Applied Mathematical Modelling 37 (2013) 2787–2801

Contents lists available at SciVerse ScienceDirect

Applied Mathematical Modelling

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

Spatial-resolution, lumped-capacitance thermal model for cylindrical

Li-ion batteries under high Biot number conditions
Rajib Mahamud, Chanwoo Park ⇑
Department of Mechanical Engineering, University of Nevada, Reno, NV 89557-0312, USA

a r t i c l e i n f o a b s t r a c t

Article history: The time-efficient yet accurate thermal modeling of the battery cells for electric and hybrid
Received 14 September 2011 electric vehicles is essential improving the performance, safety, and lifetime of the battery
Received in revised form 29 May 2012 system. This paper presents a spatial-resolution, lumped-capacitance (LC) thermal model
Accepted 6 June 2012
for cylindrical battery cells under high Biot number (Bi P 1) conditions where the classical
Available online 15 June 2012
LC thermal model is generally inapplicable because of a significant temperature variation
in the cell volume. The spatial-resolution LC model was formulated using zero- and first-
Keywords:
order Hermite integral approximations. For model validation, a one-dimensional, transient
Li-ion battery
Battery thermal management
analytical (exact) solution using Green functions was obtained for a cylindrical Li-ion bat-
Electric vehicles and hybrid electric vehicles tery cell with uniform volumetric battery heat generation of Joule and entropic heating
Lumped-capacitance thermal model under convective cooling boundary conditions. It was found from the comparison of the
High Biot number results that the spatial-resolution LC thermal model can accurately and quickly predicts
Battery duty cycles the cell temperatures (core, skin and area-averaged) under various dynamic battery duty
cycles even for high Biot numbers due to highly convective conditions such as liquid
cooling.
Ó 2012 Elsevier Inc. All rights reserved.

1. Introduction

As electric vehicles (EVs) and hybrid electric vehicles (HEVs) are poised to become a major player in the global auto mar-
ket, there is a great need of advanced thermal modeling tools for battery cells to design and operate the vehicle traction bat-
tery systems more efficiently. The traction battery systems often house several hundred to thousands of battery cells such as
Li-ion and Ni-MH, which are used under dynamic electrical loading [1–3].
Battery thermal management is critical for operating the battery systems efficiently in all driving and climate conditions
without causing thermal damage to the battery cells [1–4]. The Li-ion battery has recently been a popular choice for the vehi-
cle traction battery due to its high power density and charge/discharge efficiency [5]. However, a tight temperature require-
ment presents a big challenge to the Li-ion battery thermal management. For example, the operating temperatures of the Li-
ion battery for EVs and HEVs typically ranges from 40 to 40 °C and the cell-to-cell temperature differences are required to
be below 5 °C [1,3].
The computational fluid dynamics (CFD) models and finite element models have been widely used for the battery thermal
analysis [1,3,6,7]. Multi-dimensional CFD modeling is often a difficult task because of complex modeling steps such as solid
modeling, meshing, intense computation and post-processing. It is sometimes very time-consuming and impractical to run

⇑ Corresponding author. Address: Department of Mechanical Engineering, University of Nevada, Reno 1664 N. Virginia Street, Reno, NV 89557-0312, USA.
Tel.: +1 775 682 6301; fax: +1 775 784 1701.
E-mail address: chanwoo@unr.edu (C. Park).

0307-904X/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.apm.2012.06.023
2788 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

Nomenclature

A area [m2]
Bi cell Biot number, Bi = hfR/ks
cp specific heat capacity [J kg 1 K1]
D diameter of a battery cell [m]
dEoc/dT entropic coefficient [mV K1]
E cell potential [V]
G Green function
H Hermite integral approximation
h convective heat transfer coefficient [W m2 K1]
I battery electrical current [A]
J Bessel function of the first kind
M[kg] thermal conductivity [W m1 K1]
L length [m]
R radius [m]
Re internal electrical resistance of a battery cell [mX]
r radial dimension [m]
r0 instantaneous location [m]
S heat generation [W]
SOC state of charge [%]
s volumetric heat generation [W m3]
T temperature [K]
t time [s]
V volume [m3]
x negative electrode stoichiometry constant
y positive electrode stoichiometry constant
z axial dimension [m]

Greek symbols
a thermal diffusivity [m2 s1]
b eigenvalue
q density [kg m3]
u homogenous solution
s time or period of charging and discharging [s]

Subscripts
0 zero-order Bessel function
1 first-order Bessel function
a ambient condition
avg area-averaged
c core
e electrical
f fluid phase
i,j order of Hermite integral approximation
irr irreversible
m dummy index (1,2, . . . , 1)
o initial condition
oc open circuit
p pressure
r battery reaction
rev reversible
s heat generation and solid phase or skin
x,y electrode stoichiometry constants

Other symbols
- time-averaged
1 ambient condition or infinity
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2789

the CFD models for the battery thermal simulation using vehicle duty cycles. As a result, it is considered that the CFD tools
are not suitable for real-time calculation required for smart battery control and vehicle duty cycle analyses.
A number of mathematical and numerical models for battery electrochemical and thermal analysis have been developed
at battery system, pack and cell level [1–10]. Like the electrical battery modeling, a lumped-capacitance (LC) thermal model
is commonly used for the battery thermal modeling thanks to its simple formulation and minimal computational time. It is
well-known that the LC thermal model is reasonably accurate only for low cell Biot number conditions (Bi = hfR/ks < 0.1)
where the inherent assumption of uniform cell temperature is valid [1,3]. The Biot number can be increased by increased
heat transfer coefficient and/or large cell diameter. In fact, some of the latest HEVs and EVs employ liquid cooling [11] whose
convective heat transfer coefficient (hf) is an order-of-magnitude higher than conventional air cooling method. Some appli-
cations that use large cell diameter (D = 2R) such as marine, power storage and aerospace applications also run into the same
problem [12]. Such progresses in battery technologies could make the Biot number well exceeds the Biot number limit of 0.1
causing the classical LC model inapplicable unless the accuracy is compromised. Heavy battery electrical loadings like pulse
power also create significant battery heat in a very short period of time resulting in substantial temperature variation across
the cell. Not to mention that the battery surface temperature measured from the temperature sensor installed on the battery
skin is often used as the battery temperature for battery thermal management. It is reasonable to anticipate that the emerg-
ing technical progresses require better thermal models capable of providing a spatial resolution and applicable for high Biot
number conditions encountered in the future applications.
In this paper, an improved LC thermal model with a spatial resolution was developed for cylindrical Li-ion cells as a time-
efficient, accurate thermal modeling tool for dynamic battery cycle analysis. The Hermite integration approximations of zero
and first orders [6,13–15] were used in the spatial-resolution LC thermal model to calculate the cell temperatures at core and
skin under high Biot number conditions where the classical LC model is usually not applicable. This paper presents the com-
putational results from the spatial-resolution and classical LC models as well as an analytical solutions using Green function
as a benchmark calculated using various battery electrical loadings including vehicle driving cycles.

