Reduced-order modeling and adaptive observer design for lithium-ion battery cells

Abstract

This thesis discusses the design of a control-oriented modeling approach to Lithium- Ion battery modeling, as well as the application of adaptive observers to this structure. It begins by describing the fundamental problem statement of a battery management system (BMS), and why this is challenging to solve. It continues by describing, in brief, several different modeling techniques and their use cases, then fully expounds two separate high fidelity models. The first model, the ANCF, was initiated in previous work, and has been updated with novel features, such as dynamic diffusion coefficients. The second model, the ANCF II, was developed for this thesis and updates the previous model to better solve the problems facing the construction of an adaptive observer, while maintaining its model accuracy. The results of these models are presented as well. After establishing a model with the desired accuracy and complexity, foundational observers are designed to estimate the states and parameters of the time-varying ionic concentrations in the solid electrode and electrolyte, as well as an a-priori estimate of the molar flux. For the solid electrode, it is shown that a regressor matrix can be constructed for the observer using both spatial and temporal filters, limiting the amount of additional computation required for this purpose. For the molar flux estimate, it is shown that fast convergence is possible with coefficients pertaining to measurable inputs and outputs, and filters thereof. Finally, for the electrolyte observer, a novel structure is established to restrict learning only along unknown degrees of freedom of the model system, using a Jacobian steepest descent approach. Following the results of these observers, an outline is sketched for the application of a machine learning algorithm to estimate the nonlinear effects of cell dynamics.