You are on page 1of 15

Multi-Domain Modeling of Lithium-Ion Batteries Encompassing

Multi-Physics in Varied Length Scales


Gi-Heon Kim,
*
,z
Kandler Smith,
*
Kyu-Jin Lee,
*
Shriram Santhanagopalan,
*
and Ahmad Pesaran
National Renewable Energy Laboratory, Golden, Colorado 80401, USA
This paper presents a general multi-scale multi-physics lithium-ion battery model framework, the Multi-Scale Multi-Dimensional
model. The model introduces multiple coupled computational domains to resolve the interplay of lithium-ion battery physics in var-
ied length scales. Model geometry decoupling and domain separation for the physicochemical process interplay are valid where the
characteristic time or length scale is segregated. Assuming statistical homogeneity for repeated architectures typical of lithium-ion
battery devices is often adequate and effective for modeling submodel geometries and physics in each domain. The modularized
hierarchical architecture of the model provides a exible and expandable framework facilitating modeling of the multiphysics behav-
ior of lithium-ion battery systems. In this paper, the Multi-Scale Multi-Dimensional model is introduced and applied to a model anal-
ysis that resolves electrochemical-, electrical-, and thermal-coupled physics in large-format stacked prismatic cell designs.
VC
2011 The Electrochemical Society. [DOI: 10.1149/1.3597614] All rights reserved.
Manuscript submitted March 17, 2011; revised manuscript received May 10, 2011. Published June 17, 2011.
Lithium-ion batteries (LIBs) have been widely accepted and
used in consumer applications such as laptop computers and perso-
nal communication and entertainment devices because of their high
power and energy density. LIBs are also getting increasing attention
for electrical energy storage in electric drive vehicles (EDVs). LIBs
for EDVs are much larger in capacity and physical size than those
used in consumer applications. However, scale-up of the batteries
raises the complexity of physical phenomena that do not play a sig-
nicant role in small battery systems. Since the physical phenomena
occur over a wide range of time and length scales from atomic var-
iations to heat transfer over an entire device, it is important to under-
stand how these different mechanisms are related. For LIB cell
manufacturers, the lack of understanding of how macroscopic
design features impact local microscopic electrochemical processes
has been one of the major obstacles in scaling up their consumer
electronics cell technologies to build large automotive batteries. For
automotive companies and system integrators, the lack of a battery
model to use in their established computer-aided engineering design
processes in order to predict thermal, electrical, electrochemical,
and mechanical response of the battery under various system opera-
tion strategies and management system designs is recognized as an
urgent barrier to overcome for expediting EDV development and
production.
Model-based investigations promote theoretical understanding of
battery physics beyond what is possible from experiments only.
Modeling of a system where the response is critically affected by
interaction between the physics at varied scales, however, is chal-
lenging in terms of computational cost. In the early 1990s, Newman
and his colleagues suggested a LIB model utilizing porous electrode
theory.
1
The model solves lithium diffusion dynamics and charge
transfer kinetics to predict the electrical response of a cell in a
paired intercalation electrode system. This model has been widely
used in academia and industry to describe the performance of a LIB
based on material properties and electrode design.
Newmans approach toward paired intercalation composite elec-
trodes is fairly adequate for prediction of small battery behavior. In
large-format cells, however, non-uniformity of the electric potential
along the current collectors in cell composites and the temperature
throughout the cell volume become severe enough to impact battery
responses. The same is true for large battery packs consisting of
multiple cells. Potential and temperature imbalances cause certain
locations in a cell to be cycled more than the rest of the cell. The
local excess use generates more heat and stress, causing severe deg-
radation in the performance and life of the cell. Therefore, for a bet-
ter understanding of the behavior of a large LIB system, the trans-
port of electrical current and heat must be evaluated not only in the
composite electrode matrices at the length scale of electrode pair,
but also across the highly anisotropic cell composite medium at the
cell-dimension length scale.
Through the multi-year effort supported by U.S. Department of
Energy (DOE), the National Renewable Energy Laboratory (NREL)
has developed a modeling framework for predictive computer simu-
lation of LIBs known as the Multi-Scale Multi-Dimensional
(MSMD) model that addresses the interplay among the physics in
varied scales.
27
In this paper, we introduce NRELs MSMD model
and present an example model analysis that evaluates largeformat,
stacked, prismatic cell designs.
Multi-Domain Model Framework
It is computationally expensive to perform a predictive numeri-
cal simulation of battery performance, degeneration, and safety
response while capturing the interactive coupling among the differ-
ent physicochemical processes in varied characteristic length and
time scales in complex geometries using a single computational do-
main. The MSMD model achieves computational efciency for
resolving multi-physics interactions occurring over a wide range of
length scales by introducing separate solution domains, at the parti-
cle, electrode, and cell level. Figure 1 shows the conceptual diagram
of the multi-scale multi-domain approach used in the MSMD model.
Each domain uses its own independent coordinate system for spatial
discretization of the variables solved in that domain. Separation of
the model domain and adoption of the statistical homogeneity
assumption are enabled based on the intrinsic nature of typical LIB
systems where physics with signicant time-scale differences
interplay.
The MSMD model framework has a hierarchical structure. Solu-
tion variables dened in a lower hierarchy domain have ner spatial
resolution than those solved in a higher hierarchy domain. Conse-
quently, physical and chemical quantities of smaller length-scale
physics are evaluated with a ner spatial resolution to resolve the
impact of corresponding small-scale geometry. Larger-scale quanti-
ties are calculated with coarser spatial resolution, eliminating the
complications of the smaller-scale geometric features. In addition to
computational efciency, the MSMD approach provides a modular-
ized framework, enabling model exibility by allowing multiple
submodel options with arbitrary physical and computational com-
plexities. This model exibility comes from the fact that each level
submodel is independent from the choice of models and solver
schemes used in other domains as long as the model input from
other domains is properly transferred through the specied inter-do-
main information exchange. The modularized MSMD framework is
*
Electrochemical Society Active Member.
z
E-mail: gi-heon.kim@nrel.gov
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011)
0013-4651/2011/158(8)/A955/15/$28.00 VC The Electrochemical Society
A955
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
well suited for multi-physics investigations as one can easily add
new physics of interest or drop physics of insignicant phenomena.
Multiscale model for a lithium-ion battery. To model the per-
formance of LIBs, the MSMD model adapts the electrochemical and
thermal physics introduced by other researchers.
1,8,9
Charge transfer
kinetic reactions are solved at the electrode-electrolyte interface.
Lithium transport in active particles is modeled with a diffusion
mechanism. Lithium-ion migration and diffusion through a liquid
electrolyte are evaluated. Charge balances are resolved in the posi-
tive and negative solid matrices and in the liquid electrolyte, respec-
tively. Temperature is solved for in the overall cell and pack geome-
try. The model predicts the electrical and thermal behavior of a LIB
cell in operation for given electrical loads and thermal boundary
conditions. As mentioned earlier, the battery geometry is resolved
into three coupled domains: a) the particle domain, b) the electrode
domain, and c) the cell domain. In the following sections, the details
about the physics represented in each level domain and the inter-do-
main coupling of the solution variables for general modeling of a
LIB will be described.
Particle domain model. In conventional LIB designs, lithium-
hosting electrode materials are prepared as particulates with con-
trolled shape and size. Solid-phase lithium transport is one of the
rate-determining steps for overall LIB performance and is affected
by the particle geometry as well as thermodynamic and kinetic prop-
erties of the materials. Using a separate coordinate system for solid-
phase diffusion from that for other variables has been well practiced
with the so-called pseudo-2d model by Newman and others.
10,11
In
the MSMD model, charge transfer kinetics and solid-phase lithium
diffusion are solved for in the particle domain. The schematic in
Fig. 2a shows the solution variables and the input and the output for
the particle domain submodel. Field variables fed in from the
higher-level electrode submodel are treated as averaged lumped val-
ues in the particle domain. Those inputs from the electrode domain
submodel are liquid-phase concentration, c
e
, electrical potential at
the negative or positive electrode,
s,a
or
s,c
, and in the liquid
phase,
e
, and temperature, T. This approach is reasonable where
transport in the liquid phase is much faster than in the solid phase:
e.g., D
e
>> D
s
. Transfer current density at the particle surface, i
//
n
,
and the lithium concentration within a solid electrode particle, c
s
,
are the main solution variables in the particle domain. Surface heat
ux at the electrode-electrolyte interface, q
//
n
, including heat from
the charge transfer reaction, is a major heat source in this domain,
while other volumetric heat sources, q
///
n
, such as the heat of mix-
ing
12,13
within a particle, may also contribute to the total heat gener-
ation. Additional complexities of particle physics, such as mechani-
cal failure and surface kinetics, can be modeled in this domain as
necessary.
Electrode domain model. The active particles are typically
mixed with a conductive agent and a polymer binder, and then
coated on thin metal current collector sheets to form porous com-
posite electrodes with a good electrical network and mechanical in-
tegrity through the solid matrix. In a conventional electrode pair, a
thin polymeric porous separator is inserted in between the cathode
and anode electrode composites to prevent an electronic short
between the pair of electrodes. The pore structures of the composite
electrodes and separator, as well as wetting of the pore surfaces,
impact lithium ion transport through the liquid electrolyte. The elec-
trode domain submodel resolves the charge balance across the com-
posite electrode pair in the solid matrices of electrode composites
and in the liquid electrolyte and lithium-ion transport in the liquid
electrolyte. The schematic in Fig. 2b shows the solution variables
and the submodel input and output for the electrode domain submo-
del. Electrostatic potential along the current collectors, U

