Você está na página 1de 11

Energy Conversion and Management 87 (2014) 472482

Contents lists available at ScienceDirect

Energy Conversion and Management


journal homepage: www.elsevier.com/locate/enconman

A rapid estimation and sensitivity analysis of parameters describing


the behavior of commercial Li-ion batteries including thermal analysis
Jorge Vazquez-Arenas a,, Leonardo E. Gimenez b, Michael Fowler b, Taeyoung Han c, Shih-ken Chen c
a

Departamento de Qumica, Universidad Autnoma Metropolitana, San Rafael Atlixco 186, Mxico, DF 09340, Mexico
Chemical Engineering Department, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada
c
General Motors Company R&D Center, 30500 Mound Rd, Warren, MI 48090, USA
b

a r t i c l e

i n f o

a b s t r a c t

Article history:
Received 13 March 2014
Accepted 24 June 2014

In this work, a methodology based on rigorous model tting and sensitivity analysis is presented to determine the parameters describing the physicochemical behavior of commercial pouch Li-ion batteries of
high-capacity (16 A h), utilized in electric vehicles. It is intended for a rapid estimation of the kinetic
and transport parameters, state of charge and health of a Li-ion battery when chemical information is
not available, or for a brand new system. A pseudo 2-D model comprised of different contributions
reported in the literature is utilized to describe the mass, charge and thermal balances of the cell and porous electrodes; and adapted to the battery chemistry under study. The sensitivity analysis of key model
parameters is conducted to determine condence intervals, using Analysis of Variance (ANOVA) for nonlinear models. Also individual multi-parametric sensitivity analysis is conducted to assess the impact of
the model parameters on battery voltage. The battery is comprised of multiple cells in parallel containing
carbon anodes and LiNi1/3Co1/3Mn1/3O2 (NMC) cathodes with maximum and cut-off voltages of 4.2 and
2.7 V, respectively. Mass and charge transfer limitations during the discharge/charge of the battery are
discussed as a function of State of Charge (SOC). A thermal analysis is also conducted to estimate the temperature rise on the surface of the battery. This modeling methodology can be extended to the analysis of
other chemistry types of Li-ion batteries, as well as the evaluation of other material phenomena including
capacity fade.
2014 Elsevier Ltd. All rights reserved.

Keywords:
Commercial Li-ion batteries
First-principle modeling
Parameters
Sensitivity analysis

1. Introduction
Li-ion batteries are a promising technology in electric vehicles
and other electronic devices; however, their future relies upon
their ability to meet the performance demands and low-cost
required in commercial applications. Many endeavors have been
undertaken to commercialize different types and chemistries of
Li-ion batteries. Their selection for a determined application
depends mainly on the chemistry of the cathode and other structural factors involved in the fabrication of the cells (e.g. density
of the material, porosity, particle size in the electrodes, cell geometry). These features signicantly affect the performance of the
battery including its average potential, reversible specic capacity
and volumetric energy density. In addition to this, the life-span,
aging mechanisms and safety of the batteries are major concerns
to extend their durability. Intensive research is ongoing to develop
Corresponding author. Tel.: +52 555 804 4600x2686.
E-mail
addresses:
(J. Vazquez-Arenas).

jgva@xanum.uam.mx,

http://dx.doi.org/10.1016/j.enconman.2014.06.076
0196-8904/ 2014 Elsevier Ltd. All rights reserved.

jorge_gva@hotmail.com

more precise methods to determine and analyze these characteristics to extend the durability of the battery and reduce the costs of
its fabrication.
To date, various chemistries have been considered for the fabrication of cathode materials for Li-ion batteries, with some of the
principal cathode materials being LiMn2O4 [14], LiCoO2 [2,4,5],
LiNi1/3Mn1/3Co1/3O2 (NMC) [69], LiFePO4 [1013]. The characteristics of these cathodes strongly depend on the nature of the material and are affected by their method of preparation [7].
Modeling provides the tools to perform a complex analysis of the
performance of Li-ion batteries and reduces the amount of time
utilized to evaluate their actual conditions, e.g. aging phenomena
[14,15]. Different models have been reported in the literature
describing Li-ion batteries subjected to different conditions and
chemistries [1,1633]. Although the basis to develop these models
is similar, e.g. electric and mass balances, porous electrode and
concentrated solution theories [1], a different chemistry of the system alters the main contributions affecting the response of the battery. For instance, it is well-known that LiFePO4 cathode materials
present the formation of two phases in the electrode unlike other

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

Li-ion battery chemistries [16]. Additionally, the variation of the


chemistry of the material composites involves a modication of
the crystallographic, chemical and electrical properties which lead
to different parameters describing the phenomena occurring
across the battery, e.g. conductivities, diffusivities, densities, kinetics. Consequently, the sum of all these contributions produces
changes in the overall performance of the battery. The identication of these contributions and their quantication (e.g. kinetics
parameters) are decisive to analyze the mechanisms operating in
the battery and detect major problems deteriorating the life of
the battery. However, these factors are not easy to perceive since
the battery is a complex system where different contributions
interact, and sometimes, chemical properties cannot be estimated
whatsoever.
The determination of kinetic parameters through tting methods and corresponding sensitivity analyses are powerful tools to
describe the complex behavior of a Li-ion battery. In addition,
these tools allow a quick analysis of the mechanisms operating
within the battery and can be used to simulate non-accessible
experimental conditions. The rst one involves the determination
of the parameters and constants through the minimization of the
error in the potential of the battery at different experimental C-rate
tests, whereas the sensitivity analysis determines the importance
of the parameters or contributions under the experimental conditions of the analysis.
Although a considerable amount of models have been reported
to describe the behavior of Li-ion batteries [1,1620], most of them
have mainly focused in the analysis of button cells, which could
present a slightly different behavior compared to actual commercial batteries (due to vast differences in cell geometry, thermal
conditions, and macroscopic mass transfer regimes). Just a few
studies have undertaken the physicochemical modeling of commercial cells or batteries [16,34,35].
Many modern publications dealing with physicochemical modeling, battery SOC and performance estimation provide lumped
parameter or reduced order models [3641] for faster computation, based on existing physicochemical models. These studies
often report model parameters that are either taken from previous
literature, estimated-through tting experimental data or otherwise, and/or determined experimentally; battery manufacturers
are not always willing to provide parameters that may compromise proprietary technology or information.
With the current rate of advancement of battery materials and
technology it may be inappropriate to use parameters found in literature, particularly because the materials used in the literature
and those in the battery application may possess slight variations
signicantly affecting the properties in question. Experimental
determination methods are time consuming, may require expensive equipment and may involve cell disassembly or handling
which may change the properties of the otherwise intact battery
materials. In addition, there are some battery parameters, such as
effective electrode porosity, which are difcult to measure experimentally. Hence, estimating all or most of the kinetic parameters
through tting appropriate physicochemical models remains the
most viable option- it is relatively fast, inexpensive, and only
requires a few cycles (unless parameters related to aging are being
determined). This method is becoming faster and more technologically feasible in recent years thanks to modern computational
advances. There are few publications which estimate kinetic
parameters for full physicochemical models through tting methods with experimental validation [4244].
Whenever tting or estimation methods are used for determining parameters in physicochemical models, the sensitivity and
accuracy of the model parameters is rarely considered, so long as
the model succeeds in providing estimation of operating voltage,
capacity, and temperature. Thus, it is unclear which parameters

