Reliable state estimation of lithium-ion batteries under model and parameter uncertainties

Lithium-ion batteries have attracted significant attention as a versatile energy storage device used in various applications ranging from cell phones to space explorers. This growing attention and ubiquity make battery safety and reliability analysis an extremely important task. While there is a great amount of active research on state-of-charge (SOC) estimation of lithium-ion batteries, the existing work is mainly focused on the development of different battery models, various battery SOC estimation algorithms, and studying the influence of battery SOC estimation under aging and different temperature conditions. While aforementioned research is pivotal to achieve accurate battery SOC estimation, two important technical components are missing if the goal is to obtain reliable battery SOC estimation. The difference is that reliable battery SOC estimation considers different sources of uncertainties in the estimation while existing research focuses on the estimation under mostly deterministic conditions. The first important uncertainty source is model bias which represents the inherent model inadequacy for representing the real physical systems. Any battery model contains a certain degree of model bias and the accuracy may vary under different conditions. Hence, a reliable battery SOC method should consider such uncertainty of the model bias, or model uncertainty, in the estimation. The second important uncertainty source that is usually ignored in the literature is the battery cell-to-cell variability due to manufacturing tolerance. As such, characterized battery model parameter, or the nominal parameter, may not represent the true parameter for specific batteries if such cell-to-cell variability is not ignorable. The proposed research firstly addresses battery model bias and cell-to-cell variability in battery SOC estimation at the cell level, then further extends the study to the pack level for quantifying the influence of pack SOC estimation given both battery model and parameter uncertainty. There are five major contributions in the proposed research. Firstly, a systematic battery model bias learning framework is proposed to dramatically improve the battery SOC estimation using any available battery models. Secondly, machine learning methods are proposed to incorporate battery model uncertainty for reliable battery SOC estimation under various operation conditions. Thirdly, a probabilistic battery SOC estimation framework with effective uncertainty quantification (UQ) method is proposed to consider the effects from cell-to-cell variability so that reliable battery SOC estimation can be obtained at any confidence levels. Fourthly, a systematical framework for both battery SOC and capacity estimation while considering the cell-to-cell variability is proposed to ensure reliable battery SOC estimation even when battery degrades without knowing its true capacity. Finally, innovative battery pack SOC estimation, in which there may be hundreds or thousands of battery cells, is proposed for reliable pack SOC estimation considering the cell-to-cell variability. For each technical contribution, theoretical study, simulations, and experiments are employed to demonstrate the effectiveness of the proposed work.Ph.D.Includes bibliographical reference