Battery State of Health Monitoring via Estimation of Health-Relevant Electrochemical Variables

This dissertation explores and compares the effectiveness of estimating two health-relevant electrochemical variables, the side reaction current density and the number of cyclable Li-ions, as indicators of battery state of health (SOH) in battery management systems of electric vehicles (EV) and hybrid electric vehicles (HEV). The choice of these two electrochemical variables is based on the assumption that battery degradation is mainly caused by consumption of cyclable Li-ions. This assumption is valid for the two widely-used types of EV/HEV batteries considered herein, namely LiFePO4 and LMO-based mixture batteries. This dissertation provides formulations to estimate these two electrochemical variables from measurements of battery terminal voltage and current. Estimation is necessary here because the electrochemical variables cannot be measured on-board. Estimation of the side reaction current density is formulated as a subsystem identification problem and is solved using retrospective-cost subsystem identification. A new subsystem identification algorithm, the two-step filter, is also developed to improve the estimation accuracy of the side reaction current density under the presence of state of charge (SOC) estimation errors. On the contrary, the number of cyclable Li-ions is estimated as an unknown battery parameter using the extended Kalman filter. This dissertation also analyzes the robustness of estimation of the two electrochemical variables by providing a framework to obtain the lower bound of relative estimation errors of each of the two variables under non-ideal conditions for algorithms that estimate the variable by minimizing the error between measured voltage and estimated voltage. This framework determines that the lower bound of the relative estimation error of a variable is proportional to the error in either measurement or estimate of battery terminal voltage caused by non-ideal conditions, and inversely proportional to the sensitivity of the voltage to the variable and the magnitude of the variable itself. This framework also yields the same lower bound for the covariance of unbiased estimates as given by the Fisher information. The effectiveness of estimating the side reaction current density and the number of cyclable Li-ions as SOH indicators is also discussed through comparison. Compared to the number of cyclable Li-ions or other SOH indicators such as capacity and internal resistance, the side reaction current density is a more ideal SOH indicator when it can be estimated accurately, because it can instantaneously indicate battery degradation rate. However, estimation of the side reaction current density under practical non-ideal conditions is fundamentally difficult due to the fact that the sensitivity of the voltage to the side reaction current density and the magnitude of the side reaction current density are both low. On the other hand, the number of cyclable Li-ions is a promising SOH indicator for battery management systems in practice because it provides an indication of the remaining capacity from the first principles, can be estimated using a standard algorithm and simple models, and demonstrates high robustness to non-ideal conditions. Future extensions of this work include i) studying the impact of temperature on estimation of health-relevant electrochemical variables by including thermal dynamics in the model, ii) validating experimentally estimation of the number of cyclable Li-ions, and iii) extending estimation of the side reaction current density to other side-reaction-based battery degradation and safety problems such as Lithium plating and dendrite formation.PhDMechanical EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttps://deepblue.lib.umich.edu/bitstream/2027.42/137087/1/zhouxin_1.pd