Multiphysics-informed Machine Learning for Uncertainty Quantification on Si Anode Based Battery Performance

The increasing demand for Li-ion batteries (LIBs) has prompted a need for advancements in their design and technology. One such improvement involves utilizing Silicon (Si) as the active anode material in LIBs. However, a significant challenge hindering its widespread application is the substantial volume change it undergoes during the lithiation-delithiation process. This can lead to volumetric stress-induced capacity degradation. Three primary capacity fade mechanisms are observed in these LIBs: volumetric-stress-induced cracking and delamination, along with the growth of the solid electrolyte interface (SEI) during the charging and discharging cycles. These mechanisms are influenced by battery design and operating conditions, such as Si anode thickness, ambient working temperature, and charging rate, introducing uncertainty in the battery’s degradation rate. In this study, multiple finite element (FE) models are constructed to simulate capacity degradation resulting from these three fade mechanisms. Since running these FE models can be time-consuming, a Gaussian process regression (GPR) based surrogate model is developed for predicting capacity fade. Once validated, this GPR model is employed in an uncertainty quantification study on the battery’s design/operating conditions. This analysis aims to identify the factors that most significantly affect capacity degradation in Si anode-based LIBs.