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

Area of Science:

• Mechanical Engineering
• Signal Processing
• Condition Monitoring

Background:

• Extracting compound fault features from vibration signals in noisy environments is challenging due to weak signal entropy and mode mixing issues common in methods like EMD and EEMD.
• Variational Mode Decomposition (VMD) offers a more robust approach to mode mixing but requires adaptive determination of its decomposition level (K), as common optimization methods are computationally intensive.

Purpose of the Study:

• To propose and validate a novel compound fault feature extraction method that overcomes the limitations of existing techniques, particularly in strong noise conditions.
• To develop an adaptive VMD decomposition level determination strategy that is computationally efficient.

Main Methods:

• The proposed method utilizes Minimum Entropy Deconvolution (MED) for initial noise reduction.
• Multipoint Kurtosis (MKurt) is employed to identify periodic faults and construct a multi-periodic vector for adaptively determining the VMD decomposition level (K).
• The noise-reduced signal is then processed using VMD, and fault features are extracted via Fast Fourier Transform (FFT).

Main Results:

• The Multipoint Kurtosis (MKurt)-VMD method effectively alleviates mode mixing compared to Ensemble Empirical Mode Decomposition (EEMD).
• The method successfully extracts compound fault features, such as gear spalling (22.4 period, 360 Hz) and roller faults (111.2 period, 72 Hz), from measured signals.
• The adaptive K determination based on MKurt is computationally efficient.

Conclusions:

• The proposed MKurt-VMD method provides a superior approach for compound fault feature extraction in challenging noisy environments.
• This technique enhances the accuracy and reliability of fault diagnosis in rotating machinery.
• The adaptive VMD decomposition level determination offers a practical advantage for real-world applications.