400 Commonwealth Drive, Warrendale, PA 15096-0001 U.S.A. Tel: (724) 776-4841 Fax: (724) 776-0790 Web: www.sae.

org
SAE TECHNICAL
PAPER SERIES
2006-01-3478
Cooling Airflow Simulation for Passenger Cars
using Detailed Underhood Geometry
Ales Alajbegovic and Raja Sengupta
Exa Corporation
Wilko Jansen
Jaguar Cars Ltd./Land Rover
Commercial Vehicle Engineering
Congress and Exhibition
Chicago, Illinois
October 31-November 2, 2006
THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
The Engineering Meetings Board has approved this paper for publication. It has successfully completed
SAE's peer review process under the supervision of the session organizer. This process requires a
minimum of three (3) reviews by industry experts.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise,
without the prior written permission of SAE.
For permission and licensing requests contact:
SAE Permissions
400 Commonwealth Drive
Warrendale, PA 15096-0001-USA
Email: permissions@sae.org
Tel: 724-772-4028
Fax: 724-776-3036
For multiple print copies contact:
SAE Customer Service
Tel: 877-606-7323 (inside USA and Canada)
Tel: 724-776-4970 (outside USA)
Fax: 724-776-0790
Email: CustomerService@sae.org
ISSN 0148-7191
Copyright 2006 SAE International
Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE.
The author is solely responsible for the content of the paper. A process is available by which discussions
will be printed with the paper if it is published in SAE Transactions.
Persons wishing to submit papers to be considered for presentation or publication by SAE should send the
manuscript or a 300 word abstract to Secretary, Engineering Meetings Board, SAE.
Printed in USA
THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
2006-01-3478
Cooling Airflow Simulation for Passenger Cars using Detailed
Underhood Geometry
Ales Alajbegovic and Raja Sengupta
Exa Corporation
Wilko Jansen
Jaguar Cars Ltd./Land Rover
Copyright 2006 SAE International
ABSTRACT
Air flow in the underhood area is the primary source of
engine cooling. A quick look at the vehicle underhood
reveals exceptionally complex geometry. In addition to
the engine, there are fans, radiator, condenser, other
heat exchangers and components. The air flow needs to
have adequate access to all relevant parts that require
cooling. Due to complex geometry, the task to ensure
sufficient air cooling is not a simple one. The air flow
entering from the front grille is affected by many
components on its path through the underhood. Even
small geometry details affect the flow direction and can
easily cause recirculation regions which reduce the
cooling efficiency. Therefore, air cooling flow analysis
requires detailed treatment of the underhood geometry
and at the same time accurate air flow modeling. Recent
advances in the lattice-Boltzmann equation (LBE)
modeling are allowing both. Presented is analysis of air-
cooling flow over a passenger vehicle, Land Rover LR3,
that includes detailed external and underhood geometry.
Simulations were performed using PowerFLOW, an LBE
based flow solver. Shown are flow structures that form at
95 km/h and their effect on the overall cooling flow rate
and coolant temperatures in the heat exchangers.
INTRODUCTION
The most important goal for the new vehicle design is to
reduce the number of prototypes and being able to make
components selection and positioning early in the design
process. All this is possible only with the use of
simulation tools. Being able to have information on the
cooling performance in the concept stage allows
reduction in the number of needed prototypes and
shortening of the design cycle.
The underhood thermal management plays an essential
role in the design of vehicles. There needs to be
sufficient airflow available to adequately cool the
underhood components. The amount of airflow depends
on the positioning and size of the grilles, fan operation,
and the positioning of the underhood components. There
is a considerable ongoing effort to develop fluid flow
simulation capability coupled with heat transfer
calculations. Srinivasan et al. [1] presented an omni-tree
meshing technology for rapid mesh generation for
Navier-Stokes solvers. The cooling airflow simulations
using Navier-Stokes solvers coupled with heat
exchanger calculations for passenger cars were
presented by [2], [3], [4], and for trucks by [5], [6], [7].
Fortunato et al. [8] used the lattice-Boltzmann equation
(LBE) solver for the cold flow over the entire car
including the underhood and a Navier-Stokes solver for
the underhood flow. The velocity field on the entrance
surfaces into the underhood area calculated by the LBE
solver was used as the inlet boundary condition to the
Navier-Stokes solver calculation. The present work is
somewhat an extension of this work. The simulations
here were performed using the LBE solver to simulate
the flow over the passenger car exterior and through the
underhood area including heat transfer. This approach
leverages all advantages of an LBE solver applied to the
simulation of the underhood thermal management.
MATHEMATICAL MODEL
One of the major challenges in performing fluid simulations
in CAE is the difficulty to adequately discretize the flow
domain and at the same time resolve surface details of
complex geometries. The Navier-Stokes solvers that are
mostly used in industrial applications are very demanding
on the quality and size of the meshes for practical
calculations. These requirements are a direct
consequence of the properties of the finite volume method
(FVM) which is the most used approach in the numerical
solution of the Navier-Stokes equations. This method
employs the pressure-correction algorithm for imposing the
mass conservation. Pressure-correction is an iterative
procedure that is expensive numerically and therefore
limits the effective mesh size that can be used for practical
THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
calculations. In order to limit the number of cells,
unstructured meshes are used which allow the
discretization of complex geometries while maintaining low
cell counts. It is not uncommon for some generated cells in
unstructured meshes to have very distorted shapes that
cause calculation inaccuracies or even instabilities and
subsequent divergence of the calculation. The varying cell
shapes in unstructured meshes cause inevitable
dependence of the results on the mesh size and quality.
Mesh generation for the Navier-Stokes solvers made
significant progress in the last years and considerable part
of the process is automized. However, the preparation of
surfaces and some mesh generation procedures still
require users involvement. This part remains to be labor
intensive and depends strongly on the skill of the user to
manipulate meshes. The entire meshing process limits the
level of geometry detail that can be handled in practical
use and in most cases the geometry is simplified to reduce
the preparation time and fulfill the design timelines.
Lattice-Boltzmann equation (LBE) solvers represent an
alternative to the Navier-Stokes solvers. They dont need
any special iterative procedure and fulfill mass,
momentum, and energy conservation by design. As a
result, LBE solvers are numerically very efficient and
robust. The increased numerical efficiency allows
handling of lattices with very large number of elements
(or voxels). In addition, the properties of Boltzmann
equation allow an improved treatment of the interaction
of the fluid flow with the wall surface. Surface elements
(or surfels) are designed as active elements that interact
with the neighboring lattice elements. The combination
of both large lattices and dynamic surface treatment
allow accurate representation of surfaces without the
need for geometry simplification. This feature is
paramount for the deployment of simulation techniques
in the design process. In addition, the LBE solvers are
linearly scalable with the number of processors.
The lattice-Boltzmann equation approach to simulate
fluid flows was demonstrated by Frisch, Hasslacher, and
Pomeneau [9]. Since then, there was considerable effort
invested into the development of a lattice-Boltzmann
based flow solver and there are good reviews available
on the current status [10], [11]. Here is presented only a
very short overview of the method. The reader should
consult the provided references in case of interest for
more detailed information.
The Boltzmann equation can be written in the following
form:
f f
t x

