A machine learning framework for discovering high entropy alloys phase formation drivers.

Area of Science:

• Materials Science
• Computational Materials Science
• Alloy Design

Background:

• High entropy alloys (HEAs) are recognized for their exceptional properties, driving significant research interest.
• Machine learning (ML) has emerged as a powerful tool for phase prediction in HEAs, achieving high accuracy.
• Identifying the fundamental drivers of phase formation in HEAs remains a critical challenge, with limited exploration beyond domain-specific feature engineering.

Purpose of the Study:

• To develop an ML-based approach for predicting HEA phases solely from elemental concentrations.
• To identify the key physical and chemical variables that govern phase formation in HEAs.
• To establish simple, interpretable rules for predicting specific phase formations, such as FCC in the AlCoCrFeNiTiCu family.

Main Methods:

• Utilized machine learning models for phase prediction based exclusively on alloy composition.
• Employed pruned tree models and linear correlation to derive predictive rules.
• Formulated phase formation driver discovery as an optimization problem using self-organizing maps (SOMs) and Euclidean spaces.
• Applied genetic algorithm (GA) optimization to identify influential elemental properties.

Main Results:

• Phase prediction accuracy reached 87% for FCC phase formation in the AlCoCrFeNiTiCu alloy family using only Al and Cu concentrations.
• Genetic algorithm optimization identified electron affinity, molar volume, and resistivity as significant factors influencing HEA phase formation.
• Developed simple, primitive prediction rules derived from elemental concentrations.

Conclusions:

• Elemental concentrations alone can effectively predict HEA phases, simplifying the prediction process.
• Key intrinsic properties of constituent elements (electron affinity, molar volume, resistivity) are critical drivers of phase formation.
• The developed methodology offers a novel approach to understanding and optimizing HEA phase stability and design.