(265a) Accelerated Battery Electrolyte Discovery Using Machine Learning

Rechargeable batteries are pivotal in addressing many of society’s decarbonization goals. Unfortunately, developing new rechargeable battery chemistries can take decades. A big reason why is the infinite chemical space one must traverse to find a battery material that works. In this talk, I will focus on a battery material called the electrolyte. The electrolyte consists of salts dissolved in solvents (at different concentrations, in different mixtures); hence the chemical space is practically infinite. We have pursued a data-driven approach for electrolyte discovery. I will give vignettes about (1) our development of machine learning models that can map electrolyte composition to a property prediction. We show highly accurate models for the prediction of three important electrolyte properties (ionic conductivity, oxidative stability, and Coulombic efficiency). (2) development of active learning models paired with experimental feedback to smartly navigate the chemical space and increase the diversity of high performing electrolytes. Our approach on developing data science models to accelerate battery materials discovery is pivotal in accelerating decarbonization.