Proceedings of FEDSM’03:

th
4 ASME_JSME Joint Fluids Engineering Conference
Honolulu, Hawaii, USA, July 6–10, 2003

FEDSM2003-45136

COMMERCIAL CFD CODE VALIDATION FOR SIMULATION OF HEAVY-VEHICLE EXTERNAL

AERODYNAMICS

W. David Pointer, Tanju Sofu, and David Weber

Argonne National Laboratory
Nuclear Engineering Division
9700 S Cass Avenue
Argonne, IL 60439
[email protected]

ABSTRACT equation low Reynolds number models in the near wall region,
The issue of energy economy in transportation has or transient RANS simulations may also be included depending
grown beyond traditional concerns over environment, safety on the results of the initial simulations.
and health to include new concerns over national and In the second phase, agreement between the
international security. In collaboration with the U.S. predictions of Star-CD simulations and the predictions from
Department of Energy Office of FreedomCAR and Vehicle models using PowerFLOW is considered. Since PowerFLOW
Technologies’ Working Group on Aerodynamic Drag of Heavy is a lattice-Boltzmann based code, the extensive modeling
Vehicles, Argonne National Laboratory is investigating the options available in a finite volume type code are not available
accuracy of aerodynamic drag predictions from commercial or necessary for simulation of aerodynamic drag. Standard
Computational Fluid Dynamics (CFD) Software. In this PowerFLOW modeling options are employed for comparisons
validation study, computational predictions from two with Star-CD.
commercial CFD codes, Star-CD [1] and PowerFLOW [2], will In the final phase of the study, computational
be compared with detailed velocity, pressure and force balance predictions of local velocities, local pressures and overall drag
data from experiments completed in the 7 ft. by 10 ft. wind coefficients will be compared with experimental results for four
tunnel at NASA Ames [3,4] using a Generic Conventional geometric configurations. The nominal configuration is a
Model (GCM) that is representative of typical current- representative model of a current-generation tractor-trailer
generation tractor-trailer geometries. developed by the working group in collaboration with
representatives from major truck manufacturers. Alternate
INTRODUCTION configurations investigate the addition of a low-boy device
The results presented herein are taken from the first under the length of the trailer, a full fairing between the cab and
phase of the study. In this phase, an evaluation of existing the trailer, and the combination of the fairing and low-boy
modeling capabilities provides a roadmap for the detailed device. All wind tunnel tests considered in the validation study
numerical simulations that will be completed for four validation were completed at a Reynolds number of 1.1 million. In each
cases in the second phase. In order to develop these guidelines, experiment, three-axis Particle Image Velocimetry (PIV) and
the repeatability of results using different computational mesh 490 pressure sensors, of which 18 are dynamic, are employed
structures and different modeling options within Star-CD is to capture the velocity and pressure fields around the vehicle.
evaluated. Star-CD, a finite volume code using a predictor- A standard aerodynamic force balance is employed to capture
corrector-type solution algorithm, has a wide variety of drag force data. All cases will be treated as “blind” validation
turbulence modeling options available as part of the standard studies with no prior knowledge of the details of the
commercial software release. These studies include experimental results.
investigations of standard steady-state k-epsilon models in The results of these studies will provide a validation of
conjunction with logarithmic wall functions. Two-layer RANS capabilities within commercial CFD software for form drag
models, which use k-epsilon models in bulk flow field and one- simulations, and a set of “best practice” guidelines will be

