Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
ON THE DEVELOPMENT OF
ELECTROCHEMICAL-BASED LITHIUM-
ION BATTERY MODELS FOR BATTERY
MANAGEMENT SYSTEMS
Ngoc Tham Tran
BEng, MEng
Submitted in fulfilment of the requirements for the degree of
Doctor of Philosophy
School of Electrical Engineering and Computer Science
Faculty of Science and Engineering
Queensland University of Technology
2019
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems i
Keywords
Battery management systems;
Grid-connected battery storage systems;
Lithium ion batteries;
Battery modelling;
Reduced order models electrochemical models;
Single particle models;
Extended single particle model;
Pseudo-two-dimension model;
Padé approximate model;
Thermal models;
Degradation models;
ii On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems
Abstract
Renewable energy sources such as wind and photovoltaics have been broadly
integrated on the power network to reduce the reliance on fossil fuels and achieve
desired carbon emissions targets. However, these power generation technologies are
inherently variable. This variability presents a major challenge for grid stability. Grid-
connected battery storage systems have been widely accepted as an effective solution
to this problem. The battery storage systems are capable of providing a rapid response
to counteract the fluctuations and filter out the variabilities associated with renewable
generation and therefore stabilize grid performance and maximize system security
benefits.
Typical battery management systems are used to control the charge and
discharge of the battery systems within a safe operation window. These battery
management systems often employ battery equivalent circuit models which have
limited ability to accurately represent battery dynamics, which can lead to situations
in which batteries are not fully utilized, or used optimally. This motivates the need for
a next generation of battery management systems with advanced features. These
advanced management systems employ electrochemical models which have greater
capabilities in terms of predicting battery dynamics, states and degradation. They offer
the promise of better battery resource utilization and prolonged system life. There are
existing works on these management systems, however, major issues related to
electrochemical-based state estimation and the development of sophisticated, whilst
computationally efficient, electrochemical models embodying coupled
electrochemical-thermal-degradation have not yet been addressed.
The main objective of this thesis is to construct high-performance, reduced order,
electrochemical models incorporating thermal and degradation phenomenon that can
be used in advanced battery management systems. These models have been developed
to satisfy the requirement of computationally simple, whilst maintaining their ability
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems iii
to accurately represent the major electrochemical and thermodynamical degradation
mechanisms of lithium ion batteries. The performance of these models is validated
numerically by comparing their output with solutions of a more sophisticated (and
computationally expensive) pseudo-two-dimensional electrochemical model.
Furthermore, an electrochemical-based adaptive state estimation algorithm is proposed
and validated via experiment.
iv On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery
Management Systems
Table of Contents
Keywords ........................................................................................................................ i
Abstract .......................................................................................................................... ii
Table of Contents .......................................................................................................... iv
List of Figures .............................................................................................................. vii
List of Tables ............................................................................................................... xii
List of Abbreviations .................................................................................................. xiii
Statement of Original Authorship ................................................................................ xv
Acknowledgements ..................................................................................................... xvi
Chapter 1: Introduction ................................................................................ 1
1.1 Background .......................................................................................................... 1
1.2 Research objectives.............................................................................................. 4
1.3 Original contributions .......................................................................................... 4
1.4 Thesis outline ....................................................................................................... 7
Chapter 2: Literature Review ....................................................................... 9
2.1 Empirical equivalent circuit models .................................................................... 9
2.2 Pseudo-two-dimensional electrochemical model .............................................. 14
2.3 Reduced order models of P2D model ................................................................ 18
2.4 Physics-based equivalent circuit models for lithium ion batteries ..................... 22
2.5 Thermal model ................................................................................................... 26
2.6 Degradation model ............................................................................................. 27
2.7 Battery state estimation techniques .................................................................... 29
2.8 Conclusion and research gaps ............................................................................ 30
Chapter 3: Single Particle Models and State Estimation by Sigma-point
Kalman Filter 33
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems v
3.1 Introduction ........................................................................................................33
3.2 Extended SPM and a practical implementation scheme for electrochemical
battery models .........................................................................................................................35
3.3 Extended single particle model and sigma-point Kalman filter .........................44
3.4 Experimental result and discussion ....................................................................51
3.5 Conclusions ........................................................................................................57
Chapter 4: Padé Approximation of Linearized P2D Model .................... 58
4.1 Introduction ........................................................................................................58
4.2 Analytic Laplace domain transfer functions of linearized P2D model ...............59
4.3 Padé approximation model .................................................................................63
4.4 Result and discussion .........................................................................................69
4.5 Conclusions ........................................................................................................79
Chapter 5: Coupled electrochemical-thermal model for small-format
lithium ion cells 81
5.1 Introduction ........................................................................................................81
5.2 Simplified thermal model for small-format cylindrical cells .............................82
5.3 Coupled electrochemical thermal model for small-format cell ..........................83
5.4 Result and discussion .........................................................................................85
5.5 Conclusions ........................................................................................................91
Chapter 6: A Computationally Efficient Electrochemical-thermal-
degradation Model for Large-format Lithium Ion Cells ..................................... 92
6.1 Introduction ........................................................................................................92
6.2 Model domain and reduced orders electrochemical model in each wind ...........93
6.3 Thermal approximation model ...........................................................................98
6.4 Coupled degradation-electrochemical-thermal models ....................................107
6.5 Results and discussion ......................................................................................109
6.6 Conclusions ......................................................................................................117
vi On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery
Management Systems
Chapter 7: Conclusions.............................................................................. 118
7.1 Summary and discussion ................................................................................. 118
7.2 Directions for further research ......................................................................... 120
Bibliography .................................................................................................. 123
Appendices ..................................................................................................... 131
Appendix A ................................................................................................................ 131
Appendix B ................................................................................................................ 133
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems vii
List of Figures
Figure 2.1 (a) Current pulse technique, (b) EIS technique, (c) A widely used
equivalent circuit model for lithium ion battery [18]. Reprinted from
"Experimental investigation of the lithium-ion battery impedance
characteristic at various conditions and aging states and its influence on
the application," by W. Waag, S. Käbitz, and D. U. Sauer, Applied
Energy, vol. 102, pp. 885-897, Copyright (2013), with permission from
Elsevier. ....................................................................................................... 10
Figure 2.2 Dependence of ohmic resistance (a), charge transfer resistance (b) and
double layer capacitance (c) on temperature, SOC and battery aging
[18]. Reprinted from "Experimental investigation of the lithium-ion
battery impedance characteristic at various conditions and aging states
and its influence on the application," by W. Waag, S. Käbitz, and D. U.
Sauer, Applied Energy, vol. 102, pp. 885-897, Copyright (2013), with
permission from Elsevier. ............................................................................ 12
Figure 2.3 Dependence of direct current resistance rDCR (rDCR = r0 + rct ) on
temperature, SOC and battery aging. (a) New cell, SOC =50%. (b) Aged
cell, SOC =50%. (c) New cell, temperature =10ºC. (b) Aged cell,
temperature =10ºC. [18] Reprinted from "Experimental investigation of
the lithium-ion battery impedance characteristic at various conditions
and aging states and its influence on the application," by W. Waag, S.
Käbitz, and D. U. Sauer, Applied Energy, vol. 102, pp. 885-897,
Copyright (2013), with permission from Elsevier. ...................................... 13
Figure 2.4 Schematic of one-dimensional electrochemical cell model of a lithium
ion cell [28] .................................................................................................. 15
Figure 2.5 Single Particle Model of a Li-ion battery [4]............................................ 19
Figure 2.6 The 1D cell diagram with transfer functions in each domain [7] ............. 21
viii On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery
Management Systems
Figure 2.7 Equivalent circuit for a porous electrode comprised of three finite
regions, adapted from [44] ........................................................................... 23
Figure 2.8 P2D equivalent circuit network implementation [46]. Reprinted from
" A physically meaningful equivalent circuit network model of a
lithium-ion battery accounting for local electrochemical and thermal
behaviour, variable double layer capacitance and degradation," by M.-
T. von Srbik, M. Marinescu, R. F. Martinez-Botas, and G. J. Offer,
Journal of Power Sources, vol. 325, pp. 171-184, Copyright (2016),
with permission from Elsevier ..................................................................... 24
Figure 2.9 Cell level scheme of the simplified physics-based equivalent circuit
model [47]. Reprinted from " An easy-to-parameterise physics-
informed battery model and its application towards lithium-ion battery
cell design, diagnosis, and degradation," by Y. Merla, B. Wu, V. Yufit,
R. F. Martinez-Botas, and G. J. Offer, Journal of Power Sources, vol.
384, pp. 66-79, Copyright (2018), with permission from Elsevier .............. 25
Figure 2.10 The list of commonly reported degradation mechanisms, mode,
cause and effect, adapted from [63] ............................................................. 28
Figure 2.11 General principle of Kalman filter algorithm. This figure is adapted
from [74] ...................................................................................................... 30
Figure 3.1 Simulink model of a lithium ion cell connected to a load ........................ 39
Figure 3.2 Equivalent circuit of the SPM-3P for lithium ion cell .............................. 41
Figure 3.3 Cell voltage response during 1C discharge rate ........................................ 42
Figure 3.4 Cell voltage response during 5C discharge rate ........................................ 42
Figure 3.5 Cell voltage response during 7C discharge rate ........................................ 43
Figure 3.6 Cell voltage response during 10C discharge rate ...................................... 43
Figure 3.7 Electrolyte concentration in the separator and positive electrode
during the dynamic charge/discharge current profile. ................................. 46
Figure 3.8 Diagram of the SPKF algorithm ............................................................... 50
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems ix
Figure 3.9 Experiment setup to test the half-cell ....................................................... 53
Figure 3.10 (a) Current rate, (b) comparison between the reference SOC and
SOC reference and (c) SOC estimation error .............................................. 54
Figure 3.11 (a) Comparison of the voltage estimation and the measured voltage
and (b) the voltage estimation error. ............................................................ 55
Figure 3.12 OCP-SOC relationship of LiFePO4 positive electrode. .......................... 56
Figure 4.1 Schematic diagram of a lithium ion cell embodying the P2D model
taken from [4] and transfer function of each variable in each domain
based on [7]. ................................................................................................. 60
Figure 4.2 A low order Padé approximation model of the linear P2D model. .......... 66
Figure 4.3 Frequency response of the Padé approximants and transcendental
transfer functions of the four variables within the negative electrode. ........ 67
Figure 4.4 Current input and corresponding frequency content of a battery
storage system that is being used to smooth the power generated from
a wind farm connected to a grid. Current data sourced from [86]. .............. 69
Figure 4.5 Comparison between the Padé approximation model and the P2D
model for a pulse current profile .................................................................. 71
Figure 4.6 Comparison between the Padé approximation model and the P2D
model for the UDDS applied current profile. .............................................. 72
Figure 4.7 Comparison between the Padé approximation model with the full
(nonlinear) P2D model for the applied current profile associated with
the wind farm application, as depicted in Figure 4.4(a). ............................. 73
Figure 4.8 Comparison between two charging methods (a). Cell voltage during
charging. (b). The over potential s e at negx L ..................................... 76
Figure 4.9 Higher order Padé approximation model of the linear P2D model .......... 77
Figure 4.10 Cell voltage error obtained from low order and higher order Padé
approximation models when compared to the predictions of the P2D
model............................................................................................................ 79
x On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems
Figure 5.1 A schematic representation of the proposed cell model ........................... 84
Figure 5.2 (a) Cross-section view of a small-format cylindrical cell which shows
negligible temperature variation along the cell’s radius, (b) Cross-
section view of a large-format cylindrical cell considered in [94] which
shows significant temperature derivation along the cell’s radius ................ 85
Figure 5.3 Comparison between the proposed cell model and the rigorous 1D
radial PDE model in Comsol using1C charging/discharging of a small-
format cell (18650 cell) ................................................................................ 88
Figure 5.4 Comparison between the proposed cell model and the rigorous 1D
radial PDE model in Comsol using UDDS current profile of a small-
format cell (18650 cell) ................................................................................ 89
Figure 5.5 Temperature error between the proposed cell model and the Comsol
1D radial PDE model: 1C charging/discharging of a large-format cell. ...... 90
Figure 5.6 Temperature error between the proposed cell model and the rigorous
1D radial PDE model in Comsol using UDDS current profile of a large-
format cell. ................................................................................................... 90
Figure 6.1 (a) Illustration of a cylindrical lithium ion battery with spirally wound
design and its cross-sectional view, (b) the component layers in each
wind and (c) the domains that constitute half of each wind and the
corresponding Padé model transfer functions [59]. ..................................... 95
Figure 6.2 Illustration of cross section of the cylindrical lithium ion battery with
boundary conditions ................................................................................... 102
Figure 6.3 Parallel connection of Padé models ........................................................ 108
Figure 6.4 A schematic representation of the DET cell ........................................... 108
Figure 6.5 Simulation result of temperature distribution at different convection
coefficients from DET model during a 1C charge-rest-discharge-rest
cycle. .......................................................................................................... 110
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems xi
Figure 6.6 Simulation result of temperature distribution at different convection
coefficients from the DET model during UDDS cycles. ........................... 111
Figure 6.7 Comparison of temperature variation between DET model and the
full P2D model in COMSOL at aRr (outer wind) and
0Rr (inner
wind) at different convection coefficients during a 1C charge-rest-
discharge-rest cycle. ................................................................................... 112
Figure 6.8 Comparison of temperature variation between DET model and the
full P2D model in COMSOL at ar R (outer wind) and
0r R (inner
wind) at different convection coefficients during UDDS cycles. .............. 113
Figure 6.9 Evolution SEI resistance filmR and maximum concentration of lithium
ion s,maxc in solid phase in negative electrode of inner and outer winds (
2100W m Kh ). ........................................................................................ 116
xii On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery
Management Systems
List of Tables
Table 2.1 The matrix showing the tests used in [18]. Reprinted from
"Experimental investigation of the lithium-ion battery impedance
characteristic at various conditions and aging states and its influence on
the application," by W. Waag, S. Käbitz, and D. U. Sauer, Applied
Energy, vol. 102, pp. 885-897, Copyright (2013), with permission from
Elsevier. ........................................................................................................ 11
Table 3.1 FreedomCAR Cell Model parameters [27] ................................................ 39
Table 3.2 Half-Cell Model Parameters ...................................................................... 48
Table 4.1 Cell parameters used in the simulation ...................................................... 70
Table 4.2 Comparison of simulation time of the proposed approximation model
to compute all cell variables at all discrete spatial locations ....................... 74
Table 4.3 Comparison of simulation time of the proposed approximation model
to compute only cell voltage ........................................................................ 75
Table 4.4 Comparison of RMS error of cell voltage and the computational
workload between low order and higher order Padé approximation
models .......................................................................................................... 78
Table 5.1 Comparison of simulation time .................................................................. 86
Table 6.1 Comparison of simulation time ................................................................ 114
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems xiii
List of Abbreviations
SOC State of charge
BMS Battery management system
ABMS Advanced battery management system
UDDS Urban dynamometer driving scheduler
P2D Pseudo-two-dimensional electrochemical model
BESS Battery energy storage system
PDE Partial differential equations
SEI Solid electrolyte interphase
OCV Open circuit voltage
OCP Open circuit potential
SPKF Sigma-point Kalman filter
EIS Electrochemical impedance spectroscopy
SPM Single particle model
ESPM Extended single particle model
DRA Discrete-time realization
FVM Finite volume method
EKF Extended Kalman filter
FEM Finite element method
CC Constant current
KF Kalman filter
CV Constant voltage
MD-MS Multi-dimensional, multi-scale
xiv On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery
Management Systems
DET Degradation-electrochemical-thermal model
NMPC Nonlinear model predictive control
DAE Differential algebraic equations
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems xv
Statement of Original Authorship
The work contained in this thesis has not been previously submitted to meet
requirements for an award at this or any other higher education institution. To the best
of my knowledge and belief, the thesis contains no material previously published or
written by another person except where due reference is made.
Signature:
Date: ___21/6/2019____________
QUT Verified Signature
xvi On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery
Management Systems
Acknowledgements
I would like to express my deepest appreciation and gratitude to my supervisors,
Prof. Mahinda Vilathgamuwa, and Prof. Troy Farrell. This dissertation would not have
been possible without their expertise, continual guidance, encouragement and support.
I would like to thank them for giving me the opportunity to pursue a PhD and work on
this innovative and advanced research project. I am extremely thankful for this
incredible experience.
I would like to extend my sincere gratitude to my external advisor Prof. San
Shing Choi for his persistent guidance and valuable advises he provided for the
improvement of this dissertation and other published articles.
I would like to acknowledge the organisations that funded my research, namely,
the Queensland University of Technology for QUT Postgraduate Research Award,
QUT Higher Degree Research Tuition Fee Sponsorship, QUT Excellence Top Up
Scholarship, QUT HDR travel grant and other resources, IEEE for student travel
grants, and the Australian Research Council.
I would like to thank Prof. Peter Talbot, Prof. Jose Alarco, Dr. Jawahar Nerkar,
Assoc. Prof. Hongxia Wang and Dr. Teng Wang for their support with the
experimental results. I would like to thank QUT School of Electrical Engineering and
Computer Science, Power engineering discipline, School of Mathematics, and High
Performance Computing group for their administrative and technical support.
I am grateful to Dr. Yang Li, Joseph Teague, Prof. Colin Please, Assoc. Prof.
Paul Corry, Dr. Shawn Neilsen and Assoc. Prof. Geoff Walker for their in-depth
discussions and contributions to improve the quality of this dissertation and other
publications.
I wish to thank committee members of my confirmation and final seminars
including Prof. Gerard Ledwich, Prof. Jose Alarco, Dr Yateendra Mishra and Assoc.
Prof. Hongxia Wang for their insightful comments and encouragement.
On The Development of Electrochemical-Based Lithium-Ion Battery Models For Battery Management
Systems xvii
Last but not least, I would like to thank my parents and my brother for their love
and encouragement during all my educational endeavours. To my partner Quynh
Phuong Lai thank you for her love and patience thorough this journey. I dedicate this
dissertation to you all, my beloved ones.
Chapter 1: Introduction 1
Chapter 1: Introduction
1.1 BACKGROUND
Lithium ion batteries are now commonly used in conjunction with renewable
generation systems, such as wind farms and photovoltaic stations. The role of the
batteries is to alleviate negative impacts the renewable generators may have on power
systems due to the inherent uncertainties and intermittency of the renewable sources
[1]. A typical example is the Hornsdale Power Reserve battery storage system
(129MWh capacity and rated at 100MW of discharge power) which has been effective
in providing network security services to South Australian grid system since 2017 [2].
A Battery Management System (BMS) is essential in order to utilize battery
storage capacity optimally, achieve safe charging and discharging of the battery system
and prolong the battery lifetime [3]. In this connection, optimal control algorithms in
advanced BMSs will necessitate models that can provide insight into cell internal
electrochemical variables, thermal dynamics and degradation [4]. An accurate battery
model that can represent the complex and nonlinear internal characteristics of batteries
is therefore desirable.
Depending on the application, the capacity of grid-connected battery energy
storage systems (BESSs) can vary from kWh to MWh [1]. Given that a single cell has
typically only 7.5Wh energy capacity [5], the BESSs are usually constructed from
many battery packs, with each pack consisting of large numbers of series- and parallel-
connected lithium ion cells to achieve the required voltage and capacity level [6]. As
a result, not only is a highly accurate battery model desirable but the model itself must
also be able to be evaluated quickly, using highly efficient computational algorithm.
Current BMSs rely significantly on equivalent circuit models of the battery due
to the simplicity of the models and the inherently low computational burden required
to implement them [7]. However, the models give only limited information on the
electrochemical characteristics and degradation response of the battery [8]. In addition,
2 Chapter 1: Introduction
equivalent circuit models require extensive battery testing in order to validate them in
a particular operational range in which the battery circuit parameters have been
identified [8]. As an alternative to equivalent circuit models, the pseudo 2-dimensional
electrochemical (P2D) model, first developed by Doyle, Fuller, and Newman [9, 10],
does provide detailed insights on these characteristics [8]. However, the P2D model
consists of partial differential equations (PDEs). To solve these, spatial discretization
methods are applied to yield a system of differential algebraic equations (DAEs) in the
time domain [4]. These methods are computationally expensive and are usually
implemented on a desktop computer, often using commercial software. Therefore, in
industrial applications, the methods are not suitable for embedded BMSs which require
much lower computational workloads. For example, as reported by Lee, Chemistruck
and Plett [7], the P2D model implemented in COMSOL Multiphysics [11] requires
about 13 min on a desktop computer, to produce 25 min of simulated voltage response
to the Urban Dynamometer Driving Scheduler (UDDS) current profile [12]. Such
computational requirements render a direct implementation of the P2D model
infeasible for use in a BMS.
On thermal management, information on heat generation is fundamentally
important for managing thermal issues such as thermal runaway, electrical cell
unbalance within the battery pack and poor performance at low temperatures [13].
However, heat generation inside a cell is a complex process that requires the
knowledge of the physical characteristics of the cell during its operation. Total heat
generation is due to irreversible and reversible processes, Joule heating in the solid and
electrolyte and heating from the electrode/current collector contact resistance [8]. It is
accompanied by changes in the electrochemical properties of the cell, such as entropy,
solid and electrolyte concentrations due to chemical and electrochemical reactions and
changes in the solid and electrolyte potentials.
Another important issue is the degradation of battery performance over time and
usage. The degradation is mainly caused by the formation and growth of the solid
electrolyte interphase layer (SEI), which scavenges active lithium ions and electrolyte
materials and increases battery resistance, leading to capacity and power fades,
respectively [14]. SEI growth is also coupled with the thermal behaviour of the cell as
Chapter 1: Introduction 3
higher temperatures increase the SEI growth rate which in turn causes higher SEI
resistance and ohmic heat generation, which elevates the temperature [15].
Thermal and degradation issues become more significant in large format
cylindrical batteries because non-uniformities in temperature and degradation occur
when the batteries operate [16]. Such non-uniformities exacerbate further degradation
and temperature gradients within the batteries. Current BMSs, which rely on
equivalent circuit models, have limited insight on the electrochemical characteristics
of a battery and, given the close coupling between the two, equivalent circuit models
are not able to model battery thermal behaviour and degradation mechanism precisely
[8].
In addition, information about battery state of charge (SOC) is often used by
BMSs to predict the available energy remaining in the batteries [4]. SOC can be
tracked by observing the ratio of the average lithium ion concentration to the maximum
concentration of the solid phase. This immeasurable quantity is a standard input for
control algorithms of the BMSs. A precise SOC information of the battery is required
since it can decide the effectiveness of the control algorithms. Studies on online SOC
estimation algorithms based on equivalent circuit model are immensely popular, those
based on physics first principle battery models have not yet matured and remains a
challenge to be considered.
The main goal of this thesis is to develop new electrochemical reduced order
battery models while taking into consideration the thermal and degradation dynamics
within the cells. Furthermore, a new adaptive electrochemical model-based SOC
estimation algorithm is also proposed. The developed models and estimation algorithm
have to be accurate and computationally efficient for application in grid-connected
battery storage systems. Given this, the research objectives of this work, aimed at
facilitating this goal, are outlined in the next section.
4 Chapter 1: Introduction
1.2 RESEARCH OBJECTIVES
To achieve the aforementioned goal, this thesis will be focused on the following
four objectives:
Objective 1: Develop a new SOC estimation algorithm using an electrochemical
model and an adaptive observer. This objective leads to the contribution 1 which is
described in Section 1.3.
Objective 2: Derive a new simplified model of Li-ion battery based on the
linearized pseudo-two-dimensional electrochemical model (P2D) using Padé
approximation method. This objective leads to the contribution 2 which is described
in Section 1.3.
Objective 3: Develop a new computationally efficient electrochemical-thermal
model for small-format lithium ion cells by incorporating a simplified battery thermal
model into the Padé approximation model. This objective leads to the contribution 3
which is described in Section 1.3.
Objective 4: Develop a new sophisticated electrochemical-thermal-degradation
model for large-format lithium ion cells by coupling together the Padé approximation
model, an approximated radial thermal model and degradation model. This objective
leads to the contribution 4 which is described in Section 1.3.
1.3 ORIGINAL CONTRIBUTIONS
Working toward achieving the objective of the study, the following original
contributions have been made:
Contribution 1:
The first contribution of the thesis is the development of a new adaptive state
estimation algorithm using the combination of the extended single particle and sigma-
point Kalman filter and the development of a practical scheme to implement the single
particle model in Simulink. The feasibility of the proposed model is validated by
comparing the estimated cell voltage and state of charge of the cell obtained with that
Chapter 1: Introduction 5
from test measurements. For this contribution, the associated published conference
papers are:
N. T. Tran, M. Vilathgamuwa, T. Farrell, and S. S. Choi, "Matlab simulation
of lithium ion cell using electrochemical single particle model," in 2016 IEEE 2nd
Annual Southern Power Electronics Conference, p. 210. (IEEE student travel grant
prize)
N. T. Tran, M. Vilathgamuwa, Y. Li, T. Farrell, S. S. Choi, and J. Teague, "State
of charge estimation of lithium ion batteries using an extended single particle model
and sigma-point Kalman filter," in 2017 IEEE 3rd Annual Southern Power Electronics
Conference, pp. 1-6. (IEEE student travel grant prize)
Contribution 2:
The second contribution of the thesis is the development of a new reduced cell
model by applying Padé approximations to simplify the complicated transcendental
transfer functions of the linearized P2D model to rational polynomial transfer
functions that are amenable to rapid computation for embedded BMS applications. The
reduced model output includes the spatial and temporal variation of all of the state
variables of the P2D model. Consequently, it can be used to determine any of the state
variables at any specific spatial location. There is no need to compute the state
variables at all of the spatial locations, as is the case when using the spatial
discretization methods to solve the P2D model. This proposed model is the first
attempt at using the Padé approximation method to simplify the linearized P2D model
whilst preserving physical meaning of the model. The model overcomes the limitation
of single particle models when applied to thick electrodes high-energy lithium ion
batteries used in grid-connected applications. The proposed model is validated by
comparing to the result obtained with that based on the P2D model solved by finite
volume method and implemented using an open source code. For this contribution, the
associated published journal paper is:
N. T. Tran, M. Vilathgamuwa, T. Farrell, S. S. Choi, Y. Li, and J. Teague, "A
Padé Approximate Model of Lithium Ion Batteries," Journal of the Electrochemical
Society, vol. 165, no. 7, pp. A1409-A1421, January 1, 2018. (Q1)
6 Chapter 1: Introduction
Contribution 3:
The third contribution of the thesis is the development of a new computationally
efficient electrochemical-thermal model of small-format cylindrical lithium ion cells.
