Machine learning-enabled high-entropy alloy discovery.

Area of Science:

• Materials Science
• Metallurgy
• Computational Materials Science

Background:

• High-entropy alloys (HEAs) offer unique properties inaccessible to conventional materials.
• Designing HEAs is challenging due to vast compositional spaces and limitations of traditional thermodynamic rules.
• Discovering HEAs with specific properties often relies on serendipity.

Purpose of the Study:

• To accelerate the design and discovery of high-entropy Invar alloys.
• To develop an active learning strategy for navigating complex compositional landscapes.
• To identify HEAs with exceptionally low thermal expansion coefficients.

Main Methods:

• Integrated active learning with density-functional theory, thermodynamic calculations, and experimental validation.
• Employed a closed-loop approach for iterative material design and characterization.
• Screened millions of potential compositions using machine learning on sparse data.

Main Results:

• Identified two novel high-entropy Invar alloys.
• Achieved extremely low thermal expansion coefficients (approx. 2 × 10^-6 K^-1 at 300 K).
• Demonstrated the efficacy of the active learning strategy in a high-dimensional space.

Conclusions:

• The proposed active learning strategy enables fast and automated discovery of HEAs.
• This approach is suitable for optimizing thermal, magnetic, and electrical properties.
• The identified alloys represent significant advancements in low thermal expansion materials.