LLMB: AI Agent for Lithium Metal Battery Research Using Large Language Model

Area of Science:

• Materials Science
• Artificial Intelligence
• Electrochemistry

Background:

• Data-driven research offers significant potential for understanding material-performance relationships.
• Extracting comprehensive battery data from diverse sources remains a challenge.

Purpose of the Study:

• To introduce LLMB, an AI agent for lithium metal battery research.
• To enable accurate extraction of battery material data and cyclability performance metrics.
• To construct a comprehensive database for machine learning model development.

Main Methods:

• Utilizing a large language model (LLM) for hierarchical text mining.
• Employing an automatic graph mining tool, Material Graph Digitizer (MatGD).
• Integrating text and graph mining outputs to build a comprehensive battery cell database.

Main Results:

• Extracted composition and operating conditions for 15,398 battery cells via text mining.
• Obtained cyclability data for 10,242 cells through graph mining.
• Constructed a database of 8,074 cells, integrating component specifics and capacity.
• Developed a machine learning model to predict lithium metal battery capacities based on material information.
• Experimentally validated findings related to electrolyte properties, SEI formation, and Li plating.

Conclusions:

• LLMB agent successfully enables state-of-the-art data extraction for battery research.
• The developed machine learning model accurately predicts battery capacities.
• Weakly solvating electrolytes promote favorable SEI formation and Li plating, confirming model reliability.