473

are critical for successful model predictions (and from the manufacturers perspective, for optimizing battery performance) and
which can vary within a certain range without signicant impact
on model (or battery) performance. Examples of this model based
optimization from a manufacturers perspective is the comparison
of critical design parameters for a cathode through model based
sensitivity analysis [45], the maximization of electrode energy
density through many parameter simultaneous optimization [46],
and effects of manufacturing variations on cell performance and
aging [47].
In some of the aforementioned studies [42,44] there is often no
indication to the tting method or estimation technique used in
parameter determination, simply an indication that the parameter
was somehow estimated. The goal of this study is to provide a
method of kinetic parameter estimation given a previously developed physicochemical model [17], as well as experimental validation and sensitivity analyses on the parameters in order to
identify critical model parameters. A previous work conducted by
our research group proposed a model to account comprehensively
for the behavior of a LixC6LiyMn2O4 cell to understand their performance at both beginning of life (BOL) and end of life (EOL) [17].
Comparisons between baseline and complex models were systematically utilized to analyze different thermal and capacity fade
effects (e.g. heat generation, SEI formation, dissolution of LiyMn2O4
particles) during typical cycle-life tests. However, the previous
study did not include any experimental work to validate the contributions of the model. To the authors knowledge, the determination
of kinetic parameters and sensitivity analysis for commercial
Graphite/LiNi1/3Co1/3Mn1/3O2 batteries have not been conducted
either. Their importance is crucial to identify the mechanisms controlling the behavior of this type of batteries, design, optimization,
quantication of their rates allowing the prompt detection of problems to extend their life and reduce the costs of their fabrication.
Thus, the present study focuses on the analysis of the kinetic parameters and mechanisms controlling the behavior of a Li-ion battery at
BOL, utilizing real data collected from commercial 16 A h Kokam
batteries [48] at different C-rate tests. Rather than reporting a
new physicochemical model, particular focus is placed on more rigorous least-square tting to obtain the system parameters and a statistical analysis of the t of the model to the experimental data to
estimate the condence intervals of the parameters. A systematic
analysis is also proposed, where individual interactions are incorporated to a baseline model and subsequently evaluated based on its
signicance.
A thermal analysis is also conducted to account for the temperature rise on the surface of the battery. There are a wide variety of
models that accurately describe the thermal prole of many batteries during cycling [20,22,36,42,4952]. Further studies will be
aimed to corroborate the magnitude of the parameters using additional chemical and electrochemical measurements, as well as
evaluating aging mechanisms and more comprehensive thermal
distributions for these commercial batteries.

2. Materials and methods


2.1. Modeling
Table 1 shows the thermal model utilized in this study to account
for the response of the batteries under different C-rates. It describes
the diffusion (e.g. porous electrode theory) and conduction of Li+
ions with conservation of charge (e.g. Ohms law) in the solid and
electrolyte phases across the cell, e.g. anode, separator and cathode.
Further details of the derivation of the material and charge balances
have been described in Refs. [1,17,19,35,53]. The isothermal studies
do not consider the energy equations shown in Table 1.

474

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

Table 1
Domain equations, initial and boundary conditions involved in the lithium-ion battery model.
Region of the
cell

Balance

Governing equations

Boundary or initial condition

Anode

Material, solid
phase

@cs;n
@t

cs,n|t=0 = cn,ini

@
Ds;n r12 @r
r2

@cs;n
@r

Ds;n

U1;n 

Charge, solid phase

reff;n @ @xU21;n an Fjn

Charge, liquid phase

@
jeff;n
 @x

Material, liquid
phase

0
@
@c
ee;n @c
@t @x Deff ;n @x 1  t an jn

Energy

@ U2;n
@x

2RT1t 0 @
@ ln c
@x eff;n @x
F

an Fjn

@cs;n 
@r r0

Ds;n

@cs;n 
@r rRp;n

0
0
xA
xA=S

@ U2;n 
jeff;n @x 
0
 xA

@ U2;n 
@U 
jeff ;s @x2;s 
jeff ;n @x 
xA=S

@
@x
kn @T
@x U1;n
@ U1;n 2
@ U2;n 2
eff ;n @x eff ;n @x

r
Separator

Cathode

xA=S

cjt0 c0


Deff ;n @c
@x xA 0

@T
n C p @t

 U2;n  U n T

jn


@U 
reff;n @x1;n 

2RT1t 0
F

@U n
@T an Fjn

T n jt0



@c 
@c 
Deff ;n @x
Deff ;s @x
xA=S
xA=S



@T 

hT  T env kn @x xA=S ks @T
@x xA=S


T env kn @T
@x xA

ln c @ U2;n
jeff ;n @ @x
@x

@ U2;n 
@x xA=S

@ U2;s 
@ U2;s 
@x xA=S  eff;s @x xS=C

jeff;s

Material, liquid
phase

@
@c
e @c
@t @x Deff;s @x

cjt0 c0

Energy

2RT1t
@ U2;s
@
@T
ln c @ U2;s
qs C p @T
jeff;s @ @x
F
@t @x ks @x jeff;s @x
@x



@c 
@c
Deff;n @x
Deff;s @x
xA=S
xA=S


@c 

Deff;s @c
@x xS=C Deff;p @x xS=C




@T 
@T 
@T 

T s jt0 T env kn @T
@x xA=S ks @x xA=S ks @x xS=C kp @x xS=C

Material, solid
phase
Charge, solid phase

@cs;p
@t


cs;p t0 cp;ini

@ U2;s
@x

2RT1t 0
F

ln c
jeff;s @ @x
0

jeff;n

@
Ds;p r12 @r
r2

@ 2 U1;p
eff;p @x2

@cs;p
@r

Ds;p

@ U1;p 
eff;p @x 
xS=C

r

ap Fjp

jeff;s

@cs;p 
@r r0

Ds;p


@ U1;p 
eff;p @x 
xC

r

@cs;p 
@r rRp;p

jeff;p

@ U2;p 
@x xS=C

Charge, liquid phase

jp

Iap

U1;p jxC Ecell






@U 
@U 
jeff;s @x2;s 
jeff;p @x2;p 
xS=C
xS=C



@c
@c
@c 
Deff;p @x
0 Deff;s @x
Deff;p @x
xC
xS=C
xS=C

@ U2;p 
@x xC

Charge, liquid phase

@
jeff;p
 @x

Material, liquid
phase
Energy

0
@
@c
ee;p @c
@t @x Deff;p @x 1  t ap jp

cjt0 c0

@U p
@
@T
qp C p @T
@t @x kp @x U1;p  U2;p  U p T @T ap Fjp




@T 

T p t0 T env ks @T
@x xS=C kp @x 

@ U2;p
@x

@ U1;p 2
eff;p @x

2RT1t 0 @
@ ln c
F
@x eff;p @x

@ U2;p 2
eff;p @x

2RT1t 0
F

ap Fjp

m X
n
X
2
Emodel
 Eexperimental

cell
cell
1

xS=C



kp @T
@x xC hT env  T

@ ln c @ U2;p
eff;p @x
@x

The model was t to the experimental data of cell potential


(Ecell, refer to Fig. 5) recorded for the different C-rate plots to obtain
parameter estimates by minimization of the sum-of-squares error
between the model predictions (Table 1) and data. This involved
the use of the tness function below in conjunction with a trustregion-reective algorithm provided by the Matlab R2011b toolbox [54]:

