Gaussian process sensitivity analysis of a capacity fade model for lithium-ion batteries

Accurate battery models are important to maximise the utility of batteries which are ubiquitous in modern society from smartphones to electric vehicles. They act as digital twin of batteries and help in expeditious design optimization as well as help us explore interplay of electro- chemical phenomena. Here we present a sensitivity analysis of a pseudo-two-dimensional battery model coupled with a capacity fade model based on the formation of solid electrolyte interphase and the corresponding irreversible charge loss for Li-ion batteries. The proposed method is based on training an inexpensive differentiable surrogate Gaussian process regression model on observed input-output pairs and analysing the surrogate model to learn about the global and local sensitivities of the original system. Our results show that the proposed method is able to identify the most sensitive input parameters globally and that the same method can be used to explore local sensitivities around specic sets of inputs providing insights to the governing electrochemical process. In addition, we found a strong correlation between the growth of the solid electrolyte interphase and the irreversible charge loss especially at low current rates.