Non-aqueous battery electrolytes: high-throughput experimentation and machine learning-aided optimization of ionic conductivity

Abstract

Although data-driven optimization has been recognized as a useful approach for the advancement of liquid battery electrolytes, a high quality and large dataset is essential to avoid bias in the interpretation of results. In this work, we present the newly designed and developed platform which comprises an automated high-throughput experimentation (HTE) modular system coupled with the Liquid Electrolyte Composition Analysis (LECA) package for the data-driven modeling and analysis of ionic conductivity as a vital bulk electrolyte property. The LECA package combines popular machine learning libraries into a simplified workflow enabling easy, semi-automated processing and analysis of HTE acquired data. The package facilitates the parallel training, cross-validation and uncertainty estimation of Linear Regression, Random Forest, Neural Network and Gaussian Process Regression models. By comparatively scoring model prediction accuracy, the LECA package identifies the best performing model architecture(s) and applies them to find electrolyte compositions which maximize ionic conductivity. Overall, this comprehensive and versatile platform with automated experiments and data-driven analysis paves the way for more efficient and insightful research in liquid battery electrolyte development.