+ =

r
r
(1)
where f is the velocity probability distribution function
and is the collision operator. Boltzmann equation in
the lattice form can be expressed as a set of algebraic
equations for probability distribution at each state,
i
f :
( , ) ( , ) ( , )
i i i i
f t t x e t f t x t x + + = +
r r r r
(2)
where the collision operator is modeled using the BGK
form [12]:
(0)
( , ) ( ( , ) ( , ))
i i i
t
t x f t x f t x

=
r r r
(3)
The equilibrium distribution function,
( ) 0
i
f , is expanded
with respect to the velocity [12]:
(0)
2 2 2
( , ) (1 ( ))
2
i i
i
i p
s s s
u u e e
e u
f t x t
c c c

= + +
r
(4)
Important feature of this approach is that the mean
quantities like density, , and velocity, u
r
, are obtained
from the calculated velocity probability distribution
functions:
( , ) ( , )
i
i
t x f x t =

r r
(5)
1
( , ) ( , )
( , )
i i
i
u t x f x t e
t x
=

r r r r
r
(6)
It can be shown that the described system yields the
behavior of the Navier-Stokes equations. More precisely,
the Navier-Stokes equations represent a subset of the
behavior modeled by the Boltzmann equation in the limit
of small Knudsen numbers. A nice feature of the lattice-
Boltzmann equation solver is that the mass and
momentum conservation are automatically fulfilled by
enforcing the sum of the collision term for all states, i, to
be equal zero:
( , ) 0
i
i
t x =

r
(7)
( , ) 0
i i
i
t x e =

r r
(8)
Turbulence effects are modeled using a modified k-
model based on the original RNG formulation [13, 14]:
( )
1
t
Dk k
P
Dt x x