2. Spatial-resolution, lumped-capacitance thermal model using Hermite approximation

The cylindrical Li-ion cell is typically constructed in a spirally-wound, multilayer structure of liquid-phase electrolyte and
solid-phase electrodes as shown in Fig. 1 [5]. Because of the interfaces between the electrode layers, the thermal conductivity
of the cell in the radial direction is usually much lower than the axial direction. In actual battery packs, the battery cells are
often stacked in series to form a battery module wrapped in an electrical insulation tape or tube causing the battery ends
thermally insulated. As a result, the Li-ion cells are often treated as a radial (one-dimensional) system for thermal modeling
[1,16]. The internal heat generation of Li-ion cells typically consists of reversible (entropic heating) and irreversible (Joule
heating) parts [17–22] which were taken into account into the current analysis. The heat generation rate not only varies spa-
tially- and temporally but also is affected by the battery cell temperature and state-of-charge (SOC) condition.

Fig. 1. (a) Multi-layer structure of a cylindrical Li-ion battery cell [5] and (b) one-dimensional (-direction) computational domain used for the numerical
analysis.
2790 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

2.1. Governing equations

The energy conservation equation and boundary and initial conditions for the radial system of a cylindrical cell [23] are
given by
 
@T 1 @ @T
qs ðTÞcp;s ðTÞ ¼ rks ðTÞ þ sðr; tÞ for 0 < r < R and t > 0 ð1Þ
@t r @r @r

@TðR; tÞ
ks ðTÞ ¼ hf ðtÞ½TðR; tÞ  T 1 ðtÞ for t > 0 ð2Þ
@r

@Tð0; tÞ
¼ 0 for t > 0 ð3Þ
@r

Tðr; 0Þ ¼ T o ðrÞ for 0 < r < R ð4Þ

The battery cell is assumed to be subject to uniform convective heat transfer coefficient and ambient temperature. It is, how-
ever, possible to extend the cell-level model to a pack-level model for multiple cells by using a flow network model which
calculates the ambient fluid temperatures along the flow direction. The detailed model of the flow network can be found in
Ref. [4]. The battery heat generation is originated from the complex electrochemical reaction of Li-ion battery in non-homo-
geneous cell volume. The major contributions to the battery heat generation consist of the irreversible heat loss due to joule
heating; and the reversible heating due to entropic losses. Side reaction, corrosion reaction and parasitic losses which can
contribute to the heat generation were neglected in this study. It was also assumed that the battery heat generation was
uniform in the cell and varied as a function of the average cell temperature. Therefore, the battery heat generation can be
given by
SðtÞ ¼ sðtÞV s ¼ Q irr þ Q rev ð5Þ
where, the first term (Qirr) is due to the Joule heating based on the battery internal electrical resistance Re and is given by

Q irr ¼ IðEoc  EÞ ¼ I2 Re ð6Þ

where, the battery electrical resistance Re was experimentally determined by averaging over a range of the state-of-charge
(SOC) and is expressed in a function of the averaged cell temperature by Park and Jaura [1,3]

Re ¼ 0:0001 T 3 þ 0:0134 T 2  0:5345 T þ 12:407 ð7Þ

where, Re is the internal electrical resistance [mX] of a Li-ion battery cell (LiMn2O4/C, capacity = 3.6 Ah,
D  L = 42.4  97.7 mm2, see Table 2). The battery T is the area-averaged cell temperature [°C].
The second term (Qrev) in the battery heat generation equation represents the reversible entropic loss and is given by

dEoc
Q rev ¼ IT ð8Þ
dT
where the entropy coefficient dEoc/dT is a function of charge density, SOC and cell temperature. For hybrid electric vehicle
application, the entropic loss is often neglected, because at high charge/discharge rates, the irreversible Joule heating be-
comes more dominant than the reversible entropic heating [20]. The entropic loss can be either positive (exothermic) or neg-
ative (endothermic) depending on the charging or discharging conditions. As a result, the net heating effect of the reversible
heating could be negligible after a full cycle of charging/discharging for a relatively constant SOC condition.
The dEoc/dT can be found from experimental measurements. For a SOC range (100–0%) considered for this study, the com-
position stoichiometry ranges are y = 0.442–0.936 for the positive electrode (LiyMn2O4) and x = 0.676–0.126 for the negative
electrode (LixC6) [20]. Therefore, the entropic coefficient for the positive electrode [17,19,20] as shown in Fig. 2(a) is given by
a curve-fitting as below

dEoc
¼ 29:41SOC6  54:18SOC5 þ 20:64 SOC4 þ 8:4946SOC3  5:4224SOC2 þ 1:0674SOC  0:2057 ð9Þ
dT
The entropic coefficient for the negative electrode [17,18,20] as shown in Fig. 2(b) is given by a curve-fitting as below

dEoc expð18:12983079 SOC þ 4:163332068Þ

¼ 344:1347148  0:418400661SOC
dT 1 þ 749:0756003 expð19:13504805 SOC þ 4:503478072Þ
þ 0:109595516SOC2 þ 0:168196519 ð10Þ
The resulting entropic coefficient for the complete Li-ion cell (LiyMn2O4/LixC6) is shown in Fig. 2(c).
Most of the battery thermal management system of electrical-drive vehicles uses cabin air as the coolant fluid because of
easy access and reliable design despite the weak convective cooling ability. The conventional lumped-capacitance (LC) ther-
mal model is applicable for the low cell Biot number conditions (i.e., Bi = hfR/ks < 0.1). In other words, the thermal resistance
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2791

