Temperature-dependent discovery of BCC refractory multi-principal element alloys: Integrating deep learning and CALPHAD calculations

TL;DR

This paper introduces a deep learning surrogate model trained on CALPHAD data to rapidly predict temperature-dependent phase fractions in BCC refractory multi-principal element alloys, significantly accelerating alloy design and discovery.

Contribution

We develop a high-accuracy deep learning model that preserves thermodynamic fidelity and speeds up phase prediction by two orders of magnitude, enabling efficient exploration of alloy compositions.

Findings

Model achieves high accuracy in phase fraction predictions across temperatures.

Screened multiple elemental combinations to identify stable BCC alloys.

Provided insights for experimental synthesis of new BCC RMPEAs.

Abstract

Single-phase body-centered cubic (BCC) refractory multi-principal element alloys (RMPEAs) offer potential for developing alloys with exceptional strength. However, the compositional design space is immense. Exhaustively mapping this space with conventional CALculation of PHAse Diagrams (CALPHAD) is impractical because database coverage and run times scale poorly with millions of candidate chemistries. To address this, we train a deep-learning surrogate on CALPHAD outputs that preserves the thermodynamic fidelity while accelerating temperature-dependent phase-fraction predictions of RMPEA phases. The model achieves high accuracy in predicting phase fractions for up to eight distinct phases across different temperatures and offers a speedup of two orders of magnitude compared to CALPHAD. Using this model, we screen the Ti, Fe, Al, V, Ni, Nb and Zr elemental space for potentially stable single-phase BCC alloys at different annealing temperatures and extract design insights to guide the synthesis of new BCC RMPEAs in experiments. Finally, we develop an analytical model that enables rapid, interpretable identification of single-phase BCC RMPEAs.