= + +



r r
(9)
( )
( )
1
3
2
0
2
3
1
1 /
1
t
D
C P
Dt x x k
C C
k

= + +

+
+

r r
(10)
The values for the dimensionless coefficients are the
same as in the original model formulation [13, 14]. This
THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
LBE based description of turbulent fluctuation carries
flow history and upstream information, and contains high
order terms to account for the nonlinearity of the
Reynolds stress [15]. This is contrasted with typical
Navier-Stokes solvers, which tend to use the
conventional linear eddy viscosity based on the
Reynolds stress closure models.
Temperature evolution is calculated using the PDE in
the following form:
Pr Pr
p t p
p
t
c c
DT T
c Q
Dt x x

= + +




&
r r
(11)
Turbulence and temperature equations are solved on
the same lattice using a modified Lax-Wendroff-like
explicit time marching finite difference scheme.
Simulation presented in this work was performed using
the solver described in the following references [16],
[17], [18], [19]. The simulation tool is also available
commercially [20].

SIMULATION DETAILS
Performed is the simulation of the airflow over the
vehicle exterior and inside the underhood of the Land
Rover LR3. The simulation is done using the LBE solver
[20]. All geometry details are considered without any
simplifications. Car geometry is shown from different
perspectives in Figure 1. The geometry is clearly of
considerable complexity and includes all underhood and
underbody details. Lattice used in the simulation was
generated automatically and without any intervention
from the user. The resulting lattice size had around 8
million fine equivalent voxels (number of voxels that are
updated every time step) and approximately the same
number of surfels (Table 1).

The parameters used in the simulation are shown in
Table 1. Vehicle was placed in the digital wind tunnel
(DWT). The vehicle front area was about 2% of the
overall wind tunnel front face. The air velocity of 95 km/h
was imposed at the inlet boundary of the DWT resulting
in uniform air distribution of the upstream air. In addition
to the airflow, the heat transfer between airflow and heat
exchangers is modeled as well. The interaction is
calculated using the one-dimensional tool or 1D-tool
[22]. The 1D-tool calculates the heat transfer between
the coolant flow inside and airflow through the heat
exchanger. The calculation approach considerably
simplifies the flow configuration essentially assuming
coolant flow as one dimensional this is the reason for
the name 1D-tool.

Figure 1. Fully resolved Land Rover LR3 geometry used
in the simulation showing external body, underhood and
underbody details.


Operating Condition 95GVW
Air velocity 95 km/h
Ambient air temperature 31.3 C
Size [MFe (V/S)] 8.2/7.7
Charge Air Cooler (CAC)
m
coolant
=0.097kg/s
T
charge inlet
=195 C
Radiator
m
coolant
=1.77kg/s
Heat input=65.2kW
Condenser Heat input=12.0kW
Fan 2440 RPM
Table 1. Set-up parameters.













THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
Figure 2. Details of the simulation preparation: a.) front
grille geometry, b.) heat exchangers, and c.) fan
geometry.

The front grille geometry is shown in Figure 2a. The
grille geometry has dominant influence on the amount of
airflow entering the underhood area. After the grille, the
air proceeds towards the heat exchangers. The heat
exchanger cores are represented by the rectangular
boxes shown in Figure 2b. The fan is placed right behind
the heat exchangers as shown in Figure 2c. Used is
detailed fan geometry and the flow in the fan is modeled
using the multiple-reference-frame (MRF) approach. The
implemented MRF model has been validated extensively
and the predicted fan performance curves were well
within 5% with respect to the measured data. High air
velocity of the upstream air is limiting the fan influence
on the airflow. The detailed configuration of the heat
exchangers from the side view is shown in Figure 3. The
charge-air-cooler (CAC) is placed beneath the
condenser and the radiator.
The heat exchangers are modeled as porous media.
The resistance of the heat exchangers is imposed from
the measured relation between the air velocity and the
pressure drop through the heat exchanger. The obtained
data is correlated using Darcys law [21].
Heat transfer between the airflow and coolant in the heat
exchanger is modeled with the 1D-tool [22]. The 1D-tool
does not model all the internal details of the heat
exchanger. It models the heat flow that propagates in
the heat exchanger in the flow direction. The heat
transfer coefficients used in the calculation of the heat
transfer between the coolant and air flow is obtained in a
calorimeter. The measurement error in heat transfer
coefficients was about 5% for the majority of the data
points. The measured heat transfer coefficients as a
function of the coolant and air mass flow rates are
shown in Figure 4. The measured data is interpolated
using the sandwich formula and the interpolation curves
are shown in red. Sandwich formula is represented by
the following function:
1
1 1
a h c
a c
Htc
K D K
m m

