Electrochemical–mechanical coupled model for computationally efficient prediction of long-term capacity fade of lithium-ion batteries

This paper proposes a novel physics-based (electrochemical–mechanical coupled) capacity fade model called the inhomogeneous stress-induced fracture (ISIF) model based on electrochemical and mechanical degradation mechanisms. The ISIF model models the mechanical fatigue fracture of cathode particles, which is a key aging mechanism that contributes to nonlinear aging observed in various studies. Through this approach, the ISIF model can accurately predict long-term capacity degradation, including the region of capacity loss below 80% of the initial capacity, without the need for time-consuming calculations, unlike electrochemical-based models. Additionally, the ISIF model can predict the knee point phenomenon, where there is a sudden decrease in capacity in the later-stages of battery life. We also propose a hybrid model called SVD-ISIF model that combines the ISIF model with a data-driven method called sparse variational dropout Bayesian neural network (SVDBNN) to improve accuracy and data efficiency especially when experimental data and computational resources are abundant, which helps users flexibly choose the optimal method for various situations. The proposed method was validated using experimental long-term capacity fading data from commercial 2170 NMC cells over a wide range of cycles (3000 to 20,000) that have not been studied before, because of the difficulties of conducting long-term aging cycle experiments and reproducing the knee point, to the knowledge of the authors. Both the ISIF model and the SVD-ISIF model accurately predict the long-term capacity reduction trend, including the knee point, with a mean absolute error is around or less than 2% for all data.