A Spectral Interpretable Bearing Fault Diagnosis Framework Powered by Large Language Models

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Existing fault diagnosis methods often act as black boxes, lacking transparent reasoning.
• This limits user trust and understanding of diagnostic outcomes in industrial settings.

Purpose of the Study:

• To develop a novel, interpretable fault diagnosis framework.
• To integrate spectral feature extraction with large language models (LLMs) for transparent diagnostic reasoning.

Main Methods:

• Vibration signals transformed into spectral representations using Hilbert- and Fourier-based encoders.
• Channel attention-augmented convolutional neural network (CNN) for initial fault prediction.
• Fine-tuned LLM integrates spectral features and CNN outputs for diagnosis and reasoning.

Main Results:

• The framework achieves high diagnostic performance.
• Substantially improves the interpretability of fault diagnosis.
• Demonstrates accurate diagnosis with transparent reasoning.

Conclusions:

• The proposed framework enhances trust and understanding in fault diagnosis.
• Makes advanced fault diagnosis accessible to non-expert industrial users.
• Offers a transparent and accurate alternative to black-box diagnostic models.