and U
-
,
and temperature, T, are input from the cell domain submodel and
treated as averaged lumped values in the electrode domain. Using
the lumped value of the electrode potentials at the interface with
current collectors without spatial dependence in the electrode do-
main coordinate system is reasonable because the current collector
sheets on the boundaries of both the positive and negative electrodes
are electrically much more conductive than the composite electrode
layers. For the same reason, an electrode domain submodel is often
represented as a one-dimensional problem with the in-plane statisti-
cal homogeneity assumption. The main solution variables in the
electrode domain are lithium concentration in the liquid electrolyte,
c
e
, and electric potentials in negative- and positive-biased solid mat-
rices and liquid electrolyte,
s,a
,
s,c
, and
e
. The volumetric
Figure 1. (Color online) Conceptual diagram for the multi-scale multi-do-
main approach of the MSMD model.
Figure 2. (Color online) Summary of the model solution variables in each
computational domain and the coupling variables exchanged between the ad-
jacent length scale domains.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A956
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
reaction current at the electrode composite volume, j
///
x
, is evaluated
with averaged charge transfer current density over the active particle
surfaces,

i
//
n
, calculated in a lower- level hierarchy particle domain
submodel. Current density at the electrode composite boundary at
the current collector interface, i
//
x
, is dened either as ux of charge
out from the electrode into the positive current collector or as the
inux of charge from the negative current collector. Volumetric
heat generation in the electrode domain, q
///
x
, includes joule heat due
to current ow in solid composite electrodes and liquid electrolyte,
as well as heat sources averaged at the lower hierarchy particle do-
main, q
//
n
, and q
///
n
. Heat generation due to the contact resistance
between electrode and metal current collector foil surface can be
considered a surface heat source term, q
//
x
, in this domain.
Cell domain model. Unit assemblies of paired electrode layers
are stacked or wound to build prismatic or cylindrical cell devices.
In this stage, major cell design issues are related to thermal and elec-
trical optimization of internal cell congurations. As the cell
capacity and size increase, the spatial non-uniformity of the temper-
ature and the electric potentials becomes sufciently large to cause
the performance and life of a battery to degenerate. Building a mod-
ule or a pack with multiple individual cells is another extension of
designing a thermal and electrical conguration for a larger battery
system, often involving external controls. The schematic in Fig. 2c
shows the solution variables and the submodel input and output for
the cell domain model. The spatial distribution of temperature, T,
and electric potential at the current collectors and other passive
components, U

and U
-
, in a system must be resolved with the sys-
tem geometry and the boundary conditions. Therefore, they are
solved in the cell domain to resolve macroscopic electrical and ther-
mal transport with a given electrical load and thermal ambient con-
ditions. On the other hand, the volumetric electric current source at
the positive current collector phase, j
///
H
, in the cell composite vol-
ume, which comprises composite electrodes, separator, electrolyte,
and current collector foils, is evaluated using the average electrode
plate current density,

i
//
x
, passed from the electrode domain submo-
del. The volumetric heat generation in a cell domain, q
///
H
, includes
joule heat for carrying the electrical current in the current collector
phase in cell composites and in other passive components of a cell,
in addition to the contribution from the heat sources passed from the
lower-hierarchy electrode domain, q
//
x
, and q
///
x
.
Inter-domain coupling. In the MSMD model framework, mul-
tiple physics models dened in the corresponding domains are
solved simultaneously through two-way inter-domain couplings, as
described in Fig. 2. In general, information from a higher hierarchy
domain to a lower hierarchy domain is delivered using eld varia-
bles. When upper hierarchy domain eld variables are given as input
to a lower hierarchy domain submodel, they are considered as aver-
aged lumped values in the lower hierarchy domain. In other words,
inputs from a higher hierarchy domain do not have spatial depend-
ence in the coordinate system of the lower hierarchy domain. On the
other hand, information delivery from a lower to a higher hierarchy
domain is done through source terms. The volumetric or surface
sources evaluated in a lower hierarchy domain are averaged over
that domains geometry, eliminating the coordinate system depend-
ence before they are passed to a higher hierarchy domain. The aver-
aged sources delivered to a higher hierarchy domain are then
converted to a volumetric source term in the higher hierarchy sub-
model. For example, the transfer current density, i
//
n
, is averaged
over the electrode-electrolyte interfaces in a particle domain

i
//
n
(x; H) =
_
A
n
i
//
n
(n; x; H)dA
n
A
n
[1]
where A
n
is electrode-electrolyte interface area of the particle do-
main. The average transfer current density,

i
//
n
, is delivered to an
electrode domain submodel and then converted to a volumetric
transfer reaction current at the electrode composite volume using
Eq. 2
j
///
x
(x; H) =

i
//
n
(x; H)a
s;x
[2]
where the specic area of active interface between the electrode par-
ticles and the electrolyte in the electrode composite volume, a
s,x
, is
an electrode domain parameter. Consequently, the current density at
the electrode composite boundary, i
//
x
, is averaged over the area of
interface with the current collector in the electrode domain

i
//
x
(H) =
_
Ax
i
//
x
(x; H)dA
x

A
x
[3]
where A
x
is the electrode composite current collector interface area
of the electrode domain geometry, and

A
x
is the electrode plate area.
They are identical in many model approaches. The average elec-
trode plate current density,

i
//
x
, is delivered to a cell domain, and
then converted to a volumetric current source at the current collector
phase of the cell composite jelly volume using Eq. 4
j
///
H
(H) =

i
//
x
(H)a
s;H
[4]
where the specic electrode plate area in the cell composite volume,
a
s,H
, is a cell domain parameter. The expressions shown in Eqs. 2
and 4 are useful to keep consistency and simplicity of formulation
for coupling quantities of the current source in each domain. The
coupling of the heat sources is also done in a similar manner. While
the actual reaction current source exists only at the electrode
electrolyte interfaces, heat release and absorption occur due to vari-
ous mechanisms
12,14,15
in all length-scales. Therefore, heat sources
should be properly modeled in each hierarchical submodel. Surface
heat sources at electrode-electrolyte interfaces, q
//
n
, and volumetric
heat sources within a particle volume, q
///
n
, are averaged over the par-
ticle domain geometries and delivered to the electrode domain
submodel
q
//
n
(x; H) =
_
A
n
q
//
n
(n; x; H)dA
n
A
n
[5]
q
///
n
(x; H) =
_
V
n
q
///
n
(n; x; H)dV
n
V
n
[6]
where A
n
and V
n
are the electrodeelectrolyte interface area and the
volume of the particle domain geometry, respectively. In the elec-
trode domain, a volumetric heat source that originates from the par-
ticle domain is evaluated using Eq. 7
q
///
x;n
(x; H) = q
//
n
(x; H)a
s;x
q
///
n
(x; H)e
s
[7]
where a
s,x
is the specic area of active interface between electrode
particles and electrolyte, and e
s
is the volume fraction of active par-
ticles in the composite electrode volume. The volumetric heat for
contribution from the particle domain heat sources evaluated with
Eq. 7 is added into the net electrode domain volumetric heat source
term with other heat sources, such as joule heat due to the electronic
current ow in the composite electrode matrix and the ionic current
ow in the electrolyte
q
///
x
(x; H) = q
///
x;n
(x; H)

k
q
///
x;k
(x; H) [8]
Surface and volume heat sources at the electrode domain, q
//
x
and,
q
///
x
, are averaged over the geometries of the electrode domain and
delivered to the cell domain
q
//
x
(H) =
_
Ax
q
//
x
(x; H)dA
x

A
x
[9]
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A957
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
q
///
x
(H) =
_
Vx
q
///
x
(x; H)dV
x
V
x
[10]
where V
x
is volume of the electrode domain geometry. A cell do-
main submodel uses these terms to evaluate the heat source from the
lower hierarchical domains using Eq. 11
q
///
H;x
(H) = q
//
x
(H)a
s;H
q
///
x
(H)e
asc
[11]
where a
s,H
is the specic electrode plate area, and e
asc
is the volume
fraction occupied by the electrode composites and the separator in
the cell composite volume. The net volumetric heat source at the
cell domain volume is the sum of q
///
H;x
and other volumetric heat
sources such as electrical heating at current collector foils and other
passive conductors
q
///
H
(H) = q
///
H;x
(H)

k
q
///
H;k
(H) [12]
The schematics shown in Fig. 2 summarize the model solution vari-
ables in each computational domain and the coupling variables
exchanged between the adjacent length-scale domains.
Flexible submodel and solver choices. The MSMD model is a
generic model framework allowing the submodel and the solver in
each domain to be independently chosen. As long as the predened
coupling variables are properly transferred between the submodels
at the designated physical time, the choice of the model equations,
geometries, spatial discretization and the solution schemes, and the
time step size in each domain is independent of those in the other
domains. During the exchange of model inputs between submodels,
information regarding the geometry or the computational grid is not
transferred. In this regard, the MSMD model approach decouples
the geometries of each domain, while the physics solved in the
domains are still coupled. Therefore, the rigorousness of the submo-
del solved in each domain can be independently adjusted for the
purpose of modeling. As shown in Fig. 2, information is exchanged
only between the adjacent scale domain submodels. For example, an
MSMD cell domain submodel does not directly communicate with a
particle domain submodel. This enhances the exibility in the total
number of domains simultaneously solved in the MSMD model.
There have been efforts to solve charge transfer kinetics, particle
lithium diffusion, and transports across electrode composites in a
single computational domain to explicitly resolve the inuence of
composite electrode morphology and composition.
16,17
An MSMD
cell domain submodel could be linked to this type of single-domain
mesoscale geometry electrode submodelat the expense of high
computational costor alternately coupled with a simple circuit
model to mimic electrode voltage-current response, maximizing
computational efciency.
Submodel Choice for Large-Capacity Stacked Prismatic Cell
Performance Prediction
Macroscopic cell design features regarding the thermal and elec-
trical conguration, such as the number of unit stacks of the elec-
trode pair, area of the unit electrode stack layer, thickness of the cur-
rent collector foils, size and location of current tabs, electric bus
geometries, and external heat transfer conditions, are known to
greatly impact the microscopic electrochemical processes and deg-
radation mechanisms, and, in consequence, the overall cell perform-
ance and life, especially in large battery systems. Therefore, for
wide acceptance of LIBs in large-capacity applications such as
hybrid electric and full electric vehicles, the need to enhance knowl-
edge of heat and electric current transport in a LIB system and their
impacts on the performance, ageing, and safety behavior is critical.
In this paper, the MSMD model is employed to perform thermal and
electrical design evaluations for a large-format stacked prismatic
cell. Microscopic cell design parameters, including material compo-
sitions, electrode loading thicknesses, and porosities, are held con-
stant. Rather, the impact of large-format cell design features such as
the location and size of electrical tabs and the electrode area of the
unit stack layer are varied.
The schematic in Fig. 3 together with the data in Table I summa-
rize the four different cell designs investigated in this study.
Because the focus of the study is to evaluate the impact of design
features in the cell domain, we use computationally efcient submo-
dels and solution schemes in the particle and the electrode domains.
The cell domain submodel resolves the complexity of the three-
dimensional geometry for the cell designs investigated. The submo-
del chosen for the electrode domain is a one-dimensional porous
electrode transport model. The submodel chosen for the particle do-
main is a spherical particle model. The temperature dependence of
some of the physiochemical properties used in this study is consid-
ered using a general Arrhenius form
W = W
ref
exp
E
W
act
R
1
T
ref

