Machine learning to explore high-entropy alloys with desired enthalpy for room-temperature hydrogen storage: Prediction of density functional theory and experimental data

TL;DR

This paper demonstrates how machine learning, specifically Gaussian process regression, can predict hydride formation enthalpies in high-entropy alloys, aiding the design of materials suitable for room-temperature hydrogen storage.

Contribution

It introduces a machine learning approach to predict hydride formation enthalpies in high-entropy alloys, accelerating the design process for hydrogen storage materials.

Findings

ML predictions align with experimental and DFT data

ML models successfully predict properties of untrained alloy compositions

Proposes ML as a rapid tool for designing hydrogen storage materials

Abstract

Safe and high-density storage of hydrogen, for a clean-fuel economy, can be realized by hydride-forming materials, but these materials should be able to store hydrogen at room temperature. Some high-entropy alloys (HEAs) have recently been shown to reversibly store hydrogen at room temperature, but the design of HEAs with appropriate thermodynamics is still challenging. To explore HEAs with appropriate hydride formation enthalpy, this study employs machine learning (ML), in particular, Gaussian process regression (GPR) using four different kernels by training with 420 datum points collected from literature and curated here. The developed ML models are used to predict the formation enthalpy of hydrides for the TixZr2-xCrMnFeNi (x = 0.5, 1.0 and 1.5) system, which is not in the training set. The predicted values by ML are consistent with data from experiments and density functional theory (DFT). The present study thus introduces ML as a rapid and reliable approach for the design of HEAs with hydride formation enthalpies of -25 to -39 kJ/mol for hydrogen storage at room temperature.