Professional Documents
Culture Documents
Series Editors
Prof. Dr.-Ing. Friedrich Pfeiffer
Prof. Dr.-Ing. Peter Wriggers
The Aerodynamics
of Heavy Vehicles:
Trucks, Buses, and Trains
Rose McCallen
Fred Browand
James Ross (Editors)
~Springer
RosE McCALLEN, PH.D.
Center for Advanced Fluid Dynamics Applications
Lawrence Livermore National Laboratory
P.O. Box 808, L-098
94551 Livermore, CA, U.S.A
ISSN 1613-7736
springeronline.com
© Springer-Verlag Berlin Heidelberg 2004
Softcover reprint of the hardcover 1st edition 2004
The use of general descriptive names, registered names, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protec-
tive laws and regulations and free for general use.
Cover design: design & production GmbH, Heidelberg
Typesetting: Digital data supplied by authors
Printed on acid-free paper 62/3020Rw -5 4 3 2 1 0
Engineering Conferences
ECI International
A Polytechnic/ Engineering Conferences Foundation Partnership
Conferences Director:
Barbara K. Hickernell, CAE
Engineering Conferences International
6 MetroTech Center
Brooklyn, NY 11201
Tel: +1-718-260-3743
Fax: +1-718-260-3754
Email:info@eci.poly.edu
www.engconfintl.org
Introduction
Keynote Papers
Paul B. MacCready
Aerodynamics and Other Efficiencies in Transporting Goods______________________ 3
Kevin R. Cooper
Commercial Vehicle Aerodynamic Drag Reduction: Historical
Perspective as a Guide______________________________________________________________________________ 9
Philippe R. Spalart, Kyle D. Squires
The Status of Detached-Eddy Simulation for BluffBodies ________________________ 29
Experimental Methods
Luis Bernal, Abdullah M Al-Garni
PIV Study of the Near Wake of a Pickup Truck_______ ____ _____ _129
Mory Gharib, Francisco, Pereira, Emilio Castano Graff
Applications ofDDPIV to Studies Associated with Road Vehicles 131
M M Koochesfahani, A. C. Goh, H. J Schock
Molecular Tagging Velocimetry (MTV) and Its Automotive
Applications__ _ _________ _________ _________ _________________ 143
R. Bommels, M Machacek, A. Landolt, T. Roesgen
Quantitative Flow Visualization for Large Scale Wind Tunnels ______________ 157
Robert J Englar
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction, Safety
Enhancement, and Performance lmprovement_______________________________________ 277
Tsun- Ya Hsu, Mustapha Hammache, Fred Browand
Base Flaps and Oscillatory Perturbations to Decrease Base Drag______________ 303
Aerodynamic Experiments
jorge Martinez, Sunil fain
Development of a Wind Tunnel Model Mounting Configuration
for Heavy Duty Trucks _____ _ _________________________________________ 517
Corey Diebler, Mark Smith
A Ground-Based Research Vehicle for Base Drag Studies at
Subsonic Speeds __ _ ___________ _ 519
G. Dumas, J Lemay
Splash and Spray Measurement and Control: Recent Progress
in Quebec______ _ __ ____________ _ _____________________________________________________ 533
R. J Gaeta, R. J Englar, G. Blaylock
Wind-Tunnel Evaluation of an Aerodynamic Heat Exchanger_______ 549
Steven Shladover
Automated Driving ofT rucks and Buses: Opportunities for
Increasing Productivity and Safety While Reducing Fuel Use
and Emissions ___________ _ __ _______ _____________ _ __ 563
Abstract
Recognizing both the pollution effects of fuel use and the likely increases of
fuel cost in coming decades puts high priority on alternative energy for trucks,
buses and trains. There are still gains available in decreasing aerodynamic drag
and rolling friction, using efficient engines, and minimizing fuel waste, but it
is appropriate to explore decisions that would be suitable if fossil fuel were
deemed unattractive. One then would consider utilizing natural energy (sun-
light, wind, wave), getting energy from braking, employing hydrogen, putting
different priorities on trucks vs. buses vs. trains, exploring integration with
water deliveries and automatic local air transport, etc. Such an investigation
might illuminate early alternatives that would at least permit partial improve-
ments.
Introduction
This presentation is aimed at raising our insights about how we might fare in
providing the future now handled by standard buses, trucks, and trains. The
time scale is 15, 25, and even 50 years.
Humankind operated with limited potentials into the 19'h century. Global
population and living expectancy grew only slowly, and materials for support-
ing life with food, housing, and clothing came, as they always had, primarily
from the earth's surface. As consumption of coal grew rapidly throughout the
19'h century, and fossil fuel was added throughout the 20'h century, the global
population grew. In 1925, the year I was born, the population was 1.7 billion.
Now it's 6.3 billion, 3.7 times a large, and is likely to be over 8 billion in 25
years.
Coal and oil consumption have been an integral part of all our lives - so
much so that it is hard to think of a world without them. Yet both cause pol-
lution, especially C0 2 , with likely significant effect on the earth's future at-
mosphere. Also, the fossil fuel is nearing its limit. The U.S. sources peaked
about 30 years ago, and globally the peak is expected in just another 10 years
or so. Fossil fuel is particularly valuable for its many non-engine uses, and for
4 P.B. MacCready
the propulsion system of airplanes for which no other technology appears vi-
able. It will probably not be viable for the cars, buses, and trucks 25 years
from now, both because of its pollution and its general cost for availability.
This puts priority on thinking about how we can do the job of heavy hauling
with much more efficiency than at present, and we're looking toward other en-
ergy sources in the long run.
11
---
10
.....
.......
..._,~
·--
•
,.
Ye•ZIIt
8
•
2
......._....... , '
VOilA CIIOICE #
............ * ' -••7
·- '
~
2000
(f> 1500
z
-
0
1-
0
1000
tb
c
~
:i 500
Fig. 2. This plot of the mass of all air and land vertebrates shows that the portion devoted to
humans + livestock + pets is now about 98%, and wild nature is about 2%. As humans were
just beginning agriculture some 10,000 years ago, the 98% portion was less than 0.1%. We
humans have won, and do not even think about the course of events that has created the present
picture. Our growing global population is not sustainable with the consumption to which we in
the United States have become accustomed.
a) Helios. In 2001 this slow, solar- b) Black Widow. This 2_ oz. air-
powered plane flew at 96,863', two plane, 6" x 6", has flown with irs
miles higher than any plane had ever video camera for half an hour. A 6"
flown continuously. It is being crafted x 12" version has flown 1_ hours, as
to station keep at 65,000' continuously the energy and power from small
for months. batteries have increased rapidly.
Surface Vehicles
Gasoline Prices to Drive Car 25 Miles
(In 1989 dollars. Increase numbers by
3/2 factor to relate to 2002-3 dollars)
1929 $4.00
1940 3.00
1960 2.00
1989 1.00
2002 1.00
Fig. 4. The combined effec;t of incre~ed fuel economy and the decreasing price of gas as
viewed relative to present day costs.
The efficiency of present day cars has remained about constant recently because of the
lack of pressure from federal guidelines. Trucks, however, have improved, from 4 to 6,
and even to 8 mpg, and further improvements can be anticipated. Some buses have
improved, but many (and many trucks) rely on old diesel engines producing unaccept-
able levels of particles and pollution. The C0 2 from trucks and buses will increasingly
become more unacceptable to the public and regulators, but little will be done until
the approaching peak in global fossil fuel becomes more obvious.
Battery power, being operationally demonstrated in modern cell phones and micro-
computers, seems capable of taking over 90% of car driving in a few years - if car
companies acquire an interest in the new technology. Small and short haul trucks
could also benefit, but large, long haul trucks may be in the airliners end of the court
for which no viable substitutes have emerged.
Trains are already rather efficient aerodynamically, and increasingly are operated
electrically.
Trucks, buses, and trains are steadily improving in all aspects. For 10 years, im-
provements will be small. For 25 years, improvements must be large, especially in the
source of the power.
8 P.B. MacCready
Fig. 5. I concocted this painting a decade ago to illustrate, on a non-linear time scale, the evo-
lution of our present airlines-cars-TV-power habits, etc. The future, of natural and robotic
cockroaches, was included as a joke- because I really don't know what is coming next. I was
surprised when we got a contract a week later to make a tiny flapping flier that could be consid-
ered related to the cockroach. Our one-half ounce flapper recently flew for half an hour- a
tribute to the rapid improvement of batteries that should impact the car and small truck field.
Over billions ofyears, on a unique sphere, chance has painted a thin cov-
ering oflife - complex, improbable, wonderfUl and fragile.
We humans can wield the paintbrush in a fashion that will work for a long term fu-
ture. The challenge is to keep realizing that our use of fossil fuel can never be a per-
manent solution- both because supplies are limited, and it increases the atmosphere's
pollution. We have used it for so long that it is second nature to us. We do not un-
derstand limits and negative associations.
Inevitably the price of fossil fuel will increase as supplies decrease. Electricity is
more likely to retain its present price, and much more of it can be made available as we
develop its continued generation from solar, wind, and water resources. Ethanol made
from special crops can also provide power without increasing C0 2 • Nuclear power can
generate electricity without generating C0 2 and may deserve a resurgence in the U.S.
Many trains are now electric. Electric local buses are receiving support. Holland has a
goal of getting half its electrical energy from wind. Iceland is headed toward a reliance
on hydrogen for which it has a uniquely large supply. Taking a long range view, there
will be non-polluting transportation energies beyond our present methods, and civili-
zation will have to use them for power while simultaneously improving efficiencies and
decreasing human involvement in transporting goods. Some such approaches are in-
evitable, but will require major research and considerable changing - excitement, fail-
ures, and successes. We are just at the very earliest stage of the transportation of goods
without negative phases.
Commercial Vehicle Aerodynamic Drag Reduction:
Historical Perspective as a Guide
Kevin R. Cooper
Introduction
The aerodynamics development of commercial vehicles has evolved over many
years. Sixty-five years ago, the Labatt Brewing Company developed a
streamlined truck for advertising purposes and to provide larger capacity and
higher cruising speeds, Figure 1. The success of this effort is demonstrated by
the fact that while trucks of the day travelled at 35 mi/h, the Labatt truck
could cruise at 50 mi/h with a fifty percent larger load. The focus today is no
longer on speed, but on energy conservation. It is beneficial for a country to
minimise its energy utilisation and equally beneficial for its trucking industry
to make money while doing so.
Prior Art
In the 1950's, a serious effort to improve truck fuel consumption was
undertaken at the University of Maryland [1,2,3] through an examination of
the aerodynamics of tractors and trailers, funded by Trailmobile. This work
provided an early, detailed look at truck aerodynamics and may have triggered
the development of the air deflector in the 1960's by Seldon Saunders and
Chet Wiley of Airshield - the first successful add-on aerodynamic device.
These studies also presaged the advent of trailer streamlining, by looking at
edge rounding, rounded trailer front faces , skirts and boat-tailing. At about
the same time that Airshield was developing the cab-mounted deflector, Joe
Fitzgerald, working at Thermoking, had realised that their refrigeration units
reduced truck fuel consumption. He decided to take this concept a step
further and developed the Nose Cone trailer streamlining fairing. Thus, the
modern truck aerodynamic age was born.
The first years were difficult. Fuel was cheap and truckers did not want
those gadgets on their rigs. However, the 1970's energy crunch arrived and
the new devices were rapidly accepted. They saved fuel and made profit for
the trucker. They also reduced direct operating cost and strengthened the
competitive position of the trucking industry with the railroads.
In the late 1970s, the National Research Council of Canada (NRC) took
on the task of comparing the commercial devices of the day [4], with the
intention of convincing truckers of the benefit they provided and helping
them choose the best type of device for their operation. Smoke pictures like
those in Figure 2 made a lot of believers, as did a growing body of road
measurements of fuel savings. When truckers saw the two pairs of
photographs, they had no difficulty in making a choice.
The growing activity attracted the attention of the SAE and the US
DOT, leading to the SAE/DOT Voluntary Bus and Truck Fuel Economy
study of the late 1970s and early 1980s [5]. By this time, OEM's and after-
market suppliers in North America and Europe were actively improving fuel
consumption through aerodynamic means, resulting in the reduced-drag fleet
of today.
The SAE/DOT study was a major government/industry cooperative
venture. Its goal was to demonstrate that truck fuel consumption could be
significantly reduced. The study was centred on a set of four pairs of trucks,
two tractor-trailer combinations and two straight trucks. Each pair consisted
of a standard truck for the time and an identical partner fitted with an
aerodynamic package, advanced tyres, a fuel-saver motor, improved lubricants
etc. These trucks were track and road tested, and run in fleet service. The
trucks are shown in Figure 3.
angle with the most streamlined configurations - the truck is beginning to sail.
With the exception of the full-height skirts and streamlined tail, all the
modifications could be implemented. Even the seemingly impractical changes
can be utilized in a less extreme fashion. Partial-height skirts and simple boat
tailing can recover a significant fraction of the potential shown above.
Case 4 - COE tractor, kirt (withou t bumper) Case 5 - COE tra tor with roof fairing
-- -
Ca Co(0°) C 0 (1 C 0 (65mi L\C 0 (65mi
Configuration oo)
se #
COE tractor, 12" radius fr.ont side 0.900 1.167 1.056 0.113
2
posts
# Drag coefficients based on reference area equal to trailer roof height times tratler width.
Commercial Vehicle Aerodynamic Drag Reduction 17
1.75 ...----~--.,..---.,.---.,.---...,.----,
differential. The test period was 30 hours, the time required for a few
computations. The end result was effective, with most users finding about a
30-50 percent improvement in fuel savings compared to the original factory
chart, which was based on guesswork.
The reality is that CFD may not be the best tool for the job, at least in
the near term. Current numerical simulation physics is challenged by highly
unsteady bluff-body flows. The presence of the natural wind ensures that the
yaw angle is almost always not zero, so that a plane-of-symmetry simulation is
not representative. Because the yaw performance of a truck is important in its
average energy utilisation, it is necessary to compute a sufficient number of
yawed cases to define this behaviour. CFD can provide a great detail of
information about a flow, aiding in understanding, but its use is time
consuming and expensive when a large database is required, particularly if the
computations are unsteady.
70~----~--~-----r----~----T-----~--~
:c
ICJ) 65+---
·a;
% •
... c 60+-~-=~--=-~~~~=-~=----+-----+----~
g::;
.....
IX.!
Ql c
55~~-r~~-~~~~1F-~~==~~~
= l! 50+--=~+-~7'~--~--~~~---+-----+----~
~'
~Q
45~~~--~~~--~~~~~t===~
ts 40~--~~--~=-~-+----~----+-~--~--~
....~ 35~~~~--~~~=t==~~
20 30 40 50 60 70 80 90
Tractor-trailer Separation, in.
Fully modified truck with full gap fairing, skirts and bevelled rear panels
1.20
0.50
0.40
-25 -20 -15 -10 -5 0 5 10 15 20 25
Yaw Angle , de g.
Fig. 9: Full-scale Test of Trailer Skirts Fig. 10: Protorype Gap Seal
The gap seal is a device that was patented by Airshield and was field-
tested successfully. However, it never made the transition to market. I do not
know why, although mechanical reliability may have been a major issue. A
prototype Airshield gap seal is shown in Figure 10. It is worth revisiting.
The tractor used in this study does not have the improved shapes of
contemporary equipment. It is expected that the drag would be reduced
further, by approximately 0.05 s dC 0 (65) s 0.08 with a current tractor and
aero package. The end result would be a drag level of C 0 (65) s 0.50 with the
skirts, rear-end treatment and gap seal.
Commercial Vehicle Aerodynamic Drag Reduction 23
The base drag reduction by the beveled plates is not the only possibility.
The use of inset boat-tail plates serves a similar function through the use of a
trapped vortex. This is not a new concept, but was first suggested, to my
knowledge, by J. J. Cornish III, chief engineer, Lockheed-Georgia Company,
in 1968 [22].
This test was completed in 8 hours of tunnel operation and required 6
person days to make the new model parts. The project would have taken
much longer if the models had to be built, but they were available, as are many
other models at 1:8 scale, 1: 10 scale and larger scales at various laboratories
and companies. These models provide an inexpensive resource for future
work.
-
- ACn(65)
Case Configuration Cn(O")' Cn(lO") Cn(65*) ACn(65) re Aero
package
Aero [l.ackage
(roof fairing + cab
2 0.569 0.833 0.724 0.147 -
side extenders +
cab skirts)
Aero package,
3 0.550 0.710 0.644 0.080 0.080
front trailer skirts
Added rear
4 bevelled extension 0.511 0.660 0.600 0.044 0.124
panels
and quite low drag. Wind tunnel measurements from a 1:10-scale model of
are presented in Figure 12 and Table 3, courtesy of Prevost Car Inc. Data for
two other configurations of the articulated bus are shown also, as is the data
from Case 6 ofTable 2 for comparison. In one bus configuration, the mirrors
have been removed and in another, a more streamlined, but practical, nose and
bevelled rear have been fitted. The single bus is the front unit from the
articulated bus.
Of note is the fact that the articulated bus is 50 percent longer than the
identically shaped single bus but has only 9 percent higher drag. This point
will be revisited in the next section. The advanced articulated bus has very low
drag that is nearly constant with yaw angle and may be near a practical limit
for passive aerodynamics for a geometry having a blunt base. It is apparent
that the articulated bus is superior to the developed tractor-trailer. However,
the difference would diminish with a more rounded cab, and would diminish
further with full cab-trailer integration and skirting. At the limit, the two
vehicles should be identical.
0.8
production Prevost H5-6C
0.7 artoculated bus
i:., While RB II
·u 0.6
:e.,
0 0.5
(.)
en 0.4
"'
Q
0.3
artoculaled bu• slre~mhned front + boat laol
I
0.2
-20 -15 -10 -5 0 5 10 15 20
Yaw Angle , deg.
Fig. I I : Low-d rag An iculared Bus
rhe Prcvo; t H '5 -60 Fig. 12: Drag haracreris ric; of rht: Prevost H S-60
Compared ro :t Tracto r-T railer
--
Case Configuration Co(0°) Cn(65mi/ h)
Another Concept
As a final thought, the concept of vehicle platoons [23] applies very well to
trucks. The simplest way to decrease the aerodynamic drag of a tractor-trailer
is to add one or two more trailers. This follows the result for the single and
articulated buses just discussed. Truck trains made up of two or three trailers
have been run on selected freeways in some states and provinces. However,
the practice is not widespread. The question is, "Should it be?". There are
many safery and infrastructure issues to deal with but the returns could be
large, both from energy and road capaciry points of view.
When a second trailer is added to increase capaciry by a factor of two, the
weight does not double and aerodynamic drag increases by about 40 percent.
Thus the aerodynamic drag per ton-mile is decreased by 30 percent. It is
unlikely that any other aerodynamic technique with a single trailer will be as
effective.
As an example to illustrate this point, consider the data from a tractor
model that was tested in the NRC 2m x 3m wind tunnel with three trailer
combinations - a single 27-foot trailer, a single 45-foot trailer and a pair of
tandem 27 -foot trailers. A photograph of the tandem 27 -foot trailer
configuration is seen in Figure 13 and the measured drag results are presented
in Figure 14 and in Table 3. The baseline tractor was equipped with a full
Airshield roof fairing and cab extenders.
1.1
1.0
E.,
·c:;
0.9
.,
if 0.8
0
."'
u
0
0.7
0.6
0.5
0.4
0.0 2.5 5.0 7.5 10.0 12.5
Yaw Angle, deg.
Fig. 13: Tandem 27-foot Trailers Fig. 14: Drag Behavior of a Tractor Pulling Single
or Double Trailers
26 K.R. Cooper
- C 0 ( 65) /unit
Case Configuration Cn(0°) C 0 (8°) C 0 (65)
cargo
Conventional tractor, 0.591
1 0.515 0.659 0.591
single 27-ft. trailer
Closing remarks
This paper turned out differently than the one that was first planned, which
was a discussion of past technology. As the paper progressed, it became more
and more apparent that most of the required aerodynamic knowledge was in
hand. It also seemed that most of this work was being ignored and that the
effort to advance CFD was retarding the application of known aerodynamic
technology to trucking.
The goal of reducing the aerodynamic drag of commercial vehicles is a
worthy one. It is economically and socially valuable. The development of
advanced CFD is also technically useful and will be of benefit in vehicular
development. However, delaying the introduction of new hardware and
concepts into the fleet while waiting for the evolution of these new CFD tools
is counterproductive, especially since the major issues are not aerodynamic,
but are those of operational effectiveness and mechanical design.
Many of the major tractor and trailer manufacturers have built
demonstration vehicles that incorporate advanced aerodynamic technology
including aerodynamic cabs, completely integrated tractors and trailers, skirts
and rear-end treatment. They all had low fuel consumption. And none of
them are on the market. Why? It must be because they were not
economically viable and because they offered too many impediments to
efficient operation. These are the issues of importance. Economics will take
Commercial Vehicle Aerodynamic Drag Reduction 27
care of itself through a steady rise in fuel prices. It would be aided by the
design of effective, operationally effective and inexpensive components,
perhaps encouraged by tax incentive. The operational issues can best be
resolved by industry-wide collaboration.
A two-pronged approach to the problem of introducing improved
aerodynamics might be beneficial. CFD can be developed for long-term
application while a parallel development of existing technology for near-term
implementation, based on present knowledge and some experiment, is
pursued.
In the latter case, the effort required is that of mechanical design done in
close cooperation with industry. The target would be to select the most likely
candidate technologies for development and, using clever design and modern
materials, produce reliable, cost-effective hardware that will benefit truckers
now, and that will be acceptable to the end users. The designs would have to
capture the necessary aerodynamic benefits without causing operational
difficulties. The answers are out there; neither CFD nor the wind tunnel will
tell us how to apply them.
References
1. DOT/SAE Truck and Bus Fuel Economy Measurement Study Report
P59A. Report No. DOT/TSC- 1007, October, 1976.
2. A. Wiley Sherwood - Wind Tunnel test of Trailmobile Trailers.
University of Maryland Wind Tunnel Report No. 85. College Park, MD,
April1974.
3. A. Wiley Sherwood- Wind Tunnel test ofTrailmobile Trailers, 2nd Series.
University of Maryland Wind Tunnel Report No. 85. College Park, MD,
April1974.
4. A. Wiley Sherwood - Wind Tunnel test of Trailmobile Trailers, 3rd Series.
University of Maryland Wind Tunnel Report No. 85. College Park, MD,
April1974.
5. K. R. Cooper - A Wind Tunnel Investigation into the Fuel Savings
Available from the Aerodynamic Drag Reduction of Trucks. Article from
DME/NAE Quarterly Bulletin No. 1976(3), NRC, Ottawa, Canada,
1976.
6. SAE Wind Tunnel Test Procedure for Trucks and Buses. Recommended
Practice, SAE J1252, August 1979.
7. Road Load Measurement and Dynamometer Simulation Using
Coastdown Techniques. SAE Recommended Practice J1263, approved
June 1979.
8. Joint Rccc/SAE Fuel Consumption Test Procedure (Short Term-in-service
Vehicle) - Type I- SAE J1264. SAE Recommended Practice, approved
April1979.
9. Joint Rccc/SAE Fuel Consumption Test Procedure - Type II - SAE
J1321. SAE Recommended Practice, approved April1979.
28 K.R. Cooper
Proposed in 1997, DES was applied to an airfoil beyond stall in 1999, and
then to a range of bluff bodies. Its accuracy has often been far superior to
that of steady or unsteady Reynolds-averaged Navier-Stokes methods, and it
avoids the Reynolds-number limitations that plague Large-Eddy Simulation.
Cases fall into three classes: simple shapes such as cylinders and spheres;
transportation components such as landing gear, simplified; and full airplane
geometries. All are manageable on present computers, some even on personal
computers. Simple shapes now and then yield surprises, but DES appears
sound and reacts well to the type of boundary-layer separation (i.e., laminar
or turbulent), and to grid refinement. However, it is possible to confuse the
method by using a grid density that is both too fine for RANS and too coarse
for LES. Component studies display progress, without reaching an industrial
level of accuracy in predicting all forces. The few full-airplane predictions
have been successful, thanks to high CPU power, and partly thanks to fixed
separation lines.
Progress and a proper dissemination in industry and by CFD vendors
now depends on prudent improvements in numerics and in physical models,
preferably without losing any of the simplicity of DES, and on communication
with non-experts. The experimental database remains weak in some areas.
Numerical issues include grid generation, code performance in unsteady flows,
and numerical dissipation. Codes used for complex configurations stabilize the
calculations via upwinding and/or limiters, and assessing their effects on DES
predictions remains important. Slight improvements to the treatment of the
"RANS region" of DES (essentially comprising the boundary layers up to
separation) are also desired. Good gridding and time-stepping practices are
both costly and crucial; deficiencies have often been traced to the grid. Few
users have experience with LES, let alone with issues specific to DES, which
are clarified here. The prediction of bluff-body flows at application Reynolds
numbers will place heavy demands on the user, as numerous aspects of the
30 P.R. Spalart and K.D. Squires
DES was inspired by estimates initiated in 1997 [1] and completed in 2000
[2], which indicate that Large-Eddy Simulation (LES) is not about to become
practical for industrial flows, such as aircraft or road vehicles. This applies
for about forty years, and is true even if it is assumed that "wall modeling"
has become successful, in other words, that limits on the grid spacing in wall
units such as Llx+ have been removed within LES. Relevant to these esti-
mates is the assumption that much of the boundary layer is turbulent; an
assumption which is most reasonable in practice. The estimates have yet to
be challenged by LES proponents, and too much work remains mired at un-
challenging Reynolds numbers, from which clear conclusions cannot be made.
Empirical evidence that strongly supports the estimates is offered by the LES-
FOIL workshops [3], for which the Reynolds number was sufficiently large to
expose LES methodologies: current super-computers, even for a single sim-
ulation with a "designer grid" optimized by experts given detailed advance
knowledge of the flow, can handle only an airfoil slice of the order of 1% of
the chord, at a moderate Reynolds number and with a laminar stagnation
region [4]. In other words, the real-life problem of a wing is roughly 1,000
times larger, even before the extra difficulties of sweep (giving a thin and
turbulent attachment-line boundary layer) and of a full-size Reynolds number
are included.
When considering near-future industrial CFD methods capable in high-
Reynolds-number bluff-body flows, experience and projections lead to a strong
consensus that Reynolds-averaged Navier-Stokes (RANS) technology is indis-
pensable in the large areas of thin attached boundary layers, and to a weaker
consensus that pure RANS methods cannot provide sufficient accuracy in re-
gions of massive separation. These regions are dominated by large, inviscid,
geometry-specific eddies, which are excellent candidates for LES. As a result,
hybrid methods are receiving wide attention.
These considerations leave some leeway in designing a hybrid RANS-LES
method. The first choice is between a method that is explicitly zonal, as op-
posed to a method that solves a single set of equations. DES reflects a strong
preference for non-zonal systems and simple formulations. It involves a single
grid and a single field for velocity, pressure, and even eddy viscosity, whether
using the Spalart-Allmaras (S-A) model or the SST model as the RANS base
[5]. The choice of model is of course an element of leeway, and a welcome op-
portunity to test the sensitivity of solutions. Simplicity is favorable in terms
of repeatability between codes and users, and experience-building. The for-
mulation using a single model only leads to a discontinuity in the gradient
of the length scale that enters the destruction term of the turbulence model
DES for Bluff Bodies 31
(this discontinuity would be easily removed by rounding the min function that
determines the lengthscale). In contrast, explicitly-zonal methods often suf-
fer from steep variations near the zonal interface, and much work is expended
controlling these. The change in the lengthscale leads to a model that becomes
region-dependent in nature - in most cases a RANS model in the boundary
layers and a Sub-Grid-Scale (SGS) model within massive separation. Objec-
tions have been made to the use of the grid spacing Ll in the formulation,
but this appears absolutely natural. The foundation of LES is the existence
of a filter width that controls the end of the energy cascade, and is available
for reduction in order to increase the range of scales and therefore improve
the physics of the simulation [2]. As separate issue, present in any LES, is
whether the filter width and the grid spacing should be tied; a recent trend is
to untie them, and seek a grid-converged solution for a fixed filter width. In
all DES studies, they have remained tied; this appears to provide a balance
between numerical and SGS-modeling errors, although this balance has not
been established (which would be difficult to do beyond question, because
in LES the error measures are quite arbitrary). Thus the nature of the SGS
model inside DES is classical, and in particular its scaling is identical to that
of the Smagorinsky model. It was adjusted to extend the inertial range in a
classical Kolmogorov setting as far as possible [18]; this adjustment requires
care when generalizing it to unstructured grids. There is no obvious accuracy
advantage in solving a transport equation for the SGS eddy viscosity, instead
of using a local algebraic formula; it is done for commonality with the RANS
region.
In addition to the formulation, there is also flexibility in the design of a
DES simulation, primarily within the prediction of the boundary layer up-
stream of separation. In "natural" DES it is handled by the RANS model
but there is a trend, at least in fundamental studies, to predict parts of the
attached boundary layers with LES. The study of Nikitin et al. in a chan-
nel was conducted to explore such applications [6] and was quite successful.
Commentators have focused on the imperfection in the skin friction, instead
of the great simplicity of the approach and excellent Reynolds-number scaling
both in terms of computing cost and accuracy. Applied to an airfoil in this
limit, DES then resembles the LESFOIL exercise [3]. Keeping in mind that
the unmanageable cost of LES arises due to the resolution requirements in the
thin regions of the boundary layer, switching from RANS to LES is feasible
only after the boundary layer thickens sufficiently. In this regime, LES treat-
ment within the boundary layer is attractive since it is possible to exploit the
accuracy of the technique and the grid-refinement possibilities inherent to the
method - a proposition that is expensive, but represents an important and
valuable research activity. A key concept underlying such an approach, and
for that matter all applications of DES, is to clearly "steer" the simulation
either to RANS or to LES behavior by means of the grid density. Unfortu-
nately this can become delicate, and normally the detailed characteristics of
a flow are not known in advance as, for example, is the case for LESFOIL.
32 P.R. Spalart and K.D. Squires
Note that "natural" DES of the LESFOIL case, with its shallow separation,
is essentially a RANS, and therefore has similar accuracy. In fact, RANS pre-
dictions are more accurate than LES has been to date, presumably because
no LES has had both sufficient grid density and adequate spanwise extent.
This judgment assumes the particular experiments on that airfoil are more
accurate than CFD, which will not be true indefinitely.
The objection to incorporating the grid spacing into the turbulence model
in DES has one root in the inaccuracies encountered by careful users, as they
gradually refine the grid starting from a natural DES [7, 8, 9]. This threat was
illustrated in the initial DES paper [1]. Once the grid spacing in both direc-
tions parallel to the wall becomes smaller than about half of the boundary-
layer thickness, the DES limiter reduces the eddy viscosity, thus "corrupting"
it compared with its RANS level, but without allowing LES behavior. The
resulting solution creates insufficient total Reynolds stresses (modeled + re-
solved stress [6]). There is no solution ready for release at present, and this
adds to the burden on the DES user of checking that the RANS/LES interface
is not deep inside the boundary layer. It also conflicts with the tendency to
refine the grid at shock-boundary-layer interactions, either manually or auto-
matically. However, the only true need is to refine in the direction normal to
the shock, and therefore in a typical situation the shock-parallel spacing will
keep .Lllarge enough to avoid unintentional eddy-viscosity limiting. Some ad-
justments may be introduced, but removing the grid spacing altogether from
the DES formulation is out of the question.
Nearly all other hybrid methods currently under consideration sensitize
the model to the grid in order to achieve LES behavior [10, 11, 12, 13]. As
the exception, the SAS hybrid is free of grid spacing and has demonstrated,
visually, LES-like behavior past a circular cylinder and a wall-mounted cube
[14]. This gives much food for thought. On the other hand, it is still extremely
new with only one conference paper, has not yet demonstrated LES behavior
during grid refinement, and has simply failed to function as LES in a channel
flow. New findings may appear rapidly.
Another issue addressed in [1] is the "grey area" in which a shear layer,
after separation, needs to generate "LES content" (random eddies) which
it did not possess as a boundary layer upstream. The process of generating
LES content is most easily achieved by a thin shear layer that is rapidly
departing from the wall, typically thanks to a thin edge or sharp corner;
greater uncertainty was anticipated over smooth-wall separations. Similarly,
LES is widely expected to be more successful for bodies with sharp edges than
smooth ones (even with laminar boundary layers; for turbulent separation LES
poses a bigger challenge, and SGS models are sometimes naively used as RANS
models in the entire thickness of the boundary layers). These considerations
motivated the circular-cylinder study, which was quite re-assuring, see §2. A
more extreme case is a shallow separation, either unclosed or a bubble. There,
depending on the grid, the DES may not create any LES content. This is
not incompatible with the original motivation for the method to achieve LES
DES for Bluff Bodies 33
2 Simple Objects
The circular cylinder remains a major challenge to bluff-body CFD methods
of any type, and gives an excellent arena in which to make DES succeed,
or fail. Even shapes reputed to be less challenging such as square cylinders
cannot be considered as "solved" even with over 106 grid points [17]; however
it is plausible that a simple increase in resolution past 107 points will be
quite helpful (the workshop [17] took place in 1995). Ensuring a good balance
between the three directions of the grid and the time step will also help. It
is not clear why the thin-airfoil or thin-plate problem appeared under good
control, at several angles of attack, already with only about 2 x 105 points
[18]. In contrast to these cases, the separation off the smooth surface of the
cylinder, a textbook case well-known to strongly depend on transition, tests
CFD approaches and codes in many areas.
At Reynolds numbers that produce laminar separation (LS), DES is not
very different from LES, except that a simple LES method is not equipped to
disable the SGS model, as must be done in the boundary layers. A DES based
on the S-A model functions quite well in "tripless" mode, and the model
34 P.R. Spalart and K.D. Squires
1. 0 ~-.-.-,---.-y-"T"-,--.-,---.-, 1. 0 ,...........,-......,.--.--r-r--.--.--.---.-.....
0.5 0 .5
0 .0 0 .0 .... ·······(····
- 2. 0 ........._._......................._.__......J__._...L......o.......J - 2. 0 ..................._.........&.......O..:...&...........__.__.o.....L_.__j
0 30 60 90 120150180 0 30 60 90 120150 180
() ()
realistic imperfections of both the surface and the incoming flow. Whether
differences in these imperfections were sufficient to cause the differences in
experimental pressures is an intriguing conjecture.
Travin et al also found that very long time samples are essential to converge
the statistics, as the vortex shedding has strong modulations in DES, just
as occurs in the experiment [20]. Although this could be a "pathology" of
2D-geometry separated flows, it is a warning for any study of a flow with
large-scale shedding: simulating only a few cycles of such shedding is unsafe.
Constantinescu & Squires presented the equivalent study for a sphere [22].
0.008 o Measurements
(X= 20° DES - Fine Grid
xiL = 0.77 - - - - DES- Medium Grid 0
DES- Coarse Grid
RANS 0
0
0.006
00.004
0.002
0 o~~~3~0~~~6~0~~~9~0~~~1~2~0~~~15~0~~~180
<!>
leeward coordinate of the symmetry plane. The distributions shown are in the
aft region, at x / L = 0. 77 for which measurements show well-developed pri-
mary and secondary separation. The computations are for a freestream at 20
degrees angle of attack and at a Reynolds number of 4.2 x 106 (based on the
freestream speed and spheroid length, L). The flow parameters of the com-
putation match the experimental measurements reported by [21], including
a boundary layer trip at xj L = 0.2 which is activated in the computations
using the trip terms of the S-A model.
DES predictions on three grids are shown together with a S-A RANS re-
sult. Experiments on the spheroid have shown that minima in CJ are well
correlated with the separation locations indicated by wall streamlines, the
measured skin friction then indicating primary separation at ¢ ~ 115° and
secondary separation at ¢ ~ 145°. The S-A RANS result predicts primary
separation at a slightly more leeward location. Figure 2 shows that the mini-
mum in CJ is around 125° and that the magnitude is slightly greater than the
measured minimum. The variation in CJ around 145° in the RANS prediction
is indicative of the influence of a secondary separation, though its influence
is very weak and an outcome of the S-A model devoid of rotation correction,
with its large eddy viscosity in the core of the vortices, overly diffusing these
structures.
For the DES, with grid refinement the skin friction magnitude shows an
increasingly well-defined secondary separation, as evidenced by the C f dis-
tribution around ¢ = 145°. Increases in grid resolution results in lower eddy
viscosity, an effect that is similar to that which could be obtained using a
RANS model with R/C corrections. Nevertheless, the more sharply defined
separated structures, in this case the coherent longitudinal vortices, experience
less diffusion with increases in mesh resolution, in turn resulting in pronounced
signatures in the skin friction and (not shown here) surface pressures.
Figure 2 also shows that the influence of grid resolution on the minima
corresponding to primary separation is not weak. On progressively finer grids
the minimum shifts windward and the magnitude decreases. For the unstruc-
tured grids used for the DES predictions shown in the figure, the RANS-LES
interface is within the boundary layer upstream of separation. As discussed
in §1, this could lower the total Reynolds stress and shift the separation loca-
tion "upstream" (in this case, towards the windward direction). This is under
intense investigation.
3 Components
500.000
Fig. 3 . Vorticity contours in six axial planes along the leeward side of the GTS (10°
yaw). Surface colored by pressure.
38 P.R. Spalart and K.D. Squires
residual influence of the model, and the value of being able to routinely control
transition in CFD as provided by the S-A model.
Active Flow Control was a new arena, and proved highly challenging. The
slot was so narrow that it covered only two to three grid points, even with
very strong clustering. Thus, the slot treatment was crude. With a grid of
the order of 600,000 points, the crucial region containing the manipulated
separated shear layer was treated in URANS mode inside the DES; for an
LES treatment, orders of magnitude more grid points would be needed. Thus
the shear-layer treatment fell back on the RANS model; no data were available
to validate the model in such a flow, and the accessible grid refinement would
have been inconclusive. The model may not matter to leading order, since the
roll-up of the shear layer is nearly inviscid, but this is not proven. For the rest
of the wake, the level of confidence was fairly good, based on the agreement
with AFC off.
The pressure distribution responded to AFC, but only qualitatively. The
drag reduction was not very consistent with experiment, and in some ranges,
increasing the level of forcing caused an increase in the predicted drag. The
time samples were adequate, but not all were long enough to rule out hysteresis
or very slow modulations. Repeating the study with larger computer resources
would close some of these issues. However, the very nature of AFC creates
CFD problems containing a huge range of scales (here, from the slot width
at 0.2% of chord to the wake length at 100 chords), which are barely possible
in three dimensions even with thorough grid design. Turbulence is the realm
of wide ranges of scales, but the premise of LES is that the small eddies only
absorb information from the large eddies; here, phenomena in the tiny slot
control the entire flow. As a result, CFD will lag experiments for years in the
field of AFC for lack of CPU power, provided the experiments are done with
care and attention to scale effects.
Other component flows are the simplified landing-gear truck of Hedges et
al. [28] and the jet-fighter forebody of Squires et al. [29].
4 Full Configurations
The series of DES over the F-15 and F-16 fighter aircraft has been the most
noticed work on quite complete configurations [31] (although the engine mod-
eling remains very simple). In particular, agreement with experiment within
5% for lift and drag at 65° angle of attack is an excellent result, even once
it is recognized that the extreme angle of attack and thin wing edges mean
that the DES depended almost only on its LES mode, except for the nose
of the airplane. The forebody "component" study [29] gives confidence and
was very favorable to DES over RANS, but accommodating a grid and a time
step capable of LES behavior on the nose while including the entire aircraft
remains very challenging.
40 P.R. Spalart and K.D . Squires
Fine effects such as drag reduction by detail shape modifications have not
been demonstrated yet.
5 Algorithm Issues
6 Design of a DES
The basis of DES is that the relationship between the grid spacing and the
natural length scales of the turbulence in a given region makes the selection
between RANS and LES behavior [19]. Here, we are assuming that the time
step is short enough to allow LES behavior; since the time step is global to
the simulation, its value can present a difficult compromise. As a result of this
option, and considering the permanent need for economy of grid points, DES
grid design is not simple. The "young person's guide" (YPG) [16] introduces
terminology and guidelines, many of which also apply to RANS calculations,
even if they have not been named. Focusing on external flows, the YPG dis-
tinguishes the Euler Region, into which no turbulence will intrude, the LES
Region, and the RANS region. The latter two also contain viscous layers at
the walls. The wall regions resemble those in RANS: shallow cells with the
usual rules for first wall-normal spacing or "y+" and stretching ratio. In con-
trast, the LES and Euler regions have isotropic cells, and the LES region little
stretching. There is also necessarily a "Departure Region" in which eddies that
will never return into the LES region and impinge on the body safely become
dissipated.
Grid refinement tends to concern the LES region and the wall-parallel
spacings of the wall region, which tend to follow the LES-region spacing. For
the wall-normal distribution, there is little point in starting with y+ = 10
or a stretching ratio of 2 (as compared with the guidelines y+ ;::::; 1 - 2 and
ratio;::::; 1.2 -1.3), say, unless this is explicitly a disposable simulation used to
prime an automatic grid adaptation (it would make much more sense to do
a precursor simulation at a reduced Reynolds number than with an excessive
y+). The Euler region does not contain a large share of the points, giving
little incentive to save points and have to re-visit it. The LES region leaves
the most leeway, as it is very difficult to predict how many points are needed
to be accurate in a new case. Grid convergence is very elusive, and the order
of convergence is not simple at all. Only the circular cylinder has strongly
suggested grid convergence [19]. In addition, reaching a good accuracy level
can well happen at different levels of resolution in different parts of the flow.
The smaller features of the geometry require finer resolution, which is not a
natural tendency when generating the LES-region grid; in addition, again, the
time step is uniform. There are clear conflicts.
The recommendation is to count many weeks of work for any new case,
several cases to form a new DES user, and to pool experience both by detailed
and critical publications and within the networks of the CFD vendors.
7 Closing comments
For treating bluff bodies at useful Reynolds numbers, the consensus behind
hybrid RANS-LES methods has grown very strong, and DES is at present the
DES for Bluff Bodies 43
most recognized of such methods. It has been stable and has active commu-
nities outside the original core, particularly in Germany and France, so that
meaningful critiques have been made (as have a few mistakes, such as grid
refinement that is not consistent in the three directions). The users have been
generally quite satisfied, while recognizing the heavy challenge of designing a
DES. However, the challenge is no surprise to those familiar with LES, with
the possible exception of the issues of corrupting the eddy viscosity in a RANS
boundary layer or accidentally approximating the turbulence depletion by ro-
tation. Abuses have been committed, typically leading to essentially RANS
behavior because of a very insufficient number of grid points; no LES can
develop with 10 points in any of the directions. On a related matter, some
users expect DES to require less resolution than LES everywhere, but this
is true only in the boundary layers. For a region of massive separation, the
size of the numerical task is dictated by a "number of large eddies" which is
no different between an LES and a DES. The recent dissemination in vendor
CFD codes is welcome, but will lead to much more use by non-experts. To
help them, it is desirable for many detailed publications to appear, and again
for the approach to remain stable, simple, and clearly defined; they are urged
to visualize their solutions extensively, and to produce at least two grids.
The warnings made in this paper over the design and interpretation of a
DES, from the simplest to the most complex geometries, make it clear that it
is not a "push-button" technique; DES requires a commitment to numerical
quality, and a working knowledge of turbulence. However, the wait for a push-
button technique will be very lengthy. Many of the warnings simply reflect
the physics of this new class of flows, which is still widely considered to be out
of reach of CFD. It is out of reach of casual CFD use. Competing methods
do not appear simpler to implement or understand, and since all problems
are three-dimensional, "over-kill" grid resolution is not about to become an
option. Fortunately, we are seeing the beginning of grid adaptation, based
on precursor solutions, which makes grid refinement more rapid and more
systematic.
8 Acknowledgments
Prof. Strelets made comments on the manuscript.
References
1. Spalart PR, Jou WH, Strelets M, Allmaras SR (1997) Comments on the fea-
sibility of LES for wings, and on a hybrid RANS/LES approach. 1st AFOSR
Int. Conf. on DNS/LES, Aug. 4-8, 1997, Ruston, LA. In Advances in DNS/LES,
C. Liu and z. Liu Eds., Greyden Press, Columbus, OH, USA.
2. Spalart PR (2000) Strategies for turbulence modelling and simulations. Int. J.
Heat Fluid Flow 21, pp. 252-263.
44 P.R. Spalart and K.D. Squires
3. Mellen CP, Frolich J, Rodi W (2002) Lessons from the European LESFOIL
project on LES of flow around an airfoil. AIAA-2002-0111.
4. Mary I, Sagaut P (2002) Large-eddy simulation of flow around an airfoil near
stall. AIAA J., 40, 1139-1145.
5. Strelets M (2001) Detached eddy simulation of massively separated flows.
AIAA-2001-0879.
6. Nikitin NV, Nicoud F, Wasistho B, Squires KD, Spalart PR (2000) An Ap-
proach to Wall Modeling in Large-Eddy Simulations. Phys Fluids 12, pp. 7-10.
7. Caruelle B (2000) Simulation d'ecoulements instationnaires turbulents en
aerodynamique: application a !a prediction du phenomEme de tremblement.
CERFACS Report TH/CFD/00/50.
8. Deck S (2002) Personal communication.
9. Menter FR, Kuntz M, Durand L (2002) Adaptation of eddy viscosity turbulence
models to unsteady separated flow behind vehicles. This volume.
10. Speziale CG (1998) Turbulence modeling for time-dependent RANS and VLES:
a review. AIAA J. Vol. 36, No.2, pp. 173-184.
11. Batten P, Goldberg U, Chakravarthy S (2000) Sub-grid turbulence modeling
for unsteady flow with acoustic resonance. AIAA-2000-0473.
12. Wernz S, Fasel H (2000) Control of separation using wall jets- Numerical in-
vestigations using LES and RANS. AIAA-2000-2317.
13. Arunajatesan S, Sinha N, Menon S (2000) Towards hybrid LES-RANS compu-
tations of cavity flowfields. AIAA-2000-0401.
14. Menter FR, Kuntz M, Bender R (2003) A scale-adaptive simulation model for
turbulent flow predictions. AIAA-2003-0767.
15. Shur M, Strelets M, Travin A, Spalart PR (2000) "Turbulence modeling in
rotating and curved channels: assessing the Spalart-Shur term". AIAA J. 38,
5, 784-792.
16. Spalart PR (2001) Young person's guide to Detached-
Eddy Simulation grids. NASA CR-2001-211032.
http:/ /techreports.larc.nasa.gov /ltrs/PDF /2001/ cr /N ASA-2001-cr211032. pdf
17. Rodi W, Ferziger JH, Breuer M, Pourquie M (1997) "Status of large eddy
simulation: results of a workshop". J. Fluids Eng. 119, 248-262.
18. Shur M, Spalart PR, Strelets M, Travin A (1999) Detached-eddy simulation of
an airfoil at high angle of attack. 4th Int. Symposium on Eng. Turb. Modelling
and Experiments, May 24-26 1999, Corsica. W. Rodi and D. Laurence Eds.,
Elsevier, Amsterdam, NL.
19. Travin A, Shur M, Strelets M, Spalart PR (2000) Detached-Eddy Simulations
past a Circular Cylinder. Flow, Turb. Comb. 63, pp. 293-313.
20. Cantwell B, Coles D (1983) An experimental study of entrainment and trans-
port in the turbulent near wake of a circular cylinder. J. Fluid Mech., 136:321-
374.
21. Chesnakas CJ, Simpson RL (1997) Detailed investigation of the three-
dimensional separation about a 6:1 prolate spheroid. AIAA Journal, 35: 990-
999.
22. Constantinescu G, Squires KD (2000) LES and DES investigations of turbulent
flow over a sphere. AIAA-2000-0540.
23. Forsythe J, Hoffmann K, Dieteker JF (2000) Detached-eddy simulation of a
supersonic axisymmetric base flow with an unstructured flow solver, AIAA-
2000-2410.
DES for Bluff Bodies 45
Abstract
The results of the large eddy simulation of the flow around a simplified bus
presented in [11, 12] are used to describe this flow in detail. Using time-
averaged trace lines on the surface of the body, the patterns of the shear-
stress lines are revealed and used to identify bifurcation lines and critical
points (zero-shear-stress points) in the flow. This information is then used to
establish a complete picture of the flow on the surface of the body that can
be used for understanding soiling and accumulation of water on the surface
or in determinations of aeroacoustic noise sources. Kinematical investigations
of the flow in two symmetry planes were done to reveal the critical points
in the flow. With this it was proven that the flow resulting from numerical
simulation is kinematically possible.
1 Introduction
Our success in predicting the behavior of the vehicle in the air stream is
strongly dependent on our knowledge of the flow influencing the vehicle. The
complexity of the structures in this three-dimensional flow makes experimen-
tal studies very difficult. Furthermore, experimental studies often provide only
information on some limited partition of the flow (point, line or a plain). Com-
putational fluid dynamics (CFD) gives a description of the flow in the entire
computational domain (numerical wind tunnel). Unfortunately, the Reynolds-
Averaged Navier-Stokes equation (RANS) commonly used in prediction of this
flow suppresses too much information. This method has also difficulty in pre-
dictions of the momentum transport after separation, which makes it difficult
to rely on the prediction of separated flows.
50 S. Krajnovic and L. Davidson
0.0 H
Fig. 1. Geometry of the bus-shaped body placed above the wind tunnel wall. Time-
averaged trace lines are plotted on the surface of the body showing the roof and the
lateral vortices that extend XR1 and Xs, respectively, in the stream-wise direction.
Separation
lin
b)
Fig. 2. Time-average d trace lines on the surface of the body showing the roof vortex,
R, and the lateral vortex, L. View of the front face of the body.
from the unstable node, only to repel it in the perpendicular direction, i.e.
upstream and downstream. It is possible that the trace lines leaving node Z
in the upstream direction toward the separation line end in a stable node in
region Q in Fig. 3(a).
Similar to node Z, the unstable node G (see Fig. 4) is found close to the
lower lateral corner of the body. The surface flow in this region is complex due
to the presence of the stable focus Fe. The flow coming underneath the bus
is deflected by this focus, and the fluid particle closest to it travels in spiral
paths ending in the focus. At the same time the node is deflecting the flow.
The partition of the flow that is deflected downstream forms together with
the stream-wise flow on the underneath surface, the negative bifurcation line
(NBL) shown in Fig. 5(a). This NBL indicates separation of the flow and
results in lateral vortices (see [11, 12]). No bifurcation lines were observed on
the upper lateral edge (see Fig. 5(b)).
Flow on the rear face of the bus is very complex. The shear stresses here
are very low, making it difficult to represent the surface topology using trace
lines. Thus the surface topology on the rear of the bus is not presented in this
paper.
Besides the study of the surface topology, kinematical studies of the flow
in two symmetry planes were conducted. This was done to identify the crit-
ical points in the flow and to determine whether the flow resulting from the
simulation was kinematically possible.
The most common critical points on the border between the surface of
the body and the surrounding flow are half-saddles (see Figs. 6 - 9). The
half-saddle indicating the reattachment is always the mirror image of the
half-saddle indicating the corresponding separation (see e.g. Fig. 8( c), where
the reattachment point 8 13 is the mirror image of the separation point 8 12 ).
In the X- y plane close to the front part of the body, three half-saddles, 81' 82
and 8 3 , were detected (see Fig. 6). In this figure, half-saddle 8 1 represents the
stagnation point and 82 and 8 3 indicate points of separation and reattach-
ment, respectively, of the recirculation region on the roof of the bus (region R
in Fig. 2). One stable focus, N 1 , was also identified in the recirculation region
(see Fig. 6).
As the flow separates at the rear of the bus, the region of recirculating flow,
shown in Fig. 7, is formed. Two thin vortices B, shown in Fig. 7, are formed on
the rear surface in the x-y plane. Figures 7(b) and (c) show that the direction
of rotation of the upper-edge and lower-edge vortices is counter-clockwise and
clockwise, respectively. Separations of the flow responsible for these structures
are indicated with half-saddles 8 6 and 87 and reattachments with half-saddles
8 5 and 8s for lower-edge and upper-edge vortices, respectively (see Figs. 7(b)
and 7(c)). It is interesting to note that both vortices Bare fed with the flow
pushed away by half-saddle 8 10 in the free stagnation point. One part of this
flow coming from 8 10 passes vortices Band interacts with the separated flow
in the x direction at y = ±H/2 and forms half-saddles 84 and 89 close to
the lower and upper edges of the bus, respectively. The main part of the
54 S. Krajnovic and L. Davidson
a)
b)
c)
Fig. 3. Time-averaged trace lines on the surface of the body showing the unstable
node, z (a), and the saddle points, sd (b) and s. (c), downstream of the separation
region, Rand L, respectively. PEL are positive bifurcation lines, i.e. attachment
lines. The velocity vectors are plotted at the surface parallel to the body surface at
the wall-normal distance of 1.6 x 10- 4 H .
Exploring the Flow Around a Bus with LES and Topological Tools 55
Fig. 4. a) Time-averaged trace lines on the surface of the body. View from below. b)
Zoom of Fig. (a). The velocity vectors are plotted at the surface parallel to the body
surface at wall-normal distance of 1.6 x 10- 4 H. Fe is a focus and G is an unsteady
node [13] .
56 S. Krajnovic and L. Davidson
(1)
Exploring the Flow Around a Bus with LES and Topological Tools 57
a)
b)
Fig. 5. Time-averaged trace lines on the surface of the body. a) View of the lower
lateral edge. b) View of the upper lateral edge. NBL is the negative bifurcation line,
i.e. the separation line. The velocity vectors are plotted at the surface parallel to
the body surface at the wall-normal distance of 1.6 x 10- 4 H.
58 S. Krajnovic and L. Davidson
a)
where :L;N is the number of nodes and foci, :L;N, is the number of half-nodes
(nodes on the boundaries), :L; 8 is the number of saddles and :L; 8, is the
number of half-saddles. n is the connectedness of the surface. For a singly
connected region with no body, n = 1; with one body (as in our case) n = 2
etc.
Let us now investigate whether the critical points in Figs. 10 and 11 satisfy
the topological constraint in Eq. 1. In Fig. 10 :L;N = 5, :L;N' = 0, :L; 8 = 1
and :L; 8, = 10 so that (:L;N + 1/2:L;N') - (:L; 8 + 1/2:L; 8 ,) = 5- (1 + 5) =
-1 and in Fig. 11 :L;N = 6, :L;N' = 0, :L; 8 = 1 and :L; 8, = 12 so that
(:L;N + 1/2:L;N') - (:L; 8 + 1/2:L; 8 ,) = 6- (1 + 6) = -1. This agrees with
Eq. 1, since n = 2. Thus these figures represent a flow that is kinematically
possible.
c)
Fig. 8. a) Time-averaged streamlines projected onto plane y = 0 showing half-
saddles Sn-Sts and foci N6 and N1. b) Zoom of the stagnation point (half-saddle
Su). c) Zoom of the lateral side vortex.
This study was based on the topological and kinematical results of our LES
in [11, 12]. For additional details on the flow features and the mechanisms of
their origin, we refer to [11, 12].
60 S. Krajnovic and L. Davidson
u. 3 4
l
D
2
5
4 Numerical accuracy
6
u
...
9 ll
7
Table 1. Lengths for re-attachment on the roof (XRt), lateral walls (Xs) and behind
the bus (Xr). Xfr ee and Ytree are coordinates of the free stagnation point at the
closure of the separation bubble.
Table 2. Time-averaged pressure drag, lift and rear pressure coefficients and domi-
nating frequency (Stp) of the Cp signal (note that Cp means the integrated Cp over
the rear surface).
]!_
H
OS
§' ·:
.•.
'0
~
jC.
..·.
...
....
·•
· '
'
0.5
.,
. . ....... ,. -
•••
\.>,
.. !
....
-~
'
0
0
0.5
X
Ji = 1.63
/ ..
4.~·
....
.
..--·
...
... ..r=~. 0
0
~:
.
0
.
I
0. ;
• ... ... 0 0
-0.5
-0.5 -0.5 .·. :
•'
v.11 1u V . 11 /U v."/
Fig. 12. Time-averaged velocity profiles at three downstream locations at z = 0.
Fine grid (solid curve) ; medium grid (dashed curve); coarse grid (dashed-dotted
curve); experiment (symbols) .
Exploring the Flow Around a Bus with LES and Topological Tools 63
5 Conclusions
The dense data obtained in LES is suited for delicate topological studies of the
flow. The critical points and the bifurcation lines on the surface of the vehicle
can be displayed and provide the information on the motion of the near-surface
fluid particles. This paper demonstrates such use of LES data to create a
topological picture of the external vehicle flow (i.e. flow around a simplified
bus). The singular points were not only identified but also classified according
to their stability (stable and unstable) and their influence on the surrounding
fluid (nodes, foci and saddles). A similar distinction was made between the
negative and positive bifurcation lines in this flow. In addition to these studies,
a relation between the number of nodes and the number of saddles was used
to determine whether the simulation produced a kinematically possible flow.
This analysis may be useful when poor resolution prevents determination of
the character of a solitary singular point. Classifying the remaining points
in topological terms as nodes and saddles and using a relation between their
number, the nature of the last point can be determined.
References
[1] S. R. Ahmed, G. Ramm, and G. Faltin. Some salient features of the time
averaged ground vehicle wake. SAE Paper 840300, 1984.
[2] J. Barlow, R. Guterres, R. Ranzenbach, and J. Williams. Wake structures
of rectangular bodies with radiused edges near a plane surface. SAE
Paper 1999-01-0648, 1999.
[3] P. W. Bearman, J. P. Davis, and J. K. Harvey. Measurement of the
structure of road vehicle wakes. International Journal of Vehicle De-
sign, Technological Advances in Vehicle Design Series, SP3, Impact of
Aerodynamics on Vehicle Design, pages 493-499, 1983.
[4] C. Chandrsuda and P. Bradshaw. Turbulence structures of reattaching
mixing layer. Journal of Fluid Mechanics, 110:171-194, 1981.
[5] S. Dahlstrom and L. Davidson. Hybrid RANS-LES with additional con-
ditions at the matching region. In 4th Int. Symp. on Turbulence Heat
and Mass Transfer, Antalya, Turkey, 2003.
[6] L. Davidson and S.-H. Peng. A hybrid LES-RANS model based on a one-
equation SGS model and a two-equation k - w model. In E. Lindborg,
A. Johansson, J. Eaton, J. Humphrey, N. Kasagi, M. Leschziner, and
M. Sommerfeld, editors, The Second International Symp. on Turbulence
and Shear Flow Phenomena, volume 2, pages 175-180, Stockholm, 2001.
[7] E. G. Duell. Experimental investigation of unsteady near wakes of ground
vehicle bodies. PhD thesis, Cornell University, 1994.
[8] E. G. Duell and A. R. George. Experimental study of a ground vehicle
body unsteady near wake. SAE Paper 1999-01-0812, 1999.
64 S. Krajnovic and L. Davidson
1 Introduction
This study is motivated by the saguaro cactus and other tall arborescent (tree-
like) succulents that withstand high wind velocities in their natural habitat.
These stationary desert plants experience Reynolds number ( Re) flows up to
106 and share a common cylindrical shape modified with complex surface ge-
ometry. Because the shape of an object influences the surrounding airflow,
natural selection may favor body morphologies that reduce forces exerted by
wind gusts in their habitat. We hypothesize that the tall cacti morphology of
longitudinal cavities and spines may function to reduce wind forces such as
drag and fluctuating lift. In this paper we address this hypothesis by numerical
simulations. A concurrent experimental investigation is described in Talley et
al., 2001; some measured data are compared with the numerical predictions
herein.
There has been much speculation on the function of cavities and spines
on cacti, and their significance is still open to speculation (Geller and Nobel,
1984). Natural selection acts on the random mutations of existing structures
(traits), resulting in improved structures, novel structures, and/or multiple-
functionality of existing structures. Therefore, one function of a trait does
not necessarily preclude other functions, and many traits may contribute to
a common function. Given that the shape of an object affects the flow, it is
surprising that no studies have examined how cavities and spines on desert
succulents influences airflow.
66 G. Iaccarino, P. Durbin, and S. Talley
Because there are many species of tall arborescent succulents that vary in
body size, depth and number of cavities, and spine arrangement, we will focus
on one of the most studied of the tall arborescent succulents, the saguaro cac-
tus, Carnegiea gigantea (Fig. 1). Saguaros are long-lived and slow to mature.
They take 30 to 50 years to reach reproductive maturity and live up to 150
years of age. Adult saguaros have one main cylindrical stem ranging from 0.3
to 0.8 m in diameter (Benson, 1981) and over 8 to 15 m in height (Hodge,
1991). Ten to 30 v-shaped cavities span the length of the stem (Hodge, 1991).
The number of cavities depends on the diameter of the stem, and new cavities
can be added or deleted (Fig. 1a) to maintain a cavity depth ratio (L/ D -
depth of the cavity divided by the diameter of the cylinder) of approximately
0.07 ± 0.0015 (Geller and Nobel, 1984). Apices of the cavity junctures are
adorned with whorls of 15 to 30 spines 2.5 to 7.6 em long (Benson, 1981).
In order for wind to be a selective agent on saguaros, high wind velocities
must occur in saguaro habitats and they must affect their reproductive suc-
cess. Within the distribution of saguaros, high wind velocities were recorded
15 m above the ground for a nine-year period (Nobel, 1994). The maximum
wind velocity recorded was 38 m/s,(Re = 106 ), and velocities exceeding 22
m/s (Re = 7 x 105 ) occurred almost every month. Saguaro habitats contain
less vegetation cover than other ecosystems and, consequently, have few if
any other tall plants to shelter them from the wind (Fig. 1b). There is sub-
stantial circumstantial evidence that wind gusts exert enough force to topple
saguaros, and thus, cause their premature mortality (Fig. 1c; Benson, 1981;
Alcock, 1985; Pierson and Turner, 1998). Information on the wind velocities
required to topple large desert succulents is lacking. Consistent with the natu-
ral selection scenario, some saguaros are toppled by gusts, while many others
remain standing. Considering most tall cacti live for 150 years and take over
30 to 50 years to reach reproduce maturity, strong gusts need only to occur
every 30 to 50 years to be important in the natural selection of tall succulent
morphology.
2 Numerical Method
Numerical simulations of the flow around a cactus section are carried out by
solving the Reynolds-Averaged Navier-Stokes equations in two dimensions.
Two codes are used: INS2D (Rogers and Kwak, 1990) and Fluent (Fluent,
1999). INS2D is an upwind based, third-order accurate code for structured
(multiblock) grids; the artificial compressibility approach is used for pressure-
velocity coupling and the time integration is second-order accurate. Fluent is
an unstructured-mesh solver based on second-order accurate spatial and time
discretization; the SIMPLE technique is used for pressure-velocity coupling.
Turbulence modeling is based on the v 2 - f model (Durbin, 1995; Iaccarino,
2001).
Unsteady Flow Around Cylinders with Cavities 67
3 Computational Grids
Cylinders with v-shaped cavities (with cavity ranging from L/ D = 0.0 to
L/ D = 0.105) are considered. Several meshes have been generated to assess
the sensitivity of the solution. In Fig. 2, examples of the grids are reported.
Simulations using the structured grids (Fig. 2a and 2b) have been performed
using both Fluent and INS2D. The structured grid is generated as an 0-
type mesh wrapped around the cylinder. The cavities are slightly smoothed
to improve the orthogonality of the grid lines at the cylinder surface. The
height of the first cell is adjusted according to Re; the distance from the far
field boundary is 25D as used in Rogers and Kwak, 1990. The unstructured
meshes are generated using a quadrilateral paving technique (Blacker et al.,
1991); this approach allows flexibility in clustering the grid cells in the wake
region and close to the surface.
In Table I, results are reported for the computations performed on differ-
ent grids at a very low Reynolds number. The flow is unsteady and exhibits a
periodic vortex shedding from the cylinder; only the averaged drag coefficient
is reported. Grid convergence is achieved for the smooth cylinder L / D = 0
using both the structured and the unstructured grids, and the correspond-
ing values are extremely close. The results for the flow around the cylinders
with cavities show that grid convergence is achieved only using the unstruc-
tured grids. An increase in cavity depth requires a finer resolution to capture
accurately the in-cavity flow; in addition, the quality of the structured grid
degrades as the cavity depth increases. It is worth noting that the results ob-
tained using the finest structured grid (761 x 201) are in good agreement with
the converged results for the unstructured mesh. In the following sections only
results computed using the unstructured grids are reported.
4 Laminar Simulations
Flow simulations at low Reynolds number (Re = 100 and Re = 200) are
carried out to evaluate the effect of cavity depth (and the accuracy of the
predictions) without uncertainties related to the turbulence modeling. Two-
dimensional simulations have been performed with unstructured grids using
68 G. Iaccarino, P. Durbin, and S. Talley
6,000 to 42,000 elements (only the fine mesh results are presented but the
results appear to be already insensitive ot the mesh for a grid size of 25,000
elements). The calculations are carried out using a timestep L1tUI D = 0.01
(corresponding to approximately 35 time steps per vortex shedding period)
and for a total time of TU I D = 150. Simulations have been carried out
using a smaller time step (L1tUI D = 0.0065 and the lift and drag coefficient
changed by less than 1.5%). The time history of drag and lift coefficients at
Re = 100 are reported in Fig. 3a and 3b respectively. The statistics (time
averaged values and the Strouhal number, etc.) are computed over a period
Tav = 50D IU and are reported in Table II.