1
T
_ _ _ _
[13]
where W
ref
is the property value dened at the reference tempera-
ture, T
ref
=298 K. The activation energy, E
W
act
, determines the tem-
perature sensitivity of a general physiochemical property, W.
Figure 3. (Color online) Schematic description of the 20-Ah stacked pris-
matic cell designs investigated: (a) ND cell, (b) CT cell, (c) ST cell, (d) WS
cell.
Table I. Description of the investigated cell form factor designs for 20-Ah stacked prismatic cell.
Case Description L
x
(mm) L
y
(mm) L
z
(mm) Tab width (mm) Tab configuration
ND Nominal design 200 140 7.5 44 Adjacent tabs
CT Counter tab design 200 140 7.5 44 Counter tabs
ST Small tab design 200 140 7.5 20 Adjacent tabs
WS Wide stack-area design 300 140 5.0 44 Adjacent tabs
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A958
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
Further details of the governing equations and solution schemes
used in each domain are described in the following sections. To
improve computation speed, a model-reduction scheme
18
is adopted to
solve the particle and electrode domain submodels. This reduced-order
model describes the electrochemical/thermal governing equations as a
quasi-linear system in state-variable model (SVM) form. Appendix A
describes the SVM model-reduction procedure and the numerical setup
used in this study. Appendix B compares numerical simulation results
for the present parameters set to a higher-order solution using a com-
mercial nite-element method solver. Note that this SVM method is
not an exclusive submodel for the present MSMD model. Other possi-
ble methods for fast computation of the particle and electrode domain
submodels include proper orthogonal decomposition,
19
circuit-analog
representation,
20
or empirical polarization function.
21
Submodel choice in particle domain.The particle domain sub-
model selected in this study solves lithium diffusion in the solid par-
ticle volume and charge-transfer kinetics at the particle surface. The
Butler-Volmer kinetic expression provides transfer current density
at the particle surface, where the rate of electrochemical reaction is
driven by the overpotential, g, which is dened as the deviation of
the potential difference between the solid phase and the liquid phase
at the reaction site from the thermodynamic equilibrium potential, U
i
//
n
= i
//
o
exp
a
a
F
RT
g
_ _
exp
a
c
F
RT
g
_ _ _ _
[14]
g =
s

e
i
//
n
R
film
U [15]
In Eq. 16, the exchange current density, i
//
o
, is given as a function of
both the solid phase and the electrolyte concentrations according to
i
//
o
= k
i
c
e
( )
aa
c
s;max
c
s;e
_ _
aa
c
s;e
_ _
ac
= i
// ref
o
c
e
c
ref
e
_ _
aa
c
s;max
c
s;e
c
s;max
c
ref
s;e
_ _
aa
c
s;e
c
ref
s;e
_ _
ac
[16]
where k
i
is a kinetic rate constant and a
a
and a
c
are the anodic and
cathodic transfer coefcients, respectively. In this study, the refer-
ence exchange current density at 100% state of charge (SOC), i
// ref
o
,
is evaluated as a function of temperature using Eq. 13. The lithium
concentration in the active particles is obtained using Ficks law of
diffusion
oc
s
ot
= \
n
D
s
\
n
c
s
_ _
[17]
with a surface ux boundary condition for the rate of charge transfer
reaction at the surface
\
n
c
s

A
n
n
n
=
i
//
n
D
s
F
[18]
where n
n
is an outward surface normal unit vector at the particle
surface. The dependence of lithium diffusivity on the lithium con-
tent or on the mechanical phase of the material is not considered in
this study. Introducing the spherical coordinate system, shown in
Fig. 4a, in the particle n-domain with symmetry assumption in the
inclination and azimuth angles simplies Eq. 17. The simplied
one-dimensional form of the equation is shown in Table II, where
n
1
= r. The solid phase diffusion coefcient, D
s
, is treated invariant
with the r-coordinate, even though it varies with temperature; tem-
perature, T, is given a lumped value in the particle domain submodel
as are the electrolyte phase lithium concentration, c
e
, the solid phase
potential,
s
, and the electrolyte phase potential,
e
. The surface
heat source at the active particle surfaces is modeled as shown in
Eq. 19 considering heat from the charge transfer reaction and the
ohmic loss at particle surface layer
q
//
n
= i
//
n

s

e
U T
oU
oT
_ _
[19]
Particle volumetric heat sources such as heat from a phase change
or the heat of mixing are ignored in this study. The average current
and heat sources for inter-domain coupling parameters passed to the
electrode domain are simplied from Eq.1, Eqs. 5 and 6 into Eq. 20,
Eqs. 21 and 22 in the submodel chosen

i
//
n
= i
//
n
[20]
q
//
n
= q
//
n
[21]
q
///
n
= 0 [22]
Submodel choice in electrode domain. A computationally
efcient electrode domain submodel is a preferred choice for the
purposes of this study. A simplied electrode domain transport
model such as a single-particle model
22,23
assuming lumped poten-
tials and liquid concentration has been used and performs reason-
ably well for thin electrode cells in moderate charge and discharge
conditions. However, in the present work, high pulse response and
end of discharge behavior of different cell designs are of interest.
Therefore, the porous electrode model is adopted in the electrode
domain to resolve the charge balance in the solid and liquid phases
and conservation of lithium in the electrolyte phase. Instead, the
SVM method is applied, compensating for the computational cost
for solving the relatively complex model. The computational do-
main for this model is composed of three contiguous volumes where
the solid phase and the liquid phase are treated as superimposed
continua as shown in Fig. 4b. Lithium concentration and electric
potential in the liquid electrolyte, c
e
and
e
, are evaluated in the
continuous liquid phase in these three volumes, while negative and
positive electrode potentials in the composite electrodes,
s,a
and

s,c
, are solved in the solid phase of the anode and cathode electrode
matrix, respectively.
o e
e
c
e
( )
ot
= \
x
D
eff
e
\
x
c
e
_ _

1 t
o

F
j
///
x

i
e
\
x
t
o

F
[23]
\
x
c
e
n
x
= 0 at all boundaries
\
x
j
eff
\
x
/
e
_ _
\
x
j
eff
D
\
x
ln c
e
_ _
j
///
x
= 0 [24]
\
x

e
n
x
= 0 at all boundaries
\
x
r
eff
\
x

s
_ _
j
///
x
= 0 [25]

s
= U or U

at the interface with a current collector,


\
x

s
n
x
= 0 at other boundaries
Incorporating the in-plane statistical homogeneity assumption in
the electrode domain, the one-dimensional forms (x
1
= x) of the
governing equations and boundary conditions used in the submodel
choice of this study are summarized in Table II.
Figure 4. Choices of submodel in each domain for large-format stacked
prismatic cell design evaluation study presented.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A959
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
The volumetric source of the electric charge for the charge trans-
fer lithium ion into the liquid phase, j
///
x
, is evaluated from

i
//
n
using
Eq. 2. U
-
, U

, and T, given from the cell domain, are treated as spa-


tial invariants in the electrode domain. In the one-dimensional
model, the electronic current density at the boundary of electrode
composite is evaluated with Eq. 26.
i
//
x
= r
eff
c
o
s;c
ox

x=lalslc
=
_
lalslc
lals
j
///
x
dx
= r
eff
a
o
s;a
ox

x=0
=
_
la
0
j
///
x
dx [26]
The average electrode plate current density,

i
//
x
, the predened inter-
domain coupling term passed to the cell domain submodel, is equiv-
alent to i
//
x
from Eq. 3 for the one-dimensional electrode model

i
//
x
= i
//
x
[27]
The averaged heat sources, q
//
n
and q
///
n
, delivered from the particle
domain submodel are converted into a volumetric heat source in the
electrode domain, q
///
x;n
, using Eq. 7. The joule heat for the electric
current ow in the presence of the potential gradient in solid matri-
ces and in the liquid electrolyte is calculated as
q
///
x;X
= r
eff
\
s
\
s
j
eff
\
e
\
e
j
eff
D
\ln c
e
\
e
[28]
The averaged volumetric heat source, q
///
x
, a quantity for passing to
the cell domain submodel, is evaluated using Eq. 29
q
///
x
=
_
lalslc
0
q
///
x;n
q
///
x;X
_ _
dx
l
a
l
s
l
c
[29]
The surface heat source at the electrode domain is not considered in
this study
q
//
x
= 0 [30]
Submodel choice in cell domain. For the cell domain sub-
model in the present study, the Single Potential-Pair Continuum
(SPPC) model was developed to solve for the temperature, T, and a
pair of electric potentials, U

