Innovations in batteries take years to formulate, requiring extensive experimentation during the design and optimization phases. In this talk, an approach to the design of a battery electrolyte as a machine-learning (ML) driven optimization problem. Specifically, a robotic fully-automated test stand capable of creating arbitrary electrolyte solutions with up to ten constituents is described. The stand is controlled by Dragonfly, a Bayesian ML software package that was first used to search mixtures lithium and sodium salts for electrolytes (aqueous and organic solvent based). Dragonfly recommended electrolyte designs to test and then received real time experimental feed- back. In one case, within only 40 hours of continuous experimentation, Dragonfly discovered a novel, high-performing electrolyte system that a human-guided design process may have missed, and would have taken weeks or months to identify. This result demonstrates the possibility of integrating robotics with machine- learning to rapidly and autonomously discover novel electrochemically functional materials. An overview of the system and the approach will be given and a several set of compelling results from aqueous and organic solvent based systems will be offered.

Autonomous machine learning-informed experimentation: Optimizing complex battery electrolyte solutions with maximum efficiency

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.