Automated extraction of materials system charts using a large language model framework

Abstract

A framework leveraging large language models (LLMs) is developed to systematically extract and organize Processing-Mechanism-Structure-Mechanism-Property (P-M-S-M-P) relationships from materials science and engineering literature, with a particular focus on metallurgy. Using multi-stage prompts, our method identifies key properties, microstructures, processing methods, and associated mechanisms, then integrates them to generate comprehensive materials system charts. Additionally, the framework refines the extracted system charts for visualization, enabling the creation of informative diagrams that capture essential insights from each paper. Evaluated across 70 papers spanning multiple alloy systems and research types, the approach achieves 94 % accuracy in mechanism extraction, 87 % in information source labeling, and 97 % in the human-machine readability index for processing, structure, and property entities. The prompts and codes are provided alongside guidelines for researchers unfamiliar with coding. This framework offers an effective methodology for knowledge extraction in materials science using the P-M-S-M-P framework.