, and U

, in a three-dimensional cell
geometry. Through the statistical homogeneity assumption, the
SPPC model treats the cell composite jelly volume as a continuum
having orthotropic thermal and electrical conductivities. Therefore,
the layered geometry of the cell jelly is not necessarily resolved in a
computational mesh of the SPPC model. In typical designs, conduc-
tion transport in the cell composite volume occurs with distin-
guished in-plane and transverse diffusivities. The SPPC model
resolves the anisotropic conduction in the cell composite volume
using the conductivity tensor, ~c
c
ij
= c
t
c
p
_ _
e
t
i
e
t
j
c
p
d
ij
[31]
where c
p
and c
t
are planar and transversal conductivities, respec-
tively, and d
ij
is the Kronecker delta. The unit transverse direction
vector is determined as a function of location in the cell composite
volume
Table II. Summary of solution variables and governing equations of the submodel choices.
Domain Solution variable Governing equation
Particle Submodel Choice: 1D spherical particle model
i
//
n
(r = R
s
; x; X; Y; Z) i
//
n
= k
i
c
e
( )
aa
c
s;max
c
s;e
_ _
aa
c
s;e
_ _
ac
exp
a
a
F
RT
g
_ _
exp
a
c
F
RT
g
_ _ _ _
c
s
(r; x; X; Y; Z)
oc
s
ot
=
D
s
r
2
o
or
r
2
oc
s
or
_ _
;
oc
s
or

r=0
= 0,
oc
s
or

r=Rs
=
i
//
n
D
s
F
Electrode Submodel Choice: 1D porous electrode model
c
e
(x; X; Y; Z)
o e
e
c
e
( )
ot
=
o
ox
D
eff
e
o
ox
c
e
_ _

1 t
o

F
j
///
x

i
//
e
F
ot
o

ox;
oc
e
ox

x=0
= 0,
oc
e
ox

x=lals lc
= 0

e
(x; X; Y; Z)
o
ox
j
eff
o
ox

e
_ _

o
ox
j
eff
D
o
ox
ln c
e
_ _
j
///
x
= 0
o
e
ox

x=0
= 0,
o
e
ox

x=lals lc
= 0

s;a
(x; X; Y; Z)
o
ox
r
eff o
ox

s
_ _
j
///
x
= 0;
s;a

x=0
= U

,
o
s;a
ox

x=la
= 0

s;c
(x; X; Y; Z)
o
ox
r
eff
o
ox

s
_ _
j
///
x
= 0;
o
s;c
ox

x=lals
= 0,
s;c

x=lals lc
= U

Cell Submodel Choice: 3D Single Potential-Pair Continuum (SPPC) model


U

(X; Y; Z)
o
oX
e

oU

oX
_ _

o
oY
e

oU

oY
_ _
j
///
H
= 0
U

(X; Y; Z)
o
oX
e

oU

oX
_ _

o
oY
e

oU

oY
_ _
j
///
H
= 0
T(X; Y; Z)
o qc
p
T
_ _
ot
=
o
oX
k
p
oT
oX
_ _

o
oY
k
p
oT
oY
_ _

o
oZ
k
t
oT
oZ
_ _
q
///
H
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A960
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
e
t
(H) = e
t
1
^
i
1
e
t
2
^
i
2
e
t
3
^
i
3
[32]
where
^
i
i
is the basis of H space, and e
t
i
is the corresponding compo-
nent of the transverse direction vector. In order to apply the SPPC
model, the following conditions should be met in the system of
interest:
i) In a nite volume of cell composite jelly in H space, e
t
can be
uniquely determined.
ii) In a nite volume of cell composite jelly in H space, U

and
U

can be uniquely determined.


Aligned stack cells, which are investigated in this study,
and spirally wound cells with extended foil type continuous cur-
rent tabs are the examples of systems where the SPPC model is
applicable. The computational domain of the SPPC model can be
divided into several zones. U

is computed from charge conserva-


tion in the negative current collector phase of the cell composite
and the negative-bias passive components outside the cell
composite
\
H
~ r
eff

\
H
U

_ _
j
///
H
= 0 [33]
while U

is solved in the positive current collector phase of the cell


composite and the positive-bias passive components outside the cell
composite
\
H
~ r
eff

\
H
U

_ _
j
///
H
= 0 [34]
where j
///
H
is the volumetric electric charge source in the positive
current collector phase and is evaluated with Eq. 4. Effective elec-
trical conductivity tensors in Eqs. 33 and 34 are reduced from
Eq. 31 as
r
eff

ij = d
ij
e
t
i
e
t
j
_ _
e

[35]
r
eff

ij = d
ij
e
t
i
e
t
j
_ _
e

[36]
where e

and e

are the volume fractions of negative and positive


current collectors in the cell composite volume, and r

and r

are
the electrical conductivity values of the conductor components car-
rying electronic current. If the transverse direction vector is dened
as a zero vector (e
t
= 0) outside the cell composite jelly, Eqs. 35
and 36 are effective both in the cell composite volume and in other
passive conductor volumes. Boundary conditions for Eqs. 33 and 34
are evaluated from the electrical load by which the cell is operated,
such as output current, voltage, or power. Energy conservation
yields
o qc
p
T
_ _
ot
= \
H
~
k\
H
T
_ _
q
///
H
[37]
where the thermal conductivity tensor
~
k is given with Eq. 31. The
volumetric heat source in the cell domain submodel, q
///
H
, includes
the heat evaluated from the electrode domain submodel and joule
heat at the passive components carrying electronic current in the
cell domain
q
///
H
= q
///
H;x
q
///
H;X
= q
///
x
e
asc

~ r
eff

\
H
U

~ r
eff

\
H
U

~ r
eff

\
H
U

~ r
eff

\
H
U

[38]
The cell output current I
o
is given as
I
o
=
_
VH
j
///
H
dV
H
[39]
In this study, the SPPC model computational domain only includes
the cell composite jelly volume (i.e., complex geometries of other
cell parts are neglected), so that the structured solver can be used.
By introducing the Cartesian (X, Y, Z) coordinate system in H-
space and aligning the cell-layer normal direction to the Z axis, the
transverse direction vector becomes invariant over the entire compu-
tational domain, i.e., e
t
= (0; 0; 1). The reduced governing equations
are summarized in Table II. The differential forms of governing
equations are discretized using the nite volume method (FVM).
Structured orthogonal hexahedral meshes are constructed and used
for simulations with 22, 17, and 7 for the numbers of discretization
in the X, Y, and Z directions, respectively. The intrinsic volumetric
conservation in the FVM has the advantage in the solution of the
MSMD model.
Model Analysis and Results
Four different stacked cell designs are numerically investigated
using the MSMD model setup described in Section 3. The theoreti-
cal capacity of the proposed cell designs is 20.46 Ah, based on refer-
ence stoichiometries at 0 and 100% SOC- and the active material
content in the cells presented in Table III. However, considering ki-
netic and transport limitations, the nominal capacity of the cells is
regarded as 20 Ah in this study. The dimensions of the nominal
design (ND) cell are 200 mm in length, 140 mm in width, and 7.5
mm in thickness. It has 44-mm-wide terminal tabs located on the
same side of the cell, as shown in Fig. 3a. The counter-tab (CT) cell,
shown in Fig. 3b, is identical to the ND cell except for the congura-
tion of the electrical tab, which are located on opposite ends. The
small-tab (ST) cell (Fig. 3c) has smaller, 20-mm-wide tabs, while
the wide-stack (WS) cell (Fig. 3d) has a stack area 1.5 times larger,
and therefore has fewer stack layers, than the ND cell. With the pro-
posed cell designs, the dependence of the performance response of
stacked cells on the location and size of the tabs and the aspect ratio
of the cell stack is explored against the reference ND cell. Constant
current discharge and hybrid pulse power characterization (HPPC)
tests are simulated to compare the capacity and rate capability of the
proposed designs. A US06 driving prole test is simulated to
explore application of the proposed cell designs to a mid-size sedan
10-mile electric range plug-in hybrid electric vehicle (PHEV10)
energy storage system. A convective heat transfer boundary condi-
tion was applied on the top face of each design using 25 W/m
2
K for
the heat transfer coefcient and 25