1 Copyright © 2003 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org on 11/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 developed for each software package considered. Using these to build a partially-unstructured, primarily-hexahedral
guidelines, a more rigorous validation will be completed in computational mesh. Portions of the mesh are then cut away
conjunction with PACCAR Technical Center considering real based upon a spatially expanded representation of the vehicle’s
truck geometry complete with mirrors, exhaust pipes, door surface definition, and a brick and prism cell layer is extruded
handles and decorative details. from the inner layer of the cut hexahedral mesh to the actual
surface of the vehicle. In the generation process, a bulk near-
GENERIC CONVENTIONAL MODEL vehicle cell size and a minimum cell size for local feature-
The Generic Conventional Model (GCM) is a based refinement regions are specified. In these studies, the
simplified representation of a conventional U. S. tractor-trailer effects of variations of these two parameters on predicted drag
truck. The model is 1/8th scale and can be configured in four coefficient are evaluated. In all cases considered in this phase
different geometric configurations as shown in Figure 1. The of the study, the extrusion layer consists of two layers of cells
standard truck configuration is used in these preliminary studies where the outer layer has a thickness of 1.0 mm and the inner
to establish guidelines for development of computational surface has a thickness of 0.5 mm. Future efforts will evaluate
meshes for use in the validation phase of the GCM studies and the effects of extrusion layer parameters on drag coefficient
in future studies in collaboration with PACCAR Technical predictions.
Center. In all evaluations, the GCM is centered at zero yaw When using unstructured meshes, the evaluation of the
angle on the floor of a wind tunnel test section that is 2.1334 m suitability of a particular computational mesh may not be as
(7 ft.) tall by 3.048 m (10 ft.) wide. Since the GCM is straightforward as in block-structured mesh problems where
approximately 2.5 m long, a total test section length of 10.0 m each cell in the mesh is uniformly divided into some number of
is assumed with one truck length included upstream of the new cells. In unstructured systems generated by feature-size
model and two truck lengths included downstream of the based meshing tools, the bulk flow resolution and the near-wall
model. Based upon a Reynolds number of 1.1 x 106, a uniform resolution may be quite different, and thus require separate
velocity of 33 m/s is enforced at the inlet boundary and a consideration. The initial studies of phase one seek to establish
uniform pressure condition is applied at the outlet boundary. some preliminary guidelines for setting the near-vehicle cell
The surface of the standard configuration GCM is defined using size and minimum cell size parameters for the evaluation of
approximately 500,000 triangular surface elements based upon tractor-trailer truck geometries.
CAD data representations taken from optical scans of the actual
model. Bulk Flow Resolution
Five unique computational meshes were generated
COMPUTATIONAL MESH SUITABILITY based upon the standard GCM configuration in order to
The Star-CD external aerodynamics analysis tool ES-Aero evaluate the effects of the near vehicle cell size parameter on
uses a local-feature-size based automatic meshing system the prediction of the drag coefficient. Near-vehicle cell sizes of
16.0, 12.0, 10.0, 8.0 and 6.0 mm were considered. In each
case, the minimum cell size resulting from local feature-based
refinements is 12.5 percent of the near vehicle cell size and an
additional restriction is set so that a minimum of 16 elements
are required for the definition of a circle. In order to ensure
that the quality of the vehicle surface is maintained, the cell
Standard truck layer immediately adjacent to the surface is refined to 25
percent of the original size prior to trimming. These settings
result in the mesh characteristics shown in Table 1. A sample
mesh structure, taken from the case with the near-vehicle cell
size of 8.0 mm, is shown in Figure 2.

Low boy trailer Table 1. Summary of computational mesh characteristics

for evaluation of bulk cell size effects.
Total Number of
Near Minimum
Number of Volume
Vehicle Cell Cell Size
Volume Elements
Size (mm) (mm)
Faired truck Elements on Surface
16.0 2.0 1012338 73574
12.0 1.5 1737085 126119
10.0 1.25 2345640 175105
8.0 1.0 3282426 266666
Faired truck with low boy trailer 6.0 0.75 5695622 400382

Figure 1. Generic conventional model configurations

2 Copyright © 2003 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org on 11/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 For each case, 2000 iterations were calculated using each of the two cases in Figure 5. As in the first study,
Star-CD’s standard conjugate gradient solver and the PISO improved resolution leads to improved separation and
predictor-corrector algorithm. In these simulations, the reattachment predictions, which again improves drag
standard high Reynolds number two-equation turbulence model coefficient predictions. The further refinement in the near wall
and a logarithmic wall function are employed for prediction of region of the model using a near vehicle cell size of 8.0 mm
turbulent kinetic energy and dissipation. Convergence criteria leads to improvements beyond the case using a near vehicle cell
were set to insure that all cases would reach 2000 iterations size of 6.0 mm, indicating that the refinement of the near wall
before stopping. However all residuals reach the normal region is possibly more influential on the drag coefficient
stopping criteria of 10-3 well before the 2000th iteration and are prediction than the refinement of the bulk flow region around
no larger than 2.0 x 10-4 at the 2000th iteration. In addition to the vehicle.
the standard flow variable residual monitoring, the drag
coefficient of the vehicle is monitored as the solution develops
to insure that the drag coefficient reaches a converged value as
well. Drag coefficient convergence data is shown in Figure 3.
Predicted drag coefficients from each of the five cases
are compared with experimental data from wind tunnel tests in
Table 2. Total CPU time and the number of CPU’s used in
each case are also shown. While there is a trend of
improvement with reduction in near-vehicle cell size, the
effects that lead to non-linearity in the trend are not
immediately clear. Further reductions in near vehicle cell size
are not possible without violating the memory limits of the Figure 2. Near-vehicle region of a typical computational
32-bit system used in the development of these computational mesh construction employed in these studies.
meshes. Further increases in the near vehicle cell size will
result in cells that are two large to capture significant vehicle 0.7

