Advancing battery failure diagnosis by knowledge-augmented large language models

Area of Science:

• Materials Science
• Computer Science
• Energy Storage

Background:

• Battery failure diagnosis is critical for energy storage safety and reliability.
• Current methods like electrochemical modeling and deep learning struggle with data dependency, generalization, and interpretability.
• Limitations hinder effective battery health monitoring and predictive maintenance.

Purpose of the Study:

• To develop a novel framework, BattFailScholar, for enhanced battery failure diagnosis.
• To address limitations of existing diagnostic approaches using knowledge augmentation.
• To improve the accuracy and reliability of identifying battery failure mechanisms and causes.

Main Methods:

• Constructed a case-level battery failure knowledge graph integrating material properties, multi-source signals, and failure pathways.
• Developed a knowledge-augmented generation method for LLM diagnostic reasoning.
• Employed failure feature-aware retrieval and optimization algorithms to enhance diagnostic capabilities.

Main Results:

• Achieved a 19.7% performance improvement in LLM-based battery failure diagnosis.
• Demonstrated enhanced capability in addressing long-tail failure problems and improving failure risk assessment.
• Attained 86.2% accuracy in identifying potential failure mechanisms or causes.

Conclusions:

• BattFailScholar offers a robust solution for battery failure diagnosis, overcoming limitations of traditional methods.
• The framework shows significant potential for discovering complex failure chains and providing practical diagnostic support.
• This knowledge-augmented LLM approach advances the safety and reliability of energy storage systems.