An active learning approach to model solid-electrolyte interphase formation in Li-ion batteries

Abstract

Li-ion batteries store electrical energy by electrochemically reducing Li ions from a liquid electrolyte in a graphitic electrode. During these reactions, electrolytic species in contact with the electrode particles form a solid-electrolyte interphase (SEI), a layer between the electrode and electrolyte. This interphase allows the exchange of Li ions between the electrode and electrolyte while blocking electron transfer, affecting the performance and life of the battery. A network of reactions in a small region determines the final structure of this interphase. This complex problem has been studied using different multi-scale computational approaches. However, it is challenging to obtain a comprehensive characterization of these models in connection with the effects of model parameters on the output, due to the computational costs. In this work, we propose an active learning workflow coupled with a kinetic Monte Carlo (kMC) model for formation of a SEI as a function of reaction barriers including electrochemical, diffusion, and aggregation reactions. This workflow begins by receiving an initial database from a design-of-experiment approach to train an initial Gaussian process classification model. By iterative training of this model in the proposed workflow, we gradually extended the model's validity over a larger subset of reaction barriers. In this workflow, we took advantage of statistical tools to reduce the uncertainty of the model. The trained model is used to study the features of the reaction barriers in the formation of a SEI, which allows us to obtain a new and unique perspective on the reactions that control the formation of a SEI.