Fig. 2. Empirical entropic coefficients as a function of state-of-charge (SOC) of the half cells (LiMn2O4/Li and Li/C) and the corresponding best-fit curves (a)
for the positive electrode (LiyMn2O4) [17,19,20] (b) for the negative electrode (LixC6) [17,18,20] and (c) for the complete cell.

of the heat conduction in a radial direction in the cylindrical cell could be much lower than that of the convection heat trans-
fer, and thus the temperature in the battery cell can be assumed to be uniform [1–3]. However, recent advances in battery
technologies for HEVs and EVs such as liquid cooling could increase the convective heat transfer as much as 100 times the air
cooling resulting in Bi > 1. Such high Biot number conditions could create a significant spatial temperature distribution in the
cell. Therefore, there is a great need for a suitable LC thermal model to be applicable for such high Biot number conditions.
The area-averaged cell temperature in a radial system as shown in Fig. 1(b) is given by
Z R
1
T av g ðtÞ ¼ 2
ð2prÞTðr; tÞdr ð11Þ
pR r¼0

Therefore, the temporal variation of the spatially-averaged temperature is given by

Z R
dT av g ðtÞ 1 @Tðr; tÞ
¼ ð2prÞ dr ð12Þ
dt p R2 r¼0 @t
Integrating the energy equation Eq. (1) with respect to r, using the definition of the average temperature, and combining the
resultant equation with the boundary conditions Eqs. (2) and (3), the energy equation can be finally written as
dT av g ðtÞ 2hf ðtÞ
qs ðTÞcp;s ðTÞ ¼ ½TðR; tÞ  T 1 ðtÞ þ sðtÞ ð13Þ
dt R
Note that the above equation still preserves the spatial-resolution [i.e., Tavg(t) and T(R, t)] unseen in the equation of the con-
ventional LC thermal model. Since the temperature variation in the cell for higher Biot number conditions is significantly
large, it is essential to establish a proper relationship between the surface and average temperatures for the LC thermal mod-
eling [15] by
TðR; tÞ ffi f ½T av g ðtÞ ð14Þ
2792 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

In fact, the conventional LC thermal model only good for low Biot numbers (Bi < 0.1) is based on the assumption that the
temperature gradient is small enough so that the cell temperature is uniform, i.e., T(R, t) = Tavg(t). As a result, the conventional
LC thermal model is formulated by

dT av g ðtÞ 2hf ðtÞ  

qs ðTÞcp;s ðTÞ ¼ T av g ðtÞ  T 1 ðtÞ þ sðtÞ ð15Þ
dt R
The relationship between the surface and average temperatures can be more generalized using the Hermite-type approx-
imation [13–15] of integrals. Various accuracies of the Hermite integral approximations can be achieved with more complex
analytical involvement. In this paper, four different formulations are discussed in the following sections. For the notation of
the Hermite approximations, Hi,i/Hj,j format is used: the first part Hi,i is for the average temperature Tavg(t) and the second
part Hj,j is for the temperature gradient @T(r, t)/@r. The dummy indices i,j represent the order of the Hermite integration.

2.2. H0,0/H0,0 approximation

This approximation is based on the trapezoidal (zero-order) formulas (H0,0) for the average temperature and temperature
gradient. The equations of the lowest approximation are given by
 
1 1
T av g ðtÞ ffi Tð0; tÞ þ TðR; tÞ ð16Þ
2 2
 
R @Tð0; tÞ @TðR; tÞ
TðR; tÞ  Tð0; tÞ ffi þ ð17Þ
2 @r @r
Combining the above Eqs. (16) and (17) with the equations for the boundary conditions, Eqs. (2) and (3), the surface tem-
perature T(R, t) is given by
 1
Bi hf R
TðR; tÞ ¼ 1 þ ½T av g ðtÞ  T 1 ðtÞ þ T 1 ðtÞ; Bi ¼ ð18Þ
4 ks
The average temperature Tavg(t) is obtained by solving the energy equation (13) with the following initial condition

T av g ð0Þ ¼ T o ð19Þ
The core temperature T(0, t), and surface temperature T(R, t) are calculated using Eqs. (16) and (18).

2.3. H1,1/H0,0 approximation

This approximation is based on the corrected trapezoidal (first-order) formula (H1,1) for the average temperature and nor-
mal trapezoidal (zero-order) formula (H0,0) for the temperature gradient. The corrected trapezoidal formula for the average
temperature is given by
 
1 5 Bi
T av g ðtÞ ffi Tð0; tÞ þ TðR; tÞ þ ½TðR; tÞ  T 1 ðtÞ ð20Þ
6 6 6
The normal trapezoidal formula for the temperature gradient is given by Eq. (17). The average temperature Tavg(t) is calcu-
lated similarly as the H0,0/H0,0 approximation. Combining Eqs. (17) and (20) with the boundary conditions, Eqs. (2) and (3),
the expression for surface temperature was found which is same as Eq. (18). The core temperature T(0, t), and surface tem-
perature T(R, t) are calculated using Eqs. (18) and (20).

