Predicting Coherent B2 Stability in Ru-Containing Refractory Alloys Through Thermodynamic Elastic Design Maps

TL;DR

This paper introduces a physics-guided machine learning framework combining DFT, Random Forest, and Symbolic Regression to design Ru-based B2 alloys with high thermodynamic stability and minimal elastic strain, enabling precise microstructure engineering.

Contribution

It develops a novel integrated approach that predicts and optimizes alloy stability and elastic properties, revealing the importance of multi-component alloying for strain management.

Findings

A physical law quantifies the strain penalty effect on solvus temperature.

Maximizing thermodynamic stability alone is insufficient for alloy design.

Multi-component alloying with elements like Al and Ti can eliminate elastic strain penalties.

Abstract

Ruthenium-based B2 intermetallics are promising for refractory superalloys but are limited by the trade-off between high thermodynamic stability and elastic precipitation strain. We present a physics-guided machine learning framework integrating high-throughput Density Functional Theory (DFT), Random Forest screening, and Symbolic Regression to navigate this design space. This approach resolves the paradox where stoichiometric compounds like RuHf fail to achieve theoretical solvus temperatures. By deriving a closed-form physical law, we quantify the strain penalty: a 1% lattice misfit reduces the solvus temperature by approximately 200 degrees C. This finding confirms that maximizing thermodynamic driving force alone is insufficient. We demonstrate that multi-component alloying is structurally necessary, identifying ternary additions such as Al and Ti as essential lattice-tuning agents that zero out the elastic penalty. This framework establishes a rigorous, constraint-based protocol for alloy design, enabling the precise engineering of zero-misfit, high-stability microstructures.