Large language models for prognostic analysis in mechanical fault diagnosis

Area of Science:

• Industrial Intelligence
• Mechanical Engineering
• Artificial Intelligence

Background:

• Rotating machinery is crucial for complex systems, necessitating robust fault diagnosis and health management.
• Traditional vibration signal analysis faces limitations in complex systems, struggling with comprehensive fault feature extraction and generalization.

Purpose of the Study:

• To propose an intelligent diagnosis framework leveraging large language models (LLMs) for rotating machinery.
• To overcome the limitations of traditional methods by enhancing feature extraction and cross-scenario generalization.

Main Methods:

• Developed an LLM-based framework integrating multimodal data: raw vibration signals, time-frequency spectrum features, and fault knowledge text.
• Employed feature fusion for cross-modal joint representation of mechanical fault features.
• Integrated principle knowledge bases for enhanced diagnostic capabilities.

Main Results:

• The framework demonstrated superior performance in fault diagnosis and health management compared to traditional methods.
• Achieved excellent performance and adaptability across different industrial scenarios on bearing datasets.
• The model successfully outputs fault location, cause analysis, and maintenance strategy suggestions.

Conclusions:

• The proposed LLM-based framework offers a significant advancement in rotating machinery fault diagnosis and health management.
• Multimodal data fusion and LLM integration effectively address the limitations of traditional signal analysis.
• The framework provides interpretable reports, enhancing practical applicability in industrial settings.