2.4. H0,0/H1,1 approximation

This approximation is based on the normal trapezoidal formula (H0,0) for the average temperature and corrected trape-
zoidal formula (H1,1) for the temperature gradient. The corrected trapezoidal formula for the temperature gradient can be
written as
  " #
R @Tð0; tÞ @TðR; tÞ R2 @ 2 Tð0; tÞ @ 2 TðR; tÞ
TðR; tÞ  Tð0; tÞ ffi þ þ  ð21Þ
2 @r @r 12 @r 2 @r2

Evaluating the energy equation Eq. (1) at r = 0 and r = R, respectively, we obtain

@ 2 Tð0; tÞ 1 @Tð0; tÞ sðtÞ ks ðTÞ

¼  ; as ðTÞ ¼ ð22Þ
@r2 2as ðTÞ @t 2ks ðTÞ qs ðTÞcp;s ðTÞ
@ 2 TðR; tÞ 1 @TðR; tÞ 1 @TðR; tÞ sðtÞ
¼   ð23Þ
@r 2 as ðTÞ @t R @r ks ðTÞ
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2793

Table 1
Summary of the formulation of the Hermite-type approximations used for the spatial-resolution, lumped-capacitance thermal model.

Governing equation dT av g ðtÞ 2hf ðtÞ

qs ðTÞcp;s ðTÞ ¼ ½TðR; tÞ  T 1 ðtÞ þ sðtÞ
dt
 R 
Average temperature 1 1
H00 T av g ðtÞ ffi Tð0; tÞ þ TðR; tÞ
2 2
 
1 5 Bi
H11 T av g ðtÞ ffi Tð0; tÞ þ TðR; tÞ þ ½TðR; tÞ  T 1 ðtÞ
6 6 6
 
Temperature gradient R @Tð0; tÞ @TðR; tÞ
H00 TðR; tÞ  Tð0; tÞ ffi þ
2 @r @r
   
R @Tð0; tÞ @TðR; tÞ R2 1 @Tð0; tÞ sðtÞ 1 @TðR; tÞ 1 @TðR; tÞ sðtÞ
H11 TðR; tÞ  Tð0; tÞ ffi þ þ   þ þ
2 @r @r 12 2as ðTÞ @t 2ks ðTÞ as ðTÞ @t R @r ks ðTÞ

Substituting Eqs. (22) and (23) into Eq. (21), we have

   
R @Tð0; tÞ @TðR; tÞ R2 1 @Tð0; tÞ sðtÞ 1 @TðR; tÞ 1 @TðR; tÞ sðtÞ
TðR; tÞ  Tð0; tÞ ffi þ þ   þ þ ð24Þ
2 @r @r 12 2as ðTÞ @t 2ks ðTÞ as ðTÞ @t R @r ks ðTÞ
Differentiating Eq. (16) with respect to t, we obtain
 
dT av g ðtÞ 1 @Tð0; tÞ @TðR; tÞ
ffi þ ð25Þ
dt 2 @t @t
Therefore, the energy equation Eq. (13) is reduced to
 
1 @Tð0; tÞ @TðR; tÞ b2hf ðtÞ=Rc½T 1 ðtÞ  TðR; tÞ þ sðtÞ
þ ¼ ð26Þ
2 @t @t qs ðTÞcp;s ðTÞ
The core temperature T(0, t), and surface temperature T(R, t) are calculated by solving Eqs. (24) and (26) with the following
initial conditions.
Tð0; 0Þ ¼ T o ; and TðR; 0Þ ¼ T o ð27Þ
Finally, the average temperature Tavg(t) is calculated using Eq. (16).

2.5. H1,1/H1,1 approximation

This approximation is based on the corrected trapezoidal formula (H1,1) for the average temperature and temperature
gradient. The corrected trapezoidal formula was given by Eq. (20) for the average temperature and was given by Eq. (24)
for the temperature gradient.
Differentiating Eq. (20) with respect to t, and equating it with the energy equation (13), we have
 
1 @Tð0; tÞ 5 @TðR; tÞ Bi @TðR; tÞ @T 1 ðtÞ b2hf ðtÞ=Rc½T 1 ðtÞ  TðR; tÞ þ sðtÞ
þ þ  ¼ ð28Þ
6 @t 6 @t 6 @t @t qs ðTÞcp;s ðTÞ
Solving Eqs. (24) and (28), the core temperature T(0, t), and surface temperature T(R, t) can be found using the above initial
conditions Eq. (27). Finally, the average temperature Tavg(t) can be found using Eq. (20). The formulation of the Hermite-type
approximations discussed above is summarized in Table 1.

3. Results and discussion

3.1. Validation of spatial-resolution, lumped-capacitance thermal models

In this paper, a numerical method using a spatial-resolution, lumped-capacitance model was discussed and validated
against the analytical solution using Green function. The numerical method using a spatial-resolution, lumped-capacitance
model provides the spatial temperature distribution in the battery cell at the battery core (or center) and surface in addition
to the averaged temperature which is obtained from the conventional lumped-capacitance model. The analytical solutions
using Green function were obtained to validate the spatial-resolution, LC thermal model (refer to Appendix A). The accuracy
of the simulation results of the spatial-resolution, LC thermal model was verified by the analytical solutions for both simple
battery loadings of constant and step-wise current and two battery cycles of electrical-drive vehicles.
The dimensional, thermophysical and electrical properties of the Li-ion cell and operational conditions used for the anal-
ysis are listed in Table 2. Fig. 3 shows the comparison of the results of the classical LC thermal model and the spatial-reso-
lution LC thermal model using the H0,0/H1,1 approximation with the exact solutions as the benchmark under the conditions
of the high Biot number (Bi = 5 and To = Tf,1 = 25 °C) and a constant battery loading of 50 A which corresponds to a contin-
uous drop of the battery SOC from the fully charged (100%) condition to 22%. As shown in Fig. 3, the differences in the tem-
peratures (core, average and surface) between the exact solution and the H0,0/H1,1 approximation was very small. In contrast,
2794 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

Table 2
Dimensional, thermophysical and electrical properties of the Li-ion cell and operational conditions used for the analysis.