LID Cd Ct St LID Cd Ct St
0 1.339 ± 0.010 ± 0.330 0.160 0 1.365 ± 0.037 ± 0.664 0.175
0.035 1.304 ± 0.011 ± 0.325 0.161 0.035 1.361 ± 0.045 ± 0.713 0.172
0.070 1.309 ± 0.010 ± 0.334 0.162 0.0701.364 ± 0.057 ± 0.742 0.172
0.105 1.318 ± 0.012 ± 0.336 0.161 0.105 1.381 ± 0.049 ± 0.740 0.170
Re = 100 Re = 200
Table II. Statistics for low Reynolds number flow around cacti.
St is the Strouhal number based on the frequency of the lift oscillations, f c 1 DIU
The results indicate a small drag reduction (~ 10%) associated with the
presence of the cavities. The cavity depth Ll D = 0.05 is somewhat optimal.
The change in the unsteady lift is also small, showing that the effect of the
cavity is limited.
The results presented for the smooth cylinder at Re = 200 are in good
agreement with the numerical simulations and the experimental data reported
in Rogers & K wak ( 1990). It is worth noting that Re = 190 represent the onset
of three-dimensional flow in the wake of the cylinder.
5 Turbulent Simulations
Calculations at Re = 20,000 andRe= 100,000 (subcritical regime) are per-
formed using the v 2 - f turbulence model. The time step, the simulated time
and the averaging time are the same as before; the time history of lift and
drag are reported in Fig. 4.
As compared to the results presented at low Re, the drag reduction is now
larger (~ 25%). The strength of the unsteady motion is also greatly reduced
as reported in Table III.
Unsteady Flow Around Cylinders with Cavities 69
L/D Cd Ct St L/D Cd Ct St
0 1.683 ± 0.164 ± 1.923 0.217 0 1.644 ± 0.113 ± 1.791 0.228
0.035 1.452 ± 0.076 ± 1.562 0.221 0.035 1.464 ± 0.120 ± 1.462 0.224
0.070 1.419 ± 0.083 ± 1.245 0.224 0.070 1.401 ± 0.131 ± 1.128 0.221
0.105 1.359 ± 0.052 ± 0.987 0.223 0.105 1.325 ± 0.079 ± 0.864 0.221
Re = 20,000 Re = 100,000
Table III. Statistics for high Reynolds number flow around cacti
From the results presented in Table III, it appears that the cavity depth
has a relatively strong effect on the drag and a substantial dampening effect
on the unsteady motion.
The time averaged turbulent kinetic energy for the four geometries con-
sidered is reported in Fig. 5; the intensity very close to the cylinder decreases
with the cavity depth, but higher values are observed in the near wake.
The comparison of the computed Cd with the experimental values for the
smooth cylinder (Achenbach, 1971) shows an overprediction of about 20%.
The flow over the smooth cylinder in the subcritical regime is characterized by
a laminar boundary layer separation; turbulence is generated in the separated
shear layer and is sustained in the near wake. The smooth cylinder calculations
(L/ D = 0) are carried out with the v2 - f turbulence model switched off for
() ~ 90° in an effort to force a laminar separation in the simulations. It is
well known that RANS turbulence models typically anticipate transition and,
even with first part of the boundary layer forced to be laminar, in the present
calculations the shear layer separates with very high level of turbulent kinetic
energy. In addition, in the subcritical regime three-dimensional effects in the
wake are substantial and not account for in the present calculations.
The simulations with cavities are carried out with the turbulence model
switched on from the stagnation point (B = 0°) because it is expected that
transition occurs immediately after the first cavity. In addition, it is expected
that three-dimensional effects are less substantial in these cases (as observed
in the experiments).
A comparison of experimental and computed velocity profiles in the wake
is reported in Fig. 6. The results for the smooth cylinder confirm that the
calculation overestimate the drag (corresponding to the larger velocity defect
in the wake); on the other hand, the data for the cylinder with cavities show
a remarkable agreement.
6 Conclusions
200. The effect of the cavity is limited and only a slight drag reduction is
obtained. At higher Reynolds numbers (20, 000 and 100, 000) the effect of the
cavities is more substantial with a considerable reduction of the drag and,
perhaps more importantly, a strong damping of the oscillating lift.
(a) (c)
Fig. 1. (a) Addition of cavities (ribs) on an adult saguaro trunk (b) Saguaro forest,
and (c) Root syst em of a saguaro toppled by t he wind.
Unsteady Flow Around Cylinders with Cavities 71
Fig. 2 . Example of the computational Grids: (a-b) structured grids 241 x 100 ele-
ments; (c-d) unstructured grids~ 20,000 elements (a-c) L/ D = 0 (b-d) L/ D = 0.07
1.4 r r - - - - - - -- - - - - - - - ,
r..7
1.2
(a) (b)
Fig. 3. Time history of drag (a) and lift (b) coefficients. Re = 100. Solid line:
L/ D = 0; Dotted line: L/ D = 0.07
72 G. Iaccarino, P. Durbin, and 8. Talley
2
2
1.5
·I
·•.. /'
'······ -2
0 25 50 75 0 25 50 75
I ' /D tU/ D
(a) (b)
Fig. 4. Time history of drag (a) and lift (b) coefficients. Re = 20,000 . Solid line:
L/ D = 0; Dotted line: L/ D = 0.07
References
1. ACHENBACH , E. 1971. Influence of surface roughness on the cross-flow around
a circular cylinder. J. of Fluid Mech. 46, 321-335.
2. ALCOCK, J. 1985. Sonaran Desert Spring. The University of Chicago, Chicago.
3. BENSON, L. 1981. The Cacti of Arizona. The University of Arizona Press, Tuc-
son.
4. BLACKER T.D. , M .B. STEPHENSON & S . CANANN 1991 Analysis automation
with paving: A new quadrilateral meshing technique Advances in Engineering
Software, 56, 332- 337.
5. DURBIN , P.A. 1995 Separated flow computations with the k-E-v 2 model , AIAA
J., 33 659-664.
6. GELLER, G . N ., AND NOBEL, P. S. 1984. Cactus ribs: influence on PAR inter-
ception and C02 uptake. Photosynthetica 18 , 482-494.
7. HODGE, C. 1991. All About Saguaros. Hugh Harelson-Publisheer, Phoenix.
8. IACCARINO , G . 2001 Predictions of a turbulent separated flow using commercial
CFD codes, J. Fluids Engineering, 123, 1-10.
9. FLUENT INC. 1999 Fluent V5.3 User Manual.
10. NOBEL, P. S. 1994. Remarkable Agaves and Cacti. Oxford University Press,
New York.
11. PIERSON , E . A. , AND TURNER, R . M. 1998. An 85-year study of saguaro
( Carnegiea gigantea) demography. Ecology. 79, 2676-2693.
12. ROGERS , S. E . AND KWAK, D. 1990 An Upwind Differencing Scheme for the
Time Accurate Incompressible Navier-Stokes Equations AIAA J., 28 , 253- 262.
Unsteady Flow Around Cylinders with Cavities 73
(a) (b )
(c) (d)
Fig. 5. Time averaged turbulent kinetic energy: (a) L/D = 0; (b) L/D = 0.035; (c)
L/D = 0.070; (d) L/D = 0.105
1
~075
075
05~ .• ~~~0~~~~~
. ~~~ ·2 0
)nJ liD
(a) (b)
Fig. 6. Velocity profiles in the wake of cylinders. Solid line: simulations (Re =
100,000); Circle: experiments (Re = 125,000). (a) smooth cylinder (b) cylinder with
LID 0.07.
74 G. Iaccarino, P. Durbin, and S. Talley
Ken Wurtzler
1 Introduction
The automotive industry has a high demand for reliable simulation meth-
ods capable of tackling the complex turbulent air flow around vehicles. The
Ahmed reference model is a generic car-type bluff body with a slant back. It
is frequently used as a benchmark test case for this kind of flow. In spite of
the relatively simple geometry of the Ahmed body, the flow around it retains
some main features of the flow around real cars.
The Ahmed body, Fig. 1, was first defined and its characteristics described
in the experimental work of Ahmed et al [1]. Therein, it is shown that most
of the drag of the body is due to pressure drag, which is generated at the rear
end. The structure of the wake is very complex, with a separation zone and
counter-rotating vortices coming off the slant side edges, whose strength is
mainly determined by the base slant angle. The maximum drag was found for
a critical slant angle of 30 ° . Above this angle a sudden drop in drag occurs
which corresponds to a change in the wake structure. Below this angle, strong
counter-rotating vortices are present and the flow separates in the middle
region of the top edge and reattaches at the sloping surface. For angles above
the critical angle, the counter-rotating vortices are weaker, the separation
occurs along the entire top and the side edges and there is no reattachment.
More recently, Lienhart et al [2] performed more detailed experiments on
the same body, albeit at a somewhat lower velocity. They measured the mean
and fluctuating velocities by LDA and obtained surface oil-flow pictures for
two rear vehicle body slant angles (<p = 25 ° and <p = 35 ° ) , i.e. just below and
above the critical angle. Their results show the differences in flow behaviour
for the two angles considered.
In recent years, there have been several computational studies of this flow.
Han [3] and Gillieron and Chometon [4] using the RANS approach obtained
qualitatively good results in terms of flow structures but they do not show
velocity profile comparisons. In two recent workshops [5, 6] the flow around
the Ahmed body was a test case and the results presented there have shown
78 C. Hinterberger, M. Garcia-Villalba, and W. Rodi
1044mm 389
:
CX)
CX)
N
~z 50
y
163.5
0
L
30
The present configuration has been specifically chosen to match the experi-
ment of Lienhart et al (2]. The Ahmed body, whose shape and dimensions are
shown in Fig. 1, was mounted in a 3/4 open test section (floor, but no side
walls or ceiling). The bulk velocity was 40 mfs. This results in a Reynolds
number Re = 2.8 x 106 (based on the length of the body L = 1.044 m),
which is of the same order of magnitude but somewhat lower as the one in
the original experiment of Ahmed et al (1] (Re = 4.3 x 106 ).
Large eddy simulation of flow around the Ahmed body 79
The simulations were performed with the Finite Volume Code LESOCC 2
(Large Eddy Simulation On Curvilinear Coordinates) which is an enhanced
fully-parallelized version of the code LESOCC [9], developed at the Institute
for Hydromechanics. It solves the incompressible 3D time-dependent filtered
Navier-Stokes equations on body-fitted curvilinear block-structured grids us-
ing second order central differences for the discretisation of the convective and
viscous fluxes. Time advancement is accomplished by an explicit, low-storage
Runge-Kutta method. Conservation of mass is achieved by the SIMPLE algo-
rithm, with the pressure-correction equation being solved by the SIP proce-
dure. The momentum-interpolation method of Rhie and Chow [10] is employed
to prevent pressure-velocity decoupling and associated oscillations.
The flow around the Ahmed body was a test case in two recent workshops
[5, 6], the computational domain was defined in them. The domain extends to
five body lengths behind the body to ensure that the outflow condition does
not affect the near-body wake. The inflow plane is placed at a distance of 1.3
body lengths upstream of the body. This distance corresponds roughly to 5
body heights where blocking effects dissapear. At the inflow section a uniform
axial velocity profile is imposed. The width of 1.87 m (5 body widths) and
the height of 1.4 m (5 body heights) are taken analogous to the experiments.
At both side boundaries and at the top boundary, free-slip conditions are
applied. At the outflow of the domain, a convective boundary condition for the
velocities is used. Finally, wall functions are used because of the high Reynolds
number which does not allow a fine resolution of the near wall flow down
to the viscous sub-layer. The wall function is similar to the Werner-Wengle
[11] approach but assuming an instantaneous logarithmic profile instead of a
power law profile. It is applied at the walls of the vehicle and at the bottom
of the channel. In order to account for the unresolved subgrid-scale motions,
the Smagorinsky subgrid-scale model is employed with a model constant of
Cs = 0.13.
2.3 Grids
and streamwise direction, an outer 0-grid structure was chosen in Grid2, Fig
2(b}. In addition, Grid 2 includes more refinement in the region of the slant
back, especially close to the top and side edges, Fig 2(c-h). For both grids,
the near-wall cell centre has a wall distance on average of about 40 wall units
( .::ly p +), but it varies from approximately 10 in the separated regions along
the slant back to 150 close to the top rear edge. The spanwise and streamwise
extent of the grid cells is up to a factor of 10 larger for Grid 2, except in the
refinement regions close to the edges. For Grid 1 this factor is even larger. This
means that the boundary layer at the body surface is highly underresolved.
(g) (h)
Fig. 2. Typical grids used for the LES. {a,c,e,g) Grid 1. {b,d,f,h} Grid 2. (a,b}
Sketches of the block structure. (c-h) Slices in different planes showing the grid
point distribution in the body region. {c,d) Cut in xz-plane . (e,J) Cut in xy-plane.
{g,h} Cut in yz-plane.
Large eddy simulation of flow around the Ahmed body 81
3 Results
3.1 Time averaged profiles
For the front part, mean streamwise velocity profiles in the symmetry plane
are compared with the experiment in Fig. 3. It can be seen that the flow
upstream of the body and in the free stream above it is properly represented
in the simulation. This is because in this region the level of turbulence is so
low that the flow is nearly potential flow.
600 - l
I
l
I
I I
I I
I I
500 r- I I
I I
I I
I I
I I
400 r- I
I
I
I
\ I I
v
I I
E'
E 300 t--
"N
200 -
I-; ~~~I
':j
I
100 -
i~
I I
1 I j I
] I I I I I)
0
-1400 -1200 -1000 -800 -600 -400
x[mm)
In Figures 4 and 5, the mean streamwise velocity profiles and root mean
square velocity fluctuations are compared with the experimental results in
the rear body part and the near wake (in the symmetry plane) . The general
agreement with the experiment is reasonably good taking into account that
neither grid is fine enough to resolve adequately the boundary layer develop-
ing on the body up to the slant back. However there are some discrepancies
between the computations and the experiment mainly concerning the velocity
profiles on the slant back. In the experiment, the flow separates right at the
corner of the sloping surface and it reattaches in the middle of the surface.
In the simulation, the flow first stays attached before it separates somewhat
downstream of the corner and no reattachment occurs on the slant back. This
is most likely due to the poor resolution of both grids near the wall leading
to an incorrect prediction of the approaching boundary layer.
The main differences between the results of the two simulations are found
at the beginning of the sloping surface. There, Grid 2 is much finer than Grid
1, and it can be seen that the turbulence intensities are very close to the
82 C. Hinterberger, M. Garda-Villalba, and W. Rodi
o EXP
LES coarse grid
- LES fine grid
300
250
'E
* 200
150
100
50
0
-250 250
Fig. 4. Mean streamwise velocity proilles in the rear body and near wake (symmetry
plane)
experimental ones for Grid 2 while they are too low for Grid 1, Fig. 5. In
addition, the prediction of the separation point is closer to the experiment
in the simulation performed with the finer grid, in which the flow separates
earlier, Fig. 4.
The complex unsteady flow on the slant back leads to high fluctuation
intensities which are very difficult to predict. In fact, no RANS method has
succeeded in predicting the high values observed in the experiments [5, 6]
while the present LES calculations (Fig. 5) have.
Figure 6 shows the mean velocity vectors in the symmetry plane, Fig.
6{a-b}, and in three transverse yz-planes, Fig. 6{c-h). The streamwise loca-
tion of these transverse planes is indicated in Fig. 6{a) by dashed lines. The
colour of the vectors corresponds to turbulent kinetic energy. From the com-
parison of the experiment, Figs. 6{a,c,e,g) on the left, with the simulation,
Figs. 6{b,d,j,h) on the right, it is clear that the main flow structures are well
captured in the simulation. The size and the extent of the recirculation zone
behind the body are well predicted, Fig. 6{b). The development of the counter-
rotating vortices, which can be seen from the secondary flow vectors in the
Large eddy simulation of flow around the Ahmed body 83
450
0 EXP
400 LES coarse grid
LES fine grid
300
'E
E
'N
Fig. 5. Root mean square streamwise velocity fluctuations in the rear body and
near wake (symmetry plane)
y-z planes, is also in close agreement with the experiment. In Fig. 6{d), it can
be seen that these vortices develope half-way down the slant back, they grow
while they approach the end of the body, Fig. 6{/), and they are strong and
fully developed in the near wake, Fig. 6(h). The level of turbulence obtained
in the simulation is also in good agreement with the experiment, as can be
seen from the coloured regions in Figure 6.
The calculation results show complex time-dependent flow features in the wake
region. In Fig. 7, two typical instantaneous velocity fields in the symmetry
plane are shown. As discussed in section 3.1, the prediction of the flow on the
sloping surface is very difficult. Thus, in Fig. 7 on the left, an instantaneous
field is captured in which the flow tends to reattach on the slant back However,
in Fig. 7 on the right, the flow in that zone is completely separated. Note that
in the experiment the flow reattaches (at least in the mean). The unsteadiness
of the near body-wake is clearly well captured. The flow between the body and
84 C. Hinterberger, M. Garda-Villalba, and W. Rodi
= -------------------
~-
- - - -- -- "' h= O(mm] j ------------------~-
--------~--------- = O~m
11
350 300-~~~~~~~~~~~~~~~~~------
....
;::
;::
;::
;::
@ ;::::::
~ 300 ,_ ---------------
----~-- - -------~----~
- --~--------------- x-z
---------------- ,
1 ....
---------------
_ _ _ _ ..._..., .... _ _ _ _ _ _
~ ~ ::::
::::
:::: 250
' i :::~ ~~~~~::::::::::~:::~
:: ~~~~~~2:~:::::::
....~. ...........
......... .... , , ...................
~ § ... ,,,,.... ,
~ ~
......................... .....
... ..' ,, ,.,' ,.........................
.......................... .....
;:::nnH
~
200 .....
~ ~
::::
~
.....
.................... ....
"
~
! r
:~~: ~~~~.::::::::::
~ 150
~ ~ 100
50
0
(a) -; -
-200 - 100 200
xrnincr' 300
500
300
'"'
200
...
300
200
150 150
0 50 100 ISO 200 0 ~1001502'00
300
300
'"'
200
150
...'"'
150
100 100
50 {t) 50 (/)
o_..., -300 300 •oo o,.oo -300 -200 -100 0'11""" 100 200 300 •OO
IEXPI' .................
................ ... .
......... ,,, , .....
~mm) l
-z
·--
500
450
........
........ , ........ .
................
,
. . :[~ 200 [mmil
........................
························· -z
400
350
........ , •• 11111101 •
.................. ,111111111''''
•••
••
•
···························
........... ,.,,,,, ,, ,,,,,,, ........... .
• ' ' ' ' ' ' 0 0 0 0 0 ' 0 I I I I I I I I I I I I I 0 0 ' ~ ~'' •
.·:<~ll~~mmtp:: ·:::WH§H~~L-
.-
• ' • • ' .. " '" " ' " \ \ I I I I ~ I I I ' • • • • • • •
::w_)~:7illl1 /1 111111jtii~ ..
300
'"' 250
..·:;:::;;s~~~
::::::---~ff l'J~r:~:-··
j'j···:::::::
::;:; l\
200 200 0
150 I '_,
150
100 . '• • '': f I I:I•• :- - ' '
..
100
(h) )iH;;: ' . u !:\\! .:.
"'
0-400
(g)
o,.oo
-300 ·200 200 300 •oo -:100 .,.. 100 0 f,ll'"'lj 100 200 300 400
Fig. 6. Mean velocity vectors coloured by turbulent kinetic energy. (a,b) symmetry
plane y = 0 mm. (c,d) Close to the middle of the slant back x = - 88 mm. (e,f) End
of the body x = 0 mm. (g,h} Near-body wake x = 200 mm. (a,c, e, g) Experiment.
(b,d,f,h) Simulation performed with Grid 2.
Large eddy simulation of flow around the Ahmed body 85
the ground plate has a rather strong influence on the shape of the recirculation
zone.
4 Conclusions
A Large Eddy Simulation of the flow around the Ahmed body with a slant-
back angle of 25 ° was performed. The flow around the Ahmed body is a very
86 C. Hinterberger, M. Garcfa-Villalba, and W . Rodi
1:~
challenging problem because of the complex geometry and the high Reynolds
number. The results obtained by the simulation are promising; the comparison
with the experiments shows that the flow structures are well captured. The
agreement of the time-averaged quantities is good although some discrepancies
are present, especially in the lower part of the slant back. The results also show
that this is a good test case for further development. The calculations for the
configuration with a slant angle of 35 ° are in progress. Improvements of the
method (subgrid-scale model, discretisation, wall modelling, etc) are required
to obtain good results for the 35 ° case.
Acknowledgments
This work was funded through the EU TMR-project "LES of Complex Flows".
The calculations were carried out on the IBM RS/6000 SP-SMP high perfor-
mance computer of the University of Karlsruhe, and the assistance of Mr.
Gernert is gratefully acknowledged. The authors are also grateful to Dr. J.
Frohlich for many helpful discussions.
References
1. S.R. Ahmed, G. Ramm, and G. Faltin. Some salient features of the time averaged
ground vehicle wake. SAE paper no. 840300, 1984.
2. H. Lienhart, C. Stoots, and S. Becker. Flow and turbulence structures in the
wake of a simplified car model (Ahmed model). In DGLR Fach Symp. der AG
STAB, 2000.
3. T. Han. Computational analysis of three-dimensional turbulent flow around a
bluff body in ground proximity. AIAA Journal, 27:1213-1219, 1989.
4. P. Gillieron and F. Chometon. Modelling of stationary three-dimensional sepa-
rated air flows around an Ahmed reference model. In ESAIM proc., volume 7,
pages 173-182, 1999.
5. S. Jakirlic, R. Jester-Ziirker, and C. Tropea, editors. Proceedings of 9th ER-
COFTAC IAHR COST Workshop on refined turbulence modelling, 2001.
6. R. Manceau and J.P. Bonnet, editors. Proceedings of 10th ERCOFTAC IAHR
QNET-CFD Workshop on refined turbulence modelling, 2002.
7. S. Kapadia, S. Roy, and K. Wurtzler. Detached eddy simulation over a reference
Ahmed car model. AIAA paper no. 2003-0857, 2003.
8. R.J.A. Howard and M. Pourquie. Large eddy simulation of an Ahmed reference
model. Journal of Turbulence, 3, 2002.
9. M. Breuer and W. Rodi. Large eddy simulation of complex turbulent flows of
practical interest. In E.H. Hirschel, editor, Flow simulation with high perfor-
mance computers II, volume 52 of Notes on Numerical Fluid Mechanics, pages
258-274. Vieweg, Braunschweig, 1996.
10. C.M. Rhie and W.L. Chow. Numerical study of the turbulent flow past an airfoil
with trailing edge separation. AIAA Journal, 21(11):1061-1068, 1983.
11. H. Werner and H. Wengle. Large-eddy simulation of turbulent flow over and
around a cube in a plate channel. In 8th Symp. on Turb. Shear Flows, 1993.
Detached-Eddy Simulation of the Ground
Transportation System
Computational Fluid Dynamics ( CFD) is a useful tool for analysis and is in-
creasingly relied upon in the design process for applications within the trans-
portation industry. Emphasis on fuel efficiency, for example, will require in-
novative approaches to vehicle design and testing. For heavy trucks at typical
highway speeds the power to overcome aerodynamic drag accounts for more
than half of the total fuel consumption [1]. CFD offers a powerful approach
that should be able to efficiently screen configurations, prior to more costly
and time-consuming wind-tunnel and field tests.
Prediction of the flow fields around complex configurations and for regimes
of technological interest continues to strongly challenge CFD. Many of the
regimes in vehicle aerodynamics occur at high Reynolds numbers and for
which much of the flow around the configuration of interest is turbulent. Tur-
bulence treatments at application Reynolds numbers are one of the primary
obstacles in the advancement of CFD as a more routinely applied tool in
engineering analysis and design.
Especially problematic in predicting the flow around vehicles at high
Reynolds numbers are regions of massive separation. For many engineering
systems, high-Reynolds number predictions are obtained from solutions of the
Reynolds-averaged Navier-Stokes (RANS) equations. While the most popular
RANS models appear to yield predictions of useful accuracy in attached flows
as well as some with shallow separations, RANS predictions of massively sep-
arated flows have typically been uneven. In vehicle wakes, for example, it is
not anticipated that RANS will provide accurate predictions of back pressures
and, therefore, of the drag. In addition, in many studies that focus on tech-
niques for reducing drag, alterations to the vehicle geometry result in strong
modifications of the flow in the wake, precisely the region for which RANS
modeling is the least reliable. Other approaches that offer higher-fidelity (and
presumably greater accuracy) than RANS and are numerically feasible today
are desired for applications.
The issues outlined above provided part of the motivation for detailed
experimental studies of the Ground Transportation System (GTS) for which
a series of experimental measurements were reported by Storms et al. [3].
The GTS is a simplified tractor-trailer configuration, the geometry is rounded
without a tractor-trailer gap and without wheels. Measurements were acquired
in the NASA-Ames 7 x 10ft wind tunnel for a range of Reynolds numbers, yaw
angles ranging from -14 to 14 degrees in 2.5 degree increments, and a Reynolds
number range from 3 x 10 5 to 2 x 10 6 . Storms et al. [3] acquired pressure
measurements at several locations on the GTS, in addition to measurements of
the drag, spectra of the pressure fluctuations, and Particle-Image-Velocimetry
measurements in the wake. One of the key contributions of the study was
investigation of the influence of boattail plates mounted on the base of the
GTS and used to investigate drag reduction.
DES of the Ground Transportation System 91
The objective of the current effort is prediction of the flow field for a sub-
set of the conditions considered by Storms et al. [3]. The approach adopted
in the present investigation is prediction of the flow around the GTS using
Detached-Eddy Simulation (DES). DES is a hybrid method which has RANS
behavior near the wall and becomes a Large Eddy Simulation in the regions
away from solid surfaces provided the grid density is sufficient [2]. The for-
mulation of the model used in this work is based on a modification to the
Spalart-Allmaras one-equation model [4], referred to as S-A throughout and
described in greater detail in the next section. DES is a non-zonal technique
that is computationally feasible for high Reynolds number prediction, but
also resolves time-dependent, three-dimensional turbulent motions as in LES.
Previous applications of the method have been favorable, yielding adequate
predictions across a range of flows and also showing the computational cost
has a weak dependence on Reynolds number, similar to RANS methods yet
at the same time providing more realistic descriptions of unsteady effects (see
also [5]). This paper presents simulations for two yaw angles, 0° and 10°, using
DES. The predictions are assessed against experimental measurements [3] as
well against solutions of the steady-state flow obtained using the un-modified
Spalart-Allmaras [4] one-equation model.
2 Approach
Dv
Dt = CblS
~ v-
~ [cwdw - cb1
--;;}2 ft2
] [z;]d 2
fv2 = (1 + ..£) -
Cv2
3
(3)
- [ 1 + c~3
fw- g 6 + cw3
6
] 1/6
g = r + Cw2 ( r 6 - r) ,
g
In (5), ..:1 is the largest distance between the cell center under consideration
and the cell center of the neighbors (i.e., those cells sharing a face with the cell
in question). In "natural" applications of DES, the wall-parallel grid spacings
(e.g., streamwise and spanwise) are on the order of the boundary layer thick-
~ess and the S-A RANS model is retained throughout the boundary layer, i.e.,
d = d. Consequently, prediction of boundary layer separation is determined
in the "RANS mode" of DES. Away from solid boundaries, the closure is a
one-equation model for the sub-grid scale eddy viscosity. When theyroduction
and destruction terms of the model are balanced, the length scaled= CDEs.6.
in the LES region yields a Smagorinsky-like eddy viscosity v ex S ..:1 2 . The ad-
ditional model constant CDES = 0.65 was set in homogeneous turbulence
[6].
and viscous Jacobians is used for advancement of the discretized system. For
time-accurate computations, a Newton sub-iteration scheme is employed, the
method is second order accurate in time. The domain decomposition library
ParMETIS [9] is used for parallel implementation and provides optimal load
balancing with a minimal surface interface between zones. Communication
between processors is achieved using Message Passing Interface.
In the following, lengths are non-dimensionalized by the GTS width w, x
identifies the longitudinal axis of the GTS , y is the coordinate normal to the
lower tunnel wall, and z identifies the spanwise dimension. In terms of the GTS
width w, the total length is 7.647w and height is 1.392w. The computational
domain included the sidewalls and upper tunnel walls (including their one
degree divergence). The no-slip condition was applied on the tunnel walls and
the boundary layers were resolved. The upstream section of the computational
domain was extended approximately 15w in front of the GTS in order that
the boundary layer on the lower wall develop the same thickness as measured
at the inlet to the test section in Storms et al. [3]. The exit boundary of the
computational domain was located approximately 13.5w from the rear surface
of the GTS.
Fig. 1. Side, top, and front views of the GTS. Top view shown in lower-left frame
shows the grid in a plane for the computations of the flow at 10° yaw with a clustering
of cells into the leeward region. Front view shows grid clustering around both sides
of the GTS for the computations at 0° yaw.
The unstructured grids were generated using Gridgen [10], with prisms in
the boundary layer and tetrahedra elsewhere. The wall-normal distribution
of cells was clustered near solid surfaces such that average distance to the
first cell center from the wall was less than one viscous unit. The grids were
comprised of approximately 6 x 106 cells. Shown in Figure 1 are front, side,
94 S. Maddox et a!.
and top views of the GTS along with crinkle cuts of the grid. Each of the views
shows a clustering of grid cells in the vicinity of the GTS, with the side view
in Figure 1 showing a biasing of points into the wake region behind the GTS.
This clustering is achieved by using Gridgen's multiblock capability. The top
view in the figure shows a cut of the grid used for the computations of the
flow at 10° yaw. A wedge-shaped distribution of finer resolution is shaded into
the leeward region in order to provide improved resolution of the separated
structures along the leeward side. The front view in the figure is from the grid
used in the computations at 0° yaw, a symmetric grid clustering is evident in
the frame.
3 Results
Summarized in this section are DES and RANS predictions of the flow around
the GTS for yaw angles of 0° and 10°. The DES predictions are of the un-
steady flow , the dimensionless timestep (non-dimensionalized using the GTS
width w and upstream speed Uin) was 0.02. RANS predictions presented in
this manuscript are of the steady-state solution, obtained on the same grids as
the DES results , the calculations having been driven to convergence in compu-
tations performed using very large timesteps (corresponding to a global CFL
number of 106 ).
Fig. 2. Contours of the instantaneous vorticity in the GTS wake from the DES (left
frame) and of the steady-state vorticity from the RANS (right frame).
Fig. 3. Contours of the instantaneous vorticity in a plane midway between the lower
tunnel wall and lower surface of the GTS. DES prediction of the flow at 0° yaw.
the wake down to roughly the grid scale. Figure 3 shows vorticity contours of
the instantaneous solution predicted in the DES at 0° yaw in a plane parallel
and midway between the lower tunnel wall and lower surface of the GTS. The
posts on which the GTS was mounted were included in the computations and
as shown by the figure, the grid used for the time-dependent DES prediction
was sufficiently fine to resolve vortex shedding in the wake of the posts.
Fig. 4. Instantaneous vorticity isosurface colored by pressure from the DES, 10°
yaw.
96 S. Maddox et al.
---
Fig. 5. Streamlines (left frame) and velocity vectors (right frame) from the DES
prediction of the flow at 10° yaw. GTS surface colored by pressure in the left frame,
velocity vectors colored by the eddy viscosity ratio in the right frame.
Also apparent in Figure 4 is a separated region near the front leeward side
of the GTS. Two views through the separated region are shown in Figure 5.
The top view is a plane at y = 0.696w (midway from the lower surface to
the roof of the GTS) and shows a separation bubble with reattachment at ap-
proximately xjw ~ 0.2, substantially further along the GTS than indicated by
pressure measurements in Storms et al. [3]. The front view in the right frame
of Figure 5 shows the counter-rotating structures that "close" the separation
bubble.
Shown in Figure 6 are the time histories of DES predictions of the body-
axis drag coefficient Cd and side force coefficient C 8 • Only a portion of the
force histories from a statistically-stationary portion of the simulation are
shown (statistics were acquired for more than 80w/Uin)· The figure shows
DES of the Ground Transportation System 97
1.5
--- c.(10°y&W)
--- c.(10°yaW)
--- c.(0°yaw)
-----· c. (0° yaw)
r-------~--~---------
10 20 30
Fig. 6. Temporal evolution of the body-axis drag coefficient, Cd, and side-force
c.,
coefficient, from the DES predictions of the flow at 0° and 10° yaw.
that for both yaw angles the variations in the body-axis drag are smaller than
occurring in the side force. For 0° yaw the side-force coefficient shows a lower-
frequency meandering that occurs at a Strouhal number around 0.2. For 10
degrees yaw, a less apparent low-frequency component in Cs is observed.
Summarized in Table 1 are the time-averaged body-axis and side force
coefficients from the DES and RANS. Also tabulated are the experimental
measurements from Storms et al. [3]. At 0° yaw, the RANS prediction of the
drag coefficient is substantially larger than the measured value of 0.249 from
Storms et al. [3]. The DES prediction is only 12% over the measured value; the
realistic treatment of the separated region in the GTS wake appears to be re-
sponsible for the improved predictions of the drag force (see also Figure 7). For
10° yaw both the DES and RANS predictions of the body-axis drag coefficient
are above the measured value, slightly better agreement between simulations
and measurements is observed in the side-force coefficient.
1.5
--DES
~~ ~~ RANS
0.5 0 measurements
1.5
'
!o
I --DES
~- - RANS
0.5 0 measurements
·o
\
and Figure 8 are the symmetry plane pressure coefficients for the GTS at 0°
and 10° yaw, respectively. The pressure coefficients are plotted as a function
of the vertical coordinate in order to provide an indication of the drag. In
general, both figures show that the stagnation pressure on front of the GTS is
captured with the acceleration around the front face also accurately recovered.
The pressure distributions predicted by the DES and RANSon the front face
are essentially identical, differences between the two techniques are apparent
along the rear surface. The DES prediction of the pressure coefficient exhibits
relatively little sensitivity to the vertical coordinate and is nearly uniform,
characteristic of the uniform pressure over surfaces in the separated regions of
massively separated flows. The RANS prediction, on the other hand exhibits
DES of the Ground Transportation System 99
more substantial variation with a relatively strong suction near the upper and
lower surfaces. This in turn leads to an over-prediction of the drag .
.0.1
.0.2
u..
--DES
.0.3 ----RANS
0 measurements
.0.4
·0.50 4 6
x/w
0
0
0 p
.0.1 0 -/7
'I
r
.0.2
.0.3 --DES
- - - - RANS
0 measurements
·0.4
-o.so!--'----'-.........~l:---'---'---'-......._41:---'---'---'-......._s~---'---'-~
x/w
Pressures along the top of the GTS for 0 and 10 degrees yaw are shown in
Figure 9 and Figure 10, respectively. The pressures are plotted along the GTS
centerline (zjw = 0) on the top of the model (yjw = 1.392). The effect of the
model surface curvature above the cab results in a suction peak of about Cp ~
-0.4 in the figure. The pressure is then relatively constant and subsequently
decreases in response to the flow acceleration due to the streamline curvature
100 S. Maddox et al.
into the wake. The pressure coefficient for the GTS at 10° yaw is lower because
of the higher speed flow over the top of the GTS, associated with the side-
force and circulation generated by the geometry at sideslip. In general, both
the DES and RANS predict reasonably accurately the acceleration of the flow
over the front surface of the GTS. For 0° yaw the DES prediction of the slight
decrease in Cp near the rear surface in Figure 9 is more accurate in the DES
than the RANS. At 10° yaw, Figure 10 shows that both the DES and RANS
predictions of the pressure coefficient exhibit a much earlier and more gradual
reduction, over the second half of the model, compared to the experimental
measurements of Storms et al. [3]. Based on Figure 4, this probably results
from a discrepancy in the exact size and location of the roof vortex.
1
0.5
\r:;~
u~
~
I
I --DES
/ - - - - RANS
·0.5
11 D measurements
·1w_~~~~~~~~4~~~~~~~
x/w
Fig. 11. Pressure distribution along the side of the GTS, (3 = 0°.
Pressures along the side of the GTS are shown for both yaw angles in
Figure 11 and Figure 12. The distributions were measured at yjw = 0.696
(midway between the lower GTS surface and roof) and in the experimental
configuration on the left side of the configuration, corresponding to the wind-
ward side for the computations at positive 10° yaw. The pressure distribution
for 0° yaw in Figure 11 shows that at xjw = 0 a pressure near stagnation is
recovered, as expected. The pressure decreases very significantly due to the
acceleration around the front corner and is then relatively constant and near
freestream level to the end of the trailer. The acceleration due to the converg-
ing streamlines into the wake is captured in both the DES and RANS and in
good agreement with the measurements. Though not obvious from Figure 11,
the DES prediction is in slightly better agreement with measurements near
the end of the trailer.
For 10° yaw the pressure distributions along the side of the GTS are shown
for both the leeward and windward side along with the measurements of Cp
DES of the Ground Transportation System 101
-0.5
-1w_~~~~~~~~4~~~~~~~
x/w
Fig. 12. Pressure distribution along the side of the GTS, (3 = 10°.
which were acquired only along the leeward side of the GTS. Figure 12 shows
on the windward side of the GTS analogous features to those observed in Fig-
ure 11, i.e., a sharp reduction in the pressure as the flow turns the front corner
of the model with a roughly constant distribution to the end of the trailer and
then slight reduction due to the flow acceleration into the wake. As evident in
the figure, the RANS and DES descriptions of the pressure distribution along
the windward side are very similar. The leeward side pressure distribution
shows the largest differences near the front corner. A much stronger suction
is predicted by both simulation techniques, an effect that arises due to the
substantially larger region of flow separation predicted than measured (c.f.,
Figure 5. The discrepancy in the pressure distribution in this region in turn
leads to larger discrepancies in the predicted forces at 10° yaw compared to
the results at 0° yaw.
A plot of the skin-friction magnitude along the centerline of the top of the
GTS is shown in Figure 13 and Figure 14 for the GTS at 0 and 10 degrees
yaw, respectively. Also included in each figure is the theoretical curve for skin
friction on a turbulent flat plate, which may be expected to apply starting
near xjw = 2, in view of the pressure distribution (note also that the vir-
tual origin of the flat plate was simply set to xjw = 0). In the experiments
reported by Storms et al. [3] the boundary layers were not tripped, and in-
deed the measurements in Figure 13 show that the boundary layer is initially
laminar before the skin friction rapidly increases to values characteristic of
turbulent boundary layers following transition. In the simulations (both DES
and RANS) the eddy viscosity at the inlet to the computational domain is
non-zero (prescribed as three times the molecular value), an effect that primes
the turbulence model as the fluid enters the boundary layers. Consequently,
c 1 is larger than the experimental measurements near the front of the GTS.
Figure 13 shows that the experimental measurements are slightly above the
102 S. Maddox et al.
'\1 measurements
() RANS
D DES
- - - - correlation
)(
c.)
0 o~~~~~~~~~4~~~~~6~~~~
x/w
"V measurements
D DES
<> RANS
- - - - conelatlon
>< 4 \ 0
c.)
... ___
' , Oo Oo Ao <)[J
... ... v ;:: "V
---- ----- -----·<>
0 o~~~~~~~~~4~~~~~~~~
x/w
fiat-plate correlation, the simulations slightly below; these differences are con-
sistent with the different transition histories.
The effect of yaw is apparent in the skin friction along the GTS centerline
shown in Figure 14. For reference, the same curve for a fiat-plate turbulent
boundary layer as shown in Figure 13 is also shown in Figure 14. In the vicinity
xjw = 0, the DES and RANS predictions are similar to the distributions for
0° yaw and also in good agreement with measurements (the experimental skin
friction is here at a turbulent level, presumably as a result of transition due to
three-dimensionality in the boundary layer). At approximately the mid-region
of the GTS (slightly upstream of xjw ~ 4), the influence of the roof vortex is
apparent via the increase in CJ as shown in the figure. DES and RANS predic-
DES of the Ground Transportation System 103
tions are again similar and also in good agreement with measurements. Near
x/w = 7.2 the experimental measurements show a relatively sharp increase in
Cf that is not reflected in the calculations.
4 Summary
The work reported in this manuscript represents the findings from an initial set
of investigations aimed at gaining insight into the application of DES to vehicle
aerodynamics. The principal comparisons were to experimental measurements
of Storms et al. [3] as well as assessment of DES predictions against RANS
results obtained for the steady-state flow. Overall, the result are encouraging
in that for mean quantities DES is superior to RANS, e.g., predictions of the
drag were in substantially better agreement with measured values at 0° yaw.
At 10° yaw the differences in measured and predicted forces were more
substantial. The computations were of the fully-turbulent flow, i.e., without
the use of the trip terms in the S-A model to dictate the location of laminar-
to-turbulent transition. Some of the discrepancy in the force predictions at 10°
yaw may be connected to the laminar or turbulent state of the boundary lay-
ers. Relevant in this regard is the front corner region in which the separation
is more substantial in the simulations than indicated by measurements. The
corner radius of the GTS challenges the RANS region and transition predic-
tion - the complexity is also manifest in the hysteresis of the drag coefficient
with Reynolds number observed in the experiments. Tests with less-rounded
geometries would enable some resolution of these issues, more clear-cut from
the standpoint of flow physics but possibly more challenging of computational
aspects.
In addition to these aspects, there remain issues for which further inves-
tigations are also motivated. The role of the grid was not assessed. However,
grid generation was guided by experience in previous DES applications to
configurations of similar complexity, and a grid count near 6 million is very
defendable for a shape that is only of moderate complexity. A follow-on study
may revolve around a grid with 3 million points and another with 8.5, for
instance, providing a factor of J2 in each direction. Coupled with an inves-
tigation of the grid would be related efforts to explore the timestep. Within
the context of the present approach using the S-A model, prescribing the lo-
cation of transition is possible and would enable interesting investigations of,
for example, tripping the top surface boundary layers (which are attached)
while employing the tripless approach of Travin et al. [11] for the sides of the
GTS (which have separation). As also summarized above, aspects related to
geometrical features such as the influence of the corner radius of the model
should be investigated. Finally, the ability of boattail plates to reduce drag is
well documented by Storms et al. [3] and any future numerical efforts should
develop a simulation plan for investigation of such effects.
104 S. Maddox et al.
5 Acknowledgments
The authors are grateful for the helpful suggestions provided throughout the
project by Dr. Philippe Spalart. Dr. Bruce Storms kindly provided the ex-
perimental measurements in electronic format. Discussions of the GTS grids
with Dr. Chris Roy are also gratefully acknowledged.
References
1. McCallen R, Couch R, Hsu J, Leonard A, Brady M, Browand F, Hammache M,
Salari K, Rutledge W, Ross J, Storms B, Reineck JT, Driver D, Bell J, Zilliac G
(1999) Progress in Reducing Aerodynamic Drag for Higher Efficiency of Heavy
Duty Trucks (Class 7-8), 1999-01-2238, Gov/lndustry Meeting, Washington,
DC.
2. Spalart PR, Jou WH, Strelets M, Allmaras SR (1997) Comments on the fea-
sibility of LES for wings, and on a hybrid RANS/LES approach. 1st AFOSR
Int. Conf. on DNS/LES, Aug. 4-8, 1997, Ruston, LA. In Advances in DNS/LES,
C. Liu and Z. Liu Eds., Greyden Press, Columbus, OH, USA.
3. Storms BL, Ross JC, Reineck, JT, Walker SM, Driver DM, Zilliac GG (2001)
An experimental study of the Ground Transportation System (GTS) model in
the NASA Ames 7- by 10-ft wind tunnel. NASA/TM-2001-209621.
4. Spalart PR, Allmaras SR (1994) A One-Equation Turbulence Model for Aero-
dynamic Flows. La Recherche Aerospatiale, 1, pp. 5-21.
5. Strelets M (2001) Detached Eddy Simulation of Massively Separated Flows
AIAA 01-0879.
6. Shur ML, Spalart PR, Strelets MK, Travin AK (1999) Detached-Eddy Simula-
tion of an Airfoil at High Angle of Attack. Fourth International Symposium on
Engineering Turbulence Modelling and Measurements, Corsica, France.
7. Strang WZ, Tomaro RF, Grismer MJ (1999) The Defining Methods of Cobalt 6 o:
a Parallel, Implicit, Unstructured Euler/Navier-Stokes Flow Solver. AIAA 99-
0786.
8. Gottlieb JJ, Groth CPT (1988) Assessment of Riemann Solvers for Unsteady
One-Dimensional lnviscid Flows of Perfect Gases Journal of Computational
Physics, 78, pp. 437-458.
9. Karypis G, Schloegel K, Kumar V (1997) Par METIS: Parallel Graph Partition-
ing and Sparse Matrix Ordering Library Version 1.0. University of Minnesota,
Department of Computer Science, Minneapolis, MN 55455.
10. Steinbrenner J Wyman N, Chawner J (2000) Development and Implementation
of Gridgen's Hyperbolic PDE and Extrusion Methods. AIAA 00-0679.
11. Travin A, Shur M, Strelets M, Spalart PR (2000) Detached-Eddy Simulations
past a Circular Cylinder. Flow, Turb. Comb. 63, pp. 293-313.
CFD: Software Methods and Applications
Time Dependent vs. Steady State Calculations of
External Aerodynamics
Abstract
The paper assesses the use of the most popular standard k-E model and the most
accurate Reynolds-stress model for various simple and complex flows including
vehicles, and their impact on steady and transient RANS calculations. At the same
time, the error of the steady state approach for the flows where transient effects are
important, is analyzed. The Hybrid Turbulence Model is examined as an alternative
solution, e.g. the recent proposal of combining the Boussinesq's concept with the
second moment closure. Finally, new possibilities in the further use of steady and
averaged transient results will be addressed e.g. acoustic calculations. The paper
compiles previous and present work performed at AVL List GmbH using the in-
house commercial CFD software AVL Swift.
Introduction
Most of reported calculations of external aerodynamics have been done with Rey-
nolds-Averaged-Navier-Stokes (RANS) methods. The basis behind these methods
is decomposition of the instantaneous flow variable into a mean and a fluctuating
(random) part and then ensemble averaging of the Navier-Stokes equations. The re-
sulting new (unknown) terms, namely Reynolds stresses, require turbulence mod-
eling. This is very discouraging for some CFD users as turbulence models, despite
the recent progress, are considered to be the largest source of error in present cal-
culations. While it is not possible to eliminate this error, it can be decreased consid-
erably by the knowledgeable use of turbulence models.
A frequent 'simplification' in the use of RANS is to make steady state calcula-
tions. However, to insist on the steady state solution because it is inexpensive, is a
very risky practice which could lead to serious errors in particular cases. The first
alternative is to use the Transient Reynolds-Averaged Navier-Stokes (TRANS) ap-
proach. This is particularly important in flows which are dominated by low-
frequency periodic features such as encountered in a vortex shedding.
108 B. Basara and P. Tibaut
Cd Cl' Str
Transient RSM 2.28 1.39 0.141
Transient k-£ 1.80 - 0.119
Steady RSM 1.89 - -
Measurements 2.16-2.28 1.1-1.4 0.130-0.139
0 03 0.04
t(s)
The phase averaged data was calculated as suggested by Rodi and Ferziger 1995
for the Workshop on LES of Flows Past Bluff Bodies. The time period between
two successive maximums is divided into 20 equal intervals and the predicted
streamlines are compared with the measurements for the same phases. Neverthe-
less, Fig. 2. shows a good agreement between predictions and measurements for
the phase 01, for more details see original reference Basara 2004.
The next step was to use the RSM for the steady state calculations enforced by
employment of the symmetry plane. Predicted streamlines are shown in Fig. 3. The
separation process around the cylinder is described as one very large vortex starting
at the first front corner and finishing ten lengths behind the obstacle. The pre-
dicted drag coefficient is 15% lower than obtained by measurements showing that
this procedure is inappropriate, although the full Reynolds-stress model was used,
see Table 1.
Time Dependent vs. Steady State Calculations of External Aerodynamics 109
••
Fig. 2. Phase averaged streamlines (Phase 01): measurements (left) and predictions (right)
with the RSM (Basara 2004)
Fig. 3. Streamlines predicted by the RSM and for the steady state flow.
However, it should be noted that there is not much benefit in using transient
RANS if the turbulence model chosen for the calculations poorly reproduces tran-
sient effects e.g. the standard k-E model.
The next adequate test for the turbulence models prior to their use for the vehi-
cle aerodynamics is the vortex shedding around a cylinder placed at various dis-
tances from an adjacent wall. When the gap between the obstacle and the wall was
reduced to 0.5 of the obstacle width, the vortex shedding was predicted with the
Reynolds-stress model as reported by the measurements, quite contrary to the
standard k-E model which predicted steady flow, see Bosch and Rodi 1995, Basara
et al. 1996 etc.
These simple benchmarks were used to assess the basic differences of steady
and transient calculations using the standard k-E model and the RSM . The other
models e.g. non-linear k-E models or algebraic stress models were not tested as-
suming that the RSM represents the best turbulence modeling approach. In the last
few years, attention has been turned more to the numerical aspects of the RSM's
employment in order to make this model numerically robust for real-life vehicle
simulations. Consequently, the new hybrid modeling approach was developed pro-
viding additional options for more accurate prediction of turbulence, but at the
same time having a numerically robust model.
The basis of the modeling approach, the solution method and the results are
given in the following sections.
110 B. Basara and P. Tibaut
Fig. 4. Ford Ka. Numerical grid: upper part (left) and underbody (right).
The ACT in AVL Swift v3.1 (Swift Manual 2002) is based on a fully conserva-
tive finite volume approach. The cell-face based connectivity and interpolation
practices for gradients and cell-face values are introduced to accommodate an arbi-
trary number of cell faces. All dependent variables, such as momentum, pressure,
density, turbulence kinetic energy, dissipation rate, and passive scalar are evaluated
at the cell center. A second-order midpoint rule is used for integral approximation
and a second order linear approximation for any value at the cell-face. A diffusion
term is incorporated into the surface integral source after employing the special in-
terpolation practice. The convection is solved by a variety of differencing schemes
(upwind, central differencing, MINMOD, and SMART). The rate of change is dis-
cretized by using implicit schemes, namely Euler implicit scheme and three time
level implicit scheme of second order accuracy. The overall solution procedure is
iterative and is based on the Semi-Implicit Method for Pressure-Linked Equations
algorithm (SIMPLE). For the solution of a linear system of equations, a conjugate
gradient type of solver and algebraic multigrid are used. Special attention is given
to the implementation procedure of the RSM on arbitrary unstructured grids, see
Basara 2004.
The flow field is modeled by the ensemblemean Navier Stokes equations cou-
pled either with the eddy-viscosity k-£ model equations or with the differential
Reynolds stress model equations. In the k-£ model of turbulence, the Reynolds
stresses are obtained from the Boussinesq's eddy viscosity formulation:
Time Dependent vs. Steady State Calculations of External Aerodynamics 111
(I)
s
where k is the turbulent kinetic energy, 0 is the mean rate ofstrain tensor and
l't is the turbulent viscosity, which is evaluated from the expression:
l't
j3/2) ; "o =
= pC,. La "o ; La = (£ tU2 (2)
where Lg and v0 are characteristic length and veloc:ity seales of turbulence, re-
speetively. In fUll Reynolds stress models, the differential transport equations for
the Reynolds stresses u1uJ are solved. When closing these equations, the diffusion
is modelled by a simple gradient transport hypothesis. For the pressure-strain
term, the SSG model of Speziale, Sarkar and Gatski 1991 is used. This model does
not require the use of wall-damping terms.
As Eq. (2) suggests, the k - e models use the turbulence kinetic energy and
it•s dissipation rate to define characteristic turbulence scales. The C/J coefficient
is derived from the measured ratio u1uJ I k for the wall boundary layers and then
used in Eq. (2) as a constant value. The commonly used value is 0.09. An ap-
proach advanced by Basara and Jakirlic 2003 suggests a derivation of CP by
equalizing the production of turbulence kinetic energy predicted by the Reynolds
stress model and with the production obtained by the k-£ model, thus
(3)
Therefore. the hybrid model (HTM) solves the momentum equations in co-qjunc-
tion with Eq. (2) and (3). In addition, a set of full Reynolds stress equations is
solved to provide the Reynolds stresses u1uJ and turbulent kinetic energy
k = 0.5 u1u1 • The dissipation rate equation is also solved in a term commonly
used in the framework of the Reynolds stress model.
ln conjunction with selected "'high Reynolds nwnber'' turbulence closures, the
wall functions are used to model the near-wall effec:ts.
separation region compared well with the measurements. Trailing vortices appeared
at the C-pillars pulling the attached flow coming from the roof down to the base,
were always better predicted by the RSM. With the k-e model, a starting poposition
of these side vortices at the C-pillars is lower than the measured one. In the case of
RSM model, a description of the flow pattern over the slant is very close to the
measured one. In addition, smaller vortices which might appear on the side edge
due to interactions between two streams could be captured as well, see Basara et al.
2002 (see Fig. 5). As the separation point is crucial for accurate predictions of the
pressure distribution on a car body, the average drag and lift coefficients predicted
by the RSM model were closer to the measured values than those obtained by the
k-e model. The correctness of the flow pattern predicted by the RSM model is
definitely confirmed by distribution of the pressure coefficient on the upper side of
the car (see Fig. 5). The k-e model predictions of pressure distribution miss meas-
urements considerably. The flow pattern was also calculated correctly for Morel
body by Basara and Alajbegovic 1998. Very good agreement was obtained, espe-
cially in the case of transient (unsteady) simulations. Steady state calculations even
with the RSM model were not able to predict sudden pressure drop caused by
changing the back slant angle. On the other hand, the transient calculations showed
presence of vortex shedding and averaged drag and lift coefficients were close to
measurements.
••
. .
.! ,--, .
0.6 0.!! , ,. . . . .
-o•
·~
-I
-1.5'-------------__j
Fig. 5. Velocity vectors at the trailing edges and the distribution of the pressure
coefficient as predicted by RSM (Basara et al. 2002)
Clearly after all these calculations of the flow around simple obstacles and ide-
alized vehicles, the procedure for the reliable calculations could be established. It
consists of the following steps:
-to provide steady state results with the k-e model.
-to continue with the steady state calculations by using the RSM model ap-
proximately making the same number of iterations as performed with the k-
€ model.
-to continue with the transient calculations by using the RSM.
Therefore, after calculating SAE notchback reference body (Basara et al. 2002),
Ahmed Body (Basara and Jakirlic 2003) or Morel Body (Basara and Alajbegovic
1998) etc., the same procedure was applied for more complex models. The next ex-
ample was the first including the wheels. Prior to this simulation, separate calcula-
tions of the flow around a rotating wheel were performed (Basara, Beader and
Przulj 2000), see Fig. 6. Again good performances of the RSM were reported re-
Time Dependent vs. Steady State Calculations of External Aerodynamics 113
garding the pressure distribution as well as the prediction of the separation point,
for more details see the original reference.
c. 0.4
-o.o
I -0.4
I
-0.8
- 1.2
0 60 120 180 240 300 360
8
Fig. 6. Numerical grid (left) and predicted and measured mean pressure coefficient around
the wheel in the mid plane (right). Basara et al. 2000.
Calculating a complete vehicle with the wheels (Renault model), once more, a
difference between models performances becomes apparent. Clearly the flow pre-
dicted by the k-£ model shows less tendency to separate either on wheels or on the
slanted back comparing to the RSM results as shown in Fig. 7 and 8.
Fig. 7. Renault model. Velocity vectors around the wheel as calculated by the k-£
(left) and the RSM (right).
Transient results were obtained by the RSM (see Fig. 8 right) and the averaged
drag and lift coefficients are in good agreement with the measurements. The RSM
calculates the time-mean drag coefficient of 0.2 (-0.2% error). That compares well
with the measurements and quite contrary to the k-£ result of 0.223 (+9.8% error).
R ....
024 ~
~~.
i 02 ·v
_.) ~
.,.
I[
Fig. 8. Renault model. Velocity vectors on the slanted rear as predicted by the RSM and the
k-E model (left) and the calculated drag coefficient by the RSM (right).
114 B. Basara and P. Tibaut
The next step was to calculate a vehicle with the detailed underbody (Ford Ka),
see Fig. 4. Numerically, this was a challenging task for the calculations with the
full Reynolds-stress model. However, it was possible to perform such numerical
study as shown in Fig. 9. It should be noted that the computing time is 4-5 times
larger than when using the k-E model. Predicted pressure distribution in the wake
behind the vehicle is compared with the measurements in Fig. 10. The best fitting
shape is achieved with the RSM. The pressure coefficient distribution along the
upper central plane of the body predicted by the k-E and RSM model is shown in
Fig. II. The averaged drag and lift coefficients are given in Table 2. In general, the
results obtained by the RSM are very satisfactory.
Fig. 9. Ford Ka. Surface streamlines (left) and the velocity magnitude projected on
the surface (right).
Table 2. Ford Ka. Predictions and measurements of drag Cd and lift Cl coefficients.
Ford Ka Cd Cl
Data 0.321 0.076
k-E 0.355 0.226
RSM 0.319 0.043
n
- . . , [ .,.,.._~-- ..... ""'
l ,.
' ~ ~ ~~~·
. '
I i ' 1
..... '
', ~ ...._...
0 .0 0.) J.()
Fig. 10. Ford Ka. Predicted pressure coefficient in the wake behind the vehicle.
Time Dependent vs. Steady State Calculations of External Aerodynamics 115
- measured p ca l ul t~~e d Cp (k e)
.-l< ul .. ed Cp (RS t.l)
~ .~~-+~~~--~----------~
~
' '
Fig. 11. Ford Ka. Distribution ofthe pressure coefficient on the upper side.
Volvo VRAK shown in Fig. 12, was calculated by using the k-E model and AVL
HTM model. The obtained results highlight the importance of using a more accu-
rate model to calculate the turbulence. The HTM approach was previously tested
on many simple examples including all cases shown in the introduction (Basara and
Jakirlic 2003). It provides regularly more robust and faster solution than the RSM
and more accurate results than the k-e model. In this case, transient solution was
obtained by the HTM, see Fig. 12 right (see also Basara eta!. 2001). The averaged
drag and lift coefficients are given in Table 3. The results obtained by the HTM
agree much better with the measurements than those obtained by the k-E model.
The HTM requires half less computing time than the RSM. Note that the HTM
model can be also applied as an 'initialization' model used between the k-E and
RSM models in order to stabilize and shorten RSM calculations.
Table 3. Volvo VRAK. Predictions and measurements of drag Cd and lift Cl coefficients.
Cd Cl
Data 0.359 0.336
k-E 0.368 0.466
HTM 0.351 0.365
'"
Fig. 12. Volvo VRAK. Velocity magnitude projected on the surface (left) and the averaged
transient drag coefficient (right).
One can argue that different turbulence models may give a similar drag or lift
coefficient for the certain shapes of the car and hence, those models which need
less computing time would be preferable for such calculations. But, this may hap-
116 B. Basara and P. Tibaut
pen only in isolated cases. For example, the flow pattern can be very similar due to
the fact that the separation points are fixed by the shape of the vehicle. However,
even in such cases, the intensity and the distribution of the calculated turbulence
kinetic energy will be different and therefore, the input to some other calculations
modules could be wrong e.g. acoustic predictions, soiling etc. Fig. 13 shows sim-
plified bus and the calculated turbulence kinetic energy. Based on these results, cal-
culations of flow-induced noise are performed jointly by AVL-TNO showing an
importance of proper predictions of the turbulence kinetic energy. The method de-
veloped by TNO (e.g. Snelen et al. 2002, BJorn et al. 2001) consists of the un-
steady acoustic source generation based on the RANS data and the calculation of
the propagation of the pressure fluctuations using the linearized Euler equations.
The results are available on request.
Fig. 13. Simplified bus. Predicted turbulence kinetic energy by the k-E model (left) and the
RSM (right).
Conclusion
Calculations performed for the simple and complex real-life benchmarks show that
there is consistency in the results obtained when using different turbulence models.
Compared to the standard k-E model, the present Reynolds-stress transport equation
model continuously produces results which are in better agreement with the meas-
urements. The hybrid turbulence model is a good compromise in modeling re-
garding the accuracy and the computing time. It is necessary to perform transient
RANS calculations in order to get accurate results for the external aerodynamics.
References
A VL AST (2002) Swift Manual 3.1 . A VL List GmbH. Graz.
LynDA (1992) Ensemble-Averaged Measurements in the Turbulent Near Wake of a Square
Cylinder: A Guide to the Data, Report CE-HSE-92-6, Sc, Purdue University.
FrankeR, Rodi W (1991) 8th Syrup. On Turbulent Shear Flows. Munich .
Rodi W, Ferziger JH (1995) Proc. Of Workshop on Large Eddy Simulation of Flows past
Bluff Bodies, Germany.
Basara B (2004) Employment of the second-moment closure on arbitrary unstructured grids.
Int. J. for Numerical Methods in Fluids 44: 377-407.
Bosch, G. Rodi W (I 995). IO'h Syrup. On Turbulent Shear Flows, Pennsylvania.
Time Dependent vs. Steady State Calculations of External Aerodynamics 117
Basara B., Bachler G, Schiffermuller H. (1996) Calculation of vortex shedding from bluff
bodies with the Reynolds-stress model. In: Kutler P, Flores J and Chattot J (eds) 15th
International Conference on Numerical Methods in Fluid Dynamics, Lecture Notes in
Physics. Springer.
Speziale CG, Sarkar S, Gatski TB (1991) Modelling the pressure-strain correlation of turbu-
lence: an invariant dynamical system approach. J Fluid Mech 227: 245-272.
Basara B, Jakirlic S (2003) A New Turbulence Modelling Strategy For Industrial CFD. Int.
J. for Numerical Methods in Fluids 42: 89-116.
Basara B, Decan B, Przulj V (2000) Numerical Simulation of the Air Flow around a Rotat-
ing Wheel. The 3'd Mira International Vehicle Aerodynamics Conference, UK.
Basara B, Alajbegovic A.(l998) Steady state calculations of turbulent flow around Morel
Body, The 7th Int. Symp. on Flow Modelling and Turbulence Measurements. Taiwan.
Basara B, Przulj V and Tibaut P (2002) On the calculation of external aerodynamics: Indus-
trial Benchmarks. SAE Transactions - Journal of Passenger Cars and Mechanical Sys-
tems V110-6.
Basara B, Jakirlic S, Przulj V (2001) Vortex-shedding flows computed using a new, hybrid
turbulence model. In: Ninokata H, Wada A. and Tanaka N. (eds), Advances in Fluid
Modeling & Turbulence Measurements, World Scientific.
Snellen M, Lier L, Rops C, Janssens M, Heck J, Strumolo GS (2002) Flow-induced noise
around the A-pillar of an idealized car greenhous. AIAA paper 2002-2548.
BJorn CPA, Verhaer BT, Heijden JC, Soemarwoto BI (2001) A linearized Euler method
based prediction of turbulence induced noise using time-averaged flow properties.
AIAA paper 2001-1100.
Aerodynamics of Ground Vehicles - Toward
Reliable and Affordable CFD
Sung-Eun Kim
Alec Wong
Hudong Chen
David Gosman
The turbulent flow structure in the near wake of a pickup truck model has
been investigated experimentally using Proper Orthogonal Decomposition
(POD) analysis of Particle Image Velocimetry (PIV) data. The experiments
were conducted in the 2'x2' wind tunnel at the University of Michigan at
Reynolds numbers based on model width of 3x105. A model of a pickup truck
with extended cab, 432 mm long by 156 mm wide by 149 mm tall, was used.
PIV measurements of the velocity field in several planes of the wake including
the symmetry plane were obtained using a large sample size, sufficient to
determine the mean flow and the Reynolds stresses in the wake. The mean
flow structure in the symmetry plane consists of separated shear layers
originating at the edge of the cab and from the underbody flow. For this
particular geometry, there is a recirculating flow region behind the cab ending
upstream of the tailgate; but there is no mean recirculating flow region behind
the tailgate. The mean flow pattern in a horizontal plane behind the tailgate
suggests that this is due to trailing streamwise vorticity. POD analysis of the
data provides useful information on the unsteady large scale structures in the
wake. It is shown that only a few orthogonal modes (~20) contain a large
fraction of the fluctuation energy (~60%), as expected. But, more important,
the analysis isolates features of the unsteady large scale turbulent structures
into different uncorrelated modes. It is shown that the development of vortex
shedding in the underbody flow shear layer is described by a few modes, while
oscillations of the recirculating region are captured by other modes. This
feature of the analysis makes it very useful to the study of the structure and
dynamics of complex bluff body wakes.
Applications of DDPIV to Studies Associated with
Road Vehicles
Introduction
Aperture 1 Lens
- B
a
z
B" - B' L A
B
/') b
z
B'~
lb
B" L -
B' , '
Aperture I Lens
Image Image plane Reference
(CCD sensor) plane
Fig. 1. The defocusing concept: a) standard, one aperture system, b) defocusing, multiple aper-
ture system.
Applications ofDDPIV to Studies Associated with Road Vehicles 133
Geometric Analysis
L
P'(x' ,y")
"(Y".f")
Reference Sensor
plene plare
sensor plane, i.e. at Z = -l. The coordinates of a pixel on the imaging sensor are
given by the pair (x,y). Point P(X Y,Z) represents a light scattering source (a
particle). For Z"'L, Pis projected onto points P'(x;y') and P"(x';y''), separated
by the distance b.
The coordinates (x;y') and (x';y'') of the images P' and P" of P(X Y,Z) in the
:l :;~d~L~LL
image plane are given by the following relations:
2LY] (1)
b =Md(L- Z)
z (2)
1 1 1 1
=-(-- -) with K= MdL
K Z L
The sensitivity of the system, i.e. its ability to detect small changes of the
particle location, can be evaluated through the separation gradient
ab 1
(3)
aZ ==- Kz2
The coordinates of P in the world coordinate system are derived from the
image coordinates of the projections P' and P" (see Eq. 1)
-Y"=--
X'Oz with
x' + x''
xo = ----=--
AfL 2
y =- 'JioZ with 1Jo=
11 + y''
AfL 2 (4)
1
z
~+Kb
Assuming that the apertures are equidistant from the origin of the coordi-
nate system, the image point defined by (x0,yo) is the image of the particle if
there were a single aperture at the origin.
