Identification of Stable Intermetallic Compounds for Hydrogen Storage via Machine Learning

Abstract

Hydrogen is one of the most promising alternatives to fossil fuels for energy as it is abundant, clean and efficient. Storage and transportation of hydrogen are two key challenges faced in utilizing it as a fuel. Storing H₂ in the form of metal hydrides is safe and cost effective when compared to its compression and liquefaction. Metal hydrides leverage the ability of metals to absorb H₂ and the stored H₂ can be released from the hydride by applying heat when needed. A multi-step methodology is proposed to identify intermetallic compounds that are thermodynamically stable and have high hydrogen storage capacity (HSC). It combines compound generation, thermodynamic stability analysis, prediction of properties of the metal hydride and ranking of discovered materials based on predicted properties. The US Department of Energy (DoE) Hydrogen Storage Materials Database and the Open Quantum Materials Database (OQMD) are utilized for building and testing machine learning (ML) models for enthalpy of formation of the intermetallic compounds, stability analysis, and enthalpy of formation, equilibrium pressure and HSC of metal hydrides. The models proposed here require only attributes of elements involved and compositional information as inputs and do no need any experimental data. Random forest algorithm was found to be the most accurate amongst the ML algorithms explored for predicting all the properties of interest. A total of 349 772 hypothetical intermetallic compounds were generated initially, out of which, only 8568 compounds were found to be stable. The highest predicted HSC of these stable compounds was found to be 3.6. Magnesium, Lithium and Germanium constitute majority of the high HSC compounds. The predictions of HSC using the present models for metal hydrides that are not in the DoE database were reasonably close to the experimental data published recently but there is scope for improvement in prediction accuracy for metal hydrides with high HSC. The findings of this study will be useful in reducing the time required for development and discovery of new hydrogen storage materials and can be used to check the practical applicability of the hydride compound using the predicted properties.