A New Compound Fault Feature Extraction Method Based on Multipoint Kurtosis and Variational Mode Decomposition

Due to the weak entropy of the vibration signal in the strong noise environment, it is very difficult to extract compound fault features. EMD (Empirical Mode Decomposition), EEMD (Ensemble Empirical Mode Decomposition) and LMD (Local Mean Decomposition) are widely used in compound fault feature extraction. Although they can decompose different characteristic components into each IMF (Intrinsic Mode Function), there is still serious mode mixing because of the noise. VMD (Variational Mode Decomposition) is a rigorous mathematical theory that can alleviate the mode mixing. Each characteristic component of VMD contains a unique center frequency but it is a parametric decomposition method. An improper value of K will lead to over-decomposition or under-decomposition. So, the number of decomposition levels of VMD needs an adaptive determination. The commonly used adaptive methods are particle swarm optimization and ant colony algorithm but they consume a lot of computing time. This paper proposes a compound fault feature extraction method based on Multipoint Kurtosis (MKurt)-VMD. Firstly, MED (Minimum Entropy Deconvolution) denoises the vibration signal in the strong noise environment. Secondly, multipoint kurtosis extracts the periodic multiple faults and a multi-periodic vector is further constructed to determine the number of impulse periods which determine the K value of VMD. Thirdly, the noise-reduced signal is processed by VMD and the fault features are further determined by FFT. Finally, the proposed compound fault feature extraction method can alleviate the mode mixing in comparison with EEMD. The validity of this method is further confirmed by processing the measured signal and extracting the compound fault features such as the gear spalling and the roller fault, their fault periods are 22.4 and 111.2 respectively and the corresponding frequencies are 360 Hz and 72 Hz, respectively