Fault diagnosis of nonlinear analog circuits using generalized frequency response function and LSSVM

Area of Science:

• Electrical Engineering
• Circuit Analysis
• Machine Learning Applications

Background:

• Nonlinear analog circuits require robust fault diagnosis methods.
• Existing methods may face challenges with accuracy and efficiency.
• Accurate fault identification is crucial for system reliability.

Purpose of the Study:

• To propose a novel fault diagnosis method for nonlinear analog circuits.
• To enhance diagnostic accuracy and reduce training time.
• To develop an efficient multi-fault classifier.

Main Methods:

• Estimating generalized frequency response functions (GFRFs) using harmonic signal input.
• Directly solving GFRFs from time-domain data, avoiding Fourier transforms.
• Designing a multi-fault classifier using a simplified least squares support vector machine (LSSVM).
• Employing the conjugate gradient algorithm for efficient LSSVM training.

Main Results:

• The proposed method accurately estimates GFRFs from time-domain data.
• The simplified LSSVM classifier demonstrates high diagnostic accuracy.
• The conjugate gradient algorithm significantly reduces training time and storage requirements.
• Simulation experiments on a biquad filter circuit validate the method's effectiveness.

Conclusions:

• The combined GFRF and simplified LSSVM approach offers a powerful tool for nonlinear analog circuit fault diagnosis.
• The method achieves high accuracy with improved efficiency.
• This technique is suitable for real-time fault detection and system maintenance.