Intelligent Material Framework for Data Mining and Performance Prediction of Invar Alloy

ABSTRACT Invar alloys exhibit a low coefficient of thermal expansion (CTE) over a wide temperature range and are regarded as critical materials for precision engineering and aerospace applications. In this work, an integrated OLR‐IMF (an intelligent material framework based on optical character recognition [OCR], large language models [LLMs], and retrieval‐augmented generation [RAG]) is developed for the CTE prediction of Invar alloys. This framework synergizes OCR, LLMs, and RAG core technologies to construct a closed‐loop workflow, which encompasses literature acquisition, structured data extraction, materials information mining, and interpretable prediction. Within this framework, the structured data extraction step achieves an F 1‐score of 89.4 (a standard metric for the accuracy and completeness of literature data extraction) based on which 120 data entries of as‐cast Invar alloys were extracted from 662 Invar alloy‐related publications, and 323 potential machine learning features were collated through two rounds of RAG‐based QA. Feature selection was performed using a genetic algorithm and an optimal subset was identified, yielding a prediction coefficient of determination ( R 2 ) of 0.935 after 10‐fold cross‐validation. The approach provides a generalizable pathway for intelligent research and development in materials science.