C as the ambient temperature in


all cases, reecting single-side cooling thermal management. In the
present study, all three domain submodels were run with identical
time steps. The HPPC test simulation uses 100-ms xed time steps,
and the other cases use 1-s time steps. Solution variables over the
three domains are tightly coupled through the rst-order implicit
temporal discretization and Picard iteration addressing the nonli-
nearity of the equations. The MSMD model input parameters for the
submodel choices in this study are summarized in Table III.
Constant current discharge test simulation. Figure 5 compares
the voltage curves of the investigated cell designs for 100A
(5C-rate), 60A (3C-rate), and 20A (1C-rate) constant current dis-
charge cases. During discharge, the output voltages of the four cells
show several millivolts of variation. The discharge capacities at the
2.5 V cut-off voltage are very similar: 18.9 Ah in 5C discharge, 19.6
Ah in 3C discharge, and 20.2 Ah in 1C discharge. Even though sev-
eral millivolts difference in cell voltage can cause meaningful
changes in battery kinetics, uncertainties from typical measurement
errors and manufacturing variability make it difcult to distinguish
a few millivolts discharge voltage variability among the different
designs from experimental variability. The apparent cell output volt-
age does not seem to be affected signicantly by the design changes
between the comparable stacked cells. However, the model predicts
that the internal cell kinetics, which is not easily measured, is quite
different among the cells and closely follows the thermal and elec-
trical congurations of the cell designs.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A961
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
Table III. Summary of model parameters in the submodel choices.
Domain Parameter Value/Model
Particle Li
x
C
6
Li
y
(NCA)O
2
Ref # Maximum Li capacity, c
s,max
(mol m
3
) 2.87 10
4
4.90 10
4
Characteristic diffusion length, R
s
(m) 3e
s
/a
s
3e
s
/a
s
Stoichiometry at 0% SOC, x
0%
, y
0%
0.0712 0.98
Stoichiometry at 100% SOC, x
100%
, y
100%
0.63 0.41
[27],[25] Reference exchange current density at 100%
SOC, i
o
ref
[(A m
2
)
36.0 4.0
- activation energy, E
io
act
(J/mol) 3.010
4
3.0 10
4
Charge-transfer coefcients, a
a
, a
c
0.5, 0.5 0.5, 0.5
[26],[10] Film resistance, R
lm
(X m
2
) 0.023 0.015
[26],[10] Solid diffusion coefcient, D
s
(m
2
s
1
) 9.010
-14
3 10
15
- activation energy, E
Ds
act
(J/mol) 4.010
3
2.0 10
4
[26] Negative electrode, U
-
(V)
U

x ( ) = 0:124 1:5 exp 70x ( ) 0:0351 tanh


x 0:286
0:083
_ _
0:0045 tanh
x 0:9
0:119
_ _
0:035 tanh
x 0:99
0:05
_ _
0:0147 tanh
x 0:5
0:034
_ _
0:102 tanh
x 0:194
0:142
_ _
0:022 tanh
x 0:98
0:0164
_ _
0:011 tanh
x 0:124
0:0226
_ _
0:0155 tanh
x 0:105
0:029
_ _
Fit to [24] Positive electrode, U

(V) U

x ( ) = 1:638x
10
2:222x
9
15:056x
8
23:488x
7
81:246x
6
344:566x
5
621:3475x
4
554:774x
3
264:427x
2
66:3691x 11:8058 0:61386 exp 5:8201x
136:4
_ _
Electrode Negative electrode composite Separator Positive electrode composite
Ref # Thickness, l
a
,l
s
, l
c
(m) 70.0 10
6
25 10
6
50.0 10
6
Thickness unit stack, l
asc
(m) l
a
l
s
l
c
[25],[24] Volume fraction inert, e
f
0.09 0.6 0.19
Volume fraction electrolyte, e
e
0.4 0.4 0.4
Volume fraction active material, e
s
0.51 0.41
Specic active surface area, a
s,x
(m
2
m
3
) 3.010 10
6
0.753 10
6
[26],[24] Solid electronic conductivity, r
a,
, r
c
(S m
1
) 100.0 10
Bruggeman tortuosity exponent, p 2.0 2.0 2.0
Electrolyte concentration, c
e
(mol m
3
) 1.2 10
3
[25] Electrolyte Li

diffusion coefcient,
D
e
(m
2
s
1
)
D
e
= 5:84 10
7
exp[2870=T[(c
e
=1000)
2
33:9 10
7
exp[2920=T[(c
e
=1000)
129 10
7
exp[3200=T[
[25] Electrolyte ionic conductivity, j (S m
1
)
j = 3:45 exp[798=T[(c=1000)
3
48:5 exp[1080=T[(c=1000)
2
244 exp[1440=T[(c=1000)
Effective solid phase conductivity,
r
eff
(S m
1
)
re
s
p
Effective liquid phase diffusion coefcient,
D
e
eff
(m
2
s
1
)
D
e
e
e
p
Effective liquid phase ionic conductivity,
j
eff
(S m
1
)
je
e
p
Effective liquid phase diffusional ionic
conductivity, j
eff
D
(S m
1
)
j
eff
D
=
2RTj
eff
F
(t
0

1) 1
d ln f6
d ln ce
_ _
[25] Li

transference number, t
o

t
0

= 0:000267 exp[883=T[(c
e
=1000)
2
0:00309 exp[653=T[(c
e
=1000)
0:517 exp[49:6=T[
Thermodynamic factor, o ln f
6
=o ln c
e
0
Cell Negative current collector Positive current collector
Thickness, d

