The evolved kurtogram: a novel repetitive transients extraction method for bearing fault diagnosis

Kurtogram, a classic repetitive transients extraction method, plays an important role in bearing fault diagnosis. However, its performance is unstable since its index used for optimal sub-band component selection is sensitive to random pulse. Moreover, its sub-band component extraction is characterized by over-decomposition and under-decomposition defects. In this paper, an evolved Kurtogram (Evkurtogram) is proposed by designing a new index called the Gaussian distribution assigned Gini index (GDAG) for optimal sub-band component identification. In addition, a multi-scale empirical Fourier decomposition (MSEFD) for signal separation is proposed. GDAG is more suitable for quantifying the fault features of the signal due to its robustness of accidental pulses. MSEFD can achieve multi-scale decomposition of the signal reasonably and adaptively. The proposed Evkurtogram is compared with some relevant state-of-art algorithms by processing simulated and experimental bearing fault signals. It is demonstrated that the proposed Evkurtogram is effective and superior when compared to other approaches.