A camera system has been designed and fabricated based upon the above
concept. The specific characteristics of this instrument can be found in the pa-
per by Pereira et al. 2000.
The velocity vector field is obtained by local spatial cross-correlation be-
tween small volume elements (voxels, equivalent to windows in DPIV) con-
taining particles observed at two time steps, as shown and discussed by Pereira
et al. 2000.
Applications
The nature of the DDPIV system makes it very versatile. Its ability to resolve
complex flows and perform particle sizing in 3 dimensions makes it a useful
tool in many aspects of the design of road vehicles.
Exterior flow measurements require, in most cases, that the flow be seeded
with particles. For use in heavy fluids, such as water, there is a large selection of
appropriate solid particles. For lighter fluids, such as air, one is not limited to
solid particles. The whole system can be used in a laboratory environment (e.g.
wind tunnels, water tunnels, tow tanks) and, with relative ease, can be taken to
the field (full-scale on-road tests). A system can be designed to measure a range
of different probe volumes that requires no tedious alignment. Moreover, it is
compact and can be transported in one unit which allows for fast set-up times.
Spray-type flow measurements can be performed, in some cases, more easily
than exterior flows, since the spray itself will form the particles - there is no
need to seed the spray to perform measurements. This provides great advan-
tages, since in most cases the part or component's performance can be visual-
136 M. Gharib, F. Pereira, and E.C. Graff
ized and quantized with the actual fluid that would flow through it. For in-
stance, experiments can be performed on injector-cylinder combinations, per-
forming velocimetry and particle sizing on the actual fuel that would be used
in a final production model.
Fig. 4. A truck model in the water tunnel with the probe volume illuminated.
Other spray applications include water spray and splash from the vehicle in
wet-road conditions. Detailed information on how a vehicle spray patterns
from driving through large water puddles or evenly wet roads is fundamental
in the development of road-friendly vehicles and tire designs that do not im-
pose an unsafe, low-visibility environment on other vehicles and pedestrians.
Internal flow measurements are almost no different from those for exterior
flows, except for the main requirement that the probe volume must be opti-
cally accessible for both illumination and imaging. The DDPIV system has
great potential in vehicular components for cooling, intake, exhaust, and other
components where flow characteristics contribute in large part to the perform-
ance and efficiency of the vehicle.
To demonstrate the DDPIV system's ability to evaluate complex flows (with
both PIV and particle sizing functionality), we consider two cases of a model
propeller immersed in a water tank, one propeller having 2 blades, the other
having 3 blades.
The two-blade model propeller is immersed into a water tank. The rotation
speed is 12 rps, corresponding to a tangential velocity of 2.52 m.s·1 at the tip of
the blades. A bubble generator is placed below the propeller, producing a dense
Applications ofDDPIV to Studies Associated with Road Vehicles 137
24&&53
21 .3303
17.7752
J\:l:lOJ
A 3-D velocity field is obtained after averaging and outlier correction. Outlier
correction is a smoothing algorithm directly evolved from DPIV, whereby con-
straints can be set so that the magnitude of a particular vector is not drastically
different from that of its nearby neighbors.
By computationally injecting "mass-less" particles into the mean velocity
data set in a radial arrangement and one diameter upstream the propeller,
pathlines can be generated, providing a unique insight into this complex flow
as shown in Fig. 5. Gray level relates to the local measured velocity amplitude.
Velocity reaches a maximum of 2.49 m.s- 1 in the outer region of the propeller,
matching closely the blade tip tangential velocity.
138 M. Gharib, F. Pereira, and E.C. Graff
100 3000
..
75 Propeller at :c2500
Y=30mm
50 ~2000
E' E Y=30mm
.§.25 ~ 1500
:g
Q)
0 1000 Pop,!Pop,. 65%
:::l
vf,NH=96%
-25 co 500
-50 0 •• _........,!!'£
'"~;;~·--
150 200 250 300 350 0 250 500 750 1000
Mean Bubble Radius [J.tm] Bubble Radius [l!m]
Fig. 6. Bubble size vs. longitudinal distance (left) and bubble population vs. size (right).
The (real) bubble mean radius along the Y vertical axis of the flow (rotation
axis of the propeller) is reported in Fig. 6 (left). The mean radius increases al-
most linearly to almost 325 flm at Y=30 mm, where the propeller is located.
After the bubbles pass the immediate vicinity of the propeller, the radius is
found to follow the opposite trend, decreasing to about 200 flm. The growth
of bubbles is due to a very small extent to the decrease of the static pressure
with increasing Y. In fact, bubbles experience first the low pressure in the suc-
tion side of the propeller before reaching the high-pressure region where they
collapse.
Included in Fig. 6 (right) are the histograms calculated taking the same vol-
ume below and above Y=30 mm. The histogram peak follows the trend out-
lined before due to the pressure variations. The ratio of the upstream to the
downstream populations is 65%. However, the ratio of the respective void
fractions is close to 100%. These observations indicate that coalescence of
bubbles is the main mechanism acting here, although breakup may occur in
the immediate vicinity of the propeller.
A three-blade boat propeller is the second case. A similar configuration is used
here. The propeller is rotated at 12 rps. The velocity field represented in Fig. 7
is obtained by phase-averaging a sequence of 50 instantaneous velocity fields.
Spurious vectors can be seen on borders of the interrogation domain. A slice in
the velocity field, as shown in Fig. 8, clearly displays the high speed jet core
along the downstream section of the propeller axis. However, the isovelocity
contours displayed in Fig. 9 show a viscous wake that appears as a velocity de-
fect due to the merging of the two boundary layers from the blades. A slight
contraction of the slipstream could also be detected. The wake is found to
rapidly fade into the bulk flow.
Applications ofDDPIV to Studies Associated with Road Vehicles 139
Acknowledgements
References
Pereira, F., Gharib, M., Modarress, M., and Dabiri, D, 2000, Defocusing
DPIV: A 3-component 3-D DPIV Measurement Technique, Application
to Bubbly Flows, Experiments in Fluids, Vol. 29.
Singh, A., Optic flow computation, IEEE, Computer Society Press, 1991.
Westerweel, J., 1993, Digital Image Velocimetry: Theory and Application, Delft
UP, Delft, Netherlands.
Willert C. E. and Gharib, M., 1992, Three-dimensional particle imaging with
a single camera, Experiments in Fluids, Vol. 12, pp. 353-358.
Willert, C., and Gharib. M., Digital Particle Image Velocimetry, Experiments
in Fluids, Vol. 10, pp. 181-183, 1991.
Molecular Tagging Velocimetry (MTV) and Its
Automotive Applications
Abstract.
This work provides an overview of the technique of Molecular Tagging Ve-
locimetry (MTV) and some of its automotive applications. The various ele-
ments of MTV implementation are briefly described in terms of the available
molecular tracers, methods of tagging, detection, and processing schemes.
The automotive applications of this velocimetry technique are demonstrated in
mapping the velocity field of the intake flow into a "steady flow rig" model of
an internal combustion engine and flow mapping of cycle-to-cycle variation in
late compression of a motored IC engine.
1. Introduction
Molecular Tagging Velocimetry (MTV) is a whole field optical technique
which relies on molecules that can be turned into long lifetime tracers upon
excitation by photons of an appropriate wavelength. Typically a pulsed laser is
used to "tag" the regions of interest, and those tagged regions are interrogated
at two successive times within the lifetime of the tracer. The measured Lagran-
gian displacement vector provides the estimate of the velocity vector. This
technique can be thought of as essentially a molecular counterpart of Particle
Image Velocimetry (PIV), and it offers advantages compared to particle-based
techniques where the use of seed particles is not desirable, difficult, or may lead
to complications. Figure 1 illustrates one implementation of the technique
where a planar grid of intersecting laser beams, formed from a pulsed excimer
laser at 308 nm, is used to turn on the luminescence of water-soluble su-
pramolecules that are premixed in a water flow of a vortex ring approaching a
solid wall at normal incidence (Gendrich et al. 1997). The displacement of
the tagged regions is determined using a direct spatial correlation method.
The conventional planar imaging shown in Figure 1 provides information on
144 M.M. Koochesfahani, A.C. Goh, and H .J. Schock
two components of the velocity vector, the projection onto the viewed plane.
Stereo imaging can produce the complete three components of the velocity
vector (Bohl et a!. 200 I).
L L
i
I I
Fig. l. Typical MTV image pairs and the resultant velocity field (Gendrich et al. 1997). The
flow shown is from a vortex ring impacting on a flat wall at normal incidence. The axis of
symmetry is indicated by the dashed lines. (a) The grid imaged I r..ts after the laser pulse. (b)
The same grid imaged 8 ms later. (c) The velocity field derived from (a) and (b).
The MTV technique has advanced significantly over the past decade in
terms of the availability of new molecular tracers, methods of tagging, detec-
tion/imaging, and data processing. This paper provides only a brief descrip-
tion of such details. More complete reviews, along with more extensive list of
related references and applications, can be found in Koochesfahani et al.
(1996) and Koochesfahani (1999, 2000).
3 mm=0.025 C
Fig. 2. Multi-line tagging in the trailing edge region of an oscillating NACA-0012 airfoil. The
tagging pattern right after the laser pulse is a series of parallel lines with a spacing of 3 mm or
about 0.025 chord length.
The MTV image pairs are acquired by a pair of CCD detectors that view
the same region of interest in the flow through a beam splitter. Using a refer-
146 M.M. Koochesfahani, A.C. Goh, and H.J. Schock
ence target, the two cameras are aligned to within one pixel, and the remaining
residual displacement field between the two detectors is quantified to sub-pixel
accuracy and accounted for in subsequent data processing {for details see Gen-
drich et al. 1997). Immediately after the pulsed laser fires, the first detector re-
cords an initial image of the tagged regions. After a prescribed time delay, the
second detector records a second image of the tagged regions displaced by the
flow. As described in Gendrich et al. {1997), and Koochesfahani (1999), such
a two-image system offers advantages over the typical single-image system in
that no assumption needs to be made a priori about the intensity field in a
tagged region. For example, one can properly take into account the variations
in the initial tagging pattern {e.g. due to laser beam pointing instability, vibra-
tion of the optics, non-uniform tracer concentration, etc.), which could other-
wise be misinterpreted as flow velocity fluctuations. In situations where the
initial tagging pattern remains spatially invariant throughout the experiment, a
single detector is sufficient; the initial {or reference) tagging pattern is recorded
once, usually at the beginning of the experiment, and then the "delayed" im-
ages are acquired. In our work in liquid-phase flows, we use both non-
intensified frame transfer cameras and gated image-intensified detectors, de-
pending on the imaging requirements {e.g. field of view, time between image
pairs). In gas-phase applications, we use gated image-intensified detectors. See
Koochesfahani {1999) for several examples of applications in liquid and gas
phase flows. These detectors are nominally 512 x 512 pixel arrays operating at
30 or 60 frame/s. The images are typically digitized to 8 bits by two image ac-
quisition/ processing systems and transferred onto high capacity disk arrays in
real time.
A common method for finding the displacement of tagged lines or grids has
been to locate the center of each line through various techniques. Most of the
recent techniques use the best fit to an assumed laser line shape, for example, a
gaussian intensity distribution. A recent study {Hill & Klewicki 1996) gives
the accuracy in determining the displacement vector to be ±0.35 pixel rms.
The performance of this method will suffer when the intensity distribution of
the tagged regions cannot be assumed in advance, for example, due to non-
uniform tracer distribution, difficulties associated with laser beam transmission
through a flowing medium, bleaching effects, etc.
Our approach for finding the displacement of tagged regions is based on a
direct digital spatial correlation technique, and offers certain advantages over
the traditional line-center methods. In particular, it is a more general scheme
that is independent of the specific intensity distribution within a tagged region
and can accommodate arbitraty tagging patterns including those due to non-
uniform scalar mixing fields. The details of this approach and its performance
are described in Gendrich & Koochesfahani (1996). A small window, referred
to as the source window, is selected from a tagged region in the earlier image,
and it is spatially correlated with a larger roam window in the second image. A
well-defined correlation peak occurs at the location corresponding to the dis-
placement of the tagged region by the flow; the displacement peak is located to
sub-pixel accuracy using a multi-dimensional polynomial fit. Based on both
experiments and an extensive statistical study, it has been found that the dis-
placement of the tagged regions can be typically determined with a 95% con-
Molecular Tagging Velocimetry (MTV) and Its Automotive Applications 147
fidence limit of ±0.1 sub-pixel accuracy {i.e. 95o/o of the displacement meas-
urements are accurate to better than 0.1 pixel). This corresponds to an rms ac-
curacy of ±0.05 pixel, assuming a Gaussian distribution for error. For high
values of image SIN, the 95o/o confidence level can be as low as 0.015 pixel
(0.0075 pixel rms). An example of the application of this procedure is pro-
vided in Figure 1; the velocity vectors shown in this figure are "raw" and have
not been filtered or smoothed.
tions in the shear layer at the interface between the intake jet and adjacent
fluid can be as high as 5000 g, making it difficult to rely on the results of parti-
cle-based techniques. The details of this work can be found in Stier &
Koochesfahani (1999).
Figure 3 shows a 3 em x 3 em field of view in the nirrogen/biacetyl flow
being investigated and the regions tagged by a grid pattern. Part of the valve
body and the left wall of the cylinder are visible in the picture. The maximum
flow speed in the annular jet entering the cylinder is about 10 m/s. Also
shown is an example of the later image of the tagged regions after a 50 ~s de-
lay. For this time delay, the maximum displacement of tagged regions is about
8 pixels (- 500 ~m). Image pairs such as those in Figure 3 are used to deter-
mine the instantaneous radial and axial velocity components in this flow field.
An example of the instantaneous velocity field and the structure of the intake
flow in this geometry are shown in Figure 4 along with the average velocity
field based on 320 realizations. The instantaneous flow map shows a highly
unsteady intake annular jet, which as has an undulating appearance with oppo-
site sign large scale vortical structures on its two sides. The mean flow map in-
dicates a large scale region of recirculation in the upper left corner of the en-
gine cylinder, a feature typical of an IC engine flow field. These data have also
been used to derive other properties of the flow such as the instantaneous and
average vorticity fields and velocity fluctuations (Stier & Koochesfahani 1999).
Fig. 3 . Example of the tagging pattern for the gas flow into the steady flow rig (Stier &
Koochesfahani 1999).
Molecular Tagging Velocimetry (MTV) and Its Automotive Applications 149
va lve
:;;::::::::::: : ~;;;
, , I "' ....... - ... ' \ .. ~ •• I , I
cylinder ~~~~:::~::: ;: ;;;
I I I , ~ •• I I I ' ' I/
wall ~ I , I ••• • I j I I II
l rt •.• , , , , 11 /
I' • . • • • • •
/, •..•.• •
' '"· --,,~/
f l t ... .,.,l'//
/, ..... .. .
/ ;· · · ······
r' ' . .. ,.,// /"' .. - . , .•... ,
:~: ::;:,~/
. -- ..., . , / / / / / / "
·- ,~////// ,,_
/~~ :: :: :: ~:: ::
. - .. ' . '
.......... .
.... .
0 ' / / / / / 1 ' ;"' • , , o • •
I I I I 0 0 I
1 / / / / / / / .1'"'" • • • • • ' o o I o t
I I I I/
\I I I I
I'"'' ' •-' ' ' ' ' ' ' ' •'
I , , .. - • . • • • • • • ' •
I I I I , " • - ..•••••••
Fig. 4. Intake flow into a steady flow rig model of IC engine measured with MTV using nitro-
gen seeded with biaceryl (Stier & Koochesfahani 1999).
Fig. 5. The optically accessible research engine and a typical MTV grid tagging the fluid within
the engine cylinder.
Molecular Tagging Velocimetry (MTV) and Irs Automotive Applications 151
Spark Plug
Imaged plane at late
~
compression , 270 CAD
-1
e
~
-2
>.
-3
piston
-4
-5
-6
0 2 4 5
X (em)
Two samples of the instantaneous velocity field, for two different engine cy-
cles, are shown in Figure 7. It is clear the flow pattern is significantly different
between these two realizations, an indication of the large cycle-to-cycle vari-
ability of the flow. The maps of the ensemble-averaged velocity field and the
rms fluctuation of the horizontal velocity component are computed from 500
such realizations, see Figure 8. As expected for a flow with large cycle-to-cycle
variation, the mean velocity field has little resemblance to the instantaneous
field; the local fluctuation level can be higher than the mean by several hun-
dred percent. The data shown in Figures 7 and 8 are the first cycle-resolved
velocity measurements that use molecular tracers rather than particles to de-
termine flow velocities in a piston-engine assembly. Such measurements en-
abled by the MTV imaging technique are being used to advance the under-
standing and quantification of cycle-to-cycle variability in an internal
combustion engine, and finding methods for reducing it (Goh 2001).
! 52 M.M. Koochesfahani, A.C. Goh, and H.J. Schock
. : • " ., "
-
I I
-1
·1 5
/ I I
I
I
I
I
I
I
.
/.
'
·1
' ' / // / .
. .. ..
I / ·1.5
,• _... _.
.'
/ I /
-2 / '
i i
·2 I / / ' I
I I
I
. • . ..
/
·2 5
•I /
• ••.
j -2.5
,.. -3 .,
I I /
.
~
' '
-35
I
' ' -3.5 I ," _.
. _. ..
._ '
"
' '
6
-
5 mls
u' l
170
160
150
140
130
120
110
100
90
80
70
60
50
Fig. 8. Maps of the ensemble-averaged velocity fi eld and the rms flu ctuation of the horizo ntal
compo nent of velocity u' (in cm/s). Colo rs denote rms fl uctuation level.
Molecular Tagging Velocimetry (MTV) and Its Automotive Applications 15 3
4. Conclusions
A brief overview of the Molecular Tagging Velocimetry (MTV) technique is
given, along with two of its automotive applications. These applications in-
clude mapping the velocity field of the intake flow into a "steady flow rig"
model of an internal combustion engine and flow mapping of cycle-to-cycle
variation in late compression of a motored IC engine. The MTV approach
has continuously evolved and is now being utilized as a tool in both funda-
mental flow studies and applied engineering measurements.
Acknowledgments
This work was supported by the MRSEC Program of the National Science
Foundation, Award Numbers DMR-9400417 and DMR-9809688, Depart-
ment of Energy Grant No. DE-FC02-99EE50574, and Ford Motor Com-
pany.
References
Bohl, D., and Koochesfahani, M., and Olson, B. (2001) Development of
stereoscopic Molecular Tagging Velocimetry," Exp. Fluids, vol. 30, 302-
308.
Bohl, D. G. (2002) Experimental Study of the 2-D and 3-D Structure of a
Concentrated Line Vortex Array, PhD thesis, Michigan State University.
Cohn, R. K. (1999) Effect of forcing on the vorticity field in a confined
wake, PhD thesis, Michigan State University.
Falco, R. E. and Chu, C. C. (1987) Measurement of two-dimensional fluid
dynamic quantities using a photochromic grid tracing technique, SPIE,
vol. 814,706-710.
Gendrich, C. P. and Koochesfahani, M. M. (1996) A spatial correlation
technique for estimating velocity fields using Molecular Tagging Veloci-
metry (MTV), Exp. Fluids, vol. 22, no. 1, 67-77.
Gendrich, C. P., Koochesfahani, M. M. and Nocera, D. G. (1997) Molecular
tagging velocimetry and other novel applications of a new phosphorescent
supramolecule, Exp. Fluids, vol. 23, no. 5, 361-372.
Gendrich, C. P. (1998) Dynamic stall of rapidly pitching airfoils: MTV ex-
periments and Navier-Stokes simulations, PhD thesis, Michigan State
University.
Goh, A. C. H. (2001) Active flow control for maximizing performance of
spark-ignited stratified charge engines, MS thesis, Michigan State Univer-
sity.
Hascher, H. G., Novak, M., Lee, K., Schock, H., Rezaei, H., and Koochesfa-
hani, M. M. (1998) An evaluation of IC-engine flow with the use of
modern in-cylinder measuring techniques, AIAA Paper 98-3455.
154 M.M. Koochesfahani, A.C. Goh, and H.J. Schock
Hill, R. B. and Klewicki, J. C. (1996) Data reduction methods for flow tag-
ging velocity measurements, Exp. Fluids, vol. 20, no. 3, 142-152.
Hilbert, H. S. and Falco, R. E. (1991) Measurements of flows during scav-
enging in a two-stroke engine, SAE Technical Paper 910671.
Hu, H. and Koochesfahani, M. M. (2003) A novel technique for quantitative
temperature mapping in liquid by measuring the lifetime of laser induced
phosphorescence, journal ofVisualization, vol. 6, no. 2, 143-153.
Koochesfahani, M. M., Cohn, R. K., Gendrich, C. P. and Nocera, D. G.
(1996) Molecular tagging diagnostics for the study of kinematics and
mixing in liquid phase flows, Proceedings of the Eighth International Sym-
posium on Applications of Laser Techniques to Fluids Mechanics, July 8 - 11,
1996, Lisbon, Portugal, vol. I, 1.2.1-1.2.12; Also in Developments in La-
ser Techniques and Fluid Mechanics, Chapter 2, section 1, p. 125, Eds.
Adrian, Durao, Durst, Maeda, and Whitelaw; Springer-Verlag, Berlin,
1997.
Koochesfahani, M. M. (1999) Molecular Tagging Velocimetry (MTV): Pro-
gress and Applications, AIAA Paper 99-3786.
Koochesfahani, M. M. (Ed) (2000) Special Feature: Molecular Tagging Ve-
locimetry, Meas. Sci. Techno!., vol. 11, no. 9, 1235-1300.
Koochesfahani, M., Cohn, R, and MacKinnon, C. (2000) Simultaneous
whole-field measurements of velocity and concentration fields using com-
bined MTV and LIF, Meas. Sci. Techno!., vol. 11, no. 9, 1289- 1300.
Koochesfahani, M. M. and Bohl, D. (2002) Molecular tagging visualization
and velocimetry of the flow at the trailing edge of an oscillating airfoil,"
Proceedings of the I O'h International Symposium on Flow Visualization,
Kyoto, Japan, Paper No. F0453.
Kruger, S. and Grunefeld, G. (1999) Stereoscopic flow-tagging velocimetry,
Appl. Phys. B, vol. 69, 509-512.
Lempert, W. R., Magee, K., Rooney, P., Gee, K. R., and Haughland, R. P.
(1995) Flow tagging velocimetry in incompressible flow using photo-
activated nonintrusive tracking of molecular motion (PHANTOMM),
Exp. Fluids, vol. 18, 249-257.
Lempert, W. R., Jiang, N., Sethuram, S., and Samimy, M. (2001) Molecular
tagging velocimetry measurements in supersonic micro nozzles, AIM Pa-
per 2001-0244.
Lum, C, Koochesfahani, M. M., and McGrath, J. J. (2001) Measurements of
the velocity field with MTV during the solidification of an alloy analog
with mushy region, ASME/IMECE2001 Paper HTD-24222.
Miles, R., Cohen, C., Conners, J., Howard, P., Huang, S., Markovitz, E., and
Russell, G. (1987) Velocity measurements by vibrational tagging and
fluorescent probing of oxygen, Optics Letters, vol. 12, no. 11, 861-863.
Miles, R. B., Connors, J. J., Markovitz, E. C., Howard, P. J., and Roth, G. J.
(1989) Instantaneous profiles and turbulence statistics of supersonic free
shear layers by Raman Excitation plus Laser-Induced Electronic Fluores-
cence (RELIEF) velocity tagging of oxygen, Exp. Fluids, vol. 8, 17-24.
Pitz, R. W., Brown, T. M., Nandula, S. P., Skaggs, P. A., DeBarber, P. A.,
Brown, M. S., and Segall, J. (1996) Unseeded velocity measurement by
ozone tagging velocimetry, Optics Letters, vol. 21, no. 10, 755-757.
Molecular Tagging Velocimetry (MTV) and Its Automotive Applications 15 5
Introduction
Despite the increased use of numerical simulations in the development and
performance optimization of aerodynamic vehicles wind tunnel tests still are of
fundamental importance in the related engineering design process. In order to
keep the cost down and to increase the data return from the expensive meas-
urement campaigns, there is an interest in improving and expanding the diag-
nostic tools available.
Recent developments in modern measurement technology, especially in the
areas of digital imaging and photonics, have led to the introduction of a num-
ber of computer-based, quantitative flow visualization tools. While these tech-
niques have rapidly gained acceptance in laboratory research, their introduc-
tion into the domain of commercial testing has been slower. This may be not
only due to the cost involved but also for a number of technical reasons.
The reliability, precision and flexibility of the new methods still have to be
improved to convert them into turn-key tools. Furthermore, there are some
problems associated with the scale-up from bench-top hardware to systems op-
erating in considerably larger and less controlled environments.
The paper will address some of those scaling issues. The specific require-
ments for large scale testing will be analyzed and a number of recent flow visu-
alization and measurement techniques are described which should be well
suited for such applications.
where for example particle image velocimetry {PIV) has developed almost into
standard method within the last ten years {Kompenhans 2000).
In large scale facilities such as wind tunnels, there are a number of generic
problems which have delayed a similar progress. Although these problems are
at times simple and of a purely technical nature, they still can affect the func-
tionality and performance limits of various imaging techniques at a very fun-
damentallevel.
Most camera-based measurement methods require some form of active scene
illumination. Dependent on the capabilities of the cameras (high speed or slow
scan, direct integration or intensified, large or small dynamic range etc.) this
lighting may have to be pulsed or continuous, wide band (white) or spectrally
narrow-band and often of a considerable intensity. Incandescent lamps, lasers,
high-powered flash lamps or other, more specialized sources can usually be
found to meet these requirements. However, when arbitrary aerodynamic
models are to be inserted into the field of view, spurious surface reflections
may arise which cannot easily be controlled or eliminated. Since it is often the
flow close to the model surface which is of particular interest, the cameras' pix-
els can become saturated and the information is lost where it counts most.
Finding a locally optimized arrangement of illumination and observation an-
gles may temporarily solve the problem but it will arise immediately again if
the configuration is changed.
Another fairly common requirement in air flow diagnostics is the generation
of flow seeding. Usually micron-sized particles or droplets have to be chosen so
that the tracers can follow the flow with a sufficient fidelity. Since the back-
scatter signatures tend to be small, fairly high tracer densities have to be gener-
ated which may affect the overall visibility of the model and the optical access
in general, defeating in part the original purpose. More importantly, in many
imaging techniques the individual tracers have to be spatially resolved by the
recording device in order to extract the desired information. This puts con-
straints on the cameras' resolution and optical magnification which may in
turn severely limit the size of the field of view. In addition, there are issues re-
garding the chosen tracers' lifetime {too short, too long) and the environ-
mental and health impact of the sometimes corrosive or poisonous substances
involved.
For imaging techniques that require either long time integration periods or
comparisons with reference images, another problem may arise from a possible
lack of configurational stability. Model movements on the supports, facility vi-
bration and other uncontrolled effects may affect the precision with which the
position of the measurement points on the model surface can be determined.
In techniques such as pressure sensitive paint monitoring, where a pixel-by-
pixel referencing is necessary, even very small displacements can create signifi-
cant changes in the ratio images.
Finally, there are simple operational and financial constraints that have to be
accommodated. While the actual data taking with an imaging setup may be
done in a few seconds, the calibration and adaptation of that setup can be an
hour-long activity. This may lead to the situation where a specific measure-
ment technique is ruled out purely because of its operational complexity.
Quantitative Flow Visualization for Large Scale Wind Tunnels 159
Taking into account these technical constraints, there remain a few basic
choices which have to be made in the design and selection of a large scale,
quantitative flow visualization system.
In the laboratory, a certain preference has developed for imaging configura-
tions which exploit the Eulerian view of the flow such as in PIV. This may be
due to technical reasons (e.g. the need for strong illumination, available only in
light sheets) but is also related to the interest in spatial flow structures and gra-
dient properties. As a consequence, the flows under investigation are often
"tailored" to accommodate an essentially planar image acquisition philosophy.
Furthermore, the amount of raw data produced and the associated transfer and
storage requirements for a full 3-dimensional measurement grid would become
prohibitively large.
In flows around complex model surfaces, three-dimensionality is almost al-
ways present and a volumetric view of the measurement domain is necessary.
In order to keep the data complexity at a reasonable level, it may be more ad-
vantageous to rely on Lagrangian techniques, which track localized flow features
in time rather than providing spatially resolved snapshots.
A similar duality is present when one looks at the data processing philoso-
phies. The established approach is based on a time-global approach where all
data acquired during a measurement run are stored and utilized in a post-
processing step to extract the desired information. The need for fast, near real-
time data visualization and efficient data storage, however, does favor incre-
mental strategies where the incoming data are processed "on the fly" with only
a limited knowledge of the temporal flow evolution. Especially for Lagrangian
techniques (e.g. particle tracking methods), the choice of such an incremental
approach does have a significant impact on the overall processing scheme.
Finally, a balance has to be found between the requirements for fast data
analysis I presentation and the achievable measurement accuracy. For visuali-
zation-oriented schemes the temporal coherence of the data may be more im-
portant than the ultimate precision of each individual data point reading, cre-
ating a requirement for efficient error rejection, but not necessarily for error
correction. In other situations, the priorities are reversed in that accuracy is
more important than speed and an increased temporal effort (including aver-
aging multiple measurements or recursive processing) is considered acceptable.
Looking at these considerations, one may state that large scale diagnostic
applications create their own set of requirements which may be different from
those applicable to laboratory-style research. The measurement techniques
which are to be applied have to be adapted especially regarding their opera-
tional characteristics. At times, even the development of new methods de-
signed specifically for large scale environments may become necessary.
160 R. Bommels et al.
This facility is also used as a test bed for a number of advanced quantitative
imaging techniques which are being developed with true large scale applica-
tions in mind.
Pressure sensitive paints are being studied in cooperation with RUAG Aero-
space (CH) . Being a technique of obvious potential for wind tunnel applica-
tions, the research concentrates specifically on ways to eliminate the inherent
temperature sensitivity of the fluorescent paints used and the wind-on I wind-
off calibration procedure (see also Engler 2000).
A pulsed infrared thermography system has been installed to provide a tool
for the rapid visualization of laminar-turbulent transition and separation lines
on model surfaces. It utilizes a high power stroboscope (7500 W, approx. 7 J I
flash) in conjunction with a midwave (3-5mm) infrared camera (20 mK
NETD) to detect the subtle changes in surface heat transfer associated with
those flow phenomena. The technical challenge lies in the development of im-
aging strategies which are independent of the model's surface structure and
composition which can strongly affect the infrared signature (Le Sant 2002).
In the area of flow velocimetry, a Doppler global velocimeter is being devel-
oped based on a custom narrow band pulsed laser with two independent os-
cillators. This permits the simultaneous use as a PIV system for comparison
and the measurement of 3-dimensional velocity vectors with a single camera I
single view point arrangement.
At the same time a particle tracking velocimeter is being used to provide a
Lagrangian view of the flows under investigation. A pair of high speed cameras
tracks the motion of helium-filled soap bubbles which are used as low-inertia
Quantitative Flow Visualization for Large Scale Wind Tunnels 161
The price one has to pay for this improvement is twofold. First, the lasers
required to generate the light sheet have to be very stable and must emit a very
narrow-band radiation with an optical line width of, say, below 20 MHz. Such
lasers are available as CW systems, but the illumination in large facilities de-
mands higher intensities only available from pulsed systems (e.g. Nd:YAG la-
sers). Here a so-called injection seeder has to be used. A low power, highly sta-
ble ring laser injects seed radiation into the cavity of a Q-switched oscillator
which leads to a longitudinal mode selection and stabilization. Since the opti-
162 R. Bommels et al.
'. l .. h ... _,
J:ft:,)•/1
l'~l
The system operated at IFD is more advanced in that it operates two seeded
lasers in parallel to provide a double pulse capability as well. This facilitates the
use of the lasers in a PIV mode for comparison. As a matter of fact, if the scale-
up I resolution issue is not critical, a simultaneous DGV I PIV measurement is
possible. In such an arrangement, the out-of-plane (DGV) and in-plane (PIV)
velocity components in the light sheet can be measured with the same camera
system, creating a true 3-component velocimeter. (Normally, different velocity
components are measured with separate camera systems in an all-DGV con-
figuration) .
Besides the advantages cited so far, DGV as a large scale imaging technique
is also affected by some inherent limitations. First of all, the intensity ratioing
approach for the computation of velocities makes DGV into an essentially
analogue technique. The intensity reading of a single camera pixel is affected
Quantitative Flow Visualization for Large Scale Wind Tunnels 163
Fig. 4: DGV images or a rotating disc target: filtered (left), unfiltered (center), processed (right)
• I
Frame Grabber I, <!
--- ---
_______Wind Twmel
ate a "cloud" of reference points with a known relative distance from which
the unknown parameters can be deduced (Borghese 2000).
A typical result of a reconstructed path line field is shown in Fig. 7 . The
vortex flow behind a delta wing is clearly visible in this 30 rendering of the re-
constructed bubble tracks. The color coding of the individual tracks is an indi-
cation of the relative velocity of the helium bubbles forming the path lines.
Fig.~, llo" bc:hind .1 ddt.l wing vi,u.llitnl u'ing .)!) rc:\:Oil\ln!Unlp.uh line:'
The challenges in this type of processing are once again primarily the light-
ing and choice of seeding. In addition, fast processing hardware is required but
the rapid increase in computing power means that dedicated hardware such as
pipelined image processors can successively be replaced by general purpose
CPUs.
Fig. 8: Real-time processing of smoke trails into a cumulative streak line picture
Summary
Quantitative flow visualization based on the processing of remotely acquired
image data can significantly enhance the understanding and analysis of engi-
neering flows. Modern techniques such as Doppler global velocimetry, pres-
sure sensitive (fluorescent) paints or infrared thermography make use of ad-
vanced lasers and image detectors to provide the optical raw data that are
subsequently transformed into the desired flow information (velocity, tem-
perature, pressure, etc.). While the different principles of operation have been
successfully verified, there are still a number of generic technical issues which
have to be resolved before one can expect a routine use in commercial facilities.
Examples were given for a DGV system, a 30 particle tracking velocimeter
and a simple yet powerful real-time visualization tool. All systems are designed
for operation irrespective of the large scales involved (several meters field of
view). The control of the illumination and the seeding density remain as a
problem which is directly related to the geometrical size and complexity of the
measurement environment. The scaling issues will remain one of specific
problems in large scale flow diagnostics which are not found on the laboratory
scale and must thus be addressed separately.
References
Borghese NA; Cerveri P; (2000) Calibrating a video camera pair with a rigid
bar, Pattern Recognition 33: 81-95
Engler RH; Klein C; Trinks 0 (2000) Pressure sensitive paint systems for
pressure distribution measurements in wind tunnels and turbomachines,
Meas. Sci. Techno!. 11: 1077-1085
Quantitative Flow Visualization for Large Scale Wind Tunnels 167
Dale Satran
Abstract
Particle Image Velocimetry (PIV) measurements were acquired in the wake of the
trailer and in the gap between the tractor and trailer of the Generic Conventional
Model (GCM) truck for the US Department of Energy. The data will be used both
for validation of computational fluid dynamics (CFD) codes and for understanding
the flow physics. The GCM is a 1!8th-scale, moderate-fidelity model of a full-scale
truck. The test was performed in the Army/NASA 7x10 wind tunnel at NASA
Ames Research Center. Surface pressure and force measurements were made prior
to the PIV measurements. PIV measurements were made at two yaw angles and at
three horizontal planes for three model configurations, each at a free-stream veloc-
ity of 52 m/s (Mach 0.15), which corresponds to a Reynolds number of 1 x 106 ,
based on the width of the tractor. This paper discusses the PIV system, samples of
flow data and some of the observed features that may have contributed to the meas-
ured drag.
Introduction
A multi-agency effort led by the US Department of Energy (DOE) is underway to
help truck and trailer manufacturers increase fuel efficiency of heavy vehicles. An
important part of the program is to reduce aerodynamic drag. This part of the effort
includes the development of computational fluid dynamics codes and the perform-
ance of experiments in both wind tunnels and full-scale road tests. The GCM is a
moderate fidelity 1!8th-scale model of a modem engine-forward tractor and trailer.
The level of fidelity is high enough to represent the flow of the modem full-scale
truck, though not so detailed as to make CFD gridding prohibitive. Figure 1 is a
photograph of the model installed in the test section.
174 J.T. Heineck, S.M. Walker, and D. Satran
The PIV measurements were made on the basic model configuration (Fig. I),
then the basic model tractor with side-extensions that spanned 50% of the gap
width (Fig. 2A), and then basic model with trailer boat-tail plates (Fig. 2B). For the
basic configuration, the gap was 5 inches, The trailer back face was flat and no
skirts covered the wheels along the trailer. The balance measurements from the
aerodynamics testing phase of these experiments showed these devices reduced
drag.
A B
Fig.2. Photographs of the glass side extenders (A) and the trailer boat-tail (B).
The PIV test matrix included combinations of the parameters. There were two
flow areas of interest; the wake and the gap. Two model configurations were tested
for each area; the basic and the basic with drag reduction treatment. Then two yaw
angles for each location; 0 degree and I 0 degrees. The data set for each flow area
consisted of three vertical locations of the horizontal data planes: I ,4, 112 and 3/4 of
the trailer height of each area of interest was measured with and without the drag
reduction devices at two yaw angles. Extra data sets were acquired in response to
preliminary analysis of the drag data. A feature in the balance data showed that
drag suddenly decreases when the model was at approximately II degrees. Also, a
The Measurement of Wake and Gap Flows of the Generic Conventional Truck Mode 175
weak vortex generated off the top of the trailer was observed when the model was
yawed at angles greater than 5 deg. An extra data set was acquired with the plane
oriented in a vertical cross-stream manner, 9 inches aft of the trailer. All conditions
were for M=O.I5 (52 rn!s). The Reynolds number for this condition was I x 106 •
,' I
I , I
\ , I
View 2
View 1 \ ',
,
,/'
I
/
2
• •
3 \1 ,,,'' 1/ 2
•
3
\ ,' I
, I
I A I
\ , I
\ ,' I
\ , I
I , I 3mm 1 mm
\ , I
\ , I
'' 0 0
•
Particle Pairs 1 2 Time 1
T 11&
Fig.3. Illustration of the camera and laser plane relationship for three-component PIV.
In this illustration, three flow conditions are represented as particle pair exam-
ples. Pair I indicates a positive flow direction with a significant in-plane displace-
ment. The two camera views render that motion quite differently, which demon-
strates the perspective difference that is exploited to calculate the third component.
Pair two represents cross-plane flow. Pair 3 represents reverse (relative to Pair 1)
flow.
The geometry of the model dictated the placement of the cameras. The gap flow
could only be viewed from above and to the side to permit off-normal imaging of
the laser plane. The plane for the wake was oriented horizontally in order to capture
176 J.T. Heineck, S.M. Walker, and D. Satran
the wake convergence. Thus, the PIV system design parameters followed these
constraints.
The PIV system consisted of two Kodak ES 4.0 cameras (2048x2048 pixels),
two New Wave Gemini PIV laser systems, IDT Provision software for calibration,
acquisition and data reduction, hardware for synchronization signal timing and re-
mote focusing, sheet forming optics and a micro-particle generator. Two lasers
were used so that both cameras received forward-scattered illumination of the seed
particles. The forward scatter position receives more than 20 times the light than
the back scatter position. The cameras were unable to render the particles in back-
scatter position; therefore the second laser was necessary. The light sheet from each
laser was matched for output and formed with identical optics.
Figure 4 is an illustration of the system position relative to the model. This fig-
ure shows how the two laser sheets were aligned to each other: the sheets were
made level, then the support frames were moved vertically to place the planes at the
proper height.
• •
1::0 In
Referring to Figure 3 once again, the laser sheet thickness is shown as 3 mm.
The highest probability for image-to-image correlation occurs when the the highest
number of particle pairs from image to image remain within the interrogation area.
The particles moving through the laser plane must remain in the laser plane for this
to occur. The optimal cross-plane displacement should be 1/4 - 113 of the sheet
thickness (Willert, 1997). Therefore, the maximum cross-plane displacement of the
particle motion was limited to lmm. The actual thickness of the sheet projected for
this experiment was 3mm as well. The magnitude of the measurable in-plane parti-
cle displacements is limited by the cross-plane component. When examining the
gap data, even the worst case permits sufficient particle displacement to allow for
an accurate measurement. In that example, a 1 mm cross plane displacement is re-
corded as a displacement of 5 pixels. The same case for the wake measurement
would get recorded by slightly less than 2 pixels, due to the lower magnification;
therefore the local measurement uncertainty increases. Purely cross-plane flow oc-
curred in a small region in the wake where the wake converged. The rest of the
flow was primarily in the in-plane direction. The measurement uncertainty was ap-
proximately 2% for the in-plane components and 4% for the cross-plane component
for any instantaneous measurement.
t
)100
3000
E
1900
Flow.
E
x I 0
2Tn0
~f>n
800 600
z, mm V. m s
10 00
Average wake ortruckwith boattaii, 112 height, 0 dog 800
3100
6 00
3100 •oo
J 00
3000 000
E .00
E 1900
x· '00
11100 • 00
8 00
1700 ·10 00
Fig. 7. Streamline plots of the mean wake flows at 0 deg of the GCM without and with the
boattail. The color contour renders the vertical component, V.
t
Flow.
600
V. m s
0 00
Average of truck wake with boattall. 112 height. 10 dog
8 00
; 00
• 00
• 00
0 00
·2 00
• 00
• 00
·8 DO
·10 00
600
Fig . 8. Streamline plots of the mean wake flows at 10 deg yaw without and with the
boattail plates. Color contour renders the vertical component, V.
180 J.T. Heineck, S.M. Walker, and D. Satran
The gap flow data show similar improvements. Fig.9 contrasts the flows without
and with side extenders for the 0 deg case. Fig. I 0 are plots of the I0 deg case. Note
in both cases, the vortex strength and the V component of the flow reduces dra-
matically with the use of side extenders.
Figure II is the plot of the drag vs. yaw angle from the aerodynamics data. PlY
data were collected at II degrees, where this sudden drop in drag occurs. Figure 9
shows the flow patterns of the high drag state at 10 degrees and the low drag state
at 11 deg.
E
750
700 Flow,
t
E
X-
650
600
·100 ·100 0 100 200
z, mm v . m/s
Average of gap now with 50% side extenders. 112 height. o degyaw 25 00
10 00
15 00
750
1000
5 00
0 00
E 700
· 500
E
X- ·1000
650 ·1500
· 20 00
·25 00
600
·100 ·100 100 100
z . mm
Fig. 9, Streamline plots of 0 deg mean gap flows without and with side extenders. Color
contour plots the vertical component, V.
The Measurement of Wake and Gap Flows of the Generic Conventional Truck Model 181
Tl3ller Front
750
E 700
E
x
Flo". U
650
600
·200 0 100 200
z . mm V, m/s
Average gap flow with 50 % side extenders . 112 height 1Odeg.vaw 25 00
20 00
Tnollor Front 15 00
750
10 00
5 00
E 700 0 00
E ·5 00
x ·10 00
650 ·15 00
·20 00
·25 00
600
·200 ·100 I 00 200
z mm
Fig. 10, Streamline plots of 10 deg mean gap flows without and with side extenders .
Color contour shows vertical component, V in the same scale as Fig. 7.
0.8
0.75
__..,_ _Yaw
0.7
0.65
Cl ------ +Yaw
(.) 0.6
0.55 ____.,__ 2.5" Side
0.5 Exta1ders
0.45
0.4
-16 -12 -8 -4 0 4 8 12 16
Beta
Fig.ll, Drag polar for basic model and basic with 50% side extenders. (Satran, 2003)
182 J.T. Heineck, S.M. Walker, and D. Satran
11 deg Yaw, low-drag state
WrntM'I
2000
2000
1000
1000
500
000
~co
· 1000
· 1500
·2000
·2500
Flow, U
10 deg Yaw, high-drag state
w ""
2500
2000
1000
1000
500
000
-500
· 1000
- 1500
2000
-2500
Fig. 12. Streamline plots of mean gap flow at 1/4 -height at II deg (low drag) and 10 deg
(high drag) with vertical component V in color contour.
In Fig.l3, the two surface pressure plots from Satran, (2003) were manipulated
to offer a perspective view of the gap's pressure distribution. The streamline plot
from Fig.8 of the gap flow of the basic tractor configuration was scaled and placed
in the appropriate position, at I/2 height. This composite shows the viewer there-
lationship between the recirculation (or vortex) region to the low pressure region on
the back of the tractor.
This same style of composite is presented for the tractor side-extension case.
With both the velocity contours and pressure contours are plotted using the same
scale. It becomes clear that the reduced gap flow has reduced the pressure differen-
tial between the tractor back and the trailer front. The drag polar plot in Figure II
demonstrates the difference in measured drag with and without extension. But the
pressure reduction and flow field change may be a result of some other flow
mechanism and not a cause of the drag hysteresis. Another mechanism may cause
the change in circulation, where the change in the flow through the gap is the result
of a change in the condition along the trailer that induces axial drag. These discus-
sions are left to future articles.
The Measurement of Wake and Gap Flows of the Generic Conventional Truck Model 183
-25 -10 0 10 25
Vmean. m/s
Fig. 13. Composite illustrations of the surface pressures in the gap (without and
with side extenders) and a flow field plot at 1/2 height. Surface pressure contours in
Cp and flow field contour of Vmean, in m/s.
184 J.T. Heineck, S.M. Walker, and D. Satran
Conclusion
Flow measurements were made around a tractor-trailer model called the Generic
Conventional Model (GCM). Three-component PIV measurements of the wake and
gap flows clearly render the changes in flow patterns caused by aerodynamic drag
reduction devices. Furthermore, a vortex structure in the gap was shown to cause a
hysteresis in the drag polar measured by Satran in the first phase in these experi-
ments. The goal of creating a database for the truck engineering and research com-
munity was advanced with the addition of these measurements.
References
Willert, C., "Stereoscopic digital particle image velocimetry for application in wind
tunnel flows" 1997, Measurement Science and Technology 8, pp. 1465-1479
On the Aerodynamics ofTractor-Trailers
Abstract
Introduction
At typical cruising highway speeds, most of the fuel consumed by a large scale
road vehicle is expended to simply overcome aerodynamic drag, even in the
absence of unfavorable wind conditions. The remaining fuel expenditure is
needed to overcome the rolling resistance on tires and internal losses. The fuel
efficiency of tractor-trailers, which account for the greatest portion of heavy
vehicle traffic by a large margin, is influenced by the shape and physical
dimensions of these vehicles. Shape and scale are themselves severely
constrained by economic considerations (the parallelepiped shape of trailers is
meant to optimize volume loading) and by regulatory constraints (truck
dimensions are fixed and drag-reduction devices cannot exceed specified
limits).
A look at the evolution of tractor-trailer design over the last several
decades reveals gradual aerodynamic improvements to the front of vehicles,
namely from the front of the tractor to the front of the trailer, but very little
has been done to improve the back of trailers. Tractors have benefited from
tremendous improvements, such as the adoption of the aero-shield as an
integrated part of the cab. The problem of the gap berween tractor and trailer,
which is an important source of drag, has been partly mitigated by the
introduction of cab extenders that effectively reduce the size of the gap but do
not eliminate it entirely.
In contrast, trailers have seen little modification, aside from the rounding
of their vertical leading edges in the front. The fact that hard-shell trailers are
designed to be loaded from the back makes the implementation of drag
reduction devices in the back particularly challenging. Truck operators are
extremely reluctant to deal with any type of physical device that may interfere
with routine loading and unloading operations.
The main objective of this paper is to investigate the impact of tractor-
trailer gap width on the drag forces experienced by a truck. The study relies on
detailed measurements performed on truck models in a wind tunnel. These
models are not replicas of actual trucks that can be seen on the roads, but
rather truck-like shapes. While these shapes are simplified to the extent that
they do not incorporate any of the secondary features of real vehicles (such as
mirrors, handlebars, cab extenders, etc ... ), they do capture the first order
effects that account for virtually all aerodynamic forces acting on a real truck.
Also, it is worth noting that Reynolds number matching is not achievable in
our flow facility because of limitations on model size and flow speeds.
However, the experiments discussed here are not meant to simulate the flow
around an actual tractor trailer. Instead, the goal is to unveil relevant flow
physics that can be generalized, at least to a first order, to higher Reynolds
numbers. The simplified shapes also have the advantage of lending themselves
to numerical computations. The elimination of detailed features on the models
allows numerical computations to be conducted without the costly burden of
On the Aerodynamics ofTractor-Trailers 187
Tntiler
I
\ f
\ I
I I
''
I
\ I
I I
I
\ I
I I
'\ j,' I
\ I /
J /
' /
'
___ ___, _____ __
Figure 1. Detail of turntable and traverse mechanism inside ground plane.
RVC>;(,_,
"~"-
rLE'
~~r$~OO)
10 1.60 (4 .0)
~ 1~.31 (8.4375)
A 45.47 (1 .79) ~ ~ _l
1~) 1 307.98(12.125) 1
Units: mm (ln.) I
e.l.,,,~O ~
A 27.05 (1 .065)
I
[] 214~1 (BE~ I I i I
Units: mm (ln.) I
Figure 2. Schematics of tractor and trailer with dimensions.
As Figure 1 illustrates, the position of the tractor is fixed with respect to the
turn-table, while the trailer is mounted on a motorized traverse so that the gap
width between the two bodies can be varied continuously. The vertical leading
edges on the front face of the trailer are also rounded with a radius of
curvature of 4.55 em.
front end (larger radius of curvature) experiences considerably less drag. Here,
the data is plotted versus ReA. Plotting the same data against ReR, as shown in
Figure 4, essentially collapses the two curves on each other. Hence, the radius
of curvature appears to be a more appropriate length scale for this flow. Also,
note that the drag bottoms out at approximately ReR = 70,000. This suggests
that the flow is not prone to leading edge separation above this critical
Reynolds number, and that further rounding of the front end would not lower
the drag.
0.60
~
0.55 "---- .________. r = 1.3cm
--.....
T
Cn ~ ~
0.50
"' ..___-, r=5.1cm
~
0.45
0.60
\ \;
~
0.55
Cn
0.50
...
..
~~.
~
0.45
0 .9
Total Drag
0 .8 . . ,...
/ ...._. _._"'·"'"""--~"_
-
'
0.7 I
,.. 0 .6 Trailer Drag
~
·c I
~- -
e
Q
u
~
0.5
0.4 F---·-·
-- --. -- I-
I
Q
~
...
~
0 .3
J. I -·.
f~_ .. . ---------- . ... .. - .. - .... ----
................ - TractorDrag
0 .2
-
0 .1 f>
0
0 0.1 0.2 0.3 0.4 0 .5 0.6 0.7 0 .8 0 .9 1 1 .1 1 .2 1 .3 1.4
G
Nonnalized Gap Width
JA
Figure 5. Drag coefficient versus gap width.
S ymmelric
flow
Tra1ler
Figure 6. Time-averaged streamline patterns of the flow in the gap at zero yaw, G/VA
= 0.28.
On the Aerodynamics ofTractor-Trailers 193
Tratler
Cab
Trader
Tratler
Symmetric
flow
Trailer
Figure 10. Ensemble-averaged streamline patterns of the flow in the gap at zero yaw,
GNA = 1.0.
Hence, over a relatively narrow range of gap width, the truck experiences a
dramatic increase in drag. In addition, both drag and side forces are
characterized by a substantial increase in fluctuation levels, such that the entire
truck undergoes intermittent shaking. This is illustrated by Figure 11, which
represents time series of drag force on the trailer at sub-critical and critical gap
width. As the gap increases, note that the drag starts out relatively steady then
exhibits large-scale, low-frequency oscillations when the critical gap width is
achieved. Furthermore, the oscillation amplitude varies with time. Over the
first five seconds of data in the bottom plot of Figure 11, the gap flow is
symmetric and oscillation amplitude is low, while the next five seconds show
higher amplitude as the flow goes asymmetric.
On the Aerodynamics ofTractor-Trailers 195
5
Sub-critical gap
4
~
2
0 10 20 30 40 50 60 70
9r------r------~----~------~-----.
8 Ctitical gap
~~_pJ'v
6~----~------~----~------~----~
30 32 34 36 38 40
Tiint~ (s)
Figure 11. Time series of drag on trailer at sub-critical (top) and critical (bottom) gap
width. Bottom plot is a portion of a longer time series.
0
0 15
0.8
0 0
(a)
t 5
0
0
0.8 10
0 .8
SliM 0 0
Ytw ,liiQit
(b)
On the Aerodynamics ofTractor-Trailers 197
0
0
0 .8 tO
t 6
0.8
0 0
Yaw Angle
(c)
Figure 12. Surface maps of drag force on tractor (a), trailer (b) and entire truck (c) as a
function of yaw angle and gap width.
t'
0
0 15
(a)
198 M. Hammache and F. Browand
I ~
c
c
(b)
c
c
c 0
(c)
Figure 13. Surface maps of side force on tractor (a), trailer (b) and entire truck (c) as a
function of yaw angle and gap width.
...
Flow
'pr --
CDrotal
Skm
(fop
I
R un led
'pr::: - 3 'p, ·.ll
(b
Drotal :;:::. ;'i l
Figure 14. Two basic shapes: (a) a simple parallelepiped; (b) same parallelepiped but
with rounded vertical front edges.
The drag in isolation is defined as the drag of a body alone in the wind tunnel.
200 M. Hammache and F. Browand
_.
Fl " '
Rounded I Roun ed
Rounded I Blunt
Blunt I Blunt
Blunt I Rounded
•• ••·~#&e.
-
Ot
01
lo· ~ or I
~
.
Ot _ / ... 01·
§
'\
OJ
... ..-• # ..........
••
u
o• ..
O>
. .•
q
~
OJ
~~ 02
• ••
01
'
01
,. o
01
••
,.
Figure 16. Drag curves for various combinations of simplified bodies. o front-body
drag; ~rear-body drag;_ average drag. (a) blunt lead and trail; (b) blunt lead-
rounded trail; (c) rounded lead- blunt trail; (d) rounded lead and trail.
The total drag savings for the pair of bodies is given by the plot of average
drag ratio as defined earlier. The average drag ratios for the four configurations
are plotted on Figure 17.
The tandem configuration that experiences the least amount of total drag
is that of a rounded body in the front and a blunt body in the rear.
Conversely, when the relative positions of the two bodies are switched, the
highest possible total drag results. Note that when the two bodies are
identical, the total drag ratio is essential identical in the region of strong
interaction, regardless of whether the bodies are both rounded or both blunt.
As the gap is increased beyond SlYA= 1, the tandem of two blunt bodies
experiences further drag ratio drop and starts to behave like the lowest drag
case (rounded followed by blunt) as SlYA approaches 3.5. On the other hand,
the tandem of two rounded bodies sees an increasing drag ratio with increasing
gap spacing, and gradually converges to the highest drag case (blunt followed
by rounded). As S/VA increases beyond 2.5, the pair of rounded bodies in
tandem appears to benefit less and less from the tandem configuration; the
202 M. Hammache and F. Browand
, 1
02
0 , 0.__ _0_._
5_ -~---, 5
2 25 3 35 ..
Normalized truck separation
Figure 17. Summary of average drag of platoon for all four configurations.
Hence it is clear that drag savings depend upon the drags of individual
trucks in isolation, and upon which truck is leading. For completeness,
measurements are made with more realistic truck models, as shown in Figure
18. The trucks have wheels, and the drag of each truck is artificially increased
by introducing drag-enhancing elements, such as the netting spoken of earlier,
by providing a gap between the tractor and trailer, and by providing additional
drag enhancing "collars". The drag collars consist of a series of short,
protruding cylinders ringing the trailer at a particular station. They are meant
to provide an increase in drag by increasing the boundary layer momentum
thickness. Various combinations that either include or exclude netting collars,
On the Aerodynamics ofTractor-Trailers 203
or gap, result in truck models having drag coefficients in isolation in the range
C 0 = 0.5- 0.7.
Figure 18. Truck models in tandem configuration. With netting and "collars", the
C 0 for each truck in isolation= 0.7.
The drag data for the realistic models is plotted in Figure 19. The data
points collapse on a single curve within the range covered by the simplified
shapes. This indicates that the data total drag savings for the models in Figure
18 in tandem is independent of the choice of leading and trailing truck. Note
also that the drag ratio rises gradually with gap spacing and that the local
maxim at S/vA= 1 vanishes. The effect of the critical gap is not so dramatic for
models having a distributed drag including wheels and drag collars.
Also, the entire data set is bounded by the "blunt - rounded" and
"rounded - blunt" cases on the upper and lower end, respectively. These
limiting cases delineate the maximum and minimum drag savings for any
tandem configuration. For comparison purposes, the road test data of Bonnet
& Fritz (2000) (who used real trucks) is included, along with its computed
average. The average drag ratio for the road test appears to agree best with the
"blunt- blunt" and "rounded- blunt" where there is overlap.
204 M. Hammache and F. Browand
c
0
'T
:p 09
co
0U)
08
.!;;;
Ol 07
co
L.
0 06
19 Bonne &Fritz
!
0
t:_ D!i
field test
Ol
co
,_
0
19
Q
~ 03
No greater savings
02
Figure 19. Average drag of platoons. The circles represent data obtained by arranging
the models shown in Figure 18 in various tandem combinations.
Summary Conclusions
The present experiments reveal the impact of on the drag of a single tractor-
trailer truck. A combination of force balance and whole-field velocity
measurements demonstrates a direct link between the level of drag forces and
the nature of the fluid flow patterns within the gap. At moderate widths a
stable toroidal vortex is present in the gap and drag levels are relatively low.
The symmetry of the flow pattern breaks down at a width of approximately
half the square-root of frontal area, which leads to intermittent flow separation
off the front of the trailer and a non-linear increase of drag forces on the trailer
and on the truck as whole. Most of the drag contribution is attributed to the
trailer alone; the tractor is affected much more moderately and only in a
narrow band of gap width near the critical value of G/VA=0.5. A similar
process of drag rise also occurs at moderate angles of yaw. However, the drag
increase is more pronounced and occurs at a lower critical gap width as the
angle of yaw is increased. At more extreme angles of yaw the tractor and trailer
are rather like independent bodies that do not benefit from dose-formation
and their drag behavior is much less a function of spacing.
The benefits of dose-formation also apply to trucks in tandem. The total
drag for all tandems tested is lower than the sum of the drags of the models in
isolation, though the drag savings vary with the degree of bluntness of each
On the Aerodynamics ofTractor-Trailers 205
model and on which model is placed in front. The highest relative drag
reduction is achieved by a rounded model followed by a blunt model, while
the reverse order achieves the least drag reduction.
References
K. R. Cooper: The effect of front-edge rounding and rear-edge shaping on the aerodynamic
drag of bluff vehicles in ground proximity. SAE paper No. 850288 (1985)
A.M. Fincham & G.R. Spedding: Low cost, high resolution DPIV for measurement of
turbulent fluid flow. Exps. Fluids 23, (1997)
Abstract
Introduction
Problem Formulation
The configuration to be examined is the Ground Transportation System
(GTS) studied experimentally at the NASA Ames research center (Storms et al
2001). The GTS geometry is a simplified tractor/trailer configuration which is
mounted on four posts in the wind tunnel. A photograph of the GTS in the
NASA Ames 7xl 0 ft wind tunnel is shown in Figure 1. The GTS model is an
approximately 1/8 scale class 8 tractor/trailer configuration. The Reynolds
number based on the trailer width (W=0.3238 m) is 2 million, approximately
one-half of full scale.
RANS Simulations of a Simplified Tractor/Trailer Geometry 209
0075
0.05
·0.025
Tunnel Side Wall
-0.05
0 Expt. (yfW = 2.6)
0 Expt. (yfW = 1.2)
-0.075 - - - Computed (yfW = 2.6)
- - - - Computed (yfW = 1.2)
2 3 4 5 6 7 8
xJW
0
0 0.25
Numerical Formulation
CFD Code
The CFD code is SACCARA, the Sandia Advanced Code for Compressible
Aerothermodynamics Research and Analysis, and was developed from a paral-
lel distributed memory version (Wong et a!. 1995) of the INCA code, origi-
nally written by Amtec Engineering. This code is used to solve the Navier-
Stokes equations for conservation of mass, momentum, energy, and turbulence
transport in three-dimensional form . The governing equations are discretized
using a cell-centered finite-volume approach. A finite-volume form of Yee's
symmetric TVD scheme (Yee 1987) is employed. This flux scheme is second-
order accurate and reduces to a first-order Roe-type flux (Roe 1981) in regions
of large gradients based on a minmod limiter. The viscous terms are discretized
using central differences. The SACCARA code employs a massively parallel
distributed memory architecture based on multi-block structured grids. The
solver is a Lower-Upper Symmetric Gauss-Seidel scheme (Yoon and Jameson
1988) which provides for excellent scalability up to thousands of processors. A
number of code verification studies have been performed which give confi-
dence that the code is free from coding mistakes including comparison to es-
tablished numerical benchmark solutions and code to code comparisons (Roy
et al 2000). The fine grid and coarse grid simulations presented herein were
run in parallel on the ASCI Red parallel processing machine using 1149 and
125 processors, respectively.
212 C. Roy et a!.
Turbulence Models
Two turbulence models are examined in the current work: the one-equation
eddy viscosity transport model of Spalart-Allmaras (Spalart and Allmaras
1994) and the Menter k-w model (Menter 1994). The Spalart-Allmaras model
requires the solution of a single transport equation for the eddy viscosity. The
Spalart-Allmaras model has proven to be a numerically robust approach, and
generally good results have been demonstrated for a wide variety of flows. The
Menter k-w model is a hybrid model which uses a blending function to com-
bine the best aspects of both the k-w and the k-E turbulence models. Near solid
walls, a k-w formulation is used which allows integration to the wall without
any special damping or wall functions. Near the outer edge of the boundary
layer and in shear layers, the model blends into a transformed version of the k-
E formulation, thus providing good predictions for free shear flows. This
model also shows less sensitivity to freestream turbulence quantities than other
k-w formulations. In both cases, the turbulence models were integrated to the
wall in order to avoid model validation issues associated with wall functions.
Numerical Accuracy
Before the simulation results can be compared to experimental data, the nu-
merical accuracy of the solutions must be assessed. For the steady-state Menter
k-w computations examined herein, the solution accuracy will be judged by
examining the iterative convergence of the solutions to steady-state as well as
the discretization error. The numerical accuracy of the Spalart-Allmaras results
is expected to be comparable.
Iterative Convergence
Discretization Error
1 25
Front of Tl'aeto,
z/W . 0
~ 75 Me nti!lr k-4•J Model
- - FlnoGrid
- - - - CoarseGrld
05
025
-2 ·1
Cp
1 25
Tral1erBntt
z/W . 0
M&nler k~ " Model
1 - - FineGrid
- - - - CoarwGrid
~ 75
OS ,,
025
.()2 .()1
Cp
Surface Comparisons
Surface Pressure
The surface pressure for the two turbulence models using the fine mesh is compared to
experimental data (Storms et al 2001) on the front of the tractor in Figure 7. Both
simulations show good agreement with the experimental data. Simulation results are
presented for the base of the trailer in Figure 8. In this case, the Menter k-w model
does a reasonable job of matching the pressure levels, while the Spalart-Allmaras model
significantly underpredicts the pressure on the base.
Front of Tractor
ziW . 0
c E:lcpe rirne-" t
-- Menter~!! ... ,,
- - - - Spalar1-A.IIml!lt1S
OS
025
-2 ·1
Cp
-- - -... t D
'25 I
Trailer Ba se
ziW 0 =
0 E:~periment
- - M ento r h.-t·J
- - - - Spalart.Atlmaras
~ 75
OS
025
.OJ ·02 0
Cp
Aerodynamic Drag
Aerodynamic drag predictions using the two turbulence models as well as the
experimentally measured drag are presented in Table 2. These drag results
RANS Simulations of a Simplified Tractor/Trailer Geometry 215
(both predicted and experimental) are for the GTS model only and do not in-
clude the support posts. The Menter k-w results are approximately 7.5%
higher than the experimental value, while the Spalart-Allmaras results are
nearly 50% high. The overprediction of the drag with the Spalart-Allmaras
model is due to the poor prediction of the base pressure. Also shown in the ta-
ble is the estimated uncertainty in the experimental drag coefficient (Storms et
al 2001), and the estimated numerical error for the fine grid Menter k-w
simulation. This numerical error estimate is determined by performing
Richardson extrapolation using the coarse and fine grid drag coefficients,
0.474 and 0.298, respectively. The resulting extrapolated value for the drag
coefficient using the Menter k-w model is 0.239, which is essentially within
the experimental uncertainty bounds.