, d

(m) 10.010
6
15.0 10
6
Specic electrode plate area, a
s,H
(m
2
m
3
) (l
asc
0.5d

0.5d

)
1
Volume fraction, e

, e

0.5d

a
s,H
0.5d

a
s,H
Conductivity, r

, r

(S m
1
) 59.6 10
6
37.8 10
6
Electrode plate area, A
H,p
(m
2
) 1.33
Volume, cell composite, V
H,cc
(m
3
) A
H,p
/a
s,H
Volumetric heat capacity, cell composite,
qc
p
, (J K
1
m
3
)
2.04 10
6
Planar thermal conductivity, cell composite,
k
p
(W m
1
K
1
)
27
Transversal thermal conductivity, cell
composite, k
t
(W m
1
K
1
)
0.8
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A962
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
Figure 6 shows the contours of electrode plate current density, i
//
x
,
at the cell composite volume near the bottom plane of the cells after
6 min of 5C discharge. An electric charge capacity of 10 Ah was
discharged from each cell, and the average electrode plate current
density is constant at 75 A/m
2
. Electrical current in the current col-
lector foils converges to or diverges from the tabs of the cells.
Therefore, a fast electrostatic potential change happens in the cell
composites near the tabs, causing locally larger deviations from the
thermodynamic equilibrium. In consequence, the cell composites
nearest the tabs are preferentially discharged initially. In addition,
higher temperatures for higher currents can further energize the
local transfer reaction. This positive feedback promoting non-uni-
form discharge reaction over a cell is limited by thermodynamic
equilibrium shifts at the reaction sites. As inferred from Eq. 15,
steep changes in the open circuit potential, U, with the lithium con-
tent help to mitigate the non-uniformity caused by the thermal-elec-
trical imbalance. In the ND cell shown in Fig. 6a, the electrode plate
current density is larger near the tabs. The CT cell achieves a more
uniform electric potential difference across the positive and negative
current collector foils, and subsequently the most uniform distribu-
tion of electrode plate current density among the designs can be
compared, as seen in Fig. 6b. The electrode plate current density
eld in the ST cell, shown in Fig. 6c, is very similar to that of the
ND cell except that even higher electrode plate current densities
appear near the tabs. The most uneven kinetics is observed in the
WS cell design, shown in Fig. 6d. The WS cell is 1.5 times wider in
the stack area than the ND cell; therefore, each current collector
sheet carries a proportionally larger electric current for a longer dis-
tance. This results in an uneven electric potential eld along the cur-
rent collectors and uneven transfer reaction over a cell composite
volume.
During discharge of the cells, a cell-internal SOC imbalance
occurs as a result of non-uniform discharge in the investigated cells.
The contours of the electrode plate SOC at the cell composite vol-
ume near the bottom plane of the cells after 6 min of 5C discharge
are presented in Fig. 7. Because cell degradation involved with lith-
ium loss or active site loss is not considered in this model study, the
mean SOC of the electrode plate is directly determined from the
stoichiometry numbers averaged over particles and electrode com-
posites. The evolution of the imbalance in the electrode plate SOC
within the cell during 5C discharge in each cell design is plotted in
Fig. 8a. The internal deviation of the SOC of the simulated cells
increases until about half of the cell capacity is discharged, then it
starts to decrease and quickly converges at the end of discharge with
sharp drops. This behavior can be explained with the voltage charac-
teristics of the cell. Figure 8b shows dQ/dV, which is an inverse of
the voltage slope calculated from the 5C discharge curves shown in
Fig. 5. The curves shown in Fig. 8b mainly represent the combined
Figure 5. (Color online) Discharge voltage comparison among the investi-
gated cell designs for 100A (5C), 60A (3C), and 20A (1C) constant current
discharge events.
Figure 6. (Color online) Contour of electrode plate current density at the
cell composite volume near bottom plane of the cells after 6 min of 5C dis-
charge: (a) ND cell, (b) CT cell, (c) ST cell, (d) WS cell.
Figure 7. (Color online) Contour of electrode plate SOC at the cell compos-
ite volume near bottom plane of the cells after 6 min of 5C discharge: (a) ND
cell, (b) CT cell, (c) ST cell, (d) WS cell.
Figure 8. (Color online) (a) Instantaneous maximum difference of electrode
plate SOC in the cells simulated during 5C constant current discharge, (b)
Inverse of discharge voltage slope of the cells simulated during 5C constant
current discharge.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A963
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
effect from the open circuit potentials of the cathode and the anode
materials of the cell. The maxima of the SOC imbalance in the
middle of the discharge of the simulated cells correspond to the
maximum of dQ/dV at around 10-Ah discharge. Near the end of dis-
charge, the SOC distribution in the cells quickly converges because
of the large slope of the voltage curve. The resemblance of these
two plots implies that cycling a battery along a slowly varying volt-
age plateau can enhance non-uniformity of material use in a cell.
Increasing a voltage slope by modifying the thermodynamic charac-
teristics of the electrode materials for a better balanced use of a cell
may require considerable effort. Alternately, improving the electri-
cal conguration of a cell can greatly lessen the cell use imbalance,
as shown in Fig. 8a. In the WS cell, the difference in SOC reaches
2.89% during a constant current discharge at 5C, while the SOC dif-
ference of the CT cell only reaches 0.56%.
Heat generation rate during 5C discharge for each cell is plotted
in Fig. 9a. The average temperature evolutions of the CT, ST and
WS designs are presented with minimum and maximum cell internal
temperature range bounds, compared to the temperature response of
the ND cell in Figs. 9b, 9c, and 9d. The average temperature evolu-
tions of the CT cell and the ST cell are very similar to that of the
ND cell, since these three designs have similar heat generation
amounts and heat transfer conditions. However, the cell internal
temperature range is narrower in the CT cell and wider in the ST
cell as compared to the ND cell. The WS cell shows a distinct ther-
mal behavior. It generates higher heat during discharge, but ends up
with a lower average cell temperature at the end of discharge. In the
initial stage of discharge, when heat generation dominates cell heat-
ing behavior, the maximum temperature increases faster in the WS
cell than in the ND cell, and average temperatures evolve in a simi-
lar fashion in both designs. But as the cells heat up, the impact of
heat transfer becomes signicant. Since the WS cell has a 1.5 times
larger cooling surface and a shorter thermal diffusion distance in the
normal direction to the stack layer, the enhanced heat rejection lim-
its the temperature increase of the WS cell during the later stage of
discharge. In Fig. 10, temperature contours at nine cross-sectioned
surfaces of each cell are presented to show details of the spatial tem-
perature imbalance at the end of discharge. Due to preferential
kinetics and electric current convergence, a higher temperature is
observed near the tabs in all cell designs. Unlike the other designs,
the tabs of the CT cell are not co-located on the same end of the
cell. So, the CT cell, as shown in Fig. 10b, has the most uniform
temperature distribution among the compared designs. Its main tem-
perature gradient exists in the normal direction to the stack plate
with the lowest temperature at the top surface, which is cooled by
ambient coolant. The spatial temperature prole of the ST cell,
shown in Fig. 10c, appears similar to that of the ND cell because of
a similar distribution of kinetics over the electrode plates as implied
by the electrode plate current density distribution presented in
Fig. 6. The peak temperature, however, is higher near the smaller-
sized tabs. Thanks to the larger surface area for cooling, the average
temperature at the end of discharge of the WS cell, shown in Fig. 10d,
is lower by a few degrees Celsius than those for the other designs;
however, the difference in the internal temperature in the WS cell is
still high among the cells compared. The discharge efciency of each
cell design is evaluated using heat generated and electric energy out-
put during discharge and is summarized in Table IV.
Figure 11 shows the spatial variation of the electrode-domain av-
erage volumetric heat generation rate, q
///
x
, which is evaluated in this
study using Eq. 29 from the electrode domain submodel and appears
in the rst term on the right-hand side of Eq. 38 evaluating the
source term for the energy conservation equation solved in the cell
domain submodel. The results are shown at the cell composite vol-
ume near the cell bottom plane after 6 min of 5C discharge. This av-
erage quantity over the electrode domain volume includes heat from
the transfer reaction and ohmic loss for charge transport under
electric elds in the cathode and anode electrode composite
matrices and in the liquid electrolyte. Therefore, q
///
x
has a close corre-
lation with the electrode plate current density, i
//
x
, shown in Fig. 6. The
cell composite volume heat source, q
///
H
, calculated with Eq. 38,
includes joule heat in the current collector phase of the cell composite
volume on top of the contribution from q
///
x
. Joule heat from electrical
heating in the current collector foils contributes a relatively small
amount, just 510% of total heat generation in the simulated cell
designs. However, depending on the internal electrical path design of a
cell, this heat is highly localized, causing spatial non-uniformity in q
///
H
,
as seen in Fig. 12.
Hybrid pulse power characterization test simulation. The
pulse responses are compared through the HPPC test simulations for
the cell designs examined. Figure 13 presents the HPPC voltage
responses of the cells at 20% SOC. The HPPC prole consists of a
10-s duration 5C discharge pulse, a 40-s rest period, and a 10-s
3.75C charge pulse. The area-specic resistances for the electrode
plate are evaluated from the voltage curves and presented in the g-
ure. The lumped cell, reasonably applicable to small cells, plotted
in Fig. 13 assumes that the current collectors carry electric current
without resistance and that heat is transferred faster inside the cell
compared to the external heat exchange rate, so that the entire elec-
trode sandwich of a cell is considered as being operated with identi-
cal potential difference at the same temperature. Therefore, the dif-
ferences between the HPPC pulse resistance of the lumped cell and
those of the other three-dimensional cells (i.e., ND, CT, ST, and
WS) originate from non-ideal electrical and thermal transport. For
the stacked cells simulated in this study, the major portion of the
pulse resistance increase from the lumped cell comes from ohmic
resistance for current transport through the current collectors. The
insets in Fig. 13 are magnied views of the cell output voltage
behavior during relaxation right after the discharge pulse. The
lumped cell shows several millivolts higher initial relaxation voltage
than the other three-dimensional cells. The voltage curves merge in
the later stages of relaxation. In the lumped cell model, relaxation
resolves lithium diffusion within the particle and redistribution of
lithium through the depth of each electrode composite plate. In the
three-dimensional cells, the 10-s discharge establishes an uneven
Figure 9. (Color online) (a) Net heat generation rate of the cells simulated
during 5C constant current discharge, (b) Comparison of the evolution of av-
erage temperature (solid) with maximum and minimum temperatures (dotted)
of the CT cell against those of the ND cell during 5C constant current dis-
charge, (c) Same for the ST cell, (d) Same for the WS cell.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A964
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
SOC distribution across the electrode plates. Therefore, three-
dimensional cell models resolve an additional relaxation process not
captured in the lumped cell model; i.e., lithium redistribution
between the counter electrode composites accompanying local ion
current through the separator. Figure 14 presents the contours of
electrode plate current density 1 s after the end of discharge pulse,
showing local charging and discharging current driven by in-plane
SOC imbalance. The model predicts sub-millivolt output voltage
differences among the initial relaxation voltages of the cell designs
compared. The more uneven in-plane SOC distribution results in
lower initial voltage relaxation following interruption of the dis-
charge pulse current.
Vehicle driving test simulation. Standard cell characterization
tests such as the constant current discharge test and the HPPC test
provide useful information about the cells electrical-thermal per-
formance characteristics. The design of a cell, however, must be
evaluated with application-specic use scenarios as well because
the response of a battery system is largely affected by the attributes
of the application and the operation strategies. The cell designs
investigated here are examined for use in a battery system for a
mid-size PHEV10 sedan. Vehicle simulation was conducted over a
repetition of an aggressive vehicle speed prole, known as the US06
cycle, to determine the power demand for the vehicles battery. The
simulated PHEV10 vehicle consumes battery energy during charge-
depleting (CD) mode for the initial 16 km (10 mile) of driving, and
Table IV. Cell efciency during constant current discharge.
Discharge Rate ND (%) CT (%) ST (%) WS (%)
5C 93.74 93.78 93.64 93.43
3C 96.23 96.26 96.16 96.04
1C 98.63 98.64 98.62 98.58
Figure 10. (Color online) Contours of temperature at nine cross-sectioned surfaces in cell composite volume at the end of 5C constant current discharge: (a)
ND cell, (b) CT cell, (c) ST cell, (d) WS cell. (Dimensions in Z direction of the contour surfaces are exaggerated for clear view of quantity variation in Z
direction.)
Figure 11. (Color online) Contour of electrode plate (electrode-domain-av-
erage) volumetric heat generation rate at the cell composite near bottom
plane of the cells after 6 min of 5C discharge: (a) ND cell, (b) CT cell, (c)
ST cell, (d) WS cell.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A965
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
subsequent cycling occurs in the charge-sustaining (CS) hybrid
drive mode maintaining a steady battery charge level. Figure 15
presents the battery power prole, which is scaled for an individual
20-Ah cell in the pack and used as a cell model input for electric
load prole. During CD mode driving, which displaces petroleum
use in the vehicle, about 60% of the charge capacity of the battery is
consumed, and then the battery is cycled at around 20% SOC in CS
mode driving. As shown in Fig. 16a, the evolutions of the average
SOC of the ND, CT, ST, and WS cells over a 15-min vehicle drive
are quite similar. However, the imbalance of the electrode SOC
within each cell, shown in Fig. 16b, is greatly affected by the cell
design. In all the designs investigated, the maximum discrepancy in
SOC occurs at around 5 min, where the average cell SOCs corre-
spond to the small voltage slope in the middle section of the voltage
curve, and the cells run on continual high-rate discharge pulses. The
model results indicate that the cell response, such as SOC imbalance
implying non-uniform cycling of cell materials, is determined
through the close interaction among the various length-scale physics
and designs from material properties to system control.
Temperature control is especially important in LIB vehicle appli-
cations to extend the battery life while maximizing the electric per-
formance of a vehicle. To prevent rapid degradation, excursions of
the battery temperature to a high temperature must be limited. At
the same time, the battery temperature should be kept in a narrow
range across the system to avoid excessive non-uniform kinetics.
Limiting cell-to-cell and cell-internal temperature variation is more
critical in high-voltage battery systems where a large number of
cells are connected in series. The thermal response of the cell
designs investigated in the PHEV10 US06 driving simulations are
compared for average temperature, shown in Fig. 17a, maximum
temperature, shown in Fig. 17b, and internal temperature difference,
shown in Fig. 17c. Cumulative heat generation and heat transfer to
ambient from each cell are plotted in Fig. 17d. The WS cell gener-
ates higher heat than the other cell designs, but it also rejects heat to
the ambient environment at a much faster rate. The larger cooling
surface area of the WS cell brings the advantage of a lower average
cell temperature over the other three designs. The average tempera-
tures of the other three cells behave similarly. However, in terms of
maximum temperature and internal temperature imbalance, each
Figure 12. (Color online) Contour of net volumetric heat generation rate at
the cell composite volume near bottom plane of the cells after 6 min of 5C
discharge: (a) ND cell, (b) CT cell, (c) ST cell, (d) WS cell.
Figure 13. (Color online) Comparison of the voltage responses of the simu-
lated cell designs at 20% SOC for the HPPC pulse prole consisting of a 10 s
duration 5C discharge pulse and a 10 s duration 3.75C charge pulse with 40 s
rest period between the pulses.
Figure 14. (Color online) Contour of electrode plate current density at the
cell composite volume near bottom plane of the cells at 1 s after the HPPC
discharge pulse: (a) ND cell, (b) CT cell, (c) ST cell, (d) WS cell.
Figure 15. (Color online) Battery power demand prole produced from ve-
hicle simulation a US06 driving cycle for mid-size sedan PHEV10, scaled
for an individual 20-Ah cell in the vehicle battery pack. Aggressive accelera-
tion and high driving speed characteristics of the US06 cycle result in high
heat generation rates for the battery. The above power prole has an average
C-rate of about 2.5 and an RMS C-rate of about 5.0 for the cell designs
compared.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A966
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
design behaves differently. Increasing the electrode stack-area with-
out thickening the current collector foils, as is the case with the WS
cell, causes a larger electrical current to converge in the current col-
lectors toward the cells electrical tabs, and consequently energizes
local charge transfer kinetics and generates more heat near the tabs.
As a result, the maximum local temperature of the WS cell is high-
est among the compared cells during the most of CD mode driving.
While the average temperature increase from the initial cell temper-
ature is less than 20