=

+ +


& &
(12)
where
a
m& is air mass flow rate,
c
m& is coolant mass flow
rate, and
a
K ,
c
K ,
h
D , , are the interpolation
coefficients that need to be calculated. It can be seen
that the interpolation is quite good for the CAC, but not
as good for the radiator. Therefore, the calculated
interpolation is used only for the CAC. Bi-linear
interpolation is used for the calculation of the heat
transfer in the radiator. Condenser is not modeled with
the 1D-tool. The known condenser heat rejection is
uniformly distributed through the condenser volume.
This approach proved to provide very good results while
allowing to avoid dealing with the complexity of the
condenser physics.
Figure 3. Configuration of the underhood geometry.


THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
Figure 4. Heat transfer coefficients at different coolant
mass flow rates for a.) the radiator and b.) the charge-
air-cooler.

RESULTS
The simulation was performed on a 64 processor Linux
cluster. The simulation was run for 30,000 time steps
before starting the coupling with the 1D-tool. This time
interval is needed for the flow to settle. After that, the
coupling between the 1D-tool and the LBE solver was
done every 4,000 time steps. The entire simulation time
was 66 hours, resulting in 4224 CPU hours.
The evolution of the coolant inlet temperature (TTT),
inlet and outlet air temperatures, and air mass flow rates
through the radiator and CAC as a function of the
coupling iteration is shown in Figure 5. The calculated
values cannot be shown due to the confidentiality
issues.
It can be observed that the steady state is achieved after
only six (6) coupling iterations. Consequently, the
simulation didnt need to run as long as it did and could
be stopped approximately after 2,500 CPU hours. Of
course, the other option to reduce calculation times is to
increase number of processors. The LBE solvers are
linearly scalable with the number of processors.
Figure 5. Evolution of the air mass flow rate, coolant
entry temperature (TTT), inlet and outlet air
temperatures as a function of coupling iteration with the
1D-tool for a.) radiator and b.) charge-air-cooler.

The most important calculated parameter of interest was
the coolant temperature at the inlet into the radiator also
known as the top tank temperature. In this case, the
calculated value was exactly 1C higher than the
measurement. The overall temperature difference
between the inlet and outlet coolant temperature in
radiator was around 10C. Such accuracy is not
uncommon for the presented simulation methodology
using the LBE solver, even for much higher heat
rejections. High level of geometry resolution allows
accurate predictions of the airflow through the vehicle
underhood area and consequently accurate predictions
of the air mass flow rates through the heat exchangers.
Correct prediction of the air mass flow rate is paramount
for the correct prediction of the coolant inlet temperature.
It is interesting to observe flow structure and
temperature field in the underhood area. The two fields
are shown both on the centerline and horizontal planes
in Figure 6. Flow accelerates through the grille openings
and enters the condenser and CAC. Flow heats up in
the condenser and continues through the radiator with
further increase in temperature. Even larger heating

THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
rates can be observed in the CAC. The flow is complex,
but doesnt show any recirculation patterns that could
result in the feedback between the exit and inlet
temperatures into the heat exchangers.
One of the important features of the presented method is
the ability to resolve both underhood and external
aerodynamics flows simultaneously as shown in Figure
7. Observed can be typical characteristics of the external
air flow over the vehicle body. In this particular case the
set-up didnt use the placement of variable resolution
(VR) regions that would be necessary for the detailed
resolution of the external flow features. The reason for
this was to simulation focus on the flow in the underhood
area. However, joint resolution of the exterior and
underhood flows is possible and only increases the case
size and simulation times. At the same time, the need to
prepare only one case significantly reduces geometry
and case preparation times.
Figure 6. Velocity and temperature fields on centerline
and horizontal planes through the underhood.

Figure 7. Details of the external flow aerodynamics.

CONCLUSION
Performed was the simulation of the airflow over the
Land Rover LR3 with complete geometry details of the
external body and the underhood area. The employed
simulation method doesnt require geometry
simplifications. The generation of the simulation lattice
(flow domain discretization) is entirely automatic and
doesnt require any user intervention. The interaction
between the airflow and heat exchangers was simulated
using the 1D-tool. Special attention in the calculation
was placed in the resolution of the underhood flow. The
predicted inlet coolant temperature (top tank
temperature) was exactly 1C higher than in the
experiment.







THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM
REFERENCES
1. K. Srinivasan, Z.J. Wang, W. Yuan, R. Sun, Vehicle
thermal management simulation using a rapid omni-
tree based adaptive Cartesian mesh generation
methodology, HT-FED2004-56748, 204 ASME Heat
Transfer/Fluids Engineering Summer Conference,
July 11-15, Charlotte, North Carolina, USA.
2. B. Uhl, F. Brotz, J. Fauser, U. Krueger,
Development of engine cooling systems by
coupling CFD simulation and heat exchanger
analysis programs, SAE 2001-01-1695.
3. G. Seider, F. Bet, T. Heid, U. Hess, T. Klein, and J.
Sauer, A numerical simulation strategy for complex
automotive cooling systems, SAE 2001-01-1722.
4. H. Knaus, C. Ottosson, F. Brotz, W. Kuehnel,
Cooling module performance investigation by
means of underhood simulation, SAE 2005-01-
2013.
5. T.P. Nobel, S.K. Jain, A multidimensional approach
to truck underhood thermal management, SAE
2001-02-2785.
6. C.L.R.Siqueira, P. Vatavuk, M. Jokuszies, M.R.
Lima, Numerical simulation of a truck underhood
flow, SAE 2002-01-3453.
7. E.A. Costa, CFD approach on underhood thermal
management of passenger cars and trucks, SAE
2003-01-3577.
8. F. Fortunato, F. Damiano, L. Di Matteo, P.Oliva,
Underhood cooling simulation for development of
new vehicles, SAE 2005-01-2046.
9. U. Frisch, B. Hasslacher, and Y. Pomeneau, Lattice
gas automata for the Navier-Stokes equation,
Physical Review Letters, 56:1505-1508, 1986.
10. S. Chen and G. D. Doolen, Lattice Boltzmann
method for fluid flows, Annual Review of Fluid
Mechanics, 30:329-364, 1998.
11. S. Succi, The Lattice Boltzmann Equation for Fluid
Dynamics and Beyond, Series Numerical
Mathematics and Scientific Computation, Clarendon
Press, Oxford, 2001.
12. D. dHumieres, P. Lallemand and Y. H. Quian,
Lattice BGK models for Navier-Stokes equations,
Europhysics Letters, 17(6):479-484, 1992.
13. V. Yakhot, and S.A., Orszag, Renormalization
Group Analysis of Turbulence. I. Basic Theory J.
Sci. Comput., 1(2), 3-51, 1986.
14. V. Yakhot, V., S.A. Orszag, S. Thangam, T. Gatski,
and C. Speziale, Development of turbulence
models for shear flows by a double expansion
technique, Phys. Fluids A, 4 (7), 1510-1520, 1992.
15. H. Chen, S.A. Orszag, I. Staroselsky, and S. Succi,
Expanded Analogy between Boltzmann Kinetic
Theory of Fluid and Turbulence, J. Fluid Mech.,
519: 307-314, 2004.
16. H. Chen, H-theorem and generalized semi-detailed
balance conditions for lattice gas systems, J. Stat.
Phys. 81:347-359, 1995.
17. H. Chen and C. Teixeira, H-Theorem and origins of
instability in thermal lattice Boltzmann models,
Comp. Phys. Communication, 129:21-31, 2000.
18. H. Chen and R. Zhang, Lattice Boltzmann method
for simulations of liquid-vapor thermal flows, Phys.
Rev. E67(6): Art. no. 066711 Part 2, 2003.
19. C. M. Teixeira, Incorporating turbulence models into
the lattice-Boltzmann method,. Int. J. Modern
Physics C, 9(8):1159-1175, 1998.
20. PowerFLOW Users Guide, Release 4.0, Exa
Corporation, Boston, Massachusetts, 2006.
21. D.M. Freed, Lattice Boltzmann method for
macroscopic porous media modeling,. Int. J.
Modern Physics C, 9(8):1491-1504, 1998.
22. PowerCOOL Users Guide, Release 4.0, Exa
Corporation, Boston, Massachusetts, 2006.


CONTACT
Ales Alajbegovic
Exa Corporation
17177 N. Laurel Park Drive
Livonia, MI 48152
email: ales@exa.com
THIS DOCUMENT IS PROTECTED BY U.S. AND INTERNATIONAL COPYRIGHT.
It may not be reproduced, stored in a retrieval system, distributed or transmitted, in whole or in part, in any form or by any means.
Downloaded from SAE International by Bo Feng, Friday, April 05, 2013 07:03:28 PM