Estimated Uncer-
Drag_ Coeff., C0 tainty/Error
Experiment (Storms et al2001) 0.25 ±0.01
Menterk-w 0.298 ±0.06
Spalart-Allmaras 0.413 --
Field Comparisons
Velocity data are available from PIV measurements performed at the NASA
Ames 7x10 ft wind tunnel (Storms et al 2001). These PIV data represent a
time-averaged picture of the flow in the wake regions immediately behind the
trailer base. Figure 9 gives streamlines based on the PIV data in a vertical
streamwise cut through the wake (z/W = 0). The flow is from left to right,
with the base of the trailer shown on the left; the PIV window is also shown in
the figure. A large, counter-clockwise-rotating vortex is centered at approxi-
mately x/W = 8, y/W = 0.4. Also, the presence of a clockwise-rotating vortex is
suggested by the vertical nature of the streamlines in the upper right-hand cor-
ner of the PIV window. A similar view of the streamlines from the fine grid
computations using the Menter k-w model is shown in Figure 10. These
Menter k-w computations predict a more symmetric pair of vortices than is
indicated from the experimental data. The location of the experimental PIV
window is shown in the figure for reference. Streamlines for the Spalart-
Allmaras model are presented in Figure 11, showing a much shorter recircula-
tion zone than the Menter model. This shorter recirculation zone produces
larger velocities in the outer flow as the flow accelerates around the wake. The
higher velocities result in lower pressures and hence higher drag (see Table 2).
216 C. Roy et al.
82 84 86 88 92
xJW
04
01
62 84 86 sa 92
xJW
02 ;
82 84 86 88 92
xJW
Experimental PIV results for a horizontal streamwise cut through the wake
(y/W = 0.696) is shown in Figure 12. Two counter-rotating vortices are evi-
dent at x/W = 8.5. The computational streamlines for this case using the
Menter k-w model are given in Figure 13. In this case, the location of the vor-
tices is accurately predicted by the RANS model.
Experiment: NASA Ames
Horizontal Streamwi$e Cut: y!W ~ 0 .696
08
95
- -_;- - ~ --
04
85 95
xJW
Conclusions
Steady-state RANS simulations were conducted for the flow over the GTS ge-
ometry. The numerical accuracy of the computed flowfields was assessed by
performing the computations on multiple grids. Simulation results using the
Menter k-w turbulence model gave good agreement with the experimental
data for surface pressure, field velocities in the wake, and drag coefficient;
however, this model predicted a more symmetric pair of counter-rotating vor-
218 C. Roy et al.
tices in the vertical streamwise plane than was indicated in the experiment.
These results suggest that the Menter k-w model can accurately predict the
drag for tractor/trailer configurations when performed on highly refined
meshes using integration to the wall, although some time-averaged flow
structures are not captured. Simulation results with the Spalart-Allmaras tur-
bulence model showed good agreement with experimental data for the surface
pressure in the attached flow regions, but significantly underpredicted the
pressure in the base region. As a result, the Spalart-Allmaras model overpre-
dicted the drag coefficient by nearly 50%.
References
McCallen R, Couch R, Hsu J, Browand F, Hammache M, Leonard A, Brady M, Salari K, Rut-
ledge W, Ross J, Storms B, Heineck JT, Driver D, Bell J, Zilliac G (1999) Progress in re-
ducing aerodynamic drag for higher efficiency of heavy duty trucks (class 7-8). SAE Paper
1999-01-2238.
Menter FR (1994) Two-equation eddy-viscosity turbulence models for engineering applications.
AIAAJournal32: 1598-1605
Roe PL (1981) Approximate Riemann solvers, parameter vectors, and difference schemes. Jour-
nal of Computational Physics 43: 357-372
Roy CJ, McWherter-Payne MA, OberkampfWL (2000) Verification and validation for laminar
hypersonic flowfields. AIAA Paper 2000-2550
Spalart PR, Allmaras SR. (1994) A one-equation turbulence model for aerodynamic flows. La
Recherche Aerospatiale 1: 5-21
Storms BL, Ross JC, Heineck JT, Walker SM, Driver DM, Zilliac GG (2001) An experimental
study of the ground transportation system (GTS) model in the NASA Ames 7- by 10-ft
wind tunnel. NASA TM-2001-209621.
Wong CC, Soetrisno M, Blattner FG, Imlay ST, Payne JL (1995) PINCA: a scalable parallel
program for compressible gas dynamics with nonequilibrium chemistry. Sandia National
Laboratories Report SAND 94-2436, Albuquerque, NM
Yee HC (1987) Implicit and symmetric shock capturing schemes. NASA-TM-89464
Yoon S, Jameson A (1988) An LU-SSOR scheme for the Euler and Navier-Stokes equations.
AIAA Paper 87-0600
Computational Simulation of a Heavy Vehicle
Trailer Wake
1 Introduction
Over the past several decades, a significant amount of effort has been put
forth to reduce the aerodynamic drag of heavy vehicles. It has been shown that
the power required to overcome the aerodynamic drag at highway speeds (70
mph) is on the order of 65% of the total fuel consumed by the engine of a heavy
vehicle. Thus, a reduction in the total aerodynamic drag will have a significant
improvement on the fuel economy of a heavy vehicle. The aerodynamic drag
can be divided into two components: viscous drag and pressure drag. For
the flow over a heavy vehicle at highway speeds, the Reynolds number is
large enough such that viscous forces can be safely ignored. Consequently, the
drag experienced by a heavy vehicle is primarily due to pressure drag. The
pressure drag is comprised of the pressure forces that exist on the front and
rear of the vehicle. In recent years, the pressure drag on the front of a heavy
vehicle has been successfully reduced by streamlining the tractor, installing
air shields on the tractor roof, or optimizing the relative height of the tractor
and trailer. However, the pressure drag on the rear of a heavy vehicle, also
known as base-drag, yet remains to be reduced on a widespread basis. This
220 J.M. Ortega et al.
2 Computational Setup
~= ('#..)1/7 (1)
Uo 8
where U0 = 92.65 m/s is the freestream velocity. 8 is taken to be the turbulent
boundary layer thickness on a flat plate with a length of l = 6.26w (see Figure
1b) and is found from
222 J.M. Ortega et al.
3 Results
To visualize regions of rotational flow, a sequence of iso-surfaces of vorticity
in the wake of the truncated GTS is shown in Figure 4 for the coarse grid. As
the solution evolves, shear layers develop on the periphery of the truncated
GTS. These shear layers roll up into a vortex ring, which is advected down-
stream (Figure 4a). A secondary vortex ring begins to develop, though it is
not quite as coherent as the first. Shortly thereafter, the vortex rings break
apart and no evidence of the formation of new vortex rings is seen for the
rest of the simulation. No effort was made to determine the dependence of the
characteristics of the vortex rings on the specified initial conditions. The ini-
tial transients quickly subside and the shear layers extend downstream of the
base of the GTS. The shear layers repetitively undulate periodically in time
and frequently shed patches of vorticity into the bluff body wake (Figure 4d).
It can be seen that the massively separated wake interacts with the ground
plane downstream of the GTS, resulting in a flow separation bubble on the
ground plane (Figure 4d). The vorticity fields from the solution on the fine
grid are slightly different in appearance than those obtained for the coarse
grid. Since the turbulent length scale is dependent on the grid resolution,
the vorticity fields calculated on the fine grid (Figure 5) possess significantly
more small scale structures than those found for the coarse grid. Additionally,
a review of the animations of the iso-vorticity surfaces reveals that patches of
vorticity often burst through the shear layers in the fine grid solution, while
no such phenomena is seen to occur in the coarse grid solution.
One of the more interesting calculated quantities is the helicity field in the
wake of the truncated GTS. Helicity is defined to be u x w, where u and w
Computational Simulation of a Heavy Vehicle Trailer Wake 223
are the velocity and vorticity vector fields, respectively. The helicity can be
thought of as a measure of a flow field that has a swirling nature to it. For
instance, a two-dimensional viscous vortex with an axial velocity component
acting co-linear with the vorticity field would be an example of a flow with a
positive value of helicity. Regions of a flow field that are irrotational (w = 0)
or have a no-slip boundary condition (u = 0) have a zero value of helicity. Fig-
ures 6 and 7 show iso-surfaces of the time-averaged helicity in the wake of the
truncated GTS for the coarse and fine grids, respectively. The time averages
are calculated over the life of the simulation after the initial transients have
decayed. The surfaces denote values of helicity that are ±10,000 m/s 2 , re-
spectively. The coherent structures present in the wake of the GTS are most
clearly evident in the upstream views of the helicity iso-surfaces. It can be
seen that the top corners each spawn the formation of two oppositely-signed,
swirling flow structures. From the side views in Figure 6b and 7b, it is evident
that these structures are fairly robust and persist relatively far downstream
in the wake of the truncated GTS. The solutions on the coarse grid and, more
distinctly, on the fine grid demonstrate oppositely-signed surfaces of helicity
at the bottom corners of the GTS. However, these surfaces are smaller than
those being produced on the top corners, indicative of the influence of the
ground plane. Due to the potential upstream influence of the truncation of
the GTS geometry, the reader may be skeptical of the validity of these helic-
ity surfaces and whether or not they would exist in an actual flow. Perhaps,
they are artifacts of the computational scheme related to the sharp corners of
the truncated geometry. To address these issues, future computational studies
need to determine the sensitivity of these results to the upstream truncation
condition, the turbulence model, the boundary conditions used in the simula-
tions, and the sharp corners of the geometry.
We wanted to determine how well the simulations with the truncated geom-
etry compare with the experimental data for the full GTS geometry [7] taken
at a Rew = 2.0 X 106 and yaw angle of 0°. For this comparison, the particle
image velocimetry (PIV) data of the wind tunnel study [7] is employed. Figure
8 shows a schematic of the PIV setup for vertical and horizontal mid-plane
velocity field measurements in the wake of the GTS. Post-processing of these
velocity fields allows for the calculation of other flow quantities, such as the
vorticity or kinetic energy fields.
Figure 9 shows a comparison of the time-averaged velocity fields in the
vertical planes for the coarse grid and fine grid, where the time-averaged
quantities are calculated in the manner described in Section 3. Figure 10
shows the corresponding PIV velocity measurements in this same plane. Due
to the limited field of view of the PIV camera, only a portion of the wake
of the GTS is measured in the PIV data. There is a difference in the wake
224 J.M. Ortega et al.
structure when comparing the results of the vertical mid-plane velocity fields.
For the solutions from both the coarse and fine grids, two counter-rotating flow
structures exist in the separated wake, with the top one positioned slightly
upstream of the bottom one. However, in the PIV data, there is no evidence
for a second flow structure. Perhaps, this second flow structure exists outside
the field of view of the PIV camera. The measured wake in the PIV data looks
very similar to the near wake of a bus [13]. This distinct difference between
the simulations and experimental measurements may be due to the effects of
truncating the model or neglecting the undercarriage on the GTS geometry
(e.g., the four supporting posts on the GTS model) and the wind tunnel walls.
Figure 11 shows instantaneous fields of the y-component of vorticity in the
horizontal mid-plane for the coarse grid, fine grid, and PIV data. As expected,
there is a distinct difference between the vorticity fields of the coarse and fine
grids, namely that the fine grid captures more smaller scale structures in the
shear layers being shed by the GTS. These small scale structures are also
evident in the shear layers measured in the PIV data.
5 Conclusions
Unsteady, three-dimensional, large-eddy simulations are done on the flow
about a truncated GTS geometry located above a no-slip surface. Calculations
of the vorticity field show the unsteady behavior of the shear layers being shed
by the GTS geometry. Time-averaged iso-surfaces of helicity demonstrate the
existence of coherent flow structures that are formed from the corners of the
GTS. These structures do not readily dissipate and exist beyond the massively
separated wake. A comparison with the PIV velocity measurements made at
the same Reynolds number and yaw angle reveals differences in the structure
of the separated wake in the vertical mid-plane, perhaps indicating the influ-
ence of not including the model supports or of truncating the GTS model.
Future simulations on the GTS geometry must investigate the sensitivity of
the solutions to these effects, including the influence of the wind tunnel con-
finement, turbulent boundary layer treatment, and time averaging issues. In
regard to achieving the goal of reducing the drag of heavy vehicles, future
simulations also need to explore and to understand the flow physics of various
base-drag reduction concepts.
References
1. Lanser, W.R., Ross, J.C., & Kaufman, A.E. Aerodynamic Performance of a
Drag Reduction Device on a Full-Scale Tractor Trailer. Paper No. 912125, SAE
Transactions, 1991.
2. Virgin, B. Rounded add-on makes driving truck less of a drag, Seattle Post-
Intelligencer, http: I /seattlepi. nwsource. com/business/103544_truck10.
shtml, January 10, 2003.
Computational Simulation of a Heavy Vehicle Trailer Wake 225
3. Whitten, W.D. Collapsible Streamlined Tail for Trucks and Trailer, U.S. Patent
Number 6,309,010, September 27, 2000.
4. Bassett, D.A., Bassett, M.R., & Schroeder, J.C. Method and Apparatus for
Reducing Drag on a Moving Body, U.S. Patent Number 5,374,013, November
4, 1993.
5. Kirsch, J.W., Sabodh, K.G., & Garg, K. Airvane Device for Bluff Vehicles and
the Like, U.S. Patent Number 3,999, 797, November 26, 1975.
6. Engler, R.J. Advanced Aerodynamic Devices to Improve the Performance , Eco-
nomics, Handling and Safety of Heavy Vehicles, SAE Paper 2001-01-2072.
7. Storms, B.L., Ross, J.C., Reineck J.T., Walker, S.M., Driver, D.M. & Zilliac,
G.G. An Experimental Study of the Ground Transportation (GTS) Model in
the NASA Ames 7- by 10-ft Wind Tunnel, NASA/TM-2001-209621, February,
2001.
8. Gresho, P.M. & Sani, R.L Incompressible Flow and the Finite Element Method,
Volumes 1 & 2, April, 2000.
9. Smagorinsky, J. General Circulation Experiments with the Primitive Equations,
I, The Basic Experiment, Mon. Weather Rev. , 91, 1963, 99-165.
10. McCallen, R.C. Large-Eddy Simulation of Turbulent Flow Using the Finite El-
ement Method, Ph.D. Thesis, University of California, Davis, 1993.
11. Deardorff, J.W. A Numerical Study of Threee-Dimensional Turbulent Channel
Flow at Large Reynolds Numbers , J. Fluid Mech., 41, 453-480.
12. Piomelli, U., Ferziger, J.H., & Moin, P. Models for Large Eddy Simulations
of Turbulent Channel Flows Including Transpiration, Rept TF-32, Stanford
University, 1987.
13. Gotz, H. & Mayr, G. Commercial Vehicles, Aerodynamics of Road Vehicles,
edited by Hucho, W.H., 1998, 415-488.
truncation
line
flow direction
0.24w
(a) (b)
Fig. 1. (a) Photo of the GTS geometry in the NASA Ames 7ft x 10ft wind tunnel.
(b) Schematic of the GTS geometry and location of the model truncation. w is the
width of the GTS model (0.3238 m).
226 J.M. Ortega et a!.
~flow
(/ direction
I
3.7w
\
ground
plane
Fig. 2 . Computational domain and boundary conditions for the flow simulations.
(a) (b)
Fig. 3. Vertical mid-plane cross-sections (z = 0) of the (a) coarse and (b) fine grids
used in the computational study.
Computational Simulation of a Heavy Vehicle Trailer Wake 227
ground plane
Fig. 4. !so-surfaces of the vorticity magnitude at several times in the coarse grid
simulation.
228 J .M. Ortega et al.
Fig. 5. !so-surfaces of the vorticity magnitude at several times in the fine grid
simulation.
Computational Simulation of a Heavy Vehicle Trailer Wake 229
(a)
• ~-'\o3 ~ A x.~
-w. 6 0. ~ 0 . :1 -o ~B!
0.8 .a 0. .e
0.6 .6 0. .6
,.,,
0.4 ,q 0. .4
' R
0.:1 . :1
•3 .~
0. . :1
0 'I
.,.,- ,...J -0 .
(b)
FI K,~
0.8 0.8
0.5 0.6
0.4 0.4
A.
' ll.2
~
Rx.~
-0 . :1
z R
••S
~ -0 .:<'
' Rxo~
L.
Fig. 6. (a) Upstream and (b) side views of the time-averaged iso-surfaces of helicity
for the coarse grid.
230 J.M. Ortega et al.
(a)
(b)
.e o.e
().5 o. e.
0.4 0.'1
Fig. 7. (a) Upstream and (b) side views of the time-averaged iso-surfaces of helicity
for the fine grid.
Computational Simulation of a Heavy Vehicle Trailer Wa ke 231
GT ' model
PI\ ' camcm
\CMtcal plane hontontal plane
(a) (b)
Fig. 8. Schem atic of t he PIV setup in the NASA Ames 7ft x 10ft wind tunnel for
(a) vertical mid-plane a nd (b) horizont al mid-plane measurements.
o• (a)
03
02
0 I
~
>-
·0 I
.02
.03
·0 . ·0 2 0 02
X AJJI
•• 06 oa
•• (b)
Ol
02
01
.e
~
>-
.01
.02
.03
.o• .02 02
X All ls
•• 06 oa
0.4
0.3 (a)
0.2
§. 0.1
>-
0.0
-0.1
-0.2
-0.3
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8
x (m)
(b)
~
CO\
................................................................ , .......
Fig. 10. (a) PIV velocity field measurements in the vertical mid-plane. For clarity,
a schematic of the velocity field in (a) is shown in (b).
Computational Simulation of a Heavy Vehicle Trailer Wake 233
(a)
~
0 (b)
to)
0 -·------------·
N ...;
ci
r.
~
~
ci r
"'
·x f
~
"'
N ~
~
~
9
N
9
M ...............................
9
~
9
~ N N .0 co
9 9 ~ ci
~
0 ci ci
~
0
(c}
.:.
to)
0 .................................
'' '
N :
ci '
ci
.!a 0
f
X
ci
"'
N
~
"'!
9
M ... . . .. ....... .............
9
"':
9
"': N 0 N "': co
9 9 ci ci 0 ~ ci
Abstract
Results of drag reduction studies based on direct numerical simulations of the
separating flow aft of a rectangular forebody are reported. The study examines
the two-dimensional wake flow forming in the presence of rectangular boat-
tails (blocks or plates) attached to the base of a forebody, with special emphasis
on modifying the base pressure component of the drag. An optimal aspect
ratio for boat-tail notches at the separation point are identified, and the
underlying mechanism leading to elevated base pressures in such
configurations is revealed.
1. Introduction
Aerodynamic drag reduction has been a prominent concern, and has received
extensive investigation, in the automobile industry for many years. However,
a comparable emphasis on optimizing the drag of large transport vehicles has
peaked only recently, driven largely by increasing economic and
environmental pressures arising from their increased role in most economies.
This is especially true in the case of large tractor-trailer units used widely for
the transport of raw and manufactured goods over large distances. In most
applications employing tractor-trailer rigs, a significant fraction of fuel
expenditures is directly attributable to aerodynamic drag, and incentives exist
therefore to devise both active and passive means of reducing this penalty.
There are several primary target areas on tractor-trailer combinations that
are deemed most amenable to aerodynamic improvement. These include the
aerodynamic shaping of the frontal area of the trailer, the shielding of the gap
between the tractor and the trailer, and modifications of the separating flow in
the base region of the trailer. It has been found in some cases that minor local
modifications can yield significant reductions in aerodynamic drag. These
practical improvements have often been realized largely through empirical
238 D.R. Areas and L.G. Redekopp
2. Problem Definition
The possibility of reducing the aerodynamic drag of a bluff body through
modifications of the base geometry is investigated by means of numerical
simulation. Toward that end we assess changes in the structure and flow
properties in the wake behind a rectangular, two-dimensional body when a
class of boat-tails are attached to the trailing edge of the base configuration.
Two different boat-tail geometries are examined in this study: a closed boat-
tail obtained by the symmetric attachment of a closed box to the base of the
body, and an open boat-tail obtained by the symmetric attachment of a pair of
plates extending from the base of the body. A schematic of these base
configurations is shown in Figure 1. The closed box configuration has a width
smaller than that of the reference body, allowing for the formation of two
corner notches of non-dimensional depth d and extension length l adjacent to
the separation point. All dimensions are scaled with the forebody width W
Our purpose in particular is to examine the effect of these two geometric
parameters on the integrated base pressure over the total width of the body.
The open box configuration is characterized by the same non-dimensional
parameters, but offers further freedom to the near-wake dynamics arising from
the existence of a cavity formed between the two extension plates.
Drag Reduction ofTwo-Dimensional Bodies by Addition of Boat Tails 239
For the purposes of this study we ignore any details concerning the frontal
shape of the forebody, and the forebody length, and simply examine the flow
in the base region with a prescribed inflow to the computational domain
which is specified at a set distance upstream of the trailing edge. The inflow
consists of identical parallel streams on either side of the forebody with
nominal laminar boundary layers adjacent to the forebody. With this
formulation, the (non-dimensional) boundary layer thickness {j at the inlet
station on the forebody is an independent parameter and must also be
prescribed in addition to the Reynolds number. Consequently, the parameter
space for the problem includes the geometric parameters d and /, and the flow
parameters {j and Re, where Re is the Reynolds number based on the ambient
speed and the forebody width.
The numerical code used in this study is based on the Marker-And-Cell
(MAC) method, developed originally by Welch & Harlow (1965) . The
method is characterised by the use of a staggered grid and the solution of a
Poisson equation for the pressure at every time-step. The code solves the two-
space dimensional Navier-Stokes equations for an incompressible fluid in a
channel with impermeable, stress-free boundaries confining the flow. In this
study the outer channel boundaries were positioned a distance of four and a
half units away from each side of the forebody, and the outflow boundary was
20 units downstream from the inlet. The forebody length was three units,
leaving a distance of 17 units for the wake development between the base of
the forebody and the outflow boundary. A convective outflow boundary
condition was applied at the downstream end of the domain. The code was
validated by comparing the critical Reynolds number for onset of vortex
shedding, the frequency of vortex shedding, and the length of the recirculation
region for sub-critical Reynolds numbers, with existing data and simulations
for flow over a rectangular forebody (cf. , Eisenlohr & Eckelmann, 1988, and
Hammond & Redekopp, 1997), and also comparing similar characteristics for
flow over a backward-facing step (cf., Armaly et al., 1983, and Kim & Moin,
1985).
With the flow being confined between impermeable, stress-free
boundaries, it is important to assess the problem of blockage imposed by the
240 D.R. Areas and L.G. Redekopp
proximity of these boundaries to the wake and their influence on the base
pressure drag. It was found that, with the channel boundaries set at a distance
of 5 W on each side of the centerline for computational efficiency, the base
pressure drag was essentially independent of the forebody length when L was
greater than 3. The insensitivity of the base pressure drag to forebody length is
important since the increased blockage arising from addition of a boat-tail
could adversely influence any conclusions regarding drag reduction. The
channel width and forebody length were, therefore, held fixed at these values
throughout the study. Also, all results that follow were computed for a fixed
boundary layer thickness of 0 = 0.3 and a Reynolds number of Re = 800. The
base pressure drag, computed by integrating the base pressure distribution
across the exposed end of the forebody, yielded a base pressure drag coefficient
CDp = -0. 11 . This value of the drag coefficient, noted hereafter as cdo> is used
subsequently as a reference value in comparing the efficacy of various boat-tail
configurations for drag reduction.
The computed drag coefficient can be compared to the experimental value
of C Dp = -0.2 reported by Balkanyi & Bernal (2000) for flow with Re =
100,000. The differences are substantial, but can perhaps be rationalized on
the basis of the considerable differences in Reynolds number and the three-
dimensional nature of the flow in the physical experiment. The variation of
our computed drag coefficient C Dp for a restricted range of low Reynolds
numbers is shown to be increasing with Reynolds number in Figure 2, but the
extrapolation of the curve to full-scale values is pure speculation.
-0 .02
-0 .04
-0 .06
-0 .08
'0 -0 .1
:.>
-0 .12 ...... •·>
-0 .16
-0.18 ... '· ...
-0 .2 L . __ __.__ __..__ ___._ _._L__ _ .....__ _j
3. Discussion of Results
The sensitivity of the base pressure drag to the notch depth was investigated
first for a fixed boat-tail length of l = 0.4. The time-averaged pressure field
was computed and pressure coefficient profiles along the base of the forebody
are plotted in Figure 3 for various notch depths. The arrows indicate the
direction of increasing values of notch depth d. It is clear that an optimum
depth exists, occurring for a value d = 0.13, where the base pressure recovery
attains its highest value. Interestingly, the pressure distribution is nearly
uniform across the base for this geometry. The integration of these profiles
across the base yields base pressure drag coefficients which, when normalized
by the drag coefficient for the reference case without any boat-tail, are plotted
in Figure 4. There is a quite steep decrease in drag for small notch depths
until the optimum for the particular boat-tail length is reached. The rise in
drag for notch depths beyond the optimum is, by comparison, quite modest.
- 0 .01
- 0.05
-0 .07
-0.09
---- --------
-0.11
-0 .13
-0 .15 L-~~-~----'----~~--~-'---'
· 0 .5 · 0 .4 -0 .3 -0 .2 -0.1 0 0 .1 0 .2 0 .3 0 .4 0 .5
y
-0. 02 1~~-~~---r--~-;==~~:;']
-0.025
-0 .03
-0.035
-0 .04
- 0 .1145
-0.05
-0 .055
-0 .06
-0 .065
- 0 .07 L__~---'-~_.__i__~___;_ _.__.__J
-0 .5 -0 .4 - 0 .3 -0 .2 - 0 .1 0 0 .1 0 .2 0 .3 0 .4 0 .5
y
Figure 3. Pressure coefficient profiles along base for different notch depths. l = 0.4, Re = 800.
242 D.R. Areas and L.G. Redekopp
0 .9
0 .8 ................. \
..,~ 0 .5
u 0 .4
0 .3
0 .2
0 .1
0
0 0 .05 0 .1 0 .15 0 .2 0 .25 0 .3
d
ap V2
(1) -=p-
an R
where R is the local radius of curvature of a streamline with local speed V and
n is normal to the streamline and pointing away from the center of curvature.
Since the ambient pressure is fixed, one needs a flow with a curvature that is
locally convex relative to the wake centerline (alt. concave relative to the
freestream) in order to increase the pressure along the base of the boat-tail.
This convex curvature is found to be maximized for the fixed parameters in
this set of simulations when d"" 0.13.
One can also examine characteristics of the time-averaged near-wake
structure as the notch depth is varied. We show in Figure 6 the distance
between the end of the boat-tail and the near-wake saddle point as a function
of the notch depth d. It is observed that the near-wake length reaches a
Drag Reduction ofTwo-Dimensional Bodies by Addition of Boat Tails 243
maximum when the optimum value d = 0.13 is attained. Clearly, the presence
of an inflectional shape in the streamlines defining the boundary of the near-
wake region increases the length of the near-wake. Also, as the length of the
near-wake increases, the convex curvature of the dividing streamline will be
diminished. Based on the inviscid relation (1), this effect will also contribute
to a raising of the pressure in the near-wake region
With these results in hand, the sensitivity of the base pressure drag
reduction to boat-tail length was pursued. For this purpose, the notch depth
was fixed at the optimum valued"' 0.13 for a notch length of I"' 0.4, and the
parameter I was varied to track the variation of the pressure drag with notch
length I. As shown in Figure 7, a pronounced minimum in the normalized
drag coefficient was found for I= 0.66. Using these two sets of simulations we
suggest that the optimum aspect ratio of trailing edge notches, optimum with
respect to base pressure drag reduction, is around lid= 6 (0.66/0.133). It is
interesting to note that this particular value of the notch aspect ratio
corresponds very closely with that identified by Lancer & Ross (1991) in their
tests of a full-scale model at typical speeds (i.e., Reynolds numbers). They
reported an optimum configuration of boat-tail plates having the values of I=
0.36 and d = 0.06 (both for the vertical and horizontal notch depths on their
three dimensional body); that is, a notch aspect ratio of six. For their vehicle
with added boat-tail plates satisfying these optimum dimensions, they report a
total drag reduction of 9%.
ct-l.tl
........,
Cp
....•••.,
'"'
....•••..
.........
'"'
·• 011
Oil
'"
0 .95
0 .9
.r; 0 .85
;;.
i
..... 0 .8
c
0 0 .75
-;;
:; 0 .7
2
'¥
a: 0 .65
0 .6
0 .55 .
0 .5
0 0. 05 0 .1 0 .15 0 .2 0 .25 0 .3 0 .35
d
0
'C
Q
'C
()
0.2
o ~----~------~----~~----~
0 2 3 4
results for the base pressure drag are shown in Figure 8. It is evident that
equivalent trends of drag reduction behaviors are realized when the boat-tail is
opened. Also, the absolute level of drag reduction is greater for an open boat-
tail as opposed to one that is closed. The extra freedom in the near-wake flow
structure, particularly the absence of the stagnation point on the plane
coinciding with the end of the boat-tail when it is open, allows for greater
pressure recovery in the near-wake. Evidently, it also allows for the appearance
of a region of more pronounced reflexive curvature of the streamlines at the
edge of the boundary layer. Also, the near-wake appears to be shorter when the
boat-tail is open, which would imply greater convex curvature of the
streamlines (relative to the center-line) just downstream of the boat-tail.
Comparisons of the pressure field for open and closed configurations with the
same parameters are shown in Figure 9 to exhibit some of these differences.
0 .8
~
\
\
I= closed
open
\
\
\
I
0 .6
..,
0 \
\
\
~ \
u \
0 .4
''
' ....
0 .2
0
0 0 .25 0 .5 0 .75
I
Figure 8. Base pressure drag coefficient for different boat-tail configurations as a function of
boat-tail length. d = 0.1, Re = 800.
246 D.R. Areas and L.G. Redekopp
1=0.4
1 . 5 .----~--~--~.----""'""-----..,....-----,
>-
Figure 9. Comparison of time-averaged pressure contours for an open and closed boat-tail
configuration.
d = 0.1, I= 0.4, Re= 800.
4. Conclusions
The numerical investigation presented in this paper clearly identifies an
optimum parameter setting for maximum pressure recovery (alt., drag
reduction) along the base of a rectangular body via the attachment of boat-tails
to the trailing edge. The combination of geometric parameters d, notch depth
Drag Reduction of Two-Dimensional Bodies by Addition of Boat Tails 247
dp1:1n
~dn
Figure I 0. Sketch of the modified separating streamlines and associated pressure gradients in
the presence of boat-tails.
Acknowledgements
The first author acknowledges partial support of a Powell Graduate Fellowship
during this study, and both authors gratefully acknowledge many collegial and
beneficial discussions of this work with F. K. Browand.
248 D.R. Areas and L.G. Redekopp
References
Harlow, F.H. & ].E. Welch, 1964. The MAC Method: A computing
technique for solving viscous, incompressible, transient fluid-flow problems
involving free surfaces. Los Alamos Scientific Laboratory Report No. LA-
3425.
Abstract
The use of planar-sided boat tail plates for aft-end drag reduction on a tractor-trailer
was studied numerically, experimentally and on a full scale prototype. Parametric
wind tunnel tests utilized a 1:15 scale Peterbilt 379 tractor and 48 foot (14.6 m)
trailer with cavity plate concepts mounted perpendicular to the trailer base. Yaw
angles up to 9 degrees were examined. Qualitative numerical results confirmed a
pressure increase on the aft face of the trailer. Model drag increments, obtained at
zero yaw and a width-based Reynolds number of 230,000, based on trailer width,
indicated reductions in the drag coefficient, based on frontal area, of up to 0.075 or
about 9% of the baseline model trailer drag. Removal of the top plate degraded the
performance of all devices. Performance also decreased with yaw angle for all
plates mounted perpendicular to the trailer base, contrary to devices with angled
plates. Devices with shorter angled plates indicated better performance with the
top open rather than an open bottom. Drag reduction was more sensitive to plate
inset from the trailer edge than to plate length and a zero inset of the bottom plate
maximized performance. Two full scale prototypes were road tested, the first
utilized rigid composite sides with a flexible top and bottom and the second was an
all rigid-sided aluminum design. The former exhibited cross-country road fuel
savings of about 0.5 miles per gallon (0.2 kilometers/liter), approximately 9%, over
a 10,000 mile (16,093 km) trip, while the latter returned inconclusive results.
Estimated fuel savings for a typical 120,000 miles (193,121 km) per year traveled
were approximately 1500 gallons (5677 liters) per truck.
Introduction
The flow behind bluff bodies, and the desire to reduce the base pressure drag, has
been the subject of numerous investigations (Hoerner 1950; Mair 1965; Kentfield
1984; Rodriguez 1991). The reduction of the drag specifically associated with
tractor-trailers has also been under investigation for many years. In the early
1970's, NASA, as well as private firms, extensively examined the concept of trailer
drag by modifying the flow over the forward part of the vehicle (Montoya and
250 J.D. Coon and K.D. Visser
Steers 1974) Their results indicated that commercially available devices could
reduce drag by up to 24 percent.
Sovram et a!. (1978) reported that Mason and Beebe (1978) examined several
add-on devices to increase the base pressure of a tractor-trailer in 1976. They
included vertical and horizontal splitter plates, guide vanes, and non-ventilated
cavities as illustrated in Figure I a. The vertical splitter plates had little or no effect
on the tractor-trailer drag, while the vanes exhibited an adverse effect. The only
device that showed positive results was the non-ventilated cavity design, shown in
the bottom right of Figure I a. Mason and Beebe varied the length of the plates and
found an optimum of 22 inches (0.55 m) for a conventional48 foot (14.6 m) trailer
which yielded an overall drag reduction of 5 percent.
Splitters
Vanes Cavities
(a) (b)
Fig. 1. Drag reduction concepts a) Mason and Beebe in Sovram (1978) b) Bilanin (1985)
Similar research by Hucho (1987) on a small minivan style vehicle in the 1970's
substantiated the drag reduction behavior of non-ventilated cavities. Cooper (1985)
conducted extensive tests on detailed scaled models of a tractor trailer and a panel
truck using rounded and beveled panels on the rear end of the vehicle. His results
indicated an optimum reduction in the drag of 7-10% of the baseline model drag
and indicated it was not unreasonable to expect this at full scale. A patent was filed
by Bilanin (1985) on a variation of the cavity design concept, illustrated in Figure
I b, that inset the endplates from the trailer perimeter. His claims indicated a drag
reduction on the order of 10%. Recently patents filed by Boivin and Roberge
(2001) describe an attachment of plates to the edge of the trailer and inclination
toward the centerline, but with no bottom plate.
NASA has extensively supported testing of devices aimed to reduce the pressure
drag. A summary of the NASA work from 1973-1982 by Saltzman and Meyer
(1999) indicated that a reduction in base drag, such as by using a truncated boat
tail, will be necessary in light of the increase of base drag associated with forebody
streamlining. In 1988, full-scale tests were performed at NASA's Ames Research
Center on a full scale, inset, non-ventilated cavity device (Lanser et a!. 1991)
similar to Bilanin (1985). Inset, as well as overall plate length, were optimized.
The results showed that the highest drag reduction came with a plate length of
0.36w and an inset on the top and side plates of 0.06w , where w was the width of
Drag Reduction of a Tractor-Trailer Using Planar Boat Tail Plates 251
Numerical Simulation
Although the primary focus of this study was experimental, an effort to gain some
qualitative insight into the effect of endplates, and to provide direction for
experimental tests, was first undertaken using the numerical code FLUENT. The
use of FLUENT is questionable when attempting to estimate drag numbers or flow
structures on such an unsteady, asymmetric, separated bluff body flow, however
the purpose was a quick and "dirty" look to see if there was some time averaged
improvement on the aft trailer face. A simplified tractor-trailer geometry was
modeled three dimensionally in a large, 50ft x 50ft x 150ft, (15.2 m x 15.2 m x 45.7
m) wind tunnel as depicted in Figure 2. A total of 2.64xl0 5 nodes were used at a
Reynolds number of 4.32x 106 based on truck width. The grid was not refined in the
present study as the number of nodes were as many as the available computer
hardware could accommodate. The model was run under steady state conditions
only and utilized a k-£ turbulence model. The purpose of the numerical model was
252 J.D. Coon and K.D. Visser
simply to see if there was any pressure increase on the back of the trailer. Cases
were computed with and without plates attached perpendicular to the back of the
trailer. The plates were inset on each side an equal distance, d, from the trailer
edge for a range 0<d<36 inches (0 < d 0.91 m) full-scale in 6 inch (0.15 m)
increments. Plate length, L, was varied from 3-6ft (0.91- 1.83 m) in 1 ft (0.3 m)
increments.
(a)
' \
·v--~ ~
I
I
I
r ~ t., 1
-t ~ -
f-
r · --'"
~ ,jl
fill.l.
I
(b)
(a) (b)
Fig. 3. Numerical pressure distributions on aft face of trailer a) Baseline b) Boat tail plates
added, d = 6 inches (0.15m), L = 48 inches (1.22 m)
Experimental Study
An extensive set of wind tunnel tests were conducted on a scale model of a tractor-
trailer in the Clarkson University Low Speed Wind Tunnel. The tunnel facility is
an in-draft, open circuit type with a 4. 7: 1 contraction inlet. The test section is of 48
inch by 36 inch (1.22 m by 0.91 m) cross section with a length of 65 inch (1.65 m).
Honeycomb and fine screens are mounted across the entrance to the contraction
inlet to reduce the turbulence and flow irregularities. The maximum speed
obtainable in the tunnel was approximately 58 miles/hour (26 m/s) which
corresponds to a Re/m of about 1.5e06/m.
Four different plate cavity models were tested. The first, denoted Equal Inset
(EI), is illustrated in Figure 5a and was similar to the numerical geometries. The
second, Equal Inset, Zero Bottom Plate Inset (EI-OB), was identical to EI, but with
the bottom plate at zero inset, as suggested by previous literature (Bilanin 1985;
Lanser et al. 1991). The last two model types tested were identical to the previous
two except that the top plate was removed, and were referred to as EI-NT and EI-
OB-NT, respectively. Figure 5b illustrates an EI-OB-NT geometry. Full-scale
issues fueled the interest in top plate removal. If viable from a drag perspective, it
would be more practical as a design as it incorporates less material and would be
far easier to operate than a four-plate design. The geometries test matrix is listed in
Table 1.
(a)
Fig. 5. Wind tunnel plate cavity geometries a) Equal Inset (EI) b) Equal Inset, Zero Bottom No
Top (EI-OB-NT)
A model of the Boivin and Roberge (2001) design was also constructed,
containing three plates (top and sides) inclined at an angle towards the centerline
with no inset from the perimeter of the trailer. An estimated plate length of 18
inches (0.46 m), full-scale, and angle of 15° , based on photographs, was modeled.
A bottom plate was also added to the original design, perpendicular to the trailer
face, as an additional configuration.
A biaxial, drag and side force, balance was designed and built to take
measurements on the tractor-trailer model. It uses two IKO International linear
translators with crossed roller bearings (CRWU 80-125) mounted perpendicular to
each other. They enabled motion in the direction of flow and perpendicular to the
flow. Two Precision Transducers model PTlOOO load cells, were used to take
measurements for drag and side force on the truck model. Both have an instrument
error of 0.03% of the applied load. Voltage output data from the load cells was
signal conditioned with a National Instruments (NI) Strain Gauge Board Model
SG-2043 and acquired with aNI data acquisition (DAQ) card model PCI-6024E.
(1)
(2)
and
_
C D- Drag (3)
Dynamic pressure* Frontal area
The dynamic pressure is one half of the tunnel air density multiplied by the tunnel
velocity squared.
The experimental results for 0° yawed geometries will be presented first
followed by the yawed conditions. The method of Kline (1985) was used to
estimate the uncertainty in C0 , ~C 0 and %~C 0 . Although the typical uncertainty
error in C 0 was± 0.015 or approximately 2% of the measured value, the percentage
of error associated with ~C 0 and %~C 0 can be significantly higher depending on
the actual drag savings of the device as the uncertainty in ~C 0 can represent a
significant portion of the drag savings increment. Consider the uncertainty in ~C 0 ,
UtJ.c 0 , which is determined as:
(4)
256 J.D. Coon and K.D. Visser
(5)
U11.c 0
% error in ~C = 100-- (6)
D !'!Co
The same is true for %~C 0 • As a result, the % error in ~C 0 , typically on the
order of± 0.02, becomes about 28% for an incremental drag savings of 10% over a
baseline C 0 of, say, 0.8. A drag savings of only 5% would incur % error in ~Co of
over 55%. Percent error magnitude in %~C 0 is similar.
It is important that this error associated with these data be kept in mind when
observing the trends in the following plots. Although many repeat cases were
taken and data was acquired over significant time periods to ensure time averaged
results, consideration of the error bars obscures many of the smaller trends
observed between individual geometric details. Thus, despite the fact that the
overall drag savings of these devices, relative to the baseline geometry, were
statistically significant, the specifics of the data presented below associated with
individual geometric differences should be taken only as indicative of potential
trends and not definitive conclusions. Note that the error bars have been left off the
figures for clarity.
Equal Inset
The behavior of the EI case is plotted in Figure 6a. The data indicated an optimum
geometric range of plate length, L, of 45in. to 55in. (I. 14m to 1.40 m) and inset, d,
of 3in. to 7in. (0.076m to 0.18m) full-scale. For comparison, the values of %~C 0
are presented in Figure 6b to illustrate the danger of only presenting percentages of
drag reduction instead of ~C 0 . Since the values of %~C0 are scaled by the
baseline, an unrealistic value of the magnitude of drag of the baseline geometry can
lead to erroneous conclusions. This is especially critical at model scales and
Reynolds numbers. As an example, note the difference in perceived %~C 0
performance of the 3 foot (0.91 m) case at an inset of 4 inches (0.10 m) compared
to the values of ~C 0 . Also, the incremental values of ~C 0 can more readily be
applied to a full scale geometry, as percentages require the baseline model C0 . The
maximum %~C 0 was achieved with a geometric configuration of L=48in. (1.22 m)
and d=5.72in. (0.14 m), yielding an 8.84% reduction in drag, however in terms of~
C0 , the L = 36 inch (0.91 m), d = 3.8 inch (0.097 m) geometry performed as well as
the 48 inch (1.22 m) case.
Drag Reduction of a Tractor-Trailer Using Planar Boat Tail Plates 257
. '•
Full Scale Inset ( 1nt=-f'oe$l
.. . ,,
Ful Scale ktset (lncbttl
(a) (b)
Fig. 6. Effect of plate length and inset on drag reduction, equal inset geometry a) ~C 0 b)
%~Co
"'
~/
.... ~ -
"'
7
u
~
'" ~""'...
... FIJI $Q6t PI.J.l& llll"'gll'l
J- s '"' Otloaom
1::'..,-:1
)IMI~
,., ,,
0 e 1!1
(a) (b)
Fig. 7. Effect of plate length and inset on drag reduction, zero bottom geometry a) ~C 0 b)
%~Co
258 J.D. Coon and K.D. Visser
u
<l
0 .04 t----"=---o..,-;""F:t,.--------=----~
0 .01
10 12 14 16 18
Fig. 8. Effect of no top plate, plate length and inset on drag reduction of 4 foot ( 1.22 m) plate
length
model configurations from the zero degree tests from -3° to +9°, with an increment
of3°.
For all perpendicular plate configurations tested, and at low angles, the behavior
of the drag with yaw angle was concave down . In other words, the ilC 0
improvement decreased with increasing yaw angle. The device with maximum
ilC 0 from the zero degree tests, the EI-OB, L=4ft. (1.22 m) , d=5.72in. (0.15 m) is
illustrated in Figure 9. For all configurations tested with 4 foot ( 1.22 m) plate
length, drag benefits decreased with yaw angle. Erratic behavior was observed at
higher yaw angles for the model with L=3 foot (0.91 m). This may be attributed to
the effect of less area exposed to the flow with a shorter plate length L. Removal
of the top plate had a consistent, negative offset on drag reduction over the yaw
angle range, similar to that of the oo yaw data of Figure 8.
Yaw tests were also conducted on variations of the Boivin and Roberge (2001)
design. As previously mentioned, this design does not include a plate inset for any
of the three plates which were placed at a 15° angle with respect to the trailer top
and sides. Contrary to the perpendicular plate behavior, at low yaw angles the
trend is somewhat concave up. That is, ilC 0 increases with increasing yaw angle
over the range :t3 o. The behavior is similar for the inverted case which even
indicated better performance. The addition of an orthogonal bottom plate to the
design improved the drag yet further, but the performance remained below the 4
foot (1.22 m) optimum at 0° yaw.
It should be noted that the curves are not ideally symmetric about a yaw angle of
zero degrees. Two possible explanations can be given for this behavior. First,
although the centerline of the model was aligned with the tunnel free-stream, the
addition of the aft devices could be cause for a non-symmetric geometry. Each aft
device was carefully hand-built and attached with magnetized strips to the trailer,
however the impact of small scale construction and alignment errors at this scale
could be more serious than it was thought to be, certainly more so with a tunnel
model than the full scale device. Secondly, the asymmetries present at +/- 3
degrees can be seen to fall with in the estimated error bands of a ilC0 of about+/-
0.02, indicating that there may be no asymmetry present at all.
0.1 , . . - - - - - , - - - - - - - - - - -- - ,
3
Yaw Angle (Degrees)
At higher yaw degrees, the results became more erratic and somewhat
unexplainable, possibly due to measurement uncertainty, as seen in Figure 9. At
+6°, the inverted case continues its convex behavior (as experienced over ±3°).
260 J.D. Coon and K.D. Visser
This differs from both the original and 4-plate cases. The common element in the
original and 4-plate cases is the presence of a top plate. It is possible that a
negative effect from the top plate is occurring at +6°. This is inexplicable at the
present, however, since at +9° the trend is reversed. That is, at +9°, the 4-plate
angled outperforms both the original and the inverted Boivin and Roberge (2001)
design. It is suspected that the low Reynolds number effects are contributing to
this behavior in addition to measurement uncertainty. The general conclusion in
the yaw tests is that further data is necessary at more yaw angles and higher
Reynolds numbers to substantiate any claims made for yaw trends of the angled
plate design, or any variations.
Literature Comparison
The optimum geometrical results from the present experimental data were
compared to that of Bilanin (1985) and Lanser, et al. ( 1991) using the definitions of
Figure 10. Plate insets, G and D, and the plate length, L, were non-
dimensionalized by the trailer width, W.
The results of Lanser et al. (1991) indicate similar geometries to the present
study with identical insets of GIW=DIW=0.06. Plate length ratio differs, from
LIW=0.50 in the present study to LIW=0.36 for Lanser, et al, but recall, as
discussed earlier, that plate inset had a much stronger influence on drag reduction
than does the plate length.
Although it is tempting to speculate that an increase in the drag savings would
occur as Reynolds numbers increased into the full scale regime, other
investigations have indicated this is not always the case (Watkins et al. 1987;
Cooper 1985) provided the model scale Re is great enough A full-scale tractor-
trailer operates at aRe of about 4xl06 based on width of the trailer. Full-scale wind
tunnel tests have shown that the drag coefficient for tractor-trailers lies in the range
0.6 to 0.9 at typical road speeds (McCallen et al. 1999). It has been indicated (SAE
11252) that although theRe based on width need not match full scale for acceptable
results, it should be a minimum of 7x105 • Others have indicated a minimum of
1xl06 (Watkins et al. 1993) or 2xl06 based on the square root of frontal area
(Cooper 1985).
the standard extruded aluminum hinges were replaced with steel hinges.
Deployment required manual actuation of the two side flaps, both of which are
within reach of the truck operator from the ground
Fig. 11. Full scale prototype storage sequence a) open b) collapsing c) closed
A tensioning system was devised so that the cloth stayed taut and the side flaps
remained opened during operation. High-strength climbing rope was attached to
the flaps via eyelets on the top and bottom of each flap. A ratcheting tie-down
tensioner (a "come-along") pulled the flaps into the open position. In order to close
the device the tensioners were relaxed and the rope unhooked . A locking
mechanism latched the device to the driver side door when the doors were closed
and the trailer in motion. The flaps were secured in place when non-operational by
two keyed hasp locks.
Issues associated with fitting the device to the truck doors were alleviated by
purchasing doors from the local trucking company, La Valle Trucking Inc. (L TI),
who were going to road test the device, and installing the device on the doors
before the doors were put on the trailer. Originally, the prototype was to be
centered on the back face of the trailer, inset from all sides by about 7 inches (0.18
m). Due to complexities within the trailer doors, such as hinges and latches, the
prototype placement had to be changed. The bottom of the prototype framing was
inset about 20 inches (0.51 m), because of interference with the latches used to get
in and out of the trailer, and the sides were inset 10 inches (0.25 m). Figure 12
illustrates the installed prototype.
(a) (b)
Fig. 12. Full scale prototype configuration on test vehicle a) closed b) open
average speeds. The truck was fitted with an onboard fuel consumption and
condition monitoring computer. It should be noted that as this was a one-truck test
and not a standard SAE Type II test (SAE 11321 1981) the data collected could
have an appreciable bias due to the particular ambient atmospheric and road
conditions experienced during the trip.
The results from the test indicate that the addition of the device increased the
fuel mileage from 5.8 to 6.3 miles per gallon (mpg), or 2.5 to 2. 7 km/liter, ± 0.1 %,
for a savings of approximately 0.5 mpg (0.2 km/liter) This corresponds to an 8.3%
average increase in fuel mileage. Based on these values, an estimated savings for a
typical 120,000 miles (193,121 km) per year use would be on the order of 1500
gallons (5677 liters) per truck. Single trip and single truck bias without a control
issues aside, the preliminary data did indicate encouraging results in the right
direction. Observation of the device before the cross-country road test did indicate
an increase in pressure in the aft region of the trailer. The upper and lower canvas
surfaces bulged outwards from the stationary position, indicating a pressure
difference across the material.
A second generation device was constructed based on experience gained with
the first test. It utilized all aluminum rigid sides with a split plate design as
illustrated in Figure 13a. Unfortunately, uncertainty in results and a short 5500 mile
(8851 km) trip, combined with questionable data reporting led to inconclusive
results. In addition, geometry and fatigue issues, as illustrated in Figure 13b,
caused complications. Excessive vibration, according to the driver, was the cause
of the cracking along the hinges, breakage of a pin joint in the top flap, and
resulting geometry mismatch that led to increased cavity ventilation.
(b)
Fig. 13. Full scale aluminum prototype a) installed on trailer b) hinge fatigue
Conclusions
The wind tunnel and full scale data presented in this study verify results from the
literature that significant drag reduction can occur through the use of a cavity
device mounted on the aft face of a tractor-trailer rig. The best results were
obtained with a four foot (1.22 m) plate width yielding a drag savings of about
264 J.D. Coon and K.D. Visser
C 0 =0.075 or about 9% of the baseline drag. It should be noted that because the
incremental drag is a difference of two substantially larger values, the error in AC 0
was appreciable.
All the wind tunnel results indicated a higher sensitivity for plate inset on drag
reduction than for plate length. Setting the bottom plate inset close to the bottom of
the trailer base appears to be essential to maximizing drag reduction. The optimum
experimental plate inset, d, was found to agree very closely to that of Lanser, et al.
(1991). Removal of the top plate decreased performance by about 30% in all cases,
but there still was an observable drag reduction and this may be a key factor in
realizing the practical role of such a device on a full-scale level.
Device performance decreased with yaw angle for plates mounted perpendicular
to the trailer base. Each of the angled plate design variations, however, displayed a
concave up behavior at low yaw angles, contrary to the concave down behavior of
the perpendicular plate-cavity devices. At small yaw angles, the four-plate angled
plate device achieved the highest drag reduction, indicating the importance of the
presence of the bottom plate to maximize performance.
A full-scale, fully operational prototype was road tested and indicated that the
addition of such a cavity device increased the fuel mileage by 0.5 mpg (0.2
km/liter) or about 8%. Estimated savings for a typical 120,000 miles (193,121 km)
per year use would be on the order of 1500 gallons (5677 liters) per truck
The positive results obtained from the scale and full size tests have prompted a
continued study of the potential use of a cavity design for drag reduction, especially
the possibility of a design employing only three sides. A new series of models is
being constructed to optimize length, inset, and boat tail angle. Yaw behavior,
optimization of three plate geometries, and downstream effects obtained with seven
hole probe flow field surveys will be conducted. In addition, two more full scale
prototypes are being designed with the aim of improving functionality in addition
to obtaining full scale data.
Acknowledgements
The authors would like to thank the National Collegiate Inventors and Innovators
Alliance (NCIIA) for their support of this project through an E-Team grant and the
New York State Energy Research & Development Authority (NYSERDA) Grant
Agreement #6436 for their support. The contributions by the local trucking firm,
La Valle Trucking Inc of Potsdam, NY, of time and labor on the road are also much
appreciated. Finally, a big thank you to the enthusiastic undergraduate Clarkson
students who participated in this project: Anna Sawabini, Matthew Pausley,
Matthew Menotti, Jamison Coon, and Matthew Allen.
References
Bilanin AJ (1985) Vehicle Drag Reducer. US Patent 4682808
Boivin and Roberge (2001) Vehicle Drag Reducer. US Patent Number 6257654
Bureau of Transportation (2000) National Transportation Statistics
Coon JD (2002) The Effects of Non-Ventilated Plate-Cavity Devices on Drag Reduction of
Tractor Trailers. Masters Thesis, Clarkson University, Potsdam NY.
Drag Reduction of a Tractor-Trailer Using Planar Boat Tail Plates 265
Cooper K (1982) The Wind Tunnel Testing of Heavy Trucks to Reduce Fuel Consumption.
SAE 821285
Cooper K (1985) The Effect of Front Edge Rounding and Rear-Edge Shaping on the
Aerodynamic Drag of Bluff Vehicles in Ground Proximity, SAE 850288
Cooper KR, Mason WT, Bettes WH (1982) Correlation Experience with the SAE Wind
Tunnel Test Procedures for Trucks and Buses. SAE 820375
Davidson L, Krajnovic S (2001) Large Eddy Simulation of Flow around a Ground Vehicle
Body AIAA 2001-0 1-0702
Hoerner SF (1950) Base Drag and Thick Trailing Edges. J Aeronautical Sciences:622-628
Hucho W (1987) Aerodynamics of Road Vehicles
Hucho W, Sovran G (1993) Aerodynamics of Road Vehicles. Annual Review of Fluid
Mechanics
Kentfield, JAC (1984) Short, Multi-Step, Afterbody Fairings. J Aircraft: Vol.21 No. 5
pp.351-2
Kline SJ (1985) The Purposes of Uncertainty Analysis. J. of Fluids Eng: 153- 160
Lanser WR, Ross JC, Kaufman AE (1991) Aerodynamic Performance of a Drag Reduction
Device on a Full-Scale Tractor/Trailer. SAE 912195.
Mair WA (1965) The Effect of a Rear-Mounted Disc on the Drag of a Blunt-Based Body of
Revolution. The Aeronautical Quarterly Royal Aeronautical Society:350-360
Mason Jr WT, Beebe PS, Schenkel FK (1973) An Aerodynamic Test Facility for Scale-
Model Automobiles International Automotive Engineering Congress
Mason Jr WT, Beebe PS ( 1978) The Drag Related Flow Field Characteristics of Trucks and
Buses, Aerodynamic Drag Mechanisms of Bluff Bodies and Road Vehicles, General
Motors Research Laboratories, pp. 45-93
McCallen R, Couch R, Hsu J, Browand F, Hammache M, Leonard A, Brady M, Salari K,
Rutledge W, Ross J, Storms B, Heineck JT, Driver D, Bell J, Zilliac G (1999) Progress
in Reducing Aerodynamic Drag for a Higher Efficiency of Heavy Duty Trucks (Class
7-8). SAE 1999-01-2238.
Montoya LC, Steers LL (1974) Aerodynamic drag reduction tests on a full-scale tractor
trailer combination with several add-on devices. NASA TM X-56028
Rae WH, Pope A (1984) Low Speed Wind Tunnel Testing
Rodriguez 0. (1991) Base Drag Reduction by Control of the Three-dimensional Unsteady
Vortical Structures;. Experiments in Fluids Vol. 11 p218-226
SAE J1252 (1981) SAE Wind Tunnel Test Procedures for Trucks and Buses
SAE J1321 (1981) Joint ATA/SAE Fuel Consumption Test Procedure, Type-11
Saltzman EJ, Meyer RR ( 1999) A Reassessment of Heavy Duty Truck Aerodynamic Design
Features and Priorities. NASA/TP-1999-206574
Saunders JW, Watkins S, Hoffmann PH, Buckley FT (1985) Comparison of On-Road and
Wind Tunnel Tests for Tractor Trailer Aerodynamic Devices, and Fuel Saving
Predictions. SAE 850286
Sovran G, Morel T, Mason W (1978) Aerodynamic Drag Mechanisms of Bluff Bodies and
Road Vehicles, General Motors Research Laboratories
Watkins S, Saunders JW, Hoffmann PH (1993) Comparison of Road and Wind Tunnel Drag
Reductions for Commercial Vehicles. J. of Wind and Industrial Aerodynamics: 411-
420
Watkins S, Saunders, JW, Hoffmann PH (1987) Wind Tunnel Modeling of Commercial
Vehicle Drag Reducing Devices: Three Case Studies. SAE 870717
RANS Simulations of Passive and Active Drag
Reduction Devices for a Road Vehicle
1 Introduction
Several drag reduction devices have been studied in the literature with ap-
plication to bluff bodies [I]; in most cases they correspond to modifications
of the basic geometry and aim to generate a more streamlined body (pas-
sive devices), thus limiting viscous losses associated with separated flows and
pressure losses corresponding to incomplete recovery in the wake. In addition
to change the baseline geometry, several investigators have studied the effect
of appendices added to the body base [2]: plates, cavities, boattails, etc. are
well known as possible means to alter (and eventually eliminate) the large
low-pressure recirculating bubble in the near wake of truncated bodies.
In contrast to passive devices, actuators and suction or blowing (active
devices) modify the flow structures in the near wall region affecting mainly
the viscous contribution to the total drag. This is generally negligible with
respect to the pressure drag for bluff bodies. Strong blowing (base bleed [3]),
on the other hand, modifies the wake thus impacting directly the pressure
recovery. Sophisticated blowing systems use a jet parallel to a curved surface
(Coanda jet) to eliminate completely the base recirculation [4].
268 G. Iaccarino et al.
In this work, a CFD study of passive and active drag reduction devices is
carried out. The baseline configuration was provided by GJ\I [5] and analyzed
previously using Large Eddy Simulation [6]. In the present work two modified
configurations with base appendices (plates and boattail) are considered and
compared. In addition, a new geometry with active drag control by Coanda
jets, has been designed and analyzed.
The present investigation is a complement of the CFD work reported in
[6]; steady and unsteady three-dimensional RANS simulations are carried out
using the Immersed Boundary Technique [7] and the v 2 - f turbulence model
[8] for every geometry.
2 Numerical Method
3 Simulations
Three-dimensional steady and unsteady simulations are carried out with the
objective of measuring the performance of different drag reduction devices
applied to a baseline configuration representative of a truck or a van [5].
The results are presented in two separate sections; the first one is related
to the passive devices, the second to the investigation of a configuration with a
Coanda jet. Numerical results are reported in terms of velocity profiles in the
wake and drag data. Time averaged URANS results are presented and com-
pared to steady-state RANS and LES predictions for the baseline geometry.
For all the other configurations only time-averaged URANS results are pre-
sented. During the calculations time signals of the velocity at eight locations
270 G. Iaccarino et al.
in the wake are recorded and analyzed using Fourier transforms to evaluate
the dynamics of the wake. These signals provide indications on the base bub-
ble pumping (movement in the streamwise direction) and of unsteadiness in
the vertical and horizontal directions respectively (wake flapping).
Three geometries are considered for the study of passive devices for drag
reduction, Fig. 1. The first one is the baseline geometry used previously [5]
and [6], hereafter called Square-Back (SB). It is a hexahedron 360mm long,
140mm wide and 100mm high, with the front edges rounded and a truncated
tail. The second geometry has an appendix on the tail consisting of a solid
protrusion of 50mm and a tapered angle of 9 degrees. It will be referred to in
the following as Boat-Tail (BT). The third geometry has four plates on the
tail each 8mm thick and 50mm high forming an open cavity; this geometry
will be referred to as Cavity (CT).
In the experiments conducted by GM the models were placed in a wind
tunnel1700mm long, 710mm wide and 377mm high and mounted on a stream-
lined support with a ground/body clearance of 20 mm. In the simulations, all
the models are positioned as in the experiment with the leading edges located
at 270 mm from the flow inlet, 20 mm of clearance from the floor and centered
with respect to the side walls of the wind tunnel.
The use of the immersed boundary technique has reduced the overall time
required for the grid generation process. For the SB geometry, three struc-
tured volume grids have been generated ranging from 250,000 to 1,150,000
cells. Strong clustering has been used to obtain a good representation of the
boundary layers with an adequate clustering of the grid cells near the walls
(y+ < 10 everywhere). Grid convergence was achieved with a grid size of about
650,000 cells. All the results presented are obtained using this grid resolution.
The Reynolds number is 1.25 million based on the inflow velocity (50 m/s)
and the SB model length (360mm); a very low inlet turbulence intensity has
been used.
At the beginning of the simulation, a very large time step is employed
leading to a steady solution; later the time step is decreased (to .L\t = 0.001
seconds), and a time accurate simulation is carried out. Coherent periodic
wake motion is obtained after 1 seconds for the SB configuration; the averaged
solution is then computed gathering results in a period of .5 seconds.
The initial work was focused on the baseline geometry (SB) because de-
tailed measurements of the near wake region are available. Velocity profiles
are reported in Figure 2 and 3 in a horizontal and vertical plane respectively.
The time averaged RANS predictions are compared to a steady RANS solu-
tion and to the LES data reported in [6]. The URANS represent a substantial
improvement with respect to the steady RANS results even if the accuracy
especially very close to the base of the model is still limited compared to the
LES data.
RANS Simulations of Drag Reduction Devices 271
The predictions for the two passive devices (BT and CT) are reported
in Fig. 4. The CT configuration appears to be very similar to the baseline
whereas the boattail alters the intensity of the recirculation substantially.
This is reflected in a better pressure recovery and a lower drag (see Table 1).
The analysis of the unsteady velocity signals in the wake shows that an-
other important effect of the boattail is the strong reduction of the wake
motion. The CT on the other hand provides only a limited damping of the
unsteadiness.
A Coanda blowing device has been designed starting from the original SB
geometry following the process described in [4].
In Fig. 5, a sketch of the Coanda device is reported. The final part of the
original SB has been shelled with a thickness of 2.25mm, a width of 67mm and
a tapered angle of 5 degrees. The rounded base plate has a radius of 15 mm
to satisfy the relationship discussed in [11] and [4]. The model is positioned
as before with the leading edge of the body located at 270mm from the flow
inlet, 20mm from the floor and centered with respect to the side wall of the
wind tunnel. As for the previous configurations the geometry description is
obtained via a surface triangulation with very small elements in the region
at the Coanda jet exit. The same flow conditions specified in the previous
sections have been used.
Several different jet velocities have been considered up to 150% of the
free-stream velocity; in what follows three different cases are presented and
compared to the original SB configuration (Fig. 6). The first case corresponds
to a no jet condition (V/Vi = 0) and the results are similar to the ones
obtained with the BT geometry (even if there is a slight reduction in drag
due to the presence of the rounded corners). By increasing the jet velocity the
wake modifications are substantial and, for the highest jet velocity investigated
(V/Vi = 1.5) there is no recirculation at the base of the body. This correspond
to a substantial drag reduction.
An interesting aspect of the Coanda jet simulations is that no unsteadiness
is present in the wake even for low jet velocities (V/Vi = 0.2 was the lowest
velocity investigated).
In Table 1, the drag coefficients obtained for the passive and active drag
reduction systems analyzed are reported; the forces are computed using mo-
mentum balance and non dimensionalized using the inlet dynamic pressure
and the cross section area. The method used does not allow generating drag
breakdown (friction, base, etc.) and therefore only global values are reported.
Additional work will be devoted in the future to obtain more detailed drag
analysis from the flow calculations.
272 G. Iaccarino et a!.
4 Conclusions
A computational analysis of passive and active drag reduction devices for road
vehicles is presented. The numerical procedure used is based on the Immersed
Boundary technique and the v 2 - f turbulence model. Three configurations
have been compared to the baseline SB geometry; the first is obtained by
adding a boattail at the base, the second by appending four plates forming
a base CT and the third by including a Coanda jet system and a rounded
base plate. Unsteady three-dimensional RANS simulations were carried out
to investigate the qualitative and quantitative differences between the config-
urations both in terms of global drag coefficients and in terms of wake dy-
namics. A suppression or damping of the unsteadiness is the common element
of all the devices studied. The baseline configuration exhibits strong three-
dimensional flapping of the wake (the horizontal motion being stronger than
the vertical because of the presence of the ground) and the main shedding
frequency captured is in good agreement with available experimental data.
The BT configuration does not show a strong coherent wake motion and the
recirculating region at the base is shortened and weakened; this results in 15%
reduction of the overall drag with respect to the baseline geometry. On the
other hand, the CT shows only a 3% drag reduction and the wake structure
is very similar to the SB case; the unsteadiness is somewhat reduced by the
presence of the stagnating fluid within the cavity. The most effective drag
reduction system investigated is the one based on the Coanda jets; a very
careful design is required to force the jet to stay attached to the base surface.
The low-speed flow in the near-wake is completely eliminated and the pressure
recovery is extremely fast with a jet velocity larger than the inlet velocity; no
unsteadiness is observed.