C, the WS cell shows excessive internal tem-


perature differences that reach up to about 10

C, which would be
unacceptable in the practice of vehicle battery system integrators.
Therefore, in spite of its large stack plate area available for better
cooling, the WS cell is not the best thermal design for use in the
PHEV10 application. However, it should be noted that the thermal
imbalance can be alleviated in lower power applications or with
more precise thermal management. To achieve a better-performing
thermal design and management strategy, the thermal behavior of
LIBs should be well understood, not only the characteristics of a
cell but also the characteristics of its application.
Figure 18 shows the contours of the electrode plate area-specic
ampere-hour throughput in the simulated cells. The absolute value
of charging and discharging electrode plate current density is inte-
grated over time during the 15-min PHEV10 drive, revealing spatial
variation of the cumulative electrochemical cycling over the cell
composite volume. In general, cell composites near the tabs are
preferentially cycled in all designs, but the unevenness of electrode
cycling is also greatly affected by the cells electrical-thermal con-
gurations. The average values of ampere-hour throughput per elec-
trode plate area are similar across the four cell designs: 13.12,
13.11, 13.15, and 13.20 Ah/m
2
for the ND, CT, ST, and WS cells,
respectively. However, the relative magnitudes of their internal vari-
ation compared to the average throughputs are signicantly differ-
ent. Ampere-hour throughput imbalances are 6.0, 2.5, 6.9, and
12.7% for the ND, CT, ST, and WS cells, respectively. Non-uniform
cycling of a cell is expected to bring subsequent effects in long-term
performance degradation of a LIB system. Therefore, the impact of
the electrical and thermal design of a battery should be adequately
considered in predicting the life of large battery systems. A study on
large-format battery degradation using the MSMD modularized
multi-physics LIB model framework is ongoing.
5
Conclusions
To enhance the understanding of the interplay among the physics
occurring in LIB systems at various length scales, a modularized
multiscale model framework, the MSMD model, was developed.
The MSMD model introduces multiple computational domains for
corresponding length scale physics. While the model decouples sub-
model geometries through the statistical homogeneity assumption, it
still couples the physics between the domains using the predened
inter-domain information exchange. The MSMD model selectively
enhances spatial resolution for the physics taking place at smaller
characteristic length scales, achieving high computational efciency
compared to a single domain model approach. Thanks to its modu-
larized hierarchical architecture, the MSMD model provides a exi-
ble and expandable framework facilitating multiphysics LIB model-
ing with various levels of physical and computational complexities.
In this study, the MSMD model is applied to evaluate the thermal
and electrical design of large-format stacked prismatic LIB cells.
Four different cell designs, ND, CT, ST, and WS, are investigated to
evaluate the impacts of tab conguration and size and cell stack as-
pect ratio for identical electrode-level designs. Apparent overall cell
Figure 16. (Color online) (a) Average SOC evolution of the cells simulated
during 15-min PHEV10 drive with the US06 cycle, (b) Instantaneous maxi-
mum difference of electrode plate SOC in the cells simulated during 15-min
PHEV10 drive with the US06 cycle.
Figure 17. (Color online) Comparison of thermal response of the investi-
gated cell designs from the PHEV10 US06 driving simulations: (a) for aver-
age temperature, (b) for maximum temperature, (c) for internal temperature
difference, (d) for cumulative heat generation and heat transfer to ambient.
Figure 18. (Color online) Contour of electrode plate ampere-hour through-
put at the cell composite volume near bottom plane of the cells during 15
min PHEV10 drive with the US06 cycle: (a) ND cell, (b) CT cell, (c) ST
cell, (d) WS cell.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A967
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
responses such as cell output voltage, which are directly measurable
with typical experiments, did not show great variation among the
compared cell designs, because the cell designs simulated here are
already fairly optimized. However, the model results revealed that
the cell internal battery kinetics is signicantly inuenced by the
macroscopic cell design for electrical current and heat transport.
The thermal, electrical, and electrochemical behaviors of LIB sys-
tems, for example, non-uniform electrochemical cycling over a
cell composite volume and subsequent internal SOC imbalances,
are found to be affected by the complex interaction among the
physics at varied length scales from thermodynamics to system
control.
Acknowledgments
The authors gratefully acknowledge David Howell, Brian Cun-
ningham, and the U.S. DOE Ofce of Vehicle Technologies Energy
Storage Program for funding and support and Jeffrey Gonder of
NREL for generating the vehicle application power prole.
Appendix A: State Variable Representation of a
1-Dimensional Porous Electrode Model
A reduced-order representation of the 1-dimensional electrochemical model of
Doyle and Newman
1
improves the computational speed of the present MSMD model
by a factor of 100 to 1,000, depending on the chosen geometry. This appendix summa-
rizes the reduction method
18
used here. The governing equations are treated as a quasi-
linear system. Nonlinearities are approximated by interpolating between reduced local-
linear models at various electrode-surface stoichiometries and temperatures. The
approximation generally works well in practice but should be veried for each new
electrochemical model parameter set as in Table III. The quasi-linear treatment loses
validity, for example, when severe spatial nonlinearities arise within the electrode sand-
wich, such as in the case of electrolyte depletion.
18
A.1 Model order reduction.The reduction procedure follows Section 3.4 of
Ref. 28. High-order frequency model simulations are run to determine the output
response of electrochemical eld variables j
///
x
, c
s,e
,
e
and c
e
to a small input perturba-
tion in the applied current

i
//
x
. At discrete spatial locations across the electrode sand-
wich, each eld variable is organized into an output vector, y. Applied current, a scalar
value, is model input u. Output/input transfer functions are constructed as y(s)=u(s),
where s is the Laplace variable. The complex frequency response is calculated by sub-
stituting s = jx where j is the square root of 1. For a full-order transfer function,
y(s)=u(s), its reduced-order counterpart is dened as
y
+
(s)
u(s)
= z

n
k=1
r
k
s
s k
k
[A-1]
with steady-state vector z obtained from the full-order model as z = lim
s0
y(s)=u(s),
and eigenvalues k
k
and residue vectors r
k
numerically generated by minimizing the cost
function
J =

m
k=1

n
i=1
Re y
+
i
(jx
k
) y
i
(jx
k
)
_ _

2
Im y
+
i
(jx
k
) y
i
(jx
k
)
_ _

2
[A-2]
across the frequency range x [ [0, 2pf
c
]. The cutoff frequency, f
c
, is chosen to be
slightly faster than the smallest time step desired for the nal time-domain model. In
this manner, fast dynamics such as double-layer capacitance and electrode lm trans-
port, are automatically captured in the reduced model with steady-state representations.
A time-domain representation of the reduced frequency model in state variable
form
29
is
_ x(t) = Ax(t) Bu(t)
y
+
(t) = Cx(t) Du(t) y
0
[A-3]
where
A = diag k
1
k
n
[ [; B = 1 1 [ [
T
C = r
1
k
1
r
n
k
n
[ [; D = z

n
k=1
r
k
_ _
[A-4]
and static constant y
0
gives output y
*
the proper value at the linearization point.
A.2 Frequency submodel in the particle domain.Analytical Laplace transfer
functions are available to describe the impedance due to lithium intercalation in active
material particles, that is the perturbation of the lithium surface concentration for a per-
turbation in charge transfer current. For a spherical particle, that solid-state diffusion
impedance is Ref. 30
c
s;e
(s)
i
//
f
(s)
=
1
F
R
s
D
s
_ _
tanh(b)
tanh(b) b
_ _
[A-5]
where b = R
s

s=D
s
_
. For linear charge transfer kinetics at the particle surface, the But-
ler-Volmer equation reduces to
R
ct
=
og
oi
//
n

i
//
n
=0
=
RT
i
//
o
F(a
a
a
c
)
[A-6]
A.3 Frequency submodel in the electrode domain.With the assumption of
uniform electrolyte concentration across each electrode, analytical transfer functions
are available to describe the perturbation of electrochemical eld variables j
///
x
and c
s,e
for a perturbation in applied current

i
//
x
. Modify the dimensionless impedance variable v
from Ong and Newman
31
to include solid-state diffusion impedance
m(s) = d
1
j
eff

