UC Berkeley

Lithium-ion (Li-ion) batteries have emerged as one of the most prominent energy storage devices for large-scale energy applications, e.g., hybrid electric vehicle (HEV), battery electric vehicles (BEV) and smart grids, due to their high energy and power density, low self-discharge and long lifetime. This dissertation focuses on developing and validating real-time model-based state and parameter estimation algorithms for Li-ion battery management systems. An overview of individual chapters are provided below.

Chapter 2: This chapter examines an online battery capacity estimation scheme from a thermal perspective. Mathematically, a cylindrical battery is modeled by coupling an electrical model that describes the dynamics of state of charge (SOC) with a two-state thermal dynamics. The critical challenges, however, are that (i) only input current, battery terminal voltage, and surface temperature are measureable in real time, and (ii) the sub-models are nonlinearly coupled. Consequently, the proposed hierarchical estimation algorithm uses a combination of input-to-state stability and sliding mode observer to collectively estimate cell capacity. Furthermore, the algorithm also presents real-time estimation for SOC, core temperature, heat generation, and thermal model parameters, making the algorithm a novel methodology for combined state of charge/state of health (SOH) estimation. The results demonstrate the benefits of thermal model based battery capacity estimation against traditional equivalent circuit based estimation.

Chapter 3: An electrochemical battery cell relies on the intercalation and de-intercalation of Li-ions between electrode solid-phase and electrolyte. An important cell capacity fade mechanism is the particle fracture due to intercalation-induced stresses. Volume changes of the electrode particles due to stress may induce particle fracture if the stress exceeds the yielding stress of the electrode material. In this chapter, we design a nonlinear observer based on a Single Particle Model (SPM) coupled with intercalation-induced stress to estimate battery bulk SOC, the particle stress profile, and the anode lithium diffusivity from online current and terminal voltage measurements only. Practically, real-time monitoring of aging related parameters in battery model and internal mechanical stress enables a battery management system (BMS) to apply optimal control methods that protect against particle fracture, and consequently extend battery life.

Chapter 4: This chapter seeks to derive insight on estimation problem for battery packs. A battery pack system generally consists of hundreds or thousands of single cells connected in parallel and series in order to fulfill the requirements of high-energy and high-power applications. Mathematically, an equivalent circuit model is coupled with a thermal model to form a single cell model, which is then electrically interconnected with other cell models to form a pack model utilizing Kirchhoff’s law. For cells in parallel, the resulting model is depicted by a differential-algebraic system (i.e. a descriptor system). The first part of this chapter aims at designing a Lyapunov-based asymptotic state observer for both differential (state of charge) and algebraic (local cell current) state estimation subject to reduced sensing. On the other hand, however, when number of cells in a pack becomes large, executing estimation algorithm for each and every cell becomes intractable computationally. The second part of this chapter proposes a monotone system based interval observer while taking into account modeling and measurement uncertainties in a pack. The estimated SOC intervals (upper and lower bounds) are guaranteed to envelop all SOC trajectories in the pack. The interval observer loses the tractability of single cell states but maximize the scalability of the algorithm.

Chapter 5: Battery thermal effects have been shown to be key factors in the rate of battery degradation. In practical applications, many cases of thermal runaways leading to fire and explosion of Li-ion batteries have been reported. This chapter proposes a model-based estimation algorithm for the battery temperature relying on a reduced high-fidelity nonlinear distributed parameter thermal model using surface measurements. Theoretically, the work extends the traditional partial differential equation (PDE) backstepping technique for a nonlinear parabolic PDE state estimation problem without performing linearization and spatial discretization prior to the observer design. When modeling and measurement disturbances exist, the algorithm quantifies the estimation error bounds in terms of L2 spatial norm.

Chapter 6: The modeling for electrochemical phenomena inside a battery cell generally adopts a diffusion process. This chapter exclusively investigates a class of reaction-advection-diffusion system subject to boundary disturbance. Theoretically, we design a disturbance estimator for boundary disturbance in an unstable reaction-advection-diffusion PDE system, and derive a sufficient condition on the reaction coefficient, for which the disturbance estimator achieves asymptotic convergence. Subsequently, we propose an asymptotically convergent state estimator for the unstable reaction-advection-diffusion PDE using the estimated disturbance signal, adopting the backstepping technique. This chapter builds a solid foundation for PDE-based Li-ion concentration, and therefore SOC, estimation in a electrochemical cell subject to boundary disturbances.

Chapter 7: This chapter provides concluding remarks of the dissertation and discussions on future works.