Reweighted generalized minimax-concave sparse regularization and application in machinery fault diagnosis

Area of Science:

• Mechanical Engineering
• Signal Processing
• Computational Mathematics

Background:

• Vibration signals from faulty rotating machinery contain complex mixtures of repetitive transients, discrete frequencies, and noise.
• Accurate extraction of repetitive transients is crucial for effective machinery fault diagnosis.
• Existing methods like L1 norm penalty can underestimate signal sparsity.

Purpose of the Study:

• To propose a novel sparse regularization method, reweighted generalized minimax-concave (ReGMC), for enhanced extraction of repetitive transients.
• To overcome the underestimation deficiency of L1 norm penalty using a generalized minimax-concave (GMC) penalty.
• To introduce a new reweight strategy based on squared envelope spectrum kurtosis for improved sparsity.

Main Methods:

• Developed the reweighted generalized minimax-concave (ReGMC) sparse regularization method.
• Utilized the generalized minimax-concave (GMC) penalty within a weighted sparse representation model.
• Implemented a novel reweight strategy informed by squared envelope spectrum kurtosis.
• Processed simulated and real-world vibration signals from gearboxes and bearings.

Main Results:

• ReGMC effectively extracts repetitive transients from complex vibration signals.
• The method successfully suppresses discrete frequency components and background noise.
• Comparative studies demonstrate superior performance of ReGMC over GMC, improved lasso, and spectral kurtosis.
• Validated effectiveness on simulated data and experimental data from a milling gearbox and a bearing.

Conclusions:

• The proposed ReGMC method offers a significant advancement in extracting repetitive transients for machinery fault diagnosis.
• ReGMC provides a robust approach to handling mixed signals, outperforming existing techniques.
• This method enhances the accuracy and reliability of diagnosing incipient faults in rotating machinery.