1
r
eff
_ _1
2
R
ct

oU
oc
s
c
s;e
(s)
i
//
n
(s)
_ _ _ _

1
2
[A-7]
and dene the dimensionless electrode position z =x/l, where z =0 is the current collec-
tor interface and z =1 is the separator interface. Following the derivation,
18
the transfer
functions are
j
///
x
(z; s)

i
//
x
(s)
=
1
d
1
j
eff
r
eff
m(s)
sinh m(s)
j
eff
cosh v(s) z 1 ( ) [ [ r
eff
cosh m(s) z ( ) [ [
_ _
[A-8]
c
s;e
(z; s)

i
//
x
(s)
=
1
a
s
j
///
x
(z; s)

i
//
x
(s)

c
s;e
(s)
i
//
n
(s)
[A-9]
Equations A-8 and A-9 are written for the negative electrode. For the positive electrode,
multiply the right-hand sides of the equations by 1.
Analytical transfer functions for
e
and c
e
are unduly cumbersome. A numerical
approach is taken instead. Spatial discretization of Eq. 23 followed by Laplace transfor-
mation yields the numerical transfer matrix
c
e
(s)

i
//
x
(s)
= (K
ce
sM
ce
)
1
F
ce
j
///
x
(s)

i
//
x
(s)
[A-10]
In Eq. A-10, K
i
, M
i
and F
i
are the stiffness, mass, and forcing matrices dened by the
nite element method, and c
e
(s) and j
///
x
(s) are vectors representing eld variables
c
e
(x; s) and j
///
x
(x; s) at discrete node points x
i
. Similar treatment of Eq. 24 yields the
transfer matrix
/
e
(s)

i
//
x
(s)
= (K

j
e
)
1
K

j
D
e
c
e
(s)

i
//
x
(s)
F
e
j
///
x
(s)

i
//
x
(s)
_ _
[A-11]
A.4 Reduced-order submodel in the electrode domain.Applying the reduc-
tion procedure from Section A.1 to transfer functions Eqs. A-8, A-9, and A-11 yields
reduced-order state variable models for outputs c
+
s;e
(x; t), j
///
x
+
(x; t), and
+
e
(x; t)as a func-
tion of input

i
//
x
(t). Note that these reduced models are in the time domain in SVM
form. For a given applied current, the voltage drop across the 1-dimensional electrode
sandwich with thickness L=l
a
l
s
l
c
is thus approximated as

+
s;c
(L; t)
+
s;a
(0; t) =
+
e
(L; t)
+
e
(0; t) U

c
+
s;e
(L; t)
_ _
U

c
+
s;e
(0; t)
_ _
g j
///
x
+
(L; t)
_ _
g j
///
x
+
(0; t)
_ _
[A-12]
Appendix B: SVM Performance Comparison
The SVM that serves as a reduced-order 1-dimensional porous electrode model is cre-
ated using the procedure in Appendix A using the electrode and particle domain submo-
del parameters in Table III. Those model parameters are chosen to represent a moderate
power cell for PHEV application. The model reduction procedure, minimization of Eq.
A-2, is used to create local linear models for discrete values of temperature and stoichi-
ometry, the latter non-evenly spaced to capture features of each electrodes open-circuit
potential curve. The model order, the value n in Eq. A-1, is chosen such that reduced-
order models closely match full-order transfer functions from 0 to 10 Hz. The present
SVM uses a third-order model to approximate negative electrode dynamics (Eqs. A-8
and A-9), a third-order model for positive electrode dynamics (Eqs. A-8 and A-9), and
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A968
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp
a fourth-order model for electrolyte transport dynamics (Eqs. A-10 and A-11). Together
with one state to represent the SOC bulk dynamics, the complete SVM is an 11th order
set of ordinary differential equations.
Figure B1 compares the reduced-order SVM with a high-order model, solved using
the commercial nite element software, COMSOL. Figure B1 generally shows less
than 10-mV difference between the two models during 1C, 2C, and 4C constant current
discharges. An exception is during end-of-discharge where model differences increase
to as high as 30 mV. For the simulated cell, end-of-discharge occurs due to positive
electrode surface saturation causing a sudden drop-off in voltage. Slight differences in
the two models prediction of positive surface stoichiometry at the current collector are
responsible for the end-of-discharge model error as small differences in stoichiometry
cause large differences in voltage in this region of positive electrode operation.
In Ref. 18, a high-power HEV cell with graphite/nickel-cobalt-aluminum chemistry
was considered. Similar to that work, the present SVM results compare sufciently well
with the high-order model to serve as a surrogate porous electrode model for numerical
studies of PHEV cell design. A deciency of the SVM, however, is its linear treatment
of electrolyte transport. Figure B2 compares distributions of electrolyte-phase concen-
tration and potential between the two models at various times during the 4C discharge.
Model agreement is good as long as perturbations in electrolyte salt concentration are
small, e.g., for times of 50 s and less in Fig. B2. Later during the 4C discharge, the elec-
trolyte salt concentration gradient grows large enough that non-linear electrolyte trans-
port becomes important. At end-of-discharge, the model mismatch in the electrolyte
phase potential is 4 mV. Future extensions to the SVM are desired to better accommo-
date nonlinear electrolyte transport and electrolyte depletion.
References
1. M. Doyle, T. Fuller, and J. Newman, J. Electrochem. Soc., 140, 6 (1993).
2. G.-H. Kim, K. A. Smith, and A. Pesaran, Paper 262 presented at The Electrochemi-
cal Society Meeting Abstracts, Washington, DC, Oct 712, 2007.
3. G.-H. Kim and K. A. Smith, Paper 1295 presented at The Electrochemical Society
Meeting Abstracts, Honolulu, HI, Oct 1217, 2008.
4. G.-H. Kim and K. A. Smith, Paper 252 presented at The Electrochemical Society
Meeting Abstracts, San Francisco, CA, May 2429, 2009.
5. K. A. Smith and G.-H. Kim, Paper 255 presented at The Electrochemical Society
Meeting Abstracts, San Francisco, CA, May 2429, 2009.
6. K. A. Smith, G.-H. Kim, and A. Pesaran, Paper 247 presented at The Electrochemi-
cal Society Meeting Abstracts, Vancouver, Canada, April 29, 2010.
7. K.-J. Lee, G.-H. Kim, and K. A. Smith, Paper 1114 presented at The Electrochemi-
cal Society Meeting Abstracts, Las Vegas, NV, Oct. 14, 2010.
8. K. Kumaresan, G. Sikha, and R. E. White, J. Electrochem. Soc., 155, A164 (2008).
9. C.-Y. Wang and V. Srinivasan, J. Power Sources, 110, 364 (2002).
10. D. W. Dees, S. Kawauchi, D. P. Abraham, and J. Prakash, J. Power Sources, 189,
263 (2009).
11. P. Ramadass, P. M. Gomadam, R. White, and B. N. Popov, J. Electrochem. Soc.,
151, A196 (2004).
12. D. Bernardi, E. Pawlikowski, and J. Newman, J. Electrochem. Soc., 132, 5 (1985).
13. K. E. Thomas and J. Newman, J. Power Sources, 119121, 844 (2003).
14. K. E. Thomas and J. Newman, J. Electrochem. Soc., 150, A176 (2003).
15. V. Srinivasan and C.-Y. Wang, J. Electrochem. Soc., 150, A98 (2003).
16. Y.-H. Chen, G. Liu, X.-Y. Song, V. S. Battaglia, and A. M. Sastry, J. Electrochem.
Soc., 154, A978 (2007).
17. R. E. Garc a, Y.-M. Chiang, W. C. Carter, P. Limthongkul, and C. M. Bishop, J.
Electrochem. Soc., 152, A255 (2005).
18. K. A. Smith, C. D. Rahn, and C.-Y. Wang, Energy Convers. Manage., 48, 2565
(2007).
19. L. Cai and R. E. White, J. Electrochem. Soc., 156, A154 (2009).
20. M. Verbrugge and B. Koch, J. Electrochem. Soc., 153, A187 (2006).
21. U. S. Kim, C. B. Shin, and C.-S. Kim, J. Power Sources, 189, 841
(2009).
22. S. Santhanagopalan, Q. Guo, P. Ramadass, and R. E. White, J. Power Sources,
156, 620 (2006).
23. M. Safari, M. Morcrette, A. Teyssot, and C. Delacourt, J. Electrochem. Soc., 156,
A145 (2009).
24. P. Albertus, J. Christensen, and J. Newman, J. Electrochem. Soc., 156, A606
(2009).
25. D. W. Dees, E. Gunen, D. P. Abraham, A. N. Jansen, and J. Prakash, J. Electro-
chem. Soc., 155, A603 (2008).
26. V. Srinivasan and J. Newman, J. Electrochem. Soc., 151, A1530 (2004).
27. M. Doyle and Y. Fuentes, J. Electrochem. Soc., 150, A706 (2003).
28. K. A. Smith, C. D. Rahn, and C.-Y. Wang, ASME J. Dyn. Syst., Meas., Control,
130, 0110121 (2008).
29. G. F. Franklin, J. D. Powell, and M. Workman, in Digital Control of Dynamic Sys-
tems, 3rd ed., Addison-Wesley, Menlo Park, CA (1997).
30. T. Jacobsen and K. West, Electrochim. Acta, 40, 255 (1995).
31. I. J. Ong and J. Newman, J. Electrochem. Soc., 146, 4360 (1999).
Figure B1 (Color online) Comparison of constant current discharge voltage
predicted by reduced-order SVM (lines) and high-order COMSOL model
(symbols).
Figure B2 (Color online) Comparisonof electrolyte distributions predicted
by reduced-order SVM (lines) and high-order COMSOL model (symbols);
(a) electrolyte salt concentration, (b) electrolyte phase potential.
Journal of The Electrochemical Society, 158 (8) A955-A969 (2011) A969
Downloaded 29 Nov 2011 to 41.98.38.130. Redistribution subject to ECS license or copyright; see http://www.ecsdl.org/terms_use.jsp

You might also like