Simultaneously Low Rank and Group Sparse Decomposition for Rolling Bearing Fault Diagnosis

Area of Science:

• Mechanical Engineering
• Signal Processing
• Condition Monitoring

Background:

• Singular Value Decomposition (SVD) is widely used for bearing fault diagnosis by extracting periodic impulses.
• Existing SVD methods primarily rely on the low-rank property of Hankel matrices, which limits performance in high-noise environments.
• Effective bearing fault diagnosis is crucial for preventing catastrophic failures in rotating machinery.

Purpose of the Study:

• To introduce a novel decomposition method for bearing fault diagnosis that overcomes the limitations of traditional SVD approaches.
• To reveal and exploit the simultaneously low rank and group sparse (SLRGS) property of Hankel matrices for fault features.
• To enhance the extraction of incipient fault features in bearing diagnostics, especially under strong background noise.

Main Methods:

• Proposed a Simultaneously Low Rank and Group Sparse Decomposition (SLRGSD) method for bearing fault diagnosis.
• Formulated a regularization model based on the SLRGS property of the Hankel matrix for fault features.
• Employed an incremental proximal algorithm to achieve a stationary solution for the decomposition model.

Main Results:

• The SLRGSD method demonstrated superior performance in extracting incipient fault features compared to state-of-the-art methods.
• Enhanced fault feature extraction was validated through numerical analysis, artificial bearing fault experiments, and wind turbine bearing fault experiments.
• The method showed significant improvements in both performance metrics and visual quality of the extracted features.

Conclusions:

• The SLRGSD method effectively enhances bearing fault diagnosis by leveraging the SLRGS property of Hankel matrices.
• This approach offers a robust solution for detecting early-stage bearing faults, even in the presence of substantial noise.
• The findings suggest SLRGSD as a promising technique for condition monitoring and predictive maintenance in critical machinery.