References
1. Hucho, W. H. (1998) Aerodynamics of Road Vehicles, SAE SP-1441 Interna-
tional, Fourth Edition, Warrendale, PA
2. Mason, W. T., Beebe, P. S. (1978) The Drag Related Characteristics of Trucks
and Buses, Aerodynamic Drag Mechanism, Plenum Press
3. Bearman, P. W. (1967) Aero Quart 18:207-224
4. Freund, J. B., Mungal, M. G. (1994) J Aircraft 31:572-578
RANS Simulations of Drag Reduction Devices 273
Experiments
LES
RANS
URANS
5. Han, T., Sumantran, V., Harris, C., Kuzmanov, T., Huebler, M. Zak, T. (1996)
SAE Paper 960678
6. Verzicco, R. , Fatica, M. , Iaccarino, G., Moin, P. Khalighi B. (2002) AIAA J.
40:2447- 2455
7. Fadlun, E . A., Verzicco, R., Orlandi, P ., Mohd-Yusof, J. (2000) J Comp Physics
161:35-43
8. Durbin, P. A. (1995) AIAA J 33:3241- 3247
9. O'Rourke, P . (1998) Computational Geometry inC, Cambridge Press
10. Duell, E. G., George, A. R. , (1999) Experimental Study of a Ground Vehicle
Body Unsteady Near Wake. In: Vehicle Aerodynamics and Wind Noise, SAE
SP-1441 International, Fourth Edition, Warrendale, PA
11. Mahesh, K. (1989) Private Communication
274 G. Iaccarino et a!.
.0.5
Experiments
LES
RANS
URANS
Fig. 3. Velocity profiles in a vertical plane in the wake of the Square-Back geometry.
.0.5
Square Back
Cavity
BoatTail
Fig. 4. Velocity profiles in a horizontal plane m the wake. Passive drag reduction
devices.
RANS Simulations of Drag Reduction Devices 275
67mm
2.25mm
o SquareBack
Coamda Jet- VN; = 0
Coamda Jet- VN; = 0.5
Coamda Jet- VN; = 1.5
Fig. 6. Velocity profiles in a horizontal plane in the wake. Coanda jet device.
Pneumatic Heavy Vehicle Aerodynamic Drag
Reduction, Safety Enhancement, and Performance
Improvement
Abstract
Under contract to the US Department of Energy, Georgia Tech Research Institute
(GTRI) has developed and applied blown aircraft aerodynamic technology to en-
train separated flowfields, significantly reduce drag, and increase the fuel economy
of Heavy Vehicles. These aerodynamic improvements also lead to increases in sta-
bility and control, braking, and traction, thus enhancing safety of operation. GTRI
wind-tunnel model results on test Heavy Vehicle (HV) models demonstrated drag
coefficient reductions of 50% using only 1 psig blowing pressure in the plenums,
and over 80% drag reductions if additional blowing air were available. Addition-
ally, an increase in drag force for braking was produced by blowing different slots.
Lift coefficient was increased for tire rolling resistance reduction, while down force
could be produced for traction increase. Also, side force and yawing moment were
generated on either side of the vehicle, and directional stability was restored by
blowing the appropriate side slot. These experimental data confirmed the elimina-
tion of directional instability caused by side-winds.
The above model data formed the basis for the design and modification of a full-
scale test vehicle by prototype shop Novatek, Inc. and GTRI. Initial confirmation
road test results are presented for this patented concept applied to an HV test rig
supplied by team members Volvo Trucks of North America and Great Dane Trail-
ers. To verify fuel economy increase, anSAE Type-11 Fuel Economy test was con-
ducted at the Transportation Research Center test track in East Liberty, Ohio. Re-
sults presented in this paper include wind-tunnel data for both unblown and blown
configurations, full-scale blowing and fuel-economy data, comparisons to experi-
mental results of the smaller-scale blown Pneumatic Heavy Vehicle model, and
tunnel tests on a full-scale Pneumatic SUV.
278 R.J. Englar
Nomenclature
A Vehicle frontal area
b Vehicle width
c Vehicle total length, tractor+trailer+gap
cD Drag coefficient=Drag/(qA)
CL Lift coefficient=Lift/( qA)
eM Pitching moment coefficient about mid-length =Pitching moment/(qAc)
eN Yawing moment coefficient about mid-length =Yawing moment/(qAb)
cy Side Force coefficient=Side force/(qA)
C Jet momentum coefficient=mVj/(qA)
~
Introduction
Considerable interest has arisen recently in improving the aerodynamics of Heavy
Vehicles (HVs) as a means of improving their operating costs, performance and
safety. While devices that can reduce the HV's drag coefficient can significantly
improve the fuel economy, it is also desirable that additional capabilities result
from advanced aerodynamic devices. These could include increased stability (both
lateral and directional), reduction in side-wind sensitivity, reduction in splash and
spray, and improved aerodynamic braking. One could also include an aerodynamic
means to reduce tire rolling resistance as well. Any such devices being considered
in these applications should also be simple and robust, contain few or no moving
parts, not be hampered by weather, and not increase vehicle weight or external di-
mensions. This paper presents pneumatic aerodynamic devices currently under de-
velopment at GTRI under contract to the DOE Office of Heavy Vehicle Technol-
ogy, which possess many if not all of these desirable characteristics. The paper
first describes the basics of pneumatic aerodynamics and its application to Heavy
Vehicles, and then provides details of the wind-tunnel programs, the full-scale pro-
grams, their results, and possible future applications.
GTRI researchers have been involved for a number of years in the development of
pneumatic (pressurized air blowing) concepts to yield efficient yet mechanically
simple means to control and augment or reduce the aerodynamic forces and mo-
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 279
Figure 1 shows the basic pneumatic concept, which has become known as Cir-
culation Control (CC) aerodynamics. Here, an airfoil's conventional mechanical
trailing edge device has been replaced with a fixed curved surface and a tangential
slot ejecting a jet sheet over that surface. That jet remains attached to the curved
surface by a balance between sub-ambient static pressure on the surface and cen-
trifugal force (the so-called Coanda Effect, Reference 3). This entrains the external
flowfield to follow the jet, and thus enhances the circulation around the airfoil and
the aerodynamic forces produced by it. The governing parameter is not angle of
attack, but rather the blowing momentum coefficient:
where m is the jet mass flow, Vi the isentropic jet velocity, S is a reference wing
area (or frontal area A for a ground vehicle configuration) and q is the freestream
dynamic pressure, 0.5 p V 2 (with p being the free-stream density, not the jet's). At
lower C~ values, augmentation of the aerodynamic lift by a factor of ~C 1 I C~ = 80
has been recorded (Ref. 3), representing an 8000% return on the invested momen-
tum (which in a physical sense is also equal to the jet thrust). Familiarity with
blown aerodynamic systems will remind the reader that this is quite extraordinary:
thrust-deflecting Vertical Take-Off and Landing (VTOL) aircraft are fortunate if
they recover anything near 100% of the engine thrust expended for vertical lift.
It is because of this high return, or conversely, because of very low required
blowing input and associated pumping power required to achieve a desired lift, that
Circulation Control airfoils appear very promising for a number of applications.
The A-6/CC Wing Short Take Off & Landing (STOL) flight demonstrator aircraft
(Figure 2 and Ref. I) showed the STOL performance listed, and also suggested ca-
pabilities very useful to ground vehicles: during short takeoff, it demonstrated high
280 R.J. Englar
lift and reduced drag, while in the approach/landing mode, very high lift with in-
creased drag was shown.
These advantages led to the application of this pneumatic concept to improve the
aerodynamics of an already streamlined car (Reference 4). The curved rear of the
vehicle and the resulting large jet turning angle are shown in Figure 3. Significant
but distinctly different trends were observed depending upon which portion of the
tangential slot was blown. Blowing the entire slot produced the large jet turning in
Figure 3, and drag increases of greater than 70%, showing potential for pneumatic
aerodynamic braking. Blowing only the outside corners of the slot weakened the
corner vortex rollup, lessened aft suction, and reduced drag by as much as 35%.
Blowing the aft slot also yielded a lift increase of 170%. One can envision a simi-
lar slot applied to the lower rear surface that could yield negative lift or positive
down force instead. This concept has been patented by GTRI and verified by a
similar installation on a model of a European Formula 1 race car (Reference 5).
Figure 5. GTRI 0.065-scale Baseline GTS Model, Full Open Gap, High Cab
Tests of this unblown model configuration did, however, demonstrate the impor-
tance of cab/trailer gap and fairing treatments. These configurations were tested in
the GTRI Model Test Facility research tunnel (Refs. 5 and 8) and showed some
significant drag reductions just due to changes in unblown geometry. Figure 6
shows drag reductions of up to 25% over a low-cab full-gap vehicle if the gap is
.. f V•7DI"'Cf''
'"
1114.
,..
I •
..
q. jN
..
11 Dr.t!!' , ._ Y;1" An~;lc. u::()O, Whee l' On /'
"
,,
1:o "
II
...
...
...'"
...
"' '
Y11"
II
f\n~c. "t· d"-'1!:
Figure 8. Jet Turning 011 all -t ' ide> of Blo" 11 Trailing dgc
. -6 .~ ,6·• ·6,·· -- ~
. 0 O!
• •~ 0 0 0 All 4 Slots Blown
• •
• Ooo,. o o oi o o 0 0
0.25 PSIQ ; .. .._. ~ .. ~ 1
' .. 10 p~og
03
J(
0 .5 pstg~ 0.7S ps q: • .. i
IIAII '----'--...J'--L.....-....J.....L...-..L....~..u..~~-'---.....!.......J-~-'-._;_..:.:_~-._1
~ - . ~·
() ()() 1111! IIIIJ IICif> IIIIX IIIII 11. 1! 11. 14 II. If> ll!ll
\1umcntum Coefficient. C 1-i
When only the top slot, the bottom slot, or both of these slots were blown in the
absence of the side jets, drag initially reduced slightly, but then significantly in-
creased with the addition of blowing. This represents an excellent aerodynamic
braking capability to supplement the hydraulic wheel brakes. Blowing efficiency is
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 285
plotted in Figure 10, where ~CD is an increment from the blowing-off value (nega-
tive ~CD is reduced drag). Absolute values of ~CDIC!l greater than 1.0 represent
greater than 100% return on the input blowing CW It is seen that the 4-slotted con-
figuration generates values as high as -5.50, representing 550% of the input blow-
ing momentum recovered as drag reduction. The figure also shows the opposite
trend as well, with up to 200% of the blowing momentum from top/bottom slots re-
covered as increased drag for braking.
However, should additional air become available from an onboard source such
as an existing turbocharger or an electric blower, additional drag reduction is possi-
ble, as shown in Figure 11. Drag coefficients of less than 0.30 are shown for faired
blown HV configurations. This is in the arena of streamlined sports cars. (The
drag coefficient of a 1999 Corvette coupe is C0 =0.29). Figure 11, origin-
"' Lm' C1h. :\n Gap. RnunLI LE.0 .37 ~ R Cu\:ul.u \r~.: 'Jt) 'II::..
q:: ll ~hp, t. \ '::7Hmph . tl~l . R~ ::o: :! ..S' IU ' ' 6
:u li •
II r ~lo1 Bot tOm 51 : Only
I ~ Top 6~g~t _~~~·
HI __ § __________-_-_-_.: ~·~#~
"·
~ '<>.. Z Side Slots Only
Ill 1----\
f ~
-------------gt --- ••
I ~
' 3 o0 •
~-• :V
.••
0
o
00
-~~ 0
•
0
0
; ; L---~--~--~--~--~~--~--~--~--~--~
Il l ~ 1112 U.U\) U UO ll tl' (UIO (IO \ O(Hl 0(1') .tJI.! --tJI5
&c 0 =Co-Coo
h=UOI IU7~ " R Circular \n: 4(1 30' I :'~TE,I E & Tr Blm~mg. \\'hc.."t'h un.
C..1h 'hu1lcr(i..1p Pl~lc~ ln,l.lllt..>t.l. ljr-l:l • u-<l
..
0.
·o
• o.
0
0
••• • ••t-
(1\
00 q 1 psr
o; \·~1 mrh
Re· 'b' ·o· ·o
0/
n 111 to
0
0
·o
"15
II :'II b
0
Ul~
0
. 0
010
II Ul u :' u~ (I ..I ~~~ 11.1> 117 II t)tJ
Moml'nlum Coefficient . C ~
Figure 11. Effect of Reynolds Number and Increased Blowing Values, plus Leading-edge
Blowing and Gap Plate Effects
ally intended to show that the drag curves tend to converge onto one slope inde-
pendent of Reynolds number, also shows a measured drag coefficient of 0.13 for
the Pneumatic Heavy Vehicle model with blowing. This is about half the drag co-
efficient value of the Corvette or the Honda Insight hybrid (C 0 =0.25). Even though
not achieved in the most efficient blowing operation range, this is an 84% drag re-
duction compared to the unblown baseline configuration. Note that the tractor cab
in Figure 11 has "gap plates" (or fairing extensions) instead of the full "No Gap"
fairing of Figure 9, and is thus much closer to an actual tractor/trailer configuration.
Figure 11 also shows the reduction of drag coefficient with increasing Reynolds
number at constant lower Cf..l values
It should again be mentioned, when comparing these data to other experiments
on similar GTS models being conducted by other researchers, that these GTRI data
above and below include simulated wheels, which as Figure 6 shows, should add
about llCo = 0.18 to the non-wheeled vehicles' Co values, perhaps more, depend-
ing on how well the tunnel ground effects are treated experimentally. GTRI's
measured data are generated using test section tangential floor blowing to eliminate
the floor boundary-layer interference, as discussed in References 3 and 8.
(It should be mentioned here that this yawing moment is measured about the rigid
model's midpoint at the ground level, whereas on a real articulated tractor-trailer, it
would be experienced at the tires of the individual units. However, comparisons of
blowing on and off are being made for the same single unit, and relative benefits
should still be valid). Blowing only one side slot can easily correct the yaw insta-
bility: with the nose straight ahead ('lfl=0°), blowing the left slot at Cfl = 0.06 yields
the equivalent opposite yawing moment (CN = +2.0). With the nose yawed left (for
example, 1fJ = -8°), slightly higher blowing (Cfl = 0.063) returns this unstable
yawing moment to CN = 0.0. Then, increasing the blowing a bit more can cause the
nose to yaw in the opposite direction, to the right. The opportunity for a non-
moving-part quick-response aerodynamic control system is apparent.
.r:i
25
.0
20
o' Blow1ng Produces Elllcess
I'
10
05 Ol r:::-~:
"
~)
0
-10 0 Yaw1ng Moment Tnmmed
0 by Blowing
0
-I ~ 0
,......
l.dl ·B L-___.~__._ _ __._._ _.__....__~_.._..__...____.~__,
000 O.Ol 0 .04 0.06 O.OM 010 0 ll 111• 016 0 IM O.lO
Momentum Coefficient, C Jl
stock (reference) Great Dane trailer. Blowing tufts turning around the trailing-edge
curved pneumatic surfaces are shown in Figure 14.
Tuning Tests
Test vehicle fabrication and assembly were completed at GTRI in February 2002.
The modified trailer was then picked up by team member Volvo Trucks of North
America and moved to their test facility in Greensboro, NC, where two initial
Tuning Tests were conducted (Reference 9) in February and May 2002, Figure 15.
Figure 16 shows a rear view of the pneumatic trailer with the tufts confirming on-
road flow turning. These tests verified the test equipment and data system opera-
tions, and indicated unofficial fuel economy increases on a highway, rather than on
a test track.
Figure 17. Test and Control Vehicles in Pits at TRC Test Track
The six sets of fuel economy runs were made at different blowing rates and at
zero blowing. This allowed reference comparisons to be made when the pneumatic
test trailer was re-configured into the baseline trailer, and tested to provide refer-
ence fuel economy for the standard vehicle (all fuel economy data achieved with
the other test configurations was compared to this one to determine percent fuel ef-
ficiency increase, %FEI). Figure 19 shows %FEI as a function of blowing coeffi-
cient, C!-l, at several vehicle test speeds. The %FEI improvements shown range
from 4% to 5% (5% to 6% if the l% error band is included) above the fuel econ-
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 291
omy of the baseline standard tractor-trailer, but these are seen to reduce somewhat
as blowing increases to values beyond C!-l = 0.02 - 0.03. Whereas responses heard
from typical HV users indicate this 5-6%FEI to be quite respectable, it is less than
we had anticipated based on our smaller-scale wind-tunnel tests (Refs. 5 and 8).
Figure 20 compares this data to the predicted fuel efficiency increase that we had
expected from the blown configurations. Whereas the lower blowing values were
nearing 70-80% of the expected values, at greater blowing the payoff was reduced.
The test team of GTRI, Novatek, and American Trucking Associations identified
suspected reasons for this, and we then conducted an experimental test program to
confirm these, as discussed below.
·,,
.,.___ --···-··· !jl
\ '=55 mph .
..,.1 r.; tnM R..ukl
\ =<li~ mph
q ffi /
~ ,-~:.l\1mum R.P\ t
,/' Lamll
. ~~--~--~--L-~~~--~--~--~~
0 001 00~ 0.03 U.o.l 00~ 006 0.07 0.08 009 0.1
Blowing Momentum Coefficient, C ~
Figure 19. Measured PHV Fuel Economy Increase, With 4 Trailer Slots Blowing
292 R.J. Englar
,,
"..
\\ u.S f ll11nt"lf{unl l A . - - .... ~
"
'n
\ II ~ . 111b Hlo,..-a •#
,. ~
,.
17
"·
",,
,, .• WT ~ GTRJ mali·<CUI< W,nd Tunn<l Tesc-.
(from f 1gu~ I J
%FEI TRCz Full--.cole Tmcl Te-.1ru TRC
"
II •
10
fR{ <4i
U L-~--~~--~--~~--~~~-L--~~
o 001 om om on. om 006 001 008 009 o1 o 11
Blowing Momentum Coefficient, C Jl
A new series of wind-tunnel runs was made on the 0.065-scale PHV model, where
we began with the "Best Blown Configuration" from previous tests and sequen-
tially made the changes suspected of being detrimental. It was the intent of this
new wind-tunnel program to determine if the geometric differences between the
full-scale test vehicle and the wind-tunnel model produced the aerodynamic and
fuel consumption differences discussed above. Figure 21 shows that as the con-
figurations approached the representation of the full-scale test vehicle (Run 239),
both blown and unblown drag increased. Reference 14 summarizes this data. Fig-
ure 22 compares the percentage drag reduction due to each configuration change,
while Figure 23 shows the predicted change in percent Fuel Efficiency Increase
(%FEI) due to each configuration. A comparison of Run 239 (model most like the
blown full-scale test vehicle) with Run 36 (most like the standard tractor-trailer ve-
hicle) shows that only a 2% FEI occurs for the unblown vehicle and only 7% for
the blown one. This confirms the trends of Figures 19 and 20, and explains the
causes of the less-than-expected test results. We have since conducted further tun-
nel testing to improve the final PHV configurations in anticipation of a second on-
road fuel economy test at TRC. Note from Figure 23 that if we converted the full-
scale PHV test vehicle to a blown configuration much more like the one in Run
205, we can expect Fuel Efficiency Increases of 16% unblown and 23% blown,
which will be very significant results.
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 293
0.7
• b..
··o~r:fl
. ··-····· ·-....... .
'\.:. .... 201) 101,
··a• ....----•------ ...... I
.. •. - ......... t__.--. .
0~ •~ ... ........
- ~
0.45
O.<IOOJlO
L...~~O.L
JY2~......;•0.-04
~:...-~0•.L06-'-~•0...J.Cli!
L...~......;O_.
.I0~~·0.-'-
12~~~0.L
. I-
4 ~~0...1.1 6
Momentum Cotf!l<lut, C j.l
Figure 21. GTRI MTF065 Wind-tunnel Test Results: Drag Change with Configuration
~5 .------------=-------------------------------------,
0 CD~(I'~o
50
Lll
!:i2ru;
{fo...rCol/Co. 0 r. where
Co•ef • ({, for R36. Unblown
All Conf09s except RZOS & R36
have Lower Surface Fair1ng
JO
Run Z12·R239
1.1
~ r---------------~----------------~--------,
0 ,f."'£1. .. • uo Rm 205•R207 "'•th No
~ ret c~.~..ooM lov.ef FHlflVS. No Cab Gape
f'lQw.; Bet.t ~ COtlhgurat.on
%FEI•0.5'1 OO'(COret · Co)/Coref• ~ 207-.U09 w.th
CD<ot ·Co for R36, Unblown Wedge Fa.nngs on Lower
Ah Cutout
All Configs except R205 &R36
have Lower Surface Fatrong
!0
'1FEI
10
gle could be changed. Blowing coefficient was variable, and mass flow, pressure
and jet velocities were measured to enable on-line calculation and setting of Cf.,l.
I~J(J
I~HI
IWU
,
·-.., Domestoc Ool Productoon
·..
\
, ...
...\
I
•.....r·.-*'.~
1(1()1)
'.. Gap
\lO.O
BGV
<()0 ..•
··-·· ' "•.•"-.......,
.,'7 _1-"---
(Class 1&2a)
lUll
Repon Oate ,. t.;
Figure 24. Highway Energy Usage Comparisons (Billions of Gallons per Year) by Vehicle
Type
Figure 25. GM Suburban Vehicle Undergoing Smoke Flow and Tuft Visualization
Testing in the Lockheed 16' x 23' Wind-Tunnel Test Section
296 R.J. Englar
Ct..., u;"i
l'ncuma11c S ·v. Lnd,hcc I_. \\. Tl~~5. 10 IS 02
It"' Yol\\ S\\t.:cp. q=l::!.!'-+ fl'L \ '=71 7 mph . Run _l
Ct,n' c-nt1unal G ,\1 Suhurhan u .:u
......
Co
.
u<
H ~~
"" -
"'
II III
/
Ull"i
-H Ill
-H IIIIi
n:•u
.
f1 ~41
-I II•
' lO..CRI).!hl
' U" \n~ot l l·. 'l'· dcJ,.:
Figure 26. Resulting Aero Forces and Moments as Functions of Yaw Angle for Baseline
Suburban
Flow visualizations taken with blowing activated on the pneumatic vehicle prior to
tunnel testing showed significant attachment of flow over the curved aft surfaces
and a contracting of the jet wake behind the vehicle. The wind-on, blowing-on data
showed different behaviors for different trailing edge configurations. Greater trail-
ing-edge turning-surface angle produced greater jet turning but also greater suction
of the trailing edges, adding a downstream suction force. The resulting drag gener-
ated is shown in Figure 27 for four different blowing configurations. Notice that
for some configurations initial drag reduction reaches a minimum point, followed
by drag increase at higher C~. This drag reduction at low blowing is on the order
of 3 to 4.15 times the input blowing coefficient, representing as much as a 415%
return on the jet momentum invested. Note also that increased blowing eventually
yields a rise in drag for some of the configurations, which represents an opportunity
for an aerodynamic braking system. What is needed, of course, is an onboard con-
trol system to switch from drag reduction to braking as needed by the driver. Note
also that the configuration with a 4SO turning surface on all exposed trailing edges
continued to reduce drag with increased blowing, although at a lesser rate of reduc-
tion. Also, the blowing-off drag coefficient for these non-optimized configurations
was the same as that of the stock reference Suburban tested earlier (Figure 26), in-
dicating no blowing-off drag penalty for installing this system on a typical SUV.
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 297
D
0.50 0 ..
o·
Run I t . I (I' ·\II ~ Suk'
OAS
-·~----.
0.+1 '--~--'-~-'-----'-~----''----'-~----'---'-~----'
0 0 01 0 .02 0.0.1 0()..1 0.05 0.06 007 008
Momentum Coefficient, C14
Right
008
0.07
0.06
Cyaw reqUired to offset ~·-· 1 0 unblown
005
O.Ql
001
0 ~------------------------------~
.0.01 L.....-....L-~...L..--.J......---.Jc.-.~-.L----J
It is to be noted from the above that we have not yet achieved the optimum con-
figuration to maximize drag reduction and yaw generation while requiring mini-
mum blowing input, but we have otherwise verified that blowing on SUVs can be a
powerful means to reduce or increase drag as needed, and to increase vehicle sta-
bility, all with no external moving parts.
Conclusions
Under DOE-sponsored research programs, GTRI and its teammates on the DOE
Pneumatic Heavy Vehicle project have completed experimental model investiga-
tions as well as full-scale road and track tests to confirm and improve upon the use
of pneumatic aerodynamics on these vehicles. These concepts will reduce aerody-
namic drag; favorably modify other aerodynamic characteristics; and thus improve
the performance, stability & control, handling qualities, safety, and economics of
both Heavy Vehicles and Sports Utility Vehicles. The GTRI team has confirmed
these capabilities by designing, fabricating, and testing small-scale PHV models in
3 separate wind-tunnel tests; and designing, fabricating and conducting 3 road or
track tests of a full-scale Pneumatic Heavy Vehicle demonstrator. We have also
conducted 2 full-scale wind-tunnel tests of the technology applied to a typical
Sports Utility Vehicle. The very favorable capability to control all aerodynamic
forces and moments was shown for the PHV and PSUV configurations, as was the
ability of a non-external-moving-part pneumatic control system to restore direc-
tional stability by eliminating unstable yawing moment and providing yaw in the
opposite direction. In a related application, GTRI is also currently developing a
patented aerodynamic heat exchanger that is based on these pneumatic principles
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 299
(Ref. 17). This device can further reduce the drag associated with the conventional
vertical radiator and related cooling system, while also adding favorable aerody-
namic characteristics to the vehicle.
The above test programs and analyses have confirmed the following capabilities
for pneumatic aerodynamics applied to Heavy Vehicles or Sports Utility Vehicles:
• Pneumatic devices, using one to four blowing slots and non-moving downstream
jet turning surfaces on HVs and SUVs, have reduced drag up to 84% in tunnel
tests. Recent tunnel tests of a PHV new configuration soon to be tested full-
scale have shown anticipated drag reductions of up to 46% with resulting Fuel
Efficiency Increase of approximately 23%. This is caused by prevention of
flow separation and increase in base pressure on the rear of the vehicle.
• Specific blowing on only some of the slots can cause a deliberate increase in
drag that can be used for instantaneous aerodynamic braking without the delay
time of moving parts, or the wear of mechanical parts.
• Specific blowing on only the top or bottom slots can cause a deliberate increase
in lift to reduce rolling resistance and thus increase fuel economy, or a deliber-
ate increase in downforce and thus a further increase in load on the wheels to
increase both traction and braking.
• Specific blowing on only one side slot can cause a deliberate increase in side
force and yawing moment to overcome the directional instability of these vehi-
cles caused by side winds and/or gusts.
• Because blowing is instantaneously switchable to whichever slot it is needed in,
these devices provide a very rapidly responding pneumatic control system with-
out external moving parts. Integrated with an onboard sensor and controls, it
can thus control all aerodynamic forces and moments acting on HV s and SUV s
and increase operational safety.
• Pneumatic reduction of the vehicle's wake should reduce splash and spray from
behind and its effect on following vehicles, and thus also increase safety of op-
eration.
• Low pressure blowing required could be supplied from onboard sources such as
a turbocharger or supercharger, or from an existing auxiliary engine such as is
currently used by HV refrigeration units.
• These pneumatic aerodynamic systems can be integrated with a patented
GTRI/Novatek aerodynamic heat exchanger to further increase fuel economy by
additional drag reduction and reduced radiator size requirements for cooling
(see Reference 17).
• Non-moving external components can yield an all-pneumatic system with very
small (if any) blowing-off component drag.
• Very small-size aft trailer extensions for blowing should incur no length limita-
tions
• For safety, stability and/or economy, positive use can be made of aerodynamic
forces and moments (lift, download, side force, yaw, roll) that are not currently
employed to influence Heavy Vehicle or SUV operation.
In the first 4 years of this DOE Pneumatic HV and SUV project, reported herein,
we have confirmed that these patented blown devices are able to control the aero-
300 R.J. Englar
dynamics of these vehicles and thus favorably affect their fuel consumption, their
performance, and their safety of operation. The importance of that is confirmed in
Figure 24. It appears that it is time to approach this fuel-consumption problem for
Heavy Vehicles, Light Trucks and SUV s in order to reduce the predicted deficit
shown between oil usage and supply. The concepts demonstrated by these pneu-
matic vehicle results suggest that certain favorable options are now available to do
so. In addition to Heavy Vehicle and SUV application, the above results appear
quite promising to other forms of automotive vehicles. Clearly, buses, motor
homes, and trains are also prone to large drag values and directional stability issues
due to aft flow separation and large vertical panels exposed to side winds. Of
course, the application of improved blown aerodynamics to increase the perform-
ance, traction, braking and handling of race cars is a very related and promising
subject.
Thus all the original objectives for this Pneumatic Heavy Vehicle program have
been experimentally confirmed: drag, lift, down force, side force and all corre-
sponding moments can be significantly augmented (increased or decreased as
needed) by blowing, and improved to the point where appreciable increases in
Heavy Vehicle and SUV performance, economy, stability and safety of operation
should result.
Recommendations
The above aerodynamic data confirm the Pneumatic Heavy Vehicle and Pneumatic
SUV as viable concepts for improving the aerodynamic performance, economy,
stability, handling and safety of operation of these vehicles. Data presented has ex-
ceeded the 35% drag reductions (previously demonstrated on streamlined cars) that
the American Trucking Associations claim will result in savings of more than 1.2
billion gallons of diesel fuel per year for the US heavy trucking industry, (as much
as 1.7 billion gallons per year can be saved at the 50% drag reduction level). The
following recommendations are made to suggest a meaningful continuation of this
program:
• Additional wind-tunnel evaluations should be conducted to even further reduce
the required blowing momentum, which needs to be acquired from some air
source on board the tractor-trailer rig or SUV. These tests might include slot
height variation, improved blowing surface geometry, alternate jet turning sur-
faces, pulsed blowing, or other innovative means.
• Continued feasibility studies are needed, where the above results are transferred
to the HV and SUV industries and interactions occur with tractor, trailer, and
SUV manufacturers, as well as with engine manufacturers, turbocharger builders,
or other possible air-supply specialties.
• Additional full-scale on-the-road demonstration and development of this technol-
ogy should continue, including further study of available air supplies and any as-
sociated penalties, plus further blowing configuration optimization.
Pneumatic Heavy Vehicle Aerodynamic Drag Reduction 301
Acknowledgements
The author wishes to acknowledge and thank Dr. Sidney Diamond, Dr. Jules Rout-
bart, Mr. Richard Wares, and Mr. Rogelio Sullivan of DOE for their continued
support and encouragement of this work, as well as Mr. Victor Suski for the con-
tinued very valuable involvement of the AT A. The technical assistance of Mr. Ken
Burdges of Novatek, Inc. in wind-tunnel model and road-test vehicle design and
fabrication is also greatly appreciated, as are experimental efforts of GTRI Co-op
students Graham Blaylock, Warren Lee, Chris Raabe, Erik Kabo, and Brian Corner
of the Georgia Tech School of Aerospace Engineering. Assistance of Skip Yeakel
of Volvo, Bob Dotson of TRC and Charlie Fetz of Great Dane is also much appre-
ciated.
Contact
Robert J. Englar, Principal Research Engineer
Georgia Tech Research Institute
Aerospace, Transportation & Advanced Systems Lab
Acoustics and Aerodynamics Branch
Atlanta, GA 30332-0844
(770) 528-3222
(770) 528-7586, Wind tunnel
(770) 528-7077, Fax
bob.englar@ gtri.gatech.edu
References
I. Englar, Robert J., "Development of the A-6/Circulation Control Wing Flight Demon-
strator Aircraft," DTNSRDC Report ASED-274, January 1979; and Englar, R. J,. eta!,
"Design of the Circulation Control Wing STOL Demonstrator Aircraft," AIAA paper
No. 79-1842, August 1979.
2. Englar, R. J. and C. A. Applegate, "Circulation Control-A Bibliography of DTNSRDC
Research and Selected Outside References (Jan 1969 to Dec 1983)," David Taylor Na-
val Ship Research and Development Center Report 84/052, Carderock, MD, Sept.,
1984.
3. Englar, R. J., "Circulation Control Aerodynamics: Blown Force and Moment Augmen-
tation and Modification; Past, Present and Future," AIAA Paper 2000-2541, June, 2000.
4. Englar, R. J., M. J. Smith, C. S. Niebur and S.D. Gregory, "Development of Pneumatic
Aerodynamic Concepts for Control of Lift, Drag, and Moments plus Lateral/Directional
Stability of Automotive Vehicles," SAE Paper 960673, Feb. 26-29, 1996. Also pub-
lished in SAE SP-1145, "Vehicle Aerodynamics: Wind Tunnels, CFD, Aeroacoustics,
and Ground Transportation Systems," pp. 27-38.
5. Englar, Robert J., "Development of Pneumatic Aerodynamic Devices to Improve the
Performance, Economy and Safety of Heavy Vehicles," SAE Paper 2000-01-2208, June
20,2000.
302 R.J. Englar
6. U.S. Department of Energy, Oak Ridge National Laboratory, "Development and Evalua-
tion of Pneumatic Aerodynamic Devices to Improve the Performance, Economics, Sta-
bility and Safety of Heavy Vehicles", Contract No. 450000555, December 1, 1998 -
December, 31, 2001.
7. Gutierrez, W. T., B. Hassan, R.H. Croll, and W.H. Rutledge, "Aerodynamics Overview
of the Ground Transportation Systems (GTS) Project for Heavy Vehicle Drag Reduc-
tion," SAE Paper 960906, February, 1996.
8. Englar, Robert J., "Advanced Aerodynamic Devices to Improve the Performance, Eco-
nomics, Handling and Safety of Heavy Vehicles," SAE Paper 2001-01-2072, May 14-
16, 2001.
9. Englar, Robert J., "Preliminary Results of GTRI/DOE Pneumatic Heavy Vehicle Tun-
ing Tests," GTRI Report A-5871, Mar 14, 2002.
10. Englar, R. J., "Preliminary Results of the Pneumatic Heavy Vehicle SAE Type-11 Fuel
Economy Test," GTRI Draft Report, September, 2002.
11. Englar, R. J., "Development and Evaluation Of Pneumatic Aerodynamic Devices To
Improve the Performance, Economics, Stability and Safety Of Heavy Vehicles," DOE
Quarterly Progress Report No. 14, April1, 2002 to June 30, 2002.
12. Dotson, Robert, "SAE 11321 Class-Eight Truck Aerodynamic and Tire Comparison
Fuel Economy Tests," Transportation Research Center report, Project 20020465, Sep-
tember, 2002.
13. Englar, R. J., "Development and Evaluation Of Pneumatic Aerodynamic Devices To
Improve the Performance, Economics, Stability and Safety Of Heavy Vehicles," DOE
Quarterly Progress Report No. 15, July 1, 2002 to September 30, 2002.
14. Englar, Robert J., "GTRI Updated Wind-Tunnel Investigation of Pneumatic Heavy Ve-
hicle Road-Test Configurations," GTRI Draft Report, Projects A-5871 and A-6395,
January 10, 2003.
15. "Transportation Energy Data Book: Edition 19," DOE/ORNL-6958, September, 1999.
16. "EIA Annual Energy Outlook 2000," DOE/EIA-0383(2000), December, 1999 and "AEO
2001", Appendix A Table 11.
17. Gaeta, R. G., R. J. Englar, and G. Blaylock, "Wind Tunnel Evaluations of an Aerody-
namic Heat Exchanger," published in Proceedings of the UEF Conference "The Aero-
dynamics of Heavy Vehicles: Trucks, Buses and Trains," Monterey-Pacific Grove, CA,
Dec. 2-6, 2002.
Base Flaps and Oscillatory Perturbations to
Decrease Base Drag
Abstract
The objective of this investigation is to study possible means for reducing the base
drag of a tractor-trailer. The experiments are conducted in the Dryden wind tunnel
at the USC Ground Vehicle Aerodynamics Laboratory. A roughly 1115 scale
model resembling a trailer is utilized for the study. The model is fitted with a
shaped nose-piece to ensure attached flow over the forward portion of the model.
The model is equipped with a force balance to measure drag. In addition base
pressures are measured, and hot-wire wake surveys are conducted downstream
from the model base. The Reynolds numbers (based on the square-root of the
model cross-sectional area), range from 0.1 x 106 to 0.4 x 106 •
Drag reduction is effected by means of flaps attached along the edges of the
model base, and inclined inward to decrease the size of the downstream wake. In
addition, an oscillatory perturbation is applied at the flap origin in an attempt to
maintain attached flow for larger angles of flap inclination.
The present study has found that a simple, passive base-flap deflection-no
forcing whatsoever-produces significant drag saving. The maximum drag
reduction is 0.06-0.08 at an angle of 9-10 degrees. The magnitude of the saving
is in accord with both early and recent measurements in other laboratories.
The present results also show that oscillatory momentum addition has little
effect on drag reduction unless the net oscillatory momentum flux coefficient is
equal or greater than 0.1 %. Increasing the oscillatory momentum perturbation to a
coefficient value of 0.3% produces drag savings at angles greater than 9-10
degrees, but has very little effect upon the maximum saving at 9-10 degrees.
To further study this anomalous behavior, follow-on experiments are planned to
investigate a larger range of forcing amplitudes, and a variety forcing-function duty
cycles. In addition, Digital Particle Image Velocimetry will be used to capture the
detailed flow-field in the vicinity of the flap.
304 T.-Y. Hsu, M. Hammache, and F. Browand
Introduction
The objective of this research is to provide guidance for the reduction of base drag
of a typical tractor-trailer. Roughly speaking, the total drag of a tractor-trailer can
be broken down into a contribution from the tractor and undercarriage, a
contribution from the trailer base and a much smaller contribution resulting from
skin friction along the (long) trailer sides. As truck tractors become more
streamlined, and other sources of drag are minimized, the base drag of the trailer
will remain and assume even more importance.
As a hypothetical example, assume a tractor-trailer with a drag coefficient CD =
0.7. The drag coefficient is defined as C 0 = D , where D is the measured
_!_pU 2 A
2
drag; p is the air density; U is the wind tunnel speed, and A is the area of the
cross-sectional body perpendicular to the flow. In this case, the contribution from
the trailer base might be chased = 0.14, which is 20 percent of the total drag.
Reducing trailer base drag by 50 percent would reduce total drag by approximately
10 percent. Now imagine continued improvements to the tractor that decrease the
total drag to CD= 0.5. The base drag coefficient might actually increase somewhat
because the thinner boundary layer separating from the end of the trailer is more
efficient in entraining air from the base region. More efficient entrainment may
increase the base drag to a value Chased = 0.20. In this circumstance, trailer base
drag becomes approximately 40 percent of the total drag, and reducing trailer base
drag by 50 percent now represents a total drag reduction of 20 percent. The
observation that base drag and fore-body drag are inversely related was put forth by
Hoerner [5] based upon observations on aircraft shapes having blunt bases. The
argument has also been suggested to hold for heavy trucks (see Diebler & Smith in
this proceeding, and Saltzman & Meyer [10]).
There are two general approaches for the reduction of base drag. One approach,
we would term passive control, is to alter the geometry of the base region is some
way. The boat-tail-plate attachment studied by Lanser et al., [8], and illustrated in
Fig. 1, represents one possible alteration of base geometry. Measurements on a
full-scale tractor-trailer in the 80x120 Foot Wind Tunnel at NASA Ames indicate
total drag reductions of the order of 10 percent, and good performance at angles of
yaw (Lanser et al., [8]). Other more recent investigations of the boat-tail-plate
attachment include Khalighi et al., [6], and Storms et al., [11]. Storms et al., show
that total drag decreases from 0.263 without the boat-tail-plates to 0.215 with the
boat-tail-plates for a clean tractor-trailer geometry without wheels-the GTS
model. The total drag reduction is 18 percent.
The second approach is to attempt flow control by means of an active forcing
(such as an oscillating flap, or a blowing slot), meant to alter the boundary layer
properties- usually to avoid an unwanted separation. The blowing device
discussed by Englar in this proceeding and in [3], falls within this second category.
Our approach encompasses both a modification of the base geometry by means
of the addition of flat-panel flaps, and an additional active control by means of an
oscillatory mass flow perturbation within the boundary layer meant to delay flow
separation over the surface of the flaps. The (flat) flaps are attached to the trailer
Base Flaps and Oscillatory Perturbations to Decrease Base Drag 305
base along the trailer base edges, and are inclined to the free stream to close the
wake more efficiently, as in Fig. 2. They are attractive, because they can be folded
flat against the base when not in use.
Fig. 2 Base flaps attached to model truck base, recent test at NASA Ames
The addition of passive flaps to the base of a realistic tractor-trailer model was
first studied by Cooper [2]. Cooper shows that the drag coefficient for a typical
straight-sided truck decreases from 0.78 without flaps to 0.72 with flaps. The
optimum flap angle is within the range I 0-15 degrees. Further, Cooper
demonstrates that the effectiveness of the drag reduction increases with non-
dimensional flap length, l, defined as l = L!sqrt(A), where Lr is the flap length and
A is the cross-sectional area of a truck. He points out that most of the drag
reduction is accomplished for flap length less than a value of approximately, l =
0.18.
With regard to active flow control, Nishri and Wygnanski [9] reveal that an
oscillatory jet introducing momentum flux with a net zero mass flow is more
effective than a steady blowing jet in delaying flow separation over a flapped
306 T.-Y. Hsu, M. Hammache, and F. Browand
airfoil. They are concerned primarily with increasing lift by means of oscillatory
suction/blowing. Their oscillatory jet, generated by a loudspeaker, is located at the
origin of the flap as illustrated in Fig. 3 from Nishri and Wygnanski. They show
that the effectiveness in delaying flow separation is determined by the location of
the jet, the frequency of the induced oscillation, the net momentum flux coefficient,
and the shape size of the slot. The frequency of the induced oscillation, f, is non-
dimensionalized by the flap length and free-stream velocity, Ux. It is defined as
F+ = f* Lr . The net oscillatory momentum flux coefficient, Cl' , is defined as
u~
g u
c" 2 *- * (-')' , where g is the slot height, i.e., the gap between the flap and
=
Lr U x
the side wall, and Uj is the amplitude of the oscillatory jet fluctuation. Favorable
delays in separation are obtained for slot heights of the order of 1/3-112 of the
incoming boundary layer displacement thickness, and for cl' values less than 0.1 %.
The most effective non-dimensional frequencies appear to lie in the range p+ =
0.3-1.5.
We wish to carry over these previous insights and technologies, and to apply
them for the purpose of base drag reduction. Specifically, we wish to address the
following questions.
• For the flat-panel base flap geometry applied to our model, what is the optimum
flap angle to achieve drag reduction?
• Will the oscillatory jet delay the separation of the flow , resulting in additional
drag reduction?
• If the oscillatory jet has the positive effect of drag reduction, what will be the
magnitude of the drag reduction?
• What are the critical parameters for the application of active flow control?
G = g+THICKNESS O F F L APE R O N
Experimental Apparatus
Wind Tunnel
The USC Dryden closed-circuit wind tunnel facility consists of a settling chamber,
contraction, test section, diffuser and axial fan section. The contraction represents
an area change of 10:1. The test section-6.1 meters in length-has an octagonal
cross-section. The dimension between any two parallel sides of the octagon is 1.37
meters. For the ground vehicle studies, a porous ground-plane 1.37 m in width and
5.8 m in length is installed in the test section. A small amount of suction is applied
over the surface of the ground-plane to maintain a thin boundary layer. The
maximum free stream velocity in the wind tunnel is approximately 30 m/s.
Truck Model
A bluff-body model consisting of a parallelepiped having a rounded nose is used to
study base drag reduction, Fig. 4(a). At this point, no attempt is made to model an
actual tractor-trailer. The width of the model is 0.197 m. The ratio of width:
height: length for the parallelepiped is [1:1.4:4]. It is mounted above the ground
plane a distance 50.8 mm. To this parallelepiped is attached a rounded nose of
length 0.934 m, as shown in blue in Fig. 4(a). The nose is carefully shaped to
avoid flow separation. The cross-sectional area of the model, A = 0.0535 m2 , is
used as the reference area.
Model drag, side force and yawing moment are measured by means of a strain
gage sandwich designed specifically for this purpose. The model is mounted
directly to the upper surface of the sandwich. The lower surface of the sandwich is
anchored below the surface of the ground plane by means of the two protruding
pins seen in Fig. 4(a). Also seen within the model is a Scanivalve pressure
sequencer. There are fourteen pressure ports on each side of the model, and sixteen
pressure ports on the model base. Each of these ports is connected to the
Scanivalve and sequentially sent to a Baratron pressure transducer located outside
the wind tunnel.
The model incorporates flat-panel flaps along each edge of the base. In Fig.
4(a), the flaps are shown fabricated from acrylic plastic flat stock. To provide a
cleaner design and more reliable flap-slot geometry, we now construct the flaps
from 0.89 mm brass flat stock, as in Fig. 4(b). The flaps are soldered along the
four common edges to form a box-like structure attached to a plate mounted
directly to the base of the model. A total of eight independent sets of flaps at flap
angles 0, 4, 8, 12, 16.5, 20.5, 25 and 29 degrees are tested. Flaps can be doubled in
length by taping on the appropriate flap extension, as the second photograph in Fig.
4(b) illustrates.
A detailed picture of the model base is shown in Fig. 5. A schematic drawing
looking from the top is shown to the left. A commercial hi-fi loudspeaker made by
Vifa (frequency range from 35 to 4000 Hz) is used to produce the oscillatory
forcing. Motion of the speaker diaphragm in an otherwise sealed enclosure forces
air out through the slot gaps, g = 1mm, along all four edges of the base. Two flap
lengths, Lr = 5.4 and 10.48 em, are studied. When normalized with the
308 T.- Y. Hsu, M. Hammache, and F. Browand
v
characteristic scale, A = 0.231 m, the non-dimensional flap lengths become,
respectively, l =0.233 and 0.453).
(a)
(b)
Fig. 4 USC model equipped with rounded nose and base flaps
Base Flaps and Oscillatory Perturbations to Decrease Base Drag 309
Top''
Experimental Method
Experimental Conditions
All data sets are recorded at wind tunnel speeds between 13 and 25 m/s. Reynolds
numbers, based on the square-root of the model cross-sectional area are,
respectively 0.11 x 106 and 0.36 x 106 • A sine wave is chosen as the forcing
function. The forcing frequency is zero or within the range 40-600 Hz. The
corresponding non-dimensional frequencies are, F = 0, 0.17-3.93. The
oscillatory momentum flux coefficient, C1, ranges from 0 (no forcing) to 0.3%.
Signals from the force balance, the Baratron, and the hot-wire probe are digitized
using Labview. The nominal sampling rate for the force measurements in the wind
tunnel is 100-1000 Hz for a total sampling time of 5-8 seconds for each data point.
The sampling rate for the pressure measurements is similar. The averaging time of
5-8 seconds is sufficient to obtain accurate mean values. Previous observations of
wind tunnel performance (Zabat et al., [12]) have shown that virtually all of the
wind tunnel turbulence (including low frequency unsteadiness) lies above 0.2 Hz.
Typically, 5-second averages of the free stream velocity lie within .0015*U, of the
long-time mean. Hot-wire signals are normally sampled at about 20 samples per
cycle for a suitable number of cycles.
The rms amplitude of the unsteady velocity in the vicinity of the slot gap is
measured outside the wind tunnel by placing the hot-wire a distance of about one
gap width downstream from the slot. This measurement is needed to determine the
magnitude of the applied forcing (the value of C~'). Observations at several
positions along the length of the slot verify that the influence of the corner is
restricted about two slot widths. The slot flow is two-dimensional within ± 8%.
310 T.-Y. Hsu, M. Hammache, and F. Browand
The rms fluctuation amplitude that determines the value of Cl' is measured at
ambient pressure on a bench test, as is common practice [4], [7], [9]. The particular
level of ambient pressure has no influence on the measurement of rms fluctuation
since all surfaces are exposed to the same uniform pressure. In the wind tunnel,
however, the situation is slightly different. One side of the driving speaker is open
to the slot gap while the other side is exposed to a static pressure equivalent to a
pressure along the underside of the body. The speaker must be driven against this
pressure difference. In our case these pressures are both close to the wind tunnel
static pressure, and no correction to Cl' has been made. A more thorough in situ
calibration of forcing is planned.
Fig. 6 shows the large variation in rms amplitudes that are a consequence of the
dynamic response characteristic of the speaker plus enclosure at the different
applied frequencies. The largest rms amplitudes occur in two periods-one at
about 50 Hz and the other centered at about 140Hz.
The hot-wire is also used to determine boundary layer thickness at the trailing
edge of the model. The boundary layer there is turbulent with a displacement
thickness of approximately 2 mm at a wind tunnel speed of 16 rnls. The hot-wire is
used to measure the downstream wake of the model truck for a variety of flap
angles. For this purpose, the hot-wire is positioned at the mid-height of the model
in a cross stream plane 76.2 mm behind the flaps.
10
--.. 8 5.5
5
>
'-'
~.5
~ 7 J.~
J
c. 1.5
E l
< 6 1.5
I
o.s
5 0
Fig. 6 Strength of the oscillatory jet for I mm gap with sine wave forcing
Base Flaps and Oscillatory Perturbations to Decrease Base Drag 311
Passive Control
We first examine the addition of flat-panel base flaps in the absence of the
oscillatory jet forcing. The effect of flap angle upon drag reduction is shown in
Fig. 7. The horizontal axis represents the flap angle in degrees, and the vertical
axis represents the change in drag coefficient, ilC 0 = Conotlaps - Cotlap"" A positive
value of ilC0 corresponds to a reduction in drag. Data is shown here from three
widely different experiments. The two curves in red are from our wind tunnel
experiment on a model having an aerodynamically clean fore-body at Reynolds
numbers of 0.23 and 0.36xl06 • The experimental reliability is indicated by the
scatter of data points representing multiple trials.
The blue curve is an earlier mentioned result from Cooper [2]. Although the
flaps are similar, the model truck geometries are different. His flaps are attached to
the base of a truck model along the vertical edges and the top edge. The bottom
edge has an attached flap, but this flap is not deflected. The result shown
represents flaps on a model truck having straight sides, wheels and a trailer with
rounded nose tested at a Reynolds number of RevA:::: l.5x10 6 •
The data shown in black is from a recent experiment performed in the NASA
Ames 12-foot pressurized wind tunnel at a Reynolds number of 6x106 , which is
comparable to a Reynolds number for a full-scale truck at highway speeds. In this
case, the model is a generic truck shape having a gap between tractor and trailer,
cab extenders and simple axles with wheels. Multiple data points again give an
indication of the measurement reliability.
In spite of the different model geometries, all of the ilC 0 curves have a roughly
similar shape. Drag savings first increase and then decrease with increasing flap
angle. Thus there is an optimum angle for maximum drag saving. The maximum
saving is in the range LlC 0 = .06 - .08 for all three experiments. Overall, the three
data sets suggest a robust drag saving that is neither particularly dependent upon
the details of truck shape nor upon the Reynolds number of operation. However,
the angle for maximum saving does seem to depend upon Reynolds number. The
optimum angle seems to increase steadily from 9-10 degrees at 0.3x10 6 to perhaps
20 degrees at 6x106•
The drag reduction resulting from base flaps can also be observed in the wake
profile as a reduction in the wake momentum deficit. Wake velocity profiles are
shown in Fig. 8 for flap angles of 0 and 12 degrees. The horizontal axis represents
the span direction, non-dimensionalized by the square-root of the model cross-
sectional area. The model centerline is at the zero span position. The vertical axis
represents the non-dimensional velocities. The blue curve indicates the wake
profile for flap angle equal to zero. It can be seen that the wake width is the order
of the model width. The red curve shows the wake profile for flap angle equal to
12 degrees. The wake width is now considerably narrowed. The small velocity
increase at mid-span may actually represent a recirculation (backflow) along the
center-plane. The single hot-wire used here is not capable of determining flow
direction and may thus "demodulate" the recirculation.
312 T.-Y. Hsu, M. Hammache, and F. Browand
Active Control
Now consider the addition of oscillatory blowing/suction along the edges of the
model base, coincident with the origin of the flap. The purpose of the oscillatory
blowing/suction is to attempt to maintain attached flow over the flap for larger flap
deflection angles, and thereby to further reduce drag. Fig. 9 summarizes the best of
our results. The drag coefficient difference, .:lC 0 , is now shown for two flap
lengths- with and without oscillatory blowing. The two blue curves represent
flaps of length 0.233 and 0.453 in the absence of forcing. The drag saving peaks at
about 9 degrees as in Fig.?, and doubling the flap length has no significant effect
on the result. The results in red, for a momentum coefficients of C11 =0.15% and
0.3%, demonstrate the effectiveness of oscillatory blowing/suction. There is a
modest additional drag saving at the higher flap angles, but the application of
forcing has very little effect in the vicinity of peak saving. Thus the peak saving is
almost unchanged. The cause of this unexpected result is still under investigating.
Examination of the forced and unforced results in the vicinity of 10 degrees
suggests that the momentum addition does delay the onset of separation in some
manner. However, we had anticipated that in the presence of forcing and delayed
separation, the drag saving would continue to increase beyond the unforced peak at
9-10 degrees. Rather, the demonstrated effect of forcing is to broaden the peak
region by extension to larger angles but not to increase the maximum value of drag
saving.
For the shorter flap length, the optimum forcing frequency definitely lies in the
range F* ... 0.6 - 0.7. For the longer flap, the forcing frequency may be slightly
below optimum. Here the non-dimensional value F* ... 0.33 is dictated by the need
to stay in the frequency band near 40 Hz to achieve the highest momentum
coefficient possible.
Our experiments suggest that there is a rather broad range of non-dimensional
forcing frequencies F+ ... 0.3 - 1.0 that will effect flow modification, but that
forcing amplitude is critical. We begin to see noticeable change only at momentum
coefficients of the order of 10-3 or greater (C11 ~ 0.1 %). Nishri & Wygnanski [9]
show observable increases in lift over deflected flaps at momentum coefficients as
small as 0.01 %. Evidently drag reduction for a three-dimensional body such as
ours is a much more subtle proposition.
Base Flaps and Oscillatory Perturbations to Decrease Base Drag 313
C> 0.05
c::
·~
0
lL
0
z
'C
~ 0
Re=0.23x1 ()6 Red. 1=0 233 from USC
Re=0.36x1o6 Blue, 1=0.183f Stra1ght truck w/
Re=1 .5x1o6 rounded ront from Cooper
Re=6x1o6 Black, 1=0.25 from NasaAmes
-O OSOL--_ __._5_ __.10_ _ _1.._5_ __..20_ _ _2._5_ _3_,_0_ _~35-
Fiap Angle (degrees)
Fig. 8 Effect of base flap angle on the momentum defect in the wake
314 T.-Y. Hsu, M. Hammache, and F. Browand
0.1 r----r----.---..-----,.-----.-----.-----,
"0
5 0.05
Cll
u.
"'
Cll
"0
U5
5
0
u.
-o 0 1=0.233: f•=o. c " =O t;r
~ 1=0.453: F•=o. c " =O
1=0.233: F• =0.67, Ce=0.3%
1=0.233: f • =o.33, c "=0.15% Re=0.23x1Q6 for a ll case
1=0.453: f+:0.33, Ce=0.3%
,___
15::------'20
-0.050~--5":---1-L0--....1. _2.._ 35
_,___ __J
5 _ _ 30
Flap Angle (degrees)
Conclusions
Present Study
The present study describes the addition of flat-panel flaps to all four sides of the
downstream end of a bluff-body model resembling a tractor-trailer. The nose of the
model is rounded to prevent unwanted upstream separation (and boundary layer
thickening). It is possible to form partial answers to the questions posed at the end
of the introduction.
• The flaps-when deflected to flap angles of 9-10 degrees-produce a total drag
reduction of ~C 0 "" 0.075, as measured by a force balance within the model. The
drag reduction is broadly consistent with reductions measured earlier by Cooper
[2], and with more recent measurements acquired in the 12-foot wind tunnel at
NASA Ames.
• The amplitude of the oscillatory forcing applied at the origin of the flaps is most
important. Forcing has little effect at small values of the oscillatory momentum
coefficient, C" = 0.1 %. Forcing at amplitudes of C" = 0.3 %, does produce
modest additional drag reduction at larger flap angles, but does not alter the peak
drag reduction at 9-10 degrees.
• The forcing does appear to delay separation over the flap.
• Non-dimensional forcing frequencies in the broad range, F•=f*LrfU, "" 0.3 - 1.0,
do appear to provide the greatest effect on the flow , although this is a tentative
conclusion.
Base Flaps and Oscillatory Perturbations to Decrease Base Drag 315
To date, the results from oscillatory forcing leave several questions unanswered.
We believe oscillatory forcing should lead to greater drag savings. This view is
supported by the more recent work of Kjellgren et a!., [7], who investigated drag
changes for a two-dimensional V-22 model wing in hover mode. In this case, the
bluff-body is the wing itself with down-flow normal to the upper surface of the
wing, as would be achieved in hover mode (with tilt rotors above the fixed wing
providing the down-flow). Forcing at the leading and trailing edges near flap
junctures does delay separation to larger flap deflection angles, and appears to
provide ~C 0 of as much as 40-50 percent. What is evident from this work is that
drag improvements may require larger forcing amplitudes-of the order of c~
?:O.oi.
The present experiment will be modified in several important respects to allow
an expansion of the range of forcing parameters-particularly the forcing
amplitude. The present loudspeaker arrangement cannot achieve C~ values beyond
0.3%, when all four sides are forced. A means for forcing at higher amplitudes is
currently being investigated, and will be incorporated into the redesign of the
model.
We plan to investigate forcing-signals other than sinusoidal. For example, a step
in speaker voltage produces a blowing impulse response or a suction impulse
response as sketched in Fig. lO(a). If either the suction or the blowing portion of
the cycle should provide more effective control (there is anecdotal support for the
importance of suction), one or the other can be emphasized by means of saw-tooth
signals as in case (b).
Finally, we plan DPIV studies (Digital Particle Image Velocimetry) of the flow
adjacent to the flap with and without forcing. The intent is to answer the question
regarding the state of the boundary layer and the degree of attached/unattached
flow present.
b~~~
omp..,m-d
I
1+-------"1
1 Cycle
1 Cycle
(a) (b)
Fig. 10 New designs: (a) Example of equal blowing and suction response, (b) Saw-tooth
emphasizing either blowing response or suction response
Acknowledgement:
Support from Dr. Sidney Diamond, Chief of Technology Development, Heavy
Vehicle Systems, Office of FreedomCAR and Vehicle Technologies, is gratefully
acknowledged. Additional thanks to Dale Satran and to NASA Ames for providing
the recent experimental results for base flaps operating at full-scale Reynolds
numbers.
316 T.-Y. Hsu, M. Hammache, and F. Browand
References
DaimlerChrysler AG
During the planning stage the fundamental structure, specification, and layout
of a new vehicle are determined. Goals like fuel consumption, safety and
comfort requirements are defined. A great bandwidth of design concepts and
innovative ideas is characteristic for this period. Available data are usually
based on simplified wind tunnel test configurations or on models of
predecessor vehicles. CFD is becoming a more and more important tool for
evaluating design studies of aerodynamic components. In general, the CAD-
data that are available early in the design process are not very detailed, allowing
for fast and easy setup of the computational models. As an example, fig. I
shows the model of a tractor trailer configuration where the tractor cabin or
other relevant aerodynamic components can easily be replaced in the baseline
model. The model is scaled down to 40% of a production vehicle. It exists in
hardware as well as in CAD-format. Investigations of the various design
concepts may be done in a full scale wind tunnel as well as in a virtual CFD
wind tunnel. A good coordination of experimental and computational
investigations is important.
The drag force of the assembled configuration and the contribution of the
investigated components are the major results of both experiment and
computation. The effect of geometrical modifications on the drag values has
been well captured by the CFD results and the total drag forces differ in a
range of 5-1 Oo/o compared to the experimental results.
The CFD results also offer information about the pressure and velocity field
as well as specific flow field phenomena like vortices and separation zones. As
an example, fig. 2 shows the surface streamlines and the near wall velocity
distribution on the surface for two different COE (cab over engine) tractor
cabin studies. In both pictures, the stagnation zone at the grill and the
separation zone downstream of the a-pillar can be seen. Obviously, the
separations are much more extended for the cabin on the right side of fig. 2
where the front roof edge and the a-pillar have significantly smaller radii. For
this cabin the reattachment zone is located on the trailer, while for the cabin
with the more rounded edges the reattachment zone is located downstream of
the side door.
Use of Computational Aerodynamics for Commercial Vehicle Development 321
....
Fig. 1. 40% tractor-trailer model with different tractor cabin designs
Fig. 2. COE (cab over engine) tractor study- near wall velocity and surface streamlines for two
different cabin designs
During the styling stage, digital surface data are available from surface scans
which are periodically performed for packaging verifications. These data can
be used to set up computational models. Because of the dynamic change of the
shapes of the styling models an important requirement for the use of
computational aerodynamics is a short time lag between the surface scan and
the aerodynamic evaluation. Therefore a process has been established which
allows fast CFD model generation based on scanning data. This process will be
discussed in the subsequent section.
Again, the most important result of the computation is the total drag
coefficient. Nevertheless, the difference in the drag value between various
styling models can be explained by inspection of the 3-dimensional flow
results. As an example, fig. 3 shows the isosurface of the total pressure
coefficient Cp,=O for four different van styling models. In the interior of the
isosurface the flow velocity usually is small. Therefore the extension of the
isosurface indicates separation and wake regions as well as zones of weak flow
attachment near surfaces.
As an example fig. 3 shows that the near wall air flow between the front
wheel well and the a-pillar is much weaker for model (d) compared to model
(a). This indicates that flow separation along the fender panel will occur at
greater yaw for model (a) than for model (d) which is important in order to
reduce crosswind sensitivity. On the other hand the separation at the a- and d-
pillar is remarkably smaller for model (d) compared to (a).
In addition, the pressure distribution on the body surface is inspected.
Stagnation regions at the underbody can be identified as well as separation
lines at the a- and d-pillar and the rear roof edge. Also the base pressure
distribution is inspected in order to assess the overall aerodynamic quality of
the styling model. Differences in the base pressure usually can be explained by
geometrical differences in the models.
For vans, the styling of side mirrors needs to be evaluated with respect to
drag and foreign soiling of the side window. Also the drag reduction potential
of underbody paneling can be studied with flow computation. It can be
Use of Computational Aerodynamics for Commercial Vehicle Development 323
demonstrated that underbody paneling strongly influences the underbody flow
as well as the flow structures in the wake region. Stagnation zones at
underbody structures can be detected and minimized. The influence of full
underbody paneling on the wake flow is demonstrated with section streamlines
in the vehicle mid-plane in fig. 4. In the figure the dashed lines indicate the
extension of the wake. Obviously the full underbody paneling reduces the
extension of the wake zone (fig.4,b) and thus the total drag. It may also
influence the shape of the vortex system in the wake and therefore the soiling
of the rear window.
(a) (b)
-'
~
', '
·.
·.'
. .
·······--· ·······=· ~~ ~·-·-
Fig. 4. Streamlines in vehicle mid-plane without (a) and with (b) underbody paneling
L
! CAD surrace creation
CAD Data( CATIA) J
20 days
! surface triangulation
Ismooth surface m esh (ANSA) I
1-2 days surface triangulati on
Ipolygonal representation (POLYWORKS) I
! Ill
Fig. 6. Model generation and aerodynamic simulation during the styling stage
Use of Computational Aerodynamics for Commercial Vehicle Development 325
During the styling stage only data from surface scans are available. The
traditional work process which is necessary to obtain computational results
during this stage is depicted in fig. 6 on the left side. Digital data representing
the surface of the clay model are obtained from surface scans. A 100% clay
model is scanned during 2-3 days; a 25-40% model in approximately one day.
Based on the resulting point cloud a polyhedral with up to 10 million triangles
can automatically be generated. These data are used for packaging verification
and in-house presentations. During the following 20 work days, CAD data
(e.g. in CATIA format) are derived from the polyhedral. For the CFD
calculation with Powerflow, a closed surface mesh is required which can be
derived from the CAD data within 1-2 additional days with the assumption
that an appropriate underbody already exists. Fig. 7 (a) shows a section which
has been meshed manually with ANSA. In this mesh which consists of
approximately 100,000 triangle elements, the surface curvature is very well
represented. The analysis with EXA Powerflow takes about 2-4 days on a
Fujitsu-Siemens 16# Linux cluster. The overall time lag between the
availability of digital surface data and the computational results amounts to at
least 5 weeks which is not acceptable within a dynamic styling process.
Therefore an alternative process has been established which is depicted on
the right side of fig. 6. There the polyhedral is reduced (e.g. with
POLYWORKS) to an appropriate number of triangles whereas the surface
curvature is still sufficiently represented. This surface needs to be
complemented with an underbody structure in order to finally obtain a closed
surface mesh. Fig. 7 (b) shows the result of this procedure with approximately
the same number of triangles as in fig. 7 (a). The computational results
obtained with the different meshes differ within an acceptable range while
reducing the time lag from 5 weeks to one week with the alternative surface
meshing process.
(a) (b)
Fig. 7. Smooth surface mesh (a) and polygonal surface representation (b)
326 ]. Urban
Once the closed surface model has been generated, a partially refined
volume mesh is generated within a fully automatic procedure of Powerflow.
The refinement regions can be defined with simple hexahedral boxes or other
closed volumes of arbitrary shapes. The voxel size is halved in each spatial
direction at every refinement level. A section of such a mesh is depicted in fig.
8 for a Freightliner Century Class. In this case, hexahedral boxes have been
defined in order to resolve the flow structures around the sunvisor and the
bumper region.
During the instationary computation, the solution is periodically stored on
scratch and the total drag of the vehicle is calculated. The computation is
considered complete when the total drag reaches a stable level. A quasi-
stationary solution as well as a drag value is finally obtained by averaging a set
of solutions as shown in fig. 9.
averag1ng Interval
For the Mercedes-Benz Actros face-lift, a side flap at the a-pillar has been
designed which prevents water droplets and spray from entering the zone
between the side window and the side mirror. Three-dimensional streamlines
in fig . 10 show the generation of an extended separation bubble. As a
consequence, droplets only splash on the outer edge of the side mirror and the
spray is transported with the main flow around the mirror. The flap increases
the total drag significantly, but the water accumulation on the side window
and mirror has been significantly reduced. This has also been verified by wind
tunnel experiment (fig. 11). Final optimization of the flap shape was done in
the wind tunnel at a level of efficiency higher than that for CFD.
