Integration of Semi-Empirical and Artificial Neural Network (ANN) for Modeling Lithium-Ion Electrolyte Systems Dynamic Viscosity

Abstract

The dynamic viscosity is a key characteristic of electrolyte performance in a Lithium-ion (Li-ion) battery. This study introduces a one-parameter semi-empirical model and artificial neural network (ANN) to predict the viscosity of salt-free solvent mixtures and relative viscosity of Li-ion electrolyte solutions (lithium salt + solvent mixture), respectively. Data used in this study were obtained experimentally, in addition to data extracted from literature. The ANN model has seven inputs: salt concentration, electrolyte temperature, salt-anion size, solvent melting, boiling temperatures, solvent dielectric constant, and solvent dipole moment. Different configurations of the ANN model were tested, and the configuration with the least error was chosen. The results show the capability of the semi-empirical model in predicting the viscosity with an overall mean absolute percentage error (MAPE) of 2.05% and 3.17% for binary and tertiary mixtures, respectively. The ANN model predicted the relative viscosity of electrolyte solutions with MAPE of 4.86%. The application of both models in series predicted the viscosity with MAPE of 2.3%; however, the ANN MAPE alone is higher than this value. Thus, this study highlights the promise of using predictive models to complement physical approaches and effectively perform initial screening on Li-ion electrolytes.