Sparse representation theory for support vector machine kernel function selection and its application in high-speed...

Area of Science:

• Machine Learning
• Mechanical Engineering
• Signal Processing

Background:

• Support Vector Machines (SVMs) are powerful classification tools but their performance heavily relies on appropriate kernel function selection.
• Bearing fault diagnosis is critical for industrial machinery maintenance, demanding accurate and reliable detection methods.

Purpose of the Study:

• To propose and validate a novel kernel function selection mechanism for SVMs tailored for bearing fault diagnosis.
• To enhance the accuracy and robustness of SVM-based fault diagnosis through optimized kernel selection.

Main Methods:

• A comprehensive search of 125,150 kernel function types and parameters was performed.
• Sparse representation and the Orthogonal Matching Pursuit (OMP) algorithm were employed for kernel selection.
• Particle Swarm Optimization (PSO) was used to fine-tune parameters for classifying unknown data.

Main Results:

• The proposed mechanism successfully identified the most suitable kernel function and parameters for given datasets.
• The SVM with the selected kernel demonstrated superior performance in high-speed bearing fault diagnosis simulations and experiments.

Conclusions:

• The developed kernel function selection mechanism offers a superior approach for SVM-based bearing fault diagnosis.
• This method provides a robust and accurate solution for identifying faults in rotating machinery.