Fitness function

jeff;p

where Emodel
and Eexperimental
are the model-predicted and expericell
cell
mental cell voltages in the discharge plots (i.e. Fig. 5), respectively;
n is the number of points recorded per C-rate, and m the corresponding C-rate plot (e.g. 1C, C/2, C/5 and C/25). Charge and material balances in the solid-phase and liquid-phase as well as the
kinetic contributions were considered one at a time to t the model
parameters associated with them to the four different discharge
plots, and then their importance was determined through a sensitivity analysis shown in Fig. 4 (see details below). The individual tting of the charge balance in the solid phase, charge balance in the
liquid phase, material balance in the solid phase, material balance
in the liquid phase and kinetics involved 2, 1, 4, 2 and 9 parameters
respectively. The tting stage was completed by performing a global t involving sensitive and non-sensitive parameters, including
18 parameters from all balances and kinetics. This procedure was
carried out with the intention of observing the variability of nonsensitive parameters when sensitive parameters of other balances
are subjected to variations. The parameters determined from this
last stage and non-tted parameters are reported in Table 2. The

condence intervals of these parameters were rigorously calculated


using the statistical analysis reported in Ref. [55] for non-linear
models. This procedure involved the calculations of the covariance
matrix and the analysis of variance (ANOVA) through the residuals
vector and the Jacobian matrix yielded from the output of the trustregion-reective algorithm from Matlab [54]. Condence intervals
for insensitive parameters to the model were not determined since
they were very large (specied in Table 2), and represent non-accurate estimations within the experimental conditions of analysis of
the batteries. However, they are considered in the model to account
for steps which are not rate-controlling. As shown in the properties
and parameters utilized in the lithium-ion battery model (Table 1),
the model is able to describe the material and charge transport in
the liquid and solid phases of the battery. These mechanisms and
their interactions are well-known to occur during the operation of
a Li-ion battery [56,57]. However, this does not imply that all of
them are rate controlling during the discharge/charge of the battery, but they are necessary to construct the physics of the battery.
The rate controlling steps can vary with temperature, material
chemistry, discharge/charge rates and state of charge. Their signicance is statistically estimated in this study. The model parameters
determined by the global t were independently studied using a
Multi-Parametric Sensitivity Analysis (MPSA) in order to determine
their relative importance in the model, and corroborate the calculations established by the condence intervals. This test was conducted according to the owchart described in Fig. 4. MPSA
measures the variation of the model prediction with respect to
the experiment when the parameters are modied via a normal distribution with a variation range of 10%, situated according to values determined by the tting. The coefcients (ddev) determined

475

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

Table 2
Properties and parameters utilized in the lithium-ion battery model determined through tting or from the literature. ddev is the coefcient representing the degree (low,
intermediate and high) of sensitivity of the parameter to the model, determined through a Multi-Parametric Sensitivity Analysis.
Description

Anode

Separator

Maximum concentration in intercalation materialctn = 26,390 mol m3 [1]

ctp 53,284 mol m3

Initial state of charge


Solid phase Li-diffusivity in the particles

SOC n;ini 0:66


Ds;n 3:9  1014 m2 s1 [1]

Volume fraction of electrolyte phase

ee,n = 0.303a
ddev = low
es;n 1  ee;n  ep;n  ef ;n = 0.200 0.001
ddev = high

SOC p;ini 0:07


Ds;p 1.64  1014 m2 s1a
ddev = low
ee;p 0:127a
ddev = low
es;p 1  ee;p  ep;p  ef ;p 0.338 0.005
ddev = high

Volume fraction of solid active material phase


Separator porosity
Specic surface area
Salt diffusivity in the electrolyte

e = 1 [1]
an

3es;n
rp;n

= 8.153  105 m2
54
0:22x103 c
T2295x103 c

4:43

ap
2

Ionic conductivity of the electrolyte

je = (10.5c + 0.668  103c2 + 0.494  106c3 +

Intercalation/deintercalation rate constant


Anodic transfer coefcient for lithiation
Cathodic transfer coefcient for lithiation
Intercalation/deintercalation current density

3es;p
r p;p

= 1.236  105 m2

[63]

r0,p = 0.023 S m1 [64]


reff;p r0;p e1:5
s;p

0.074cT  1.78  105c2T  8.86  1010c3T  6.96 


105cT2 + 2.80  108c2T2)2  104 S m1 [63]
jeff,s = jeee,n
jeff,s = jees

jeff,p = jeee,p

t0 0:57a
ddev = low
kn = 3.67  106 m2.5 mol0.5 s1a
ddev = low
a_Aneg = 0.5a
ddev = low
a_Cneg = 0.5a
ddev = low


a Aneg

cs;n 
jint=dein kn ct  cs;n 
n

ca
Resistance of SEI formation

m s

1

Electronic conductivity of the solid phase


Effective conductivity of the solid phase

Effective ionic conductivity


Li transference number

x10

4

De 10
2 1
Deff De;n e1:5
e m s
r0,n = 100 S m1 [1]
reff;n r0;n e1:5
s;n

Effective salt diffusivity in the electrolyte

Cathode

c0 = 2000 mol m3


[1]

Initial electrolyte salt concentration

Aneg

rRp;n

a Cneg

rRp;n

0:5F
exp0:5F
RT gn  exp RT gn

kp = 1.30  106 m2.5 mol0.5 s1


ddev = low
a_Apos = 0.36 0.035
ddev = intermediate
a_Cpos = 0.46 0.048
ddev = intermediate


a Apos
a
jpint=dein kp ctp  cs;p rRp;p
cs;n rRp;p




a Apos
0:5F
0:5F
c
exp RT gp  exp RT gp

Cpos

RSEI = 0.035 X m2a


ddev = low

The condence interval was not determined since the parameter had a low sensitivity to the model.

from the sensitivity analysis are also included in Table 2, and denote
the degree (low, intermediate and high) of sensitivity of the parameter to the model. Further details of the MPSA can be consulted in
Ref. [55].
2.2. Experimental set-up
Four different discharge plots (C/25, C/5, C/2 and 1C) were collected utilizing commercial Kokam batteries (Model SLPB
75106205) with a rated capacity of 16 A h. Note that the maximum continuous discharge rate allowed for the operation of these
batteries is 1C (16 A). The maximum pulse discharge rate is 5C
(80 A), but it can be sustained only during a 10 s pulse. Presumably, the rate limitation for these batteries is associated with the
polymer-gel electrolyte since the application of high C-rates leads
to a signicant temperature rise producing the eventual degradation and capacity fade of the batteries. The cells of these batteries
are comprised of a certain number of mesocarbon microbeads
(MCMB) anodes, separators and NMC cathodes. Other cell geometry details are not revealed in brevity to maintain manufacture
condentiality. Cell specications of these batteries can be seen
in reference [58]. The discharge proles were collected using a
Maccor battery cycler utilizing the following protocol: (a) The
battery was rested for at least ve hours, (b) then the battery
was charged at a constant current of 16 A followed by a constant
voltage charging stage at 4.2 V until a cut-off current of 0.8 A, (c) a
rest period was applied to attain a constant open circuit voltage,
(d) followed by the constant current discharge of the battery at
16 A until a cut-off voltage of 2.7 V, (e) the battery was newly

rested until constant voltage, (f) proceeding the charge of the battery at the C-rate of interest to 4.2 V, then constant voltage charging at 4.2 V until a cut-off current of 0.8 A (g) rest until constant
voltage, (h) discharge of the battery at C-rate of interest to cutoff voltage of 2.7 V. Stages (a) to (e) correspond to a pre-test
chargedischarge cycle in order to attain the same state of charge
(SOC) in the batteries before running the C-rate tests and to verify
that the performance of the battery is consistent with previous
cycles in terms of capacity at 1 C.
During the tests, the batteries were placed in a Cincinnati SubZero MicroClimate thermal chamber with the temperature controlled by forced air convection at 25 0.2 C. Within the
chamber, the batteries were fastened in a jig that consists of two
separate aluminum plates, each one encased in a larger acrylic
plate which can be fastened to the other with screws (see Fig. 1).
The jig was built in order to prevent convective cooling of the battery faces within the chamber; the aluminum plates aid in the heat
distribution and compression of the batteries (representative of
pack conditions) and allow the measurement of heat ux through
the jig. Eight T-type thermocouples were taped in different locations on the jig, four were placed approximately 1 cm below each
of the two tabs on both faces (at these locations, the highest surface
battery temperature was observed), and four directly across the
enclosing aluminum plate on each outer side (to estimate onedimensional heat ux). Fig. 1 indicates the placement of four of
these thermocouples (the other four are symmetrically placed on
the reverse side of the battery). The temperature of all 8 thermocouples was read every second by using a FieldPoint System from
National Instruments controlled using LabVIEW software.

