Selection of ionic liquid electrolytes for high-performing lithium-sulfur batteries: an experiment-guided high-throughput machine learning analysis

Abstract

The polysulfide (PS) shuttle mechanism (PSM) is one of the most significant challenges of lithium-sulfur (Li-S) batteries in achieving high capacity and cyclability. One way to minimize the shuttle effect is to limit the PS solubilities in the battery electrolyte. Ionic liquids (IL) are particularly suited as electrolyte solvents because of their tunable physical and chemical properties. In this work, thousands of ILs are screened to narrow down potentially viable candidates to be used as electrolytes in Li-S batteries. To that end, the COnductor-like Screening Model for Realistic Solvents (COSMO-RS) calculations are performed over more than 36,000 ILs. An extensive database containing PS solubilities and other relevant properties is constructed at 25 °C. First, the effectiveness of the COSMO-RS calculations is experimentally tested with six different ILs having a wide range of solubility and viscosity values; a strong correlation between the PS solubility and battery performance is obtained. After specifying the target limits for promising ILs using the experimental battery performance data, machine learning (ML) tools are used to predict and identify the relationship between IL properties and PS solubilities and structural and molecular descriptors of ILs. The extreme gradient boosting (XGBoost) method successfully predicts the solubility and property values. Association rule mining (ARM) and the feature importance analysis show that anion descriptors are more dominant, whereas cations have less impact on the solubilities and properties of ILs. Finally, the imidazolium and pyridinium ILs with bis_imide and borate anion groups are identified as the most promising ones.