Electro-thermal simulation of Lithium Ion battery for

EV/HEV applications

L. Kostetzer; S. Nallabolu; E. Rudnyi; L. Krüger, CADFEM GmbH,

Grafing bei München, Germany
M. Geppert; D. Quinger, LION Smart GmbH

Abstract
Highly power energy storage systems are necessary for the new age of electric
vehicles. Lithium ion batteries have the advantage of high energy density, good aging
characteristics and high efficiency, but at the same time the thermal range of operation is
limited. By temperature under 0°C, the power capacity of the lithium battery is reduced about
70% and by temperature over 40°C irreversible damage (over 70 °C also thermal runaway)
could happened. Hence an efficient and accurate thermal management is necessary.
A battery cooling system is analyzed and designed with aid of numerical tools; we
propose a modeling methodology starting with CFD and ending at a low dimensional but
accurate compact thermal model for system level simulation. This model is ready to be
coupled with other physical domains like electrical, chemical and mechanical. Detailed CFD
simulation is performed to analyze the air flow over corrugated channels. Heat transfer
coefficients are calculated with CFD and then used in a finite element (FEM) thermal model
of a Li-Ion battery pack with one-dimensional flow (FLUID116). After that model reduction
produces a compact thermal model for system level simulation.
Our electrical Li-ion battery model is based on work [9] and it can be classified as
semi-physical. It is based on electrochemical equations developed by the group of Prof
Newman [13] but with simplified assumptions to speed model simulation. The parameters of
the model still have some physical meaning but they should be determined during a
parameterization procedure. The battery model is developed in VHDL-AMS that make it
simple to use it in many different environments.

Introduction. Electro thermal simulation at system level

Electro thermal simulation describes the interaction between electrical and thermal
components. Electrical circuits are temperature dependent and at the same time the circuit
produces power dissipation. On the other side, the thermal subsystem takes power
dissipation and evaluates temperatures in the system according to the heat transfer laws
(see Fig 1).

Fig. 1 Simple electrothermal simulation

As a simple example, let us take a temperature dependent resistance shown on the

left in Fig 1. When the current passes through the resistor, it generates heat that in turn goes
through a thermal mass and thermal resistor (on the right in Fig 1). The temperature of the
system goes back to the electrical model of the resistor and thus we have two-way coupled
electrothermal simulation.
The thermal model in Fig 1 is pretty simple and the question arises how one can
develop it in the general case. Finite element modeling would be the best solution because
with available software thermal modeling has already become a routine procedure. In
principle, one can directly convert a thermal model into a circuit model. After the
discretization by the finite element method one obtains an equation that can be considered
as an electrical network where the vector T will be equivalent to unknown voltages, the
matrix E will be a capacity matrix and the matrix K the resistance matrix.

ET KT F (1)
The problem along this way is that the vector T is usually high-dimensional, say
several hundred thousand degrees of freedom, and hence the Eq 1 is not compatible with
system level simulation because of its large dimensionality.
 Fig. 2 The idea of model order reduction

Model reduction [2] allows us automatically to reduce the dimension of Eq 1

preserving at the same time good accuracy of the dynamic response (see Fig. 2). The
reduced thermal model now can be used in a system level simulation to integrate to another
domains e.g. electrical cell models, vehicle dynamics and power electronics. We apply model
order reduction in order to develop a compact thermal model of a real battery pack in this
work.
In a battery pack the cooling process is mainly done by conduction and convection
phenomena. With a thermal model described by equation (1) the convection is defined by a
boundary condition with the convection heat transfer coefficient or simply the film coefficient.
The film coefficient can be obtained by using correlations, yet it may not be accurate in the
case of complex geometry. In this work we use computational fluid dynamics to evaluate the
film coefficient.
The finite element of the battery pack is briefly described in the next section. After that
a CFD methodology is presented followed by a comparison with a finite element thermal
model of the cooling channel. Finally the finite element thermal model is reduced to a system
level by model order reduction and coupled with an electrical model of the battery cell model
in Simplorer.

Battery pack thermal model

The finite element model of a battery pack has been developed by the company Lion
Smart in ANSYS [3][4] (see Fig. 3). The model consists from 33 individual batteries that have
been cooled through pipes modeled by means of one dimensional FLUID116 elements in
ANSYS. The finite element model has the dimension roughly 50000 degrees of freedom and
its transient simulation for 100 timesteps takes 40 min.
 The convection boundary condition connects the fluid elements with the surface
elements and consequently to solid elements that represents aluminum plates and battery
cells. The film coefficient was evaluated by a CFD analysis of the detailed fluid flow geometry
according to the methodology described in the next section.