476

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

Fig. 1. Position of the thermocouples. (a) On the aluminum plate once the jig is closed, (b) on the battery face.

3. Results
3.1. Isothermal studies
The rst simulations were conducted utilizing the parameters
reported in reference [1] for a LixC6LiyMn2O4 cell (e.g. previous
to the application of tting methods), with the intention to present
the original deviations between the model and the four experimental C-rate tests. Fig. 2 shows these results where the poor quality of
the cell voltage (Ecell) prediction is clearly evidenced. This is not
surprising, given the variation existing between the battery components (e.g. cathode material) and the magnitude of the phenomena operating within it. The next step involved the determination
of the kinetic parameters, which was sequentially carried out following the stages described in the owchart shown in Fig. 3.
Fig. 5 shows the experimental and modeling cell voltage as a function of the depleted capacity at four different C-rates: C/25, C/5, C/2
and 1C. As observed in this plot, the quality of the ts is better than
the one shown in Fig. 2. This indicates that even when the model
describes properly the physicochemical contributions of the system, parameters and constants must be appropriately selected
and tted to describe the magnitude of these phenomena in the
model.
In order to reduce the uncertainty of some of the constants and
parameters utilized by our model described in the plots shown in
Fig. 5, the research program has tried to measure them through
data contained in the literature or by using our own experimental
measurements whenever possible. One of these variables is the
open circuit voltage (OCV) of electrode reaction depending on
the local state of charge (h) at a reference temperature (Uj,ref).

Equations describing the OCV in each electrode are difcult to


derive due to the dependence of this value upon SOC. Thus, in
the present work these expressions were empirically estimated
for each electrode by tting their cell voltage as a function of
SOC, from discharge plots conducted at C/25 (not shown). The
expression for the cathode material (NMC) was estimated to be:
U p;ref 0:48 3:44  106  tanh1:29  1016
 4:63  1016  SOCp  1:67  1014


1:12  106

 exp1:03  104  10:57 SOCp

Baseline model
Charge balance in the solid-phase
Are the parameters

sensitive?
NO

YES

Charge balance in the liquid-phase


Are the parameters

sensitive?
NO

YES

Material balance in the solid-phase


sensitive?
NO

YES

Least-square fit to
determine parameters
Incorporation of the
parameters to the
model

Material balance in the liquid-phase

Ecell/ V

0:56

3:04  106  SOCp

1:03  103
 1:70  106  exp0:036  SOC0:0096
p

Are the parameters

4.5

Are the parameters

sensitive?
NO

YES

3.5

Kinetics
3

Are the parameters

NO

YES

1C

2.5

sensitive?

C/2
C/5
C/25

10

15

Capacity/ A h
Fig. 2. Computed and experimental cell voltages as a function of capacity for
Kokam batteries characterized at four different discharge rates. Symbols describe
the experiments and continuous lines represent the simulations calculated using
the parameters reported in reference [1].

Overall
Fitting
Set of
Kinetic Parameters
Fig. 3. Flowchart used to determine the parameters of the model accounting for the
behavior of the Kokam batteries.

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

477

while the OCV of the anode material (graphite) is described by the


following expression determined from our battery:

U n;ref 0:028  6:33  exp3:52  SOCn


9:67  103  exp0:87  SOCn 9:64  103
 exp1:03  103  SOC n 0:001  exp0:001  SOCn 3
The local states of charge are respectively dened for the cathode and anode materials as:

Fig. 4. Flowchart of the Multi-Parametric Sensitivity Analysis (MPSA) used to


estimate the sensitivity of the parameters (Table 2) to the model.

4.5

Ecell/ V

3.5

3
1C

2.5

C/2
C/5
C/25

10

15

Capacity/ A h
Fig. 5. Computed and experimental cell voltages as a function of capacity for
Kokam batteries characterized at four different discharge rates. Symbols describe
the experiments and continuous lines represent the simulations calculated using
the parameters reported in Table 2.

SOCn

cs;n
ctn

SOCp

cs;p
ctp

These values are difcult to measure accurately at fully charged


state since there is a loss of cyclable lithium during the manufacture of the batteries to form the solid electrolyte interface (SEI)
[18]. Therefore, they were determined from tting initial experimental discharge curves to yield the values shown in Table 2.
These values are close to those reported by other researchers for
LixC6LiyMn2O4 cells (0.53 and 0.17). The maximum concentration
in the electrodes was determined using the density and the molecular weight of the material. Values of 26,390 and 30,555 mol m3
Li+ have been reported by Doyle et al. [1] and Ning et al. [18],
respectively for the anode material. These values are similar to
some extent for most of the batteries since most of the Li-ion cells
contain mesophase microbeads of carbon (LixC6). Values of
53284 mol m3 were estimated for the maximum concentration
in the cathode utilizing the density of the NMC material
(4.77  106 g m3) [59] and its molecular weight (89.52 g mol1).
A solid state diffusion coefcient of 3.9  1014 m2 s1 (Ds,n) has
been reported by Doyle et al. for carbon electrodes [1], whereas a
value of 1.64  1014 m2 s1 (Ds,p) was found for the cathode
through tting. Ds,p differs by almost one order of magnitude
(Ds;Liy Mn2 O4 = 1  1013 m2 s1) for LiyMn2O4 [1], while diffusion
inside LiFePO4 particles has been shown to be even slower than
in NMC cathodes, e.g. 1.18  1018 m2 s1 [16]. These differences
are a clear evidence of the modication in the transport of Li+ ions
inside the electrode particles due to changes in the crystal structure of the cathode. It is known that Ds,p relies on different properties of the cathode including electrode composition, porosity,
morphology, etc. Likewise, it is a strong function of the techniques
utilized to fabricate the cathode and used to determine its value
(e.g. different time domain conditions). Ds,p values in the order of
1014 m2 s1 have been reported for LiNi1/3Mn1/3Co1/3O2 electrodes
fabricated by wet chemical methods and characterized using electrochemical impedance spectroscopy (EIS) [60], 10141015 m2 s1
for similar electrodes synthesized at high temperature via
precursor methods and characterized using CV [61], 1015,
1014 m2 s1 and 1015 m2 s1 for LiNi0.36Mn0.29Co0.35O2 electrodes
fabricated using co-precipitation methods and characterized
through CV, GITT and PITT, respectively [62]. However, despite
this variation Ds,p is averagely found within the range of
1014 m2 s1 as the value determined by our tting (Table 2), and
3.3  1014 m2 s1 via potentiostatic intermittent titration technique (PITT) (details not shown) as part of a related study. The
low sensitivity calculated for this parameter denotes its poor inuence in the model. Although, this effect is evaluated for the overall
behavior of the battery voltage, a signicant behavior will be
observed at certain SOC values (i.e. depending on capacity) for
the diffusive processes occurring in the battery (e.g. inside the particle electrodes). This effect is further discussed below.
The volume fractions for the electrolyte phase, the current conductive llers and the polymer phase (e.g. NMC contains a polymer
electrolyte) for the cathode and anode materials are also reported