In order to reduce self-induced soiling of the side door handle and other
cabin entry contact areas, a vent has been designed which is depicted in fig. 12.
This side vent prevents flow separation at the lower region of the a-pillar and
creates a jet flow which reduces road spray from the front wheels. The
streamlines in fig. 12 show the downward direction of the net flow. The
twisting of the streamlines represent the swirl which is generated at the sharp
edges of in the vent. Nevertheless there remains soiled regions at surface areas
where the flow is weak or separated from the surfaces. This has been
demonstrated with wind tunnel testing as depicted in fig. 13.
328 ]. Urban
4 Conclusions
During recent years, computational aerodynamics has become a powerful tool
within an efficient design process. The delay between the first available digital
data and the flow results has been significantly reduced with an innovative
model generation method and enlarged computational resources. It has been
shown that computational aerodynamics offers an assessment of various design
aspects of the exterior.
It needs to be noted that during all stages of the design process, a
synchronization of computational investigations and testing activities is
important in order to obtain the best available design within a limited time
frame.
Numerical Simulation of the Flow about a Train
Model
Summary
This paper presents results from flow simulations about a simple train model
to be used in an upcoming experiment. The simulations are made with the un-
structured Navier-Stokes solver TAU employing hybrid grids. The influence of
wind tunnel walls as well as a ground plate with and with out a boundary layer
is examined. The results obtained so far indicate that the influence of the wind
tunnel wall can be neglected in the computation, while the boundary layer on
the ground plate might become critical in the experiment. Furthermore the
influence of cross wind on the train is studied by yawing the train up to 30° to
the oncoming flow.
1. Introduction
Amongst different kinds of trains used in Germany are the fast trains called
ICE (Inter City Express). While the front and rear cars with the engines were
quite heavy in the first generation trains, they are quite light weighed in the
new generation since the engines are now distributed over all bogies. Travelling
at very high speeds while encountering strong side winds can be hazardous
causing derailment of the train especially at the instance of exiting a tunnel or
crossing a bridge. Therefore the study of high speed trains under different cross
wind angles is of great significance. The objective of this particular preliminary
analysis has been the determination of suitable wind tunnel conditions for fu-
ture tests and studies of stability and derailment of the ICE trains. Therefore a
simple geometry of the train without bogies is considered to examine both the
influence of the tunnel walls in the experiment and of the boundary layer de-
veloping on a ground plate below the model. Once the computer simulation
and the wind tunnel test results to be available in the near future are within
reasonable agreement, more complex geometries will be studied.
330 S. Barakat and D. Schwamborn
2. Numerical approach
In this study the DLR-TAU Code which solves the Reynolds-averaged Navier-
Stokes equations (RANS) on hybrid grids (Gerhold and Evans 1999;
Schwamborn et al. 1999) is used. The three major modules of this code are:
the pre-processing for the unstructured hybrid grid, the flow-solver which
works on a dual grid with an edge-based data structure and the adaptation
which allows for an automatic adaptation of the grid to the solution. Both pre-
processing and solver are fully parallelized for distributed memory machines
such as the LINUX cluster used in this exercise. Parallelizing the adaptation is
currently under development.
The pre-processing supports tetrahedral, prismatic, pyramidal and hexahe-
dral elements thus enabling the use of efficient and flexible modern grid gen-
erators for the generation of meshes about complex geometries within days.
Besides producing the dual grid, the pre-processing also takes care of the do-
main decomposition for the case of parallel computation, the agglomeration of
coarser grids to be used in the multi-grid acceleration and the optimisation of
the data structure for vector or cache-based computers.
TAU is based on a finite volume technique, and it integrates the solution in
time using a Runge-Kutta method. It employs different central and upwind
discretisations for the convective fluxes and a number of one- and two-
equation turbulence models. In the current investigation, central discretisation
and a Spalart-Allmaras one-equation model (Edwards 1996) is used. For time-
accurate calculations a dual-time stepping approach can be enabled.
The resolution of flow features can be taken automatically into account by
the adaptation which allows for local refinement of the hybrid grids based on
different refinement sensors. Additionally a redistribution of grid points in
structured sub-layers (composed of prisms or hexahedrons) is possible in order
to adapt the mesh along wall-normal rays for an improved boundary-layer
resolution.
3. Computer Simulation
A geometry representing a German ICE train is used in the simulation which
considers only the leading car where the engine is located plus a connector
section and the front part of a second car. The physical specifications and
condition of the train are given below:
The commercial grid generation tool CENTAUR was used in creating the
hybrid unstructured grid. The near wall grid of the train consists of prismatic
layers to resolve the boundary layer, while the remainder of the computational
domain is filled with tetrahedrons except for the area of the ground plate be-
low the train where the grid was also prismatic to allow the simulation of a vis-
cous ground plate
In the case of the viscous ground plate, a no-slip condition was used along
the ground below the train starting at one reference length (considered train
length 45m) in front of the nose. The prismatic layers on the ground plate are
Numerical Simulation of the Flow about a Train Model 333
In the case of a viscous ground; however, the stagnation point at the nose of
the train is shifted upwards (fig.3) due to the displacement effect of the on-
coming boundary layer. In Figure 4 it can be seen that the streamlines tend to
diverge along the bottom due to the increase in boundary layer thickness along
the ground.
The influence of the boundary layer on the pressure distribution in the
symmetry plane of the train can also be seen in Figure 5 where the cases with
and without ground plate are compared for free-air boundary conditions in the
far-field . Due to the displacement effect of the boundary layer the stagnation
point is moved upward, and the pressure on the upper nose is increased. At the
same time the suction peak below the nose is reduced. Thus the average pres-
sure on the frontal area rises resulting in an increase of the pressure drag.
From this it is concluded that the effect of the wind tunnel can be neglected
at the low blockage for the case of zero yaw. Although a nonzero yaw angle
would increase the blockage effect to a certain extent, this increase would not
be substantial. Although this has still to be proven in a future calculation, it is
334 S. Barakat and D. Schwamborn
concluded for the moment that the wind tunnel simulation would not be
negatively influenced by wall effects. Thus it seems not necessary to simulate
the wind tunnel in future computations.
Table 1. Drag Coefficient Comparisons between the Train inside the Wind Tunnel, Outside the
Wind Tunnel and Viscous Ground
1 5
- - ~ VISCous ground plate
- NOf'l scous ground pla;e
I
Trai n ~· o n lour
/
~
0.5
c.
'1
....__
J
0
- 05
I
!
0 0 .2 0 .4 0.6
Normalized Length
However, the ground plate might play a critical role in the experiments. In
the current calculation the running length of the boundary layer in front of the
train was equal to the simulated length of the train. Since the front edge of the
plate will be much closer to the nose of the train (approximately one train
height) in the experiment, the effect of the boundary layer would be much
smaller at least at the nose of the train and at zero yaw. However, at higher yaw
angles the running length of the boundary layer is increased again. Addition-
ally the boundary layer height will then vary along the train causing effects in
the flow about the train that cannot be neglected.
to cover the worst case scenario which is unlikely to occur in reality at train
speeds above 200 km/h.
Fig.6. Streamlines distribution on the train seen from the windward side for yaw angles of 0°,
15 o, 30° (top to bottom)
Figure 6 depicts streamlines on the surface of the train as the yaw angle
changes. (Please note that the train in the picture is see-through, i.e. the
streamlines from both sides of the train could be viewed). Looking at this fig-
ure it can be concluded that separation occurs along both the top and bottom
leeward edge of the train in all cases of nonzero yaw. This separation leads to
the formation of leeward vortices which are visualized in Figure 7 for the case
of25° yaw in a cut 11m behind the train nose using lines of constant vorticity.
It can also be seen from Figure 6 that the top separation is moving upstream
as the yaw angle increases, finally arriving at the very nose of the train.
336 S. Barakat and D. Schwamborn
.---- - -. ------.:::=~:-
:· .... '
Fig.?. Vortices separating from top and bottom of the train indicated by vorticity contours in a
cut a 11m behind the train nose
2 ~----~------~----~------.------.------,
.-· . . ...
0 .5 . . ..··:=-~:--~ , . .
..........
_... --
- 0.5
- 1
0 5 10 15 20 25 30
Yaw Angle (Degrees)
4 ~----~------~------,------,----.----~.
- - CS - Pressure
- · - CS - Viscous
···· ······ CS- Fiow Direction
3 ---- CS- Train Direction
- - CS-Exp ., Re=5.9e6
-1
0 5 10 15 20 25 30
Yaw Angle (Degrees)
Fig. 9 Side force coefficient in wind and body fixed coordinates
338 S. Barakat and D. Schwamborn
References
Edwards JR, Chandra S (1996) Comparison of Eddy Viscosity Transport Tur-
bulence Models for Three-Dimensional Shock-Separated Flow Fields.
AIAAJ. Vol. 34, pp 756-763
Gerhold T, Evans J (1999) Efficient Computation of 3D-Flows for Complex
Configurations with the DLR TAU-Code Using Automatic Adaptation.
In: Notes on Numerical Fluid Mechanics, Vol. 72, Vieweg
Loose S (2002) Private communication
Schwamborn D, Gerhold T, Hannemann V (1999) On the validation of the
DLR TAU-Code. In: Notes on Numerical Fluid Mechanics, Vol.
72,Vieweg
Adaptation of Eddy-Viscosity Turbulence Models to
Unsteady Separated Flow Behind Vehicles
Ansys-CFX
Introduction
Turbulence model development for aerodynamic applications has for many
years concentrated on improving the capabilities of CFD methods for separa-
tion prediction. Validation studies of turbulence models in the '80'h have
clearly shown that most turbulence models were not capable of predicting the
development of turbulent boundary layers under adverse pressure gradient
conditions. Based on that observation, new models were developed to specifi-
cally meet this challenge, resulting in a series of models capable of capturing
boundary layer separation in good agreement with experimental data Qohnson
and King 1984, Menter 1993, Spalart and Allmaras 1994).
From the experience with the SST turbulence model (Menter 1993), the
present authors would argue that the capability of the model with respect to
the prediction of the onset of separation is within the accuracy of the available
experimental data and that no systematic deviation between the simulations
and the data is observed. Based on the experimental evidence (which is admit-
tedly limited for three-dimensional flows), there is currently little need for
model improvements for that type of flows. Off course, this does not imply
that aerodynamic flows can be predicted within experimental uncertainty lev-
els, as these flows involve other effects, which pose additional challenges to the
turbulence model. The main areas of concern are the behavior of the flow
downstream of the separation line, including the flow recovery after re-
attachment Qohnson et al. 1994), the proper simulation of vortex flows and
questions related to laminar-turbulent transition. Particularly the flow devel-
opment downstream of separation is of major importance from an aerody-
namic standpoint and can have a significant effect on the characteristics of
aerodynamic bodies. This is particularly true for ground vehicles, as they gen-
erally exhibit significant regions of separated flows, even at design conditions.
From a modeling standpoint, it has been observed for a long time that
RANS turbulence models underpredict the level of the turbulent stresses in the
detached shear layer emanating from the separation line Qohnson et al. 1994).
This in turn seems to be one of the main reasons for the incorrect flow recov-
ery predicted by the models downstream of reattachment. The issue is some-
340 F.R. Menter and M. Kuntz
times masked by the tendency of models of under-predicting the onset and
therefore the strength of the separation zone, which in turn results in an ac-
ceptable agreement in the recovery region. However, the improvement is only
the result of a cancellation of errors, as one cannot trade separation prediction
capabilities against improved velocity profiles in the recovery region. The de-
layed recovery of the boundary layer downstream of the reattachment line can
lead to a premature separation under a second adverse pressure gradient. A
second, and even more disturbing uncertainty resulting from the incorrect
prediction of the detached shear-layer, concerns the flow topology downstream
of the separation line. Current turbulence models cannot reliably answer the
question, whether the flow is forming a closed separation bubble, or a fully
stalled flow regime. This question is of major importance for the prediction of
the aerodynamic characteristics of automobiles, which almost always exhibit
regions of separated flow. The topology of these regions has a strong influence
on the drag and more pronouncedly on the lift of the car.
In an attempt of improving the predictive capabilities of turbulence models
in highly separated regions, Spalart (1997) proposed a hybrid approach, which
combines features of classical RANS formulations with elements of Large Eddy
Simulations (LES) methods. The concept has been termed Detached Eddy
Simulation (DES) and is based on the idea of covering the boundary layer by a
RANS model and of switching the model to a LES mode in detached regions.
Ideally, DES would predict the separation line from the underlying RANS
model, but capture the unsteady dynamics of the separated shear layer by
resolution of the developing turbulent structures. Compared to classical LES
methods, DES saves orders of magnitude of computing power for high Rey-
nolds number flows, due to the moderate costs of the RANS model in the
boundary layer region, but still offers some of the advantages of an LES
method in separated regions.
There are two main concerns with the current DES formulation. The first is
how quickly the unsteady turbulent structures develop after the model has
switched from the RANS to the LES mode. This is of significance for the pre-
diction of separated shear layers, as a delayed onset of resolved turbulent
structures would aggravate the underprediction of the turbulent stresses due to
a reduction of the unresolved turbulence level by the DES formulation.
The second concern is with the switching mechanism employed by the cur-
rent DES methods. In order to prevent the activation of the DES limiter in
attached boundary layers, it is typically required to ensure a lower limit on the
local surface grid resolution. If this condition is violated, the integrity of the
RANS model is severely compromised resulting in most cases in grid induced
separation. This issue will be addressed in the section on turbulence model
formulation. For an alternative of a hybrid turbulence model without an ex-
plicit grid dependency see Menter et al. (2003).
The capabilities and limitations of advanced aerodynamic RANS and DES
turbulence models for automotive applications will be evaluated and discussed.
Alternatives to the current RANS/DES switch will be discussed. The models
will be applied to the flow around a simplified generic car shape, known as the
Ahmed car body (Ahmed et al. 1984, Lienhart and Becker 2003). The RANS
Adaptation of Eddy-Viscosity Turbulence Models to Unsteady Separated Flow 341
simulations have been presented at the ERCOFTAC workshop on Refined
Turbulence Modelling (Durand et al. 2002) in a comparison study of different
CFD methods for specific testcases. A detailed report is available from the
authors upon request.
All simulations have been computed with the commercial CFD method
CFX-5 of ANSYS.
Flow Physics
The geometry of the Ahmed car body is shown in Figure 1. It consists of a box
with rounded edges and a slanted back. The angle of the slant is adjustable and
is the main variable model-parameter in the experimental investigations. Two
sets of experiments have been carried out for this geometry (Ahmed et al.
1984, Lienhart and Becker 2003).The present comparison is mainly based on
the data of Lienhart and Becker (2003). In this experiment, the Reynolds
number with respect to the length of the car was ReL=2.6x10 6• Slant angle of
25° and 35° were investigated. The emphasis of the experiments was on the
flow structure in the slant region and downstream of the body.
o.• .--- - - - - - - -
288
[ -1
D-og
Cc€ hcienf
viscous drag
front part
back part
slant part
40°
It was observed by Ahmed et al. (1984) that a change in the slant angle gave
a significant change in the drag coefficient, as shown in Figure 1 taken from
Ahmed et al. (1984). An increase in the slant angle from zero to around 30°
results in a gradual increase in the drag coefficient with a significant shift in
342 F.R. Menter and M. Kuntz
the contributions from the different surfaces. While the viscous drag forces
show little sensitivity to the slant angle, there is a distinct increase in the drag
portion coming from the slant surface combined with a decrease in the contri-
bution from the back surface. At a critical slant angle of around 30°, the drag
coefficient drops significantly, mainly due to a reduction in the contribution
coming from the slant. At higher slant angles, the drag coefficient stays virtu-
ally constant. The change in drag coefficient is due to a change in the flow to-
pology in the aft region of the car body (Lienhart and Becker 2003) .
Figure 2: Flow topology in the aft region of Ahmed car body (left: 25° and right: 35°). Courtesy
Lienhart, LSTM Erlangen.
Turbulence Model
RANS simulations have been carried out with the standard k-E model (Laun-
der and Spalding, 1974) with wall functions and the SST model (Menter
1993) with automatic wall treatment (Esch et al., 2003). In addition, the SST-
DES model proposed by Strelets (2001) has been modified and applied to
overcome some of its deficiencies in the RANS regions.
The practical reason for choosing the maximum edge length in the DES
formulation is that the model should return the RANS formulation in at-
tached boundary layers. The maximum edge length is therefore the safest esti-
mate to ensure that requirement. The DES modification of Strelets can be
formulated as a multiplier to the destruction term in the k-equation:
with CoES equal to 0.61, as the limiter should only be active in the k-E model
region. The numerical formulation is also switched between an upwind biased
and a central difference scheme in the RANS and DES regions respectively.
Figure 3: Regions of negative U-velociry for NACA4412 simulation for SST model (left) and
SST-DES model by Strelets (right)- separation point indicated by arrow
FssT=O recovers the Strelets et al. model. F1 and F2 are the two blending
functions of the SST model. F2 shields more of the boundary layer and is
therefore the prefered default. It should however be noted that even F2 does
not completely eliminate the problem, but reduces it by an order of magni-
tude, /). s < co
where c is now of the order of 0 .1.
Figure 4 shows the same simulation as for the standard SST-DES formula-
tion computed with the SST-CFX-F 2 model. The influence of the DES limiter
Adaptation of Eddy-Viscosity Turbulence Models ro Unsteady Separated Flow 345
is avoided and the DES model does not affect the separation point. It can be
seen that even a more severe grid refinement does not lead to separation (right
picture). However, refinement of the surface grid below !l 5 = 0.10 should be
avoided.
Numerical Results
Simulations for the two experimental cases with 25° and 35° slant angle have
been computed using three different grids with 0.65, 1.3 and 2.6 million
nodes for the half-model. The results were grid converged to plotting accuracy
on the medium grid. For the 35° case the k-£ and the SST model predict a
fully separated flow at the slant region, which is in good agreement with the
experimental data. In the following, the flow for the 25° slant angle is ana-
lyzed, as it is the more challenging configuration. A detailed report for the 25°
and 35° slant angle, including a grid refinement procedure, can be obtained
upon request from the authors. Note that both cases have been computed by
several authors for a recent ERCOFTAC workshop (http:/ /labo.univ-
poitiers.fr/informations-lea/Workshop-Ercoftac-2002/lndex. html) .
W)
§ \!(1
I • '
f :"
I /
<::
:J , i i !' : I . .'. ,'
.- ; \00
!:'II
~I ~~'~<I
ll"
lt.' •·" 1.!-' 1•1, l f)\ n !\ ' u n
- a is [mmj
Figure 5: Velocity profiles computed with different turbulence models for the 25° case
Adaptation of Eddy-Viscosity Turbulence Models to Unsteady Separated Flow 34 7
III
""'
""'
l.&O
..• ·.
E '~
E
--' lt:.:l
"' • Experiment
· ~ !~1 - ~ -e
- - S T
~611
T-DES
,.~~~
~-~ :!!' :ttl un lt.1 tH ·~' 10~ !'' -t:~• ""'' ..!' ' 11 u
X - a i [mm)
Figure 6: Profiles for the turbulent kinetic energy for the 25 ° case
From the standpoint of the steady state flow topology, it appears that the
k-E model is in closer agreement with the experimental data than the SST
model. For the 2) 0 , the experiments show a topology associated with high lift
and strong trailing vortices. The confined separation bubble does not alter the
overall flow topology. RANS models, which predict no separation, are there-
fore in closer agreement with the experiments. This does however not imply
that the k-E model is as superior turbulence model for this type of flow. The
agreement in flow topology is a result of the failure of the model to predict the
separation and not of its ability to accurately predict reattachment. The ability
of a turbulence model of predicting the onset of separation is still the first pri-
ority for aerodynamic flow simulations and cannot be traded against other
model characteristics.
- Drag force
lO ide force
LifL force
30
Figure 8: Flow structure on the slant for SST (left) and SST-DES-CFX-Fl (right)
More details can be seen in Figure 9 showing the velocity profiles at a plane
at 180 [mm] (the half width of the body is 194 [mm]). In the symmetry plane
(see Figure 5 and Figure 6), the advantage of the DES simulation is not very
pronounced, but in the off-symmetry plane, the change in flow topology and
the improvement in the predicted results is apparent. The unsteady simula-
tions give a first indication of the underlying mechanism of the large-scale un-
steadiness of the flowfield. It is observed that the separation zone has a strong
lateral movement, which at some instances interacts with the side vortices,
leading to a vortex-breakdown. This is shown for a certain instance in time in
Figure 10. The right side vortex is unaffected by the separation, whereas on the
left side of the body the separation interacts with the vortex leading to a vor-
tex-breakdown. The time value correlates with a maximum of the side force
(compare Figure 7). While it is likely that this is the main mechanism for the
experimentally observed global unsteadiness, it is not (yet) of sufficient
strength to reproduce the experimental fluctuation level. It is not clear if this is
a question of an insufficient length of the time integration, or a shortcoming
of the DES model.
350 F.R. Menter and M. Kuntz
!6,.\ -1•1 -!!\ -~' -181 lflol ·l-ll -In -103 -3J ~\ ..... , ·21 ·l 11 )I
X - axi [mml
Figure 9: Comparison of velociry profiles at Y=l80[mm] for SST and SST-DES-CFX-FI model
for the 25 o case
chances of capturing the essential flow features. For the present application,
the flow separation takes place at a corner - it is therefore not of major conse-
quences if the DES limiter is activated already in the region upstream of the
separation line, as the boundary layer there is not exposed to an adverse pres-
sure gradient. Grid induced separation is therefore not a problem for the pre-
sent geometry.
However, for a general car geometry, where the separation can be induced
by an adverse pressure gradient from a smooth surface, standard DES would
face severe difficulties. One choice is generating a fine grid, which would allow
the activation of the DES formulation in the separation zone, at the danger
that the DES limiter will change the RANS part and produce grid-induced
separation. The second choice is the use of a coarser grid and thereby delay the
DES impact far downstream of the separation line and miss the physics of the
flow. As the separation line is not known during grid generation, it is difficult
to imagine how a suitable DES-grid could be generated for such a flow.
The use of the proposed zonal DES formulation, based on the SST model
blending functions, will at least reduce the risk of a grid induced separation
occurrence, however, it still has to be tested if the blending function will
switch quickly enough from RANS to DES to activate the DES mode for a
pressure induced separation. Nevertheless, this approach is preferred, as it re-
duces the influence of the user (grid) on the solution.
Conclusions
CFD simulations have been carried out for the generic Ahmed car body at
25°. Strong turbulence model differences where observed for this case. The k-f
model produced an attached flow over the entire slant, whereas the SST model
predicted a fully stalled flowfield. Both solutions are in disagreement with the
experiment, where the flow separates and reattaches at about 50% of the slant.
DES simulations based on a modified version of the SST-DES formulation,
resulted in a significant improvement of the solution compared to the SST-
RANS model. Instead of a fully stalled flow, the time-averaged DES solution
shows a confined separation zone and strong trailing vortices, associated with
the experimentally observed flow topology. The DES solution also gives first
insight into the mechanism driving the strong unsteadiness of the flow, as ob-
served in the experiments. The most likely explanation is a strong lateral
movement of the separation zone, which interacts with the side vortices, lead-
ing to periodic vortex-breakdown. It could not be determined if the full un-
steady effect could be obtained by further continuation of the simulation, due
to constraints in computing power.
352 F.R. Menter and M. Kuntz
Acknowledgment
This work was supported by research grants from the European Union under
contracts GRDl-2001-40199 (Flomania) and EVGl-2001-00026 (EXPRO).
The authors want to thank H. Lienhart from the University of Erlangen for
the provision of some of the figures concerning the physical interpretation of
the flow.
Literature
Ahmed S.R., Ramm G. and Faltin, G., 1984, "Some salient features of the
time-averaged ground vehicle wake" SAE technical paper 840300.
Durand, L., Kuntz, M. and Menter, F., 2002, "Validation of CFX-5 for the
Ahmed Car Body", CFX Validation Report, CFX-Val13/1002.
Esch, T., Menter, F.R. and Vieser W., 2003" Heat transfer predictions based
on two-equation turbulence models", 6'h ASME-JSME Thermal Engi-
neeringJoint Conference, March 16-20, 2003, Hawaii.
Johnson, D.A. and King, L.S., 1984, "A new turbulence closure model for
boundary layer flows with strong adverse pressure gradients and separa-
tion", AIAA Paper 1984-0175.
Johnson, D.A., Menter, F.R., and Rumsey C.L., 1994, "The status of turbu-
lence modeling for aerodynamics", AIAA Paper 1994-2226.
Launder B.E. and Spalding D.B., 1974, "The numerical simulation of turbu-
lent flow", Comp. Meth., in Appl. Mech. and Engng., 3, pp 269-289.
Lienhart H. and Becker S., 2003, "Flow and turbulence structures in the wake
of a simplified car model", SAE Technical Paper 2003-01-0656, Michi-
gan.
Menter, F .R., 1993, "Zonal two-equation k-w turbulence model for aerody-
namic flows", AIAA Paper 1993-2906.
Menter, F. R. Kuntz, M. and Bender R., 2003, "A scale-adaptive simulation
model for turbulent flow predictions", AIAA Paper 2003-0767.
Spalart, P.R. and Allmaras, S.R., 1994, "A one-equation turbulence model for
aerodynamic flows", La Rech. Aerospatiale, V.l, pp.5-21.
Spalart, P.R, Jou, W.-H., Strelets, M. and Allmaras, S.R., 1997, "Comments
on the feasibility of LES for wings, and on a hybrid RANS/LES ap-
proach", 1st AFOSR Int. Conf. On DNS/LES, Aug.4-8, 1997, Ruston,
LA. In Advances in DNS/LES, C. Liu & Z. Liu Eds., Greyden Press, Co-
lombus, OH.
Spalart, P.R., 2003, Private communication.
Strelets, M., 2001, "Detached eddy simulation of massively separated flows",
AIAA Paper 2001-0879.
Simulation of vehicle aerodynamics using a
vortex element method
Summary. Recent developments of the 3-D Lagrangian vortex element method for
bluff body flows are presented. In this approach attached boundary layer regions are
modelled using infinitely thin vortex sheets while Lagrangian vortex elements are
used for the separation regions and the wake. Preliminary results for the flow past
a simplified generic truck geometry are presented. Further developments, aimed at
the development of a hybrid Eulerian-Lagrangian solver, are briefly introduced.
1 Introduction
The numerical method used here is a combination of a Lagrangian vortex ele-
ment method (VEM) and a boundary element method (BEM). Vortex meth-
ods are based on the vorticity formulation of the Navier-Stokes equations and
on the fact that, for incompressible flows, it is sufficient to follow the evolution
of the vorticity field (the velocity can be recovered from the vorticity).
For bluff body flows, the vorticity is only present in the boundary layers
and in the wake. As vortex methods only require particles to carry the vortic-
ity, they require much less computational elements than grid-based methods.
It was only recently that direct numerical simulations (DNS) of 3-D bluff
body flows were performed using a vortex method (Ploumhans et al. [6]). Such
simulations were made possible because of the important advances made by
the "vortex methods community" in the past ten to fifteen years.
We present some recent developments on going aimed at the extension of
the vortex method as a tool for Large Eddy Simulations of bluff body flows.
First, we review the modifications that we introduced for computing 3-D flows
where the attached boundary layer regions are modelled using infinitely thin
354 G. Daeninck et al.
vortex sheets while the separation regions and the wake are modelled using
vortex blobs. We then present some results using this approach applied to
the flow past the Ground Transportation System (GTS, Gutierrez et al. [2]).
Finally, a new hybrid Eulerian-Lagrangian approach is briefly presented: the
near-wall regions are resolved using a grid based method while the vortex
element method is used for the convection-dominated part of the flow.
We here consider simulations where the thin attached boundary layer regions
are modelled using infinitely thin vortex sheet panels, while the vorticity in
regions where separation is allowed is captured using vortex particles. In prac-
tice, the body surface mesh is tagged in order to define the "attached" regions
and the "separation" regions.
Let us consider a set of M panels, discretizing the body surface, and N
vortex particles. Let Patt be the subset of "attached" -tagged vortex panels
and Pdiff the "diffusing" -tagged panels (i.e. where separation is allowed).
One time step then goes as follows:
1. Compute the velocity field, u, and its gradient, V'u, at particle locations
from w, the vorticity carried by the vortex particles and"'(, the vorticity
carried by the subset of attached vortex panels, Patt·
2. Update the particles' position, xi, and strength, ai, for the time step, 6t,
(1)
where the Particle Strength Exchange (PSE) scheme is used to treat the
viscous diffusion.
3. Enforce the no slip boundary condition at the wall:
a) Compute the slip velocity, lls!ip, underneath all vortex panels (Patt U
Pdiff) from the vorticity carried by the vortex particles and Patt·
b) Compute the vortex sheet, 6"'(, that cancels the slip velocity at the
solid boundary.
Simulation of vehicle aerodynamics using a vortex element method 355
c) For panels belonging to Patt, the new panel strength is set to /;+6.1;·
For panels in Pdiff, the vortex sheet is equivalent to a vorticity flux that
must be emitted into the flow for the time 6t: v ~';: = !f{- (where v
is the molecular viscosity, and n is the local outward-pointing normal
to the body surface). This amounts to modifying the strength of the
particles close to these panels as in [6].
4. Redistribute the particles (when necessary): the old set of particles is
replaced by a new one, in which the particles are on a regular lattice.
fs J~ Wpart dz dS
Tpart = S
For each panel, we approximate this quantity as the discrete integral of Wpart
over an hemisphere of radius R centered on the panel divided by an "equiva-
lent" surface Seq:
Ldi <R 0 i
Tpart rv S
eq
where d; is the distance of a particle to the panel center, ai is its strength,
and Seq was chosen equal to 1r R 2 . It is clear that R should be greater than the
local boundary layer thickness, but also remain fairly low in order to obtain
a good approximation of the local boundary layer strength.
Fig. 1. Boundary layer strength carried by the vortex particles (r part) and attached
to the vortex panels (r panel).
If one wants to perform simulations where ground effects are taken into ac-
count, ground panels are necessary in order to enforce the no-slip boundary
condition there. Furthermore, the wake vorticity has to interact in a viscous
way with the ground. This is true even if one is not interested in the wake-
ground effect (as in the far wake for example), due to the fact that the inter-
action of wake vorticity with a non-viscous ground rapidly leads to numerical
blowup. This observation leads to the conclusion that it is necessary to use
"diffusing" ground panels over the whole wake region.
Simulation of vehicle aerodynamics using a vortex element method 357
In order to avoid the extra cost of ensuring a few layers of particles above
the whole wake region at all times (to capture the diffused panel vorticity),
we introduce "automatic" diffusing panels: as soon as vortex particles travel
above these panels, their vorticity is diffused onto them (they act as "diffusing
panels"); otherwise the panel vorticity remains attached.
The vortex sheet 6.1, necessary on the body surface in order to cancel the
slip velocity, is found as the solution of a boundary integral equation. The dis-
cretization of this boundary integral equation results in a system of 2M linear
equations for the vortex panels (i.e. find 6.ri for each panel) which is solved
iteratively. The use of multi pole expansions for vortex panels reduces the cost
of each iteration to O(M log M). The efficiency of this iterative "multipole-
based" solver was improved as follows:
• Choice of a good initial guess: the use of 6.ri = -2 nix u~lip (exact solu-
tion for coplanar panels) has proved to lead to faster convergence compared
to the former method where the previous time step solution is used. This
is mainly because the vortex sheet is only a small correction at each time
step dominated by high frequency variations as vortex particles come lo-
cally close to panels or not. These high frequency variations call for local
corrections which are well estimated by the exact solution for coplanar
panels.
• Let 6.1r be the corrective vortex sheet at iteration r. In order to evalu-
ate 6.1r+ 1 one must compute the slip velocity, uslipr, induced by 6.rr.
Now, instead of computing the slip velocity induced by the whole vortex
sheet 6.1r at each iteration we can compute the slip velocity induced by
(6.rr - 6.1r- 1 ) and simply add it to uslipr-I (linear problem). This is
advantageous because of the use of multipole expansions: as we converge
to the solution, (6.ri- 6.1~- 1 ) gets smaller and thus each iteration takes
less time to compute as we can make a more extensive use of multipole
expansions.
• When using "attached" panels, compute the vortex sheet correction 6.1
only, even for "attached" panels (instead of solving the system for 1).
This correction is much smaller compared to 1 (panel's attached vorticity)
which, again, is advantageous because of the use of multipole expansions.
These modifications reduced the global computational time for the panel
solver by a factor 2 (up to 5 in some cases).
For a physical flow, one must ensure that the total vorticity remains equal to
zero for all times. As the global time integration scheme is not conservative,
358 G. Daeninck et a!.
the sum of the particle strengths is set back to zero at each time step, by
distributing the difference over all particles.
However, when "attached" vortex panels are used, this simple scheme can-
not be used anymore as one must take the "attached" vorticity into account.
Several approaches for enforcing the total vorticity where investigated.
• Our first approach was based on the fact that the vortex flux at the wall
must be divergence free. At a global level, this leads to the following con-
1
straint :
61 dS = 0.
Sbody
We track the total amount of circulation diffused onto the vortex particles
(Fshed) over time. The particle strengths are then adjusted to Fshed at each
time step.
Despite the sound physical foundation of this approach, it performs very
poorly in time. Actually, the basic hypothesis, Is
body
61 dS = 0, is true
only for a divergence free vorticity field, which is not the case in practice.
Practically, as we modify 61 in order to satisfy the total vorticity con-
straint, the vortex sheet no longer cancels the slip velocity and particles
soon penetrate inside the body which finally leads to numerical blowup.
• In the second approach we do not place any constraint on the vortex
sheet. We simply ensure, at each time step, that the total vorticity remains
equal to zero by setting the sum of the particle strengths to minus the
total "attached" vorticity, - Isbody
1 attached dS (carried by the "attached"
vortex panels).
This approach produces good results, even for long time simulations.
2.6 Vorticity flux at the wall for under resolved boundary layers
When performing simulations where the boundary layers are no longer well
resolved, the characteristic diffusion distance becomes small compared to the
local particle size. This can become critical as the panel diffusion step must en-
sure that the total flux of vorticity is distributed to neighbor vortex particles.
The use of the classical diffusion scheme with the correction for conservation
leads to a very noisy distribution of the vorticity on the neighbor particles.
This situation degrades even further for simulations with ground effect where
a decreasing resolution is used in the wake.
Actually, in these under-resolved computations, one "simply" wants to
transfer the whole vorticity flux to the first layer of particles above the wall.
We therefore "relax" the local Reynolds number at the wall with respect to
the local grid size to ensure that the vorticity flux can be "sufficiently well"
captured. This scheme gives produces results because of its diffusive nature
which helps to remove the high-frequency noise present in the computed vor-
ticity flux.
Simulation of vehicle aerodynamics using a vortex element method 359
The vortex particle method has no built-in control for keeping the vorticity
field divergence-free as time evolves. When the computation is well resolved
(as it has been shown in various DNS results for the flow past a sphere), the
divergence of the vorticity field appears to be kept "naturally" to a fairly low
value, even for long time simulations. However, as soon as one wants to push
the resolution towards its limits, the divergence problem becomes a major
issue.
In fact, one can state that an increasing divergence of vorticity field is the
sign that the computation is not well resolved locally and that energy is ac-
cumulating in small scales of the flow: subgrid-scale stress modelling becomes
necessary.
It is clear that the redistribution step plays a major role here, as it is
essential to keep a good representation of the vorticity field. One can go even
one step further by using the redistribution scheme as a form of subgrid-
scale model: it can easily be shown that the low order A 1 scheme (which only
conserves moments of order zero and order one of the vorticity field) has a
viscous-like behavior. The diffusion effect introduced by a A1 redistribution
scheme applied every n time steps is roughly equivalent to an effective viscosity
given by:
where h is the local grid size and 6.t the time step.
In the vicinity of the body, the resolution is h/ D "' 0.035t. Although
quite crude, the method enabled us to perform long time simulations for the
flow past a hemisphere (T = 75) with a relatively low number of particles
(Fig. 2). The present results show that the vortex method appears to be able
to capture the dynamics of the complex vortex structures with a reduced
number of computational elements and to reproduce the qualitative behavior
of a turbulent flow past the hemisphere.
However, due to its highly diffusive nature which is only controlled by the
local grid size, the redistribution interval and the time step, one cannot rely
on a low order A1 redistribution scheme to perform satisfactory subgrid scale
t For information (although they should not be compared), the case of DNS past
a sphere at Re = 300, requires a grid twice as fine in all directions. This case
was studied in [6] where it was shown that the "mesh Reynolds number", Reh =
lwlh2 /v, never exceeded 6 in the far wake and 2 in the boundary layer and near
wake regions. For the viscous vortex method, this corresponds to a well resolved
DNS in the far wake and a very well resolved DNS near the sphere surface.
Furthermore, the results also compared favorably to those obtained by other
authors using different numerical methods. For further discussion about the DNS
method, as well convergence studies of the method, refer to [5] and [6].
360 G. Daeninck et al.
Fig. 2. F low past a hemisphere using the 1h redistribution scheme. Only particles
with Ja; I above an arbitrary threshold are shown. Top: view of the vortex particles
and their strength at an early stage of the flow (T = 6). One can see the vortex ring
behind the body becoming unstable. Bottom: view of t he developed turbulent flow
at T = 75.
AUached panels
ProgreSSiVely diffUSing panets
• Diffusing panels
Fig. 3. Left : "Numerical setup" for the flow past the GTS. Right: Zoom on body
surface discretization (vortex sheet panels) ; panels are colored by diffusion ratio.
All the recent developments presented in the previous sections where tested
on the flow past the GTS. This simple geometry was used because modelling
the flow using infinitely thin attached boundary layers along the truck and
allowing separation on the back-face is a good approximation of the physical
flow (at a zero degree yaw angle). The "real" ground effect was also taken into
account in this simulation: the truck is travelling above a fixed ground.
As can be seen in Fig. 3, the whole front part of the truck was tagged as an
attached boundary layer region. Then progressively diffusing panels are used
in order to ensure a smooth transition from the infinitely thin boundary layers
to the region where diffusing panels are used. A close up view of the panel
discretization is shown in Fig. 3. One can see that a variable panel size is used
for computational efficiency: coarse panels can be used in the attached regions,
whereas the diffusing (and progressively diffusing) panels' size is chosen equal
to the local particle size. On the ground, the panels were tagged as attached
except for the wake region: the near wake region uses diffusive panels (with a
transition region), and the far wake uses auto-diffusing panels.
We use a redistribution mapping which smoothly goes from a constant
lattice (in the near body region) to an exponentially growing lattice (in the
far wake). At each redistribution, the center of the mapping is "displaced"
randomly around its nominal position: this reduces the influence of the ar-
bitrary intersection between the redistribution lattice and the body. In the
near body region, the particle size is h = 0.023 W (W is the width of the
truck). The time step is 6T = 0.01 W/UcTS (UcTs is the velocity of the
GTS). Redistribution is performed every 5 steps (A 3 scheme). The relaxation
scheme for the divergence of the vorticity field is applied every 50 steps. The
simulation was carried out up to T = 16, at that time the number of particles
362 G. Daeninck et a!.
Fig. 4. Flow past the GTS with viscous ground effect in the wake. Top: The
regions of non zero vorticity are shown in three slices behind the truck: each slice is
colored by the y-component of vorticity. Bottom: Zoom on the vertical midplane
slice behind the truck.
was "' 1 000 000. The computation ran 50 hours on 8 Pentium 4 processors at
2.4 GHz (Beowulf Linux cluster).
Defining a Reynolds number for these under-resolved computations has
little meaning (no quantified subgrid scale modelling, infinitely thin boundary
layers). We can however mention that the viscosity coefficient used for the PSE
scheme was VpsE/(UcysW) = w- 4 .
In Fig. 4 one can see the vorticity field in a slice behind the GTS: it
clearly shows the development of a truly turbulent flow in the wake. These
results feature much more small structures when compared (qualitatively) to
simulations where the A1 redistribution scheme was used.
Simulation of vehicle aerodynamics using a vortex element method 363
Lagrangian subdomain
Eulerian subdomain
The results for the flow past the GTS are certainly preliminary; no con-
vergence studies have been carried out yet. We believe however that these
results already capture a significant fraction of the turbulent wake dynamics,
including the interaction with the ground and with the back face of the GTS.
Hence we hope that they will allow to obtain global quantities, such as the
mean pressure drag, at a fair level of accuracy.
In the previous section, an "over-simplified" approach was used for the bound-
ary layers: they were modelled by infinitely thin vortex sheets in a priori de-
fined regions. It is clear that this approach can only provide physical results in
simple cases. In more general situations, the evolution of the boundary layer
has to be simulated in order to determine where the separation dynamically
occurs.
The VEM suffers from two drawbacks in the boundary layer regions which
prevent it from working efficiently in these regions:
• The VEM uses isotropic computational elements whereas, in the boundary
layer regions, the strong gradients in the direction normal to the wall would
allow highly anisotropic elements for computational efficiency.
• Although the VEM performs particularly well for flows dominated by con-
vection (due to the implicit treatment of the convective term), it is less
suited to flexible and accurate treatment of the no-slip boundary condition.
These drawbacks currently limit the applicability of the VEM to relatively
low Reynolds number flows or flows where an excessive degree of modelling
must be introduced in the boundary layer regions. Approaches along the lines
364 G . Daeninck et a!.
Fig. 6. Vorticity field for the flow past cylinder at Re = 3000. One can see the
Eulerian subdomain covering the near-body region with the fine grid, and the coarse
Lagrangian subdomain covering the whole computational domain (the near wall-
region is hidden by the Eulerian subdomain).
Eulerian subdomain uses a fine grid in order to capture well the detailed dy-
namics in the near-wall region. A good behavior of the algorithm was observed
in terms of robustness and information transfer between subdomains. Further
investigations will focus on quantitative validations of the method. The next
steps will be to extend the method to 3-D, to introduce a DES approach in
the Eulerian subdomain, and a LES model in the Lagrangian subdomain.
5 Conclusions
Several contributions to the 3-D vortex and boundary element method where
presented. The main feature of the resulting method is its capability of mod-
elling "high" Reynolds number bluff body flows with prescribed attached I
separation regions at a reduced computational cost. Preliminary results for
the flow past a simplified truck geometry showed the promising capabilities
of the method.
An important issue in vortex methods is the vorticity field divergence. It
was addressed here through the use of a relaxation scheme. This relaxation
scheme also acts as a filter which eliminates the energy that gets accumulated
in the small scales. The next step will be to introduce a more involved subgrid-
scale model I procedure.
Finally, a new hybrid Eulerian-Lagrangian vortex method was presented.
In this method, the Eulerian subdomain is used as a near wall flow solver
"feeding" the Lagrangian domain.
We believe that the future of vortex methods for bluff body flows could be a
combination of the "simplified" 3-D vortex method (as presented in Section 2)
using a grid-based approach to solve the boundary layer and early separation
regions: a type of detached eddy simulation (DES) approach.
References
1. Cottet G-H, Koumoutsakos P (2000) Vortex Methods: Theory and Applications.
Cambridge Univ. Press, Cambridge, UK
2. Croll R, Gutierrez W, Hassan B, Suazo J, Riggins A (1996) Experimental In-
vestigation of the Ground Transportation Systems (GTS) Project for Heavy
Vehicle Drag Reduction. SAE paper 960907
3. Leonard A, Shiels D, Salmon J, Winckelmans G, Ploumhans P (1997) Recent
Advances in High Resolution Vortex Methods for Incompressible Flows. Proc.
13th AIAA Computational Fluid Dynamics Conf., Snowmass Village, CO, June
29 - July 2, 1997, AIAA #97-2108
4. Ould-Salihi M, Cottet G-H, El Hamraoui M (2000) Blending finite-difference
and vortex methods for incompressible flow computations. SIAM J Sci Comp
22:1655
5. Ploumhans P, Winckelmans G (2000) Vortex methods for high resolution sim-
ulations of viscous flow past bluff-bodies of general geometry. J Comput Phys
165:354
366 G. Daeninck et al.
Thomas Gielda
Ronald Dupree
Abstract
To gain insight into the ventilation needs for an enclosed engine compartment
of an off-road machine, a prototypical test-rig that includes an engine and
other installation hardware was built. Well controlled experiments were con-
ducted to help understand the effects of ventilation air flow on heat rejection
and component temperatures. An assessment of 1-D and 3-D simulation
methods was performed to predict underhood ventilation air flow and compo-
nent temperatures using the experimental data. The analytical work involved
development, validation, and application of these methods for optimized ven-
tilation air flow rate in the test-rig. A 1-D thermal-fluid network model was
developed to account for overall energy balance and to simulate ventilation and
hydraulic system response. This model was combined with a 3-D CFD model
for the ventilation air circulation in the test rig to determine the flow patterns
and the distributed surface heat transfer. The tests conducted at Caterpillar
and the complementary analyses performed at Argonne provide an opportu-
nity to understand the isolated effect of ventilation air cooling on underhood
thermal management.
Introduction
Construction equipment and other types of heavy vehicles have common un-
derhood thermal management challenges: restrictive enclosures and ever-
increasing variety of heat sources. But off-road machines have rather unique
additional underhood thermal management issues such as
Experimental Study
A U .S. EPA Tier II emis-
sions level engine (Fig 2)
was installed into a
mockup representing a
typical medium size off-
highway machine with a
full engine enclosure sepa-
rated from the cooling fan
by a solid wall [1] . The en-
closure was constructed
from sheet metal and
tightly sealed at all seams,
but was not insulated. The
CAD model shown in Fig 2. Engine setup and enclosure frame without walls.
Fig.3 provides a perspective
on the enclosure and in-
let/outlet locations with respect to engine components. Consistent with a typi-
cal off-highway machine with this size engine, the enclosure dimensions were
100x140x140-cm3 • The 30x30-cm 2 inlet opening in front of the crank shaft
was used to supply ventilation air into the enclosure. A 30-cm diameter open-
ing at the top was connected to a variable capacity blower to draw air from the
enclosure, and the total flow rate throughout the enclosure was measured.
Analytical Studies
Computer simulations can improve the understanding of interactions between
the engine subsystems[2]. The main purpose of this study has been an assess-
ment of simulation methods that could be used in predicting underhood ven-
tilation air flow field and temperatures for an off-road machine. The work in-
volved development and validation of combined 1-D and 3-0 simulation
models of the Caterpillar test-rig. A 1-D thermal-fluid network model was de-
veloped to account for overall energy balance and simulate cooling system re-
sponse using the commercial software Flowmaster[3]. A 3-D underhood
model of the complex test rig was built using the commercial CFD software
Star-CD[4] to determine the flow paths for the ventilation air system and the
surface heat transfer coefficient.
In order to capture the ventilation air flow distribution at the enclosure in-
let accurately, a large inlet plenum (not shown in Fig.4) was also included in
the model to represent ambient conditions (pressure and temperature). The
desired flow rate through the enclosure was assured by imposing a proportional
uniform flow field at the plenum inlet as the boundary condition. The enclo-
sure outlet pipe was considered much longer than what is shown in Fig.4 and
Measurement and Analysis ofUnderhood Ventilation Air Flow 377
its top end was treated as a standard outlet boundary. The final CFD model
consists of 1.34 million fluid cells, with a 3 mm thick extrusion layer sur-
rounding the engine and enclosure surfaces to give a maximum y+ value of 200
for airflow ratio of 1.5.
Fig.4. CFD mesh of the test rig (a) cutaway view of the surface mesh (b) a cross section of the
volume mesh.
The ventilation air flow field in the test rig and the convective heat transfer
coefficient for the solid surfaces were obtained using the commercial CFD
software Star-CO. An initial parametric study for inclusion of the buoyancy
force in the thermal-fluid calculations revealed that the effect of density varia-
tions on the overall flow and temperature fields is negligible. Thus, the venti-
lation air flow field was simulated as a steady incompressible flow with energy
equation using the high-Re number k-epsilon turbulence model with loga-
rithmic wall functions.
As the most basic two-equation model, k-epsilon model is believed to pro-
vide a reasonable approximation of the time-averaged flow distribution over
the surface of the engine and its components in the test rig. A set of transient
calculations were also studied to investigate temperature fluctuations observed
during the experiments and assure that the calculated flow field is steady with
no oscillations. The results indicated negligible difference between the tran-
sient and steady state solutions. Five different inlet locations, each for five air-
flow ratios, were studied with the CFD model; however, only the results of
front inlet configuration (shown in Fig.4) are discussed here. The calculations
were performed on a linux cluster.
378 T. Sofu et al.
Key
Cc- !ant
Engmt
Au
Fig.5. 1-D network flow model of the test rig for front-inlet configuration.
Air flow paths in the 1-D model are based on 3-D simulation results. In the
air loop, the entering ventilation air is considered to gain heat as it passes
through individual surface points on the engine as shown in Fig.5. In the oil
Measurement and Analysis ofUnderhood Ventilation Air Flow 379
loop, after losing heat through the oil pan, the flow splits into three separate
branches (the turbo, the cylinder head, and the engine block) before returning
to the sump. In the coolant loop, the water cools the lubrication oil in the oil
loop and circulates inside the engine block and the cylinder head. The radiator
is simply modeled as a source with constant flow rate and with known inlet
temperature.
Interface between the 3-D CFD and 1-D Network Flow Models
Fig.6 shows the schematic of the sequential analyses with the 1-D network
flow and 3-D CFD models. The 1-D model requires flow rates and inlet tem-
peratures as the boundary conditions in the air and coolant loops and oil pump
speed in the oil loop to account for overall energy balance and predict the en-
gine component temperatures. In the 3-D thermal analysis, these predictions
are prescribed as surface temperature boundary conditions for various engine
components and enclosure walls, and they are used to calculate ventilation air
flow field and temperatures. The results of the 3-D CFD analysis are, in re-
turn, provided back to the 1-D model to improve component temperature
predictions by modifYing the air flow paths and heat transfer coefficients be-
tween the engine components and ventilation air. The typical values of esti-
mated heat transfer coefficients between the engine components and ventila-
tion air are found to vary in the range from 10 to 50 W/m 2 -K.
Mnt1PI Tr
,,,......,
+ I
Boundary Conditions: 1-D Network Flow Model Output:
Coolant flow rate and r------- using FLOWMASTER vSurface temperatures
inlet temperature Air temperatures
(All four loops)
Oil and coolant temps.
Oil pump speed
--
/
3-D CFD Model Output:
Boundary Conditions:
using STAR-CD Ventilation air flow paths
Air flow rate and inlet f----o (only for ventilation air and heat transfer rates
temperature flow inside enclosure) between engine and air
Energy Balance
X
•
total energy) is attributed to E
the energy convected from ex- ~0.7
X
•
-o
terior of the enclosure walls. (J
~0.6
X
A comparison of the meas- E 0 •
urements and 1-D model pre- z 0.5
dictions for the enclosure outlet • X X
Fig.9. The calculated ventilation air flow field and temperature distributions on a vertical plane
that intersects the front inlet.
~0.8 ~0.
2 3
~06 ~0.6
=
~ ~
"8N 0.4 ]OA
-;::; -;::;
:: 0 E
§0.2 5 0.2
z z
0
... ... .,... ...u
e u u 0.
<0 cJ.)
E <0
=
0
U:
"'
u
c:.:: "'u
c:.:: <
~
<
~
t::
·co ' Oi,) 0 0
t::
·;::: E
::>
u
Ll.
u u ~ u
t::
tlJ t.5
0
u u0
"'
u Cll
g
"(i);;;
0 CI:l
~ :=! ~ u ~ g E g E g
Cll Cll Cll t.ll 0::: ... 0 0
6
e
u
6 .!:;:
.,...
:;: u -"' 3 .!:;:
::>
~
:::>
.::c: 6
"'
.<::
X
= "
.<::
>( Ll. ~ ;; 6
t.ll t.ll ~
Fig. II. Comparison of temperatures between measured data and model predictions: (a) ventila-
tion air temperatures, (b) coolant and oil temperatures.
Although the discrepancies are generally small, the attempts to resolve them
are part of the overall modeling effort to provide a better description of the
underhood system. For example, based on the CFD results, the discrepancy for
the exhaust-side rear ventilation air temperature is attributed to a local recir-
culation zone in that region. However, since the estimated temperature is small
and its impact on overall temperature distributions is negligible, a modification
to the network flow model for the front inlet configuration is not considered
to be essential.
Measurement and Analysis of U nderhood Ventilation Air Flow 383
Conclusions
Experiments were conducted to gain insight into the ventilation air flow needs
for an enclosed engine compartment of an off-road machine. These laboratory
experiments were well controlled to provide good accuracy and to draw im-
portant conclusions on minimum ventilation flow requirements for maintain-
ing acceptable underhood temperatures. About 96% of the total fuel energy
was accounted for during the test. Underhood temperatures in the areas of
concern are found to be generally stabilized near an airflow ratio of two. Data
obtained were also used to provide boundary conditions and validation infor-
mation for simulation methods.
A combined 1-D and 3-D simulation methodology was developed for op-
timization of engine compartment ventilation air flow. The air flow field and
the rate of heat transfer between engine and ventilation air inside the enclosure
were determined with the 3-D CFD simulations. A 1-D network model was
built by discretizing the various fluid paths and the solid metal structure in the
system. Once the ventilation air flow paths and heat transfer coefficients were
determined with CFD, the 1-D network model with reduced complexity was
used to simulate thermal interaction of the engine structure with the air, cool-
ant, and oil flow. The results indicate that the temperatures and distributed
heat rejection rates can be estimated within reasonable accuracy when 3-D and
1-D models are used in combination.
Acknowledgements
This work was completed under the auspices of the U.S. Department of En-
ergy Office of FreedomCAR and Vehicle Technologies. The submitted manu-
script has been created by the University of Chicago as Operator of Argonne
National Laboratory ("Argonne") under Contract No. W-31-109-ENG-38
with the U.S. Department of Energy.
References
Robert F. Kunz
Narneer Salman
Abstract
Computational Fluid Dynamics (CFD) analyses were performed for an ar-
mored tank engine compartment cooling flow. Large hybrid unstructured
meshes (2.5-3.0x10 6 cells) were constructed using the ICEM-CFD grid gen-
erator. The flow and convective heat transfer field were computed using an in-
house CFD code, NPHASE. The commercial software package,
RADTHERM was utilized to incorporate radiation heat transfer within the
simulations.
Two steady operating conditions and one engine-off cool-down transient
were analyzed. Specifically, the conditions analyzed were open throttle (here-
after OT), Tac Idle (TI) and engine off soakback (SB). OT and TI were run
with and without convection heat transfer employed in the radiation assess-
ments to provide best-estimate and conservative peak temperature predictions
respectively. SB was run transiently using fixed heat transfer coefficients ob-
tained from NPHASE analysis.
Results are presented for the simulations performed, with emphasis placed
on peak temperatures of several design critical elements.
Software Tools
ICEM-CFD (2000) is a commercial geometric modeling and mesh generation
package that has been widely employed in the automotive industry to accom-
modate the very complex geometries associated with underhood thermal man-
agement analysis.
The NPHASE CFD code was developed by the author and several col-
leagues and is described in detail in Kunz et. al. (2001), Antal et. al. (2000)
386 R.F. Kunz and N. Salman
and Yu et. al. (2001). The code is fully unstructured and supports arbitrary
element types (the meshes employed here utilize hexahedra, tetrahedra, prisms
and pyramids). A parallel implicit, pressure-based segregated solution proce-
dure is employed. The code can predict steady state and time dependent flows
and employs higher order temporal and spatial discretization. A range of
physical models are implemented in the code. Those implemented in the pre-
sent work are:
• High Reynolds number k-£ turbulence model
• Perfect gas compressibility (with buoyancy)
• Porous media
• Turbomachinery capability (including body force modeling).
• Specified temperature, heat flux, heat transfer coefficient and various spe-
cialized conjugate heating boundary conditions
NPHASE does not contain an on-board radiation heat transfer modeling
capability. For this, the commercial package RADTHERM
(http:/ /www.thermoanalytics.com) was employed. This software is widely used
for underhood thermal management analysis in the automotive industry. As
described below, heat transfer coefficients and fluid temperatures, obtained
from NPHASE analyses, were imported into RADTHERM. Based on speci-
fied material properties, RADTHERM employs a view-factor based algorithm
to determine the convection-conduction-radiation heat balance on all solid
surfaces in the domain, resulting in the final temperature predictions of inter-
est.
ICEM-CFD generates a "CGNS" file that includes all vertex, edge, face and
element data defining the hybrid unstructured mesh, as well as "patch family"
designations that can be used to define boundary conditions, as well as volume
family identifiers that can be used for localized element based treatments
within the flow solver (e.g., body forces within the fan). A CGNS file reader is
available for NPHASE, which accommodates this richness afforded by the
CGNS format.
The ICEM/NPHASE meshes employed for the CFD analyses carried out
here have just under 5.0xl 0 5 triangular wall faces. This very fine resolution is
consistent with fine grid requirements for the fluid-thermal CFD analysis.
However, this is far more than necessary for requisite accuracy in the
RADTHERM analysis, and would require processor weeks to even run in
RADTHERM (which is not currently parallelized). Accordingly, a procedure
was developed to "coarsen" the CFD surface so as to the reduce fidelity of
viewfactor and radiation patch simulation. This process involves: I) Using
ICEM to coarsen surface meshes (while retaining good resolution on impor-
tant parts), and to generate a Patran Neutral file defining the new surface
mesh, 2) Running the Thermoanalytics (vendors of RADTHERM) tool, map-
convbc, to "interpolate" (closest point) the fine mesh CFD surface solution
(heat transfer coefficients and film temperatures) onto the coarser model, and
to output a Patran Neutral file with these interpolated values, 3) Importing
these interpolated H, T 61 m in the RADTHERM analysis to define convection.
Flow Field and Thermal Management Analysis 387
Modeling Details
The general configuration and flow path of the engine compartment is shown
in figure 1. In figure 2, two views of the ICEM model are shown. Note that
for TI and OT an artificial exit extension was installed to accommodate the
highly recirculating fan exhaust flow within the cooling fan housing.
Grids
For OT and TI, a 2,951,279 element mesh was constructed. The mesh em-
ployed 2,804,579 tetrahedra and 146,700 prisms. A detail of the ICEM mesh
in the vicinity of the engine is shown in figure 3a, illustrating the complexity
and high geometric resolution of the model. Further indication of the high
resolution of the present mesh is observable in the part surface grid plots given
in figure 13. For the SB case, all elements downstream of the top of the heat
exchanger pack were deleted and the face on the top of the heat exchanger
pack itself was re-designated as a wall (see figures 1 and 8 for reference). This
resulted in a somewhat smaller mesh of 2,424,411 tetrahedra and 240 prisms.
By virtue of the relatively low Reynolds numbers encountered in the engine
compartment, tetrahedra were deemed adequate for resolution of most wall
layers. An average wall spacing (i.e., wall adjacent element volume centroid to
wall face centroid distance) for all wall faces was 1.6mm. This gave rise to an
average y• value for all wall adjacent cells of 73 and 39 for OT and TI runs re-
spectively, consistent with the high-Reynolds number shear and thermal wall
function turbulence modeling employed. For the SB case, the average value of
y• was approximately 7. In all simulations, those elements for which y• drops
below 10 employ a two-layer wall-function treatment consistent with the pres-
ence of the cell centroid in the laminar sublayer.
Domain Decomposition
ICEM outputs a CGNS file which is read into a sequence of front-end utilities
which, among other tasks, implements domain decomposition using the freely
available METIS (2001) partitioning software. All of the NPHASE simula-
tions were performed on a LINUX cluster of 1GZ Pentium IV processors, us-
ing 24 processors. The 24 domain METIS partitioning for the OT and TI
simulations is shown in figure 3b.
nolds number is determined based on L and the bulk velocity of the engine air
based on nominal air-box cross-sectional area and given engine air mass flow
rates at OT and Tl respectively.
The exhaust duct is treated as a constant temperature surface for OT and
TI. A number of the compartment parts are "2-sided", that is, they have flow
on both sides. These "sheet-metal" pieces are treated as infinitely thin internal
boundaries. A conjugate heat transfer boundary condition is applied for these
pieces where heat flux and wall temperature are constant on each side of such
faces. These pieces include the cooling fan housing and the heat exchanger
housings. Adiabatic boundary conditions were employed for all other surfaces.
Bulk modeling is employed for the heat exchangers and cooling fan. As il-
lustrated in figure 8, design values of pressure drop and temperature rise are
available for each of the three heat exchangers. Each of these devices is meshed
independently using tetrahedra, such that each element is uniquely defined as
being within one of the three coolers. Body forces are added to the momentum
equations to establish the correct pressure drop and flow straightening. Specifi-
cally, in the flow direction a force is added: FY = m1ocal;(Vrer/2), where
rillocal is the local cell's mass flow rate and; is a loss coefficient. Generally, ; is
determined from empirical correlations for loss mechanisms associated with
inlet, core, acceleration, and exit losses, using a suitable definition of reference
velocity, Vrer· In the present work, since Llp and rillocal are known, ; was itera-
tively determined through several solution restarts such that the desired Llp was
matched to requisite accuracy. In the two cross flow directions, the loss factor
was increased by a factor of 10 to "straighten" the flow. This is illustrated with
a velocity vector plot in figure 9. The accuracy with which the design pressure
drops were matched is shown in figure 11.
Heat addition to the air flow in the three coolers was accommodated in a
consistent fashion. Specifically, local heat addition sources were added to the
enthalpy equation based on a local energy balance: q 1ocal = Ih1ocaiECpi\THx'
where L\THX = Tair - Tcoolant , Cp is the specific heat of the air and E is an un-
known cell "efficiency". For each exchanger we have available approximate
values for q,orai and Ll T HX> which when substituted into the equation above
yield estimates for E for each of the three coolers. As with the loss coefficients,
E was then iteratively refined through several solution restarts such that the de-
sired LlT was matched to requisite accuracy. The accuracy with which the de-
sign temperature rises were matched is shown in figure 12.
The fan was also modeled using a bulk representation. Specifically, ap-
proximate machine rotation rates and pressure rise across the fan were avail-
able. Since the mass flow rate and flow path are fixed, a suitable body force
distribution could be designed. A design code at Penn State Applied Research
Lab was utilized to generate tangential, axial and radial forces through the me-
ridional plane of the fan. These were then distributed onto the NPHASE grid
using bilinear interpolation. Elements of this procedure are illustrated in figure
10. The accuracy with which the design pressure rises were matched is shown
in figure 11.
390 R.F. Kunz and N. Salman
Results
As already indicated, three sets of runs were made: OT, TI, and SB. In this
section we summarize the results obtained. First, details of the CFD simula-
tions are presented, followed by the RADTHERM results.
Conclusions
The analyses carried out under the present effort involved the application of state-
of-the-art grid generation, CFD and radiation heat transfer analysis software. Large
high quality hybrid unstructured meshes were constructed using ICEM-CFD. The
NPHASE CFD code was employed to solve the flow and convective thermal trans-
port. The commercial radiation analysis package, RADTHERM, was employed in
assessing the radiation dominated peak temperatures in the system.
References
Kunz, R.F., Yu, W.S., Antal, S.P., Ettorre, S.M. "An Unstructured Two-Fluid
Method Based on the Coupled Phasic Exchange Algorithm," AIAA Paper 2001-2672,
Proc. 15th AIAA Computational Fluid Dynamics Conference, Anaheim, CA, June,
2001.
Antal, S.P., Ettorre, S.M., Kunz, R.F., Podowski, M.Z. "Development of a Next-
Generation Computer Code for the Prediction of Multicomponent Multiphase
Flows," presented at the International Meeting on Trends in Numerical and Physical
Modeling for Industrial Multiphase Flow, Cargese, France, September 27, 2000.
Yu, W.S., Kunz, R.F., Antal, S.P., Ettorre, S.M. "Unstructured Rotor Stator
Analysis of Axial Turbomachinery Using a Pressure-Based Method", ASME Paper pre-
sented at the International Mechanical Engineering Congress and Exposition, New
York, NY, November, 15,2001.
Kreith, Frank Principles of Heat Transfer, Harper and Row, New York, 1973.
Schlicting, Hennann Boundary Layer Theory. McGraw-Hill, New York, 1968.
Touloukian.. Y.S., DeWitt, D.P. (eds.) Thermal Radiative Properties: Metallic
Elements and Alloys, Vol. 7 of Thermophysical Properties of Matter, Plenum Press,
New York, 1970.
METIS Version 4.0 documentation, 2001.
ICEM CFD Software User Manual v4.1, ICEM CFD Engineering, Berkeley, CA,
2000.
396 R.F. Kunz and N. Salman
Front
Airflow
~~~~~---- Pa
lntak1
Grille
Fig. l. Sketch of engine compartment configuration and cooling air flow path for OT & T I.
Fig. 3. a) View of the ICEM engine compartment model illustrating the complexity of the con-
figuration . b) View of24 domain METIS partitioning for engine compartment model.
Flow Field and Thermal Management Analysis 397
mas flow 05
to'
• .li"
to'
" .,~
pressure -l '8
5
25
.J
Fig. 4. Representative NPHASE convergence history for engine compartment analyses. Shown
are pressure and velociry residuals, and global mass flow rate through compartment (outflow-
inflow).
Auld uD
"Cbhf" b..."1.rkside
lhsiltttU
Fig. 7. Illustration of airbox and specialized conjugate hearing boundary condition employed in
NPHASE.
Transmission o il cooler :
\pi .p 1 : 1 31
Fig. Sa. Illustration of three heat exchangers and design specifications for their respective pres-
sure drop, l1p/l1p"r' at OT & TI operation.
