Multi-Objective Optimization Employing Knowledge Graph-Embedded Large Language Model to Strategize Battery Recycling Technology Selection

This study presents two AI-driven frameworks to support sustainable battery recycling: (1) the Battery-LLaMA module, which integrates a fine-tuned LLaMA-2-7B model with a battery recycling-informed knowledge graph (BR-KG) for domain-specific question answering, and (2) a multi-objective optimization (MOO) framework for process selection based on internal energy consumption at the molecular level and revenue generation from the recycling process. The BR-KG, built from over 10k+ Elsevier abstracts using named entity recognition, captures structured recycling knowledge and enhances the performance of the Battery-LLaMA model (F1-score: 0.821 vs. 0.701 for ChatGPT-4o). The MOO framework reveals that hydrometallurgical processes, particularly those combining leaching, roasting, and regeneration, offer the most economically and energetically favorable pathways for recycling 100 kg of Li-ion batteries with >90% efficiency.