Battery technology: Li-ion battery (cathode: LiMn2O4, anode: carbon)

Battery cell specifications Baseline conditions
Capacity [Ah] 3.6 Current [A] 50(=13.88C),
D [mm] 42.4 90 (=25C)
L [mm] 97.7 Tf,1 [°C] 25
M [kg] 0.3 hf [W m2 K1] 151, 1509,
Vs [mm3] 137,948.2 7547, 15,095
qs [kg m3] 2007.7 Bi 0.1, 1, 5, 10
ks[W m1 K1] 32.2 To [°C] 25
cp,s [J kg1 K1] 837.4

Fig. 3. Comparison of the cell temperatures [core (Tc), skin (Ts) and average (Tavg)] from the analytical solution and various lumped-capacitance thermal
models for a constant battery current (I = 50 A) and Bi = 5.

the average temperature from the classical LC thermal model is also quite different from the average temperatures of the
other sources showing that the classical LC model is significantly inaccurate for high Biot number conditions. Furthermore,
it means that the averaged temperature from the classical LC model is too under-predicted to be used as a representative
battery temperature considering the temperature at the core of the cell is even higher than the average temperature.
This significant error in the simulation results of the spatial-resolution LC thermal model can arise when the ambient
temperature differs from the initial cell temperature despite the use of higher-order Hermite Integral formulation. Because
the first-order approximation for the average temperature [Eq. 20] uses the difference between the ambient and battery skin
temperatures T(R, t)  T1(t) in the equation, there is an inevitable error in the calculation of the average, skin and surface
temperatures, especially at the initial period of the calculation when the temperature difference is the greatest.
Fig. 4(a)–(c) shows the comparison of the results from the Hermite approximations with the exact solutions for high Biot
number (Bi = 5) using the constant battery loading as shown in Fig. 3 and lower ambient temperature than the initial tem-
perature (i.e., Tf,1 = 20 °C and To = 25 °C). As shown in Fig. 4, the significant errors in the temperatures occur during the first
few seconds from the initial time, which we called the ‘‘initial error’’ except from the H0,0/H1,1 approximation giving the best
agreement with the exact results. It is clearly explained by the fact that the zero-order approximation H0,0 for the average
temperature in Eq. (16) does not have the ambient temperature, therefore, the initial error problem does not exist. It is also
interesting to see that the higher-order approximations do not always give more accurate results, especially when the ambi-
ent temperature is different from the initial temperature. It is also worth noting that the H0,0/H0,0 and H1,1/H0,0 approxima-
tions predict the identical results, since the same differential equations were used for those approximations [Eqs. (17) and
(18)]. Because of the best result from the H0,0/H1,1 approximation, this approximation was chosen as the baseline formulation
throughout this study. It is also shown in Fig. 4(b) that the temperature (Tavg, average temperature) from the classical LC
model is always lower than the spatial-resolution LC model and rather close to the surface temperature [T(R, t)] of the spa-
tial-resolution LC model.
The additional comparison of the results of the LC thermal models using a step-wise battery current profile consisting of a
periodic charge and discharge as shown in Fig. 5 was performed. The SOC change and battery heat generation are also shown
in the figure. The charging and discharging rates were 25C (=90 A), respectively. The periods of the charging and discharging
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2795

Fig. 4. Comparison of the cell temperatures [(a) core (Tc), (b) average (Tavg) and (c) skin (Ts)] from the analytical method and the spatial-resolution lumped-
capacitance thermal model based on four different Hermite approximations (H0,0 /H0,0, H0,0/H1,1, H1,1/H0,0, and H1,1/H1,1) when T1 = 20 °C and T0 = 25 °C.

Fig. 5. Stepwise battery duty cycle and the corresponding heat generation and state-of-charge (SOC) variation.
2796 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

were 144 s so that the SOC variation ranged from 100% to 0%. Fig. 6 shows the comparison of the average temperature results
of the H0,0/H1,1, and H1,1/H1,1 approximations, the classical LC model and the exact solutions for a cell Biot number of 5 using
the step-wise current profile and the same ambient and initial temperatures. As shown in the figure, results from the Her-
mite approximations match well with the exact solution except the classical LC thermal model. The core, averaged and skin
temperatures from the H0,0/H1,1 approximation were also compared with the exact solutions in Fig. 7. It is clearly shown that
the temperature results from the H0,0/H1,1 approximation match reasonably well with the exact solutions.
The computational accuracy and time needs to be taken into consideration when selecting the suitable thermal modeling
for battery thermal management. Table 3 lists the computational times of three different computational methods (two-
dimensional CFD model, spatial-resolution lumped-capacitance model, finite-difference model) based on the same battery

Fig. 6. Comparison of the average cell temperatures (Tavg) from the spatial-resolution lumped-capacitance model based on two different Hermite
approximations (H1,1/H1,1, and H0,0/H1,1) and the classical lumped-capacitance model with the exact solution for the stepwise battery duty cycle and Bi = 5.

Fig. 7. Comparison of the cell temperatures [core (Tc), skin (Ts) and average (Tavg)] from the analytical method and the spatial-resolution lumped-
capacitance thermal model based on H0,0/H1,1 approximation for the stepwise battery duty cycle and Bi = 5.

Table 3
Computational time of various computational methods.

Computational methods Time [min]

Computational fluid dynamics (CFD) 360
Exact solution 40
Finite volume method (FVM) using 11 temperature nodes 15
Spatial-resolution, lumped-capacitance thermal model 2

Simulation time (2000 s = 33.3 min) used for the battery loading shown in Fig. 5.
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2797

duty cycle as shown in Fig. 5. It is not arguable that the CFD model is the most preferred choice for the battery thermal mod-
eling because of its accuracy, versatility and sophisticated modeling options [1,3,6] excluding the fact that the CFD model
requires a very long computation time because of many temperature nodes used for the simulation. It is very clear from
the comparison of the results that the spatial-resolution LC thermal model is both computationally efficient and accurate
enough to perform real-time thermal analyses requiring spatial temperature information for smart battery thermal manage-
ment and control.