Electronics cooler:
'\T' T =100
Transmission o il cooler:
\T T =
9.12
Fig. Sb. Illustration of three heat exchangers and design specifications for .their respective tem-
perature rise, !1T/!1T"r' at OT & TI operation.
Flow Field and Thermal Management Analysis 399
Fig. 9. Predicted velocity vectors just above and through the three heat exchangers, illustrating
the straightening due to resistance modeling employed.
• (m
Fig. 10. Elements of fan body-force modeling employed. a) View of fan vicinity, b) output of
throughflow code.
Fig. 11. Errors in NPHASE vs. design pressure changes (dp NPHASE-dPoESIGN) across heat ex-
changers and fan .
400 R.F. Kunz and N. Salman
i + 0.0 °F
+ 0.0 "F
Fig. 12. Errors in NPHASE vs. design temperature changes (LlT NPHASE-Ll T DESIGN) across
heat exchangers.
Fig. 13. NPHASE and RADTHERM surface meshes for generator and airbox.
·--
~-- i.. ~ ·-
;·- r--- ~ r-
~ =- ,§~-~qo=u
,,_,_~· ~ ·A< -•<..•--•
Fig. 14. RADTHERM interface with view of engine compartment model (grey-scale).
Flow Field and Thermal Management Analysis 401
Fig. 15a. 3-layer transmission part treatment for soakback RADTHERM analyses.
Fig. 15b. ICEM-CFD family split employed for the transmission to distinguish between regions
with sump oil and those without under soakback conditions.
J l':ilyer tt-•gir•e
pAri
Fig. 16. Sketch of 3-layer engine part treatment for soakback RADTHERM analyses.
402 R.F. Kunz and N. Salman
Fig. 17. Elemenrs of OT NPHASE simulation. Streamlines shaded by temperature, and tem-
perature (T/T"r = 0.82) iso-surface (lighter surface enshrouding engine region).
Fig. 19. Rear-view in-plane velocity vectors, shaded by velocity magnitude (scale in VIV"r), for
OT (left) and TI (right) NPHAS E simulations.
Flow Field and Thermal Management Analysis 403
Fig. 20. Top-view in-plane velocity vectors, shaded by velocity magnitude (scale in V/V"r), for
OT (left) and TI (right) NPHASE simulations.
Fig. 21. Two cross-sectional views of density contours (p/p,mb;,.,), illustrating weak thermally
induced perfect gas density variations within the engine compartment.
1 1013)3
p ( h> )
101 3_')
Fig. 22. NPHASE SB simulation. Pressure contours, illustrating weak gravitational head rise
within engine compartment.
404 R.F. Kunz and N. Salman
Fig. 23. NPHASE SB simulation. Streamline field illustrating the complex thermally induced
free-convection field.
Ttr"'
.111
Fig. 24. NPHASE SB simulation. T/T"rcontours in vicinity of generator and exhaust duct.
Fig. 25. NPHASE SB simulation. TIT..r contours in vicinity of engine and air box.
Flow Field and Thermal Management Analysis 405
Fig. 26. NPHASE SB simulation. Free-convection induced velocity vectors in x-y plane in vi-
cinity of transmission, generator and exhaust duct.
Fig. 27. NPHASE SB simulation. Free-convection induced velocity vectors in y-z plane in vi-
cinity of engine and transmission.
T ,.,r'Trt: 0.686
•611 !----!:----+.,.~-~---!
llnl
Fig. 30. Sequence of predicted airbox surface temperature fields for SB case. Temperature con-
tour range is from 120°F to 240°F. tlc"rO, 2.5, 5, 10, 15, 20, 25, 30. Time history of peak pre-
dicted airbox surface temperature.
Flow Field and Thermal Management Analysis 407
.
.
718
~ Soakbatc:k Flow
.... No Soakback Flow
!2
....
§
E
]
"
.&n a 10 1~ lO ~ ~ ~
l 'tnr
Fig. 31. Sequence of predicted bulkhead surface temperature fields for SB case. Temperature
contour range is from l20°F to 225°F. t/t.,r =0, 2.5, 5, I 0, 15, 20, 25, 30. Time history of peak
predicted bulkhead surface temperature.
Fig. 32. NPHASE predicted total velocity vectors in an x-y plane midway between smallest re-
cuperator-bulkhead gap for SB cases. NPHASE predicted temperature contours (T/T"rl on
bulkhead for SB cases.
408 R.F. Kunz and N. Salman
....
t;
,!
~
~
~
..2
!L
~
e
So akbKk Flow
~
::< No SoabKk Flow
.•35"
"
,.
Vt..r
Fig. 33. Sequence of predicted engine compartment cover surface temperature fields for SB case.
Temperature contour range is from 115°F to 185°F. t/t"r =O, 2.5, 5, 10, 15, 20, 25, 30. Time
history of peak predicted engine compartment cover surface temperature.
SoMbacll now
No ao•ba<:k Flow
Sa .. bade now, Of'ig. RADTt£RM Model
' '·
.611 ,. .. ..
Vt..r "
Fig. 34. Sequence of predicted engine and exhaust duct surface temperature fields fot SB case.
Temperature contour range is from 120°F to 550°F. tlt"r =0, 2.5, 5, 10, 15, 20, 25, 30. Time
history of peak predicted engine surface temperature.
Flow Field and Thermal Management Analysis 409
........
r ' 1- 1: r
----
I
I I I I
.711
l
J
\ \
-Flow
No Solld:taciCRow
.750 of-'-~..-...,!10,_...._"'"";\,.,......~""-;!:;,.~~~..
llt..r
Fig. 35. Sequence of predicted generator surface temperature fields for SB case. Temperature
contour range is from 120°F to 290°F. t!t"r =0, 2.5, 5, 10, 15, 20, 25, 30. Time history of peak
predicted generator surface temperature .
... i
.ru
t::.
.~t .!
s•
SoMbadl Flow
No So*bac:k Flow
' --
e ·~ ~~c~Fiow
No Solllcblck Flow ~
!! !!
I
1!
:it ---------- !
.
... .
L'\,.r
.,
"
~t!
.
Fig. 36. Sequence of predicted transmission surface temperature fields for SB case. Temperature
contour range is from 120oF to 280°F. t!t"r =0, 2.5, 5, 10, 15, 20, 25, 30. Time history of peak
predicted transmission surface temperature. Hot spots considered here are directly across from
the engine burner (left) and directly attached to the engine (right).
Aerodynamics of High Speed Trains
Experiments and CFD in Train Aerodynamics: A
Young and Turbulent Association Full of Potential
Remi Gregoire
The advent of the modern high-speed train system may be dated back to 1964
with the opening of the first Shinkansen high-speed line in Japan, at an initial
revenue service speed of 210 km/h. At same time it associated very fast rolling-
stock and dedicated high-speed track infrastructure, it opened a new era for
the aerodynamic design of ground transportation systems. On the one hand,
the aerodynamic portion of the train running resistance was going over two
third of the global running resistance. On the other, new kind of aerodynamic
problems needed to be faced, such as those going together with train travels in
tunnel and passing of opposite high-speed trains on adjacent tracks.
These new problems could not anymore be grasped solely through
conventional full-scale or even wind-runnel reduced-scale test methods due to
(i) the absence of already built high-speed tracks and tunnels, (ii) the transient
nature of the airflows and boundary conditions in case of a train moving
relatively to another train or a tunnel. As a consequence, the usefulness of the
newly born CFD techniques started to emerge in the seventies with first
applications made by the French National Railways (SNCF) with dedicated
steady and unsteady 3-dimensional Panel Method programs. The way was
then paved for the very diverse CFD investigations the design of high-speed
train systems could benefit.
The objective of this paper is thus to address, through a few examples, the
way the need and the use of the CFD techniques have evolved these last
decades in complementarity with experiments. The paper also focuses on
present and future speed-up of CFD applications due to both ever increasing
computational, CAD and CFD resources and appearance of new troublesome
railway aerodynamic issues related to HSTs' cross-wind sensitivity as well as
aerodynamic noise emergence over 320 km/h.
Recent Studies of Train Slipstreams
1 Introduction
When a train travels in the open air it displaces the air around and over it
forming a slipstream alongside the train and a wake behind. The air at the
surface of the train moves at the speed of the train, whilst far from the train
the air moves at the ambient air speed. Therefore, there is a region near the
train sides where the air can be moving at speeds comparable to that of the
train. In this region the air is very turbulent and, depending on the aerody-
namic roughness of the train, may contain complex and interacting vortices.
After the train passes there is a wake flow, which decays as the train moves
away.
The turbulent and highly chaotic air flows in the train's slipstream and wake
impact on people and objects near the passing train. The adverse effects in-
crease with the aerodynamic roughness and speed of the train. Certain trains
can create airflows sufficiently large that they pose a danger to people either
working at the trackside or to passengers waiting on station platforms.
With increasing freight and passenger train speeds in Europe, there has been
a strong interest in understanding the nature of slipstreams with a view to
minimising the potential danger and producing common European standards
to control the risk. Recent studies of high-speed train slipstreams have been
carried out in a European collaboration and are briefly reviewed in this paper.
• Children's pushchairs, usually with luggage in and the brakes on, have been
drawn into the trains' slipstreams and set into motion or destroyed by the
train.
• Three people have been almost knocked off their feet by passing trains.
The train slipstream is generated by the viscous shearing effect of the train
moving through the air. Gust components and turbulence are created by sur-
face roughness elements and discontinuities in the train surface, such as ex-
posed bogies, intercar gaps and gaps between containers. Aerodynamically
rough trains such as freight trains, especially freightliners and open car carriers,
generate very strong slipstream flows, which peak as the train is passing. For
streamlined trains the peak air speeds occur in the wake of the train after it has
passed.
Fig. 1 shows typical measured averaged resultant slipstream air speeds
caused by a German ICE high-speed passenger train passing a station platform.
The air speeds have been normalised by the train speed and the time base has
been adjusted to be a distance base by multiplying by train speed. Measure-
ments were made using anemometers measuring the longitudinal and lateral
horizontal components of the air velocity at a height of 1.335 m above the
platform and at 0.5 m, 1.0 m and 1.5 m from the platform edge.
At each position, the resultant air speeds show similar characteristics:
• A sudden peak and subsequent fall in speed as the train nose passes (at about
50 m) .
• A maximum value just after the tail of the train passes (just after 400 m) .
• A steady rise in mean air speed between the nose and tail passing.
• A gradual decay in air speed after the train has passed for a distance similar to
that of the train length.
0.3
0.25
Platfom1 1.5 m
c
·u
0 0.2
v
.,> 0. 15
"
.!::!
-.;
e 0. 1
z0
0.05
Distance (m)
0.6
Fig. 2. Mean, 5'h and 95'h percenrile slipstream velocities from full-scale tests
However, at the measurement position closest to the train, the peaks associ-
ated with the nose and tail passing are significantly greater than at the other
rwo positions. Also, the air speed as the train passes is much higher and ex-
hibits larger fluctuations at the closest measurement position. The larger
fluctuations have been found to correlate with the passing of the coaches of the
train.
Fig. 2 shows the spread of the data around the mean slipstream history for a
typical series of data, consisting of 10 train passes and taken with very low am-
bient wind speeds. This demonstrates the wide spread of the data even when
ambient conditions are favourable.
The main factors influencing slipstream strengths are the train speed; the
distance away from the side of the train; the shape and surface finish of the
train and the ambient wind speed and direction. It has been observed in tests
that strong cross-winds significantly enhance slipstream effects on the lee-side
of the train.
It is worth noting that improvements in the design of trains from the point
of view of aerodynamic drag are usually also beneficial in reducing slipstream
effects.
Finally, a train slipstream is a truly unsteady and almost chaotic phenome-
non when viewed from a stationary viewpoint. This has sometimes led to
confusing and contradictory results being obtained from full-scale test meas-
urements in the past.
418 T. Johnson, S. Dalley, and J. Temple
3 Safety Standards
Despite the concerns held by the European railway operators, there is currently
no single safety standard for train slipstreams in Europe. However, the Euro-
pean standardisation body, CEN, will shortly be working to determine one.
Nevertheless, each railway company has its own rules for controlling the risk
based on notional limit values. For instance, in the UK a maximum peak slip-
stream value of 17 m/s has been applied in the past, although more recently a
comparative air speed criterion has been applied on the basis that existing op-
erations are 'safe'. In France, the French national railways apply limits to the
maximum force exerted by a train's slipstream on a cylindrical instrument; the
limit value being derived by comparison with exerting traffic. The German
railways also apply limits to maximum slipstream speeds using existing opera-
tions to determine limit values.
The risk from the slipstreams is then controlled by applying safe standing
distances for people near the track, indicated by a yellow line on the platform
for passengers; limiting train speeds or banning people from passing high
speed trains altogether. Safe standing distances vary from one European
country to another.
What is puzzling is that, although the physics of slipstreams must be invari-
ant, there is no international consensus about the most important characteris-
tics of the turbulent slipstream and wake as they affect people. This has led to
different characteristics being measured, i.e. air speeds or air forces on a repre-
sentative body, rather different limit values being adopted and differing meas-
ures to control the risk.
There are several possible reasons for this lack of consensus.
1. There is a genuine uncertainty about which physical quantity is the most
important for the stability of people.
2. The use of limit values determined from the assumption that the current
situation is safe will highlight differences between national railways e.g. op-
erating practices and train speeds, existing control measures, train patterns
etc.
3. Ensuring the safety of people from train slipstreams is not entirely an aero-
dynamics issue as there is a strong element of subjective response involved,
with other factors being involved such as the person's perception of danger,
their preparedness for the event among several.
It is for the reasons discussed above the recent studies of train slipstreams
were made at a European level.
Fig. 3. ERA gust anemometer array at the trackside during full-scale tests
420 T. Johnson, S. Dalley, and J. Temple
Fig. 4. Force measurement cylinder at trackside. (Note the height of the ballast shoulder)
The MMR has been used extensively for determining train aerodynamic
characteristics in tunnels and in the open air (pressures). Pressure measure-
ments in both types of tests have been validated against full-scale data.
Further information about the MMR can be found in Johnson and Dalley
1999.
it was found that there was a large variability between runs in the time histories
of the velocity components at each measurement position. It was decided,
therefore, to ensemble-average the results.
The runs were sorted according to train speed, with two target values, and
the data grouped when the speed varied within ±2 m/s of the group mean
speed. The time base was converted a distance base by multiplying by train
speed, and the slipstream velocity magnitudes formed and non-
dimensionalised by the train speed. The time series data was then aligned to
event markers, particularly the train nose passing. This enabled the data to be
meaningfully averaged at each measurement position.
Following detailed analysis, it was found that 10 runs were sufficient to
form stable ensemble averages of means, but at least 20 runs were needed to
form stable ensemble averages of standard deviations.
This result is important and explains some of the non-repeatability between
results observed in previous full-scale tests obtained with the few measure-
ments that were practicable. This seems to imply that the slipstream is a cha-
otic process requiring a statistical description. This point will be returned to
later.
~
OA
-2 -1.5 ·I
-· gl
.0.5 0 0 0.05 0.1 0.15 0.2 0.25
0.4 0.4
~
0.3 0.3
0.2 0.2
0.1 0.1
I
2 0 6 8 10
At the surface of the train the air will move at train speed and will reduce
with lateral distance from the train side. The train boundary layer thickness
may be defined by the lateral distance from the side of the train at which the
air velocity is 1o/o of train speed. In practice, it is the region of fast moving air
close to the train side which is on most interest for the safety of people at
trackside. For the MMR it appears as if this region of fast moving air is thin-
ner than at full-scale. It is not clear if this is a Reynolds' number effect, an ar-
tefact of the reduced train length used at model scale or caused because the
high track ballast shoulder at the full-scale test site was not simulated at model
scale.
0
·;:;
0
">
E
"'
"
t:
"' 0.20
~
"0
0.15
"
.;!l
-;;;
E
0 0.10
c
c
"'
"
:::E
0.05
0.00
0 200 400 600 00 1000
Distance (m)
Fig. 7. Comparison of model slipstream measurements with full-scale results for the ICE service
trains and the ICE2 test train.
The slipstream velocities in the wake initially compare favourably, with the
decay well captured. However, in the far wake region there is a poorer com-
parison. This has been attributed to the use of a linear calibration for the hot
film sensors, which introduces an increasing error as velocities become smaller.
It was concluded that model tests are useful for examining train slipstreams,
but, until the discussed discrepancies are satisfactorily resolved, they should be
conservatively used for making comparative studies, eg freight trains with and
without streamlining devices, rather than absolute measurements.
Recent Studies of Train Slipstreams 425
Fig. 8. Transient flow around the rear section of an ICE2. The approximate time step between
frames is 0.15 seconds
426 T. Johnson, S. Dalley, and J. Temple
This phenomenon has been observed in Japan and has led to complaints
from passengers of a type of motion sickness. For this application, however,
CFD was used to provide some insight into the transient nature of the slip-
stream and wake behind a high-speed passenger train.
The method used was a fully three-dimensional transient RANS approach,
with a RNG k~ _ turbulence model. The method was embodied in the
FLUENT 5.4 CFD software.
The transient flow around a horizontal section through the ICE was exam-
ined. Fig. 8 shows a series of instantaneous solutions from the CFD simula-
tion, depicting the flow pathlines. The periodic nature of the can be observed
in the pathline patterns.
It was observed by inspecting the evolution of the pathlines from the train
frame of reference, that the flow behind the train has a periodic nature and
vortices are shed from alternate sides of the train with a frequency of about
1.4 Hz. This means that when the train passes a single velocity measurement
device, it may see only one of a number of different velocity time-history pos-
sibilities. From the train frame of reference the predicted slipstream behaviour
is more orderly than appears from a fixed passed point. This goes some way to
explaining why such variability has been observed in slipstream velocity data
and why a statistical approach was necessary to analyse the data. It also implies
that the use of many along-track measurement positions, rather than multiple
train passes may be a practical way of capturing the data in future with the
consequent reduction in costs.
d 2e (1)
I - - - F(t){bcose + 2asin8)- mg( acose- bsine)
dt 2 -
where:
I is the body moment of inertia about the line of rotation
F(t) is the time-dependent aerodynamic force due to the wind load
a is the body half-width
b is the body half-height
m is the body mass
g is the acceleration due to gravity
tis time
The aerodynamic force is related to the wind speed V)t) in the usual way:
where p is the density of air, A is the area of the body facing the wind and C 0
is the aerodynamic drag coefficient.
r
2b
F(t)
1
Timet= 0 Timet
Fig. 9. Idealised human body rotating under the influence of an aerodynamic force.
20 - - -
15 ~
/ '\
"'E
......
-o
0
10
/
0
c..
~
</)
~
I::
~ 5
0
0 0.1 0.2 0.3
Time, s
'Falling down' is defined as occurring when the body tilts to the angle at
which the mass moment acts to continue the rotation rather than resisting it
within 0.35 s. An examination was then made of when falling down occurs
with different gust amplitudes and pulse widths. Such an examination is
shown in Fig. 11 for an object with the characteristics of a typical man of
85 kg weight.
This example shows three curves corresponding to the falling down wind
speeds associated with three values of mean wind speed on which the wind
pulses have been superimposed. It shows how the combined wind pulse and
mean wind speed needed to cause falling down increases as the pulse width be-
comes smaller (effectively as its frequency increase), as might be expected. The
effect of the mean wind is to reduce the total wind speed needed to cause fal-
ling down, as it provides a pre-load to the body. However, the relative effect of
the pre-load is not large unless the pulse frequency is high.
This model appears capable of providing useful insights into the aerody-
namic effects on people before they have a chance to respond, and it will be
used to compare with published results for human stability in natural winds
and in the building environment.
Recent Studies of Train Slipstreams 429
29 \
27
\
,...,
--f/)
25
\\
\\"
E
--
.........
-ci 23
'~
0)
0) \
a.
Cl) ... '
"0 21 ...
'~
c: ...
3 19 ...
. ......
• .... . . ~ r-- I
17 .. - -... -r---
.. - - .
15
0 0. 1 0.2 0.3 0.4
Pul e Width, I
Fig. 11. Wind speeds necessary to cause falling down for a man of weight 85 kg for different
wind pulse widths and mean wind speeds. (Heavy line, mean wind=O m/s; chained line, mean
wind=5 m/s, dotted line, mean wind= 10 m/s)
5 Conclusions
A brief review has been made of recent studies undertaken on the slipstreams
of high-speed passenger trains. This work was carried out in the European re-
search project RAPID£. The studies included:
• Full-scale test measurements of slipstream velocities generated by German
ICE trains.
• A model scale investigation of slipstream velocities at the side and above the
roof of a model ICE.
• CFD modelling of the flow around the rear of an ICE.
• Estimating the wind speeds necessary to make humans fall over before they
have time to react using a simple mathematical model.
The model tests showed that a stochastic approach is necessary for slip-
stream analysis. Single measurements of slipstream velocity give only one pos-
sibility, as the slipstream is highly transient and appears chaotic from the
ground frame of reference. It is necessary to have at least ten ensemble-
averaged runs to obtain stable mean values of velocity, and about twenty for
stable standard deviations. This may be achieved either by having either mul-
tiple runs of the train and/or having multiple simultaneous measurements for a
single train pass.
430 T. Johnson, S. Dalley, and J. Temple
Model tests may offer a cheaper approach to full scale testing, but some is-
sues related to boundary growth on short train models have not been resolved.
CFD has shown that the flow structure behind a train may exhibit more or-
derly behaviour when viewed from the train reference frame. The use of CFD
as a useful flow visualisation tool has been demonstrated.
The simple mathematical modelling of wind speeds necessary to cause fal-
ling down shows some promise in helping to understand human behaviour
before subjective response takes over. However, more subjective data is needed
for track workers and people in stations, including the effects of; wind speed,
train noise, distraction and surprise. Until this information is available and
correlated with the physical effects, it is difficult to determine how much the
safety of people from slipstream effects depends solely on aerodynamics.
Another outstanding issue is that the effect of cross winds not been reliably
established, although there is some evidence that they can significantly aug-
ment slipstream velocities in the lee-side of trains.
A new European project is proposed to address some of these outstanding
issues, particular subjective response.
References
Baker CJ, Dalley S, Johnson T, Quinn A, Wright NG (2001) The slipstreams and
wake of a high-speed train. Proc Instn Mech Engrs, Part F, 215: 83-99.
Fukuchi G (1961) Field Measurement of Train Drafts. Permanent Way No. 11, (Vol.
4, No. 2), Translated from "Tetsudo Senro", bulletin of Permanent Way Society
of Japan, Vol. 9, No.4, Permanent Way Society of Japan
Johnson T, Dalley S. (1999) An experimental facility for the investigation of aerody-
namic effects. World Congress on Railway Research, Tokyo, 19-23 October.
Glockle H, Gawthorpe RG (1991) Railways: consideration of the environment. Aero-
dynamics- Trackside Safety Environment, Seminar S992, London, October, Inst
Mec4 Engnrs.
Peters JWH (1999) Air Movements and human postural stability Arbo Management
Groep\ Ergonomics 2109-E (AM) Project Organisation High Speed Line South
Infrastructure (HSL South Infra), Utrecht, Netherlands
Poulton E., Hunt JCR, Mumford JC, Poulton J (1975) The Mechanical Disturbance
Produced by Steady and Gusty Winds of Moderate Strength: Skilled Performance
and Semantic Assessments, Ergonomics, Vol. 18, No.6, pp 651-673
Schulte-Werning B, Marschke G, Gregoire R, Johnson T (1999) RAPID£: A project
of joint aerodynamics research of the European high-speed rail operators. World
Congress on Railway Research, Tokyo, 19-23 October
Soligo MJ, Irwin PA, Williams CJ, Schuyler GO (1998) A comprehensive assessment
of pedestrian comfort including thermal effects. Journal of Wind Engineering and
Industrial Aerodynamics, Vol. 77/78, pp. 753-766, Elsevier
Aerodynamic Effects in Railway Tunnels
as Speed is Increased
Abstract
The aerodynamic effects occurring in a tunnel as a train moves into or through
it are totally different from those observed in the open air and their amplitude
and severity grow as the train speed is increased. The flow in the whole tunnel
needs to be considered in the same time as the flow in the vicinity of the vehi-
cle. Aerodynamic forces, pressure waves and acoustics have a strong impact on
safety and comfort issues.
When a train enters into a tunnel, a compression wave is generated, propa-
gates through the tunnel and is reflected at the tunnel extremity. During the
reflection process a part of the wave is transmitted outside the tunnel in the
form of a micro-pressure wave, which may generate a "sonic boom" problem,
depending on the shape of the incident wave, in particular the gradient of the
wavefront. The shape of the wave changes as it propagates through the tunnel
under the influence of the unsteady viscous effects (in particular skin friction
at the tunnel wall), the non-linear effects and the presence of material and
components in the tunnel (for example, ballast or niches). Measurements of
the skin friction behind a pressure wave are presented.
Introduction
speed and the building of more and more tunnels of increasing length. The es-
sential advantage of a tunnel is that the environmental impacts observed in the
open air are reduced. A new one, the "sonic boom", may appear and impair
the comfort of people living in residential area close to the tunnel portals.
The presence of the tunnel confines the air flow to a finite domain and con-
strains the air perturbations generated by the vehicle to propagate along the
axis of the tunnel only. The general flow can be divided in two different flow
domains, which are the near-field flow (in the vicinity of the vehicle) and the
far-field flow (far from the vehicle, corresponds to the flow in the tunnel).
Both domains are strongly dependant to each other.
The near-field flow is strongly influenced by the blockage ratio 15 defined as:
(1)
1 The annular space (also referred to as annulus) is defined as the volume between the vehi-
cle and tunnel walls: following the flow, it starts at the end of the nose and ends at the start
of the tail.
Aerodynamic Effects in Railway Tunnels as Speed is Increased 433
dium length, the unsteadiness of the flow is due to the train-tunnel entry pres-
sure waves. The transmission of these pressure waves inside coaches generates a
comfort problem since the internal ear of the human being is very sensitive to
pressure level changes (amplitude and rate). This problem can be a limiting
factor while defining speed and blockage ratio. The sealing efficiency and the
stiffness of coaches can be increased to limit the pressure variations at the ear
of passengers during tunnel travel. In long tunnels, these pressure waves are
strongly attenuated and, after a given travel time, the unsteady effects are es-
sentially associated with the motion of the train. In this situation, the drag-
related effects become dominant.
The oncoming of a high-speed train in an underground station can induce
high air velocities, which must be controlled by an appropriate design in order
to satisfy the comfort and safety criteria for the passengers on the platforms
and in the station. Ventilation is another key aspect to preserve the air quality
and to maintain the temperature within acceptable values.
tunnel. The initial gradient of the wavefront is directly related to the nose
shape, the portal shape, blockage ratio and the speed of the train.
A typical shape of the pressure signal recorded in a tunnel, as a train enters
into it, is shown in the following figure (from the TRANSAERO project, see
e.g. [11]):
L
Prt.o urt
Prelimi1wry pre.l.m re
ri;e due to the j70II'
ahead of the frainJwse
The maximal amplitude generated as the train enters into the tunnel de-
pends essentially on the speed and the blockage ratio. State-of-the-art numeri-
cal methods and the use of a facility such as the MMR allow an accurate pre-
diction of the pressure wave profile during the generation process [24].
Experimental results
a distance of 10.19 m from the diaphragm. The second one (IB-2) is located at
a distance of 50.95 m from IB-1. IB is an acronym for Instrumentation Block.
P1essr.sre {Pal
o~oor---~--~--~---.----~--.----r---,
1&000 ~
no lal step ol 16'ti00 Pa
fOCDtleO 18-1
14000
12000
SOOO
1000
Fig. 4. Pressure histories for three different pressure levels at 50.95 meter dis-
tance, see [ 11]
The pressure histories recorded close to the diaphragm (IB-1) shows a typi-
cal step shape for a shock tube experiment. The pressure histories recorded at
the IB-2 location are characterized by slight differences close to the wavefront,
as well as lower amplitude due to dissipative effects. A "rounding" of the "cor-
ner" can be noticed on the rwo low-amplitude pressure steps (6'100 & 7'800
Pa). This is due to the unsteady friction. The flow just behind the pressure
step is laminar and the value of the unsteady friction is very high close to the
pressure step and decreases until the start of the transition to turbulence. As
the flow is getting turbulent, higher values of the skin friction are reached.
This unsteady skin friction profile behind the pressure wave has a significant
impact on the shape of the pressure wave close to the wavefront as the wave
travels over long distances (kilometers in the case of railway tunnels).
The definition of the transitional Reynolds number Re" can be done as fol-
lows:
Weak pressure
wave hypothesis:
Re,,=-w
u~ s }
Uo << Uw ;
Uw
-"' Re"""
oVr Wo
0 .35
0.3
Laminar
0.25
0 .2
0.15
0.1 \
\._,
0.05
Transition
0
Time [s]
Fig. 5. Measurement of skin friction 10.19 m away from the diaphragm . The
pressure step amplitude is 5'800 Pa (shape is similar as what was recorded on
IB-1 location, see Fig. 4) .
It is difficult to define at what time the flow is fully turbulent on this graph.
The expansion wave limits the recording time. A longer pressure step would be
required to analyze the transition and turbulence zone in deeper details. This
can be obtained by an increase in length of the facility.
It is interesting to note that the shape of the first two peaks in the first part
of the transition zone (up to 0,02 s) is different from the following ones. This
can be compared to the observations of "turbulent bursts" by Dillon and Na-
gamatsu, 1984, [28]. The precision of the measurement technique is approx.
5%. But it must be stated that the flow is very sensitive to small geometrical
perturbations close to the transducer, as well as difference in level between the
transducer and the tube wall. Attention has been paid to minimize these per-
turbations, bur a systematic error may appear and it is difficult to evaluate it
precisely. Based on wind tunnel experience and the shock tube peculiarities,
Aerodynamic Effects in Railway Tunnels as Speed is Increased 439
the uncertainty was estimated at 40o/o. Further development of the measure-
ment technique is necessary to evaluate more precisely and reduce this uncer-
tainty.
Considering the railway systems, the pressure waves generated by a train is
of slightly lower amplitude and the wavefront is not a pressure step, but a pres-
sure wave. Additional experiments are necessary to evaluate the effects of a
gradual pressure increase on the shape of the unsteady skin friction profile. It
appears nevertheless that the wavefront is probably influenced at least by the
laminar and transitional part of the unsteady skin friction profile. The model-
ing of the unsteady laminar and turbulent skin friction in numerical codes
have been studied by many authors [12],[23],[29]. Improving these models
also requires experimental data, in particular to account for the transition and
to model the turbulent part more accurately.
The aerodynamic effects appearing in railway tunnels have been described and
the preliminary results of a study on the propagation of pressure waves in tun-
nels have been presented. The context of this work is the research effort done
to understand and model tunnel flows for railway systems and new transporta-
tion systems, such as Swissmetro or ET3. The importance of the unsteady skin
friction behind a pressure wave appears clearly from the measurements, in par-
ticular, the importance and peculiarities of the transition from a laminar to a
turbulent flow.
This research has been motivated by the importance of improving models to
predict the propagation of pressure waves for the design and operation of rail-
way tunnels. Another area in which similar effects can be observed is during
the repressurisation (for safety reasons) of an initially evacuated tunnel, for the
case of vacuum-based transport system in development. The constraints due to
aerodynamic effects have a significant impact on the costs of both tunnels and
trains. Aerodynamics is as well a key issue in the performance of high-speed
systems, in terms of energy consumption and performance.
Research projects such as TRANSAERO have provided an opportunity to
develop methodologies and tools for the improvement of the performances of
existing railway systems. On the other hand, projects as Swissmetro or ET3
show that there is a potential for new transport systems to address sustainabil-
ity issues and opening new fields for aerodynamic research.
The results presented in this paper have shown the limits of the Starlet fa-
cility. In order to investigate in more details the skin friction behind a pressure
wave, it is necessary to improve the following points: 1) to increase the length
of the facility to be able to record pressure histories over a longer time behind
the pressure wave; 2) to generate different shape of pressure waves in order to
measure the influence of the gradient on the transition; 3) to investigate in
deeper details the transition; 4) to improve the unsteady skin friction meas-
urement and to apply other techniques as well.
440 V. Bourquin, C. Beguin, and P.A. Monkewitz
Masahiro Suzuki
Abstract
1. Introduction
As the maximum speed of Shinkansen trains in Japan increases, vibration of
the trains has recently become a subject of discussion especially concerning
riding comfort. This phenomenon has the following characteristics (Fujimoto
et al. 1995). (1) The vibration amplitude of the train in tunnel sections is
more noticeable than in open sections (fig. 1), (2) it gradually increases from
the head toward the tail of the train set, and (3) the yawing vibration is more
prominent than the other vibrations.
F:
~ Open section Tunnel section
~~ ~ § 0.1
~~ _o.o. ~
·- Q) 0.1
~ 0 ~~~~~~~~~--~L_~~~~~~
~ )6 0 5 10 t[s] 15
Fig. 1. Time history of yawing angular acceleration of a train entering a tunnel (train speed: 300
km/h).
Since mountains account for about 70 % of Japan's total land area, there
are a large number of tunnels in its railway system. For example, half of the
Sanyo Shinkansen line, which connects Osaka and Hakata (622.3 km), is in
tunnels. So, the riding quality in tunnel is critical for service. Therefore, in-
tensive studies have been carried out to solve the problem of vibration in tun-
nels.
Track irregularity was considered at the initial stage as one of the factors
causing the phenomenon. However, there was no correlation between the vi-
bration and the track irregularity in the tunnel sections (Takai 1989). Another
factor, namely aerodynamic force, has also attracted attention. As for the
aerodynamic force, the effect of Karman-like vortices on the vibrations of the
train had been suggested. But, no mechanisms had been clarified in detail at
this first stage. Thus, we have been extensively investigating flow around the
trains in tunnel by running tests, wind tunnel experiments and computer
simulations.
The paper describes the phenomenon, its mechanism and countermeasures
from our studies.
Fig. 2. Time history of aerodynamic yawing moment acting on a train entering a tunnel (train
speed: 300 km/h) .
-
'2..
800
t - --Open section ........
>Q)
..c ~
600
-
<I>J2 -Tunnel section ~
-.r ---"
c:: 400
0 .2
-o E
~
.... C!l
c:: 200 .,a,-
s:
0~
0
....
Q)
0
~·
C!l
-200
0 3 6 t [s] 9
Fig. 3. Time histories of work done by aerodynamic force (train speed: 300 km/h).
the tunnel wall side is much larger than that on the tunnel center side, the
pressure difference is mainly dependent on the pressure on the tunnel wall
side.
Tunnel section
Cross sectional area
oftun nel: 63.4m2
"w-_,___,___....Joo..rv-.-...J-... Cross sectional area
~:::::1:::::s::e:::c·~-~--+----+------1 of train : 10-13m 2
Pressure on tunnel center side(A)
,. .. ro_
Pressure on tunnel wall side (B)
0...1
Pressure difference (8 -A) l[s ]
Fig. 4. Time histories of pressure on each side of a train entering a tunnel (train speed: 300
km/h)
To investigate the pressure on the tunnel wall side, we set pressure sensors
on the sides of two consecutive cars, the 4th and 5th cars. Figure 5 shows
pressure as a function of time. The pressure fluctuation that travels leeward
while keeping its shape does not decay even when it passes the gap between
two cars. The speed of this propagation is equivalent to approximately 80 per-
cent of the train speed.
Running test data of various series of trains were analyzed to find that there
is coherence in the data. The pressure fluctuation does not appear locally but
develops along the whole length of the train (the typical train length was 400m
with 16 cars) irrespective of the train types. Figure 6 displays pressure fluctua-
tion developing along the whole set of train. The pressure fluctuation in-
creases from the head to the 6th car (125~ 150m from the head), then remains
constant and finally drastically increases at the tail of the train set. The peak
frequencies of the pressure fluctuation which are recognized after the 3rd car
(50 ~ 75m) decrease from the 3rd car toward the 6th~ 8th cars (125~200m) and
remain at the same level to the tail of the train set.
Flow-induced Vibration of High-Speed Trains in Tunnels 447
Runn ing direction
I 4th oac
• • II • •
5th ca r
.-\ R (' I)
vo uu 00 u u
-: r
~
_... 0.1
~ 0 0. J\ .J\ - ...J'\.. "...r ,.._,._ 1\ f,
-~ lTv~vv ~V"
~20 -10 f 60
-
-- 0
"V'vv~-
\T \.1 v
M (\ ~ {\ D tJC:3. ,..
to :. :~o V:'>o 10 (>O
·0.1 f
:0.
1rr::u~~··~;~
- () YVVW"'J"
·O. I 10 :W :30 -10 . f GO
0. I
--
.r- 0
-·0. 1
f
Fig. 5. Time histories of pressure on the tunnel wall side of two consecutive cars (train speed:
296 km/h, t' indicates non-dimensional time based on train speed and train width) .
0.3
~0. 15 ~02
Q)
~ 0.1 _a.
(.) -o.1
~005
Fig. 6 . Development of pressure fluctuation on the whole set of train (f indicates non-
dimensional frequency based on train speed and train width).
From the above, the following are presumed. Some large organized pat-
terns exist in the space between the tunnel wall and the train. These flow pat-
terns develop from the head toward the 6th~8th cars and become steady there-
448 M. Suzuki
after to the tail of the train set. The flow separates at the tail with a large pres-
sure fluctuation.
3. Mechanism
In this section, some mechanisms of generating these aerodynamic forces are
explained.
To clarifY the flow field around a train tail, a three-dimensional unsteady Na-
vier-Stokes simulation was carried out with a short train model that has a
length of 2.5 cars (Suzuki et al. 1996). The simulation successfully obtained
unsteady flow separation on the rear nose, which causes fluctuations of the
yawing moment of the tail car. In the tunnel section, the simulation proved
that the tunnel wall makes the flow separation asymmetric and that the expan-
sion of the effective flow area along the rear nose causes a greater pressure
fluctuation.
whole side of the train after the 4th car. The unsteady aerodynamic force is
generated on the side of the train when the vortices pass.
Vertical plane
Honzontal plane
Tunnel wall
(a) z-directional vorticity distribution on the horizontal plane under the train .
c
(b) Vorticity magnitude distribution on the vertical plane on the side of train .
4. Countermeasures
Some of countermeasures to solve this problem are presented in this section.
Since nose shapes are crucial to flow separation, nose shapes were tested first.
Five different types of nose shape were prepared (fig. 8). These are a two di-
mensional short shape, a rwo-dimensionallong nose, a three-dimensional short
nose, a three-dimensional long nose and square cornered nose. The two-
dimensional nose is a so-called wedge-shaped nose. Sides of the three-
dimensional noses are rounded. The result shows the three-dimensional short
nose is the worst. This is because the flow separates around the sides of nose
and reattaches again. These separation and reattachment points fluctuate.
Thus pressure around the nose vibrates and the yawing moment changes.
0 .20
0 .15
§"'
~ 0 .10
~
~ 0 .05
0 .00
Square Short Short Long
cornered 2-D 3-0 2-D
~
Fig. 8. Effects of nose shapes. (Cy,w is a coefficient of aerodynamic yawing moment.)
As described in section 3.3, there are vortices generating around the floor of
the train. The shapes of the train bottom were supposed to be critical for re-
ducing the aerodynamic force on middle cars. Here, rwo kinds of shapes were
prepared; a train with rounded bottom corners and one with fins under the
body. The effects of these shapes are illustrated in figure 9.
Flow-induced Vibration of High-Speed Trains in Tunnels 451
0.075
§ 0.050
J'"i 0.025
0.000
Norma I shape Rounded
lower caners Fins
rulS5J~
Fig. 9. Effects of rounding lower corners and installing fins. (Cy,w is a coefficient of aerody-
namic yawing moment.)
5. Conclusions
The flow-induced vibration of the high-speed trains in tunnels was investi-
gated by the running tests, wind tunnel experiments and numerical simula-
tions. The running test revealed the development of coherent flow patterns
along the whole set of the train. The wind tunnel experiment confirmed that
the train vibration in tunnels is a forced vibration by aerodynamic force. The
computation demonstrated the vortices on the train side and the sudden ex-
pansion of flow area at the tail generate the aerodynamic force. The wind tun-
nel experiment showed the long nose, rounding the lower section of the car,
and installing fins under the train, which decrease the aerodynamic force, are
effective countermeasures.
References
Jean-Luc Peters
1. Introduction
Although trains may be considered the safest existing ground vehicles, there
has always been the occasional derailment due to strong side winds, mostly on
islands such as Japan or the British Isles [1]. According to meteorologists, the
current global warming leads to ever stronger winds meaning such accidents
are more likely to happen not only in coastal areas but even in the very heart of
larger continents. This is how a commuter train was blown over in Austria
recently (Figure 1), only 5 years after a similar accident happened in Belgium.
Figure I. Light train overturned by a 30 m/s wind gust, Uttendorf/Austria, November 17, 2002
454 J.-L. Peters
The present tendency of reducing the weight of trains obviously contributes
to their increased side wind sensitivity. The dramatic increase of the train
speeds that we have witnessed in Europe and Japan in the last decade further
exacerbates the problem, as shown in Figure 2. Finally, the long and slender
noses that are required for high speed trains are unfortunately particularly
sensitive to cross winds, as has been demonstrated by many wind-tunnel tests
[2]. This is due to the large areas of negative pressures on the leeside of such
long noses, as can be seen in Figure 3.
In view of these facts, and having been responsible for the aerodynamic
development of the soon to be introduced three fastest operational trains in
the world, the Chinese Transrapid maglev, the Spanish Talgo 350 and ICE
350 trains, Siemens Transportation Systems has started a comprehensive R&D
program to reduce the cross wind sensitivity of its trains.
1'-.
" ~
"
"' ""
DERAILMENT!
~
........
'- ............. ..........
SAFE
I I '--......!
.............
20 50 80 110 140 170 200 230 260 290 320 350
Vtr [km/h]
Figure 2. Typical HST leading car wind speed Vw vs. train speed Vtr overturning characteristic
How to Reduce the Cross Wind Sensitivity of Trains 455
WIND >
Figure 3 . Pressure distribution on slender nose shape (dark gray areas on the leeside = negative
pressures)
3. Selected results
Figure 4. ICE3leading car with optimized nose shape in SIO wind tunnel at 30° yaw.
The 6 components (drag coefficient Cx, side force coefficient Cy, lift
coefficient Cz, rolling moment coefficient Cmx, pitching moment coefficient
Cmy and yawing moment coefficient Cmz) acting on the complete cars were
measured initially by means of the wind tunnel underfloor automobile
balance, and then by means of a Siemens owned, adapted internal balance
manufactured by Aerotech (UK), up to a yaw angle of 90°. Furthermore, the 6
How to Reduce the Cross Wind Sensitivity of Trains 457
Shape optimization
The shape of the present Siemens high speed train (HST), the ICE3 (Figure
5), has been conceived more according to design than to aerodynamic criteria.
On the one hand, its extremely rounded cross section certainly has a positive
effect on its cross wind stability, as demonstrated by basic investigations of the
effect of roof corner radius carried out in the early eighties [2]. On the other
hand, its undoubtedly good looking nose shape is very far from being
optimized with regard to actual aerodynamic requirements about tunnel entry
phenomena, head pressure pulse and cross wind sensitiviry.
A maximum reduction of these last three aerodynamic effects had on the
contrary, the highest priority during the development phase of the power car
for the Spanish Talgo 350 HST, conceived and built by the Locomotives
division of Siemens Transportation Systems. To reduce cross wind sensitivity,
positive experience gained in the 1970s by experimenting with "duck bill" like
nose shapes on maglev trains (Figure 6) was used as a starting point. Adapting
that type of shape to a conventional power car while optimizing it with regard
to head pressure pulse and tunnel entry phenomena by means of CFD, led to
the present concave shape of the Talgo 350 nose (Figure 7).
Figure 6. 1Oo/o scale wind tunnel model ofTransrapid maglev vehicle with "duck bill" nose.
Figure 7. TAV 350 power car with concave, "aerodynamic" nose shape.
How co Reduce the Cross Wind Sensitivity of Trains 459
Selected results
If we take the upper level of the leeside rail as the longitudinal axis in our
reference system, then the rolling moment coefficient Cmx contains all the
information we need to calculate an overturn characteristic as shown on Figure
2. Under these circumstances, the aerodynamic stability of a vehicle under
cross wind conditions will be adequately represented by its Cmx.
Figure 8 shows a comparison of the rolling moment coefficients as a
function of yaw angle of the original ICE3 leading car with its "design-nose"
and of the same car with an optimized "aero-nose" of the TAV 350 type. On
average, the rolling moment coefficient is reduced by some 10 %. This means
that a "duck bill" nose shape would allow the ICE3 leading car to sustain 2.5
m/s or 5 knots higher cross wind speeds. This aerodynamic improvement is
also the equivalent of 6 tons of ballast, the only other way of reducing cross
wind sensitivity.
The additional benefits of the "aero-nose" should also be mentioned [4]: A
20 % lower head pressure pulse and thanks to the reduction of the tunnel
entry pressure gradient, a 10 % lower sonic-boom, should the conditions for
its appearance be met [5].
10 20 30 40 50 60 70
6 [")
Figure 8. Leading car rolling moment coefficient Cmx as a function of yaw angle g and nose
shape.
Figure 9. Standard Talgo 350 power car roof cross section (right model template) and rounded
roof section (left model template).
section, the maximum yaw angle was limited to 30°. The IC£3 model was
then tested in 2000 by Bombardier Transportation in the wind tunnels of the
Universities of Berlin and Dresden. Both facilities are quite similar to the
FlowScience wind tunnel, with closed test sections of approximately 5 m 2 cross
sectional area. According to Bombardier, the results were quasi identical to
those obtained in Manchester, so that they won't be mentioned any further
here.
In the same year, German Rail (DB) measured the model in the Porsche
wind tunnel with its 25 m 2 open test section. Siemens Transportation futher
tested the model of the original ICE3 several times in the years 2000 and 2002
in the S 10 wind tunnel described in the preceding section, as well as in the
CSTB (Centre Scientifique et Technique du Batiment) wind tunnel in
Nantes, France. The closed test section of this latter facility has a cross
sectional area of 35 m 2 • The ground clearance (to the bottom of the wheels) for
all the models tested by Siemens was 25 mm plus the displacement thickness
of the wind tunnel.
Figure 10. Effect of wind tunnel/balance on rolling moment coefficient Cmx vs. yaw angle~
(ICE3 leading car) .
7
- Aerotech balance, 912002
6
IAT balance, 10/2000
5
~
..'..... 4
X
E
0 3
Figure 11. Effect of balance and mounting on rolling moment coefficient Cmx vs. yaw angle
B measured in IAT wind tunnel (ICE3leading car).
7
~4~~~==~~~~~~~~~
0 3+f;;~~~~~~~==~====~--~--+-~
0 10 20 30 60 70 80 90
Figure 12. Effect of wind tunnel/balance on rolling moment coefficient Cmx vs. yaw angle B
with improved IAT tests (ICE3leading car).
V"' I
v. /
X
E / I
/
() ~
/ v
~
v I
0 10 20 30 40 50 60
R [o)
Figure 13. Comparison of calculated (triangles) with measured (line) rolling mo ment
coefficients Cmx vs. yaw angle K
Figure 14. Power car wind tunnel model with wedge shaped nose.
How to Reduce the Cross Wind Sensitivity ofT rains 465
080
0.40
0.00
- 0.40
-0.80
- 1.20
- 1.60
- 2.00
Figure 15. FLUENT model of power car with wedge shaped nose.
4. Efficiency considerations
A comparison of the efficiency of the experimental and the numerical methods
is made in Table 1. Computer costs were calculated on a 4 year amortization
period. As to the total time, the estimate takes into consideration the fact that
development engineers or scientists at Siemens (as in probably any other
similar company) can devote only a maximum of approximately 40 o/o of their
working time to research.
The result is surprising for two reasons. First of all, the overall costs are
comparable, essentially due to the very high time that has to be invested in
CFD modeling. More importantly, CFD that is meant to supersede wind
tunnel testing because it is not only supposed to be cheaper but also much
faster, is not yet competitive when time is considered.
466 J.-L. Peters
5. Conclusion
It has been demonstrated that the cross wind sensitivity of train leading cars
can be noticeably reduced by optimizing their roof profile and their nose
shape.
If an internal 6 component balance is used, the wind tunnel and the
Reynolds-numbers are large enough, and a few additional precautions
particularly with regard to model mounting are taken, wind tunnels can
provide vety reliable results.
For this type of investigation, CFD calculations are not yet an alternative.
Firstly, they cannot to date represent the physical phenomena with sufficient
accuracy, secondly, they do not actually offer any financial advantage and
thirdly, they demand much more time than the experimental method, at least
in an industrial context.
References
[1] Hucho W. H., "Aerodynamik der stumpfen Korper", Vieweg Verlag,
September 2002.
[2] Peters J.L., "Aerodynamics of high-speed trains and maglev vehicles: State
of the art and future potential", Int. J. of Vehicle Design, SP3, 1983.
[3] Peters J.L., "Windkanal 54 im lnstitut Aerotechnique von Saint-Cyr",
ATZAutomobiltechnische Zeitschrift, 80, no.7/7, 1978.
[4] Peters J.L., "Tunnel optimized train nose shape", Proc. of IO'h Int. Symp.
on Aerodynamics and Ventilation of Vehicle Tunnels, Boston, USA,
October 2000.
How to Reduce the Cross Wind Sensitivity of Trains 467
[5] Maeda T., Matsumara T., Iida M., Nakatani K., Uchida K. "Effect of train
nose on compression wave generated by train entering tunnel", Int. Conf.
on Speedup Technology for Railway and Maglev Vehicles, JSME,
Yokohama, Japan, 22-26 Nov. 1993.
Acknowledgement
The author would like to thank A. Rueter of the Trains Division of Siemens
Transportation Systems for providing the CFD results.
CFD Study of Side Wind Effects on a High Speed
Train
This paper deals with numerical studies in relation with cross-wind effects on
high-speed trains. 3-dimensional steady RANS simulations were performed for
the ALSTOM Transport company with the off-the-shelf STAR-CD software.
The purpose of the study was to investigate the aerodynamic performances of
three different designs of very high-speed trainsets with airflow yaw angles
ranging from 0 to 90 degrees. Each trainset was composed of five simplified
vehicles resting over a flat ground.
New meshing techniques were used in order to drastically cut down the
time required to prepare the CFD simulations. Moreover, a very careful
attention was given to the CAD preparation of the virtual aerodynamic train
models in order to create a straightforward and fully automatic meshing
process at reception of the CAD output. That led to unordinary fast meshing
times and accurate description of the boundary layer around the trains' surface.
The simulations were classically performed with the K-Epsilon RNG
turbulent model. The convergence was based on stabilization of both residuals
and aerodynamic global forces and moments on each vehicle.
Local aerodynamic coefficients and air velocities were automatically video-
like post-treated to allow fast and efficient examination of the large quantity of
results being produced. The sensitivity of the airflows and resulting forces and
moments for a given train was also investigated relatively to (i) different spatial
discretization scheme orders, (ii) different ground conditions, that is with and
without rails, with and without slip-wall conditions.
Finally, some useful conclusions could be drawn from the viewpoint of
trainset design. Moreover, important recommendations for further CFD
studies and complementary wind-tunnel reduced-scale tests were issued.
CFD Calculations by Various Methods (continued)
Commercial CFD Code Validation for Heavy-
Vehicle External Aerodynamics Simulation
Introduction
Standard truck
Faired truck
the length of the trailer, a full fairing between the cab and the trailer, and the
combination of the fairing and low-boy device.
Computational Model
The computational model employed in these studies was developed using the
ES-Aero tool for aerodynamic drag simulation that is available as part of the
Star-CO software package. The mesh is developed using a semi-automated
process that progresses in seven stages:
1. A three-dimensional hexahedral mesh is created that completely fills the
volume of the wind tunnel.
2. The mesh is refined in successively smaller zones surrounding the vehicle
until the mesh in a small region surrounding the vehicle reaches the pro-
476 W.D. Pointer, T. Sofu, and D. Weber
Turbulence Modeling
Fig. 3. Velocity magnitude profiles along the centerline of the GCM for computational
cases using the standard high-Reynolds number k-_ model, the Menter k-_ SST model, and
the renormalization group (RNG) model. Wall functions are used in the near wall region
for all cases.
Simulation Results
verged to less than lxl0-4 by the 3000'h iteration. Since velocity and pressure
data from the wind tunnel experiments will not be available for comparison
with the computational studies until the "blind" validation phase of the pro-
jects, comparisons with the measured experimental pressure and velocity data
have not yet been made. Only comparisons between the three computational
cases can be made for the pressure and velocity distributions at this time.
However, comparisons with experimental drag coefficient measurements are
made herein since this data provides little information that could potentially
be used to produce "tuned" solutions.
A sample surface pressure distribution taken from the simulation using the
standard k-£ model is shown in Fig. 4. Data shown in the contour plot of sur-
face pressure was generated by reflecting the calculated values across the sym-
metric boundary plane of the half-vehicle model. The surface pressure data
clearly captures that stagnation of flow in front of the vehicle. Also seen are
the regions of separated flow along the A-pillar, along the top of the cab fair-
ing, along the sides of the cylindrical wheels, and along the expansion feature
that would be located just downstream of the cab doors. Consistent with ex-
pectations, negative pressures are predicted along the rear surface of the cab
and the trailer, but no large pressure gradients are seen observed within those
regions. The largest pressure gradients are observed along the leading edge of
the sides of the tires and along the recirculation region just under the front
bumper. Significant stagnation and negative base pressure regions are ob-
served on all tires and axels in this simulation. This is consistent with the
static nature of the GCM model, but this distribution would be altered for real
.....
1000
0 7000
-01000
·0 1075
-ouso
05500 -Ol22S
04000 -01300
OZ'SOO -0 IJ7S
01000
-o 14SO
-0 SOOOE-01 -01525
-ozooo -01600
-03500
-o 1675
-05000 -01750
·06500 -01825
-08000
-a 19oo
-ogsoo -a 1975
- 1100 -02050
-1 2:50
-OZlZS
-1400 ·OZZOO
- 1 550 -02:275
-1700
-02350
-185(1
-OZ4ZS
-zooo ·02:500
Cab Front View
o-
·05000E-01
-·-
-01000
-01*
·OMOG
· • 3000
·03$00
·0 4000
-0 4500
-·-
-nooo
-o ssoo
·05500
---·-··--
-07000
-07500
Fig. 4. Predicted pressure distributions on the surface of the GCM model as seen in isomet-
ric, front, side and back views. Data is taken from simulations using the k-E model and
data has been reflected across the symmetric centerline boundary to generate a data set rep-
resenting the full vehicle.
Commercial CFD Code Validation 481
tractor-trailer vehicles while in service as a result of the rotation of these ele-
ments.
Pressure distributions along the front of the tractor and the base of the
trailer at the model centerline are shown in Fig. 5 for each of the turbulence
models employed. As expected, there is very little deviation between the stan-
dard k-E model and the SST model. In a manner consistent with the devia-
tions observed in the velocity field between the simulations using the standard
k-E model and the RNG model, the pressure profiles from the RNG case show
larger negative pressures in the separation zones. While a small difference is
seen between the RNG model and the other models in the predictions along
the trailer base, larger contributions to the difference in total body drag are
found along the components in the under body flow.
High-Reynolds number
0.402 1.00
k-epsilon model
1.5
1.25
•• • ...
I •• •
• • ••
• • ·--- .... ••
.......
··~
.... -:. -:.-,.,. r
{ 0.75
~
0.5
I ...."
0.25 r--- . High-Reynolds Number k-epsilon Model
~ Menter's k-omega SST Model
0
• Renormalization Group (RNG) Model
Cp
(a) Front of Tractor
1.5 ~ ....
~· ?• •
••
1.25 • ••
• • ••• . ,._i
••• ~ .f •
.
{ 0.75
• ••
...
\. • • ••
••
0.5
t .. (
•• ••••
• • ·:.. ·~
0.25
. High-Reynolds Number k-epsilon Model
-
~ Menter's k-omega SST Model
• Rinormalization Group (RNG) Model
i . I
....~
$:) ·
Cp
(b) Base of Trailer
Fig. 5. Predicted pressure distributions along the centerline of the tractor front and trailer base
surfaces of the GCM for each turbulence model considered.
Commercial CFD Code Validation 483
References
Abstract
Aerodynamic characteristics of a ground vehicle affect vehicle operation in many
ways. Aerodynamic drag, lift and side forces have influence on fuel efficiency, ve-
hicle top speed and acceleration performance. In addition, engine cooling, air con-
ditioning, wind noise, visibility, stability and crosswind sensitivity are some other
tasks for vehicle aerodynamics. All of these areas benefit from drag reduction and
changing the lift force in favor of the operating conditions. This can be achieved
by optimization of external body geometry and flow modification devices. Consid-
ering the latter, a thorough understanding of the airflow is a prerequisite.
The present study aims to simulate the external flow field around a ground vehi-
cle using a computational method. The model and the method are selected to be
three dimensional and time-dependent. The Reynolds-averaged Navier Stokes
equations are solved using a finite volume method. The Renormalization Group
(RNG) k-£ model was elected for closure of the turbulent quantities.
The external aerodynamics of a heavy truck is simulated using a validated com-
putational fluid dynamics method, and the external flow is presented by computer
visualization. Then, to help the estimation of the error due to two commonly prac-
ticed engineering simplifications, a parametric study on the tires and the moving
ground effect are conducted on full-scale tractor-trailer configuration. Force and
pressure coefficients and velocity distribution around tractor-trailer assembly are
computed for each case and the results compared with each other.
Introduction
The fluid flow in and around a ground vehicle in motion may be grouped into the
following two major categories. The external flow includes the undercarriage flow,
the flow in the gap between the tractor and the trailer(s) and the wake behind the
truck. It generates the wake that the nearby road vehicles experience and carries
Method
The set of equations solved for the present study are the time-dependent, Reynolds-
averaged Navier-Stokes equations in their conservation form. Reynolds-averaged
quantities are obtained through a time-averaging process as follows. For example,
a velocity U may be divided into an average component, U, and a time varying
component, u,
Computational Parametric Study on External Aerodynamics of Heavy Trucks 487
- - 1
U=U+u where U=-JUdt (1)
At/
where L1t is a time scale, which is large relative to the turbulent fluctuations, but
small relative to the time scale to which the equations are solved. In the following
equations, the bar will be dropped for time-averaged quantities, except for the
products of the fluctuating quantities. By substituting the time-averaged quantities,
the Reynolds averaged equations then become:
ap +V•{pU)=O (2)
at
apu +V•(pU®U)=V·~-u®u)+sM (3)
at
apE +V•(pUE)=V•~VE-puE)+SE (4)
at
The continuity equation has not been altered but the momentum and scalar
transport equations contain turbulent flux terms in addition to the molecular diffu-
---- --
sive fluxes. These are the Reynolds stress, pu ® u , and the Reynolds flux, pu¢ .
These terms arise from the non-linear convective term in the non-averaged equa-
tions. They reflect the fact that convective transport due to turbulent velocity fluc-
tuations act to enhance the mixing over and above that caused by the thermal fluc-
tuations at the molecular level. At high Reynolds numbers, turbulent velocity
fluctuations occur over a length scale much larger than the mean free path of ther-
mal fluctuations, so that the turbulent fluxes are much larger than the molecular
fluxes. Therefore, to achieve these aerodynamic simulations within the currently
available computer resources and the project milestones, the effects of turbulence
needed to be "modeled." It was realized, however, that none of the existing turbu-
lence models was developed for unsteady flows. Therefore, the present time-
accurate, finite-volume CFD methodology with its RNG k-£ turbulence model was
previously benchmarked using a series of well-documented flows (Han 1989,
Baysal and Bayraktar 2001).
The RNG k-£ model uses an eddy viscosity hypothesis for the turbulence and in-
troduces two new variables into the system of equations, k and £.The effective vis-
cosity, !Leff' is taken as the sum of molecular and turbulent viscosities. Then, for ex-
ample, the momentum equation is written as,
(7)
488 I. Bayraktar, 0. Baysal, and T. Bayraktar
given by, 2
P = f-t,VU • {vu + vur }--v • U{p,,V • U + pk)
3 (8)
Equations (2)-(4) and (6)-(7) are solved by a finite volume method. This ap-
proach involves discretization of the integral form of the governing equations,
which are solved over a number of (finite) volumes within the fluid domain. Each
node is surrounded by a set of surfaces, which comprise the finite volume. All the
solution variables and fluid properties are stored at the element nodes.
Iterative solvers, such as, the incomplete lower upper (ILU) factorization tech-
nique used herein, by themselves tend to rapidly decrease in performance as the
number of computational mesh elements increases, or if there are large element as-
pect ratios present. Therefore, the performance of the solver was greatly improved
by employing a multigrid technique. The further details of the computer code are
given in (AEA Tech 1999), and its implementation for ground vehicle aerodynam-
ics is given in (Bayraktar et al. 2002, Baysal and Bayraktar 2000, Baysal and Bay-
raktar 2001).
Computational Procedure
Tractor-trailer geometry was modeled at true scale with the dimensions of 19.5m x
2.5m x 3.9m. The size of the computational domain is shown in Fig. 1. Computa-
tional domain for tractor-trailer simulation was selected with the dimensions of
71.0m x 1l.Om x 12.5m. The distances between the model and farfield domain
boundaries are carefully chosen to minimize the spurious boundary effects. Thor-
ough investigation of farfield boundary and mesh size influence on drag coefficient
was given in (Bayraktar 2002 and Baysal and Bayraktar 2001).
A computer-aided-design (CAD) model of the truck is developed with the afore-
mentioned dimensions, and then a domain mesh is generated (Sorrells 1999). After
importing these solid surfaces into a mesh generator, the volume between the sur-
faces and the outer boundaries is discretized using 16 million cells of hybrid shapes
containing tetrahedra, prisms and hexahedra, and the surface mesh size is kept un-
der 1.8 em. A view of the surface mesh is presented in Fig. 2. Because of the
boundary layer growth on the solid surfaces, this hybrid mesh has stretched pris-
matic elements close to the body, which are, in turn, connected to the tetrahedral
cells off the surfaces. Far from the body, hexahedral elements have been used all
the way to the outer boundaries.
As this is a simulation of the external flows, the size of the computational do-
main, shown in Fig. 1, delineated by its outer boundaries, is a compromise between
accuracy and computational efficiency. Fig. 3 shows boundary conditions in the
computational domain. The domain is bounded by the ground plane, the flow inlet
boundary, the flow outlet boundary and three free-slip wall boundaries (two sides
and the top). The conditions imposed at these boundaries are required to represent
the effect of the events outside of the domain. The surface of the tractor and the
trailer provides the internal boundaries (walls).
The inlet plane is located at about one-half body length ahead of the model and
Computational Parametric Study on External Aerodynamics of Heavy Trucks 489
be normal to this boundary (as in a wind tunnel). Here, a uniform velocity profile
is prescribed, that is, the boundary layer thickness is assumed to be zero. The pre-
scribed condition at an open boundary allows for the fluid to cross the boundary
surface in either direction. For example, all of the fluid might flow into the domain
at the opening, or all of the fluid might flow out of the domain, or a mixture of the
two might occur. The velocity of the fluid on the surface of the tractor and the
trailer is set to zero to satisfy the no-slip condition. Also, scaleable wall junction is
used for turbulence model wall treatment (Grotjans and Menter 1998 and Launder
and Spalding 1974).
On the ground boundary, the velocity of the flow is set to be equal to the flow at
the inlet boundary. This emulates the ground moving with respect to the truck, as
is the case on the road. In the case of wind tunnel testing, it emulates a moving
conveyor belt floor. Although the rotating tires influence on the local flowfield, in
order to simulate common wind tunnel testing conditions, tires on the tractor-trailer
configuration is not rotated.
Results
In the aerodynamic simulation of tractor-trailer assembly, two commonly practiced
engineering simplifications, tire and moving ground affects, were investigated.
First, the external flow past the tractor-trailer assembly was computed with tires
and moving ground boundary condition. Then, stationary ground relative to the
truck (Case 2) (see, e.g., Bayraktar and Landman 2003, Summa 1992, Fukuda et al.
1995, Horinouchi et al. 1995) and other results from wind tunnels without moving
belts) and truck without the tires (Case 3) (see, e.g., Perzon et al. 1999 for this sim-
plification) were simulated. Table 1 shows case descriptions for each tractor-trailer
configuration.
Sample results are presented in Fig.s 4 and 5, which can be contrasted to observe
the effects of tires and the moving ground. As expected, the undercarriage flow is
significantly different when the tires are removed. Interestingly, the flow in the gap
between the tractor and the trailer is also dramatically altered. Because of the gap,
there is a significant pressure loss in that region. Even more significant differences
are clearly observed in the regions, where tires are located (Fig. 5). Different pres-
sure coefficient distributions in between Case 1 and Case 3 present that tire effect
on undercarriage flow even effective on longitudinal symmetry plane. In addition,
undercarriage flow is also getting affected from ground motion. When the ground
is stationary with respect to the truck (Case 2), the boundary layer on the ground
thickens to alter the entire undercarriage flow. The velocities in this region are less
than 10% of the freestream. The trailer wake is now skewed and driven towards
the ground.
Pressure coefficient distributions on the longitudinal symmetry plane of compu-
tational domain in Fig. 4 are reduced on tractor-trailer assembly symmetry surface
and plotted in Fig. 5. Although the values for different configurations collapse on
each other, the values for Case 3 are slightly differs on lower surface because of the
tire effect. The biggest pressure jump in the symmetry plane occurs at the tractor-
trailer gap region causing huge expansion and recompression on pressure coeffi-
cient values.
