From Small Data Modeling to Large Language Model Screening: A Dual-Strategy Framework for Materials Intelligent...

Area of Science:

• Materials Science
• Machine Learning
• Computational Materials Design

Background:

• Small data sets in materials science impede accurate machine learning model development.
• This limitation hinders the adoption of data-driven intelligent design for new materials.

Purpose of the Study:

• Introduce the Dual-Strategy Materials Intelligent Design Framework (DSMID) to address small data and screening challenges.
• Enhance material characterization and property prediction with limited data.
• Streamline the identification and evaluation of numerous alloy candidates.

Main Methods:

• Adversarial domain Adaptive Embedding Generative network (AAEG) for data transfer and property prediction with minimal data (90 points).
• Automated Material Screening and Evaluation Pipeline (AMSEP) using large language models for efficient candidate identification.
• Integration of AAEG and AMSEP within the DSMID framework.

Main Results:

• Successful identification and preparation of a novel eutectic High Entropy Alloy (EHEA), Al14(CoCrFe)19Ni28.
• Achieved high performance: 1085 MPa tensile strength and 24% elongation in as-cast condition.
• Demonstrated superior plasticity and comparable strength to existing eutectic HEAs like AlCoCrFeNi2.1.

Conclusions:

• The DSMID framework effectively tackles small data limitations and extensive screening challenges in materials design.
• This approach reduces costs and increases efficiency in discovering new materials.
• Offers a viable path for intelligent material design, especially with scarce data.