Accelerating Data-driven Discovery of Materials for Hydrogen Storage and Generation

Speaker: Matthew Witman, Sandia National Laboratories

Abstract
Recent industry and government commitments to hydrogen R&D and deployment are indicators of its maturing role in the clean energy transition. But the roll-out of a global “hydrogen powered economy” is still being blunted by technical challenges and costs associated with storing, transporting, and generating green hydrogen. Among a multitude of fundamental and engineering obstacles, our work is particularly focused on the fact that sufficiently high-performing candidates for materials-based hydrogen storage and generation, that could otherwise enable more rapid widespread industrial adoption, remain undiscovered. The traditional approach to novel materials discovery has typically relied on researchers’ significant domain expertise and trial-and-error experimentation in order to narrow down the vast space of possible materials, and the advent of high-throughput computational screening with modern supercomputers has rapidly accelerated this process. Building upon these approaches, a recent paradigm involves the use of machine learning (ML) techniques to derive surrogate models that can more efficiently screen materials’ performance properties. This application focused talk will survey some of the data-driven materials discovery efforts within Sandia’s hydrogen programs, from metals and high entropy alloys for hydrogen storage, to liquid metals and novel oxides for hydrogen generation via water-splitting. Each necessitates its own flavor of ML based on various problem constraints and data availability. In some applications, we focus on simpler interpretable ML techniques to elucidate previously unknown structure-property relationships for facile targeting of novel materials with desired properties. In other applications, state-of-the-art graph neural networks are utilized/extended for either direct materials’ property predictions or to feed sampling-intensive first-principles methods. Critically, all data-driven discovery exercises are being performed in in close collaboration with experimental groups to validate and close the materials discovery loop.

Bio
Matthew Witman obtained his Ph.D. in Chemical Engineering from the University of California, Berkeley in 2019 in the Berend Smit group where he focused on computationally guided discovery, simulation, and high-throughput screening of porous materials for clean energy applications. Now a Senior Member of the Technical Staff at Sandia National Laboratories, he completed a post-doctoral appointment in 2021 under the mentorship of Mark Allendorf, Vitalie Stavila, and Anthony McDaniel, and currently helps lead diverse data-driven materials discovery efforts across HyMARC and HydroGEN, including alloys, liquid metals, oxides, and porous materials for hydrogen storage and/or generation.