This proposed model takes advantage of the aforementioned Padé approximation
model by computing only electrochemical variables at specific spatial locations. The
Padé electrochemical model is then combined with a simplified thermal model of
small-format cylindrical lithium ion cells to form a new model which requires low
computational burden. The proposed model is validated by comparing to the results
obtained with those using the full, 1-dimensional, radial implementation of the P2D
model implemented in Comsol Multiphysics®. For this contribution, the associated
accepted conference paper is:
N. T. Tran, M. Vilathgamuwa, T. Farrell, S. S. Choi, Y. Li, and J. Teague, "A
Computationally-Efficient Electrochemical-Thermal Model for Small-Format
Cylindrical Lithium Ion Batteries," the 2018 IEEE 4th Annual Southern Power
Electronics Conference, Singapore. (IEEE student travel grant prize)
Contribution 4:
The fourth contribution of the thesis is the development of a new and
sophisticated electrochemical-thermal-degradation model for large-format spirally
wound, cylindrical lithium ion cells. It takes into account the non-uniform degradation
and thermal behaviour within the cells. This proposed model is able to provide insight
into the variation and evolution of the local temperature and degradation rate of each
individual wind along the cell radius. The performance of the proposed model is
validated by comparing to the full, 1-dimensional, radial implementation of the P2D
model implemented in Comsol Multiphysics®. For this contribution, the associated
manuscript to be submitted to a journal is:
N. T. Tran, M. Vilathgamuwa, T. Farrell, S. S. Choi, and Y. Li, "A
Computationally Efficient Electrochemical-Thermal Model for Large Format
Cylindrical Lithium Ion Batteries," 2019. (Under review, the Journal of the
Electrochemical Society, Q1)
Chapter 1: Introduction 7
1.4 THESIS OUTLINE
The remainder of this thesis is organized as follows.
In Chapter 2, the key literature related to battery empirical equivalent circuit
models, electrochemical models, degradation, thermal models and battery state
estimation techniques are critically analysed. Areas where further research are needed
are identified at the end of this chapter.
Chapter 3 is aimed at developing an adaptive electrochemical model-based
battery state estimation algorithm. A practical scheme to implement the
electrochemical model in Simulink is also developed. This scheme is then used in
sophisticated models proposed in Chapters 5 and 6 where thermal and degradation
models are coupled with reduced order electrochemical models. Chapter 3 addresses
the Objective 1 of the thesis.
Chapter 4 focuses on simplifying the complex transcendental transfer functions
relating the electrochemical variables of the linearized P2D model into rational transfer
functions. Padé approximation is used to substantially reduce the complexity of the
P2D model whilst still maintaining a reasonable level of accuracy. This approximation
model overcomes limitations of single particle models in high energy cells of thick
electrodes where the current density is non-uniform. This chapter addresses the
Objective 2 of the thesis.
In Chapter 5, a computationally efficient electrochemical-thermal model for
small-format lithium ion cells is developed. This model is based on the Padé
approximation model proposed in Chapter 4, but incorporates a simplified thermal
model of the cells. Temperature variations within the small-formatted cylindrical cells
are ignored. The practical scheme proposed in Chapter 3 is also used to implement the
proposed model in Simulink. Chapter 5 addresses the Objective 3 of the thesis.
Chapter 6 provides the details on the development of a novel electrochemical-
thermal-degradation model for large-format lithium ion cells. The Padé approximation
model proposed in Chapter 4 is coupled with a battery degradation model and a
simplified radial thermal model. This is in order to efficiently represent the dynamics
and the non-uniform temperature and degradation in the cylindrical cells. The practical
8 Chapter 1: Introduction
scheme of Chapter 3 is also used here to implement this sophisticated model in
Simulink. Chapter 6 addresses the Objective 4 of the thesis.
Chapter 7 concludes the main findings of this thesis and presents suggestions
for future works.
Chapter 2: Literature Review 9
Chapter 2: Literature Review
This chapter provides an overview of published works on lithium ion battery
modelling. Sections 2.1 to 2.7 will review topics in the literature related to the
modelling of lithium ion batteries. In Section 2.8 conclusions from this review will be
drawn noting the research gaps and discussing their connection to the objectives of
this thesis.
2.1 EMPIRICAL EQUIVALENT CIRCUIT MODELS
Empirical equivalent circuit battery models have been used extensively in BMS
due to their simplicity and low computational requirement. The battery models can be
represented by an Open Circuit Voltage (OCV) in series with battery impedance
parameters including an ohmic resistance 0r , a parallel circuit consisting of the charge
transfer resistance ctr and a double layer capacitance
dlC , as depicted in Figure 2.1(c)
[17]. The battery output voltage can be calculated by the OCV and the voltage drop
across the battery impedance when an input current is applied. The parametric values
of these battery impedance parameters depend significantly on the battery SOC,
operating current rates, state of cell degradation and temperature [18]. These
dependencies can be determined from tests using such techniques as the current pulse
test [18] and/or the Electrochemical Impedance Spectroscopy (EIS) test [18-20], as
depicted in Figure 2.1(a) and Figure 2.1(b), respectively. In the current pulse technique
[18], the battery voltage drop 0V at the moment of applying the input current occurs
due to the ohmic resistance. The subsequent voltage decay 1V occurs across the
parallel ctr and
dlC circuit of the charge transfer resistance. In the EIS test [18], a small
signal excitation current ( )i f at a given frequency f is applied to the battery and the
voltage response of the battery ( )v f is measured. The complex impedance of the
battery at the frequency f can be expressed as ( )( )
( )v f
z fi f
. By applying a range of
frequencies, the battery impedance spectrum can be obtained. As shown in Figure
10 Chapter 2: Literature Review
2.1(b), 0r is determined at the point where Im ( ) 0z f . The resistance
dr is
determined at the minimum local point of Im ( )z f . The resistance ctr is the
difference between dr and
0r (ct d 0r r r ). The time constant τ (
ct dlr C ) is determined
by the maximum local point of Im ( )z f ( max2 f ). Based on these techniques,
the information of the battery impedance dependency on different input current rates,
SOC, degradation and temperature can be obtained [18].
Figure 2.1 (a) Current pulse technique, (b) EIS technique, (c) A widely used
equivalent circuit model for lithium ion battery [18]. Reprinted from "Experimental
investigation of the lithium-ion battery impedance characteristic at various conditions and
aging states and its influence on the application," by W. Waag, S. Käbitz, and D. U. Sauer,
Applied Energy, vol. 102, pp. 885-897, Copyright (2013), with permission from Elsevier.
The work of W. Waag et al. [18] showed an experimental investigation of battery
impedance dependency on various operating conditions and degradation levels by
using both current pulse technique (CP) and EIS technique (IS). The test method and
results reported by W. Waag et al. [18] are shown in Table 2.1, Figure 2.1, Figure 2.2
and Figure 2.3. Table 2.1 shows the matrix of CP and IS tests at different temperature
Chapter 2: Literature Review 11
and SOCs. In each CP test, the test is repeated with ten different current rates ranging
from 0.25C to 4.25C. After testing the battery, the battery impedance is extracted from
the measured current and voltage data. Figure 2.2 and Figure 2.3 show the result of
these tests in which the battery impedance dependency on current rates, SOC,
degradation and temperature is obtained. By using these tests, full characterization of
the battery impedance can be parameterized and stored in look-up tables for battery
modelling and state estimation in the BMS, as did numerous works [21-23]. However,
the accuracy of the empirical equivalent circuit models strongly relies on operating
conditions in which the batteries have been tested and parametrized [8]. In other words,
the batteries need to be tested in various operation conditions to determine the change
in battery impedance as can be seen in the mentioned example. Thus, many tests which
are costly and time consuming need to be conducted in order to broaden the model
validity range [8]. Furthermore, the same tests need to be repeated if a different battery
type is to be modelled.
Table 2.1 The matrix showing the tests used in [18]. Reprinted from "Experimental
investigation of the lithium-ion battery impedance characteristic at various conditions and
aging states and its influence on the application," by W. Waag, S. Käbitz, and D. U. Sauer,
Applied Energy, vol. 102, pp. 885-897, Copyright (2013), with permission from Elsevier.
10% 20% 30% 50% 70% 90% 100%
-10°C IS IS IS IS IS IS IS
0°C IS, CP IS IS, CP IS, CP IS, CP IS, CP IS
10°C IS, CP IS IS, CP IS, CP IS, CP IS, CP IS
25°C IS, CP IS IS, CP IS, CP IS, CP IS, CP IS
40°C IS IS IS IS IS IS IS
12 Chapter 2: Literature Review
Figure 2.2 Dependence of ohmic resistance (a), charge transfer resistance (b) and
double layer capacitance (c) on temperature, SOC and battery aging [18]. Reprinted from
"Experimental investigation of the lithium-ion battery impedance characteristic at various
conditions and aging states and its influence on the application," by W. Waag, S. Käbitz, and
D. U. Sauer, Applied Energy, vol. 102, pp. 885-897, Copyright (2013), with permission from
Elsevier.
Chapter 2: Literature Review 13
Figure 2.3 Dependence of direct current resistance rDCR (rDCR = r0 + rct ) on
temperature, SOC and battery aging. (a) New cell, SOC =50%. (b) Aged cell, SOC =50%.
(c) New cell, temperature =10ºC. (b) Aged cell, temperature =10ºC. [18] Reprinted from
"Experimental investigation of the lithium-ion battery impedance characteristic at various
conditions and aging states and its influence on the application," by W. Waag, S. Käbitz, and
D. U. Sauer, Applied Energy, vol. 102, pp. 885-897, Copyright (2013), with permission from
Elsevier.
Alternatively, battery impedance can be estimated online by using parameter
estimation algorithms such as weighted recursive least squares [24, 25] and Dual
Extended Kalman filter [26]. These methods use measurable quantities such as battery
current as an input and battery voltage as a reference to adaptively update the battery
impedance values so that the errors between the battery voltages calculated by the
model and the measured voltages are minimized. Battery SOC can also be estimated
using the same approach and will be discussed in detail in Section 2.7. However, the
14 Chapter 2: Literature Review
parameter estimation algorithm becomes complicated and its complexity level could
be as high as the electrochemical model itself (the model which will be reviewed in
the later part) [22]. Importantly, the empirical equivalent circuit model does not
provide insight into internal electrochemical variables of the batteries which are the
key to explain the battery degradation [7].
2.2 PSEUDO-TWO-DIMENSIONAL ELECTROCHEMICAL MODEL
The pseudo-2 dimensional electrochemical model (P2D) for lithium ion
batteries, first developed by Doyle, Fuller, and Newman [9, 10], does provide insight
into battery electrochemical behaviour. The P2D model is applied to four domain
elements of a lithium ion battery, namely, the negative electrode, the separator, the
electrolyte and the positive electrode [4]. During discharge, lithium ions in the negative
electrode diffuse from the interior to the surface of the spherical solid particles that
constitute the porous electrode, where they undergo an electrochemical reaction and
are transferred into the electrolyte phase. Thereafter, the ions travel through the
electrolyte solution to the positive electrode where they again react at the surfaces of
the solid particles (constituting that porous electrode) and are intercalated into these
particles through the solid electrolyte interphase (SEI) between the electrolyte solution
and the solid particles. These processes are shown schematically in Figure 2.4 along
with the definition of the modelling domain in which the negative electrode (neg) is
defined from 0x to negx L ; the separator (sep) is defined from negx L to
neg sepx L L and the positive electrode (pos) is defined from neg sepx L L to totx L .
The reverse of the depicted discharge process occurs in the charge process, whereby
lithium ions flow from the solid phase of the positive electrode to the solid phase of
the negative electrode. As the separator forms an electrically insulated barrier between
the electrodes, the flow of electrons associated with the charge and discharge processes
occurs via an external circuit connecting the two compartments, doing useful work as
they do so [27].
Chapter 2: Literature Review 15
Figure 2.4 Schematic of one-dimensional electrochemical cell model of a lithium ion
cell [28]
The P2D model that describes the above processes consists of five state variables
including the electric potential ,s x t in the solid electrode, the electric potential
,e x t in the electrolyte, the lithium concentration of the active material , ,sc x r t of
the positive and negative electrodes, the lithium concentration ,ec x t in the
electrolyte, and the molar fluxes ,j x t of the charge that flows between the active
material in each electrode and the electrolyte. The governing equations of the model
are given as follows [4, 7] with the definitions of the symbols shown in the equations
being given in the Appendix B.
The potential in the solid phase of each (negative and positive) electrode is derived
from the principle of conservation of charge and is given by (for i neg, pos ),
i
seff,i
app e
,,i
x tI i x t
x
, (2.1)
with the boundary conditions
i i
s seff,i eff,i0, ,tot
app
t L tI
x x
.
i i
s s2
s2
, , , ,1c x r t c x r tD r
t r rr
negi posi sepi
i i
s s2
s2
, , , ,1c x r t c x r tD r
t r rr
e¯appI
Li
Li
e¯
appI
Li
Li
i
seff,i
e
,,i
app
x tI i x t
x
i
seff,i
e
,,i
app
x tI i x t
x
e i i
s
,,
ii x ta F j x t
x
i i ie e eeff,i 0
e e
, , 1,i
c x t c x tD t i x t
t x x F x
SEI
Electrolyte
Solid
D
i
e eeff,i eff,i
e
, ln ,,
i
ix t c x t
i x tx x
e ,ii x t
Negative electrode Separator Positive electrode
0x neg sepx L L neg sep pos totx L L L L negx L
tot
cell s s, 0,V t L t t
16 Chapter 2: Literature Review
The electrolyte current is defined along the whole cell domain and is given by,
(for i neg, sep,pos and sep , 0j x t ):
e i i
s
,,
ii x ta F j x t
x
(2.1)
The electrolyte potential is defined along the whole cell domain and is given by
(for i neg, sep,pos ),
D
i
e eeff,i eff,i
e
, ln ,,
i
ix t c x t
i x tx x
, (2.2)
where,
eff,i eff,i 0 /D i
e
ln21 1
ln
c ad fRTt
F d c
,
with the boundary conditions
i i
e e0, ,0
tott L t
x x
.
Lithium ions are transported in the solid particles of each electrode by diffusion and
their concentration is given by (for i neg, pos ),
i i
s s2
s2
, , , ,1c x r t c x r tD r
t r r r
, (2.3)
where the boundary conditions and initial condition are given by
i
s
s
,0,0
c x tD
r
,
i
s s
s
, ,,i
c x R tD j x t
r
and i i
s s,0, ,0c x r c .
The equation for the lithium ion concentration in the electrolyte phase (across the
whole cell domain) is derived from the principle of conservation of mass in which the
change of electrolyte concentration is due to the diffusion and migration of the ions and
their transfer at the solid/solution interfaces (for i neg, sep,pos and sep , 0j x t ),
namely,
i i ie e eeff,i 0
e e
, , 1,i
c x t c x tD t i x t
t x x F x
. (2.4)
Chapter 2: Literature Review 17
The boundary conditions and initial condition are
i i
e e0, ,0
totc t c L t
x x
and i i
e e,0,0c x c , respectively.
The transfer of charge at the solid/electrolyte interfaces in each electrode is given by
(for i neg, pos ),
11i eff,i i i i i
e s,max s,e s,e
i i
, , , ,
1exp , exp ,
j x t r c x t c c x t c x t
F Fx t x t
RT RT
, (2.5)
where, i
s,ec is the surface concentration of lithium in a spherical electrode particle.
The overpotential of the intercalation reactions in each electrode is given by (for
i neg, pos ),
i i i i i i i
s e OCP s,e film, , , , ,x t x t x t U c x t F R j x t . (2.6)
The potential difference between the positive and the negative current collectors
yields the cell voltage, namely,
tot
cell s s, 0,V t L t t . (2.7)
Equations (2.1) – (2.7) constitute a system of coupled PDEs and algebraic
equations that represent the P2D model for a lithium ion battery. Under galvanostatic
conditions, the input of the model is the applied current appI t and the output of the
model is the cell voltage cellV t . In the separator ( neg neg sepL x L L ), the model
involves two coupled PDEs which are the electrolyte concentration given by Eq. (2.4)
and the electrolyte potential given by Eq. (2.2). In the electrodes ( neg0 x L ,
neg sep totL L x L ), the model involves the coupled PDEs for electrolyte concentration,
given by Eq. (2.4), electrolyte potential, given by Eq. (2.2) and solid phase potential,
given by Eq. (2.1) on the macro (cell) scale (x) and Eq. (2.3) on the micro (particle)
scale (r) (which exists at every x in the electrode domains) [29]. The nonlinear nature
of the P2D model means that it must be solved numerically. As exemplified by the work
of Farrell and coworkers [30, 31], this is generally achieved using a desktop computer
18 Chapter 2: Literature Review
running finite volume or finite element software. As noted earlier, this complexity
prevents the implementation of the P2D model in embedded real-time applications
such as BMSs.
2.3 REDUCED ORDER MODELS OF P2D MODEL
The P2D model introduced above is an accurate model to describe battery
operation [9, 10, 32]. With this model, the battery terminal voltage can be accurately
predicted. However, the computational cost associated with solving the coupled, non-
linear partial differential equations that define the model is very high in comparison
with the battery equivalent circuit models. A simplified electrochemical model that
has lower computational overheads, whilst maintaining its precision in a specific range
of battery operation, would therefore be ideal in order to facilitate the accurate, real-
time resolution. Indeed, there are several reduced-order electrochemical models
reported in the literature [4, 29].
2.3.1 Single particle models
A common approach, which results in the development of single particle model
(SPM), is one of the practical solutions used to reduce the complexity of the pseudo-
2D electrochemical model from a system of coupled five PDEs and one algebraic
equation into a single PDE and an algebraic equation [33]. It is based on the
assumptions that the applied current to the battery is small and the conductivity of the
electrolyte is large. Each electrode compartment is represented by a single particle, as
depicted in Figure 2.5. The electrolyte concentration is assumed to be constant and that
the current in the electrolyte does not vary spatially [4]. By applying these
assumptions, the lithium ion cell model is simplified from one consisting of the PDEs
in Eqs. (2.1), (2.3), (2.4) and the algebraic equation (2.5) to the following two PDEs
(one for each electrode in Eq. (2.9)) and the algebraic equation (2.13) [4].
Chapter 2: Literature Review 19
Figure 2.5 Single Particle Model of a Li-ion battery [4]
The governing equation of lithium concentration in the solid particle of each
electrode is (for i neg, pos ):
2
( , ) ( , )10
i i
s ss
c r t c r tD
t r r r
(2.8)
Initial condition and the boundary on the particles is given by:
0,( ,0) ( )i
s sc r c r (2.9)
0
0i
s
r
c
r
(2.10)
( )
( )
p
iis
i i
r R
c I tj t
r Fa L
(2.11)
Output voltage of the lithium ion cell is:
0 0
( ) θ ( ) θ ( ) ( )pos neg
s s collectorV t t t I t R (2.12)
where:
20 Chapter 2: Literature Review
0 0
,max
( )2 ( )θ ( ) asinh ( ( ))
2 ( )( ( ) ( ))
i
fi i i
s ss i ii i i i i
eff e ss s ss
R I tRT I tt U c t
F a La L r c c t c t c t
(2.13)
and the Open Circuit Potential, ( ( ))i i
ssU c t , is a function of surface concentration in
the solid phase ( )i
ssc t .
The advantage of SPM is that it can be evaluated more rapidly, which is essential in
real-time applications. However, this model has inherent limitation due to its low
current rate validity range [34].
2.3.2 Discrete-time realization method (DRA) and reduce-order state space
model
An alternative to the SPM is reported by Lee et al. [7] wherein the authors applied
truncated Taylor series expansions about a set point to each of the variables of the P2D
model to obtain a linear form of the model, around the desired point
*
s,e OCP s,0 s,e s,0 e e e ,0{ ( ), , ( , ) (0, ), }ep U c c c x t t c c [7]. Transcendental transfer
functions of each variable in each domain are then formed by Laplace transform. Figure
2.6 depicts a set of internal electrochemical variables of a lithium ion cell in each
domain. The electric potential s ( , )z t in the solid electrode, the lithium concentration
on the surface of the active material s,e ( , )c z t , the molar ion fluxes n ( , )j z t between the
active materials in the electrodes and the electric potential s ( , )z t in the solid electrode
are functions of the dimensionless spatial variable z [7]. The variable z is defined only
in the electrodes and equal to / negx L in the negative electrode and equal to
( ) /tot posL x L in the positive electrode. The electric potential e ( , )x t in the electrolyte,
the lithium concentration e ( , )c x t are not scaled with z as is the case with the other
parameters. e ( , )x t and
e ( , )c x t are the function of the spatial location x across the cell.
Chapter 2: Literature Review 21
Figure 2.6 The 1D cell diagram with transfer functions in each domain [7]
Lee et al. [35] then introduced a discrete-time realization algorithm (DRA) to reduce
those transcendental transfer functions of the cell electrochemical variables into an
optimal reduced order state space form. The DRA contains the approximation of the
discrete-time impulse response for each transfer function H(s) of the electrochemical
variables [35]. Ho-Kalman algorithm is used to produce a state space form of the model
to yield an optimized, reduced-order, discrete-time, state-space model. It reduces the
computation time requirements by a factor of more than 5000, compared to the P2D
model. However, the parameters of such state space models have no direct physical
meaning and the practical use of the DRA algorithm in real-time applications, where
rapid updating of the parametric values of the parameters is required, is still
questionable.
2.3.3 Padé approximations for SPMs
Padé approximations provide a way of greatly simplifying complicated
transcendental transfer functions in order to produce simpler rational polynomial
transfer functions that can be used in embedded applications [36]. Previously, Padé
approximations have been applied to the transcendental transfer functions that model
22 Chapter 2: Literature Review
lithium ion diffusion in a single, solid, spherical particle in the SPM [37, 38] for lithium
ion batteries. Marcicki et al [39], Zhang et al [40] and Yuan et al [41, 42] used Padé
approximations for an extended version of the SPM, which incorporates concentration
change within the electrolyte solution while maintaining the uniform current density
assumption of the SPM. These authors report that their Padé approximate model is
computationally more than 50 times faster computationally than the P2D model and
attains reasonable accuracy in comparison to the SPM PDE models [42]. However, the
applications of Padé approximations reported in the existing literature are limited to
the SPMs only. Therefore, related research gaps in reduced order electrochemical
models will be further discussed and dealt with in the following chapters.
2.4 PHYSICS-BASED EQUIVALENT CIRCUIT MODELS FOR LITHIUM
ION BATTERIES
A number of reported works proposed in the literature to transform the P2D
model into equivalent circuit models while retaining the physical meaning of the
electrical elements associated with the models. Newman [43] reported a physics-based
equivalent circuit model constructed using a network of resistors and voltage sources.
In this network, high resistance values represent the low conductivity of the electrolyte
and separator and low resistance values represent the high conductivity in the solid
phase. Farrell et al. [44] proposed a physics-based equivalent circuit model that
comprised of a number of finite circuit loops as shown in Figure 2.7. Each circuit loop
includes resistances which represent the conductivities of the solid and solution phase
of the electrodes. Kirchhoff’s current and voltage laws are applying to the circuit
yielding a set of linear algebraic equations for the currents in the circuit. The members
of the coefficient matrix in the linear algebraic equations are determined from the
ohmic resistances corresponding to the conductivity of the solid and solution phase of
the electrodes and electrolyte, respectively [44]. This model provides a useful physical
interpretation of the differential equations associated with the electrochemical model
to investigate the current and reaction distributions within the electrodes in the first
moment after applying the current to the battery.
Chapter 2: Literature Review 23
Figure 2.7 Equivalent circuit for a porous electrode comprised of three finite regions,
adapted from [44]
S. Raël et al. [45] used equivalent circuit model to represent mass and charge
transport in lithium ion batteries. M.-T. von Srbik et al. [46] developed a detailed
physics-based equivalent circuit model which translate all variables of the P2D model
into the electrical circuit components shown in Figure 2.8. In addition, M.-T. von Srbik
et al. [46] considered the double layer capacitor, which represents the electrochemical
charge transfer process takes place at the interface between solid particles and the
solution. The degradation and thermal models are also coupled in this equivalent
circuit model. This detailed model provides a full physical interpreting of the P2D
model into electrical circuit analogy which can be easily implemented and used in
optimal control algorithms. Yu Merla et al. [47] developed a simpler version of the
detailed model proposed by M.-T. von Srbik et al. [46]. This simplified equivalent
circuit keeps most of the features of the aforementioned detailed model such as the
transport in the solid phase, double layer capacitance, degradation as shown in Figure
2.8. However, the lithium ion transport in solution phase is modelled by only resistors
and the diffusion phenomenon occurred in this phase and thermal model are omitted.
This simplified model is also easily implemented and modified due to its modular
structure as shown in Figure 2.9(a) and (b).
