Early Fault Feature Extraction of Rolling Bearings Based on CEEMD and Adaptive MCKD

It is difficult to extract the early fault of rolling bearings and the noise reduction effects of maxim correlated kurtosis deconvolution (MCKD) are affected by filter length L. A rolling bearings fault diagnosis method was put forward based on complementary ensemble empirical mode decomposition (CEEMD) and adaptive maxim correlated kurtosis deconvolution (AMCKD). First, a set of intrinsic mode function (IMF) components were decomposed by CEEMD, and the kurtosis was then used to select the IMF components that need to reduce noise. Then, the step size search method was used to select the best MCKD filter length according to the permutation entropy, and the components were processed by adaptive MCKD. Finally, the denoised components and other components were reconstructed, and the fault characteristic frequency was extracted according to the envelope power spectrum of the signal. The simulation signal and experimental data prove the effectiveness and advantages of the proposed method.