Multiphysics-Informed Machine Learning for Mechanical-Induced Degradation of Silicon Anode

Abstract Silicon (Si) anode based lithium-ion batteries (LIBs) are being developed and used in various portable electronic technologies because of their better life cycle performance and safety. These Si anode based LIBs also provide a better capacity due to the unique intercalating mechanisms of lithium (Li) into Si. However, due to this unique mechanism, volumetric changes upto 300% have been observed in these batteries that leads to the development of internal stresses in the Si anode which ultimately results in cracking and delamination in it. These two cracking and delamination failure modes along with the growth of the solid electrolyte interface (SEI) on the exposed surface of Si anode leads to loss in the overall capacity of the battery. The capacity degradation can be simulated using FE models but these models take a long time to run and are computationally expensive. Hence, in this study, we develop a physics-informed machine learning technique for the capacity degradation of the Si anode based LIBs. 3D finite element (FE) models are built to understand the volumetric stresses induced cracking and delamination along with the capacity loss due to the growth of the SEI layer. The outputs from these FE models are then used to train the Gaussian process regression (GPR) surrogate model which can be used for the design of LIBs towards application-oriented properties such as high energy storage, fast charging or optimal life time by quickly and accurately predicting the capacity degradation in the battery.