Fig. 3 The battery pack model (http://www.lionsmart.de/)

CFD of cooling channels

Cooling channels of the present battery pack have a sinusoidal cross section with
metallic walls, see Fig. 4. Due to periodical and symmetrical pattern a simplified CFD domain
is chosen, dashed area in Fig. 4. Battery (solid material) is also included to account for heat
conduction.
Hybrid mesh with prismatic and hexahedron elements is generated in ANSYS
meshing with small size hexahedrons in the boundary layer for higher accuracy in the film
coefficient predictions (see Fig. 4 right).
 Fig. 4. CFD domain and mesh

A typical operation condition is with velocities around 10m/s for air cooling. The
correspondent Reynolds number falls inside the turbulent region; in the order of 3000. The
SST turbulence model is used due to good resolution of boundary layer effects and stability
in the free stream region.
Heat generation is present in the battery region. The heat conductivity in the battery
cells was modeled as orthotropic with lower value in the vertical direction of the cross
section. Uniform temperature and velocity profiles are used for air inlet condition. The outlet
is average constant pressure.
A steady state model with high resolution of advection terms is solved in ANSYS-CFX
v12. The fluid flow and the thermal boundary layer are already fully developed in the first 1/3
of the domain, see Fig. 5 right.
Sinusoidal walls influence strongly the cooling effect of the air because the heat
transmitted through the sinusoidal walls creates additional contact area for convection. This
is helped by the good thermal conductivity of the aluminum. Just 30% of the heat generated
is transferred to air by the upper and lower surfaces; inclined surfaces correspond to the
complementary energy, which quantify the thermal effects of these walls.
 Fig. 5. Temperature field in the outlet (mirrored results) and temperature of air along the
domain (channel center).

The definition of the film coefficient is an engineering decision and in this case
influences the finite element model that receives the film coefficient information (battery pack
model and channel model). The effect of the sinusoidal walls is included in the film coefficient
and than a discretization of those walls is not necessary in the FE model. Film coefficient is
calculated by Equation 2 where the heat transfer of the wall is defined in the selected areas
in the Fig. 6. An average in area is used for the heat flux. Temperature of the wall uses the
same area and average. The fluid temperature is volume averaged. To exclude boundary
layer developing effects the averaging formulas are applied in the selected region near the
outlet, see Fig. 7. The average film coefficient is 164.4 W m-2 K-1.

qWall
hAve (2)
TWall TFluid

Fig. 6 Area definition of the film coefficient

Fig. 7. Volumetric fluid location of averaging procedures for film coefficient calculation
Thermal modeling (FEM)
A comparison of the finite element model of the cooling channel with the CFD is
needed to validate the film coefficient numerical experiment and to allow further use of the
methodology in the battery pack thermal model.
The mathematical domain in the finite element thermal model is a simplification of the
CFD domain, Fig. 4. Here the fluid flow is modeled by one dimensional elements and without
sinusoidal walls. The convection boundary conditions are used with the film coefficient
calculated in the CFD analysis. Air mass flow, inlet temperature and cell heat generation are
identical to the CFD simulation. ANSYS FEM solver is used to solve the steady-state
simulation.
Maximum and minimum temperatures in the cell material are selected to validate the
cooling model. Results showed excellent agreement of the FE model and CFD (see Tab. 1).

Tab. 1 Thermal results CFD and Thermal model comparison

CFD Thermal
Battery, Tmax 29.41 29.41 (<0.1% )
Battery, Tmin 27.58 27.75 (0.6%)

Computational effort is drastically reduced in the FE model with the one dimensional
CFD approach, here just energy equation is solved for the fluid and not conservation of
momentum is accounted. FE mesh has 14.103 elements and the CFD 312.103. With smaller
mesh and simpler model in the FE case it is expected a reduced CPU time, in this case is
15s. For the CFD simulation is necessary 4h and 40 min in the same hardware configuration.

Model Order Reduction

Model reduction is an area of mathematics that in other words can be referred to as
approximation of large scale dynamical system [5]. Model reduction starts after the
discretization of governing partial differential equation when one obtains ordinary differential
equations (1). In order to use model reduction, Eq (1) is rewritten as
ET KT Bu
(3)
y CT
The difference is 1) splitting of the load vector to a product of a constant input matrix
B and a vector of input functions u and 2) the introduction of the output vector y that
contains some linear combinations of the state vector that are of interest in system level
simulation. Model reduction is based on an assumption that the movement of a high
dimensional state vector can be well approximated by a small dimensional subspace (Fig. 8
left). Provided this subspace is known the original system can be projected on it (Fig. 8 right).

Fig. 8 Model reduction as a projection of the system onto the low-dimensional subspace

Theory of model order reduction is found in [2]. The dimension of the reduced model
during the model reduction process is controlled by the approximation error specified by the
user. Although the model reduction methods based on the Padé approximation do not have
global error estimates, in practice it is enough to employ an error indicator [6]. In our
experience it is working reasonably well for a variety of finite element models.
In order to employ model reduction in practice one needs software. The software
MOR for ANSYS [7][8] reads system matrices from ANSYS FULL files, runs a model
reduction algorithm and then writes reduced matrices out (see Fig. 8). The process of
generating FULL files in Workbench is automated through scripting. The reduced matrices
can be read directly in MATLAB/Simulink, Mathematica, Python, Simplorer and other system
level simulation tools.

