High-throughput screening of novel hydrogen storage materials – ML approach

Hydride formation in metals is a widely studied and applied phenomenon necessary to transition to clean energy solutions and various technological applications. We focus on three perspective applications of these materials, namely near-ambient hydrogen storage, hydrogen storage compressor materials, and alkali metal conversion electrodes, to demonstrate acceleration in the research achieved by utilizing a data-driven approach. Graph neural network was developed using a transfer learning approach from the MEGNet model and data related to the thermodynamics of hydride formation obtained in experimental work. Based on the crystal structure and composition as input features, we apply the MetalHydrideEnth model developed in our previous work to predict hydride formation enthalpy in intermetallic compounds. In this work, we focus on demonstrating how this approach, combined with available crystal information obtained from density functional theory calculations, can be applied for fast and extensive searches of novel metal hydride materials, having in mind the above-listed applications.

Hali tarjima qilinmagan