3.2. Effect of cell Biot number

Advances in battery technologies such as liquid cooling (higher hf) and large cell (larger linear dimension R) could lead to
high Biot number (Bi = hfR/ks) conditions. The convective heat transfer coefficient (hf) of liquid cooling is roughly an order-of-
magnitude higher than the conventional air cooling [11,24]. Fig. 8 shows the effects of the cell Biot number (or convective
heat transfer coefficient) on the battery core ðT c Þ, area-averaged ðT av g Þ and skin ðT s Þ temperatures which were time-averaged
over three consecutive cycles (3s = 864 s) of the battery current loading (as shown in Fig. 5). After a quasi-steady state was
reached, the time-averaging was performed by
Z 3s
T¼ Tdt=3s ð29Þ
0

The convective heat transfer coefficient in Fig. 8 was varied ranging from weak forced convection of air cooling
(75 W m2 K1) to highly convective liquid cooling (7547 W m2 K1) conditions.
Fig. 8 shows that the temperature difference ðT c  T s Þ between the core and skin was increased with the Biot number (or
heat transfer coefficient) and was about 1.2 °C at Bi = 1.0. It was also shown that the average temperature from the classical
LC thermal model follows the skin temperature from the spatial-resolution LC thermal model. This is because the same stea-
dy-state energy balance between the convection heat transfer and battery heat generation) at the battery skin is used for the
two LC thermal models. As a result, the average temperature from the classical LC thermal model was always lower than the
spatial-resolution LC thermal model. It is clearly expected that the temperature difference between battery core and skin
would be higher for the larger cell diameters. In fact, it is reported that a scaled-up battery (LiCoO2/C) with a 100 Ah cell
capacity has a 7 °C temperature difference between the core and skin with the Biot number of as low as 0.6 [12]. Therefore,
there is a legitimate need for accurate battery thermal modeling for high Biot number conditions when the large tempera-
ture difference between the core and skin is unavoidable and the classical LC thermal model is inaccurate not to mention that
the high battery electrical loading worsens the temperature problem.

3.3. Case studies using electrical-drive vehicle power cycles

Two plausible battery duty cycles for electrical-drive vehicles were considered for further validation of the spatial-reso-
lution LC thermal model: (i) HEV power cycle of charging and discharging as shown in Fig. 9(a), and (ii) EV power cycle of
discharging as shown in Fig. 9(b). Fig. 9 also shows the SOC variation of each battery duty cycle. Fig. 10 shows the average cell

Fig. 8. Effect of the cell Biot number on the time-averaged core, averaged and skin temperatures (The heat transfer coefficient was also shown
corresponding to the Biot number).
2798 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

Fig. 9. Battery current and SOC profiles for the (a) HEV and (b) EV power cycles.

Fig. 10. Comparison of the average cell temperatures (Tavg) for three different Biot numbers using (a) the HEV power cycle [Fig. 9(a)] and (b) the EV power
cycle [Fig. 9(b)].

temperature results of the H0,0/H1,1 approximation for the battery duty cycles [Fig. 10(a) for the HEV power cycle as shown in
Fig. 9(a); and Fig. 10(b) for the EV power cycle as shown in Fig. 9(b)] for three different Biot numbers ranging from 0.1 to 10.
The cell Biot number Bi (=hfR/ks) can be changed by varying convective heat transfer coefficient, cell diameter, and/or thermal
conductivity. In this study, only the convective heat transfer coefficient was changed to vary the Biot number. The convective
heat transfer coefficients of 151, 1509, 7547 and 15,095 W m2 K1 correspond to the Biot numbers of 0.1, 1, 5, and 10,
respectively. The large heat transfer coefficients could be achieved by highly convective cooling such as liquid cooling
[11]. Fig. 10 shows that the average cell temperatures are increased as the Biot number is decreased due to the weak con-
vective cooling (e.g., air cooling). It is also shown from Fig. 10(b) that the EV power cycle increases the battery cell temper-
ature significantly during the large current drawing period at the end of the cycle, especially for the low Bi number
conditions (Bi = 0.1).
Fig. 11 shows the core, skin and average temperatures of the spatial-resolution LC thermal model for the HEV power cycle
under an intermediate Biot number of 1. As shown in Fig. 11(a), the temperature variation was not significant due to the
highly convective cooling. Fig. 11(b) shows the magnified view of the temperatures as shown in Fig. 11(a). The differences
between the surface, core and average temperatures for the intermediate Biot number were rather significant as compared to
the low Biot number. It can be concluded from the simulation results based on the vehicle power cycles that the spatial-res-
olution LC thermal model based on the H0,0/H1,1 approximation provides the spatial resolution of the cell temperatures (core
and skin temperatures) necessary for accurately monitoring the battery thermal history affecting battery health unlike the
classical lumped thermal model on a minimum computation [1–3]. The analytical method of solution can be considered to
solve the problem; however, the improved lumped thermal model is most computationally efficient considering real time
temperature prediction and control.
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2799

Fig. 11. (a) Cell temperatures [core (Tc), skin (Ts) and average (Tavg)] from the spatial-resolution, lumped-capacitance thermal model based on H0,0 /H1,1
approximation for the HEV power cycle [Fig. 9(a)] and Bi = 1 and (b) magnified view of the cell temperatures.

4. Conclusion

The spatial-resolution, lumped-capacitance (LC) thermal models based on zero- and first-order Hermite integral approx-
imations were developed for the application of cylindrical rechargeable cell condition by assuming constant thermal prop-
erties, which are applicable for a wide range of Biot number conditions including Bi P 0.1 where the classical LC thermal
model is inapplicable because of the significant temperature variation in the cell. The cell Biot number could be increased
by using highly convective cooling (e.g., liquid cooling) and/or large cell diameters. It is found that a significant temperature
difference between the core and skin would exist which is not linear. The analytical (exact) solutions based on Green func-
tion for a cylindrical Li-ion battery cell with time-dependent and volumetrically-uniform battery heat generation of Joule
and entropic heating under convective cooling boundary conditions were obtained as the benchmark solutions for the val-
idation of the numerical results from the spatial-resolution LC thermal models. The benchmark solution using analytical
method of solution is computationally accurate and computationally efficient than CFD model for a simple or complex bat-
tery duty cycle. It was found from the comparison of the numerical results with the exact solutions using various battery
loadings including two electrical-drive vehicle battery duty cycles that the spatial-resolution LC thermal model accurately
predicts the cell temperatures at the battery core and skin as well as the area-averaged temperature regardless of the cell
Biot numbers unlike the classical LC thermal model. It was also found that the average temperature from the classical LC
thermal model was always lower than the spatial-resolution LC thermal model and rather close to the surface temperature
of the spatial-resolution LC thermal model. It was concluded from the comparison of the temperature results that the spatial-
resolution, LC thermal model was very accurate and capable of fast real-time temperature forecast and control for smart bat-
tery thermal management.

