Feature Extraction of Gear Tooth Surface Fatigue Failure in Reducers Based on Vibration Signals

Extracting periodic fault pulses caused by gear surface fatigue in reducers is often hindered by transmission path interference and strong background noise. Moreover, the traditional Variational Mode Decomposition (VMD) and Maximum Correlation Kurtosis Decomposition (MCKD) method rely on manual parameter selection, which limits its practicality. To address these issues, this paper proposes a parameter-adaptive VMD-MCKD method based on vibration signals for extracting gear surface fatigue fault features. Using the reciprocal of the peak indicator squared of decomposed signals as fitness functions, the method employs the global search capability of the Sparrow Search Algorithm to adaptively select optimal VMD-MCKD configurations. The optimized VMD-MCKD method is applied to decompose gear surface fatigue fault signals, effectively filtering out noise while highlighting periodic fault pulses caused by gear fatigue. Envelope demodulation is then performed to extract characteristic frequency components of gear surface fatigue faults. Experimental results demonstrate that the proposed method can adaptively extract periodic fault pulse components from strong noise environments, achieving a 2-fold improvement in signal kurtosis and enhanced robustness.