Designing Imidazolium-Mediated Polymer Electrolytes for Lithium-Ion Batteries Using Machine-Learning Approaches: An Insight into Ionene Materials

Over the past few decades, lithium-ion batteries (LIBs) have gained significant attention due to their inherent potential for environmental sustainability and unparalleled energy storage efficiency. To enhance the performance of lithium-ion batteries, electrolytes have garnered considerable attention as a key component of these batteries. Meanwhile, polymer electrolytes have gained popularity in several fields due to their ability to adapt to various battery geometries, enhanced safety features, greater thermal stability, and effectiveness in reducing dendrite growth on the anode. In general, polymer electrolytes are composed of polymer matrices and lithium salts, mainly categorized as solid polymer electrolytes (SPEs) and gel polymer electrolytes (GPEs), which provide higher energy densities while maintaining structural integrity and safety. Despite many advantages, offering relatively lower ionic conductivity as compared to liquid electrolytes, polymer electrolytes are limited to advanced applications. This limitation has led to recent studies revolving around the development of poly (ionic liquids) (PILs), particularly imidazolium-mediated polymer backbones as novel electrolyte materials, which can increase the conductivity with fine-tuning structural benefits, while maintaining the advantages of both solid and gel electrolytes. There have been various structural conformations explored in the design of multiple PILs, and the accurate measurement of conductivity is typically performed in laboratories, which can be both costly and time-consuming. Therefore, in this study, we aimed to develop intelligent models for the accurate estimation of ionic conductivity in exclusive imidazolium polymeric ionic liquids (PILs). For this purpose, a dataset consisting of 120 datapoints, including 8 different polymers, encompassing all the imidazolium-based PILs reported to date, was compiled from the literature. Most importantly, this study foresees the benefits of newly integrated PIL substructures, so-called ionenes, toward the performance of LIB applications. Four machine learning (ML) models of CatBoost, RF, XGBoost, and LightGBM were developed in this study by incorporating chemical structure and temperature as the models’ inputs. The results indicated the superior performance of the CatBoost model compared to other models with R2, RMSE, and MAE of 0.986, 0.000187, and 0.0000952, respectively. The importance of features in predicting conductivity was investigated using the CatBoost model. The results indicated that temperature plays the leading role, followed by chemical descriptors such as, PIL_BCUT2D_MRLOW, PIL_SMR_VSA6, and PIL_EState_VSA8. Moreover, the best-performing model (CatBoost) was used to predict conductivity for three novel ionenes, paving the way for a new approach to utilizing innovative polymer architecture toward LIB applications.