Explainable Machine Learning for Hydrogen Diffusion in Metals and Random Binary Alloys
Authors:
Grace M. Lu,
Matthew Witman,
Sapan Agarwal,
Vitalie Stavila,
Dallas R. Trinkle
Abstract:
Hydrogen diffusion in metals and alloys plays an important role in the discovery of new materials for fuel cell and energy storage technology. While analytic models use hand-selected features that have clear physical ties to hydrogen diffusion, they often lack accuracy when making quantitative predictions. Machine learning models are capable of making accurate predictions, but their inner workings…
▽ More
Hydrogen diffusion in metals and alloys plays an important role in the discovery of new materials for fuel cell and energy storage technology. While analytic models use hand-selected features that have clear physical ties to hydrogen diffusion, they often lack accuracy when making quantitative predictions. Machine learning models are capable of making accurate predictions, but their inner workings are obscured, rendering it unclear which physical features are truly important. To develop interpretable machine learning models to predict the activation energies of hydrogen diffusion in metals and random binary alloys, we create a database for physical and chemical properties of the species and use it to fit six machine learning models. Our models achieve root-mean-squared-errors between 98-119 meV on the testing data and accurately predict that elemental Ru has a large activation energy, while elemental Cr and Fe have small activation energies. By analyzing the feature importances of these fitted models, we identify relevant physical properties for predicting hydrogen diffusivity. While metrics for measuring the individual feature importances for machine learning models exist, correlations between the features lead to disagreement between models and limit the conclusions that can be drawn. Instead grouped feature importances, formed by combining the features via their correlations, agree across the six models and reveal that the two groups containing the packing factor and electronic specific heat are particularly significant for predicting hydrogen diffusion in metals and random binary alloys. This framework allows us to interpret machine learning models and enables rapid screening of new materials with the desired rates of hydrogen diffusion.
△ Less
Submitted 26 October, 2023; v1 submitted 15 August, 2023;
originally announced August 2023.
Inverse Design of Ultralow Lattice Thermal Conductivity Materials Via Lone Pair Cation Coordination Environment
Authors:
Eric B. Isaacs,
Grace M. Lu,
Christopher Wolverton
Abstract:
The presence of lone pair (LP) electrons is strongly associated with the disruption of lattice heat transport, which is a critical component of strategies to achieve efficient thermoelectric energy conversion. By exploiting an empirical relationship between lattice thermal conductivity $κ_L$ and the bond angles of pnictogen group LP cation coordination environments, we develop an inverse design st…
▽ More
The presence of lone pair (LP) electrons is strongly associated with the disruption of lattice heat transport, which is a critical component of strategies to achieve efficient thermoelectric energy conversion. By exploiting an empirical relationship between lattice thermal conductivity $κ_L$ and the bond angles of pnictogen group LP cation coordination environments, we develop an inverse design strategy based on a materials database screening to identify chalcogenide materials with ultralow $κ_L$ for thermoelectrics. Screening the $\sim$ 635,000 real and hypothetical inorganic crystals of the Open Quantum Materials Database based on the constituent elements, nominal electron counting, LP cation coordination environment, and synthesizability, we identify 189 compounds expected to exhibit ultralow $κ_L$. As a validation, we explicitly compute the lattice dynamical properties of two of the compounds (Cu$_2$AgBiPbS$_4$ and MnTl$_2$As$_2$S$_5$) using first-principles calculations and successfully find both achieve ultralow $κ_L$ values at room temperature of $\sim$ 0.3--0.4 W/(m$\cdot$K) corresponding to the amorphous limit. Our data-driven approach provides promising candidates for thermoelectric materials and opens new avenues for the design of phononic properties of materials.
△ Less
Submitted 3 April, 2020;
originally announced April 2020.