478

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

in Table 2. The volume fractions of the polymer phase and the conductive llers for NMC electrodes were not revealed by the manufacturer. In this work, all the terms were incorporated into one
term (e.g. volume fraction of solid active material, es) in order to
simplify this parameter in the model. Sensitivity analyses conducted for the volume fractions reveal that the solid phases present
a considerably higher sensitivity compared to the electrolyte
phases, which are insensitive in the range of experimental conditions evaluated. Apparently, this behavior is observed since the
mechanisms controlling the behavior of the battery are associated
with the solid phase (i.e. electronic conduction) and not with the
electrolyte inputs (i.e. mass-transfer in the liquid phase). Moreover, this high sensitivity could also be connected with the importance of the active surface area and the particle size of the
materials. The particle radii of the electrodes were found in the
range of 106 m, which are within the typical range of particles utilized for the fabrication of Li-ion batteries. These parameters were
found to be non-sensitive in the model. As aforementioned, this is
most likely due to a statistical correlation between the volume
fraction of the solid phase and the active surface area of the electrodes (which is sensitive within the model), i.e. ai = 3es,i/rp,i.
The expressions describing the salt diffusivity (De) and the ionic
conductivity of the electrolyte (je) as a function of the Li+ concentration and temperature across the cell were taken from the work
reported by Valoen and Reimers [63]. These values were obtained
for LiPF6 in a propylene carbonate/ethylene carbonate/dimethyl/
carbonate mixture, and were estimated as a function of temperature and LiPF6 concentration. The electronic conductivity of the
carbon has been typically reported to be 100 S m1 (r0,n), whereas
for NMC materials the conductivity was found for each of the following compositions: LiNi0.475Co0.05Mn0.47O2 (0.023 S m1) [64]
and LiNi0.4Co0.4Mn0.2O2 (0.0140.068 S m1 from 21 to 100 C)
[65]. These values should be close to the electronic conductivity
of the LiNi1/3Mn1/3Co1/3O2 electrode used in the batteries tested
in this work. Likewise, it is known that r0,p is lower than the LiCoO2
material (10 S m1) but higher than the LiFePO4 material
(0.005 S m1) [19].
The Li transference number in LiPF6 for our batteries was estimated to be 0.57 from the tting of the parameter. This value is
close to that one (0.363) reported for a concentration of 1.2 M LiPF6
in lithium polymer cells [66] and was found to be non-sensitive in
the model. On the other hand, the intercalation/deintercalation
rate constants were found to be in the order of 106 m2.5 mol0.5
s1. These values are similar to those reported for other chemistries, LiMn2O4 [1], LiCoO2 [18]. In this regard, the values of the

transfer coefcients in the cathode vary from the typical values


(0.5) reported for other positive electrodes (values shown in
Table 2), suggesting kinetic changes for the intercalation of Li+ ions
for NMC materials. This nding is corroborated by the sensitivity
analysis revealing that the transfer coefcients in the cathode are
sensitive in the model. This conrms that the behavior of the battery can be also controlled by the insertion process in one of the
electrodes, and not only by electrical conduction or mass-transfer
phenomena. A different situation occurs for the transfer coefcients of the negative electrode, which are insensitive in the model.
Although, it could be suggested that this occurs due to the fact that
one cannot assess the signicance of these values in the anode
independently during the discharge of the battery. More experimental evidence is required to analyze this effect, particularly
the engagement of charge proles which will be evaluated further
in this paper. Note that the transfer coefcients are more sensitive
in the model than the rate constants. A similar nding was
reported above, and it has been particularly reported for other
electrochemical systems. This situation can be explained in terms
of the correlation existing between the transfer coefcients and
the rate constants, given the exponential form of the ButlerVolmer equation [67,68].
The resistance due to the SEI formation at the anode side was
calculated to be 0.035 X m2, which is found within the range for
carbon electrodes [1,18]. In the present study, it is assumed that
the commercial Kokam batteries utilized in the experimental
characterization were initially cycled to form a pseudo steadystate SEI during their manufacture with the intention to avoid
the loss of Li+ ions from the electrolyte during subsequent operation. This agrees with reports in literature where the SEI formation
is considered within the rst one or two charge/discharge cycles
[69,70]. It is also known that for some particular anodes, this lm
continues growing but its effects of the additional growth can be
considered negligible [71], unless it undergoes major damage or
long-term degradation which requires a signicant loss of Li+ ions.
As observed in Table 2, RSEI was found to be non-sensitive in the
model, perhaps because its effects are known to occur over longterm cycling or at the end of life of the battery [17,18] and not in
the range of testing life undertaken in this work.
The capabilities of a model describing the physics of a Li-ion
battery can be evaluated considering discharge and charge proles
at different C-rates. To date, only discharge proles have been considered in this work. In order to assess the capabilities of prediction
of the model and the magnitudes of the parameters and constants
(refer to Table 2) determined through tting the discharge plots,

4.5
1C

2120

C/25
C/2 C/5

0.11
1.56
3.11

2080

4.67
6.22

c / mol m-3

Ecell/ V

7.78

3.5

9.33

2040

10.89
12.44
14
15.55

2000

15.96 A h

2.5

1960
2

10

15

Capacity/ A h

CC/A
Fig. 6. Computed and experimental cell voltages as a function of capacity for
Kokam batteries characterized at four different charge rates. Symbols describe the
experiments and continuous lines represent the simulations calculated using the
parameters reported in Table 2.

Separator

Anode

1920

Cathode

x / dimensionless

C/CC

Fig. 7. Concentration of Li+ in the electrolyte (c) across the battery plotted at
different capacity values. CC/A and C/CC represent the interfaces located in the
current collectors between the anode and the cathode, respectively. 1C test.

479

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

2120

(a)

c|x=cc/A / mol m-3

2080

2040

2000

1960

10

15

20

15

20

Capacity/ A h
2040

c|x=C/cc / mol m-3

charge proles were simulated at four different C-rates: C/25, C/5,


C/2 and 1C, maintaining all the parameters described in Table 2
constant, only the initial concentration in negative and positive
electrodes were varied to estimate SOCn,ini and SOCp,ini. The potential of the cell as a function of depleted capacity is shown in Fig. 6
for the charge proles conducted at different C-rates. As observed,
the quality of the simulation is good considering that the kinetic
parameters determined from the tting to the discharge proles
were used for this purpose. Only small deviations are observed at
the beginning of the charging period for the simulation carried
out at 1C. This may be due, at least in part, to the heat generated
by the battery at 1C (which is more signicant than at lower Crates), as well as a higher polarization at faster rates. As previously
mentioned in the discussion of the discharge proles, the transfer
coefcients of the negative electrode were insensitive during the
sensitivity analysis conducted for the discharge proles. A similar
situation occurs in the case of the charge proles shown in Fig. 6,
where it was found that they were not sensitive either, whereas
the transfer coefcients of the positive electrode are sensitive. This
occurs even if the transfer coefcients are allowed to vary during
the tting. This nding suggests that there is a kinetic control in
the positive electrode at determined moment of the discharge
and charge of the battery. In order to get further insights of the
mechanisms controlling the operation of the commercial batteries,
some variables associated with the parameters were calculated.
Calculations of variables such as the concentration of the Li+
across the battery and inside the electrodes, and the electric potential offer further evidence of the phenomena controlling the behavior of the battery. From time to time, some of these phenomena
cannot be measured experimentally and therefore, numerical evaluations need to be performed. Additionally, numerical estimations
can provide the evolution of these variables as a function of time or
battery capacity, which allows determining critical points of the
cycling of the battery. Simulations for different variables were conducted at 1 C-discharging at different locations of the cell. Fig. 7
shows the Li+ concentration in the electrolyte (c) across the cell
at different depleted capacities (refer to Fig. 5 for cell voltage values). Similar proles have been reported by Doyle et al. for cells
containing LiMn2O4 cathodes [1]. As observed, the concentration
proles always drop in the direction of the cathode, since this is
the ow of the Li+ concentration during the discharge. Note that
during the rst 25 s (Q = 0.11 A h), the concentration has the largest increase among the other proles shown in the Figure. This
effect occurs as a result of the kinetic control that operates in the
battery during the rst seconds of discharging. This phenomenon
is more clear to observe if the Li+ concentration is plotted as a function of time in the interfaces located between the current collector/
anode (Fig. 8a) and cathode/current collector (Fig. 8b). Fig. 8a
shows that there is a rise in c (i.e. concentration of Li+ in the electrolyte) produced by the deinsertion of Li+ ions from the anode and
their transference to the electrolyte. Note that the concentration
plotted in Fig. 8 is the concentration of Li+ in the electrolyte and
not in the solid phase, which presents a different behavior. On
the other hand, the initial decay observed in Fig. 8b results from
the concentration available to be inserted into the cathode. The
concentration is higher at the beginning of the test since there
are no mass-transfer limitations in the electrolyte, but once this
occurs the concentration drops to a steady-state value. The same
behavior is observed for the concentrations shown in Figs. 7 and
8a. Apparently, this pseudo-steady state prole is a consequence
of the time constant domain for the diffusion in the battery, which
is long enough to attain this steady state condition [1]. After this
stage, the Li+ concentration proles shown in Fig. 7, particularly
in the anode, show again a rise around 2100 s (Q = 9.93 A h). In
general, this rise is not presented for low-capacity cells or batteries
and could be associated with a faster deinsertion in the anode

