Title:Data-Driven Approach to Learning Optimal Forms of Constitutive Relations in Models Describing Lithium Plating in Battery Cells

Abstract:In this study we construct a data-driven model describing Lithium plating in a battery cell, which is a key process contributing to degradation of such cells. Starting from the fundamental Doyle-Fuller-Newman (DFN) model, we use asymptotic reduction and spatial averaging techniques to derive a simplified representation to track the temporal evolution of two key concentrations in the system, namely, the total intercalated Lithium on the negative electrode particles and total plated Lithium. This model depends on an a priori unknown constitutive relations of the cell as a function of thestate variables. An optimal form of this constitutive relation is then deduced from experimental measurements of the time dependent concentrations of different Lithium phases acquired through Nuclear Magnetic Resonance spectroscopy. This is done by solving an inverse problem in which this constitutive relation is found subject to minimum assumptions as a minimizer of a suitable constrained optimization problem where the discrepancy between the model predictions and experimental data is minimized. This optimization problem is solved using a state-of-the-art adjoint-based technique. In contrast to some of the earlier approaches to modelling Lithium plating, the proposed model is able to predict non-trivial evolution of the concentrations in the relaxation regime when no current isapplied to the cell. When equipped with an optimal constitutive relation, the model provides accurate predictions of the time evolution of both intercalated and plated Lithium across a wide range of charging/discharging rates. It can therefore serve as a useful tool for prediction and control of degradation mechanism in battery cells.