490 I. Bayraktar, 0. Baysal, and T. Bayraktar
After summation of the force data on the surfaces of the tractor-trailer assembly,
time averaged drag coefficient values are presented in Fig. 6. The results show that
the computed drag value at Case 3 is about 13.3% less as a result of removing the
tires. Drag difference occurs in Case 2 because of the stationary ground (simulates
wind tunnel without a moving belt), thus, the total computed drag value reduces by
4.8%. In addition, total drag coefficient is split up to its components to analyze the
local drag force on the body, the tires and the mirrors (Fig. 6). As expected, most
of the drag (82.9%) comes from tractor-trailer body. Tires and mirrors contribute
12.5% and 4.6% respectively, of the total drag coefficient. Although the effect of
the local components on drag coefficient depends on the overall vehicle design,
present study shows that presence of tires and moving ground increase the drag co-
efficient. The similar results were also obtained in the literature (Hucho 1998).
The wake flow, which is one of the most important features of bluff body aero-
dynamics, is presented in Fig. 7 and Fig. 8. Superimposed in Fig. 7 are the instan-
taneous velocity streamlines in the computational domain and the pressure coeffi-
cient contours on the model surface and the floor. When steady ground (Case 2)
and moving ground (Case 1) cases are compared, it is observed that moving ground
generates a larger wake region while the other wake vanishes on the steady ground.
On the other hand, because of the relatively higher undercarriage velocities, the
wake region is more off the ground in the case without the tires (Case 3) than it is
with the tires (Case 1). This phenomenon is also clearly seen in Fig. 8.
In order to visualize complex wake flow behind the tractor-trailer assembly, ve-
locity vectors in the wake region are plotted on cross-section planes. The first at
cross-section (x=21 m) is taken just before the rear end of the trailer, and all of the
others follow at one-meter intervals. A total of six cross-section planes are plotted
for each case, and each raw in Fig. 8 represents a different case.
The first thing that attracts attention is the wake structure, which is completely
three-dimensional in all cases. Even the formation and dissipation of side vortices
are clearly seen, especially in Case 1 and Case 3. Because of the sudden expansion,
the secondary circulations regions are remarkably noticeable. In addition, the
steady ground boundary condition unveils itself when closer to the ground in Case
2. Case 1 shows no boundary layer region on the ground, while the lower veloci-
ties exist in Case 2 and 3. Another interesting feature is noticed in Case 3. After 10
meters behind the rear end of the model, the wake regions in Case 1 and Case 2
start to dissipate onto the ground. However, wake flow in Case 3 holds off ground
with the help of stronger undercarriage flow.
Concluding Remarks
In the computations of external aerodynamics of heavy trucks, two commonly
practiced engineering simplifications, removal of tires and moving ground effects,
were investigated. In order to compare their influence on drag coefficient, the ex-
ternal flow of the tractor-trailer assembly was computed with and without the tires,
then with or without ground motion. It was concluded that differences were -8.5%
for the tires and --4.8% for steady ground. From the surface pressure distributions,
it was noted that tractor-trailer gap caused big pressure losses, and even tires on the
side of the body had significant affect on the pressure in the longitudinal symmetry
Computational Parametric Study on External Aerodynamics of Heavy Trucks 491
plane. When drag values were investigated, it was shown that most of the drag
force (82.9%) come from tractor-trailer body. Tires and mirrors contributed 12.5%
and 4.6%, respectively, of total drag.
References
AEA Tech (1999) CFX-5 Solver and Solver Manager. AEA Technologies, Pittsburgh, PA
Bayraktar I, Landman D, Baysal 0 (2002) Experimental and Computational Investigation of
Ahmed Body for Ground Vehicle Aerodynamics. SAE Transactions: J of Commercial
Vehicles 110:2:613-626
Bayraktar I, Landman D (2003) Ground Influence on External Ground Vehicle Aerodynam-
ics. IMECE2003-41224, 2003 ASME International Mechanical Engineering Congress
and R&D Exposition, Washington, DC
Bayraktar, I (2002) External Aerodynamics of Heavy Ground Vehicles: Computations and
Wind Tunnel Testing. Ph.D. thesis, Old Dominion University
Baysal 0, Bayraktar I (2001) Unsteady Wake Behind a Bluff Body in Ground Proximity.
FEDSM2001-18208, ASME Fluids Engineering Division Summer Meeting, New Or-
leans, LA
Baysal 0, Bayraktar I (2000) Computational Simulations for the External Aerodynamics of
Heavy Trucks. SAE Paper 2000-01-3501, International Truck and Bus Meeting& Expo-
sition, Portland, OR
Bradley R (2000) Technology Roadmap for the 21" Century Truck Program, A Government-
Industry Research Partnership. DOE Technical Report 21CT-001
Fukuda H, Yanagimoto K, China H, Nakagawa K, (1995) Improvement of vehicle aerody-
namics by wake control. JSAE Review 16:151-155
Grotjans H and Menter FR (1998) Wall functions for general application CFD codes.
ECCOMAS 98 Proceedings of the Fourth European Computational Fluid Dynamics
Conference, 1112-1117, John Wiley & Sons
Han T (1989) Computational Analysis of Three-Dimensional Turbulent Flow Around a
Bluff Body in Ground Proximity. AIAA J 27:9-1213-1219
Horinouchi N, Kato Y, Shinano S, Kondoh T, Tagayashi Y (1995) Numerical Investigation
of Vehicle Aerodynamics with Overlaid Grid System. SAE Paper 950628, SAE Inter-
national Congress, Detroit; MI
Hucho WH (1998) Aerodynamics of Road Vehicles. SAE Publishing, Warrendale, PA
Launder BE and Spalding DB (1974) The numerical computation of turbulent flows. Comp
Meth Appl Mech Eng, 3:269-289
Perzon S, Janson J, Hoglin L (1999) On Comparisons Between CFD Methods and Wind
Tunnel Tests on a Bluff Body. SAE Paper 1999-01-0805, International Congress and
Exposition, Detroit, MI
Sorrells MC (1999) Private communications. Volvo Trucks of North America, Greensboro,
NC
Summa JM (1992) Steady and Unsteady Computational Aerodynamics Simulations of the
Corvette ZR-1. SAE Paper 921092, SAE International Congress, Detroit; MI
Contact
Ilhan Bayraktar, PhD
Research Engineer
492 I. Bayraktar, 0. Baysal, and T. Bayraktar
Case 1 2 3
Tires Yes Yes No
Movin2 2round Yes No No
Fig. 1. Computational domain for the tractor-trailer simulations (all units are meters).
cp
.:.... ;..., 0 0
"'0 "'
00
....,
"' ...
00 0
00
0
0
1
10
1
(a)
1 20
X(m)
10
..
9
7
0
6 rv k S ur"fOCO
Co s e 1
5 -1 C s a 2
0 0 3
4
3
-2
2
(b)
0 -3
0 5 10 15 20
X(m)
Fig. 6. Drag coefficients and their components for each tractor-trailer configuration case.
Computational Parametric Study on External Aerodynamics of Heavy Trucks 495
cp Velocity [m sA · 11
"'
0 "'"' "' ..."' 0
"'
0
0
0
0
0
0 g 0
0
0
0
(a)
cp Velocity lm sA · 11
w N 0 0 0 .,.. ....
"' ~
...
0
g g
"' "'"' "'"'
0 0 0
0 0 0
"'
0 0 0
0
(b)
cp Velocity [m S"" - 1]
.
0 "'
w N 0 ~ !'" 0
.... 00 g g g g
"'0 "'"' "' 0 0 0
0 0
"' ""
(c)
x=19m
x=19m
x=19m
Fig. 8. Instantaneous velocity vectors in the wake region of tractor-trailer assembly at differ-
ent distances from the model base. First row: Case I, Second row: Case 2, Third row: Case
3 (continued).
Computational Parametric Study on External Aerodynamics of Heavy Trucks 497
x=21 m
x=21 m
x=21 m
Fig. 8. Instantaneous velocity vectors in the wake region of tractor-trailer assembly at differ-
ent distances from the model base. First row: Case I, Second row: Case 2, Third row: Case
3 (continued).
498 I. Bayraktar, 0. Baysal, and T. Bayraktar
x=23 m
x=23m
x=23 m
Fig. 8. Instantaneous velocity vectors in the wake region of tractor-trailer assembly at differ-
ent distances from the model base. First row: Case I, Second row : Case 2, Third row: Case
3 (continued).
Computational Parametric Study on External Aerodynamics of Heavy Trucks 499
x=25 m
x=25 m
x=25 m
Fig. 8. Instantaneous velocity vectors in the wake region of tractor-trailer assembly at differ-
ent distances from the model base. First row: Case I, Second row: Case 2, Third row: Case
3 (continued).
500 I. Bayraktar, 0. Baysal, and T. Bayraktar
x=27m
x=27m
x=27 m
Fig. 8. Instantaneous velocity vectors in the wake region of tractor-trailer assembly at differ-
ent distances from the model base. First row: Case 1, Second row : Case 2, Third row: Case
3 (continued).
Computational Parametric Study on External Aerodynamics of Heavy Trucks 501
x=29m
x=29m
x=29m
Fig. 8. Instantaneous velocity vectors in the wake region of tractor-trailer assembly at differ-
ent distances from the model base. First row: Case I, Second row : Case 2, Third row: Case
3 (concluded).
Applicability of the Vortex Methods for
Aerodynamics of Heavy Vehicles
Abstract.
This paper describes recent works of practical applications of vortex element
methods to study of aerodynamics of heavy vehicles, carried by the authors'
group, explaining the mathematical basis of the method based on the Biot-
Savart law. It is pointed as one of the most attractive features of the vortex
method that the numerical simulation using the method is considered to be a
new and simple technique of large eddy simulation, because they consist of
simple algorithm based on physics of flow and it provides a completely grid-
free Lagrangian calculation. As typical results of aerodynamics of heavy vehi-
cles, unsteady flows around a heavy vehicle model such as a tractor-trailer with
different gap lengths and unsteady aerodynamic characteristics of a tractor-
trailer with meandering motion are explained.
1 Introduction
The aerodynamic force and noise on road vehicles are the result of complex
interactions between the flow separation and the dynamic behavior of the vor-
tical wake. In order to design suitable shapes of vehicles, it is necessary to pre-
dict a physical mechanism of a flow separation and an interaction with vortical
wake. Recently, computational fluid dynamics (CFD) is becoming an indis-
pensable tool for vehicle design because of the advances in numerical methods
and the remarkable progress in the computer performance. A variety of nu-
merical methods have been applied to simulate the flows around a vehicle.
However, the numerical simulation of automotive flows still is not so easy
504 K. Kamemoto and A. OJima
from the view of engineering applications. The flow around a vehicle is an es-
sentially unsteady flow originated from the large scale separations of the
boundary layer. The applicability of the conventional turbulence models of
time-mean type seems questionable, as far as unsteady separated flows are con-
cerned. And the large eddy simulation (LES) of Eulerian type inevitably meets
crucial difficulties in its application to flows of higher Reynolds numbers, be-
cause the scheme essentially needs fine grids to obtain reasonable resolution of
turbulence structures.
On the other hand, the vortex methods have been developed and applied
for analysis of complex, unsteady and vortical flows in relation to problems in
a wide range of industries, because they consist of simple algorithm based on
physics of flow. Therefore, the vortex methods may be the means to provide
one of the most suitable techniques for the prediction of unsteady aerody-
namic characteristics of heavy vehicles.
Leonard (1980) summarized the basic algorithm and examples of its appli-
cations. Sarpkaya (1989) presented a comprehensive review of various vortex
methods based on Lagrangian or mixed Lagrangian-Eulerian schemes, the
Biot-Savart law or the vortex in cell methods. Kamemoto (1995) summarized
the mathematical basis of the Biot-Savart law methods. Various studies related
to the simulation of three dimensional unsteady flows around a bluff body
with vortex methods have been reported. Gharakhani et al. (1996) applied a
three-dimensional vortex-boundary method to the simulation of the flow
around tractor-trailer. Bernard et al. (1999) applied a vortex tube and sheet
method to the simulation of higher Reynolds number flows around a prolate
spheroid. Ojima and Kamemoto (2000) simulated unsteady vortical wakes
behind a sphere and a prolate spheroid by using an advanced vortex method.
Cottet and Poncet (2002) calculated the unsteady vortex features shedding
from a circular cylinder by using a vortex-in-cell method. Ploumhans et al.
(2002) applied a vortex method with parallel tree codes to the simulation of
unsteady flows past a sphere.
As well as many finite difference methods, it is a crucial point in vortex
methods that the number of vortex elements should be increased when higher
resolution of turbulence structures is required, and then the computational
time increase rapidly. In order to reduce the operation count of evaluating the
velocity at each particle through a Biot-Savart law, fast N-body solvers that re-
duce the operation count from O(N2) to 0 (N log N) are proposed (Green-
gard et al. 1987).
This paper describes the governing equation and the numerical method of
the Biot-Savart law vortex methods developed and examined up to this time by
the group of the present authors. As application examples, the numerical
simulation of unsteady flows around heavy vehicles; a simplified heavy vehicle
model such as a tractor-trailer with a gap length and a tractor-trailer with me-
andering motion, are explained.
Applicability of the Vortex Methods for Aerodynamics of Heavy Vehicles 505
The pressure in the field is obtained from the integration equation (Uhl-
man 1992), instead of the finite difference calculation of the Eq. (2.2) as fol-
lows.
+ fsc n · ~7 ds + v Jn ·
8 (VG x (l))dsj
Here, 13 =1 in the flow field and 13 =112 on the boundary S. G is the fun-
damental solution given by Eq.(5). H is the Bernoulli function defined as
follows.
p 112
11=-+- (2.7)
p 2
The values of H on the boundary surface are calculated from Eq. (2.6) by
using the panel method. Mter the pressure distribution around the boundary
surface is calculated from Eqs. (2.6, 2.7), integration of the pressure acting on
the body surface yields the force acting on the body.
Here, u,i=ui · n,i and ~Si=h · ~li. Where, ~SP , ui and n,i denote the panel
area, the velocity vector and the normal unit vector on a side sectional plane of
an element of the vorticity layer, respectively. On the other hand, the vorticity
of the thin layer diffuses through the panel into the outer flow field with a
diffusion velocity. In order to consider this vorticity diffusion, the diffusion
velocity is employed in the same manner as the Vorticity Layer Spreading
Method (Kamemoto 1994), which is expressed as the following equation.
c2v
vd =--:;;;·
(c=I.I36) {2.9)
{I) =
f I'
lO dv
(2.10)
"'' V + I/ V<JT
Every nascent vortex element which is far from the solid surface can be re-
placed with an equivalent discrete vortex element. The discrete vortex element
is modeled by a vortex blob which is a spherical model with a radially symmet-
ric vorticity distribution. The i-th vortex blob is defined by the position ri=(r.,
ry, rz), its vorticity Wi=(W,, Wy, w.) and its core radius Ei. The vorticity distri-
bution around the vortex blob is represented by the following equations.
m,(r)=m1 p(lr-t; 11&1 ) &1- 3dv, (2.14)
(2.15)
p(~) is smoothing function (Winckelmans and Leonard 1988).
The motion of the discrete vortex element is represented by Lagrangian
form of a simple differential equation iJr/iJt = u. Then, the trajectory of the
discrete vortex element over a time step is approximately computed from the
second-order Adams-Bashforth method. On the other hand, the evolution of
vorticity is calculated by Eq. (2.1) with the three-dimensional core spreading
method, in which the core radius increases with time. In this study, stretch
and diffusion terms in Eq. (2.1) are separately considered. The change of core
radius due to the stretch is calculated from the following equations.
'· I"Q)I
"'dt =rw:r dt
(2.17)
(-dE)
(2.19)
=c -v , (c=2.242)
2
dt diffusion 2E t
Taking two factors into account, each value of a new element is obtained
l
from the following equations.
(2.20)
E
t+!u
=E
t
+ [( -dE)
dt stretch
+-
dt
(dE) diffusion
·M
dl (2.21)
lt+l!.t = l, + - ·11t
dt
Applicability of the Vortex Methods for Aerodynamics of Heavy Vehicles 509
tractor and trailer with the experimental results. In Fig.5, "2-00 Cab Exp."
shown by broken lines indicates that the side corner radius is twice the length
of that of SNL model. It can be observed that the calculated results are closer
to the experimental results of 2-00 Cab indicated by broken lines. It is consid-
ered that these results were caused by the beveled shape of the side corners of
the tractor used in this calculation.
Furthermore, the aerodynamic features of a model of tractor-trailer with
meandering motion were investigated. Tractor and trailer are connected by a
pin joint at the position of x/S 112 =0.382 away from the front of the trailer.
The tractor turns to the left after turning 90 degrees to the right in the turning
radius R/ S112 =6.4. As a typical flow field, Fig.6 shows instantaneous flow
patterns represented by discrete vortex elements. Here, the maximum number
of vortex element is 331,321. The complex and unsteady vortex structure is
formed behind the tractor-trailer according to the meandering motion. Figure
7 shows the time history of drag force and side force coefficients acting on the
tractor and the trailer during meandering motion. In Fig.7, it is shown that
the drag acting on the meandering trailer becomes negative intermittently. It
is revealed from the calculation that considerable fluctuations of aerodynamic
forces inevitably act on both the tractor and the trailer as a result of unsteady
interaction of the flow separated from the tractor with the trailer. Although
the detail of the flow characteristics are left as study in the future, the present
method can be useful for the investigation of the unsteady and complex vorti-
cal flow around the tractor-trailer and for the effect of the deformation of the
gap on the characteristics of the drag acting on the tractor-trailer.
4 Conclusions
In this paper, recent works concerning the development of the methods were
overviewed, and the mathematical background and numerical treatment of a
vortex method developed by the group of the present authors were explained.
And it became dear that the vortex methods have so much interesting features
that they consist of simple algorithm based on physics of flow and provide
easy-to-handle and completely grid-free Lagrangian calculation of unsteady
and vortical flows without use of any RANS type turbulence models.
From the results of recent works of application, it has been confirmed that
the vortex method is available and useful for research and development of
aerodynamics of heavy vehicles. Finally, the present authors would like to
state that the advanced vortex method is one of the most capable methods to
contribute to the new generation of computational fluid dynamics (CFD) and
it yields a promising way to a grid-free Lagrangian large eddy simulation of un-
steady and complex flows of higher Reynolds numbers.
Applicability of the Vortex Methods for Aerodynamics of Heavy Vehicles 511
References
Bernard PS, Dimas AA, Collins JP (1999) Turbulent Flow Modeling Using a
Fast, Parallel, Vortex Tube and Sheet Method. ESAIM Vol?, pp 46-55
Cottet GH, Poncet P (2002) Particle Methods for Direct Numerical Simula-
tion of Three-dimensional Wakes J Turbulence 3-038
Gharakhani A, Ghoniem AF (1996) Massively Parallel Implementation of a
3D Vortex-Boundary Element Method. ESAIM Vol I, pp 213-223
Greengard L, Rohklin V (1987) A Fast Algorithm for Particle Simulations. J
Comp Phys 73, pp 325
Gutierrez WT, Hassan B, Croll RH, Rutledge WH (1996) Aerodynamics
Overview of the Ground Transportation System (GTS) Project for Heavy
Vehicle Drag Reduction. SAE 960906, pp 219-236
Kamemoto K (1994) Development of the Vortex Methods for Grid-free La-
grangian Direct Numerical Simulation. 3rd JSME/KSME, Sendai, pp
542-547
Kamemoto K (1995) On Attractive Features of the Vortex Methods. In: M
Hafez and K Oshima (eds) Computational Fluid Dynamics Review,
JOHN WILEY & SONS, pp 334-353
Leonard A (1980) Vortex Methods for Flow Simulations. J Comp Phys 37, pp
289-335
Nakanishi Y, Kamemoto K (1992) Numerical Simulation of Flow around a
Sphere with Vortex Blobs. J Wind Eng and lnd Aero Vol 46 & 47, pp
363-369
Ojima A, Kamemoto K (2000) Numerical Simulation of Unsteady Flow
around Three Dimensional Bluff Bodies by an Advanced Vortex Method.
JSME International} B43-2, pp 127-135
Ploimhans P, Winckelmans GS, Salmon JK, Leonard A, Warren MS (2002)
Vortex Methods for Direct Numerical Simulation of Three-Dimensional
Bluff Bodies Flows. J Comp Phys 178, pp 427-463
Sarpkaya T (1989) Computational Methods with Vortices- the 1988 Freeman
scholar lecture. J Fluids Engng 111, pp 5-52
Uhlman JS (1992) An Integral Equation Formulation of the Equation of Mo-
tion of an Incompressible Fluid. Naval Undersea Warfare Center T.R., 10,
086
Wu JC, Thompson JF (1973) Numerical Solutions of Time-Dependent In-
compressible Navier-Stokes Equations using an Integra-Differential For-
mulation. Computers & Fluids Vol1, pp 197-215
Winkelmans G, Leonard A (1988) Improved Vortex Methods for Three-
Dimensional Flows. Proc. Workshop on Mathematical Aspects of Vortex
Dynamics, Leeburg, Virginia, pp 25-35.
512 K. Kamemoto and A. OJima
I r .tCI+•I 1t ..,,l
T~ot( l ••l ••·. ,t
Tt.uk••r',lll T I . H.I~! I' ". d I
T•'l.tlat ',-.J ~ T+•l.dn",t.l •
1)4LI~~~~~~.....,.
l ie-)~~-'--:1'-;-~~~~111
ll 1, S 1 ~
0
6
0
(l"'i
---------- ~ --------- - -·
(l ..J l 2 I -1
-0 ..j
-0 6
-0 ~ -0.8
0 II) 15 20 25
til., s' ~
1 Abstract
Existing models suggest base drag is dependent upon forebody drag, however,
these models do not provide accurate predictions when applied to large-scale
vehicles. This paper describes preliminary investigations into a new base drag
model and the feasibility of minimizing total drag by optimizing the forebody-
drag to base-drag relationship.
2 Nomenclature
'This work was prepared as part of the author's official duties as an employee of the U.S. Govern-
ment and in accordance with 17 U.S.C. 105, is not available for copyright protection in the United
States. NASA is the owner of any foreign copyright that can be asserted for the work.
520 C. Diebler and M. Smith
3 Introduction
Experiments are currently underway at NASA Dryden Flight Research Center
(Edwards, California) to investigate a method of reducing vehicle base drag,
and thus total aerodynamic drag, by increasing forebody drag. Several types of
vehicles are designed with truncated base areas. Trucks, buses, and motor
homes are designed such because of their need for large internal volumes. Be-
cause of the necessity of the rocket engine at the base, most proposed reusable
launch vehicles also have truncated base areas. For all these vehicle types, base
drag comprises a major part of the vehicle total drag. The large drag forces that
arise decrease the fuel efficiency of ground vehicles or result in steep glide
slopes that limit the crossrange and downrange of reentry vehicles. Any reduc-
tion in drag is desired because it could greatly increase the overall performance
of the vehicle. Drag reductions can reduce the fuel consumption of trucks,
buses, and motor homes and lower the required glide slopes for reentry vehi-
cles, making the energy management task considerably easier. Because base
drag is such a large element of vehicle total drag, some of the greatest savings
could be made in that area.
Hoerner developed an equation [1] to predict base drag based on the
fore body drag coefficient, CD for•, for an object. Such a relationship shows that
vehicle base drag can be varied by manipulating forebody drag. Hoerner's
equation also suggests that for certain vehicle configurations, the total vehicle
drag can be reduced by increasing its forebody drag. This report discusses the
ongoing research into Hoerner's equation at Reynolds numbers to a maximum
of 3 X 10 7 • Results from initial tests and future plans also are discussed.
A Ground-Based Research Vehicle for Base Drag Studies at Subsonic Speeds 521
4 Background
Base drag originates at the aft end of an object with a blunt base, or at the
trailing edge where the flow separates from the object and a region of low pres-
sure is created. The fast-moving air going past the base acts as a jet pump,
pulling air away from the base region, resulting in pressure over the base sur-
face of the object to be reduced [1].
The boundary layer originating along the surface of a vehicle, or along
any object with a blunt trailing edge, has been demonstrated to act as an insu-
lator from the jet pump effect [1] This insulating effect reduces the effective
dynamic pressure of the outer flow, thereby weakening the jet pump effect and
resulting in reduced base drag. The thicker the boundary layer, the greater the
insulating effect. Because the boundary-layer thickness is dependant on the
flow upstream of the base, Hoerner quantifies this effect on two- and three-
dimensional objects by relating base drag to forebody drag [1]:
Two-dimensional equation:
0.135
c Dbase
(4.1)
= ;JcDJ""
Three-dimensional equation:
c = 0.029
Dba>e rc;-- (4.2)
~ '-' D fore
.25
.20
'j !1 :
c0
Three-dim. ensional: =
-reo 0.029
1 !.
1 1 : base
1 ' ' ' "'V -ufore
.5 ................... 1··················-i ·········"""•••"··f··················+·····..'"'··············-········
' I
0 .1 .2 .3 .4 .6 .7 .8 .9 1.0
020541
.8 ------------------------------------------------------- --
.6 ---·-·---------·-·---·-----------------------·-----
Co =Co +Co
total fore base ~
0 ~--------------------~----------------------~
.01 1.00
020542
Fig. 2. Base and total drag compared with forebody drag (theory)
To better understand the relationship between fore body drag and base drag, a
ground research vehicle (GRV) has been constructed. The following sections
describe the vehicle, its instrumentation, and the test approach.
5 Vehicle Description
The GRV has been designed as a facility to study the relationship between
forebody and base drag. The vehicle has been built to a scale that can ap-
proximately duplicate the Reynolds numbers experienced by reentry vehicles
along their glidepath and by trucks, buses, and motor homes. The GRV is 40-
ft long, 9-ft tall, 83-in. wide, and weighs 10,440 lb. The aft end of the under-
body has been tapered upwards, giving the GRV a square base area of 83 X 83
in 2 (47.84 ft \The base-to-wetted area ratio is approximately 3.8 percent.
The GRV has square leading-edge corners and longitudinal edges
along the top, giving the vehicle a box-like geometry. The GRV is capable of
speeds exceeding
524 C. Diebler and M. Smith
6 Instrumentation
To measure the pressures acting on all of the vehicle surfaces, 125 pressure
ports have been installed on the GRV. Two boundary-layer rakes have been
mounted on the top aft end of the vehicle to collect boundary-layer informa-
tion. The smaller of the two rakes (Fig. 4) has a dense grouping of pressure
probes near the GRV surface to provide an excellent profile of the lower
boundary layer [6]. The larger of the two rakes (Fig. 5) extends 12 in. from the
surface to capture the outer portion of the boundary layer. The GRV has been
outfitted with a noseboom to provide free-stream static and total pressure.
Four pressure modules have been installed inside the vehicle, each
with 16 individual differential pressure transducers, allowing 64 differential
pressures to be simultaneously measured. The pressure transducers are capable
of providing differential pressure measurements accurate to within ±0.004
lbf/in 2 • An absolute pressure transducer has also been installed to measure the
differential reference pressure to within ± 0.015 lbf/in 2 •
A Ground-Based Research Vehicle for Base Drag Srudies at Subsonic Speeds 525
l
2.940 ln.
0.188 in.
_L ,..,..-----=:b~~~
T
Fig. 4. The small boundary-layer rake
All pressures are recorded onto a laptop computer. A differential global posi-
tioning system (GPS) receiver internally records data during testing. The data
are later downloaded to obtain groundspeed (accurate to within 0.5 ft/sec), al-
titude (accurate to within 6 in.), and time. A handheld GPS unit serves as the
speedometer and is used as a time stamp to synchronize the differential GPS
data with the pressure data.
526 C. Diebler and M. Smith
7 Test approach
forward of the base area, C Dp. The C Dp includes the pressure forces acting on
the front face of the GRV as well as pressure drag originating from wheel wells
and the exposed underbody. The CD,;" includes skin friction as well as drag
caused by forebody separation. The CDb"" is defined as drag resulting from the
low-pressure region at the base of the vehicle.
Two types of tests were devised to obtain the drag components for the
GRV. A coast-down method was used to obtain the CD,owt, and constant
speed tests were conducted to break the C Dw,at into its constituent parts. A
portable weather tower measured ambient temperature and windspeed. Test-
ing was not conducted in winds greater than 5 kn.
Indicated airspeed was obtained from noseboom total and static pres-
sure measurements. Noseboom static pressure measurements were corrected
for position errors using a calibration curve that was generated using the GPS
altitude and ambient pressure data. The position error calibration (Fig. 6) cor-
rects the measured static pressure as a function of indicated airspeed. True air-
speed then was calculated using calibrated static pressure, total pressure, and
ambient temperature. All tests were conducted on the north base runway,
which is approximately 1-mi long, at Edwards Air Force Base (California) .
.05r---~----~----~--~----~--~~--~----~---.
I I I I I I : :._ .( • I
~P =(7 x1o-6)v.
.03 -----------~-----------i---1-~-------~-----------~--~~------~-----------~--:-~=-·~_i.-;:.... ~----------
: : : : : : : ..=>t- .. :
~p• .02
lbflin 2 -I· T T~~~~:;0~'~:+_: ~-
.01 -----------r----------+---------::r-- >I --~ r----------+----------+----------~----------
.......
:
I
: ..... ;;.
0
: I
:
o
:I
I I I 0
1 t I 0
I I I I
I I I I
0 +-----+--"""":!~
-.01 -'-----------------~---------------..i-.-....J
0 10 20 30 40 50 60 70 80 90
Vind• ft/sec 020546
8 Coast-Down Tests
In one dimension, the equation of motion for the GRV during a coast-down
test is as follows:
(8.1)
where m is vehicle mass, ax is axial acceleration, Aref is the vehicle base area
used as the reference area for these tests, W is the GRV weight, P is air den-
sity, Vis the true airspeed, and (} is the runway slope. The runway slope was
reconstructed using altitude data from the differential GPS unit so (} could be
computed at each point along the runway. Mechanical drag, Fmech, encom-
passes all of the nonaerodynamic losses of the vehicle, including rolling fric-
tion, wheel inertia, and mechanical losses. Mechanical losses were significantly
reduced by disengaging the drive train before the coasting portion of the test,
allowing the wheels to rotate freely without any drag caused by turning the en-
gine or transmission.
Using equation (3), the CD'"'"' of the GRV can be estimated by meas-
uring its deceleration while it coasts. Coast-down tests were performed by ac-
celerating the GRV to a target speed, such as 65 mi/h; disengaging the drive
train; and allowing the vehicle to coast to a stop. This technique is commonly
used to obtain the aerodynamic drag of vehicles [7,8,9]. Tests were conducted
in both directions along the runway. Ten or more coast-down tests were per-
formed for each configuration.
by subtracting the C DP and the C D~x,_ from the estimated CD'""'' from the
coast-down tests:
(9.1)
L
•
.,
L-.. :1: •
:.::
G:lll4t IXQ;48
(a) Front locations (b) Ra e locarions
Fig. 7. Pressure port locations
.I I
I
0
;
l
•
-..1 •
•
Fig. 8. Surface fit to front pressures
A Ground-Based Research Vehicle for Base Drag Studies at Subsonic Speeds 529
.1
Separated
0 region
-.1
-.2
-.3
cP
- .4
-.5
-.6
-.7
-.8
0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1.0
Length , percent
020551
Fig. 10. Centerline pressure profile along top of GRV showing flow separation
over approximately forward 30 percent of vehicle.
c Dtotal
CD p
c Dvisc
c Dfore
c Dbase
1.8
1.4
A 2
1.2
Total aerodynamic drag
1.0 .. ~ • • • • • • !Jc,• • • " . ·-~ (J-! c;.>~r:ll~r:~ t~~;'!i!I2~1::!~L0!1.al. •,., •• ·-~ •••••.•• "' •• ,., .•.•
.6
.4
.2
provides the most accurate model for predicting base drag on large-scale vehi-
cles, or if a new model is needed.
11 Concluding Remarks
Preliminary tests conducted at NASA Dryden Flight Research Center on two
configurations of a ground research vehicle (GRV) provide support for the hy-
pothesis that a new base drag prediction model is needed for large-scale vehi-
cles. Plans currently are underway to begin testing on a third GRV configura-
tion that will include rounded leading-edge corners. This new configuration
will allow the boundary layer to remain attached over the entire length of the
vehicle and is expected to provide data near the minimum total aerodynamic
drag point. Subsequent GRV configurations will increase the surface rough-
ness by adding various-sized boundary-layer trip strips to control the bound-
ary-layer thickness. These future configurations will provide additional data
and further studies of effects of forebody drag on base drag. Such test results
will either confirm or refute the need for a modified version of Hoerner's base
drag equations that applies to large-scale vehicles. Upcoming tests will also
provide insight into the feasibility of reducing drag by manipulating the
boundary layer.
References
1. Hoerner SF (1965) Fluid-dynamic drag: practical information on aerody-
namic drag and hydrodynamic resistance. Self-published work, Library of
Congress Card Number 64-19666, Washington, DC
2. Saltzman EJ, Wang KC, Iliff KW (1999) Flight-determined subsonic lift
and drag characteristics of seven lifting-body and wing-body reentry vehi-
cle configurations with truncated bases. AIAA 99-0383 (Also published as
NASA TP-1999-206573)
3. Saltzman EJ and Meyer RR Jr (1999) A reassessment of heavy-duty truck
aerodynamic design features and priorities. NASA TP-1999-206574
4. Whitmore SA and Moes TR (1999) A base drag reduction experiment on
the X-33 linear aerospike SR-71 experiment (LASRE) flight program.
NASA TM-1999-206575
5. Whitmore SA, Hurtado M, Rivera J, Naughton JW (2000) A real-time
method for estimating viscous forebody drag coefficients. NASA TM-
2000-209015
6. Bui TT, Oates DL, and Gonsalez JC (2000) Design and evaluation of a
new boundary-layer rake for flight testing. NASA TM-2000-209014
7. Lynn OK et al. (1979) Determination of vehicle rolling resistance and
aerodynamic drag. Los Alamos Scientific Laboratory, LA-UR-78-3229
8. Montoya LC and Steers LL (1974) Aerodynamic drag reduction tests on a
full-scale tractor-trailer combination with several add-on devices. NASA
TM X-56028
9. Saltzman EJ, Meyer RR Jr, Lux DP (1974) Drag reductions obtained by
modifying a box-shaped ground vehicle. NASA TM X-56027
Splash and Spray Measurement and Control:
Recent Progress in Quebec
Summary. This paper presents results of a series of road tests measuring the spray
clouds generated by a truck/trailer traveling on a watered roadway with and with-
out specifically-configured splash guards covering all the wheels. The tests were
conducted with a carefully-arranged set of laser transmissometers, and data show
significant reductions in spray density at some fixed positions from the truck-trailer
and the roadway. Data is also given showing the effect of adding a drag-reducing
air shield on the truck. Emphasis is put on the full-scale test procedures and on the
reliability /repeatability of the measurements.
1 Introduction
Fig. 1. A mild case of reduced visibility due to the splash and spray of a typical
heavy truck.
is evacuated away before it gets a chance to reform large droplets that would
quickly fall back to the ground.
It should be emphasized here that the formation mechanism of spray and
its later dispersion involves quite complex physics and aerodynamics as well
as a large variety of parameters which can be classified as "truck-centered" or
"situational" [2]. For example, in the former category, one would include the
tires treads, the speed and load of the truck, its general shape and underbody
details, while in the latter category, one would list the pavement texture, the
thickness of the water film on the road, and the magnitude and direction of
the ambient wind.
Strangely, the important progress made in the 1970's and 1980's on several
of these aspects has not brought us much closer to practical, real life solutions
of the splash and spray problem in the 1990's and up to now. In our view ,
the main reason for that is twofold: first, no system or devices tested so far
seem to provide the desired overall level of effectiveness at a minimum cost
and with minimal adverse effects (on the drag, the brakes, the maintenance,
etc.), and second, perhaps more importantly, the lack of recent development
turns out to be fundamentally linked to the difficulties associated with the
measurement of the phenomenon itself.
Indeed, virtually all large scale efforts in this area of research were aban-
doned when in 1988, the National Highway Traffic and Safety Administration
(NHTSA) announced [3] that it was terminating its pending rulemaking (or-
dered in 1982) to require splash and spray suppression devices on a ll new
truck tractors, trailers and semitrailers because it determined that no avail-
able technology had been demonstrated to
" ... significantly reduce splash and spray from truck tractors, semitrailers and trail-
ers, and significantly improve visibility of drivers, as demonstrated during testing
on highways, at test facilities , and in laboratories to take into account possible
wind and rain conditions." [3]
The major problem for the NHTSA was a lack of consistency and too much
variability in the full-scale road test data that had been produced to establish
the effectiveness of a wide variety of splash and spray suppression devices and
Splash and Spray Measurement and Control 535
In the SAE recommended procedure [9], two types of approach are proposed.
Both of them involves stationary test setups, meaning that the test-truck is
driven through a wet test section with the measuring instruments installed
on both sides of the track. An alternative setup attached to the truck has
been proposed in Europe [10], but it seems more appropriate for local splash
measurements than for global spray clouds characterization that is of interest
here.
The first technique recommended in the SAE standard is called the "video
digitizing method", and it uses large checkerboards (see Fig. 2a) and image
processing techniques to measure the loss of contrast when the spray cloud
happens to be between the camera and the target board. Although interesting
in principle in terms of providing some discrete spatial information over entire
planes, this approach was not selected here due to the post-processing delays
involved and the significant crosswind sensitivity expected.
The second, and more attractive method in our opinion, is called the
"laser transmissometers method" and it measures the loss of light transmis-
sion through the mist, i.e., the opacity of the spray cloud. According to the
standard, two lasers are used on each side of the track (see Fig. 2) and are set
at a height above the road corresponding to a typical car driver's line of sight.
The four corresponding photometers are positioned in front of the lasers at
least 15 m (50 ft) away. The signal from each of the photocell is continuously
recorded during the passage of the test vehicle.
Assuming here the "SAE low crosswind scenario", i.e., crosswind compo-
nent less than 5 km/hr (3 mph) during the run of a given test truck config-
uration on the wet track, the procedure J2245 defines the "measurement for
that run" as being the arithmetic average of the four photocell values. Each
of the sensor measurement is expressed as a percentage of light transmittance
reduction (% obscured) called "Figure of Merit" (FOM) which is calculated
as
Laserc::> I a Photocell
(
c:>
~,66m(12ft)
D
c:> D
(b) Anemometer
Fig. 2. SAE's recommended practice J2245. (a) View from behind showing the 4
lasers and the checkerboards; (b) Plan view of the test section. Adapted from [9].
Splash and Spray Measurement and Control 537
where Vmax is the voltage before the passage of the vehicle (unimpeded beam),
Vmin is the minimum voltage recorded during the run, and Vo is the "zero
voltage" (offset) measured when the beam is occluded. A 100% means to-
tal opacity of the spray cloud while 0% means total transmittance, i.e., no
visibility impairment.
In the present work, the authors have preferred to use the terminology
"Opacity Index" (OJ) in place of Figure of Merit to facilitate interpretation
and to emphasize the difference put forward with respect to Vmin as is dis-
cussed further. However, one should note for the moment that the SAE's FOM
will correspond to the maximum instantaneous value (peak value) of the OJ
proposed in this work.
The overall FOM of a test truck configuration, equipped or not with an
anti-spray device or system, is finally obtained according to the SAE's stan-
dard by ensemble averaging over a minimum of 8 different runs at a given
wind condition such as, for example, low head wind or low left crosswind [9].
The lower the score, the better the given configuration is classified in t erms
of visibility impairment and safety hazard.
For our first road test campaign in 1997, an original technique inspired of
the SAE's laser method was conceived with a particular aim to provide a
richer information about the lateral distribution of opacity through the spray
cloud. Monotonic decay of cloud opacity with respect to lateral distance from
the truck seemed to be taken for granted in previous works [2]leading to the
SAE's standard. It was believed by the present authors that, for the purpose of
comparing or developing spray suppression devices, more detailed data about
the spatial distribution of the mist was important. Information closer to the
side of the vehicle (sensors are at 1.22 and 2.13 m in SAE's J2245; Fig. 2)
was also desired since it represents a crucial visibility region for a following
motorist in the process of deciding or not to overtake the heavy vehicle in
front of him on a two-way road.
(a) (b)
Fig. 3. First road campaign in 1997. (a) Ramp of 15 sensors within their protective
tubes inside the photometers housing; (b) Laser and oscillating mirror.
538 G. Dumas and J. Lemay
5-axle tractor-trailer
truck {20 m, 65 ft)
Photocell
housing
Laser housing
(a)
Spnnklers
/?,
61 m(200 II) Wet
sectoon
Trigger
""=II
=
Anemometer Rampof7 -
(b) 15 pholocells
Fig. 4. Test track of the first road campaign in Blainville, Quebec. (a) General view
of the site prior to the tests; (b) Sketch of the setup about the wet track.
Splash and Spray Measurement and Control 539
Extracted signals
Vnm·V: : No m1sl s1~nal 1
0 0.01 0.02 0.03 0 04 0 05 0.06 0.07
Time (s)
Fig. 5. Typical sensor signal among 15 recorded during a run. The time interval
between t2 and t3 is the data analyzing window which corresponds to the truck
entering and exiting the central 30m between laser and sensors. The non-vanishing
signals in the unimpeded portion and in the analyzing window (clean and perturbed
gaussian responses of the swept-by sensor) are then regrouped for processing.
(a) (b)
Fig. 6. Two of the truck configurations tested in the first campaign. (a) The
"base configuration" (Base) of the 5-axle tractor-trailer test vehicle with its stan-
dard European-type fenders; (b) The "reddaway configuration" (RW) of the truck
equipped with some homemade Reddaway-type of fenders [1][2].
540 G. Dumas and J. Lemay
0 /1 (%) 100
I
--- RW
..... :-, -- Base -
- .....
80
60
- ~ ..._
-,
' ~
40
'
SAE Laser #1J
~ ''
20 ....
l SAE Laser ~r---. .... .....
2 3 4 5
Position (m)
Fig. 7. Example of results from the first campaign: Lateral distribution of opacity
index OJ 1(y) at a height of 1.14 m for the base and reddaway configurations. The
lateral position y is defined from the side of the truck's trailer. The curves plotted
here represent ensemble averages over 8 runs in each case, all under SAE's condition
of low wind.
Plotting the opacity index data of all sensors on the ramp against the
lateral position Yi of each sensor - measured with respect to the actual po-
sition of the test vehicle on the track for that particular run - , one obtains
the "Lateral Distribution of Opacity" OI 1 (y) for that given run and truck
configuration. The distributions obtained during this first campaign for the
case of the "base truck" and for the "reddaway" configuration (Fig. 6) are
shown on Fig. 7.
These results clearly suggest that simple systems such as reddaway fenders
can have measurable and positive impact on spray clouds in terms of both,
their peak opacity and their spatial extent. From previous studies [1] [2], one
can infer that opacity levels of 70% and above are perceived by human eyes as
"very bad" (no useful visibility). One would therefore require that an efficient
anti-spray device yields much lower levels (still to be determined) across the
whole width of the cloud, and in particular in the first 2 m from the side of the
truck. It is also noted from Fig. 7 (and other results of this campaign) that in-
deed opacity distributions appear to be reasonably monotonic and decreasing
functions, except perhaps very close to the vehicle.
Repeatability of the results provided by full-scale road tests being at the
heart of the problem in splash and spray research, dispersion in the raw data
and statistical convergence of the final results have to be considered. First, it
was noted that the dispersion in the OI 1 data was larger than expected: for
the base configuration, about ±12 for the inner part of the cloud (first 2 m)
Splash and Spray Measurement and Control 541
and up to ±20 in the outer part (last 2 m). Large external flow intermittence
may explain in part the difference between the two regions, but it is most
likely attributable to what has been termed the "triangle effect", and that is
described on the sketch of Fig. 8. The phenomenon tends to add more noise
to the calculated indices in the outer portion of the cloud which makes it
more challenging to reach good statistical convergence, and is therefore an
undesirable weakness of this approach.
In any case, significant dispersion in the 01 1 data calculated by (2) is to
be expected due to intrinsic aerodynamics unsteady effects (air turbulence
and large-scale eddies) to which one must add the effects of varying winds
(intensity and direction) between runs. Even for "low wind" conditions (Uw <
5 km/hr) considered here, important impact of the wind is expected as will
be addressed further. Furthermore, non-repeatability and non-uniformity in
the water film are important issues to be minimized, which was not rigorously
the case in this first campaign where watering was provided by distributed
sprinklers along the track.
Most importantly, one must realize that in this approach as well as in the
SAE's recommended procedure, the only way to improve statistical conver-
gence of the final results , and thus repeatability, is by doing more runs in
the same - as much as possible - conditions. Indeed, the only averaging used
is ensemble averaging which, according to us , requires many more than the
8 runs (per truck configuration) recommended and that were realized here.
Four times as many runs would reduce uncertainty of the mean distributions
by a factor of two which may be viewed as a minimum. Considering different
wind conditions - as many as 8 according to the SAE's classification - as well
as different speeds of the test vehicle and different depths of the water film,
would add up quickly to prohibitive proportions in terms of both time and
testing cost.
~Approximate
/ spray cloud
p----'_...1111(.
Laser sweeping
Ramp of
laser beam photocells
(Length= 61 m)
(Width= 4 m)
Fig. 8. Exaggerated sketch showing the "triangle effect" that contributes to add
noise in the readings of the sensors which are not directly in front of the laser
source. While the truck moves through the central part of the test section (analyzing
window, 31 m), and considering a given photocell on the ramp, this effect yields
varying optical paths of the laser beam with respect to the cloud.
542 G. Dumas and J. Lemay
For our second campaign in 1999 (see [7] for a detailed report) , we used a
small t est track in the region of Quebec City. The track had a fairly uniform
1% slope across the pavement which allowed us to produce a quite uniform
and repeatable film of water (Fig. 9a), 1.5 mm t hick, over the entire 61 rn
(200 ft) of the test section. A 4-inch perforated pipe running along t he t rack
(a)
Lasers
Wet pavement
61 m (200 fl)
F ig . 9. Second road campaign. (a) Upstream view of the t est t rack and of the
continuously fed water film; (b) Sketch of the setup about the wet track.
Splash and Spray Measurement and Control 543
(a) (b)
Fig. 10. Second road campaign setup at run time. (a) Upstream view showing the
3 lasers at the forefront; (b) Downstream view with the 3 photocells at the forefront.
and fed by a regulated fire hydrant was used to that end. For all practical
purposes, contribution of the water film to the final dispersion of the data
was reduced to negligible proportion.
In this campaign, in order to eliminate the undesirable "triangle effect"
and to provide for some temporal averaging of the signals, each photocell
used had to have its own dedicated laser aligned with it, in much the same
way as in the SAE's recommended practice J2245 discussed previously. The
continuous signal recorded for each sensor while the test vehicle is passing
through the "analyzing window" would permit the temporal averaging needed,
which departs in a fundamental way from the standard J2245. Based on the
results of the first campaign - relatively monotonic decay of lateral opacity
distributions - it appeared that three sensors (Fig. 9b) would be sufficient
to properly characterize the opacity distribution. The positions retained were
selected as: 0.61, 1.22 and 2.44 m (2, 4 and 8 ft; nominal values wjr to the
side of the vehicle's trailer). This reduc~d number of photocells allowed to get
rid of the large and perturbing protecting housings of the first campaign as
can be seen on the pictures of Fig. 10.
Typical signals recorded at run time by the three photocells are shown
on Fig. 11. Note that sensor #1 is the inner one, closest to the truck, while
#3 is the outer one at 2.44 m (8 ft) which would correspond to the middle
of the neighbor lane next to the truck's lane. The data indicates that the
outer part of the spray cloud takes about just under 59 m (193 ft) to build
up from zero and stabilizes. Of course, wetting of the track upstream of the
test section would easily fix this problem. This is unfortunate in the spirit of
the present method since it severely limits the time window over which one
can average the signals (assuming a common window for the 3 sensors). The
last 0.10 second (last 2 m) before the vehicle started exiting the test section
has usually been selected in this campaign.
Nonetheless, this modest temporal averaging, combined with the usual
ensemble averaging, can significantly improve convergence of the mean results,
544 G. Dumas and J. Lemay
v
0.5
0.5
0.5
;! Rh ,,m~'="''-+----1.tjj
0 o~~~~~~1~~----~2~--------3~~--~~~4~~~--~5
Time (s)
Fig. 11. Signals recorded for each of the 3 sensors and the triggers during a typical
run of the second road campaign.
even though dispersion itself may still remain important, as can be seen in
Fig. 12. In this campaign, one now defines the OJ measured by each of the 3
photocells in a given run as
0 o~~~_L~~~~~~2~~~~J3
Position (m)
Fig. 12. Raw results of the Opacity Index OJ according to (3) for the base truck
configuration in the second road campaign ("low crosswind" runs). The 95% con-
fidence bands of the three mean values are also shown. The line connecting these
values is only suggestive of the actual distribution of opacity.
Our second road campaign also provided the opportunity to test the effect
of a simple drag shield installed on the tractor cab. As expected, such air
shield tends to keep the mist closer to the truck thus increasing opacity in
that area. However, combined with conventional anti-spray systems such as
reddaway fenders, it tends to improve their performances in the mid and outer
portion of the cloud as can be inferred from the results presented in Fig. 13.
·-
100~~~~---------------,
0/ (%)
..... RW meanOI
·~
A ~.mean()j
a
s
c
-: (b)
0 o~----~------~2------~3
Fig. 13. (a) Mean opacity results from the second campaign for the RW (reddaway)
versus the AS-RW (air shield + RW) configuration- low wind condition, V = 70
km/hr; (b) Test vehicle equipped with a basic air shield on the tractor cab and
reddaway-type of fenders on the trailer.
546 G. Dumas and J. Lemay
The experience gained from our two road campaigns can be summarized by
the following recommendations:
Fig. 14. Sketches of approximate spray cloud envelopes under no wind condition
(left), and under some "low crosswind" condition, e.g., Uw ~ 4 km/hr (right).
There remains the challenge of how to properly take into account the effect
of crosswinds in the overall evaluation of spray suppression devices. It seems
to these authors that a finer grading system than the one proposed in the
SAE's procedure is required. As suggested in Fig. 14b, it takes very little
crosswind to significantly alter the distribution of spray around the vehicle.
This represents a critical issue which still needs to be addressed.
Splash and Spray Measurement and Control 547
Acknowledgement
The authors would like to acknowledge the financial support of the Min-
istere des Transports du Quebec, contracts MTQ/1220-97-RGOl and 1220-
98-RG02. Help and encouragement from the project manager at DoT, M.
Mario Bussieres, is also gratefully acknowledged. Special thanks to the wel-
coming and professional staff of the CFTC (Centre de Formation en Transport
de Charles bourg; www. cftc. qc. ca) where our second road campaign was re-
alized.
References
1. Weir DH, Strange JF, Heffley RK (1978) Reduction of Adverse Aerodynamic
Effects of Large Trucks. Technical Report FHWA-RD-79-84, Federal Highway
Administration, USA
2. Weir DH (1980) Truck Splash and Spray - Full Scale Tests and Alleviation
Devices. AIAA Paper 80-0203, 18th Aerospace Sciences Meeting, Pasadena,
CA, 14-16 January, 1980
3. NHTSA (1988) Splash and Spray Suppression Devices - 49 CFR Part 584:
Termination of rulemaking. May 25, 1988. U.S. Federal Register, Vol. 53, No.
101: 18861-18868
4. NHTSA (1994) Splash and Spray Suppression: Technological Developments in
the Design and Testing of Spray Reduction Devices for Heavy Trucks. Report
to Congress March 1994, National Highway Traffic and Safety Administration,
U.S. Dept. of Transportation
5. NHTSA (2000) Update on the Status of Splash and Spray Suppression Tech-
nology for Large Trucks. Report to Congress March 2000, National Highway
Traffic and Safety Administration, U.S. Dept. of Transportation
6. Dumas G, Lemay J, et. a!. (1998) Comparative evaluation of splash guard
systems for heavy vehicles (in French; English translation also available). Con-
tract report MTQ 1220-97-RGOl for the Quebec Dept. of Transport, Aprill998,
Fluid Mechanics Laboratory, Laval University, Quebec City
7. Dumas G, Lemay J (2000) Study of spray cloud reduction on heavy trucks by
aerodynamic control: Part I- Full-scale road tests (in French). Contract report
MTQ 1220-98-RG02 for the Quebec Dept. of Transport, March 2000, Fluid
Mechanics Laboratory, Laval University, Quebec City
8. Dumas G, Lemay J (2002) Study of spray cloud reduction on heavy trucks by
aerodynamic control: Part II- Wind tunnel tests (in French). Contract report
MTQ 1220-98-RG02 for the Quebec Dept. of Transport, January 2002, Fluid
Mechanics Laboratory, Laval University, Quebec City
9. Johnson WA, Weir DH (1994) Recommended Practice for Splash and Spray
Evaluation. 1996 SAE Handbook, Vol. 3, SAE Recommended Practice J2245
10. Hucho WH (1998) Aerodynamics of Road Vehicles, 4th ed., Section 9.7.4, SAE
International, Warrendale, Pennsylvania
Wind-Tunnel Evaluation of an Aerodynamic Heat
Exchanger
Abstract
GTRI has recently been developing pneumatic aerodynamic concepts for ap-
plication to Heavy Vehicles under a Department of Energy contract through
the Oak Ridge National Laboratory (ORNL). A related application under de-
velopment is a novel heat exchanger known as the Aerodynamic Heat Ex-
changer (AHE). This patented device employs airfoil/wing aerodynamic pres-
sure differences to induce large mass flows across a radiator installed inside a
wing. GTRI has recently completed an in-house wind tunnel test of this con-
cept. The objective of this proposed effort was to perform a wind-tunnel
evaluation of the AHE and establish the feasibility of the concept. A 2D wing
was fabricated with a removable center section. A radiator core was integrated
into this section of the wing. A conventional radiator core (based on a Visteon
design) and two cores made from carbon foam were tested. The carbon foam
cores were designed and provided to GTRI by ORNL. Hot water was allowed
to pass through the inside of the wing while freestream, wind tunnel air passed
over (and through) the wing. Heat rejected by the radiator was measured as
well as lift and drag. Results indicated that the concept is feasible and can pro-
vide an effective means to reduce vehicle drag by reducing the drag due to
conventional radiators.
Introduction
Motivation
Current engine heat exchangers for heavy vehicles and automobiles are neither
aerodynamic nor fuel efficient. This is evident visually as the radiator grill is
550 R.J. Gaeta, R.J . Englar, and G. Blaylock
usually the most distinctive feature on a tractor rig as seen from the front end.
Passenger automobiles get their distinctive look from the radiator grill design.
Furthermore, ram scoops on high performance race cars are necessarily evils:
they are needed for cooling air flow, but are drag generators. A common fea-
ture in radiator design is the need for large amounts of air to flow through the
heat exchanger. The solution for over a century has been to place this heat ex-
changer normal to the oncoming air stream. This is a very effective means of
producing heat transfer but it is also an effective means of creating drag [see
Figure I]. A novel approach to an alternate cooling solution is embodied in
the notion of housing the heat exchanger in a low drag envelope, namely, a
wmg.
Figure 1 Large Heavy Vehicle and passenger car: radiator presents large frontal area to
flow. thus significant drag. Fl & Indy cars: pods or ram scoops are necessary evils- speed
draining drag sources
Flat 6.P at
Plate expense of
high drag
... .n
Wing
~P with
·. low drag
Figure 2. A wing has an order of magnitude lower drag coefficient than a nat plate AND has
a mechanism to produce a pressure differential needed for heat transfer.
A proof-of-concept model was built and tested in a low speed wind tunnel at
GTRI to see if this concept was a feasible solution. Several radiator core con-
figurations were tested and the heat removal was measured. The remainder of
this paper presents the testing approach and set-up along with results and dis-
cussiOn.
..
Pomu' Hro:ll Conductmg Materia l
High Velocity
Jet Sheet
- - - Phenomenal
Increase in AP
Figure 4. Controllable lift, thus controllable heat transfer with pneumatic flow
control.
AHE Configurations
The airfoil shape chosen for the model AHE concept was elliptical with a
round trailing edge, similar to that shown in Figure 4. The aerodynamic refer-
ence configuration was a non-porous center section. Three porous center sec-
tions were fabricated which represented three different types of radiator cores.
ORNL supplied a radiator that had the same external dimensions as the Vis-
teon radiator, but was made from solid pieces of carbon-graphite foam mate-
rial. This material has phenomenal heat conductivity properties. Although it
is porous, the bulk density is such that it has a significant pressure drop. Brass
tubes were press fit into the foam to carry the coolant through the material for
heat exchange.
C.onvemloha1·· Radiator -
V/1toon
" Dense" Carbon Graphite Core
- ORNL
Figure 5. Radiator core configurations used for small scale wind tunnel model.
Experimental Set-up
The elliptical wing with the radiator core was installed vertically in the wind
tunnel and was attached to the force balance. The airfoil was connected via
flexible hoses to a 3-phase electric 3600W water heater. Water was heated and
pumped into one side of the radiator. It was allowed to settle in the inlet res-
ervoir before moving through core and settling in the outlet reservoir. The
outlet water was then allowed to return to the water heater closing the coolant
loop. Coolant mass flow was measured with a water flow meter. Thermocou-
554 R.J. Gaeta, R.J. Englar, and G. Blaylock
pies were placed in both inlet and exit reservoirs to monitor the temperature
drop across the core. All fluid measurements were fed into a Labview program
running on a Windows platform, which stored mass flow and temperatures.
Figure 6 shows the wing installed in the wind tunnel with the radiator core
removed. Piping and thermocouples are visible. Figure 7 shows the Visteon
radiator configuration and the dense carbon-graphite radiator installed in the
wind tunnel.
Blowing
Slot
Tufts
Pressure
Taps
Figure 7. Two radiator core configurations for the AHE installed in wind tunnel.
A typical run for a given radiator configuration would include a "sweep" of slot
blowing pressure at constant angle of attack (a) and tunnel speed (dynamic
pressure) , to record and evaluate aerodynamic characteristics. Then, for each
Wind-Tunnel Evaluation of an Aerodynamic Heat Exchanger 555
radiator core, the coolant lines were added (these would have caused balance
tares during the aero runs) and temperature data were taken at constant cool-
ant flow rates for variable blowing pressures. Variation in tunnel speeds was
also conducted for the radiator airfoils at constant flow rates while varying
blowing pressures. For reference, the conventional Visteon radiator was evalu-
ated without blowing or airfoil frame but perpendicular to the freestream flow
so as to simulate a standard radiator's cooling characteristics.
c mV·
=--} (1)
t-t qS
It should be noted that it is assumed that the AHE radiator airfoil was to be
mounted on race car (a GT Motorsports SAE car), intended also to give down
force for cornering and traction. Thus the model airfoil is mounted inverted
in the tunnel, with negative lift (positive downforce) towards the ground as the
lifting side of the airfoil is towards the road, and negative angle of attack a is
leading edge downward.
A simple heat balance was used to quantify heat rejection of the coolant to the
air passing through the airfoil into the tunnel. The heat transferred from the
coolant can be expressed as:
(2)
Aerodynamic Results
20 AlrfoU
r---+--11 Radiator ore
TE Blowing
The aerodynamic portion of the tests (i.e., the radiator installed and blowing
applied, but no coolant hoses connected and no coolant flowing) was con-
ducted first to evaluate the effects of porous sections in the center of the lifting
airfoil.
Figure 10 shows how blowing and porosity affect the lift and drag. These re-
sults conform directly to the pressure differences in Figure 9. As porosity in-
creases, lift decreases and drag increases, but still increased blowing is very ef-
fective. As downforce (-CL ) increase due to blowing, the high circulation
·10
••
•
-15 • •• a CJn"oornt.oni
bd'
.....
• • \1 • ton
••
•.l .O
•
-45
•• • a
I De...
1 tU
-5CI
0.20
Co
Figure 10. Drag polar for AHE with several configurations.
around the airfoil causes the leading edge to separate (or a bubble to form
there), and thus the discontinuity in the lift curves to occur. This can be cor-
rected by improving the leading edge shape. There is still improvement to be
had: the 20% Elliptic AirfoiP is a thicker airfoil (i.e. has a greater LE radius)
version of the current baseline blown ellipse airfoil, and it shows no sign of
separation, continuing on as CL reaches -8 or more. Thus great downforce
potential is confirmed with blowing (no increase in airfoil angle of attack is
necessary here) and this carries over into the heat transfer evaluations con-
ducted with coolant flowing through the blown radiators.
558 R.]. Gaeta, R.J. Englar, and G. Blaylock
Results for the conventional or Visteon radiator core indicated that a maxi-
mum coolant temperature drop of about 5 degrees Fahrenheit was achieved for
a flow rate of 5 gallons/min with a 64 mph freestream velocity. Figure 11
shows coolant temperature drop for the Visteon core as a function of blowing
coefficient, Cm, and coolant mass flow. Note that for the smaller coolant mass
onHrllinna l Radialor ore ]\'istcon]
\ '" = 64 mph
. . -.--.-.._• ........
• • ..,.. · ( fHtlilnl llu" Katr • I-t~ ~wlmm
·20
.:.:'"':.: :.!.: :..:.... .. • -.
t. ·30 ...\
~ .• -4.0 r----:::---..::
0
o:___~e
u
0
·50 0
0 0
·60
·7 0
(
.
Figure II . Effect of radiator temperature drop with respect to blowing coefficient.
kW so
6.0
...,.• ...... .
40
30
. .. ·:- ·-·
..........•...- .
·
_,_ -·-- --I
...
20 0 .... ""-_ ...
10 ~ ...... - 'V' ..rr--
('
Figure 12. Low coolant flow rate results of AHE; note effect of heat removal with
respect to blowing coefficient.
Wind-Tunnel Evaluation of an Aerodynamic Heat Exchanger 559
flows, larger temperature drops are realized. This is due to the longer exposure
of the coolant to the heat exchanger (longer residence times). It should be
stated that due to fabrication anomalies, some (42%) of the coolant flow tubes
were blocked off so the Visteon radiator was not flowing in a evenly distrib-
uted manner and it is quite likely that its performance was inhibited to some
degree. Figure 12 shows the corresponding heat removal as a function of Cm
and coolant mass flow for the various radiator configurations of the AHE. A
low coolant flow rate is shown. Note that the effect of the pneumatic lift
augmentation (the increasing Cm) is to increase the heat removal rate.
15a
14.a
-<>-Grnphi1e Rodi010< # I Jllieh D<•n,i1)J
13.a
- • · ( .raphite H~dhuor ;;~ z l' ert>cmin JJ'oro~il' )
12 a • • ,.. • \ l~rC"on "( ·onH·ntional" Racllucor
110
1aa
ga
,_._.. ___ .
aa
kW ~
.· ... . ,. . . -
?a
~ .:':"'········ ~-
v ~ ...
6.0
5.0
4a
3.0
2a
1a
a.a
a oas 0 1 a.15 02 0.25 03
c
•
Figure 13. Heat rejection from AHE for high coolant flow rate; note independence of heat
removal with respect to blowing coefficient for high density foam core.
cruising at highway speeds. Note that the high density graphite core performs
as well as the Visteon core which is somewhat surprising since it has little or no
airflow through the core. Due to the superior conductive performance of the
foam, almost all of the heat transfer takes place in the form of forced convec-
tion along the surface of the airfoil (both upper and lower) . This result was
intriguing and suggests that the heat removal can be varied by simply varying
the turbulence level of the flow over the wing surface. There are many meth-
ods (active and passive) which can accomplish this. This configuration (high
density graphite) also was the best performer in relation to the aerodynamics of
the device.
560 R.J. Gaeta, R.J. Englar, and G. Blaylock
Figure 14. AHE preparing for installation in GT Motorsports SAE Formula car.
Conclusions
Acknowledgements
The authors would like to acknowledge James Klett and April McMillan of ORNL for
being receptive to the concept of the AHE and providing funds and material for a part
of this work.
References
1. Englar, R. J. Circulation Control Pneumatic Aerodynamics: Blow Force and Moment
Augmentation and Modification; Past, Present, and Future. AlAA2000-2541 presented at
the A1AA Fluids 2000 Conference, Denver, Co., 19-22 June 2000
2. Burdges, Kenneth P. and Englar, Robert J. Vehicle Heat Exchanger System and Method for
a Vehicle that Augments and Modifies Aerodynamic Forces US Patent number 6179077,
January 30, 2001..
3. Englar, Robert J., Experimental Investigation of the High Velocity Coanda Wall Jet Applied
to Bluff Trailing Edge Circulation Control Airfoils, published as M.S. Thesis, University of
Maryland, Department of Aerospace Engineering Oune 1973).
Automated Driving ofTrucks and Buses:
Opportunities for Increasing Productivity and Safety
While Reducing Fuel Use and Emissions
Steven Shladover
Automation technology can expand the performance envelope for heavy vehicles by
eliminating the limitations imposed by driver performance. Accurate automatic
steering control makes it possible for the vehicles to operate safely, and over a full
speed range, within lanes only slightly wider than the vehicles themselves.
Accurate control of vehicle speed and spacing makes it possible for the
vehicles to operate closer together than they can today, without increasing the risk of
crashes. The sensor and control technologies can respond to disturbances more quickly
and accurately than drivers can, they are not vulnerable to fatigue or other
impairments, and they can incorporate continuous fault diagnostics as well. These
features enable the operation of heavy trucks or buses in electronically coupled 'virtual
trains', with significantly lower aerodynamic drag than if the vehicles were operated in
the conventional, uncoupled manner.
This presentation reviews the progress that has been made toward automated
driving of heavy vehicles within the past decade and shows results of some of the
experiments that have been conducted to demonstrate the performance of the vehicle
control systems.
Author Index
Page Nos.