EXPERIMENTAL STUDIES ON THE VALUE OF INFORMATION IN LITHIUM-ION BATTERY STATE AND PARAMETER ESTIMATION
Open Access
- Author:
- Mendoza Galvis, Sergio Andres
- Graduate Program:
- Mechanical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- November 17, 2016
- Committee Members:
- Hosam Fathy, Dissertation Advisor/Co-Advisor
Zoubeida Ounaies, Committee Chair/Co-Chair
Chris Rahn, Committee Member
Donghai Wang, Committee Member
Constantino Lagoa, Outside Member - Keywords:
- lithium ion batteries
identifiability
parameter identification
SOC estimation
optimal experiments
Fisher Analysis - Abstract:
- This dissertation presents a set of experimental studies on the value of information in lithium-ion battery state estimation and parameter identification. The research in this dissertation focuses on state and parameter estimation for cells employing graphite negative electrodes and lithium iron phosphate (LFP) positive electrodes, but its fundamental contributions are broadly applicable to other chemistries. This research is motivated by two critical challenges: first, electrochemical battery models generally suffer from poor parameter identifiability, in the sense that it can be difficult to estimate their parameters accurately from input/output cycling data; and second, the phase changes that occur in LFP battery electrodes cause open-circuit battery voltage to be quite flat with respect to state of charge (SOC), thereby jeopardizing state observability. In a broad sense, this dissertation is motivated by the growing need for model-based battery monitoring and control, particularly for lithium-ion batteries. Lithium-ion batteries are quite attractive for many energy storage applications, thanks in part to their good power and energy densities. However, they are also vulnerable to aging/degradation/damage mechanisms such as lithium plating, dendrite formation, solid electrolyte interphase (SEI) layer growth, current collector corrosion, the isolation/loss of active electrode materials, etc. Existing work in the literature supports the benefits of model-based battery control. Some of these benefits include: (1) improved battery power density without compromising battery life, (2) extended battery life, (3) optimized fast-charging current trajectories, (4) diminished battery degradation effects including solid electrolyte interface (SEI) layer growth, (5) extended use of active material during charge/discharge cycles, (6) improved accuracy in state of charge and state of health estimation, and (7) optimized experimental protocols for model parameter identification. However, these benefits are subject to: accurate modeling of the (1) electrochemical and (2) thermal dynamics, as well as the coupling between them, (3) identifying model parameters, and (4) developing battery state estimators. The implementation of effective model-based control, requires: accurate modeling of the (1) electrochemical and (2) thermal dynamics, as well as the coupling between them, (3) identifying model parameters (e.g., capacity), and (4) developing battery state estimators (e.g., SOC). This dissertation provides evidence supporting the hypothesis that battery state/parameter estimation accuracy improves significantly when battery experiments are optimized for information-theoretic objectives such as Fisher identifiability. These arguments are developed using five different studies. This is an important conclusion because it makes it possible to optimize battery cycling for state/parameter estimation accuracy: a fact that has been explored in the literature, but mostly using theoretical/simulation studies. The research in this dissertation contributes to the two challenges described above through the following novel contributions. First, this dissertation presents an experimental method for estimating two lithium-ion battery cell parameters from experimental data: (i) the cell's reciprocal of the thermal time constant and (ii) its entropy coefficients for different states of charge (SOC). The experiment involves using an environmental chamber to apply a periodic ambient temperature profile to the given battery cell, measuring the resulting battery terminal voltage, and finally fitting the desired thermal parameters to experimental data using least squares estimation. Second, this dissertation extends thermal dynamic experiments and studies the problem of optimizing the experimental thermal cycle used for estimating the entropy coefficient of a battery cell. The goal is to maximize the Fisher identifiability of this entropy coefficient. Optimizing Fisher information enables improved estimation of battery entropy coefficients over the existing experimental protocols from the literature. This improvement is measured both by reducing lab time and by maintaining the variance of the identified entropy coefficients within the values obtained from two benchmark experiments. Third, this dissertation presents a study that derives and validates a current input trajectory to identify the parameters of a combined thermal and electrochemical lithium-ion battery model simultaneously. Specifically, the study optimizes a current input trajectory that maximizes the identifiability of the set of parameters for Fisher identifiability. The optimized current trajectory is validated in (i) simulation, and (ii) experimentally. Fourth, the dissertation presents an analysis of the effect of measurement uncertainty and model order mismatch on state of charge (SOC) estimation error in battery applications. Two different sources of measurement uncertainty are examined, namely, (i) uncertainty in voltage measurement (both noise and bias) and (ii) current measurement bias. Moreover, this research examines the SOC errors resulting from mismatch between the order of the battery model used for SOC estimation on the one hand, and the order of the true battery dynamics on the other. Analytic expressions for the SOC estimation (i) bias and (ii) variance induced by the above three sources of estimation error are derived. The analytic expressions are validated in simulation. The most valuable contribution of this research is to quantify the effect of measurement uncertainty and model order mismatch on SOC estimation. Fifth, the dissertation examines the relative contributions of bias and noise from measurement uncertainty to the overall SOC estimation error in a lithium-ion battery. The error analysis developed in this research is based on the analytical framework in study No. 4 and extends it to a second-order equivalent-circuit model. The validation of the analytical expressions is done experimentally. This study shows that in a practical laboratory setting, SOC estimation errors are much closer to our predictions of estimation bias as opposed to the Cram'er-Rao predictions of estimation noise.