Data-Driven Acceleration of Battery Reactivity Models from Molecules to Discovery

The wealth of data generated in battery research is a wellspring for advancing fundamental discovery, performance prediction, and management via modeling methods. Yet the wide variety of modeling methods coupled with the wide variety of battery types and chemistries require refining the problem to a manageable task. Our approach focuses on a data-generating workflow using the self-driving lab concept. Here, we define a specific challenge in battery science that will benefit from automation, robotics, data handling, and engineering for ultimate autonomous materials discovery.<br/><br/>Redox flow batteries are an energy storage technology still nascent when compared to the ubiquitous lithium-ion batteries in everyday life. Compared to the traditional batteries, flow batteries possess a starkly different configuration based on liquid electrolytes that flow through electrode stacks. There is ripe opportunity for the advanced discovery of the electrolyte via predictive methods. Indeed, many redox-active materials, the charge-storing species of the battery, have been studied, both in the laboratory and in silico. Further, many of these actives have been engineered to dissolve in nonaqueous solvents, attractive due to their wide electrochemical windows for improved energy density. Yet, compared to the actives, new solvent reagents have seen minimal development. To understand which solvents could be good candidates for a flow battery, their reactivity must be mapped with the active species, which led to our goal of building a reactivity model for active-reagent pairs using data-driven modeling methods. We identified over 500 possible battery-relevant solvent reagents for testing with a charged redox-active molecule. Liquid combinations of the reagents and active were prepared using a liquid handling robot in a nitrogen-filled glovebox. As the charged active is highly colored, its decay kinetics can be monitored by UV-vis spectroscopy. Decay kinetics of the active species were measured based on changes in optical density and fed to adaptive sampling methods. Overall, 20% of the reagent space was sampled, under 300 conditions, for a total of &gt; 2500 experiments. From these predictive methods, several new reagent molecules were picked as stable materials and validated by experiment testing.<br/><br/>This research success includes (i) the general metamodels for predicting reactivity and (ii) the use of these metamodels to identify slower reacting reagents that can be potential new electrolyte solvents, with further funneling of these candidates. This research program is driven by Argonne’s Autonomous Discovery Laboratory – molecules to materials discovery platform. It also brings together an interdisciplinary team of chemists, materials scientists, robotics experts, and computer scientists for complete execution.<br/><br/>This material is based upon work supported by Laboratory Directed Research and Development (LDRD) funding from Argonne National Laboratory, provided by the Director, Office of Science, of the U.S. Department of Energy under Contract No. DE-AC02-06CH11357. The U.S. Government retains for itself, and others acting on its behalf, a paid-up nonexclusive, irrevocable worldwide license in said article to reproduce, prepare derivative works, distribute copies to the public, and perform publicly and display publicly, by or on behalf of the Government.