Acknowledgments

C.P. Author thanks the Office of the Vice President for Research at the University of Nevada, Reno for the partial financial
support by Junior Faculty Research Grant.

Appendix A. Analytical Solution Using Green function

The energy equation considered has a space/time-dependent, volumetric battery heat generation s(r, t), and a time-depen-
dent convective thermal boundary condition hf(t)[T(R, t)  T1(t)]. The time-dependent battery heat generation s(r, t) allows
using arbitrary dynamic battery loading conditions [25–27]. The exact solution was solved using Green function method
due to its certain advantage over other methods of solution such as separation of variables or Duhamel’s theorem. The Green
function method is very effective to solve the energy equation with time- and spatially-dependent heat generation and time-
dependent boundary conditions.
The Fourier series expansion and Green function were used to solve the energy equation for the aforementioned general-
ized conditions. The Green function method uses the summation of each boundary condition that contributes to the resulting
temperature. Therefore, the solution of the boundary-value problem was obtained in terms of the temperatures for initial
condition, battery heat generation, and convective cooling shown below
Tðr; tÞ ¼ T o ðr; tÞ þ T s ðr; tÞ þ T a ðr; tÞ ðA:1Þ
2800 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801

The derivation of the Green function for a particular problem can be found using several methods and some solutions are
discussed in the literatures [23,25–27]. The most suitable method is by solving the homogenous parts of the problem. The
solution of the homogenous part of this problem can be solved by the method of separation of variable and is given by
2 3
Z R X
1 2 0
2 bm J 0 ðbm r=RÞJ 0 ðbm r =RÞ5 0
uðr; tÞ ¼ T o ðr Þ4 0
exp½a 2
s bm t=R    2pr0 dr ðA:2Þ
2 2 2 2
r 0 ¼0 m¼1 pR bm þ Bi J0 ðbm Þ
s¼0

where, Bi (=hfR/ks) is the cell Biot number which is the ratio of conduction thermal resistance to convective thermal resis-
tance [24]. J0 and J1 are the Bessel functions of the first-kind. bm is the solution of the eigenvalue equation below

bm J 1 ðbm Þ ¼ BiJ 0 ðbm Þ ðA:3Þ

Again, the homogenous solution of the problem can be written in terms of the Green function as
Z R
0
uðr; tÞ ¼ T o ðr 0 Þ½Gðr; t; r 0 ; sÞs¼0 2pr 0 dr ðA:4Þ
r 0 ¼0

Equating Eqs. (A.2)–(A.4), the Green function G(r, t; r 0 , s)js=0 is given by

X
1 h i b2 J ðb r=RÞJ ðb r0 =RÞ
Gðr; t; r0 ; sÞjs¼0 ¼ exp as b2m t=R2 m 20 2m o m
ðA:5Þ
m¼1 pR ðbm þ Bi2 ÞJ20 ðbm Þ
The Green function G(r, t; r 0 , s) is obtained by replacing t in the above Eq. (9) by t  s and is given by

X
1 h i b2 J ðb r=RÞJ ðb r 0 =RÞ
Gðr; t; r0 ; sÞ ¼ exp as b2m  ðt  sÞ=R2 m 0  m 0
m ðA:6Þ
m¼1 pR2 b2m þ Bi2 J20 ðbm Þ
where, the Green function G(r, t; r0 , s) represents the temperature distribution in this region, which is initially at zero temper-
ature, and subjected to homogeneous boundary condition, due to an impulsive point heat source of strength unity located at
any location r ¼ r 0 and releasing its spontaneously at time s [23]
Therefore, the contribution from the initial temperature to the solution is given by
Z R
0
T o ðr; tÞ ¼ T o ðrÞ½Gðr; t; r 0 ; sÞs¼0 2pr0 dr ðA:7Þ
r0 ¼0

Similarly, the contribution from the battery heat generation is given by

Z t Z R
as 0
T s ðr; tÞ ¼ Gðr; t; r 0 ; sÞsðr 0 ; sÞ2pr0 dr ds ðA:8Þ
s¼0 r 0 ¼0 ks
Finally, the contribution from the convective cooling is given by
Z t
as
T a ðr; tÞ ¼ ½Gðr; t; r0 ; sÞ2pr 0 r0 ¼R hf ðtÞT 1 ðtÞds ðA:9Þ
s¼0 ks
Therefore, the resultant solution in a general form is given by
Z RX1 h i b2 J ðb r=RÞJ ðb r 0 =RÞ
0
Tðr; tÞ ¼ T o ðrÞ exp as b2m t=R2 m 0  m 0
m 2pr0 dr
r 0 ¼0 m¼1 pR2 b2m þ Bi2 J20 ðbm Þ
Z t Z R
as X1
b2 J ðb r=RÞJ 0 ðbm r0 =RÞ 0 0
þ exp½as b2m  ðt  sÞ=R2  m 0  m  sðr ; sÞ2pr 0 dr ds
s¼0 r0 ¼0 ks m¼1 2 2 2 2
pR bm þ Bi J0 ðbm Þ
Z t X h i
2as 1
b2 J ðb r=RÞ
þ exp as b2m  ðt  sÞ=R2  m 0 m  hf ðtÞT 1 ðtÞds ðA:10Þ
s¼0 ks m¼1 2
R b2m þ Bi J 0 ðbm Þ