24 Chapter 2: Literature Review
Figure 2.8 P2D equivalent circuit network implementation [46]. Reprinted from " A
physically meaningful equivalent circuit network model of a lithium-ion battery accounting
for local electrochemical and thermal behaviour, variable double layer capacitance and
degradation," by M.-T. von Srbik, M. Marinescu, R. F. Martinez-Botas, and G. J. Offer,
Journal of Power Sources, vol. 325, pp. 171-184, Copyright (2016), with permission from
Elsevier
Chapter 2: Literature Review 25
Figure 2.9 Cell level scheme of the simplified physics-based equivalent circuit model
[47]. Reprinted from " An easy-to-parameterise physics-informed battery model and its
application towards lithium-ion battery cell design, diagnosis, and degradation," by Y.
Merla, B. Wu, V. Yufit, R. F. Martinez-Botas, and G. J. Offer, Journal of Power Sources,
vol. 384, pp. 66-79, Copyright (2018), with permission from Elsevier
Overall, these models have the advantages of easier implementation in well-
accepted software package such as Simulink® which provides set of solvers to deal
with the ordinary differential equations in the model. However, these physics-based
equivalent circuit models are developed based on finite volume or finite element
methods which may require expensive computation load when a high number of
discretised node points is chosen.
26 Chapter 2: Literature Review
2.5 THERMAL MODEL
Battery thermal management is a critical function of a BMS in order to ensure
safe operation of the battery safety and to enhance battery life time [48]. Coupled
thermal and electrochemical models have been presented previously in the literature
[49, 50]. Pals and Newman [51] presented a one-dimensional, coupled thermal-
electrochemical model that calculates the heat generation rate and temperature of a
single cell and then in a second paper [52] they used their model to determine the
temperature distribution within a cell stack. Wang et al [53] modelled the thermal
performance of a battery pack with various cell configurations in the pack and with
different cooling strategies. However, due to their complexity and computational cost
these aforementioned models are implemented using powerful commercial software,
embodying integrated finite element and computational fluid dynamical packages [54-
56] such as COMSOL Multiphysics® [57] or ANSYS® Fluent [58]. Guo and White
[12] reported that a 2-dimensional, coupled, thermal-electrochemical model of a
cylindrical cell required approximately 20 hours to simulate a 30-minnute constant
discharge profile. In the same work, a proposed 1-dimensional, radial model of a
spirally wound cell required 20 minutes to simulate the same discharge profile. It is
noted that those reported simulation times are in the case of a constant discharge
current. For dynamic current profiles such as UDDS cycles or grid-connected battery
currents, much higher computational time per sample are required [59]. Thus, these
methods are not suitable for embedded BMSs, which require much lower
computational workload. An approximate thermal-electrochemical model is therefore
required for real-time applications.
A number of approximate thermal-electrochemical models have been reported
in literature [50, 60]. For example, the lumped thermal model, which is based on the
assumption that the whole interior temperature of the cell is uniform [60], has been
used in battery thermal management [8]. This model is simple and requires low
computational overhead and is useful in the thermal management of battery packs
where individual cells within the pack are considered as simple heat sources. However,
this approach is not able to compute non-uniform temperature distributions such as
those that occur in large format cylindrical cells, especially in the radial direction [61].
Chapter 2: Literature Review 27
Kim et al. [62] proposed a second order polynomial approximation to the temperature
of the cell in the radial direction and in which the temperature at the center of the cell
is at the maximum. However, due to the hollow core in spirally wound cells the
maximum will not occur at the center of the cell and in large format cylindrical cells
this approximation is likely to be unreasonable. In addition, previous works use only
a single value of material density [62], specific heat and thermal conductivity for the
whole cell and incorporate a fitting approach to estimate these parameters and
incorporate a Kalman filter to estimate the cell’s temperature. The computational
requirements of such an approach are likely to be relatively high. High efficient
thermal modelling for cylindrical cells is examined further in Chapter 5 and 6 where
the inclusion of thermal model on electrochemical models are discussed.
2.6 DEGRADATION MODEL
Degradation in the lithium ion battery is affected by a variety of factors such as
cycling numbers, storage time, current rates, SOC and mechanical stress. A list of
commonly reported degradation mechanisms, mode, cause and effect is as shown in
Figure 2.10 [63]. Three main degradation modes are classified in [63]. Degradation
mechanisms that link to the loss of lithium inventory include Solid Electrolyte
Interphase (SEI) growth, SEI decomposition, electrolyte decomposition and lithium
plating [63]. Degradation mechanisms that link to the loss of active material in
negative and positive electrode include binder decomposition, structural disordering,
loss of electrode contact, electrode particle cracking and corrosion of current collectors
[7]. The end result is battery capacity fade and power fade. In order to prevent
degradation and optimally operate the battery using health-conscious optimal control
schemes, physics-based degradation models are required.
28 Chapter 2: Literature Review
Figure 2.10 The list of commonly reported degradation mechanisms, mode, cause and
effect, adapted from [63]
Various works have been reported on the modelling of battery degradation. These
works mainly focus on the most dominant degradation mechanisms such as the growth
of SEI layer [63, 64]. SEI growth is attributed to 50% of the battery degradation [63].
Up to this time, SEI is still considered as “the most important but least understood
solid electrolyte in rechargeable Li batteries” due to the complexity of chemical and
electrochemical reactions involved and its physical properties [14, 65]. Ramadass et
al. [66] proposed a model of SEI layer growth which is represented by a continuous
film formation over the surface of the solid particles. It in turn increases the surface
resistance of the particles and the irreversible capacity loss. Safari et al. [67] developed
an isothermal, multimodal, aging model for SEI growth at the carbonaceous anode
material by considering the solvent decomposition kinetics and solvent diffusion
through the SEI layer. Henrik and G¨oran [68] proposed a degradation model based on
linear combination of continuous SEI layers growth in microscopic scale and SEI
layers cracks in macroscopic scale due to the expansion of the solid particles that are
covered by the SEI layers.
An important aspect should be noted is that the degradation rates are strongly
affected by cell’s temperature as higher temperature increases the SEI film growth rate
Chapter 2: Literature Review 29
leading to higher loss of active material and higher SEI resistance [63]. This material
loss and resistance increment in turn induces higher temperature. Moreover, during
operation, non-uniformities in local temperature and degradation exacerbate high
degradation and temperature gradients within large format cylindrical batteries [16].
Unfortunately, coupled electrochemical-thermal-degradation models considering non-
uniformities have not yet been well-studied in the literature due to the complexity
associated with these sophisticated models. An additional challenge is that under
embedded applications there is the need to solve the coupled problem in a fast
computational manner whilst still being able to produce an accurate outcome. These
challenges are examined further in Chapter 6 where the thermal and degradation
models are efficiently included on electrochemical models for large format cylindrical
batteries.
2.7 BATTERY STATE ESTIMATION TECHNIQUES
Precise and reliable battery SOC estimation is important in order to optimize and
ensure safe battery operation and extended battery lifetime. Battery SOC can be also
determined using Coulomb counting method which is based on integration of
measured battery current [69]. In real applications, the SOC initial value is unknown
and current measurement errors exist. This method is also not suitable and therefore
an adaptive observer is required to predict the battery SOC.
A number of model-based approaches have been used to adaptively estimate
battery SOC and generally divided into two categories: equivalent circuit-based and
physics-based estimation algorithms. Equivalent circuit based estimation algorithms
rely exclusively on equivalent circuit models of the battery due to the simplicity of
these models implementation [70]. However, as mentioned above, equivalent circuit
models do not provide insights into the electrochemical characteristics of the battery
during its operation [28]. A limited number of reported works use a physics-based
battery model, like P2D or SPM, to determine SOC.
Among these works, Kalman filters (KF) are attractive approaches and
favourably used in combination with the electrochemical model(s) for state estimation
30 Chapter 2: Literature Review
[70]. Generally, KF algorithms include two main steps in each measurement interval:
state update and measurement update. In the state update step, the present state, the
output of the model are predicted. Then, in the measurement step, the model state is
corrected using a weighted correction factor. This factor is the product of an adaptive
Kalman gain Lk and the error between the measured output of the real system and
estimated one of the model as shown in Figure 2.11. A standard KF with a reduced
form of the P2D model is used to estimate SOC in [71]. In comparison with the
standard KF, extended Kalman filter (EKF) is a highly efficient adaptive filter and is
preferable for estimating the state of the battery [72]. In the EKF, the nonlinear
equations of the system need to be linearized at each time step by using the first order
Taylor series expansion. The linear approximation can reduce the accuracy and result
in an unstable filter [72]. Sigma-point Kalman filter (SPKF) is an alternative algorithm
for state estimation and has some advantages such as its ability to operate without
linearizing the original system functions and the need to calculate their derivatives [72,
73]. SPKF algorithm is discussed in greater detail in Chapter 3 where this algorithm is
adapted to the SOC estimation study using simplified electrochemical models.
Figure 2.11 General principle of Kalman filter algorithm. This figure is adapted from
[74]
2.8 CONCLUSION AND RESEARCH GAPS
Based on this literature review, an essential requirement of advanced BMS is
that it should be able to precisely provide insight into battery electrochemical
dynamics. This function is necessary in order to control the battery operation to
Chapter 2: Literature Review 31
achieve overall technical and economic benefits. In large scale battery storage systems,
another requirement is that the battery model needs to be computationally highly
efficient. This allows the battery model to operate effectively in embedded real time
application. Battery equivalent circuit models have been widely used to represent the
battery dynamic in real time application due to its simplicity and robustness. However,
this model is not able to give insight into battery electrochemical characteristic due to
its limitation on representing physical phenomena occurring inside the battery [7]. The
pseudo-2D battery model can overcome this problem by providing full information of
battery electrochemical characteristic accurately [4]. However, high complexity of the
model prevents its implementation in embedded real-time application. Since high
computational workload is required to solve coupled PDEs, pseudo-2D battery model
is often solved with the aid of a desktop computer and powerful commercial finite
element software. Many works have been reported to reduce the order of the high
fidelity P2D model, however, they still have some limitations discussed above and
need to be improved. High performance coupled electrochemical-thermal-degradation
models considering battery geometry are not yet well developed. Furthermore,
estimating battery SOC in real time is mandatory in BMS. In real applications, the
SOC initial value is unknown and current measurement errors exist. Consequently, the
research problem at hand is to develop a computationally efficient electrochemical
battery model and an adaptive electrochemical model-based estimation algorithm
suitable for a modern BMS of a large battery storage system. The developed scheme
shall precisely and efficiently evaluate battery electrochemical variables such as
lithium ion concentrations in solid phase and solution phase, cell state of degradation,
temperature and be able to predict the battery state of charge. In order to address this
research challenge, there are several gaps from this literature review that are
considered for this thesis as follows.
The combination of the ESPM model and the SPKF technique for battery SOC
estimation has not yet been reported. The ESPM accounts for the electrolyte
concentration and electrolyte potential which are not considered in SPM models
leading to a more accurate electrochemical model compared with SPM. Concurrently,
the SPKF theoretically outperforms EKF and have been demonstrated in previous
32 Chapter 2: Literature Review
works related to battery SOC estimation. This research challenge will be addressed in
Chapter 3 to achieve the Objective 1 of the thesis.
The Padé approximation method has not yet been applied to approximate the
pseudo-2D electrochemical model. This method can reduce the complicated
transcendental transfer functions corresponding to lithium ion electrochemical
parameters to rational transfer functions. These transfer functions will be simpler and
applicable for the embedded BMS application. This research gap will be fulfilled in
Chapter 4 to achieve the Objective 2 of the thesis.
Coupled electrochemical-thermal models of lithium ion batteries are often
associated with high complexity and computational cost. They are implemented using
powerful commercial software, embodying integrated finite element and
computational fluid dynamical packages. A highly computationally efficient,
simplified electrochemical-thermal model is essential in real-time applications of
small-format cylindrical lithium ion cells where temperature variation along cell’s
radius can be negligible. This research challenge will be addressed in Chapter 5 to
achieve the Objective 3 of the thesis.
For large format lithium ion cells, a more sophisticated electrochemical-thermal-
degradation model would be more suitable as it accounts for non-uniform temperature
and degradation distributions such as those that occur in these cells, especially in the
radial direction. Such type of model has not been available in the literature. This
research gap will be closed in Chapter 6 to achieve the Objective 4 of the thesis.
Advanced control of the battery storage systems requires the insightful
information of battery electrochemical variables, degradation and thermodynamics. In
this thesis, a systematic approach to derive high-performance reduced-order
electrochemical models will be presented in Chapters 3-6. These models are capable
of efficiently computing the variations of cell electrochemical variables, temperature
and the degradation of grid-connected lithium-ion battery banks.
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 33
Chapter 3: Single Particle Models and State
Estimation by Sigma-point
Kalman Filter
3.1 INTRODUCTION
From Chapter 2, it has been explained that large-scale lithium ion battery energy
storage systems are widely used in conjunction with renewable resources such as
electrical power generated from wind and solar. The batteries are to act as energy
buffers so as to alleviate the inherent intermittency of the renewables. The advantages
of batteries are their low self-discharge rates, long life and high power and energy
density [3]. In conjunction with these applications, it is essential to employ a suitable
BMS to ensure the batteries operate optimally with respect to safety and service life. A
battery model that yields accurate SOC estimation is an important requirement in the
design of advanced control strategies in the BMSs [4, 28]. Most estimation algorithms
require an accurate battery model that can precisely describe the voltage response of
the battery during the battery charge/discharge process. Therefore, battery modelling
has become an integral part in the development of an effective BMS.
Simple equivalent circuit models are normally applied in model-based methods.
In such connection and as described in Section 2.1, the battery is represented by an
Open Circuit Voltage (OCV) source connected in series with an internal resistance and
a parallel circuit that includes a charge transfer resistance and a double layer
capacitance [17]. This circuit model is very simple and it requires minimal
computation cost to solve. However, batteries have highly nonlinear electrochemical
characteristics that depend significantly on external and internal conditions. Simple
circuit models have only limited capability in accurately accounting for the dynamic
electrochemical behaviour of lithium ion batteries.
34 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
The electrochemical model, which accounts for phenomena such as the transport
of charge within the cell, the intercalation processes that occur within the solid phase
electrode materials and the electrochemical kinetics occurring at the charge transfer
interfaces inside the battery during its operation, is an alternative to battery modelling
[9, 10, 32]. With this model the battery voltage can be accurately predicted. However,
as explained in Section 2.2, the computational cost associated with solving the
coupled, non-linear partial differential equations that define the model is very high in
comparison with equivalent circuit models.
A simplified electrochemical model that has low computational overheads,
whilst maintaining its precision in a specific range of battery operation, would
therefore be ideal in order to facilitate the accurate, real-time resolution and control of
individual cells within a battery energy storage system. There are several reduced
electrochemical models reported in the literature [4, 29]. The SPM that is reviewed in
Section 2.3.1 is one of the practical solutions to reduce the complexity of the P2D
model. An improvement of SPM is the extended SPM (ESPM) proposed in [75, 76],
which takes into account the variations in electrolyte concentration and potential. The
ESPM in [75] approximates the distribution of the electrolyte concentration and
potential within the electrodes with cubic functions of time and space whereas a
quadratic function is used to describe the electrolyte concentration and electrolyte
potential in the separator. The ESPM in [76] approximates the lithium concentration
in the solid and electrolyte phases by quadratic functions of time and space while the
electrolyte potential is solved analytically based on the assumption that the applied
current passed through the electrode is distributed uniformly over the electrode.
Regarding SOC, in order to ensure safe battery operation and to extend the battery
lifetime, precise and stable battery SOC estimation is essential [22, 28]. Examples
include different optimal control strategies suggested in [8] [77] to improve the battery
lifetime by charging/discharging the battery optimally based on the SOC information.
A number of model-based SOC estimation techniques has been reviewed in the
Chapter 2. Among these techniques, Sigma-point Kalman filter (SPKF) outperforms
standard Kalman filter and extended Kalman filter (EKF) algorithm. Therefore, SPKF
algorithm is chosen in this study for SOC estimation algorithm.
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 35
In the first part of this chapter, a simple electrochemical model based on the SPM
[33] is developed and in which the governing PDE is approximated by two differential
algebraic equations and one algebraic equation. This is done in a manner similar to that
given in [29] where the lithium concentration in the single particle that comprises each
electrode can be expressed by a polynomial function of the particle radius. The model
is implemented in Simulink in such a way that it can be used in all types of system
simulations. The feasibility of the proposed model is validated by comparing the
predicted cell voltage obtained from this model with that of the SPM. It is found that
the proposed model significantly reduces the computation time for simulation whilst
producing a reasonable approximation to the SPM output. Hence the first part of this
chapter is mainly to introduce a practical scheme that is used to convert an
electrochemical model into a Simulink cell block.
In the second part of this chapter, the accuracy of the simplified SPM model is
improved by taking into account the evaluation of the electrolyte concentration
distribution as did Han et al. [76]. This improvement yields an extended SPM, so called
ESPM. A new SPKF-ESPM approach is then proposed, which is a combination of the
ESPM and the use of the SPKF technique to estimate the battery SOC. The feasibility
of the proposed model is validated by comparing the predicted cell voltage and SOC
from this algorithm with data obtained from laboratory experiments. The proposed
method is shown to be capable of estimating the SOC of the battery with high accuracy
and computational efficiency.
3.2 EXTENDED SPM AND A PRACTICAL IMPLEMENTATION SCHEME
FOR ELECTROCHEMICAL BATTERY MODELS
3.2.1 Three parameter approximation model for the lithium solid concentration
A highly efficient approximation for the SPM can be achieved by reducing the
PDE in Eq. (2.8) of the lithium diffusion in particles (see Section 2.3.1) to differential
algebraic equations. Two- and three-parameter models that do this are presented in
[29], with the three-parameter model found to be more accurate at representing the
solutions of the PDE. Here the same three- parameter model reduction approach as
36 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
given in [29] is applied to Eq. (2.8) of the SPM presented in Section 2.3.1. A summary
of the model reduction approach is as follows [29].
The lithium concentration is assumed to be a polynomial function of the single
particle radius and given as,
2 4
s
p p
( , ) ( ) ( ) ( )r r
c r t a t b t d tR R
. (3.1)
By using this assumption, the first boundary condition ( 0r ) in Eq. (2.10) is
satisfied and the second boundary condition ( pr R ) in Eq. (2.11) is changed into Eq.
(3.2):
s sn
p p
2 4( ) ( ) ( )
D Db t d t j t
R R . (3.2)
The volume-averaged concentration and volume-averaged concentration flux
are expressed in Eq. (3.3) and Eq. (3.4), respectively:
p 2
s s2
p p0
( ) 3 ( , )
R
r
r rc t c r t d
R R
, (3.3)
p 2
s s2
p p0
( ) 3 ( , )
R
r
r d rq t c r t d
R dr R
. (3.4)
Substituting Eq. (3.1) into Eq. (3.3) and Eq. (3.4) and evaluating, one obtains
3 3
( ) ( ) ( ) ( )5 7
sc t a t b t d t , (3.5)
3 2
( ) ( ) ( ) ( )2
s
p p
q t a t b t d tR R
. (3.6)
Evaluating Eq. (3.1) at the surface condition ( pr R ) yields
( ) ( ) ( ) ( )sc t a t b t d t . (3.7)
The three coefficients ( )a t , ( )b t and ( )d t can be obtained by manipulating
equations Eq. (3.8), Eq. (3.9) and Eq. (3.10) to obtain:
39 35
( ) ( ) ( ) 3 ( )4 4
ss s s pa t c t c t q t R , (3.8)
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 37
( ) 35 ( ) 35 ( ) 10 ( )ss s s pb t c t c t q t R , (3.9)
105 105
( ) ( ) ( ) 7 ( )4 4
ss s s pd t c t c t q t R . (3.10)
Substituting Eq. (3.8), Eq. (3.9) and Eq. (3.10) into Eq. (3.1),
2
4
39 35( , ) ( ) ( ) 3 ( ) ( 35 ( ) 35 ( ) 10 ( ) )
4 4
105 105( ) ( ) 7 ( )
4 4
s ss s s p ss s s p
p
ss s s p
p
rc r t c t c t q t R c t c t q t R
R
rc t c t q t R
R
. (3.11)
Volume averaging the governing equation (2.8) produces
2
2 2
0
( , ) ( , )13 0
pR
s ss
p pr
c r t c r tr rD d
R t r r r R
. (3.12)
Substituting Eq. (3.11) into Eq. (3.12), the volume-averaged concentration then
becomes:
( )
( ) 3 ns
p
j tdc t
dt R . (3.13)
Differentiating Eq. (2.8) and volume averaging yields
22
2
0
( , ) ( , )1
3 0
p
s sR s
p pr
c r t c r tD
t r r rr rd
R r R
. (3.14)
Substituting Eq. (3.11) into Eq. (3.14), the volume-averaged concentration flux
is then governed by:
2 2
( )45( ) 30 ( ) 0
2
s ns s
p p
D j tdq t q t
dt R R . (3.15)
The surface concentration of lithium in the particles ( )ssc t is then obtained by
substituting Eq. (3.9) and Eq. (3.10) into Eq. (3.2) to give:
8
( ) ( ) ( ) ( )35
p p
ss s s n
s
R Rc t c t q t j t
D . (3.16)
Equations (3.13), (3.15) and (3.16) represent the SPM-Three parameter model
for the lithium concentration at the surface of the single particle that constitutes each
38 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
electrode in the SPM presented above. As stated above, one has reduced a single PDE
in Eq. (2.8) into an algebraic equation (3.16) and two ordinary differential equations
(Eq. (3.13) and Eq. (3.15)). The initial conditions for Eq. (3.13) and Eq. (3.15) are
given by the initial state of the cell. For a single discharge of a new cell these values
will be 0,( ,0) ( )s sc r c r and (0) 0sq . To obtain the cell voltage using this model, the
lithium surface concentration ( )ssc t from Eq. (3.16) is substituted into Eq. (2.13), which
are then substituted into Eq. (2.12).
3.2.2 Implementation of SPM-Three parameter model for lithium ion cell in
Simulink
In this investigation, the SPM-Three parameter model (henceforth, labelled as
SPM-3P model) that is presented in the previous section (Eq.s (2.12), (3.13), (3.15)
and (3.16)) has been implemented in the Matlab/Simulink environment. In Simulink,
it represents a “lithium ion cell block” which can be applied to various applications in
power electronics including an electrochemical cell within a large battery pack. Since
the SPM-3P model consists of three differential algebraic equations, the computation
requirement is dramatically reduced from that of the SPM and full electrochemical
models. Consequently, the cell block can be connected in series and parallel to achieve
the required battery pack voltage and capacity of large battery energy storage system.
Figure 3.1 shows an example of the cell model in a simple application in
Simulink. The cell is connected to a Controlled Current Source that represents a
constant current load. The lithium ion cell is continuously discharged by different
constant current rates ranging from 1C to 10C. The parameters used in the model are
those from the FreedomCar cell and are given in [27] and listed in Table 3.1.
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 39
Figure 3.1 Simulink model of a lithium ion cell connected to a load
Table 3.1 FreedomCAR Cell Model parameters [27]
Symbol Units Negative
Electrode Separator
Positive
electrode
410 cm 50 25.4 36.4
pR 410 cm 1 - 1
,maxsc
3 -310 molcm
16.1 - 23.9
,0ec
3 -310 molcm
1.2 1.2 1.2
0i
3 -210 Acm
3.6 - 2.6
collectorR mΩ 1.9 - 1.9
sD
2 -112 cm1 s0
2.0 - 3.7
fR mΩ 0 - 0
L 410 cm 50 25 36.4
eD
2 -16 cm1 s0
2.6 2.6 2.6
s - 0.58 - 0.50
0% , 0%
- 0.126 - 0.936
100% ,
100%
- 0.676 - 0.442
40 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
Symbol Units Negative
Electrode Separator
Positive
electrode
s - 0.58 - 0.50
The Open Circuit Potential of the negative and positive electrodes are functions
of stoichiometry and are adapted from [27] as follows:
neg neg neg neg 1/2 4 neg 1
5 neg 3/2 neg
neg
( ) 8.00229 5.0647 12.578( ) 8.6322 10 ( )
2.1765 10 ( ) 0.46016exp[15.0(0.06 )]
0.55364exp[( 0.92)]
U x
x
(3.17)
pos pos pos 6 pos 5 pos 4 pos 3
pos 2 pos pos 115
( ) 85.681( ) 357.7( ) 613.89( ) 555.65( )
281.06( ) 76.648 0.30987exp(5.657( ) )
13.1983
U
(3.18)
where, stoichiometry ss s,max( ) / ( )c t c t is the ratio between the lithium ion
concentration at the surface of the solid particle and the maximum lithium ion
concentration.
Figure 3.2 shows the detailed equivalent circuit of a designed electrochemical
cell named Electro_cell_SPM3. The cell model consists a SPM-3P model block, a
Controlled Voltage Source block, a resistance and a current sensor. The SPM-3P
model block contains Matlab Functions blocks and Integrator blocks used to calculate
the cell voltage based on Eq.s (2.11), (2.12), (3.13), (3.15), and (3.16). The input of
the SPM-3P model block is the charge/discharge current from the current measurement
block. In order to avoid the algebraic loop error, an initial condition block (IC block)
is added into the SPM-3P model block to set the initial condition of the current input.
The output of the SPM-3P model block then forms an input of the Controlled Voltage
Source block which converts the input signal into an equivalent voltage source. The
resistor collectorR represents the cell current collector contact resistance.
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 41
Figure 3.2 Equivalent circuit of the SPM-3P for lithium ion cell
Figure 3.3, Figure 3.4, Figure 3.5, and Figure 3.6 depict the comparison between
the voltage response of the SPM-3P model, a Two-parameter version of the SPM
(labelled as SPM-2P, details not given here) and the SPM (solved by using the “pdepe”
solver in Matlab) for discharge current rates of 1C, 5C, 7C and 10C. SPM-2P is built
using similar approach used for SPM-3P. The difference is that the lithium ion
concentration in the solid phase is approximated using a quadratic function of the
single particle radius. It is observed that the difference between the SPM-2P model
and SPM-3P model increases as the current rate increases. Furthermore, the SPM-3P
model and the SPM give very similar results.
