Combination of high-throughput phase field modeling and machine learning to study the performance evolution during lithium battery cycling

Abstract

This study introduces a phase field (PF) model of a full-cell during galvanostatic cycling, taking into account dead lithium formation. A step function is used to distinguish between the ‘active’ and ‘dead’ states of localized lithium metal. The galvanostatic conditions are described using Ohm's law. The relationship between voltage, current density, and internal resistance is also established. High-throughput (HTP) PF simulations of Li-Li symmetric cells show the effects of current density and ion diffusion coefficient on the growth of lithium dendrites. The morphological changes are quantitatively characterized by introducing the perimeter-to-area ratio. It is found that higher current densities and lower diffusion coefficients accelerate dead lithium accumulation, hindering ion transport and leading to an increase in potential. The accumulation of dead lithium directly causes battery capacity degradation. Finally, by combining HTP-PF simulations with machine learning, the study establishes the relationship between battery parameters (e.g., current density, diffution coefficient, and number of cycles) and performance (e.g., battery life and Coulombic efficiency (CE)). This approach aims to address the issue of high computational cost associated with applying PF methods to predict battery performance under cycling conditions. It is expected that combining HTP-PF simulation with machine learning, along with experimental validation in the future, will provide new ideas for accelerating the development of lithium batteries.