(b)

2000

1960

1920

10

Capacity/ A h
Fig. 8. Concentration of Li+ in the electrolyte as a function of the capacity. (a)
Current collector/Anode interface and (b) Cathode/Current collector interface. 1C
test.

compared to the insertion in the cathode or heating in the interior


electrodes. As observed, in Fig. 8b, after the pseudo-steady regime
of concentration in the cathode, the concentration remains around
the value obtained for this previous stage, which suggests that it is
not consumed fast enough as it arrives to the surface of the cathode
particles, where it builds up in this section of the battery. Note in
Table 2 that the maximum concentration in intercalation material
is higher for the cathode than the anode, thus, there are no limitations regarding the number of active sites available for lithiation.
This phenomenon conrms the nding reported in the sensitivity
analysis, where the kinetics of the cathode controlled the lithiation
process. Thus, more Li+ ions cannot be transferred from the anode
to the cathode since they are inserted more slowly, whereby ions
are accumulated in the electrolyte as observed in 8a for the range
of depleted state of charge located between 9.33 and 12.44. At the
end of discharge, a drop in Li+ concentration is observed (Fig. 8a) at
the current collector/anode interface as a result of the concentration gradient created inside the anode particles and the moving
front of the Li+ insertion in the cathode. Likewise, these effects produce an increase in the ohmic drop, reected in the potential drop
in the electrolyte (/2) at the end of discharge (not shown).
To date, this work has only analyzed the phenomena occurring
in the electrolyte phase across the cell. Valuable information can
also be collected from the analysis of the solid-phase in the electrodes since this permits to determine the phenomena governing
the behavior inside the particles, as well as nding out the most
reactive zones of the electrodes. Fig. 9 shows the Li+ concentration
at the surface of particles as a function of the length of the anode
(Fig. 9a) and the cathode (Fig. 9b). As observed in Fig. 9a at low

480

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

depleted capacities (i.e. low depth of discharge (DOD)), most of the


concentration drop occurs in the region close to the separator
(Length = 1) as a result of the utilization of the anode mainly at this
zone of the electrode [1]. At intermediate depleted capacity values,
other portions of the electrode are utilized as gradients of concentration are also generated in the electrolyte. At the end of discharge, the concentration proles in the anode present a steadystate behavior as a result of the diffusion control dictating the
mass-transfer across the particles. On the other hand, Fig. 9b shows
that the concentration proles in the surface of the cathode particle do not develop any drop as a function of the length of the electrode. This again conrms the kinetic limitations present in the
cathode particles to insert Li+ ions.
3.2. Thermal analysis
The role of temperature is critical on the performance and
safety of the batteries. Particularly for automotive applications,
information concerning the thermal distribution across the cell is
necessary to identify local spots that cause thermal overheating,
and may lead to battery failure or permanent material degradation.
More importantly, the performance and aging of the battery are
signicantly affected by temperature. In order to provide maximum range for as long as possible the battery should be cycled
close to room temperature. Accordingly, one of the motivations
of this work is to explore through experiments and model

(a)

20000

0.11
6.22
12.44

1.56
7.78
14

3.11
9.33
15.55

4.67
10.89
15.96 A h

cs / mol m-3

16000

12000

8000

4000

0.5

CC/A

(b)

32000

Length of Anode / dimensionless


0.11
6.22
12.44

1.56
7.78
14

3.11
9.33
15.55

A/S

development, the temperature rise produced on the surface of


the battery during a typical discharge of the battery at 1 C. This
analysis will not only allow the determination of the individual
thermal contributions generated by one battery, but also to investigate the phenomena generating this input. The thermal model is
presented in Table 1, including the energy balances for each section of the battery (cathode, separator and anode). For the energy
balances of the cathode and anode sections, the term on the left
side describes the heat accumulation, the rst term on the right
side corresponds to the heat conduction, the second term represents the heat effect due to electrode reactions, while the third
to fth terms account for the Joule heating in the solid active material and electrolyte phases, respectively.
Fig. 10 shows the experimental and modeled temperature proles as a function of depleted capacity during a typical discharge of
the battery at 1 C. As described in the experimental section, four
thermocouples were placed below the tabs on the surface of the
battery (see Fig. 1). The prole shown in Fig. 10 describes the maximum surface temperature rise. This also agrees with the 1-D
model (e.g. cross-section of the battery) used in this work to model
the behavior of the Li-ion battery. The thermal properties utilized
in the model are reported in Table 3, and they were determined
through tting the model to the experimental data. The parameters
describing the thermal properties were found to be insensitive to
the model, and as such the condence intervals could not be evaluated. The parameters reported in Table 2 associated with the
mass and charge balances in the liquid and solid phase of the battery were maintained xed during the tting of temperature. However, their contributions were allowed to vary with temperature,
e.g. De, je. Additionally, some of the parameters (e.g. k, Ds,i) were
converted in terms of the form described by Arrhenius equation
[20]. As observed in Fig. 10, a temperature rise of 4.5 K is produced
for the battery at the end of discharge. Although, this increase is
not signicant for one battery, this situation can become crucial
when the heat generated for each battery is integrated for the
entire battery pack and the batteries are found under quasi-adiabatic conditions.
It is important at this point to highlight two experimental
observations: (a) that, as previously mentioned, the temperature
increase reported was the maximum recorded surface battery temperature which always occurs near the current collectors (i.e. highest current) and that other regions of the battery (those located
away from current collectors) may experience half or less of this
rise in temperature (further studies will present more detailed

4.67
10.89
15.96 A h

304
303

Temperature/ K

cs / mol m-3

24000

16000

302
301
300

8000
299
298

0
1.5

2.5

S/C

Length of Cathode / dimensionless

C/CC

Fig. 9. Concentration of Li+ inside the particles (cs) across the lengths of the: (a)
anode and (b) cathode at different capacity values. CC/A, A/S, S/C, C/CC represent the
interfaces Current collector/Anode, Anode/Separator, Separator/Cathode and Cathode/Current collector, respectively. 1C test.

10

15

20

Capacity/ A h
Fig. 10. Computed and experimental temperature proles as a function of capacity
for Kokam batteries characterized at a 1-C rate. Symbols describe the experiment
and the continuous line represent the simulations calculated maintaining the
parameters reported in Table 2 xed, and tting the temperature properties shown
in Table 3.