42 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
Figure 3.3 Cell voltage response during 1C discharge rate
Figure 3.4 Cell voltage response during 5C discharge rate
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 43
Figure 3.5 Cell voltage response during 7C discharge rate
Figure 3.6 Cell voltage response during 10C discharge rate
The SPM-3P model requires 0.011 second to simulate the 3600 second discharge
profile on a desktop computer, which is about 3.06 µsec/sample. On the other hand,
44 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
the “pdepe’’ solver in Matlab requires 0.478 second to simulate the same profile on
the same desktop computer, which is about 0.133 msec/sample. Thus, the SPM-3P
model is approximately 40 times faster than using the SPM with one governing PDE.
Consequently, the SPM-3P model is a possible candidate for real-time control
applications since it requires low computation while maintaining high accuracy
compared to the SPM.
3.3 EXTENDED SINGLE PARTICLE MODEL AND SIGMA-POINT
KALMAN FILTER
3.3.1 Extended single particle model ESPM
The ESPM is an approach to improve the accuracy of the SPM [75]. It is based on
the assumptions that each electrode compartment is represented by a single particle
and the electrolyte concentration in it varies with time and space. The concentration is
computed as shown below.
The output voltage of the lithium ion cell is expressed in Eq. 3.19 using ESPM
approach. The detailed derivation can be found in [78, 79].
pos
f1
pos pospos pos pos 0 pos pos pos
eff e ss s,max ss
neg
f1
neg negneg neg neg 0 neg neg neg
eff e ss s,max ss
pos p
e tot e col ss
2sinh
2
2sinh
2
, 0,
I t R I tRTV t
F a La L r c c c t c t
I t R I tRT
F a La L r c c c t c t
L t t I t R OCP c
os neg neg
sst OCP c t
. (3.19)
In Eq. (3.19), the electrolyte potential difference ttL ,0, etote is approximated as in
Eq. (3.20), based on the assumption that the molar flux is the same in each electrode
[78]. The difference between the electrolyte potential at the positive electrode-current
collector boundary and negative electrode-current collector boundary is
0
neg sep pos e tot
e tot e neg sep pos
eff eff eff e
2 12 ,, 0, ln
2 0,
RT tL L LI t c L tL t t
A F c t
. (3.20)
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 45
The governing equations of electrolyte concentration txc ,e in the negative electrode,
separator and the positive electrode are expressed as in Eq. 3.21 [75].
neg
e e eeff 0
e,neg s neg
sep
e e eeff
e,sep neg nm
pos
e e eeff 0
e,pos s nm tot
, ,1 , ,0
, ,,
, ,1 , ,
c x t c x tD t a j x t x L
x x x
c x t c x tD L x L
x x x
c x t c x tD t a j x t L x L
x x x
. (3.21)
In the ESPM, the electrolyte concentration is then approximated as quadratic
functions of position x along the electrodes and separator [76]:
2
1 2 3 neg
2
4 5 6 neg nm
2
7 8 9 nm tot
, 0
,
,
e
a t x a t x a t x L
c a t x a t x a t L x L
a t x a t x a t L x L
. (3.22)
Using six boundary conditions and three equations of total amount of lithium ion as
derived in [76], the nine coefficients 91 aa governing the electrolyte concentration can
be obtained through solving the simultaneous equation Eq. (3.23).
7 tot 8
2
neg 3 neg
e 1 neg 3 neg e
sep 3 3 2 2 sep
e 4 nm neg 5 nm neg 6 sep e
pos 3 3 2 2 pos
e 7 tot nm 8 tot nm 9 pos e
2 2
1 neg 3 4 neg 5 neg 6
2
4 nm 5 nm
2 0
0
2 6 6
2 3 6 6
2 3 6 6
a t L a t
a t
a t L a t L Q t
a t L L a t L L a t L Q t
a t L L a t L L a t L Q t
a t L a t a t L a t L a t
a t L a t L
2
6 7 nm 8 nm 9
neg eff sep eff sep eff
1 e e,neg neg 4 e e,sep neg 5 e e,sep
sep eff sep eff pos eff pos eff
4 e e,sep nm 5 e e,sep 7 e e,pos nm 8 e e,pos
2 2
2 2
a t a t L a t L a t
a t D L a t D L a t D
a t D L a t D a t D L a t D
. (3.23)
Figure 3.7 depicts the electrolyte concentration distribution in the separator and the
positive electrode during a particular charge-discharge process of a half-cell
constructed in the Queensland University of Technology’s (QUT) battery research
facility. The current profile is shown in Figure 3.10(a). The negative electrode of the
half-cell is a lithium foil and so the electrolyte concentration at the boundary between
the foil and the separator is considered constant throughout the charge-discharge
46 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
process. On the other hand, in the separator and in the positive electrode, the electrolyte
concentration is shown to have non-uniform profiles as the dynamic charge-discharge
proceeds. The electrolyte concentration varies spatially and temporally in these two
domains which clearly invalidates the constant electrolyte concentration assumption
made in developing the SPM. This is one of the reasons why the results obtained from
the SPM deviate from that obtained from the P2D model [75]. Therefore, by
introducing the variation of electrolyte concentration in the ESPM, one can obtain a
more accurate determination of the cell terminal voltage.
Figure 3.7 Electrolyte concentration in the separator and positive electrode during the
dynamic charge/discharge current profile.
3.3.2 Sigma-point Kalman filter algorithm
In order to estimate the SOC of the battery with high precision and stability, the
SPKF algorithm reported in [73] is used in this work. For the problem at hand, the cell
ESPM can be represented in the discrete-time state space form:
, , 1 11 1 1 1
,max ,max ,max
, , , 1 3s k s k k
k k k k k
s s s p
c c j tx f x u w k w
c c c R
. (3.24)
where, , , ,x u w k are state of the system, input current, process noise and time
step, respectively.
The cell SOC can be calculated using the transformation shown in Eq. (3.25). It
is noted that the SOC given in Eq. (3.25) is calculated in terms of the physical
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 47
parameters of the cell such as average lithium concentration ,s kc and the maximum
lithium concentration in the solid particle ,maxsc . SOC is calculated based on the ratio
of the average lithium concentration in the solid phase to the maximum possible
concentration in the electrode.
,
0%
,max
100% 0%
s k
s
k
c
cSOC
. (3.25)
where, 0% 100%, are stoichiometric limit at SOC 0% and SOC 100%,
respectively.
The model output cell voltage is given as Eq. (3.26).
1
pos pos pos 0 pos pos pos
eff e ss s,max ss
pos
f1
pos posneg neg neg 0 neg neg neg
eff e ss s,max ss
neg sep pos
neg sep
eff eff eff
2, , , sinh
2
2sinh
2
2
2
k k k k
I tRTy g x u v k
F a L r c c c c t
I t R I tRT
F a La L r c c c c t
L L LI t
A
0 neg
e tot f
pos neg neg
e
pos pos neg neg
ss ss col
2 1 ,ln
0,
k
RT t c L t R I t
F c t a L
OCP c t OCP c t I t R v
(3.26)
Within the first term on the RHS of Eq. (3.26), the reaction rate effr is fitted as an
empirical function of the input current and the normalized value of surface
concentration in the solid phase. This fitting of the reaction rate contributes to
compensate the weakness of the kinetic expression used in the Butler-Volmer equation
which cannot represent correctly the specific phenomenon related to a phase transition
electrode material [80]. For the particular half-cell used in this study, the reaction rate
pos
effr in Eq. (3.26) is fitted using measured voltage and current data from experiment
and shown in Eq. (3.27). The detail of the experiment is discussed in the next section.
48 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
pos 8 6 6 2 3
eff
2pos pos pos
ss ss ss6 8 8
pos pos pos
s,max s,max s,max
pos pos
ss ss5 2 6
pos
s,max s
2.025 10 1.156 10 7.141 10 0.01484
2.048 10 2.462 10 3.059 10
3.546 10 3.772 10
r I t I t I t
c t c t c tI t
c c c
c t c tI t I t
c c
2
pos
,max
. (3.27)
Regarding the second term on the RHS of Eq. (3.26), the Open Circuit Potential
(OCP) of the negative and positive electrodes are functions of stoichiometry . The
analytical expression to describe the function can be obtained by fitting the curve
relating the measured OCP against the stoichiometry or a lookup table can be used.
This OCP-stoichiometry relationship and the form as well as the coefficients of the
function depend on the material that is used in each electrode. In the case of the
particular half-cell that is used in experiments carried out in this study, the OCP of the
negative electrode which is made by lithium foil is considered to be zero and the
positive electrode (LiFePO4) is a function of stoichiometry as adopted from [81] as
follows:
neg neg 0OCP , (3.28)
pos pos pos 1.3198
6 pos 3.8003
6 pos 3.7995
3.4323 0.8428 exp 80.2493 1
3.247 10 exp 20.2645 1
3.2482 10 exp 20.2646 1
OCP
. (3.29)
Of the remaining terms on the RHS of Eq. (3.26), the values of the various
parameters of the half-cell used in this study are listed in Table 3.2.
Table 3.2 Half-Cell Model Parameters
Symbol Units Negative
Electrode Separator
Positive
electrode
pR 910 m - - 53
s,maxc -3molm - - 21,900
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 49
Symbol Units Negative
Electrode Separator
Positive
electrode
e,0c -3molm 1,000 1,000 1,000
colR 2m 0.001 - 0.001
sD 18 2 110 m s
- - 8
fR 0 - 0
L 610 m 750 25 30
eD 10 2 110 m s
0.22 0.22 0.22
s - - - 0.48
0% , 0%
- - - 0.99
100% , 100% - - - 0.01
The schematic diagram of the ESPM-SPKF algorithm is depicted in Figure 3.8.
The lithium ion cell used in this investigation has its applied current (input) and its
terminal voltage (output) measured at every time step. The unmeasurable battery state
of the cell (either the cell SOC or its average lithium ion concentration in the solid
particle) is unknown and needs to be estimated. Thus, the battery model and the ESPM-
SPKF estimator will have the same battery current as input, and the estimated output
voltage ˆkV as the output of the estimator. The estimated state of the model is processed
by the SPKF within each time interval via two steps: state update and measurement
update. The output of the state update step ˆkx
is calculated based on the prior estimated
state 1ˆ
kx
of the previous time step. ˆkx
is then used as an input to compute the estimated
voltage output of the model. The error between the output of the battery and the model
is multiplied by the SPKF gain and the resulting signal is then used to update the
estimated state again in the measurement update step. This final updated state estimate
50 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
value ˆkx at each time step is exported to the BMS and is used to calculate the state
estimate of the next time interval. In this way, the ESPM-SPKF is used to estimate the
SOC of the half-cell.
State update
Lithium ion cell
Measurement update
Z-1 SPKF gain
+
++ +
k-1 k-1 k-1f(x ,u ,w ,k - 1), , )k k kg(x ,u v k
ˆk-1
x
1ku
ˆk
x
ˆk
x
kvkw
cellVcellI
errorV
ˆkV
Real system
Battery model and estimation by ESPM-SPKF
Figure 3.8 Diagram of the SPKF algorithm
A summary of the relevant equations in the SPKF algorithm adopted from [73] is as
follows:
; ;a T T T
k k k kx x w v ,
; ;T
a x w T v T
k k k k
,
2 dim a
kp x ,
Initialization at k=0:
0 0x̂ x Ε , T
, , ,
,0 0 0 0 0 ,0ˆ ˆ diag , ,a a a a a
x x w vx x x x
Ε ,
T
,0 0 0 0 0ˆ ˆ
x x x x x
Ε , , T
0 0 0ˆ ˆ ; ;a ax x x w v
Ε ,
For k = 1, 2, …
Step 1: State update
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 51
, , , , , ,
1 1 1 , 1 1 , 1ˆ ˆ ˆ, ,a a a a a a
k k k x k k x kx x x
,
, , ,
, 1, 1 1,, , , 1x x w
k i k i k k if u k
,
( ) ,
0 ,ˆ p m x
k i i k ix
,
T
( ) , ,
, 0 , ,ˆ ˆp c x x
x k i i k i k k i kx x
,
Step 2: Output estimate
, ,
, , 1 1,, , ,x v
k i k i k k iY f u k
,
( )
0 ,ˆ p m
k i i k iy Y ,
Step 3: Estimator gain matrix
T( )
, 0 , ,ˆ ˆp c
y k i i k i k k i kY y Y y ,
T( ) ,
, 0 , ,ˆ ˆp c x
xy k i i k i k k i kx Y y
,
, ,k xy k y kL ,
Step 4: Measurement update
ˆ ˆ ˆk k k k kx x L y y ,
T
, , ,x k x k k y k kL L .
where, the detailed definition of the symbols shown in these steps is given in
[73].
3.4 EXPERIMENTAL RESULT AND DISCUSSION
In order to verify the proposed method for estimating the battery SOC using
ESPM-SPKF, the half-cell alluded to earlier is assembled and used for an experimental
study. Since the half-cell is built in the QUT’s battery research facility, it is possible
to obtain cell’s physical parameters such as electrode thickness, electrode plate area,
and active material volume fraction from measurements. The half-cell contains a
LiFePO4 porous electrode, a lithium foil as a reference electrode, a porous separator
52 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
and electrolyte. The nominal capacity of the half-cell is 1.679 mAh. The thickness of
the lithium foil used in this half-cell is 0.75mm which is 25 times thicker than the
thickness of the positive electrode which is 0.03mm. Thus, the lithium foil can present
itself as an “inexhaustible lithium source” to the positive electrode. Therefore, the
lithium concentration at the separator and lithium foil boundary can be maintained
constant during the half-cell charge/discharge operations.
After the assembly process, the half-cell is subjected to a preconditioning
charge-discharge operation at low current rate (C/10). This is to allow the formation
of a stable solid electrolyte interphase (SEI) layer. After this preconditioning step, the
half-cell is ready for the main test which will also be emulated in the simulation study.
The half-cell is subjected to a sequence of dynamic current pulses, with the current
rate varying from 1C to 5C, as is shown in Figure 3.10(a). The duration for each
charge/discharge pulse is two minutes. This current pulse profile used in this
experiment will be helpful for verifying the ESPM model for predicting the cell
voltage. The discharge amount is higher than the charge amount in this profile. It
reflects the operation of the battery where the overall discharging is higher than
charging when the battery is to power electric vehicles on the road.
A VMP-300 battery test station from BioLogic Science Instruments, running
EC-Lab Software was used to cycle (charge and discharge) the half-cell. The Galvano
Profile Importation (GPI) function in the EC-Lab Software enables the user to import
the charge-discharge patterns designed with dynamic current pulses as a text file. The
half-cell is kept in a Coin Cell Holder and connected to one channel of the VMP-300
battery test station via a low current probe. A computer controls the charge/discharge
process based on the imported current profile and automatically records cell current
and voltage every second. The experiment setup is shown in Figure 3.9. The proposed
algorithm is implemented in a Matlab program. The measured data is imported into
this program whereby the current data is used as an input to the model and the voltage
is used as the reference for the state estimator. SOC is also calculated by the Coulomb
counting method which integrates the measured current over the time [69]. Since the
initial SOC is known and the error accumulated over the short operation period is
assumed to be negligible, the outcome of the Coulomb counting method can be used
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 53
as the reference or the so-called true SOC. In the simulation, incorrect initial SOC
value (60%) is intentionally chosen as the initial value for the SOC estimation to check
whether the proposed algorithm is able to converge to the true SOC value.
Figure 3.9 Experiment setup to test the half-cell
Figure 3.10(b) gives the progress of the SOC reference and the estimated SOC
values during the validation experiment, plotted over time. The estimated SOC by the
ESPM-SPKF algorithm starting at SOC 60% can converge to the SOC reference within
180s and thereafter able to track the SOC reference. This can be seen in Figure 3.10(c),
where the estimation error is shown. Similarly, in the case of estimating the battery
voltage, the voltage estimate starting with a wrong initial voltage value can (due to
wrong initial SOC) converge and track the measured voltage, as illustrated in Figure
3.11(a). The voltage estimated error is shown in Figure 3.11(b).
54 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
Figure 3.10 (a) Current rate, (b) comparison between the reference SOC and SOC
reference and (c) SOC estimation error
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 55
Figure 3.11 (a) Comparison of the voltage estimation and the measured voltage and
(b) the voltage estimation error.
The open-circuit potential (OCP) vs SOC characteristic of LiFePO4 based
battery exhibits a flat profile over a wide range of SOC, as shows in Figure 3.12. The
OCP-SOC relationship of LiFePO4-based positive electrode presents nearly constant
OCP over the range 20% to 90% of the SOC. OCP of positive electrode is a dominant
part of the full cell OCP. Theoretically it should be straightforward to infer from a
given SOC the corresponding OCP value. However, in practice, the shallow gradient
in the OCP-SOC characteristic introduces difficulty in estimating the SOC of the
battery from a given the OCP value. Traditional method reliant on OCP-based
estimation is thus unsatisfactory for this type of battery material. For example, when
the controller initializes the system or after the system is reset, the terminal voltage of
the battery which is in equilibrium after a resting period is considered as the OCP of
the battery. This value is used to deduce the SOC value from OCP-SOC relationship.
56 Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter
For LiFePO4-based battery, this method is not feasible. A sustained SOC error can be
introduced, without the help of the SPKF algorithm. The proposed method is therefore
most useful for this type of battery material.
Figure 3.12 OCP-SOC relationship of LiFePO4 positive electrode.
Importantly, the computation time of the ESPM-SPKF algorithm is small. The
algorithm is implemented in Matlab and is computed in a desktop computer with Intel
Core i7-6700 CPU running at 3.4GHz and 16GB RAM. The total computational time
required is 2.977s to complete 8,900s of the study period, which is about 0.33ms per
sample. If the sampling time is 1s in real-time application, the proposed ESPM-EKF
algorithm is sufficiently fast to compute for 3,000 cells within one sampling time. In
practice, the cells can be connected in series or parallel to form a battery module. A
battery pack can be configured by many modules and the ESPM-SPKF algorithm can
be used to estimate the SOC of each module instead of each cell, with the assumption
that every cell in the battery module is identical. By doing so, the computational speed
of the proposed algorithm can be increased sufficiently for use in a large battery bank.
Therefore, there is scope for the ESPM-SPKF algorithm to be adopted in real-time
BMS applications for grid-connected battery energy storage systems.
Chapter 3: Single Particle Models and State Estimation by Sigma-point Kalman Filter 57
3.5 CONCLUSIONS
This chapter presents a simple electrochemical SPM-3P model, its extension
ESPM and the incorporation of it into a state estimation scheme based on the sigma-
point Kalman Filter (SPKF) algorithm for lithium ion cells. Specifically, the
implementation of the SPM-3P model as a Simulink block has been carried out. This
Simulink block could be a useful tool for the simulations of battery systems such as
those used in electric vehicles or renewable energy applications. Through ESPM, the
accuracy of the SPM-3P is improved by taking into account the variation of electrolyte
potential and current. Finally, a state estimation algorithm for estimating battery SOC
has been proposed by combining the ESPM model with the SPKF estimation scheme.
The results shows that the proposed ESPM-SPKF algorithm can effectively track the
reference SOC and voltage whilst enjoying low computational overheads.
This contribution closes the research gap and meets the requirement from
industry on the management of modern electrical distribution networks which have
high penetration of renewables and battery storage systems. Such an application
requires low computational-burden battery models which can at the same time provide
insightful information of battery SOC and electrochemical variables, as was explained
in Chapter 2.
58 Chapter 4: Padé Approximation of Linearized P2D Model
Chapter 4: Padé Approximation of Linearized
P2D Model
4.1 INTRODUCTION
The batteries used in storage systems for grid applications are designed to have
a high energy density; as opposed to cells that are specifically designed to deliver high
power (like those used in electric vehicle applications [4]). Such high energy density
cells are generally made with thicker electrodes [4], which results in non-uniform,
lithium ion concentration and current density profiles across the electrode [4].
However, in both the SPM and extended SPM the current density is assumed to be
constant, spatially as discussed in Chapter 3. Thus, these models can yield inaccurate
predictions when applied to the thick electrodes of high energy density cells [4].
Accordingly, this chapter describes the attempt to improve on the SPM and
ESPM using on Padé approximation technique. To do this, the transcendental transfer
functions of the linearized P2D model and the corresponding corrections for nonlinear
behaviour given by Lee, Chemistruck and Plett [7] are adopted, with the view to
improve on the accuracy of the linearized P2D model. Padé approximations to these
complicated transcendental transfer functions are applied to develop simpler, rational
polynomial transfer functions, which form a simpler model of cell operation that is
amenable to rapid computation and therefore embedded BMS applications. It is
believed that this is the first attempt at using the Padé approximation method to
simplify all transcendental transfer functions of the linearized P2D model. The
proposed reduced model output includes the spatial and temporal variation of all of the
variables of the P2D model. In addition, it can be used to determine any variable at
any specific spatial location without the need to compute variables at all of the
discretized spatial locations; as is the case when using spatial discretization methods
to solve these models. Importantly, the physical meaning of the approximation model
can be preserved since the coefficients of these rational polynomial transfer functions
Chapter 4: Padé Approximation of Linearized P2D Model 59
directly link to the physical parameters of the cell such as, SEI layer resistance, particle
size of the active material, lithium ion diffusivity and solid and solution phase volume
fractions. If the operating conditions of the cell change (for example, due to cell aging),
which in turn changes the values of the physical parameters, these coefficients are
readily updated as they themselves are in computationally simple algebraic forms.
4.2 ANALYTIC LAPLACE DOMAIN TRANSFER FUNCTIONS OF
LINEARIZED P2D MODEL
Truncated Taylor series expansions about a set point can be applied to each of
the variables of the P2D model to obtain a linear form of the model. The details of this
are given by Lee, Chemistruck and Plett [7]. The set-point, p*, used by these authors is
given by the initial conditions of the cell, defined as
*
, ,0 , ,0 ,0p , , , 0s e OCP s s e s e eU c c c c c j , where s,e s e . These authors also
define a set of transfer functions for the cell, shown in their respective domains in
Figure 4.1, where the dimensionless distance z is related to the spatial variable x via the
transformations, neg/z x L ( neg0 x L ) and tot pos/z L x L ( neg sep totL L x L ).
Subsequently, the “~” notation is appended to the variables , , ,0s e s e OCP sU c , and
, , ,0s e s e sc c c . The transcendental transfer functions are formed by first linearizing the
governing equations of the P2D model around the set point p* and then taking Laplace
transforms of these linearized equations. The details of these transfer functions are as
follows.
60 Chapter 4: Padé Approximation of Linearized P2D Model
Figure 4.1 Schematic diagram of a lithium ion cell embodying the P2D model taken
from [4] and transfer function of each variable in each domain based on [7].
For the phase potential difference neg
s,e in the negative electrode:
neg
negneg negeffs,e
neg negnegapp
neg
eff
1cosh
,
1sinhcosh 1
s zz s L
I s A s ss z
, (4.1)
where:
s,0
12
neg neg negneg neg negs s OCP s
totneg neg neg
eff eff s,e s
tanh
tanhc
a a U Rs L R
c FD
, (4.2)
neg
s neg
s
sR
D (4.3)
and s is the frequency parameter in the Laplace domain.
For the molar flux in the negative electrode:
Chapter 4: Padé Approximation of Linearized P2D Model 61
neg neg
neg neg neg negapp s eff eff
neg neg neg neg
eff eff
neg
,
cosh cosh 1
sinh
J z s s
I s a FAL
s z s z
s
. (4.4)
For the over potential in the negative electrode:
neg neg
neg
ct
app app
, ,z s J z sFR
I s I s
. (4.5)
For the solid phase Li+ concentration on the surface of the particles in the
negative electrode:
neg neg neg
s,e s
neg neg neg neg negapp s s eff eff
neg neg neg neg
eff eff
neg
, tanh
tanh
cosh cosh 1
sinh
C z s s R
I s a FAL D
s z s z
s
, (4.6)
For the potential in solid phase of the negative electrode:
neg neg neg negnegeffs
neg neg neg neg negapp eff eff eff
neg neg neg neg neg
eff
neg neg neg neg neg
eff eff eff
cosh cosh 1,
sinh
1 cosh sinh
sinh
L s s zz s
I s A s s
L s z s s
A s s
. (4.7)
The transfer functions for the electrolyte potential in the positive electrode
pos
s,e app, /z s I s , the molar flux in the positive electrode app, /posJ z s I s , the solid
concentration on the surface of the particle in the positive electrode pos
s,e app, /C z s I s
and the potential in the solid phase of the positive electrode pos
s app, /z s I s , are given
by multiplying the corresponding transfer functions in the negative electrode by (-1)
and substituting the corresponding parameters of the positive electrode.
For the Li+ concentration in the electrolyte:
*
,
1app app
,;
Ke ke
k
k
C sC x sx
I s I s
, (4.8)
where,
62 Chapter 4: Padé Approximation of Linearized P2D Model
* ** neg pos
e,k k k
app app app
1
k
C s J s J s
I s s I s I s
, (4.9)
Here k is the kth eigenvalue, and ; kx the corresponding eigenfunction, for
app, /eC x s I s and K is the order of the expansion [7]. The transfer functions,
*neg
k app/J s I s and *pos
k app/J s I s as well as those for the electrolyte potential in
each domain are given in the Appendix A.
