Predicting Capacity Fading Behaviors of Lithium Ion Batteries: An Electrochemical Protocol-Integrated Digital-Twin Solution

Abstract

The capacity degradation and occurrence of safety hazards of lithium ion batteries are closely associated with various adverse side electrochemical reactions. Nevertheless, these side reactions are non-linearly intertwined with each other and evolve dynamically with increasing cycles, imposing a major barrier for fast prediction of capacity decay of lithium ion batteries. By treating the battery as a black box, the machine-learning-oriented approach can achieve prediction with promising accuracy. Herein, a numerical-simulation—based machine learning model is developed for predicting battery capacity before failure. Based on the deterioration mechanism of the battery, numerical model was applied to test data from only 25 batterie to extend 144 groups data, resulting in the digital-twin datasets, which can reliably predict the maximum total accumulative capacity of the lithium ion batteries, with an error less than 2%. The workflow with iterative training dramatically accelerates the capacity prediction process and saves 99% of the experimental cost.