Key Applications of State and Health Estimation

In the previous two chapters, we have developed theoretical foundation of data-driven methods for lithium-ion batteries. In this chapter, we present test cases to demonstrate applicability and capabilities of these methods for lithium-ion battery state estimation. First, we explore the recursive Bayesian framework for state of charge estimation. In this chapter, we compare the unscented Kalman filter and particle filter for state of charge estimation. Functionality of these algorithms is demonstrated for a commercial NCA/C cell state estimation at different operating conditions including constant current discharge at room and low temperatures, hybrid power pulse characterization (HPPC), and urban driving schedule (UDDS) protocols. In addition to accurate voltage prediction, the electrochemical nature of ROM enables drawing of physical insights into the cell behavior. Advantages of using electrode concentrations over conventional Coulomb counting for accessible capacity estimation are discussed. In addition to the mean state estimation, the framework also provides estimation of the associated confidence bounds that are used to establish predictive capability of the proposed framework. Next, we demonstrate applicability of the machine learning algorithms for lithium-ion battery state of health estimation. For this, we present a novel method that utilizes both the classification and regression flavors of the machine learning algorithms. For demonstration purpose, we consider SVM for classification and regression; however, other approaches can be similarly used. For this demonstration, we have used a publicly available battery life testing dataset for training the SVM/R algorithm and subsequently tested our approach on a different subset of the dataset (Figures and discussions reproduced with permissions from Elsevier.).