Phase Stability Through Machine Learning

Abstract

Understanding the phase stability of a chemical system constitutes the foundation of materials science. Knowledge of the equilibrium state of a system under arbitrary thermodynamic conditions provides valuable information about the types of phases that are likely to be synthesized and how to get there. Accessing the phase diagram in a materials system provides one with the information necessary to design materials and microstructures with optimal properties. While the materials science community has long been focused on exploiting this knowledge to navigate the materials space, recent advances in machine learning (ML) and artificial intelligence (AI) have provided the community with novel ways of interrogating the materials thermodynamics space. This work presents some of the most recent advances in ML/AI applied to phase stability and thermodynamics of materials. Prof. John Morral always had a passion for understanding and teaching the fundamental characteristics of phase diagrams. This review is written to honor his memory.

Design and Selection of High Entropy Alloys for Hardmetal Matrix Applications Using a Coupled Machine Learning and Calculation of Phase Diagrams Methodology

“…Several studies utilizing ML for the design of HEAs have been reported in the literature and the reader is referred to the comprehensive reviews by Liu et al [39,40] Arróyave [41] also provided a detailed discussion on ML for phase prediction and stability. The majority of early ML studies focus on the prediction of a single property, either phase formation or hardness.…”

Chemistry Informed Machine Learning-Based Heat Capacity Prediction of Solid Mixed Oxides

“…Another option is to incorporate information related to changes and instability in the system phases into the models. There has been work on the prediction of phase diagrams and phase stability using ML methods, which could pave the way for the creation of phase change-informed models for heat capacity prediction, , which will be explored in the future. Nonetheless, the current ML approach already predicts C p ( T , x ) with an error that is less than the experimental uncertainty and state-of-the-art models (ideal, semiempirical), presenting a significant advance and useful tool for the screening and C p prediction for mixed oxides, which can be generalized to higher-order systems.…”