This is the general solution of the problem which includes three parts for time- and spatially-dependent heat generation
s(r, t), time-dependent heat transfer coefficient hf(t), and time-dependent ambient temperature T1(t). The exact solution
could be extended for multi-dimensional problems, if necessary. By releasing the non-required transient boundary condi-
tion, the solution can be simplified. The resultant equation was solved by a commercial solver (MATLAB) using a 1.1 mm
as the spatial interval (twenty divisions of the cell radius R) for the spatial temperature integration and one second as the
time interval for the transient temperature integration based on the trapezoidal integration formula. Because the accuracy
of the exact solutions from the analytical method can be affected by the number of nodes used for a spatial integration, the
node dependency was minimized using twenty integration nodes.
 R. Mahamud, C. Park / Applied Mathematical Modelling 37 (2013) 2787–2801 2801

References

[1] C.W. Park, A.K. Jaura, Dynamic thermal model of Li-ion battery for predictive behavior in hybrid and fuel cell vehicles, SAE 2003-01-2286.
[2] A.A. Pesaran, S. Burch, M. Keyser, An approach for designing thermal management systems for electric and hybrid vehicle battery, in: Fourth Vehicle
Thermal Management Systems Conference and Exhibition, London, UK, May 24–27, 1999.
[3] C.W. Park, A.K. Jaura, Transient heat transfer of 42 V Ni-MH batteries for an HEV application, SAE 2002-01-1964.
[4] R. Mahamud, C. Park, Reciprocating air flow for Li-ion battery thermal management to improve temperature uniformity, Journal of Power Sources 196
(2011) 5685–5696.
[5] D. Linden, Handbook of Batteries, second ed., McGraw-Hill, New York, 1994.
[6] R. Mahamud, C. Park, Spatial-resolution lumped-capacitance thermal model for battery power cycle analysis, SAE 2011-01-1362.
[7] G. Guo, B. Long, B. Cheng, S. Zhou, P. Xu, B. Cao, Three-dimensional thermal finite element modeling of lithium-ion battery in thermal abuse application,
Journal of Power Sources 195 (2010) 2393–2398.
[8] S. Al-Hallaj, H. Maleki, J.S. Hong, J.R. Selman, Thermal modeling and design consideration of lithium-ion batteries, Journal of Power Sources 83 (1999)
1–8.
[9] Y. Inui, Y. Kobayashi, Y. Watanabe, Y. Watase, Y. Kitamura, Simulation of temperature distribution in cylindrical and prismatic lithium ion secondary
batteries, Energy Conversion and Management 48 (2007) 2103–2109.
[10] K. Smith, G.H. Kim, E. Darcy, A. Pesaran, Thermal/electrical modeling for abuse-tolerant design of lithium ion modules, International Journal Energy
Research 34 (2010) 204–215.
[11] R. Parrish, K. Elankumaran, M. Gandhi, B. Nance, P. Meehan, D. Milburn, S. Siddiqui, A. Brenz, Voltec battery design and manufacturing, SAE 2011-01-
1360.
[12] S. Al-Hallaj, H. Maleki, J.S. Hong, J.R. Selman, Thermal modeling and design considerations of Li-ion, Journal of Power Sources 83 (1999) 1–8.
[13] M.C. Hermite, Sur la Formule d’Interpolation de Lagrange, J. Crelle 8 (1878).
[14] J. Mennig, T. Auerbach, W. Halg, Two point hermite approximations for the solution of linear initial value and boundary value problems, Computer
Methods in Applied Mechanics and Engineering 39 (1983) 199–224.
[15] R.M. Cotta, Improved lumped-differential formulation of diffusion problems, in: B. Sunden, M. Faghri (Eds.), Modeling of Engineering Heat Transfer
Phenomena, vol. 2, Computational Mechanics Publications, United Kingdom, 1999.
[16] T.I. Evans, R.E. White, A thermal analysis of a spirally wound battery using a simple mathematical model, Journal of The Electrochemical Society 136
(1989) 2145–2152.
[17] V. Srinivasan, C.Y. Wang, Analysis of electrochemical and thermal behavior of Li-ion cells, Journal of The Electrochemical Society 150 (2003) A98–A106.
[18] S. Al-Hallaj, R. Venkatachalapathy, J. Prakash, J.R. Selman, Entropy changes due to structural transformation in the graphite anode and phase change of
the LiCoO2cathode, Journal of The Electrochemical Society 147 (2000) 2432–2436.
[19] K.E. Thomas, C. Bogatu, J. Newman, Measurement of the entropy of reaction as a function of state of charge in doped and undoped lithium manganese
oxide, Journal of The Electrochemical Society 148 (2001) A570–A575.
[20] K. Smith, C.Y. Wang, Power and thermal characterization of a lithium-ion battery pack for hybrid-electric vehicles, Journal of Power Sources 160 (2006)
662–673.
[21] K. Smith, C.Y. Wang, Solid-state diffusion limitations on pulse operation of a lithium ion cell for hybrid electric vehicles, Journal of Power Sources 161
(2006) 628–639.
[22] M. Doyle, J. Newman, Comparison of modeling prediction with experimental data from plastic lithium ion cells, Journal of The Electrochemical Society
143 (1996) 1890–1903.
[23] N. Ozisik, Heat Conduction, second ed., John Wiley & Sons, Inc, 1993.
[24] F.P. Incropera, D.P. Dewitt, Fundamentals of Heat and Mass Transfer, fifth ed., John Wiley & Sons, 2001.
[25] H.S. Carslaw, J.C. Jaegar, Conduction of Heat in Solids, second ed., Oxford university press, 1959.
[26] J.V. Beck, K.D. Cole, A. Haji-Sheikh, B. Litkoui, Heat Conduction using Greens Functions, Hemisphere publication, 1992.
[27] M.D. Greenberg, Application of Green’s Functions in Science and Engineering, Prentice-Hall, Inc, New Jersey, 1971.