481

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482


Table 3
Temperature properties in the lithium-ion battery model determined through tting.
Description
Density of the material
Heat capacity at constant pressure
Thermal conductivity
Heat transfer coefcient

Anode

Separator
3

qn = 3420 kg m

kn = 2.51 W m1 K1

Cathode
3

qs = 1930 kg m
Cp,s = 980 J kg1 K1
ks = 1.23 W m1 K1
h = 0.025 W m2 K1

qp = 1760 kg m3
kp = 5.63 W m1 K1

The condence intervals were not determined since the parameters had a low sensitivity to the model.

thermal proles for the battery); and (b) that, while the connections to the battery themselves may be a heat source or vector
for removal of heat from the batteries, the temperature at various
points during the charge/discharge periods was probed with a
Fluke Infra-red thermometer gun and it was consistently found
that the current-delivering wires were only slightly (0.5 to
1 K) colder than the surface of the battery underneath the tabs,
but the surface of the battery away from the tabs was at an even
lower temperature (23 K) than the wires. At this point it is known
that there is a region within the batteries close to the current collector tabs which experiences the highest current density and this
region is therefore the main source of heat (due to high Joule heating). This is due to the contact resistance between the wire and the
tab and is the reason why companies are looking for various methods to reduce the amount of heat generated. Further studies will
corroborate this and investigate other possible mechanisms of heat
generation/propagation, as well as extend the model to a 3D
simulation.
Note that at the end of discharge, there is an abrupt rise of temperature that is not well-predicted by the current model. There are
several possible reasons why this effect might be observed; it is
important to highlight that power discrepancies between model
and experiment is not a valid reason as the thermal model predicts
voltage response to a reasonable degree of accuracy and the current is input into the simulation exactly as it happened on the
experiment. The variation of reaction rates with SOC has also
already been accounted for in the model and is therefore unlikely
to be a cause of this effect. Upon analysis of the model parameters
it was found that the internal resistance of the battery does not
vary according to expectations. Typically, the internal resistance
decreases in the region between 30% and 70% depth of discharge
(DOD), increasing slightly toward 0% DOD and signicantly toward
100% DOD [72,73]; this being the main reason why companies tend
to operate their batteries within the 2080% DOD range.
In the model, the T slope in the entropy term was found to be a
signicant factor in determining the behavior of the internal resistance with respect to SOC, and it is believed that this term is
responsible for the unexpected variation observed. Due to convergence issues, a more accurate T term that matches both the temperature and voltage proles could not be found; however, it is
believed that the T term should, in theory, be replaced with a Taylor series expansion [53] in this model.
Another observation is that charging and discharging processes
(at the same C-rate) exhibit signicant hysteresis in terms of heat
generation. This contribution is somewhat lower during charging
at the same C-rate (not shown). This gives an insight into the
importance of heat generation terms due to electrode reactions:
if the majority of the heat were simply due to a Joule heating effect
then the temperature proles would be expected to be similar
(with the slightly higher temperature rise occurring during the
charging step as slightly more energy is delivered to the battery
in this step). Thus it can be concluded from this that during the
charging process there is some endothermic reactions while during
discharge these reactions are exothermic. It is possible that this

observation is due to the similar magnitude of endothermic reactions in comparison to Joule heating effects and exothermic reactions at (a maximum of) 1 C, i.e. relatively low C-rate. It is
expected that this will be less observable at high C-rates (5 C), as
the relative magnitude of the Joule heating will overshadow endothermic reactions. A more detailed thermal analysis will be the
motivation of a forthcoming study.
4. Conclusions
This study presented a methodology to rapidly determine the
parameters of physicochemical models utilized to account for the
behavior of commercial high capacity (16 A h) pouch Li-ion batteries (Kokam), such as the pattern of cells (e.g. SOC, State of Health)
that would be used in the automotive industry, when chemical
information is not available, or for a brand new system. A pseudo
2-D model comprised of different contributions reported in the literature was utilized to describe the mass, charge and thermal balances of the cell and porous electrodes; and adapted to the battery
chemistry under study. The methodology was based on combined
tting, calculation of condence intervals using the Analysis of
Variance for non-linear models and individual multi-parametric
sensitivity analysis as an efcient method to estimate the phenomena governing the battery voltage. The model was validated with a
battery comprised of carbon anodes and LiNi1/3Co1/3Mn1/3O2
(NMC) cathodes. It was found that the kinetics of Li+ insertion in
the cathode controls mostly the battery voltage despite mass and
charge transfer affect the performance of the batteries. A thermal
analysis was also conducted to account for the temperature rise
on the surface of the battery. This methodology will be useful for
analysis and understanding of changes in materials in a commercial cell, and it can be extended to the analysis of other types of
Li-ion batteries, as well as the evaluation of other phenomena
including capacity fade.
Forthcoming studies will be oriented to measure the possible
kinetic parameters of the pouch Kokam batteries through chemical and electrochemical measurements, with the intention to evaluate the accuracy of the values obtained by the present model.
Acknowledgments
The authors are indebted to the CONACYT (Grant No. 2012183230) and NSERC Automotive Partnership Canada for their
nancial support to carry out this work.
References
[1] Doyle M, Newman J, Gozdz AS, Schmutz CN, Tarascon JM. J Electrochem Soc
1996;143:1890.
[2] Winter M, Besenhard JO, Spahr ME, Novak P. Adv Mater 1998;10:72563.
[3] Thackeray MM. Prog Solids Chem 1997;25:171.
[4] Dahn JR, Fuller EW, Obrovac M, Von Sacken U. Solid State Ionics
1994;69:26570.
[5] Wang H, Jang Y-II, Huang B, Sadoway DR, Chiang Y-M. J Electrochem Soc
1999;146:47380.
[6] Hwang BJ, Tsai YW, Carlier D, Ceder G. Chem Mater 2003;15:367682.

482

J. Vazquez-Arenas et al. / Energy Conversion and Management 87 (2014) 472482

[7] Li DC, Muta T, Zhang LQ, Yoshio M, Noguchi H. J Power Sources


2004;132:1505.
[8] Oh S, Lee JK, Byun D, Cho W, Cho BW. J Power Sources 2004;132:249.
[9] Kweon HJ, Park JJ, Seo JW, Kim GB, Jung BH, Lim HS. J Power Sources
2004;126:156.
[10] Dominko R, Bele M, Gaberscek M, Remskar M, Hanzel D, Pejovnik S, et al. J
Electrochem Soc 2005;152:A60710.
[11] Padhi AK, Nanjundaswamy KS, Goodenough JB. J Electrochem Soc
1997;144:118894.
[12] Chung S-Y, Bloking JT, Chiang Y-M. Nat Mater 2002;1:1238.
[13] Huang H, Yin S-C, Nazar LF. Electrochem Solid State Lett 2001;4:A1702.
[14] Omar Noshin, Monem Mohamed Abdel, Firouz Yousef, Salminen Justin,
Smekens Jelle, Hegazy Omar, et al. Lithium iron phosphate based battery
assessment of the aging parameters and development of cycle life model. Appl
Energy 2014;113:157585.
[15] Omar Noshin, Daowd Mohamed, van den Bossche Peter, Hegazy Omar,
Smekens Jelle, Coosemans Thierry, et al. Rechargeable energy storage
systems for plug-in hybrid electric vehiclesassessment of electrical
characteristics. Energies 2012;5:295288.
[16] Safari M, Delacourt C. J Electrochem Soc 2011;158:A112335.
[17] Vazquez-Arenas J, Fowler M, Mao X, Chen S-K. Modeling of combined capacity
fade with thermal effects for a cycled LixC6LiyMn2O4 cell. J Power Sources
2012;215:2835.
[18] Ning G, White RE, Popov BN. Electrochim Acta 2006;51:201222.
[19] Srinivasan V, Newman J. J Electrochem Soc 2004;151:A151729.
[20] Gu WB, Wang CY. Lithium batteries. In: Surampudi, S, Marsh, RA, Ogumi Z,
Prakash J, editors. The electrochemical society proceedings series; 2000. p.
748.
[21] Tsang KM, Sun L, Chan WL. Identication and modelling of Lithium ion battery.
Energy Convers Manage 2010;51:285762.
[22] Jeon DH, Baek SM. Thermal modeling of cylindrical lithium ion battery during
discharge cycle. Energy Convers Manage 2011;52:297381.
[23] He H, Xiong R, Guo H, Li S. Comparison study on the battery models used for
the energy management of batteries in electric vehicles. Energy Convers
Manage 2012;64:11321.
[24] Rao Z, Wang S, Zhang G. Simulation and experiment of thermal energy
management with phase change material for ageing LiFePO4 power battery.
Energy Convers Manage 2011;52:340814.
[25] Smith KA, Rahn CD, Wang C-Y. Control oriented 1D electrochemical model of
lithium ion battery. Energy Convers Manage 2007;48:256578.
[26] Xie Y, Li J, Yuan C. Mathematical modeling of the electrochemical impedance
spectroscopy in lithium ion battery cycling. Electrochim Acta
2014;127:26675.
[27] Taheri P, Mansouri A, Yazdanpour M, Bahrami M. Theoretical analysis of
potential and current distributions in planar electrodes of lithium-ion
batteries. Electrochim Acta 2014;133:197208.
[28] Latz A, Zausch J. Thermodynamic derivation of a ButlerVolmer model for
intercalation in Li-ion batteries. Electrochim Acta 2013;110:35862.
[29] Fu R, Xiao M, Choe S-Y. Modeling, validation and analysis of mechanical stress
generation and dimension changes of a pouch type high power Li-ion battery. J
Power Sources 2013;224:21124.
[30] Yuanyuan Xie JL, Yuan Chris. Multiphysics modeling of lithium ion battery
capacity fading process with solid-electrolyte interphase growth by
elementary reaction kinetics. J Power Sources 2014;248:1729.
[31] Ral S, Hinaje M. Using electrical analogy to describe mass and charge
transport in lithium-ion batteries. J Power Sources 2013;222:11222.
[32] Paul S, Diegelmann C, Kabza H, Tillmetz W. Analysis of ageing inhomogeneities
in lithium-ion battery systems. J Power Sources 2013;239:64250.
[33] Xun J, Liu R, Jiao K. Numerical and analytical modeling of lithium ion battery
thermal behaviors with different cooling designs. J Power Sources 2013;47
61:4761.
[34] Al Hallaj S, Maleki H, Hong JS, Selman JR. J Power Sources 1999;83:18.
[35] Ramadass P, Haran B, White R, Popov BN. J Power Sources 2003;123:
23040.
[36] Schmidt AP, Bitzer M, Imre AW, Guzzella L. Experiment-driven
electrochemical modeling and systematic parametrization for a lithium-ion
battery cell. J Power Sources 2010;195:507180.
[37] Marcicki J, Canova M, Conlisk AT, Rizzoni G. Design and parametrization
analysis of a reduced order electrochemical model of graphite/LiFePO4 cells for
SOC/SOH estimation. J Power Sources 2013;237:31024.