The nonlinear corrections for the internal variables are computed as follows [7]:
ct app
final
s
app
11 1eff
e s,max s,e s,e
, ,
2
2 , , ,
R I tx t z t
a LA
I tRT
F c x t c c x t c x t
, (4.10)
, , ,0, ,s e s e sc x t c z t c , (4.11)
,0, ,e s ec x t c z t c , (4.12)
s,e s,e ,, , ,OCP s ex t z t U c x t , (4.13)
e e e1 1, , 0,x t x t t , (4.14)
0
e
e e 1e
2 1 ,, , ln
0,
RT t c x tx t x t
F c t
. (4.15)
Finally, the cell voltage is calculated from:
pos tot neg
cell film film
pos tot neg
OCP , OCP s,e
tot tot
e
, 0,
, 0,
, , 0,
s e
final final
V t F R j L t R j t
U c L t U c t
L t L t t
. (4.16)
The advantage of these transfer functions is that they model the P2D variables at
any spatial location in the cell domain based only on the applied current appI s . The
transfer functions given in Eq.s (4.1) – (4.9) and Eq.s (7.1) – (7.5) are complicated and
computationally expensive to calculate which, like the P2D model from which they
are derived, makes them not amenable to real time application. The Discrete Time
Chapter 4: Padé Approximation of Linearized P2D Model 63
Realization algorithm (DRA) proposed by Lee, Chemistruck and Plett [35] can reduce
the transcendental transfer functions into an optimized, reduced order, discrete-time,
state-space model. In the DRA, Lee et al. first compute the discrete-time pulse response
of the transcendental, continuous-time, transfer functions (Eq.s (4.1) – (4.9) and Eq.s
(34) – (38)). They then apply the Ho-Kalman algorithm to produce the state-space
realization from this discrete-time impulse response. These authors show that the
resulting simple, state-space model produced from the DRA can reduce the
computation time requirements by more than 5000 times, compared to the P2D model.
However, the DRA requires about 11 min on a desktop computer to produce the
reduced state-space model from the transcendental transfer functions for a given set of
parameters (e.g. filmR , effr ,
sD , eD ,
e and s ). There have been two major
improvements to the classical DRA in terms of reducing the computation overhead, the
first by Gopalakrishnan et al. [82] who implemented an improved singular value
decomposition scheme on a virtual Hankel matrix to speed up the DRA. This resulted
in a reduced order model that was approximately 100 times faster than that using the
classical DRA. The second was by Rodríguez et al. [83] who applied variation of
parameters to speed up the computation of the electrolyte concentration transfer
function. In this case the modified transfer function can be computed 3800 times faster
than the previous electrolyte concentration transfer function given by Lee, Chemistruck
and Plett [7]. The speed up of an overall reduced order model using this modified
transfer function is not considered by Rodríguez et al [83], however. Regardless of these
improvements however, when the values of the parameters change, for example due to
cell aging, the DRA needs to be re-run to regenerate another state-space model. This
results in an increased run time for DRA based reduced order models that may prevent
their practical use in real-time applications where rapid parameter updating is required.
4.3 PADÉ APPROXIMATION MODEL
In order to simplify the above transfer functions, Padé approximations is applied.
The aim of this approach is to reduce the transcendental nature of the above transfer
functions to ones that contain only rational functions of simple polynomials in the
64 Chapter 4: Padé Approximation of Linearized P2D Model
Laplace domain [36]. It is noted that an (M, N) order Padé approximation of a given
transfer function, iG s , centered at 0s has the form [36],
0
0
Mi m
mi m
Ni n
n
n
a s
P s
b s
, (4.17)
where the superscript i associates the quantity with the corresponding transfer
function, namely, phase potential, molar flux, overpotential, lithium concentration and
solid phase potential as given by Eq.s (4.1) – (4.9) above. In Eq. (4.17) the numerator
has order M with M coefficients i
ma and the denominator has order N with N
coefficients i
nb .
The M+N+1 equations used to solve for the coefficients i
ma and i
nb , of iP s can be
determined from the polynomial equation [36],
2
0 0 0
0N M N M
i k i n i m
k n m
k n m
s b s a s
, (4.18)
where the coefficients i
k can be determined from the power series expansion of
the transfer function ( )iG s , namely,
0
1( )
!
ki i
k k
s s
dG s
k ds
. (4.19)
Equation (4.19) generates a polynomial of order N(N+M+2) in s [36]. Its right hand
side equals zero for all s, therefore, the coefficients of s equal zero [36]. Consequently,
a set of M+N+1 linear equations given in Eq. (4.20) are created, which includes N
equations dependent on i
nb and M+1 equations dependent on i
ma and i
nb , namely,
0
0
0 , 1, ...,
, 0, ...,
Ni i
n M j k
n
ki i i
m k m k
m
b k N
b a k M
. (4.20)
Chapter 4: Padé Approximation of Linearized P2D Model 65
where the zeroth order term, 0
ib , is assumed to be equal to 1 to normalize the
solutions [36]. Solving the equations in Eq. (4.20), the M+1 coefficients,
, ( 0,..., )i
ma m M , and N coefficients, , ( 1,..., )i
nb n N can be obtained [37, 84].
There are several readily available routines that automate the above methodology.
In this chapter the 0PadeApproximant , , ,iG s s M N function in Wolfram
Mathematica [85] is used to determine the Padé approximations to the transfer
functions.
As a specific example it is noted that the transfer function of the molar flux in the
negative electrode, app, /negJ z s I s , given by Eq. (4.4), can be approximated to first
order by the Padé approximation,
neg neg
neg neg
neg
1 0
app 1 0
, J J
J J
J z s a s a
I s b s b
, (4.21)
where,
neg
neg
s,0neg neg neg
eff s s totneg
s,e1 s
neg neg neg neg neg neg 2
eff eff eff eff eff eff
2neg2
effneg neg neg 2
s s sneg neg
eff eff
( )4 5
3
2 6 3
4 5 23
13 10 4 1 5 2
J
neg
OCP cR D FR
ca R
z z
z zR a D FL
z z
z z
, (4.22)
neg
neg
,0 neg neg neg neg neg neg neg neg 2
0 s eff eff eff eff eff eff eff
,
( )180 2 6 3
sJ
s e
OCP ca D z z
c
, (4.23)
neg
neg
,0neg neg neg
eff s s totneg neg
,1 s s
neg neg neg neg neg neg 2
eff eff eff eff eff eff
n
effneg neg neg neg 2 neg neg
s s s s eff eff
( )12 5
2 6 3
s
J negs e
neg neg
OCP cR D FR
cb Aa FL R
z z
Aa FL R a D FL
2eg 2
neg 2 2
eff
8 15 2
7 15 2
z z
z z
, (4.24)
neg
neg negnegeff effs,0neg neg neg neg neg
0 s s eff eff neg neg neg 2
s,e eff eff eff
2( )180
6 3
JOCP c
b Aa D F Lc z z
. (4.25)
66 Chapter 4: Padé Approximation of Linearized P2D Model
Similarly, the remaining transcendental transfer functions, introduced above, can
be approximated by their rational polynomial, Padé approximants, given in Figure 4.2.
In this way a system of transfer functions is developed that represent the variables of
the P2D model for a lithium ion battery and which have the advantage of being easy
to implement and compute in a real-time application. In addition, the physical meaning
of the approximation model can be preserved since the coefficients of these rational
polynomial transfer functions directly link to the physical parameters of the cell such
as filmR , effr ,
sD , eD ,
e and s . If the operating conditions of the cell and hence the
values of the physical parameters change, these coefficients are readily updated as they
themselves have computationally simple algebraic forms as shown in Eq.s (4.22) –
(4.25). This is in distinct contrast to the DRA approach discussed above. It is noted
that the outputs of the Padé approximants are used within their corresponding
nonlinear correction term, given by Eq.s (4.10) – (4.15), in order to improve the
accuracy of the approximations to the purely linear P2D model [7].
Figure 4.2 A low order Padé approximation model of the linear P2D model.
Figure 4.3 compares the frequency response of the Padé approximant with their
transcendental transfer function counterparts for the four internal electrochemical
variables within the negative electrode. It is noted that the frequency response of each
Chapter 4: Padé Approximation of Linearized P2D Model 67
set of transfer functions (where a set consists of a number of curves corresponding to
the range of z values between z=0 and z=1) depicted in Figure 4.3 is for a particular
set-point p*. A range of p* values for each set is considered, resulting in a family of
sets, with each member of this family corresponding to a different p*. From this family
of sets, the actual p* value that was used to produce a given set shown in Figure 4.3
was the value that maximized the root mean squared error (when compared to the
linearized P2D model). This ensures that the frequency responses of the Padé
approximations shown in Figure 4.3 are the worst case scenarios (i.e. the ones with
maximum error when compared to the linearized P2D model).
Figure 4.3 Frequency response of the Padé approximants and transcendental transfer
functions of the four variables within the negative electrode.
It can be observed that the error between the Padé approximants and the
transcendental transfer functions increases as the input frequency is increased.
However, it is noted that in grid applications low frequencies dominate the battery
68 Chapter 4: Padé Approximation of Linearized P2D Model
input current profile as shown in Figure 4.4. Figure 4.4(a) depicts the current profile,
reported by Li et al. [86], of a battery storage system that is being used to smooth the
power generated from a wind farm connected to a grid. Figure 4.4(b) shows the fast
Fourier transform of the current signal in Figure 4.4(a). Figure 4.4(c) plots the Power
Spectrum Density (PSD) and identifies 99% occupied bandwidth [87], which is
calculated by determining where the integrated power crosses 0.5% and 99.5% of the
total power in the spectrum [87]. As shown in the Figure 4.4(c), 99% of the occupied
bandwidth of the power signal have frequencies less than 8.225 mHz. Comparing the
Padé approximants with the linearized P2D model outputs in the mHz frequency in
Figure 4.3 it is observed that the two responses show an acceptable match, suggesting
that it is reasonable to apply the low order Padé approximants for grid scale
applications.
Chapter 4: Padé Approximation of Linearized P2D Model 69
Figure 4.4 Current input and corresponding frequency content of a battery storage
system that is being used to smooth the power generated from a wind farm connected to a
grid. Current data sourced from [86].
4.4 RESULT AND DISCUSSION
Comparisons of Padé approximants of various cell variables with those obtained
from the full P2D model under various applied current conditions are given in Figure
4.5, Figure 4.6 and Figure 4.7. The P2D model has been implemented using the open
source code supplied by Torchio et al. [87]. The cell parameters used for these
simulations were obtained from [7] and are listed in Table 4.1.
70 Chapter 4: Padé Approximation of Linearized P2D Model
Table 4.1 Cell parameters used in the simulation
Symbol Units Negative
electrode Separator
Positive
electrode
Li μm 128 76 190
sR μm 12.5 _ 8.5
A 2m 1 1 1
1Sm 100 _ 3.8
s _ 0.471 _ 0.297
e _ 0.357 0.724 0.444
brug _ 1.5 1.5 1.5
max
sc 3molm
26,390 _ 22,860
,0ec 3molm
2000 2000 2000
,mini _ 0.05 _ 0.78
,maxi _ 0.53 _ 0.17
sD 2 1m s 3.9E-14 _ 1.0E-13
eD 2 1m s 7.5E-11 7.5E-11 7.5E-11
0t _ 0.363 0.363 0.363
k 1/2 5/2 1mol m s 1.94E-11 _ 2.16E-11
_ 0.5 0.5 0.5
filmR 2m 0 _ 0
eff eff brug eff brug
e e e, ,s eD D
2 4 7 2 10 3 14 4
e e e e4.1253x10 5.007x10 4.7212 10 1.5094 10 1.6018 10ec c c c c
neg
OCP 0.16 1.32exp 3 10.0exp 2000U
pos
OCP 0.4924656
6
14.19829 0.0565661 tanh 14.5546 8.60942 0.0275479 1.90111
0.998432
0.157123 exp 0.04738 0.810239 exp 40 0.133875
U
Figure 4.5 compares the electrolyte concentration ,e totc L t , 0,ec t and the surface
concentration in the solid phase , ,s e totc L t , , 0,s ec t and cell voltage for a pulse current
profile. Figure 4.5 shows a similar comparison when the Urban Dynamometer Driving
Schedule (UDDS) current profile is applied. Figure 4.6 shows the comparison when
Chapter 4: Padé Approximation of Linearized P2D Model 71
the input current is that of the wind farm, grid application obtained from [86], as
depicted in Figure 4.4(a).
Figure 4.5 Comparison between the Padé approximation model and the P2D model for
a pulse current profile
72 Chapter 4: Padé Approximation of Linearized P2D Model
Figure 4.6 Comparison between the Padé approximation model and the P2D model for
the UDDS applied current profile.
Chapter 4: Padé Approximation of Linearized P2D Model 73
Figure 4.7 Comparison between the Padé approximation model with the full
(nonlinear) P2D model for the applied current profile associated with the wind farm
application, as depicted in Figure 4.4(a).
In all of these cases the Padé approximations compare very well to the P2D model
outputs. The computation times (in seconds) for each of these applied current profiles
is shown in Table 4.2. It is noted that all simulations were implemented in MATLAB
R2016b and run on the same desktop computer with an Intel Core i7-6700 CPU
running at 3.4GHz and 16GB RAM. In addition, the Padé approximants are
determined at the same discrete spatial locations as given by the FVM grid used to
74 Chapter 4: Padé Approximation of Linearized P2D Model
solve the P2D model. It is observed that the Padé approximations require
approximately 9.6s to simulate the pulse current profile in comparison to the P2D
model, which takes approximately 12s. The time savings of the Padé approximations
are more evident in the cases of the UDDS and grid current profiles, with the Padé
approximation model solving in approximately 7 and 42.5s, respectively and the P2D
model taking approximately 565 and 22,494s, respectively. It is this feature of the
reduced model coupled with the fact that the form of the Padé approximations, once
obtained, only require relatively simple rational functions to be evaluated, which leads
to the significant computational savings observed in Table 4.2.
Table 4.2 Comparison of simulation time of the proposed approximation model to
compute all cell variables at all discrete spatial locations
Simulation time (s) Pulse current
profile UDDS
Grid current
profile
Proposed
approximation model 9.546 s 7.198 s 42.519 s
P2D model 15.316 s 565.823 s 22494.1 s
Like the P2D model, the Padé approximation model can be used to predict how
important cell variables will evolve both spatially and temporally. However, another
significant advantage of the Padé approximation approach is that variables can be
evaluated at discrete locations within the domain, without the need to compute all
values of the variable at all discrete locations, as is the case with the FVM
implementation of the P2D model. This can result in further significant time savings.
For instance, as shown in Eq. 4.16, the transfer functions of
, e, , , , , , ,s e ej x t c x t x t c x t and ,x t at only two locations, totx L and 0x , are
required to compute the cell voltage. If only the cell voltage is calculated then, as
shown in Table 4.3, the computation time of the Padé approximation model reduces to
approximately 4.7s for the pulse current profile, and approximately 4.5 and 29.1s for
the UDDS and grid-scale current profiles, respectively.
Chapter 4: Padé Approximation of Linearized P2D Model 75
Table 4.3 Comparison of simulation time of the proposed approximation model to
compute only cell voltage
Simulation time (s) Pulse current
profile UDDS Grid current profile
Proposed
approximation model
4.726 s 4.5322 s 29.063 s
P2D model 15.316 s 565.823 s 22494.1 s
The application of the proposed reduced model to battery charging is now
considered. The traditional constant current-constant voltage (CC-CV) charging
method is widely used to charge lithium batteries [88]. As the name suggests, this
method applies constant current (CC) to charge the battery to a preset voltage at which
time the charging mode switches to a constant voltage (CV) regime until the current
reduces to a preset value. The CV regime is often lengthy resulting in a long charging
time and in addition, as the battery ages, the preset voltage limit for the CC regime
often results in the cell being overcharged [4]. When the cell is being overcharged, it
can be damaged and the risk of explosion is potentially increased [89]. A faster and
safer method of charging has been reported by Chaturvedi et al. [89], and is based on
observing the value of the potential difference s e at the negative
electrode/separator boundary ( negx L in Figure 4.1). It is noted that at this point,
s e , will be a minimum within the negative electrode during charging [90]. Thus,
when s e at negx L is negligible (point B shown on Figure 4.8(a)), charging should
be stopped to avoid overcharge and this is the reason for monitoring this point in the
approach presented by Chaturvedi et al [89]. Figure 4.8(a) depicts the cell voltage and
Figure 4.8 (b) depicts the over potential, s e , as given by the Padé and P2D models
for a single charge step at a rate of 1C. In the voltage limit method, the cell voltage is
monitored and the charging process is switched from constant current charging to a
constant voltage mode when the cell voltage reaches 4.2V. From Figure 4.8(a), at the
point of switching (point A, 4.2V) this corresponds to a state of charge (SOC) of
93.2%. This SOC will continue to increase under the constant voltage regime, but will
76 Chapter 4: Padé Approximation of Linearized P2D Model
do so very slowly. From Figure 4.8(b), however, at the point of switching the potential
difference s e at negx L is greater than 0V. Therefore, under the method proposed
by Chaturvedi et al. the charger would continue to charge at a constant current until
0,s e yielding a 6.8% extra SOC. The additional time required for gaining 6.8%
extra SOC in this constant current regime is 243s, which is much faster than the time
required to gain the same amount of charge if the charger had switched to a constant
voltage regime (which would take on the order of hours). The pivotal point here is that
the values of s e at negx L predicted by the Padé approximation model compare
very favorably to those predicted by the P2D model but have the advantage of being
easily computed. An accurate and fast prediction of the s e values at negx L , which
cannot be easily measured empirically, means that the Padé approximation method
could be useful for control purposes in fast charging.
Figure 4.8 Comparison between two charging methods (a). Cell voltage during
charging. (b). The over potential s e at negx L
Chapter 4: Padé Approximation of Linearized P2D Model 77
Finally, it is noted that the accuracy of the proposed Padé approximation model
can be further improved by increasing the order of the rational functions used in the
approximations. However, higher order models result in higher computational
requirements. In order to investigate this tradeoff, the approximation order is increased
as shown in Figure 4.9.
Figure 4.9 Higher order Padé approximation model of the linear P2D model
Figure 4.10 depicts the voltage error (when compared with the P2D model
predictions) for the low order approximations given in Figure 4.2, and the higher order
approximations given in Figure 4.9 for the three previously applied current profiles
(i.e. the pulse current profile, UDDS, grid-scale application). Table 4.4 shows the root
mean square value of the voltage error over the length of each current profile. From
these figures it can be observed that the accuracy of the higher order approximation
model increases by some 5.4% and 1.1% for the pulse current profile and the UDDS
profile, respectively. In the pulse current profiles and the grid-scale profile, there are
negligibly small differences between the low and higher order approximations models.
However, as can also be seen in Table 4.3, the computational time requirement of the
higher order approximation model increases by approximately 20% for the pulse and
UDDS current profiles and 7.7% for the grid current profile. It can be observed that
78 Chapter 4: Padé Approximation of Linearized P2D Model
the higher order approximation model does not result in significant improvements for
any of these applied current profiles and therefore, increasing the approximation order
is unnecessary.
Table 4.4 Comparison of RMS error of cell voltage and the computational workload
between low order and higher order Padé approximation models
Pulse current
profile
UDDS current
profile
Grid current
profile
Order selection 1 (low
order)
0.0098 V 0.0091 V 0.00822 V
Order selection 2
(higher order)
0.0093 V 0.0090 V 0.00824 V
Voltage error reduce 5.4% 1.1 % -0.24%
Order selection 1 (low
order)
4.726 s 4.532 s 29.063 s
Order selection 2
(higher order)
6.053 s 5.5 s 31.486 s
Computation increase 21.9 % 17.6 % 7.7 %
Chapter 4: Padé Approximation of Linearized P2D Model 79
Figure 4.10 Cell voltage error obtained from low order and higher order Padé
approximation models when compared to the predictions of the P2D model.
4.5 CONCLUSIONS
In this chapter, a reduced order model for a lithium ion battery is presented in
which Padé approximants were used to simplify complex, transcendental, transfer
functions associated with the linearized, pseudo 2-dimensional (P2D) electrochemical
80 Chapter 4: Padé Approximation of Linearized P2D Model
model of the battery. The resulting transfer functions take the form of simple, rational
polynomial functions, which can be used to compute any variable at any location
within a one-dimensional representation of the battery domain. Corrections for
nonlinear behaviour are also incorporated into the reduced model. The proposed model
substantially reduces the complexity of the P2D model whilst maintaining the
functionality of predicting all of the variables of that model. The form of the Padé
approximations, once resolved, only require relatively simple rational functions to be
evaluated, which leads to significant computational savings when solving the reduced
model. The results demonstrate that the reduced model predictions match very closely
with those obtained from the full (nonlinear) P2D model but in a fraction of the
computational time. Importantly, variables can be evaluated at specific discrete
locations within the domain, without the need to compute all values of the variable at
all discrete locations, as is the case with the FVM implementation of the P2D model.
This can result in further significant time savings and enhance the flexibility of the
model for a variety of applications. This contribution closes the research gap and fulfils
the requirement on grid-connected battery storage systems. The developed
computationally efficient models provide insightful information of electrochemical
variables which therefore fulfils the need of industry. In the next two chapters, this
proposed model and its advantages will be adapted to couple with thermal and aging
effects.
Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells 81
Chapter 5: Coupled electrochemical-thermal
model for small-format lithium ion
cells
5.1 INTRODUCTION
One of critical functions of a battery management system is the thermal
management. Information of evolution of cell temperature is essential for handling the
thermal issues including electrical cell unbalance within a battery pack due to
temperature difference and thermal runaway [13].
Previous chapters are concerned with the modelling of the electrochemical processes
within the battery. Thermal issues have not been considered there. In this chapter, the
focus is on developing a simpler electrochemical-thermal model for small-format
cylindrical cells. This model takes advantage of the Padé approximation model that was
developed in Chapter 4. This model can compute the cell voltage by using only rational
polynomial transfer functions at specific locations in the cell’s domain, without the need
to compute all the variables at all discrete locations, as in the case with the finite volume
method implementation of the P2D model [59]. Due to the uniform temperature over
the radius of small-format cylindrical lithium ion cells, in particular the popular type
18650 lithium ion cell, the spatial temperature dependence can be assumed to be
negligible [8]. The simplified thermal model, which neglects the spatial distribution of
temperature along the cell radius, is then coupled with the Padé approximation model
developed in Chapter 4. Consequently, the resulting model has low computation burden
whilst it can perform reasonably well in predicting cell voltage, specific
electrochemical variables at specific cell locations and the evolution of temperature
over the duration of the cell operation. The proposed model is verified by comparing
the results obtained based on this model with that obtained from solving the 1D radial
electrochemical-thermal battery model contained in Comsol.
82 Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells
5.2 SIMPLIFIED THERMAL MODEL FOR SMALL-FORMAT
CYLINDRICAL CELLS
The one dimensional radial equation governing temperature distribution along
the cell radius is obtained from [91] as:
( , ) 1 ( , )
( , )p gen
T r t T r tc r q r t
t r r r
, (5.1)
where, 3[ ]kg m is the material density, pc [ ]J kgK is the specific heat and
[ ]W mK is the thermal conductivity. The left hand side of Eq. (5.1) models the
accumulation of thermal energy per unit volume. The first term on the right hand side
of Eq. (5.1) models the heat flux due to the thermal conductivity of the material.
The last term genq is the heat generated per unit volume calculated using Eq. (5.2) [92],
respectively.
gen chem cq q q , (5.2)
where,
cellchem cell cell
c c
OCVI Iq V OCV T
V V T
,
and,
2
collectorc
s c
R Iq
A V .
Assume a
( , )0 , 0
T r tr R
r
(
aR is the outer radius of the cylindrical cell) as
the temperature distribution along the radius in small-format cylindrical cell is
negligible [8]. This yields the simplified thermal model Eq. (5.3).
( , )
( ) ( )p gen conv
T r tc q t q t
t
, (5.3)
where, the rate of convective heat removal per unit volume is given as
c amb
conv
c
hA T Tq
V
. (5.4)
Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells 83
During battery operation, the temperature changes and this in turn impacts the
model parameters such as solid phase diffusion coefficient, reaction rate, electrolyte
diffusion coefficient and the conductivity [8]. The changes in these physicochemical
properties can be calculated based on the Arrhenius’s law on temperature dependency
Eq. (5.5) [8].
1 1
exp aref ref
E
R T T
, (5.5)
where, represents the changing parameter at the temperature T. ref is the
reference value of that parameter at the reference temperature refT . For the purpose of
this study, the estimated values of these parameters can be adopted from [93].
The temperature dependence open circuit potential (OCP) of the battery is
approximated by a first-order Taylor series expansion with respect to temperature, as
given by Eq. (5.6) [91]:
ref
ref
OCPOCP OCP T T
T
. (5.6)
5.3 COUPLED ELECTROCHEMICAL THERMAL MODEL FOR SMALL-
FORMAT CELL
The proposed model consists of the aforementioned reduced order
electrochemical model using the Padé approximation method and the simplified
thermal model. The interplay between the electrochemical and thermal models is as
shown in Figure 5.1: the reduced order electrochemical model calculates the variations
of the total heat generation which are inputted to the simplified thermal model.
Rational transfer functions of electrochemical variables at totx L and 0x are
computed to determine the cell voltage and the open circuit voltage OCV . The heat
generation rate is then calculated using Eq. (5.2). Simultaneously, the thermal model,
which is coupled with the electrochemical model, computes the evolution of the cell
temperature which is in turn fed back to the reduced-order electrochemical model. The
model is then implemented in Simulink using a similar approach proposed in Chapter
84 Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells
3. It is noted that the present model is simpler than the one that will be discussed in the
next chapter for the large-format cell geometry depicted in Figure 5.2(b). In that work,
each wind is represented by an electrochemical model.
