Accelerated Discovery of Refractory High-Entropy Alloys via Interpretable Machine Learning

Area of Science:

• Materials Science
• Computational Materials Science
• Alloy Design

Background:

• Refractory high-entropy alloys (RHEAs) are critical for extreme environments due to their thermal stability and strength.
• Existing design methods for RHEAs are often computationally intensive and slow.
• There is a need for efficient and accurate methods to accelerate RHEA discovery.

Purpose of the Study:

• To develop an integrated computational framework for designing RHEAs.
• To enhance the accuracy and reduce the cost of atomic-scale simulations for RHEAs.
• To enable rapid prediction and interpretable analysis of RHEA performance.

Main Methods:

• Integration of machine learning potentials (UNEP-v1), supervised regression, and multiobjective optimization.
• High-throughput molecular dynamics simulations to generate property data (melting point, tensile strength).
• Application of interpretability techniques (SHAP, PDP, ALE, ICE) for model understanding.

Main Results:

• The framework significantly improves simulation accuracy while reducing computational cost.
• Accurate prediction of melting points and tensile strengths at 1000 K for various RHEA compositions.
• Successful validation in the TiVCrZrMo alloy system, demonstrating design of high-strength, high-temperature alloys.

Conclusions:

• The developed framework provides a precise and interpretable predictive modeling paradigm for RHEAs.
• This approach promotes the integration of atomic-scale simulations with data-driven methods.
• It accelerates the design and discovery of advanced RHEAs for demanding applications.