[38] Lee JL, Chemistruck A, Plett GL. One-dimensional physics-based reduced-order


model of lithium-ion dynamics. J Power Sources 2012;220:43048.
[39] Kumar VS, Gambhire P, Hariharan KS, Khandelwal A, Kolake SM, Oh D, et al. An
explicit algebraic reduced order algorithm for lithium ion cell voltage
prediction. J Power Sources 2014;248:3837.
[40] Kumar VS. Reduced order model for a lithium ion cell with uniform reaction
rate approximation. J Power Sources 2013;222:42641.
[41] Lin X, Perez HE, Mohan S, Siegel JB, Stefanopoulou AG, Ding Y, et al. A lumpedparameter electro-thermal model for cylindrical batteries. J Power Sources
2014;257:111.
[42] Ye Y, Shi Y, Cai N, Lee J, He X. Electro-thermal modeling and experimental
validation for lithium ion battery. J Power Sources 2012;199:22738.
[43] Boovaragavan V, Harinipriya S, Subramanian VR. Towards real-time
(milliseconds) parameter estimation of lithium-ion batteries using
reformulated physics-based models. J Power Sources 2008;183:3615.
[44] Xu M, Zhang Z, Wang X, Jia L, Yang L. Two-dimensional electrochemicalthermal coupled modeling of cylindrical LiFePO4 batteries. J Power Sources
2014;256:23343.
[45] Du W, Gupta A, Zhang X, Sastry AM, Shyy W. Effect of cycling rate, particle size
and transport properties on lithium-ion cathode performance. Int J Heat Mass
Transfer 2010;53:355261.
[46] De PWCNS, Ramadesingan V, Model-based simultaneous optimization of
multiple design parameters for lithium-ion batteries for maximization of
energy density VRSubramanian. J Power Sources 2013;227:16170.
[47] Kenney B, Darcovich K, MacNeil DD, Davidson IJ. Modelling the impact of
variations in electrode manufacturing on lithium-ion battery modules. J Power
Sources. 2012;213:391401.
[48] http://www.dowkokam.com/cell-specications.php.
[49] Li X, Xiao M, Choe S-Y. Reduced order model (ROM) of a pouch type lithium
polymer battery based on electrochemical thermal principles for real time
applications. Electrochim Acta. 2013;97:6678.
[50] Baba N, Yoshida H, Nagaoka M, Okuda C, Kawauchi S. Numerical Simulation of
Thermal Behavior of Lithium-ion Secondary Batteries using the Enhanced
Single Particle Model. J Power Sources. 2014;252:21428.
[51] Zhu C, Li X, Song L, Xiang L. Development of a theoretically based thermal
model for lithium ion battery pack. J Power Sources. 2013;223:15564.
[52] Samba A, Omar N, Gualous H, Firouz Y, Van den Bossche P, Van Mierlo J, et al.
Development of an Advanced Two-Dimensional Thermal Model for Large size
lithium-ion pouch cells. Electrochimica Acta. 2014;117:24654.
[53] Thomas KE, Newman J. J Power Sources. 2003;119:8449.
[54] http://www.mathworks.com/products/statistics/.
[55] Huerta Garrido ME, Pritzker MD. J Electroanal Chemistry. 2006;594:11832.
[56] Advances in Lithium-ion batteries. in: W. A. van Schalkwijk, B. Scrosati, (Eds.).
1st ed. Kluwer Academic Publishers, New York, 2002.
[57] Lithium Batteries Science and Technology. in: G-A. Nazri, G. Pistoia, (Eds.). 1st.
ed. Springer, New York, 2003.
[58] http://www.kokam.com/product/product_pdf/high_energy_density/EL201_SLPB75106205_16Ah_Grade.pdf.
[59] http://www.emc-mec.ca/phev/Presentations_en/S13/PHEV09-S135_JeffDahn.pdf.
[60] Zhanga X, Mauger A, Lub Q, Groult H, Perrigaud L, Gendron F, et al. Electrochim
Acta 2010;55:64409.
[61] Gao P, Yang G, Liu H, Wang L, Zhou H. Solid State Ionics 2012;207:506.
[62] Hao L. Electrochemical performance of Li(NMC)O2 cathode materials for Li-ion
batteries. City University of Hong Kong, Hong Kong: Materials Engineering;
2010.
[63] Valena LO, Reimersa JN. J Electrochem Soc 2005;152:A88291.
[64] Kang S-H, Kim J, Stoll ME, Abraham D, Sun YK, Amine K. Power Sources
2002;112:418.
[65] Ngala JK, Chernova NA, Ma M, Mamak M, Zavalij PY, Whittingham MS. J Mater
Chem 2004;14:214.
[66] Doyle M, Fuentes Y. J Electrochem Soc 2003;150:A706.
[67] Vazquez-Arenas J. Electrochim Acta 2010;55:35509.
[68] Vazquez-Arenas J, Pritzker M. J Electrochem Soc 2010;157:D28394.
[69] Alliata D, Kotz R, Novak P, Siegenthaler H. Electrochem Commun 2000;2:436.
[70] Yuqin C, Hong L, Lie W, Tianhong L. J Power Sources 1997;68:187.
[71] Wang CS, Appleby AJ, Little FE. J Electroanal Chem 2001;497:33.
[72] Wei X, Zhu B, Xu W. International conference on measuring technology and
mechatronics automation. ICMTMA; 2009.
[73] Lou T, Zhang W, Guo HY, Wang JS. Adv Mater Res 2012;455:24651.

Você também pode gostar