Figure 5.1 A schematic representation of the proposed cell model
Neg Sep Pos
app
,iJ z s
I s
app
,i z s
I s
s,e
app
,iC z s
I s
e
app
,iC z s
I s
q t
T t( )
( )p
T tc q t
t
FET cell model
Applied current Iapp,
Convection coefficient h,
Ambient temperature Tamb
Variation of specific internal
electrochemical variables and
temperature with time
e
app
,i z s
I s
Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells 85
Figure 5.2 (a) Cross-section view of a small-format cylindrical cell which shows
negligible temperature variation along the cell’s radius, (b) Cross-section view of a large-
format cylindrical cell considered in [94] which shows significant temperature derivation
along the cell’s radius
5.4 RESULT AND DISCUSSION
In order to verify the accuracy of the model, the results obtained using the
proposed model are compared with that using the 1D radial electrochemical thermal
PDE model implemented in Comsol Multiphysics software. Figure 5.3 and Figure 5.4
show the outcome of the comparison on the computed temperature variation, the
voltage and other specific variables such as lithium concentration in the solid phase
and electrolyte phase for 18650 cell. The test cases correspond to that during a 1C
charge/discharge cycle and Urban Dynamometer Driving Schedule (UDDS) cycle
[12]. The time-step used in the simulation is 1s.
86 Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells
In both cases, the results obtained using the proposed model compare favorably
with that using the Comsol 1D radial electrochemical-thermal PDE model. All the
simulations were carried out on the same desktop computer with an Intel Core i7-6700
CPU running at 3.4GHz and 16GB RAM. The computation times for each of the test
cases are shown in Table 5.1. It is observed that the proposed model requires
approximately 2s to simulate a charge/discharge cycle, in comparison to the Comsol
model which takes approximately 6 minutes. The time saving of the proposed model
is even more significant in the case of the UDDS test whereby the proposed model
took approximately 2s to complete the simulation while the Comsol model required
some 40 minutes. It is also noted that the reported computation time of the Comsol
model shown in Table 5.1 is based on extremely coarse physics-controlled meshes,
with built-in parameter sets in Comsol. The computational time will be even higher in
the case of finer meshes. It is this feature of the proposed model which combines the
reduced order electrochemical model with the Padé approximations and the simplified
thermal model which leads to the significant computational savings. In addition, as
shown in Figure 5.3 and Figure 5.4, it is observed that the relative small temperature
differences radially under different convection coefficient h values (h=10W/m2/K
represents natural convection and higher values represent forced convection conditions
[95]). This observation confirms that the assumption of uniform temperature is
reasonable. Therefore the proposed model can be used for small-format cylindrical
cells.
Table 5.1 Comparison of simulation time
Simulation time (s) Proposed
model
1D radial P2D
model
1C discharge-rest-charge h=5 W/m2/K 2.21s 6min 14s
h=10 W/m2/K 2.44s 6min 23s
h=100 W/m2/K 2.16s 6min 26s
UDDS h=5 W/m2/K 2.48s 39min 37s
Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells 87
Simulation time (s) Proposed
model
1D radial P2D
model
h=10 W/m2/K 2.43s 33min 17s
h=100 W/m2/K 2.29s 40min 33s
In the case of a large-format cell such as that with the geometry reported in [94]
where the cell diameter is 45mm, the present study shows that the proposed model
fails to predict the cell temperature accurately. High convection coefficients induces a
large divergence of temperature between the inner and outer winds of the cell [94], as
illustrated in Figure 5.2(b). This fact contradicts the assumption made in developing
the proposed model in that a uniform temperature profile is assumed along the cell
radius. Therefore, large error in the predicted temperature can be observed in case of
large format cell, as shown in Figure 5.5 and Figure 5.6. Consequently, for small-
format cylindrical cell, particularly the 18650 cell, the proposed model would be useful
as it provides favourably accurate temperature prediction whilst requires low
computation burden. For large format cell, a more sophisticated electrochemical-
thermal model would be more suitable.
88 Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells
Figure 5.3 Comparison between the proposed cell model and the rigorous 1D radial
PDE model in Comsol using1C charging/discharging of a small-format cell (18650 cell)
Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells 89
Figure 5.4 Comparison between the proposed cell model and the rigorous 1D radial
PDE model in Comsol using UDDS current profile of a small-format cell (18650 cell)
90 Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells
Figure 5.5 Temperature error between the proposed cell model and the Comsol 1D
radial PDE model: 1C charging/discharging of a large-format cell.
Figure 5.6 Temperature error between the proposed cell model and the rigorous 1D
radial PDE model in Comsol using UDDS current profile of a large-format cell.
Chapter 5: Coupled electrochemical-thermal model for small-format lithium ion cells 91
5.5 CONCLUSIONS
This chapter presents a computationally efficient electrochemical-thermal model
developed for small-format cylindrical lithium ion cells. It is shown that the proposed
model is able to predict the behaviour of the cells satisfactorily. In comparison to the
1-D radial PDE model, the proposed model enjoys a much reduced computational
burden which enhances its applicability for a variety of real-time battery management
applications. It is noted that the application of the proposed model in this chapter is
limited for small-format cells due to the assumption of uniform temperature over the
cell radius. In large-format cylindrical lithium ion cells, the temperature profiles are
no longer uniform and hence, the assumption used in this chapter is no longer valid.
It is also noted that the main contribution of this chapter is to develop a low
burden computation electrochemical-thermal model for small-formatted cells. This
model can allow users to avoid the use of a more complex model that is developed and
discussed in the next chapter. In that chapter, a sophisticated electrochemical-thermal-
degradation model will be developed, which accounts for the variation of the
temperature and degradation along the radius of large-format cylindrical cells.
92 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
Chapter 6: A Computationally Efficient
Electrochemical-thermal-degradation
Model for Large-format Lithium Ion
Cells
6.1 INTRODUCTION
Chapter 5 discussed a simple model for small-format cylindrical cells, however,
it is not suitable for large format cylindrical batteries in which non-uniformities in
temperature and degradation occur within the cells [16]. Furthermore, such non-
uniformities exacerbate further degradation and temperature gradients within the
battery. Therefore, a sophisticated model is required to represent these dynamics.
Coupled thermal, degradation and electrochemical models are uncommon,
however, Tanim et al. [15] do present a SPM coupled with thermal and SEI models.
The result is a 1D spatial model that ignores non-uniform distributions of temperature
and degradation along the radius of the cell, which, given their dimensions, does not
seem reasonable for large-format cylindrical cells. Smith et al. [16], present an
empirical degradation model coupled with a multi-dimensional, multi-scale (MD-MS)
cell model for large format cylindrical batteries. However, PDE-based models are
computationally expensive and are used in battery design rather than real-time, optimal
control situations. These models are too complex to be implemented in real time for
most control algorithms [96].
In this chapter, an efficient 1-dimensional, radial, coupled degradation-
electrochemical-thermal model of a spirally wound, cylindrical lithium ion battery,
noted hereafter as the DET model, is proposed. The DET model couples previous
reduced order, electrochemical model proposed in Chapter 4 with an approximate
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 93
thermal profile for the cell, based on the steady state solution of the one-dimensional,
radial, heat equation, and an SEI growth model [66]. The model accounts for the central,
electrolyte filled, hollow core and the spirally wound material. The proposed DET
model is computationally efficient whilst maintaining a high degree of accuracy
(compared with that obtained when using the P2D model) in simulating the radial
temperature and degradation distributions within the cell over time. It is a predictive
modelling tool that can be applied to a wide variety of battery applications. The model
is novel in that it applies to large-format, spirally wound, cylindrical batteries and
accounts for non-uniform degradation and thermal behavior coupled to electrochemical
phenomena. The proposed model is able to provide insight into the variation and
evolution of the local temperature and degradation rate of each individual wind along
the cell radius. For the first time, this insight information is computed using the
underlying physics of degradation, rather than the fitting of empirical models.
This chapter is organized in the following way. The next section introduces the
model structure and briefly describe the P2D model and the previous reduced order,
electrochemical model. In the third section, the approximation for the radial
temperature distribution in the cell is developed. Then, the fourth section shows how to
couple this approximation to the reduced order model of each wind in the cell. Finally,
results are presented in fifth section where the proposed DET model is compared to the
full, one-dimensional, radial implementation of the P2D model implemented in Comsol
Multiphysics®.
6.2 MODEL DOMAIN AND REDUCED ORDERS ELECTROCHEMICAL
MODEL IN EACH WIND
6.2.1 Model domain
A schematic of a cylindrical lithium ion cell and its cross-section is given in Figure
6.1(a). The cell has an outer radius aR and inner radius
0R . The cell structure includes
a centrally located hollow core, which is filled with electrolyte. The wound (or “jelly
roll”) structure of the cell consists of repeated electrode, separator and current collector
94 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
layers [61] as shown in Figure 6.1(b). A single “wind” is also defined in Figure 6.1(b) and
consists of a sandwich of 8 layers beginning with the negative current collector and
ending with the negative electrode.
+
- +
neg
s
app
( , )
( )
z s
I s
neg
s,e
app
( , )
( )
C z s
I s
neg
app
( , )
( )
J z s
I s
sep
e
app
( , )
( )
C z s
I s
sep
e
app
( , )
( )
z s
I s
pos
s
app
( , )
( )
z s
I s
pos
s,e
app
( , )
( )
C z s
I s
pos ( , )
( )app
J z s
I s
Neg Pos Pos Neg Neg Pos Pos Neg
neg
e
app
( , )
( )
C z s
I s
neg
e
app
( , )
( )
z s
I s
pos
e
app
( , )
( )
z s
I s
pos
e
app
( , )
( )
C z s
I s
wind ith
Neg Sep Posz=0
...
Ra
R0
r
+- - +
SEI layer
,iT r t ,i kT r t
neg ( , )iq r tsep ( , )iq r t pos ( , )iq r tcollector ( , )iq r t
collector ( , )iq r t
( , ) 1 ( , )( , )p
T r t T r tc r q r t
t r r r
(a)
(b)
(c)
z=1 z=0 z=1 z=1z=0
wind ith+k
Padé model
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 95
Figure 6.1 (a) Illustration of a cylindrical lithium ion battery with spirally wound
design and its cross-sectional view, (b) the component layers in each wind and (c) the
domains that constitute half of each wind and the corresponding Padé model transfer
functions [59].
The proposed model framework is multi-scale and includes a macroscopic
domain, characterized by the radial variable, r (a0 Rr ), in which the temperature
distribution is solved and a microscopic domain, characterized by z ( 0 1z ), in which
the coupled electrochemical and degradation models are solved. In the negative
electrode domain, 0z at the electrode-current collector boundary and 1z at the
electrode-separator boundary. In the separator domain, 0z at the negative electrode-
separator boundary and 1z at the separator-positive electrode boundary. In the
positive electrode domain, 1z at the separator-electrode boundary and 0z at the
electrode-current collector boundary.
In this chapter, temperature, ,T r t , is assumed to vary along the radial
coordinate, r, of the cell in such a way that the temperature within each wind (at the
microscale) is uniform in space, however, the temperature across winds (at the
macroscale) can vary. As the temperature in a wind is assumed to be uniform in space,
and given that the applied current density is identically symmetric over each half
domain of the wind, the electrochemical process occurring inside the three main
domains in the first half of the wind (namely, the negative electrode, the separator and
the positive electrode as shown in Figure 6.1(c)) will be identical to the corresponding
ones in three “mirrored” domains in the other half of the wind (namely, the positive
electrode, the separator and the negative electrode). Therefore, it is only required to
solve for the electrochemical variables in one half of the wind as these values are then
“mirrored” in the remaining half of the wind.
For each wind, an electrochemical model and degradation model in sub-domain
z (the microscopic domain) associated with the corresponding ,T r t is solved.
Specifically, the input of each wind is the temperature ,T r t , which affects the
electrochemical processes including the transport of ions, reaction kinetics and SEI
96 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
growth. The output of the electrochemical model are a series of heat generation rates,
which in turn form the input of the thermal model in the domain r (macroscopic
domain).
The electrochemical characteristics within each wind is solved using a Padé
approximation based, reduced order, P2D model that was previously developed in
Chapter 4 for five state variables; including the electric potential s ,z t in the solid
electrode, the electric potential e ,z t in the electrolyte, the lithium concentration of
the active material s p, ,c z r t of the positive and negative electrodes, the lithium
concentration e ,c z t in the electrolyte, and the molar fluxes ,j z t of the charge that
flows between the active material in each electrode and electrolyte. In this model Padé
approximations are used to simplify the complicated transcendental transfer functions
that result from the Laplace transform solution of the linearized P2D model [7, 97].
Corrections for nonlinear behavior, as given by Lee, Chemistruck and Plett [7], are
incorporated in the model. This approach reduces the complex nature of the above
transfer functions to ones that contain only rational functions of simple polynomials in
the Laplace domain [36]. In this way a system of transfer functions, that represent the
variables of the P2D model for each wind, is developed. These transfer functions have
the advantage in that they can be easily implemented and are computationally efficient.
Here, they are used to calculate the voltage of each wind and the volume average heat
generation rate in each of the domains shown in Figure 6.1(c). The coupling of these
transfer functions with the temperature and degradation models will be discussed in
Section 6.4, but first these models are introduced in the following sections.
6.2.2 Degradation model for each wind
A SEI layer on the surface of the particles that form the negative electrode is one
of the major causes of capacity loss and impedance rise for lithium ion battery [14]. In
this chapter cell degradation is accounted for by including model equations, based on
those developed previously by Ramadass et al. [66], which describe the growth of the
SEI layer during charging.
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 97
The irreversible side reaction, which forms a film at the solid-electrolyte interface
of the negative electrode, is given by,
S 2Li 2e P , (6.1)
where, S is the solvent of the electrolyte and P is the “product” that forms the SEI layer.
The ohmic resistance in the SEI film is related to the thickness of the film,
film ,z t , namely,
film
film film
,, ,0
p
z tR z t R z
. (6.2)
where,
film
film
,,
p
p
Mz tJ z t
t
, (6.3)
and pM is the molecular weight of the SEI film, p is the density of the film, p is the
conductivity of the film.
Here, film ,J z t denotes the molar flux of the irreversible side reaction Eq. (6.1)
and is expressed as,
0,film filmfilm film, exp ,
i FJ z t z t
F RT
. (6.4)
where 0,filmi is the exchange current density for the side reaction Eq. (6.1), R is the
universal gas constant and F is the Faraday’s constant. The overpotential, film ( , )z t ,
of the irreversible side reaction Eq. (6.1) is calculated by,
film s e neg SEI
film film
, , , ,
, , ,
z t z t z t OCP z t OCP
FR z t J z t J z t
. (6.5)
In this chapter, this degradation model is solved at the microscopic scale and is
coupled with the Padé model and the thermal model. Details of this coupling will be
discussed in Section 6.4.
98 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
6.3 THERMAL APPROXIMATION MODEL
In this section an approximation for the radial temperature distribution within the
cell depicted in Figure 6.1 is derived. In Section 6.3.1, the governing equation of heat
conduction, suitable for a 1-dimenasional radial domain, r , is recalled and the heat
generation terms in each microscopic domain, z , are briefly describe. These generation
terms are calculated using the electrochemical variables mentioned in Section 6.2.1. In
Section 6.3.2, an average thermal conductivity for the wound material within the cell is
introduced. This conductivity is used to approximate the governing thermal equation as
shown in Section 6.3.3.
6.3.1 Governing equation of the thermal model and heat generation terms
It is known that temperature variations occurring along the radius of cylindrical
cells are caused by heat conduction in the radial and azimuthal directions [60]. Chen et
al [95] reported that the heat flow in the azimuthal direction of cylindrical lithium ion
cells that have a high number (20 or more) of winds can be negligible [95]. They
showed that when the number of winds is high, heat flow is predominantly radial and
therefore a 1-dimensional, radial model can be used in preference to a 2-dimensional
model to accurately represent the temperature distribution of such cells [95]. In this
chapter, that finding is adopted and a 1-dimensional, radial, thermal model is used to
calculate the temperature distribution within the cell.
The governing equation for the radial temperature distribution within the cell is
given by [91],
( , ) 1 ( , )
( , )p
T r t T r tc r q r t
t r r r
, (6.6)
where, 3[kg m ] is the material density, pc [ J kgK] is the specific heat, [ W mK]
is the thermal conductivity. Boundary conditions for Eq. (6.6) are shown in Figure 6.2
as BC1, BC2 and BC3.
The average heat generation rate per unit volume ( , )q r t 3[ W m ] in each domain
of each wind consists of reversible and irreversible heating, Joule heating in the solid
and solution and resistive heating at the electrode/current collector interface. These
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 99
heating terms are calculated using the electrochemical variables that have been solved
using Padé model at the microscopic scale as described in the Section 6.2.1.
Irreversible heating, iq , is due to electrochemical chemical reactions and is given
by [8],
i s ( , ) ( , )q a Fj z t z t . (6.7)
where,sa is the specific interfacial area.
Reversible heating, rq , results from a change in entropy, namely [8],
r s ( , )OCP
q a Fj z t TT
. (6.8)
where OCP is the Open Circuit Potential of the of the electrode.
Joule heating in the solid, s
Jq , is a result of ohmic resistances and is given by [8],
2eff
s
2
( , )s
Ji
z tq
zL
. (6.9)
where, eff
s is the effective solid conductivity, is the solid conductivity,s is
the volume fraction of the solid phase, iL is the thickness of the electrode, and
neg,posi .
Joule heating in the electrolyte, e
Jq , is ohmically related to the variation of
electrolyte concentration and potential, namely [8],
eff
2effe De e eJ 2 2
e
( , ) ( , ) ( , )1 1
( , )i i
z t c z t z tq
z c z t z zL L
. (6.10)
where, eff Brug
e is the effective electrolyte conductivity, is the electrolyte ionic
conductivity, e is the electrolyte volume fraction, Brug is the Bruggeman coefficient,
eff
D is defined as eff
D eff 0
e
d ln1 1
d ln
fRTt
F c
, 0t is the transference number, f is the
mean molar activity coefficient, and neg, sep, posi .
The heat generation rate per unit volume, collectorq , from current collector is [8],
100 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
2
collectorcollector
c c
( )( , )i
I t Rq r t
AV . (6.11)
where, collectorR is current collector resistance,
cA is the cell surface area, and cV is the
cell volume.
Given the above equations the average heat generation term in each electrode is
calculated as,
1
neg neg neg s, neg e, neg
i
0
( , ) di r J Jq r t q q q q z , (6.12)
1
pos pos pos s, pos e, pos
i
0
( , ) di r J Jq r t q q q q z . (6.13)
whilst the average heat generation term in the separator is calculated only from the
Joule heating in the electrolyte, namely,
1
sep e, sep
0
( , ) di Jq r t q z . (6.14)
The average heat generation rate per unit volume of a wind, q , in Eq. (6.6) is
calculated by summing the products of the average (volumetric) heat generation in
each domain (collectorq for current collector, posq for positive electrode, negq for negative
electrode or sepq for separator) with the volume of that particular domain and dividing
this sum by the total volume of each wind.
6.3.2 Average thermal conductivity for the wound material
In order to simplify Eq. (6.6) in Section 6.3.3, an average thermal conductivity wound
[ W mK] for the wound material (0 aR Rr ) is first introduced as shown in Figure 6.2.
The average conductivity, wound , is expressed in Eq. (6.15) and is based on the concept
of discrete thermal resistances (in each domain of Figure 6.1(c)) in series [98].
Given the temperature difference between the points at layer
i jr r and ir as shown
in Figure 6.2 , the heat transfer rate can be expressed as [98]:
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 101
layer
transfer
layer
cell
layer
2
ln
i i j
i
j
i j
i
T r T r rq
H
r r
r
, (6.15)
where, cellH is the cell height, layer
j is the thermal conductivity of each of the domains,
ir is the radius of each wind and layer
jr is the thickness of each of the domains as shown
in Figure 6.2.
Expression (6.15) has a similar form to Ohm’s law in which the potential
difference ( layer
i i jT r T r r ) across a conductor (layer
layer
cell2 lni j
j
i
r rH
r
) between
two points ( layer
i jr r and ir ) is directly proportional to the current ( transfer
iq ) through it.
Therefore, the heat transfer problem in a composite cylinder, which consists of
multiple domains, each with a different thermal conductivity layer
j , can be represented
as a system of thermal resistances connected in series as shown in Figure 6.2. The
thermal resistance of a single domain in the composite cylinder is given as [98]:
layer
wound
layer
ln
2
i j
ii
cell j
r r
rR
H
. (6.16)
The total resistance of the wound material, woundR , can then be expressed as [98]:
layer
wound
layer1
ln
2
i j
ni
i cell j
r r
rR
H
. (6.17)
Treating the whole wound material region as a single layer as shown in Figure
6.2 the average thermal conductivity, wound , can be defined as:
a
wound 0
wound
cell
Rln
R
2 H R
. (6.18)
The average thermal conductivity wound can therefore be obtain as:
102 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
a
wound 0
layer
layer1
Rln
R
1ln
ni j
i j i
r r
r
. (6.19)
Figure 6.2 Illustration of cross section of the cylindrical lithium ion battery with
boundary conditions
6.3.3 Approximation of the thermal model
In order to solve the governing PDE in Eq. (6.6), some methods such as finite Fourier
transforms [99] and spatial discretization methods [57, 58] can be used; however, they
0
0 ,T r t
R0Ra
00
0
( , )BC1: 0
r
T r t
r
a
woundwound wound
a
( , )BC2: ( , ) amb
r R
T r th T R t T
r
woundwound0 0 0
0
wound
0 0 0
( , ) ( , )
BC3:
( , ) ( , )
T R t T R t
r r
T R t T R t
wound ,T r t
layer
layer
cell
ln
2
i j
ii
j
r r
rR
H
wound
ir
layer
i jr r
layer
i jT r r
iT r
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 103
require high computational workloads [4]. For the feasibility of using DET model in
control design, an approximation profile for Eq. (6.6) is derived, which is based on its
steady state solution.
At steady state ( ( , ) 0T r t t ) and assuming a uniform constant source term q , Eq.
(6.6) becomes,
1 ( )T r
r qr r r
, (6.20)
which has the general solution [98],
2
1 2 0 a( , ) ( )ln ( ), R R4
qT r t r C t r C t r
. (6.21)
Noting Eq. (6.21) an approximation of the temperature within the cell is proposed
and given by,
0 0
wound 2 2
a a 0 a
( , ) ( ), 0 R( , )
( , ) ( ) ( ) R ( ) ln ln R , R R
amb
amb
T r t T D t rT r t
T r t T A t B t r C t r r
. (6.22)
where the initial temperature along the cell radius is equal to the ambient temperature,
ambT , namely,
amb( ,0)T r T . (6.23)
It is noted that the approximation given in Eq. (6.22) satisfies a symmetry boundary
condition at 0r (BC1), namely,
00
0
( , )0
r
T r t
r
. (6.24)
The boundary condition at aRr (BC2) describes the convective transfer of heat at
the outer surface of the cylindrical battery, namely,
a
woundwound wound
a amb
( , )(R , )
r R
T r th T t T
r
, (6.25)
where h 2[W m K ] is the convection coefficient.
104 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
Thermal continuity is assumed at the boundary 0Rr (BC3 in Figure 6.2), thus,
wound
wound0 0 00
(R , ) (R , )T t T t
r r
, (6.26)
and
wound
0 0 0(R , ) (R , )T t T t . (6.27)
Substitution of the approximate form of ,T r t given in Eq. (6.22) into the boundary
conditions Eq. (6.25) to Eq. (6.27) gives,
0
0
2 2
0 a 0 a
1 a
a
10 ( )2R ( )
R
( ) ( ) ( ) R R ( ) ln R ln R
1( )2R ( ) ( )
R
B t C t
D t A t B t C t
B t C t hA t
(6.28)
Solving for ( )B t , ( )C t and ( )D t in terms of ( )A t , yields,
2
0
2 2
0 a 0 a
wound a
2
0 a
( ) ( ) ( )2
( ) 1 R R ln R ln R ( ) ( )
( ) ( ) ( )R 1
R R
B t A t A tR
D t A t A t
hC t A t A t
(6.29)
Now, integrating both sides of Eq. (6.6) gives,
0 a 0
0
a a
0 0
R wound
0 00
0 0
woundwound
( , ) ( , )( , ) 1
1 ( , )
R R
p p
R
R R
domain
R R
T r t T r tT r tc rdr c rdr r rdr
t t r r r
T r tr rdr q rdr
r r r
. (6.30)
Substituting Eq. (6.22) and Eq. (6.29) into Eq. (6.30) and integrating, namely,
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 105
0
0
0
0
0
2 2
0 ,0 , a a
0
2 2 2N M20
0 ,0 , a a
1 1
( , ) ( , )LHS of Eq.(6.30)
( ) ( )1 ln ln
( ) ( )1 ln ln
2 2 2 2
a
a
j
j
i j
R R
p p
R
R R
p j p
R
p j p
i jr L
T r t T r tc rdr c rdr
t t
dA t dA tc rdr c r R r R rdr
dt dt
RdA t dA t r rc c R r R
dt dt
1
2 2
1 1 2
, a 1 a
N M
0 ,0 0 2 21 1
1 2
, a 1 a
1 ln ln2 2 2
( ),
1 ln ln2 2 2
i j
j
j
r L
i j i j
j p i j
p
i ji j i j
j p i j
r L r Lc R r L R
dA tc R
dt r L r Lc R r L R
0 a a
0 00
2 2N M
1
1 a a
1 1a
1 ( , ) 1 ( , )RHS of Eq.(6.30)
1( ) 2 , .
2 2
R R R
domain
R R
i j i jdomain
j i
i j
T r t T r tr rdr r rdr q rdr
r r r r r r
r L r LA t R R q r t
R
Now defining , , and as:
2 2
1 1 2
, a 1 a
N M
0 ,0 0 2 21 1
1 2
, a 1 a
1 ln ln2 2 2
1 ln ln2 2 2
j
j
i j i j
j p i j
p
i ji j i j
j p i j
r L r Lc R r L R
c R
r L r Lc R r L R
,
wound
a a
a
12R R
R
,
and
2 2N M
1
1 1
,2 2
i j i jdomain
i
i j
r L r Lq r t
, gives,
( )
( )dA t
A tdt
, (6.31)
Here domainq is the heat generation rate of each layer in each wind and is calculated
in Eq. (6.11) - Eq. (6.14), collector, neg, sep, posdomain , N is the number of winds,
neg neg neg neg neg sep pos pos pos sep neg
1 2 collector 3 collector 9 collector collector0, , , ....,L L L L L L L L L L L L L L L ,
M is the number of layers in a wind, neg neg sep pos pos
collector collector, , , , andL L L L L are the thickness of
106 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
the negative current collector, negative electrode, separator, positive electrode and
positive current collector, respectively.
