Bearing Fault Diagnosis Using a Particle Swarm Optimization-Least Squares Wavelet Support Vector Machine Classifier

Area of Science:

• Mechanical Engineering
• Machine Learning
• Signal Processing

Background:

• Bearing health monitoring is critical for rotating machinery.
• Early fault detection prevents catastrophic failures and downtime.
• Existing methods may lack precision in complex fault diagnosis.

Purpose of the Study:

• To develop a novel classifier for accurate bearing fault diagnosis.
• To enhance classification precision using a new wavelet kernel function.
• To improve the performance of bearing health monitoring systems.

Main Methods:

• Feature extraction using Nonlocal Means (NLM) and Empirical Mode Decomposition (EMD).
• Feature selection via Minimum Redundancy Maximum Relevance (mRMR).
• Classification using a Particle Swarm Optimization-Least Squares Wavelet Support Vector Machine (PSO-LSWSVM).

Main Results:

• The proposed PSO-LSWSVM achieved higher classification accuracy.
• Demonstrated effectiveness on a benchmark bearing dataset.
• Outperformed existing methods in bearing fault diagnosis.

Conclusions:

• The novel PSO-LSWSVM classifier offers a significant advancement in bearing fault diagnosis.
• The integration of PSO, least squares, and wavelet kernel SVM is effective.
• This approach enhances the reliability of rotating machinery health monitoring.