Battery pack thermal model at system level

The battery pack thermal model with film coefficient calculated by CFD is reduced to
a system level with model order reduction. The system level model is coupled with an
electrical battery cell model in Simplorer (see Fig. 9 left). For this case only three cells are
coupled with the battery pack, other cells just produce constant 1W of heat.
A sub-circuit model describes the cell model (see Fig. 9 right). The electrical model is
implemented in VHDL-AMS language based in [9]. The Lithium-ion cell is used and the
model is described according phenomena described by Newman model (physical model) [10]
using assumptions that reduce the final computational effort and the number of parameters
[11], which we name by semi-physical model.
 Kinetics of electrochemical reactions are modeled that translates to an overpotential
wich reduces the cell voltage (loss of energy), finally resulting in heat generation of the cell.
The transport of Lithium ions inside the electrolyte is described by one dimensional diffusion-
migration. Lithium diffuses inside electrodes diffusion limits can occur that translates in
current limits or cell depletion.
The heat produced is mainly by joule effect in the electrodes and chemical reaction
[12]. Physical proprieties are temperature dependent and values are taken from Dual Foil
standard input [13].
The effect of discharge current is evaluated in 25Ah cells connected in series in 1,5
and 10 C-rates, see Fig. 10 and Fig. 11. Losses effects are present in higher currents
showing a reduced total capacity and lower voltage, what can be observed in the voltage
profiles, Fig. 10. Energy losses are transformed into heat and removed by cooling channels;
the cooling conditions are made constant and the additional heat generated in higher c-rates
is illustrated by higher temperatures in the battery cells, Fig. 11.

Fig. 9 Eletro-thermal battery coupling in Simplorer

 Ansoft LLC VoltageTotal 6_System1

12.25

11.00
-I2.V [V]

9.75 10C 5C 1C
Curve Info
-I2.V
8.50 TR
-I2.V
TR
7.25 -I2.V
TR

6.00
0.00 5.00 10.00 15.00 20.00 25.00
Time*currentin/3600 [s]

Fig. 10 Total voltage in constant current discharge

Ansoft LLC Temperature 6_System1

320.00
Curve Info
TCell1.T
10C TCell2.T
315.00
TCell3.T
TCell1.T
TCell2.T
Y1 [kel]

310.00 5C
TCell3.T
TCell1.T
TCell2.T

305.00 TCell3.T

1C

300.00
0.00 5.00 10.00 15.00 20.00 25.00
Time*currentin/3600 [s]

Fig. 11 Cell temperature response in constant current discharge cases

Conclusion
A solution of an accurate battery pack thermal model in the system level is presented.
Spatial effects of cooling are taken to account through model order reduction techniques of
the thermal finite element model. The FEM model is numerically validated by comparing to
CFD simulation which provides heat transfer coefficients for FEM models. The battery pack
model is integrated with an electrical semi-physical model at a system level simulator
(Simporer).

Acknowledgment
The research was partially funded by the Federal German Ministry for Economy and
Technology based on a resolution of the German Government.
References
[1] E. B. Rudnyi. Effective Electrothermal Simulation for Battery Pack and Power
Electronics in HEV/EV. 3rd Virtual Vehicle Symposium in Graz, GSVF, 6-7 May, 2010.
[2] T. Bechtold, E. B. Rudnyi, J. G. Korvink. Fast Simulation of Electro-Thermal MEMS:
Efficient Dynamic Compact Models, Springer 2006, ISBN: 978-3-540-34612-8.
[3] Lion Smart GmbH, http://www.lionsmart.de/
[4] Daniel Quinger, Batteriemanagementsysteme und Algorithmen zur
Zustandserkennung von Lithium-Ionen-Batterien, Diploma thesis, TU-München, 2009
[5] A. C. Antoulas, Approximation of Large-Scale Dynamical Systems. Society for
Industrial and Applied Mathematic, 2005, ISBN: 0898715296.
[6] T. Bechtold, E. B. Rudnyi and J. G. Korvink, Error indicators for fully automatic
extraction of heat-transfer macromodels for MEMS. Journal of Micromechanics and
Microengineering, v. 15, N 3, pp. 430-440, 2005.
[7] E. B. Rudnyi and J. G. Korvink. Model Order Reduction for Large Scale Engineering
Models Developed in ANSYS. Lecture Notes in Computer Science, v. 3732, pp. 349-356,
2006.
[8] E. Rudnyi, From ANSYS to System Level Simulation: MOR for ANSYS, Mechatronik,
N 10, p. 34-36, 2008, http://ModelReduction.com.
[9] H. J. Bergveld, W. S. Kruijit, and P.H.L. Notten. Battery Management Systems,
Design by Modelling, volume vol. 1. Kluwer Academic Publishers, 2002.
[10] J. Newman and K.E. Thomas-Alyea. Electrochemical Systems. Electrochemical
Society Series, 2004.
[11] D. Danilov and P.H.L. Notten. Mathematical modeling of ionic transport in the
electrolyte of li-ion batteries. Electrochimica Acta, 2008.
[12] K. Smith, C. Y. Wang, Power and thermal characterization of lithium-ion battery pack
for hybrid-electric vehicles. Journal of Power Sources 160 (2006) 662-673.
[13] J.Newman, http://www.cchem.berkeley.edu/jsngrp/fortran.html 1998.