Equation (6.31) is the first order ordinary differential equation (ODE) for ( )A t . Solving
Eq. (6.31) for ( )A t then allows for ( )B t , ( )C t and ( )D t to be determined from Eq. (6.29),
which in turn determines the radial temperature distribution within the battery
according to Eq. (6.22). The proposed approach considers the thermal properties such
as material density, the specific heat and the thermal conductivity of different layer in
the wounded material; however, it requires low computational workload since there is
only one ODE function of ( )A t needed to be solved.
The parameters within the Padé model are temperature dependent. These
parameters include the solid phase diffusion coefficient sD , the reaction rate , the
electrolyte diffusion coefficient eD and the conductivity . These temperature
dependencies can be calculated using an Arrhenius law [8], namely,
1 1
exp aref ref
E
R T T
, (6.32)
where, represents one of the parameterssD , ,
eD , and , at temperature, T, and
ref is the reference value of that parameter at the reference temperature refT [93].
The temperature dependence in the open circuit potential OCP of the electrode is
approximated by a first-order Taylor series expansion, namely [91],
ref ref OCPOCP OCP T T
T
, (6.33)
where, refOCP is the OCP of the electrode at the reference temperature refT .
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 107
6.4 COUPLED DEGRADATION-ELECTROCHEMICAL-THERMAL
MODELS
This section shows how to couple the thermal approximation to the Padé model
and degradation model of each wind in the cell. To illustrate how the DET model is
developed, the cell dimension of the large-format cell reported by Lee et al. in [61] is
used. The cylindrical cell has the inner radius 0R 4mm and the outer radius
aR 22.5mm . The parameters are obtained from [7] and are given in Table 4.1. The
radial geometry of the cylindrical battery consisting of 20 winds with 8 layers in each
wind as illustrated in Figure 6.1. As mentioned in Section 6.2, due to the assumption
that temperature within one wind is uniform in space, electrochemical process in each
wind is represented by one Padé model. Since the negative electrodes and positive
electrodes of the 20 winds connect to the same negative current collector and positive
current collector, respectively, this is modelled by connecting 20 Padé models in
parallel as shown in Figure 6.3. The output voltage and the input current of each Padé
model must satisfy Eq. (6.34) according to Kirchhoff’ circuit laws, namely,
app, app, all
1
cell,1 cell,2 cell,N
( ) ( )
( ) ( ) ... ( )
N
i
i
I t I t
V t V t V t
. (6.34)
The coupling of the degradation, Padé and thermal models is shown
schematically in Figure 6.4. It can be seen that the degradation model at each point z
of each wind is coupled with the Padé model. The heat generation rates collectorq , posq ,
negq and sepq calculated from the electrochemical variables of the Padé model are used
to simulate the temperature variation ,T r t at the macroscopic scale r . In turn the
temperature ,T r t in each wind coupled with the Padé model and the degradation
model. In the large format cell, consisting of 20 winds, there will be 41 coupled “sub-
systems” of equations, 20 of which are Padé sub-systems, 20 are degradation sub-
systems and one of which is a thermal sub-system as shown in Figure 6.4.
108 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
Figure 6.3 Parallel connection of Padé models
Figure 6.4 A schematic representation of the DET cell
Wind 1 Wind 2 Wind nth
Iapp,1
Iapp,all
T1 T2 Tn
Vcell,NVcell,1 Vcell,2
Padé model
for wind 1
Padé model
for wind 2
Padé model
for wind nth ......Vcell,1 Vcell,2 Vcell,N
Iapp,all
Iapp,2 Iapp,N
Iapp,1 Iapp,2 Iapp,N
Neg Sep Pos
1
.
q
1T
.
nq
nT
( , ) 1 ( , )p
T r t T r tc r q
t r r r
…
…Wind 1 Wind nth
Neg Sep Pos
DET model
film
film film
film
film,
,, ,0
,,
p
p
k
p
z tR z t R z
Mz tJ z t
t
film,1 1,J T
Degradation model wind 1
Thermal model
Padé approximation model 1 Padé approximation model nth
app
,iJ z s
I s
app
,i z s
I s
s,e
app
,iC z s
I s
e
app
,iC z s
I s
e
app
,i z s
I s
app
,iJ z s
I s
app
,i z s
I s
s,e
app
,iC z s
I s
e
app
,iC z s
I s
e
app
,i z s
I s
film,1 ,max,1, sR c
… …
Degradation model wind nth
film
film film
film
film,
,, ,0
,,
p
p
k
p
z tR z t R z
Mz tJ z t
t
film, ,max,,n s nR cfilm, ,n nJ T
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 109
6.5 RESULTS AND DISCUSSION
This section demonstrates the performance of the DET model by comparing its
outcomes to those of a full P2D model that has been modified to include degradation
and thermal behavior. In the full P2D model, instead of approximating the PDE
governing equations as presented in the Sections 2.2, and 6.3.1, the coupled governing
PDE equations are directly solved using COMSOL Multiphysics® [57]. Specifically,
the P2D model coupled with the degradation model mentioned in Section 2.2 is solved
to simulate the electrochemical variables on the microscopic scale. Simultaneously,
Eq. (6.6), which governs the temperature on the macroscopic scale, is solved. Same
parameters, that are used for the DET model simulation implemented in
MATLAB/Simulink® [100], are used in the COMSOL simulation. These parameters
are adapted from [7] and are listed in Table 4.1.
Figure 6.5 and Figure 6.6 show simulation results from the DET model, for the
temperature distribution along the cell radius (here expressed in terms of the wind
number) for different values of the convection coefficient h , plotted over time. Figure
6.5 is for an input current of 1C for a single charge-rest-discharge-rest cycle and Figure
6.6 is for an input current associated with a UDDS (Urbane Dynamometer Driving
Schedule) profile [12]. Both of these current profiles are given explicitly in [59, 101].
From each figure it is observed that larger convection coefficient values h , result in
more non-uniform temperature profiles, spatially. In Figure 6.5 it can be seen that the
temperature difference between the outer wind and the inner wind can be as high as
4°C. Alternatively, low values of h , result in higher cell temperatures over time,
however, the temperature distribution along the cell radius is nearly uniform in these
cases.
110 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
Figure 6.5 Simulation result of temperature distribution at different convection
coefficients from DET model during a 1C charge-rest-discharge-rest cycle.
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 111
Figure 6.6 Simulation result of temperature distribution at different convection
coefficients from the DET model during UDDS cycles.
Figure 6.7 and Figure 6.8 show the temperatures at 0Rr and aRr and the cell
voltage over time resulting from the charge-rest-discharge-rest cycle and the UDDS
cycle, respectively, for the DET and P2D models. Curves for different convection
coefficient values are shown. It can be seen that in all of these cases the DET model
results compares favorably to the full P2D model results.
The computation times for each of these applied current profiles is shown in
Table 6.1. All simulations were carried out on the same desktop computer with an Intel
Core i7-6700 CPU running at 3.4GHz and 16GB RAM. I can be observed that the
DET model requires approximately three minutes to simulate the 1C charge-rest-
discharge-rest cycle profile in comparison to the reference model, which takes
approximately eight minutes. The time savings of the DET model are more evident in
112 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
the case of the UDDS profiles, with the DET model solving in approximately 90
seconds, and the reference model taking approximately two hours.
Figure 6.7 Comparison of temperature variation between DET model and the full P2D
model in COMSOL at aRr (outer wind) and
0Rr (inner wind) at different
convection coefficients during a 1C charge-rest-discharge-rest cycle.
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 113
Figure 6.8 Comparison of temperature variation between DET model and the full P2D
model in COMSOL at ar R (outer wind) and
0r R (inner wind) at different
convection coefficients during UDDS cycles.
114 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
Table 6.1 Comparison of simulation time
Simulation time (s) Proposed model 1D radial P2D model
1C discharge-rest-
charge
h=5 W/m2/K 179 s 526 s
h=10 W/m2/K 179 s 526 s
h=50 W/m2/K 187 s 527 s
h=100 W/m2/K 187 s 534 s
UDDS cycles h=5 W/m2/K 91.6 s 2h 12m 8s
h=10 W/m2/K 90.31 s 2h 11m 42s
h=50 W/m2/K 88.6 s 2h15m 6s
h=100 W/m2/K 90.6 s 2h 10m 18s
Figure 6.9 shows the evolution of the degradation at the inner and outer winds
for a periodic 1C charge-rest-discharge-rest current profile and 2100W m Kh . Figure
6.9(c) shows the SEI resistance at 0Rr and
aRr , which can be seen increase over
time. It is also noted that the SEI resistance of the inner wind is higher than that at the
outer wind. This is due to the non-uniform temperature distribution across the cell. The
higher temperatures at the inner radii facilitate the degradation reaction leading to
thicker SEI layers and thus, higher resistances. Figure 6.9(d) shows the maximum
lithium ion concentrations at 0Rr and
aRr , which decrease over time as lithium
ions are removed by the degradation reaction. Furthermore, consistent with
observations for Figure 6.9(c), the maximum concentration in the inner wind decreases
faster than that for the outer wind. Over extended charge cycles it is observed that both
the SEI film resistance and the maximum lithium concentration of the inner wind and
outer wind diverge significantly.
The DET model has the capability of predicting spatial and temporal temperature
and degradation distributions for a spiral wound cell. Based on this information,
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 115
optimal control algorithms that minimize, over time, distributions that degrade the
cell’s state-of-health, can be designed. Examples of such algorithms have been given
in the literature for situations where cell temperature is spatially uniform. Offline
trajectory optimization is reported in [102] and online Nonlinear Model Predictive
Control reported in [77, 103] for these simpler cases. The development of such
algorithms for large format cells is beyond the scope of this thesis but will be addressed
in the future works.
116 Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for
Large-format Lithium Ion Cells
Figure 6.9 Evolution SEI resistance filmR and maximum concentration of lithium ion
s,maxc in solid phase in negative electrode of inner and outer winds ( 2100W m Kh ).
Chapter 6: A Computationally Efficient Electrochemical-thermal-degradation Model for Large-format
Lithium Ion Cells 117
6.6 CONCLUSIONS
This chapter presents a new and computationally efficient one-dimensional, radial,
coupled degradation-electrochemical-thermal model of a large format, spirally wound
cylindrical cell. The Padé approximant model proposed in Chapter 4 is employed to
compute the variation of electrochemical variables and heat generation terms in each
wind and couple it with a degradation model. An approximate model for the radial
temperature distribution of the cell is proposed that is, in turn, coupled with the Padé
and degradation models. The model is able to return accurate predictions when
compared to those of a full P2D model, but in a fraction of the computation time.
The development in this chapter helps to fulfil the requirement on battery models
used in designing the management schemes of battery storage systems. The developed
models are seen to be computationally efficient and can provide insightful information
of battery degradation, thermodynamics and electrochemical variables.
118 Chapter 7: Conclusions
Chapter 7: Conclusions
This chapter provides a summary of the thesis, followed by possible research
directions for future work based on the outcomes of this thesis.
7.1 SUMMARY AND DISCUSSION
In this section, the objectives of this thesis, that are outlined in Chapter 1, are
restated and detail provided on how they were achieved throughout the thesis.
Objective 1: Develop a new adaptive electrochemical model-based adaptive
estimation algorithm.
Chapter 3 has described a practical scheme to implement the three-parameter
SPM in Simulink and an adaptive state estimation algorithm. The scheme also has
provided a novel way to implement other reduced order models and sophisticated
coupled models in Simulink as demonstrated in Chapter 5 and 6. For the first time the
ESPM-SPKF algorithm was proposed for adaptive SOC estimation which is a
combination of ESPM and SPKF. An experiment using a half-cell that was built by
the QUT battery lab with lithium foil and LiFePO4 positive electrode was used to
verify the proposed ESPM-SPKF algorithm. The results shown that the ESPM-SPKF
algorithm was able to converge to the reference SOC within a reasonable time and
tracked the reference SOC thereafter. Both predicted cell voltage and SOC from this
algorithm show excellent agreement with the measured data from the experiments.
These contributions fulfilled the research gaps on implementing electrochemical
models in Simulink and using them in battery state estimation.
Objective 2: Develop a new approximate model of the linearized P2D model.
Chapter 4 has developed a new reduced cell model for a lithium ion battery in
which Padé approximants are used to simplify all complex, transcendental, transfer
functions associated with the linearized, pseudo 2-dimensional (P2D) electrochemical
model of the battery. The resulting transfer functions took the form of simple, rational
polynomial functions, which can be used to compute any variable at any location
Chapter 7: Conclusions 119
within a one-dimensional representation of the battery domain. Corrections for
nonlinear behaviour were also incorporated into the reduced model. This proposed
model was validated by comparing results based on the developed model with those
obtained from the full (nonlinear) P2D model solved by finite volume method and
implemented using an open source code reported by Torchio et al. [87]. The results
obtained using the reduced model compare favourably to those from the P2D model
and the computational time required to produce these results was significantly reduced.
Importantly, the form of the reduced model means that variables can be evaluated at
specific discrete locations within the cell domain, without the need to compute all
values of the variable at all discrete locations, as is the case with the spatial
discretization methods most commonly used to implement the P2D model. This results
in further significant time savings and enhances the suitability of the model for a
variety of applications.
Objective 3: Develop a new computationally efficient coupled
electrochemical-thermal model for small-format lithium ion cells.
A computationally efficient coupled electrochemical-thermal model for small-
format cylindrical lithium ion cells has been developed in Chapter 5. The special
feature of the aforementioned Padé approximation model proposed in Chapter 4 was
taken advantage of by coupling it to the thermal model intended for small-format
cylindrical lithium ion cells in which temperature was assumed to be uniform along
the cell radius. This proposed model was validated by comparing the results based on
this model with those obtained using the full, 1-dimensional, radial P2D model
implemented in Comsol Multiphysics®. The results demonstrated that the objective of
developing a fast computational model while maintaining a reasonable level of
accuracy has been achieved. The application of the proposed model is limited for
small-format cylindrical cells, however.
Objective 4: Develop a new sophisticated coupled electrochemical-thermal-
degradation model for large-format lithium ion cells
For the first time a sophisticated coupled electrochemical-thermal-degradation
model for large-format cylindrical lithium ion cells has been developed in Chapter 6.
The model accounts for non-uniform degradation and thermal behaviour coupled to
120 Chapter 7: Conclusions
the electrochemical phenomena. The Padé approximation model proposed in Chapter
4 was used in this chapter to represent the dynamics of electrochemical variables in
each wind. The SEI film growth model reported in [66] was adapted to represent the
degradation process in each wind. A new approximation of the thermal model was also
proposed. This approach considers the thermal properties of different layers in the
wound material. However, there is only one ODE function which needs to be solved,
resulting in a significant computational saving. The practical scheme proposed in
Chapter 3 was used in this chapter to implement the model in Simulink which is
essential for computationally scaling up the cell model to a battery pack system.
Consequently, the proposed coupled electrochemical-thermal-degradation model was
able to provide insight into the variation and evolution of the local temperature and
degradation rate of each individual wind along the cell radius. The results from the
proposed model matched very closely with those obtained from the full, 1-
dimensional, radial implementation of the P2D model implemented in Comsol
Multiphysics® but only require a fraction of its computational time.
7.2 DIRECTIONS FOR FURTHER RESEARCH
Notwithstanding the encouraging progress made so far, there are a number of
fruitful areas for further works.
Direction 1: Identify parameters for electrochemical models
Battery manufacturers do not disclose information of battery physical
parameters for battery users [104]. Using cell tear down and physical tests to measure
these parameters is notoriously difficult and costly. Importantly, specific subsets of
parameters such as health related or kinetic parameters gradually changed as batteries
getting degraded. Therefore, online estimating these parameters presents an important
advance in the field. The identified parameters can be used to update the battery
electrochemical model characteristics based on the real-time operating conditions.
Such these ‘adaptive’ battery electrochemical models can be used to develop robust
fault diagnosis, degradation models which are essential in health-consciously optimal
controls. Future works on identifying specific subsets of parameters, such as health
Chapter 7: Conclusions 121
related or kinetic parameters can be focussed and explored. Hyper-heuristics employed
Particle Swarm Optimization [105] and Simulated Annealing [106] are possible
approaches to identify cell parameters. The cost function of the parameter
identification algorithm is the error between the outputs of the electrochemical model
and the measurable cell input current, output voltage, temperature and impedance.
Using these real-time, easily measurable quantities is a feasible way to implement this
proposed identification algorithm in online applications.
Direction 2: Virtually parallel-like connection for serial-connected batteries
In a battery pack, lithium ion cells are connected in series and parallel to achieve
desired voltage and power levels. Cell-to-cell unbalancing can occur due to various
factors such as cell manufacturing and assembly process, parameter variation during
battery operation, and inhomogeneous aging stress factors [107]. Unbalanced serial-
connected cells in a string without appropriate control schemes can exacerbate further
imbalance and inhomogeneous degradation when the same current rate is applied to
every cell. A novel virtually parallel-like connection can be used to control the current
rate for each serial-connected cell in a string. In a virtually parallel-like connection of
series connected batteries, the series batteries are mathematically modelled as if they
are connected in a parallel connection. These simulated current rates are then used to
control the output currents of cell-level converters so that the lithium ion cells receive
the same current rates that they may receive if they are connected in parallel. By
virtually connecting these serial-connected cells in parallel, their ability to stay
balanced throughout the life of the battery pack can be enhanced. Therefore, the usable
capacity of the battery pack can be fully utilized. It is noted that the high computational
burden may be imposed when dealing with large-capacity battery packs in the
centralized control scheme of the traditional BMS in which a master controller is used
for sensing and computing all cells in the pack. However, this issue can be addressed
by using the decentralized control scheme that currently attracts interest in research
and industry [108, 109]. The impact of this decentralized control scheme is that the
cost may increase as additional Cell Management Units comprising of sensors and
microcontrollers for individual cells are added to the systems. However, benefits can
122 Chapter 7: Conclusions
be gained from the improvement the life time and safety of the battery pack. This trade-
off can be also investigated as part of the future work.
Direction 3: Apply Padé approximation model in fast charging applications
One of significant advantages of the reduced order model using Padé
approximation approach proposed in Chapter 4 is that variables can be evaluated at
discrete locations within the domain. This feature can be used in fast charging
applications since the dynamics of interested variables can be computed and monitored
with low computational burden without the need to compute all values of the variable
at all discrete locations. Future investigation on fast charging scheme can be carried
out to minimize the charging time while ensuring the potential voltage 0s e . This
approach is expected to reduce significantly charging time in comparison to traditional
CC-CV charge method whilst ensuring safe operation of the batteries. In this thesis,
the degradation is modelled by the SEI growth during charging. Fast discharging has
yet to be considered. One challenge in the fast discharging application will be the
higher computational requirement, as a more comprehensive degradation model needs
to be used. This challenge may be addressed in further research.
Direction 4: Optimally control large-format cylindrical cells
The proposed model in Chapter 6 can be used in optimal control scheme such as
optimally controlling of large format batteries in which the cell’s hot spots can be
observed. This is because the proposed model is able to provide information of the
radial distribution of cell temperature and degradation which is essential for such those
applications. By optimally controlling the battery based on the information of the cell
hot spots, the batteries life can be extended and safety of the battery during its
operation can be ensured. Offline trajectory optimization reported in [102] and online
nonlinear control algorithms such as Nonlinear Model Predictive Control reported in
[77, 103] can be possible approaches. Furthermore, the implementation structure of
the model in form of coupled sub-systems makes it feasible to add additional
mechanism such as mechanical stress on solid particles which can also be developed
in this future work.
Bibliography 123
Bibliography
[1] C. Patsios et al., "An integrated approach for the analysis and control of grid
connected energy storage systems," Journal of Energy Storage, vol. 5, pp. 48-
61, 2//.
[2] A. E. M. Operator, "Initial operation of the Hornsdale Power Reserve Battery
Energy Storage System," 2018.
[3] A. Fotouhi, D. J. Auger, K. Propp, S. Longo, and M. Wild, "A review on
electric vehicle battery modelling: From Lithium-ion toward Lithium–
Sulphur," Renewable and Sustainable Energy Reviews, vol. 56, pp. 1008-1021,
Apr. 2016.
[4] N. A. Chaturvedi, R. Klein, J. Christensen, J. Ahmed, and A. Kojic,
"Algorithms for Advanced Battery-Management Systems," IEEE Control
Systems, vol. 30, no. 3, pp. 49-68, 2010.
[5] A123-Systems, Nanophosphate® High Power Lithium Ion Cell
ANR26650M1-B datasheet, 2012.
[6] M. A. Roscher, R. M. Kuhn, and H. Döring, "Error detection for PHEV, BEV
and stationary battery systems," Control Engineering Practice, vol. 21, no. 11,
pp. 1481-1487, 11/ 2013.
[7] J. L. Lee, A. Chemistruck, and G. L. Plett, "One-dimensional physics-based
reduced-order model of lithium-ion dynamics," Journal of Power Sources, vol.
220, pp. 430-448, 2012/12/15/.
[8] A. M. Bizeray, S. Zhao, S. R. Duncan, and D. A. Howey, "Lithium-ion battery
thermal-electrochemical model-based state estimation using orthogonal
collocation and a modified extended Kalman filter," Journal of Power Sources,
vol. 296, pp. 400-412, 11/20/ 2015.
[9] M. Doyle, T. F. Fuller, and J. Newman, "Modeling of galvanostatic charge and
discharge of the lithium/polymer/insertion cell," Journal of the
Electrochemical Society, vol. 140, no. 6, pp. 1526-1533, 1993.
[10] T. F. Fuller, M. Doyle, and J. Newman, "Simulation and optimization of the
dual lithium ion insertion cell," Journal of the Electrochemical Society, vol.
141, no. 1, pp. 1-10, 1994.
[11] COMSOL Multiphysics®, URL: https://www.comsol.com/comsol-
multiphysics, 2018.
[12] UDDS, EPA Urban Dynamometer Driving Schedule, URL:
https://www.epa.gov/emission-standards-reference-guide/epa-urban-
dynamometer-driving-schedule-udds, 2019.
[13] T. M. Bandhauer, S. Garimella, and T. F. Fuller, "A critical review of thermal
issues in lithium-ion batteries," Journal of the Electrochemical Society, vol.
158, no. 3, pp. R1-R25, 2011.
[14] A. Wang, S. Kadam, H. Li, S. Shi, and Y. Qi, "Review on modeling of the
anode solid electrolyte interphase (SEI) for lithium-ion batteries," npj
Computational Materials, vol. 4, no. 1, p. 15, 2018/03/26.
124 Bibliography
[15] T. R. Tanim and C. D. Rahn, "Aging formula for lithium ion batteries with
solid electrolyte interphase layer growth," Journal of Power Sources, vol. 294,
pp. 239-247, 2015/10/30/.
[16] K. Smith, G. Kim, and A. Pesaran, "Modeling of Nonuniform Degradation in
Large-Format Li-ion Batteries (Poster)," National Renewable Energy
Laboratory (NREL), Golden, CO.2009.
[17] M. A. Roscher, O. S. Bohlen, and D. U. Sauer, "Reliable State Estimation of
Multicell Lithium-Ion Battery Systems," IEEE Transactions on Energy
Conversion, vol. 26, no. 3, pp. 737-743, 2011.
[18] W. Waag, S. Käbitz, and D. U. Sauer, "Experimental investigation of the
lithium-ion battery impedance characteristic at various conditions and aging
states and its influence on the application," Applied Energy, vol. 102, pp. 885-
897, 2// 2013.
[19] S. Buller, M. Thele, E. Karden, and R. W. De Doncker, "Impedance-based non-
linear dynamic battery modeling for automotive applications," Journal of
Power Sources, vol. 113, no. 2, pp. 422-430, 1/27/ 2003.
[20] S. Buller, M. Thele, R. W. A. A. D. Doncker, and E. Karden, "Impedance-
based simulation models of supercapacitors and Li-ion batteries for power
electronic applications," IEEE Transactions on Industry Applications, vol. 41,
no. 3, pp. 742-747, 2005.
[21] A. Farmann and D. U. Sauer, "Comparative study of reduced order equivalent
circuit models for on-board state-of-available-power prediction of lithium-ion
batteries in electric vehicles," Applied Energy, vol. 225, pp. 1102-1122,
2018/09/01/.
[22] A. Farmann and D. U. Sauer, "A comprehensive review of on-board State-of-
Available-Power prediction techniques for lithium-ion batteries in electric
vehicles," Journal of Power Sources, vol. 329, pp. 123-137, 2016/10/15/ 2016.
[23] X. Hu, S. Li, and H. Peng, "A comparative study of equivalent circuit models
for Li-ion batteries," Journal of Power Sources, vol. 198, pp. 359-367,
2012/01/15/ 2012.
[24] C. Fleischer, W. Waag, H.-M. Heyn, and D. U. Sauer, "On-line adaptive
battery impedance parameter and state estimation considering physical
principles in reduced order equivalent circuit battery models part 2. Parameter
and state estimation," Journal of Power Sources, vol. 262, pp. 457-482, 9/15/
2014.
[25] V.-H. Duong, H. A. Bastawrous, K. Lim, K. W. See, P. Zhang, and S. X. Dou,
"Online state of charge and model parameters estimation of the LiFePO4
battery in electric vehicles using multiple adaptive forgetting factors recursive
least-squares," Journal of Power Sources, vol. 296, pp. 215-224, 2015/11/20/.
