this paper proposes a variable scale multipoint kurtosis deconvolution (VSMKD) method to extract bearing fault features accurately. A variable scale multipoint kurtosis is defined as target function to construct the optimal filter. A variable scale position weight vector is constructed and optimized in the fluctuation range to determine the fault fluctuation period. The fault characteristic frequency is extracted by analyzing the envelope of the deconvolution signal. The effectiveness of the proposed VSMKD is validated by experimental data which are difficult to diagnose. Results verify the advantages of the proposed VSMKD by comparing it with advanced methods.