[26] G. L. Plett, "Extended Kalman filtering for battery management systems of
LiPB-based HEV battery packs: Part 3. State and parameter estimation,"
Journal of Power Sources, vol. 134, no. 2, pp. 277-292, 8/12/ 2004.
[27] K. Smith and C.-Y. Wang, "Solid-state diffusion limitations on pulse operation
of a lithium ion cell for hybrid electric vehicles," Journal of Power Sources,
vol. 161, no. 1, pp. 628-639, 10/20/ 2006.
[28] R. Klein, N. A. Chaturvedi, J. Christensen, J. Ahmed, R. Findeisen, and A.
Kojic, "Electrochemical model based observer design for a lithium-ion
Bibliography 125
battery," IEEE Transactions on Control Systems Technology, vol. 21, no. 2,
pp. 289-301, 2013.
[29] V. R. Subramanian, V. D. Diwakar, and D. Tapriyal, "Efficient macro-micro
scale coupled modeling of batteries," Journal of the Electrochemical Society,
vol. 152, no. 10, pp. A2002-A2008, 2005.
[30] T. W. Farrell, C. P. Please, D. McElwain, and D. Swinkels, "Primary Alkaline
Battery Cathodes A Three‐Scale Model," Journal of the Electrochemical
Society, vol. 147, no. 11, pp. 4034-4044, 2000.
[31] S. Dargaville and T. W. Farrell, "Predicting active material utilization in
LiFePO4 electrodes using a multiscale mathematical model," Journal of the
Electrochemical Society, vol. 157, no. 7, pp. A830-A840, 2010.
[32] A. Farmann, W. Waag, A. Marongiu, and D. U. Sauer, "Critical review of on-
board capacity estimation techniques for lithium-ion batteries in electric and
hybrid electric vehicles," Journal of Power Sources, vol. 281, pp. 114-130,
2015/05/01/ 2015.
[33] A. Bizeray, S. Duncan, and D. Howey, "Advanced battery management
systems using fast electrochemical modelling," in Hybrid and Electric Vehicles
Conference 2013 (HEVC 2013), IET, 2013, pp. 1-6.
[34] A. M. Bizeray, S. Zhao, S. R. Duncan, and D. A. Howey, "Lithium-ion battery
thermal-electrochemical model-based state estimation using orthogonal
collocation and a modified extended Kalman filter," Journal of Power Sources,
vol. 296, pp. 400-412, 2015/11/20/.
[35] J. L. Lee, A. Chemistruck, and G. L. Plett, "Discrete-time realization of
transcendental impedance models, with application to modeling spherical solid
diffusion," Journal of Power Sources, vol. 206, pp. 367-377, 2012/05/15/ 2012.
[36] Y. Shi, G. Prasad, Z. Shen, and C. D. Rahn, "Discretization methods for battery
systems modeling," in Proceedings of the 2011 American Control Conference,
2011, pp. 356-361.
[37] J. C. Forman, S. Bashash, J. Stein, and H. Fathy, "Reduction of an
Electrochemistry-Based Li-Ion Battery Health Degradation Model via
Constraint Linearization and Padé Approximation," no. 44182, pp. 173-183,
2010.
[38] G. K. Prasad and C. D. Rahn, "Model based identification of aging parameters
in lithium ion batteries," Journal of Power Sources, vol. 232, pp. 79-85,
2013/06/15/.
[39] J. Marcicki, M. Canova, A. T. Conlisk, and G. Rizzoni, "Design and
parametrization analysis of a reduced-order electrochemical model of
graphite/LiFePO4 cells for SOC/SOH estimation," Journal of Power Sources,
vol. 237, pp. 310-324, 2013/09/01/.
[40] X. Zhang, J. Lu, S. Yuan, J. Yang, and X. Zhou, "A novel method for
identification of lithium-ion battery equivalent circuit model parameters
considering electrochemical properties," Journal of Power Sources, vol. 345,
pp. 21-29, 2017/03/31/.
[41] S. Yuan, L. Jiang, C. Yin, H. Wu, and X. Zhang, "A transfer function type of
simplified electrochemical model with modified boundary conditions and Padé
approximation for Li-ion battery: Part 2. Modeling and parameter estimation,"
Journal of Power Sources, vol. 352, pp. 258-271, 2017/06/01/.
126 Bibliography
[42] S. Yuan, L. Jiang, C. Yin, H. Wu, and X. Zhang, "A transfer function type of
simplified electrochemical model with modified boundary conditions and Padé
approximation for Li-ion battery: Part 1. lithium concentration estimation,"
Journal of Power Sources, vol. 352, pp. 245-257, 2017/06/01/ 2017.
[43] J. Newman, "Optimization of Porosity and Thickness of a Battery Electrode by
Means of a Reaction‐Zone Model," Journal of The Electrochemical Society,
vol. 142, no. 1, pp. 97-101, January 1, 1995 1995.
[44] T. W. Farrell, D. L. S. Mcelwain, and D. A. J. Swinkels, "A finite difference
equivalent circuit approach to secondary current modelling in annular porous
electrodes," Journal of Applied Mathematics and Decision Sciences, vol. 5, no.
2, pp. 119-132, 2001.
[45] S. Raël and M. Hinaje, "Using electrical analogy to describe mass and charge
transport in lithium-ion batteries," Journal of Power Sources, vol. 222, pp. 112-
122, 2013/01/15/.
[46] M.-T. von Srbik, M. Marinescu, R. F. Martinez-Botas, and G. J. Offer, "A
physically meaningful equivalent circuit network model of a lithium-ion
battery accounting for local electrochemical and thermal behaviour, variable
double layer capacitance and degradation," Journal of Power Sources, vol.
325, pp. 171-184, 2016/09/01/.
[47] Y. Merla, B. Wu, V. Yufit, R. F. Martinez-Botas, and G. J. Offer, "An easy-to-
parameterise physics-informed battery model and its application towards
lithium-ion battery cell design, diagnosis, and degradation," Journal of Power
Sources, vol. 384, pp. 66-79, 2018/04/30/.
[48] N. Baba, H. Yoshida, M. Nagaoka, C. Okuda, and S. Kawauchi, "Numerical
simulation of thermal behavior of lithium-ion secondary batteries using the
enhanced single particle model," Journal of Power Sources, vol. 252, pp. 214-
228, 4/15/ 2014.
[49] Y. Inui, Y. Kobayashi, Y. Watanabe, Y. Watase, and Y. Kitamura, "Simulation
of temperature distribution in cylindrical and prismatic lithium ion secondary
batteries," Energy Conversion and Management, vol. 48, no. 7, pp. 2103-2109,
7// 2007.
[50] M. Guo and R. E. White, "Mathematical model for a spirally-wound lithium-
ion cell," Journal of Power Sources, vol. 250, pp. 220-235, 3/15/ 2014.
[51] C. R. Pals and J. Newman, "Thermal modeling of the lithium/polymer battery
I. Discharge behavior of a single cell," Journal of the Electrochemical Society,
vol. 142, no. 10, pp. 3274-3281, 1995.
[52] C. R. Pals and J. Newman, "Thermal modeling of the lithium/polymer battery
II. Temperature profiles in a cell stack," Journal of the Electrochemical
Society, vol. 142, no. 10, pp. 3282-3288, 1995.
[53] T. Wang, K. J. Tseng, J. Zhao, and Z. Wei, "Thermal investigation of lithium-
ion battery module with different cell arrangement structures and forced air-
cooling strategies," Applied Energy, vol. 134, pp. 229-238, 12/1/ 2014.
[54] T. Yang, N. Yang, X. Zhang, and G. Li, "Investigation of the thermal
performance of axial-flow air cooling for the lithium-ion battery pack,"
International Journal of Thermal Sciences, vol. 108, pp. 132-144, 10// 2016.
[55] S. Basu, K. S. Hariharan, S. M. Kolake, T. Song, D. K. Sohn, and T. Yeo,
"Coupled electrochemical thermal modelling of a novel Li-ion battery pack
thermal management system," Applied Energy, vol. 181, pp. 1-13, 11/1/ 2016.
Bibliography 127
[56] P. Amiribavandpour, W. Shen, D. Mu, and A. Kapoor, "An improved
theoretical electrochemical-thermal modelling of lithium-ion battery packs in
electric vehicles," Journal of Power Sources, vol. 284, pp. 328-338, 6/15/
2015.
[57] COMSOL Multiphysic, URL: https://www.comsol.com/comsol-multiphysics,
no. May 15, 2019.
[58] ANSYS Fluent, URL: https://www.ansys.com/products/fluids/ansys-fluent, no.
May 15, 2019.
[59] N. T. Tran, M. Vilathgamuwa, T. Farrell, S. S. Choi, Y. Li, and J. Teague, "A
Padé Approximate Model of Lithium Ion Batteries," Journal of The
Electrochemical Society, vol. 165, no. 7, pp. A1409-A1421, January 1, 2018.
[60] S.-C. Chen, Y.-Y. Wang, and C.-C. Wan, "Thermal analysis of spirally wound
lithium batteries," Journal of the Electrochemical Society, vol. 153, no. 4, pp.
A637-A648, 2006.
[61] K.-J. Lee, K. Smith, A. Pesaran, and G.-H. Kim, "Three dimensional thermal-
, electrical-, and electrochemical-coupled model for cylindrical wound large
format lithium-ion batteries," Journal of Power Sources, vol. 241, pp. 20-32,
2013/11/01/.
[62] Y. Kim, J. B. Siegel, and A. G. Stefanopoulou, "A computationally efficient
thermal model of cylindrical battery cells for the estimation of radially
distributed temperatures," in 2013 American Control Conference, 2013, pp.
698-703.
[63] C. R. Birkl, M. R. Roberts, E. McTurk, P. G. Bruce, and D. A. Howey,
"Degradation diagnostics for lithium ion cells," Journal of Power Sources, vol.
341, pp. 373-386, 2017.
[64] Y. Li, M. Vilathgamuwa, S. S. Choi, T. W. Farrell, N. T. Tran, and J. Teague,
"Development of a degradation-conscious physics-based lithium-ion battery
model for use in power system planning studies," Applied Energy, vol. 248, pp.
512-525, 2019/08/15/.
[65] M. Winter, "The Solid Electrolyte Interphase – The Most Important and the
Least Understood Solid Electrolyte in Rechargeable Li Batteries," in Zeitschrift
für Physikalische Chemie vol. 223, ed, 2009, p. 1395.
[66] P. Ramadass, B. Haran, P. M. Gomadam, R. White, and B. N. Popov,
"Development of First Principles Capacity Fade Model for Li-Ion Cells,"
Journal of The Electrochemical Society, vol. 151, no. 2, pp. A196-A203,
February 1, 2004.
[67] M. Safari, M. Morcrette, A. Teyssot, and C. Delacourt, "Multimodal Physics-
Based Aging Model for Life Prediction of Li-Ion Batteries," Journal of The
Electrochemical Society, vol. 156, no. 3, pp. A145-A153, March 1, 2009 2009.
[68] H. Ekström and G. Lindbergh, "A Model for Predicting Capacity Fade due to
SEI Formation in a Commercial Graphite/LiFePO4 Cell," Journal of The
Electrochemical Society, vol. 162, no. 6, pp. A1003-A1007, January 1, 2015
2015.
[69] K. S. Ng, C.-S. Moo, Y.-P. Chen, and Y.-C. Hsieh, "Enhanced coulomb
counting method for estimating state-of-charge and state-of-health of lithium-
ion batteries," Applied Energy, vol. 86, no. 9, pp. 1506-1511, Sep. 2009.
128 Bibliography
[70] W. Waag, C. Fleischer, and D. U. Sauer, "Critical review of the methods for
monitoring of lithium-ion batteries in electric and hybrid vehicles," Journal of
Power Sources, vol. 258, pp. 321-339, 2014/07/15/ 2014.
[71] K. A. Smith, C. D. Rahn, and C. Y. Wang, "Model-Based Electrochemical
Estimation and Constraint Management for Pulse Operation of Lithium Ion
Batteries," IEEE Transactions on Control Systems Technology, vol. 18, no. 3,
pp. 654-663, 2010.
[72] W. Weizhong et al., "Comparison of Kalman Filter-based state of charge
estimation strategies for Li-Ion batteries," in 2016 IEEE Transportation
Electrification Conference and Expo (ITEC), 2016, pp. 1-6.
[73] G. L. Plett, "Sigma-point Kalman filtering for battery management systems of
LiPB-based HEV battery packs: Part 1: Introduction and state estimation,"
Journal of Power Sources, vol. 161, no. 2, pp. 1356-1368, Oct. 2006.
[74] G. L. Plett, "Extended Kalman filtering for battery management systems of
LiPB-based HEV battery packs: Part 1. Background," Journal of Power
Sources, vol. 134, no. 2, pp. 252-261, 2004/08/12/ 2004.
[75] S. Khaleghi Rahimian, S. Rayman, and R. E. White, "Extension of physics-
based single particle model for higher charge–discharge rates," Journal of
Power Sources, vol. 224, pp. 180-194, 2013/02/15/.
[76] X. Han, M. Ouyang, L. Lu, and J. Li, "Simplification of physics-based
electrochemical model for lithium ion battery on electric vehicle. Part I:
Diffusion simplification and single particle model," Journal of Power Sources,
vol. 278, pp. 802-813, Mar. 2015.
[77] C. Sun, F. Sun, and S. J. Moura, "Nonlinear predictive energy management of
residential buildings with photovoltaics & batteries," Journal of Power
Sources, vol. 325, pp. 723-731, 2016/09/01/.
[78] E. Prada, D. Di Domenico, Y. Creff, J. Bernard, V. Sauvant-Moynot, and F.
Huet, "Simplified electrochemical and thermal model of LiFePO4-graphite Li-
ion batteries for fast charge applications," Journal of the Electrochemical
Society, vol. 159, no. 9, pp. A1508-A1519, 2012.
[79] N. T. Tran, M. Vilathgamuwa, T. Farrell, and S. S. Choi, "Matlab simulation
of lithium ion cell using electrochemical single particle model," in Power
Electronics Conference (SPEC), IEEE Annual Southern, 2016, pp. 1-6: IEEE.
[80] M. Safari and C. Delacourt, "Mathematical modeling of lithium iron phosphate
electrode: galvanostatic charge/discharge and path dependence," Journal of the
Electrochemical Society, vol. 158, no. 2, pp. A63-A73, 2011.
[81] M. Mastali Majdabadi, S. Farhad, M. Farkhondeh, R. A. Fraser, and M. Fowler,
"Simplified electrochemical multi-particle model for LiFePO4 cathodes in
lithium-ion batteries," Journal of Power Sources, vol. 275, pp. 633-643, Feb.
2015.
[82] K. Gopalakrishnan, T. Zhang, and G. J. Offer, "A Fast, Memory-Efficient
Discrete-Time Realization Algorithm for Reduced-Order Li-Ion Battery
Models," Journal of Electrochemical Energy Conversion and Storage, vol. 14,
no. 1, pp. 011001-011008, 2017.
[83] A. Rodríguez, G. L. Plett, and M. S. Trimboli, "Fast computation of the
electrolyte-concentration transfer function of a lithium-ion cell model,"
Journal of Power Sources, vol. 360, pp. 642-645, 2017/08/31/.
Bibliography 129
[84] J. Kallrath, "An Overview On Rational Function Techniques and Pade
Approximants," ed: Report, Ludwigshafen, Germany, 2002.
[85] Wolfram Mathematica, URL: https://www.wolfram.com/mathematica/, 2018.
[86] Y. Li, S. S. Choi, D. M. Vilathgamuwa, and D. L. Yao, "An improved
dispatchable wind turbine generator and dual-battery energy storage system to
reduce battery capacity requirement," in 2016 IEEE 2nd Annual Southern
Power Electronics Conference (SPEC), 2016, pp. 1-6.
[87] M. Torchio, L. Magni, R. B. Gopaluni, R. D. Braatz, and D. M. Raimondo,
"LIONSIMBA: A Matlab Framework Based on a Finite Volume Model
Suitable for Li-Ion Battery Design, Simulation, and Control," Journal of The
Electrochemical Society, vol. 163, no. 7, pp. A1192-A1205, 2016.
[88] J. W. Yang, "Secondary battery charging method and device," ed: Google
Patents, 2012.
[89] N. A. Chaturvedi, R. Klein, J. Christensen, J. Ahmed, and A. Kojic,
"Algorithms for Advanced Battery-Management Systems," IEEE Control
Systems Magazine, vol. 30, no. 3, pp. 49-68, 2010.
[90] K. Smith and C.-Y. Wang, "Power and thermal characterization of a lithium-
ion battery pack for hybrid-electric vehicles," Journal of Power Sources, vol.
160, no. 1, pp. 662-673, 2006/09/29/.
[91] K. Kumaresan, G. Sikha, and R. E. White, "Thermal model for a Li-ion cell,"
Journal of the Electrochemical Society, vol. 155, no. 2, pp. A164-A171, 2008.
[92] C. Forgez, D. Vinh Do, G. Friedrich, M. Morcrette, and C. Delacourt, "Thermal
modeling of a cylindrical LiFePO4/graphite lithium-ion battery," Journal of
Power Sources, vol. 195, no. 9, pp. 2961-2968, 2010/05/01/ 2010.
[93] W. Fang, O. J. Kwon, and C.-Y. Wang, "Electrochemical–thermal modeling of
automotive Li-ion batteries and experimental validation using a three-electrode
cell," International Journal of Energy Research, vol. 34, no. 2, pp. 107-115,
2010.
[94] K.-J. Lee, K. Smith, A. Pesaran, and G.-H. Kim, "Three dimensional thermal-
, electrical-, and electrochemical-coupled model for cylindrical wound large
format lithium-ion batteries," Journal of Power Sources, vol. 241, pp. 20-32,
2013/11/01/ 2013.
[95] P. M. Gomadam, R. E. White, and J. W. Weidner, "Modeling heat conduction
in spiral geometries," Journal of The Electrochemical Society, vol. 150, no. 10,
pp. A1339-A1345, 2003.
[96] M. Torchio et al., "Real-time model predictive control for the optimal charging
of a lithium-ion battery," in 2015 American Control Conference (ACC), 2015,
pp. 4536-4541.
[97] K. Smith and C.-Y. Wang, "Solid-state diffusion limitations on pulse operation
of a lithium ion cell for hybrid electric vehicles," Journal of Power Sources,
vol. 161, no. 1, pp. 628-639, 2006/10/20/.
[98] Theodore L Bergman, Adrienne S Lavine, Frank P Incropera, and D. P.
DeWitt, Fundamentals of Heat and Mass Transfer. John Wiley and Sons Ltd,
2011.
[99] D. W. Trim, Applied partial differential equations. PWS-Kent Pub. Co., 1990.
[100] MATLAB/Simulink, URL: https://au.mathworks.com/?s_tid=gn_logo, no.
May 15, 2019, 2019.
130 Bibliography
[101] N. T. Tran, M. Vilathgamuwa, T. Farrell, S. S. Choi, Y. Li, and J. Teague, "A
Computationally-Efficient Electrochemical-Thermal Model for Small-Format
Cylindrical Lithium Ion Batteries," in 2018 IEEE 4th Southern Power
Electronics Conference (SPEC), 2018, pp. 1-7.
[102] S. J. Moura, J. C. Forman, S. Bashash, J. L. Stein, and H. K. Fathy, "Optimal
Control of Film Growth in Lithium-Ion Battery Packs via Relay Switches,"
IEEE Transactions on Industrial Electronics, vol. 58, no. 8, pp. 3555-3566,
2011.
[103] S. Lucia, M. Torchio, D. M. Raimondo, R. Klein, R. D. Braatz, and R.
Findeisen, "Towards adaptive health-aware charging of Li-ion batteries: A
real-time predictive control approach using first-principles models," in 2017
American Control Conference (ACC), 2017, pp. 4717-4722.
[104] S. Park, D. Kato, Z. Gima, R. Klein, and S. Moura, "Optimal Input Design for
Parameter Identification in an Electrochemical Li-ion Battery Model," in 2018
Annual American Control Conference (ACC), 2018, pp. 2300-2305.
[105] J. Kennedy and R. Eberhart, "Particle swarm optimization," in Proceedings of
ICNN'95 - International Conference on Neural Networks, 1995, vol. 4, pp.
1942-1948 vol.4.
[106] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, "Optimization by Simulated
Annealing," Science, 10.1126/science.220.4598.671 vol. 220, no. 4598, p. 671,
1983.
[107] M. Baumann, L. Wildfeuer, S. Rohr, and M. Lienkamp, "Parameter variations
within Li-Ion battery packs – Theoretical investigations and experimental
quantification," Journal of Energy Storage, vol. 18, pp. 295-307, 2018/08/01/.
[108] S. Steinhorst, M. Lukasiewycz, S. Narayanaswamy, M. Kauer, and S.
Chakraborty, "Smart Cells for Embedded Battery Management," in 2014 IEEE
International Conference on Cyber-Physical Systems, Networks, and
Applications, 2014, pp. 59-64.
[109] D. F. Frost and D. A. Howey, "Completely Decentralized Active Balancing
Battery Management System," IEEE Transactions on Power Electronics, vol.
33, no. 1, pp. 729-738, 2018.
Appendices 131
Appendices
Appendix A
The electrolyte concentration transfer functions
The transfer functions used to compute the electrolyte concentration is given as
[7],
* 0 * * neg neg neg negneg1 neg neg eff effk
2 2* neg neg neg neg
appneg eff eff
20 neg neg * neg
1 eff eff neg
2 2* neg neg neg
neg eff eff
1 sin cosh
sinh
1 cos
k t L L s sJ s
I s FA L s s
k t L s
FA L s
, (7.1)
and,
* 0 * * pos pos pos pospos6 pos tot eff effk
2 2* pos pos pos pos
apppos eff eff
0 * * pos neg pos pos
5 pos nm eff eff
2 2* pos pos pos pos
pos eff eff
0
6
1 cos cosh
( ) sinh
1 sin cosh
sinh
1
k t L L s sJ s
I s FA L s s
k t L L s s
FA L s s
k t L
* * pos neg pos pos
pos nm eff eff
2 2* pos pos pos pos
pos eff eff
0 * * pos pos pos pos
5 pos tot eff eff
2 2* pos pos pos pos
pos eff eff
0 pos * pos
5 eff tot eff
cos cosh
sinh
1 sin cosh
sinh
1 cos c
L s s
FA L s s
k t L L s s
FA L s s
k t L
2* pos
nm
2 2* pos pos pos
pos eff eff
20 pos * pos * pos
6 eff tot eff nm
2 2* pos pos pos
pos eff eff
os
1 sin sin
L s
FA L s
k t L L s
FA L s
, (7.2)
where, *
neg sep /n k nL L D , *
pos pos /p k pL L D , *
nm neg sep /p k pL L L D ,
*
tot tot /p k pL L D .
For the electrolyte potential in the negative electrode ( neg0 x L ):
132 Appendices
negnegneg neg negeff
neg neg nege eff1
neg neg neg negapp eff eff
neg neg
neg neg
neg
neg neg neg neg
eff eff
1 cosh sinh,
sinh
cosh cosh
sinh
x sL x s s
x s L
I s A s s
L x sL s
L
A s s
, (7.3)
For the electrolyte potential in the separator ( neg neg sepL x L L ):
negnegneg negeff
sep negsepe eff1
sep neg neg negapp eff eff eff
1 tanh, 2
sL s
x s L x
I s A A s
, (7.4)
For the electrolyte potential in the positive electrode ( neg sep totL L x L ):
negnegneg negeff
pos negsepe eff1
sep neg neg negapp eff eff eff
negpos negeff
neg
eff
neg neg neg neg
eff eff
neg seppos
po
1 tanh, 2
1 cosh
sinh
cosh
sL s
x s L
I s A A s
L s
A s s
L L xL
L
neg
s
neg neg neg neg
eff eff
neg totpos negeff
neg pos
eff
neg neg neg neg
eff eff
neg sep
neg neg
eff eff
sinh
cosh
sinh
s
A s s
L xL s
L
A s s
L L x
A
. (7.5)
Appendices 133
Appendix B
List of symbols
iL Thickness of the layer (m)
sR Particle radius (m)
sa Specific interfacial area ( -1m )
maxs,c Maximum concentration of lithium in the solid phase ( -3mmol )
s,0c Initial concentration of lithium in the solid phase ( -3mmol )
e,0c Steady-state concentration of lithium in the electrolyte phase ( -3mmol )
es,c Surface concentration of lithium in a spherical electrode particle ( -3mmol )
effeD Effective electrolyte diffusivity ( -12 sm )
sD Solid diffusivity ( -12 sm )
f Mean molar activity coefficient
filmR Solid electrolyte interphase layer film resistance ( -1m )
R Universal gas constant (8.314 -1Jmol )
0t Transference number for the anion
sD Solid phase lithium diffusion coefficient ( 12sm )
eD Electrolyte phase lithum ion diffusion coefficient ( 12sm )
eQ Total amount of lithium ion in each domain ( m ol )
pr Radial coordinate across the solid particle radius (m)
effr Reaction rate constant ( 5.1-11-2 smmolsmmol )
A Electrode plate area ( 2m )
T Temperature (K)
F Faraday’s constant (96,478 -1Cmol )
s , e Active material, electrolyte phase volume fraction
Solid phase conductivity ( -1Sm )
Electrolyte phase ionic conductivity ( -1Sm )
134 Appendices
eff Effective electrolyte conductivity ( -1Sm )
a , c Charge transfer coefficients
s s,maxc c Stoichiometry of electrode
min,i Stoichiometry of electrode at 0% state of charge
max,i Stoichiometry of electrode at 100% state of charge
pos/neg/sep Pertaining to the positive/negative electrode/separator
N Number of winds
M Number of layers in a wind
The material density ( 3kg m )
pc The specific heat ( J kgK )
The thermal conductivity [ W mK]