features such as the rear wheel axels. Detailed data from each
0.6
of the five cases are shown in Figure 4, where pressure
distributions along the centerline of the vehicle are illustrated. 0.5
These results indicate that increasing resolution leads to
Drag Coefficient

improved prediction of separation and reattachment, which in 0.4

turn improve drag coefficient predictions.
0.3

Near Wall Resolution Near Vehicle Cell Size = 16.0 mm

In an effort to understand the significance of near wall 0.2
Near Vehicle Cell Size = 12.0 mm
resolution versus bulk flow resolution, a new computational Near Vehicle Cell Size = 10.0 mm
0.1
mesh was generated based upon the case in which the near Near Vehicle Cell Size = 8.0 mm

vehicle cell size parameter was set to 8.0 mm. In the coarser Near Vehicle Cell Size = 6.0 mm
0
mesh, the minimum cell size was set as 12.5 percent of the near 0 200 400 600 800 1000 1200 1400 1600 1800 2000
wall cell size. All cells adjacent to the vehicle wall were Iteration

refined to 25 percent of the near vehicle cell size prior to the

cutting of the hexahedral mesh. Further refinements are Figure 3. Drag coefficient convergence curves from
completed as required by local features. For example, an investigation of bulk flow region resolution effects on drag
additional restriction required a minimum of 16 volume coefficient prediction.
elements to define the boundary of a circle. In the modified
case, the minimum cell size is reduced to 3.125 percent of the Table 2. Summary of results from evaluation of effects of
near vehicle cell size, and cells adjacent to the vehicle surface bulk resolution on drag coefficient prediction.
are refined to the same level as in the coarser case. Additional
restrictions are activated in the modified case to require that at Near Number Total
Predicted Percent
least 2 volume elements be placed between any two surfaces. Vehicle of CPU
Drag Error in
The total number of volume elements is increased from Cell Size CPU’s Time
Coefficient Prediction
3,282,426 to 4,264,232, and the number of volume elements on (mm) Used (hours)
the surface is increased from 266,666 to 284,650. Experiment -- -- 0.3983 --
As in the previous studies, 2000 iterations are completed,
16.0 8 77.2 0.4491 12.0
and the convergence of the drag coefficient is monitored. The
further local refinement that is based on surface features results 12.0 8 148.8 0.4437 10.8
in a reduction of the predicted drag coefficient from 0.4154 to 10.0 8 184.0 0.4174 4.67
0.4002. Thus, the difference between the experimental
8.0 16 317.5 0.4154 4.19
measurement and the numerical prediction is reduced from 4.19
percent to 0.47 percent. Centerline pressure data are shown for 6.0 24 437.5 0.4033 1.25

3 Copyright © 2003 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org on 11/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 (a) (b) (c)

(d) (e)

Figure 4. Pressure distribution along the top (blue line) and bottom (red line) centerlines of the vehicle. Distributions are
shown for near vehicle cell sizes of (a) 16.0 mm, (b) 12.0 mm, (c) 10.0 mm, (d) 8.0 mm, and (e) 6.0 mm.

4 Copyright © 2003 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org on 11/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 flow distortion of the dissipation. All simulations use the
refined mesh from the previous study, with a near vehicle cell
size of 8.0 mm and a minimum cell size of 3.125 percent of the
near vehicle cell size. As in previous studies, 2000 iterations
are completed for each case and the convergence of the drag
coefficient is monitored. Results of the study are summarized
in Table 3. More detailed pressure data is shown in figure 5,
where the similarity of the k-epsilon and SST models is
apparent. The pressure profiles from the k-epsilon and RNG
simulations show good agreement along the top of the vehicle.
However, the RNG model appears to exaggerate the separation
of flow under the trailer in comparison with the k-epsilon
model.

CONCLUSION
These studies have been completed to provide preliminary
(a) guidelines for the prediction of drag coefficients for tractor-
trailer trucks using standard features of the commercial CFD
software package Star-CD. These guidelines will be employed
in future studies using this software package for the evaluation
of more advanced turbulence modeling capabilities.
Furthermore, these guidelines will help define future studies
using alternate software packages.
The local refinement of the computational mesh based
upon local feature sizes is shown to enable the prediction of
drag coefficients for a generalized tractor-trailer geometry
using simple two-equation turbulence models. Based upon
these studies, one may conclude that the infinite refinement of
the mesh may lead to infinite improvements in the solution.
However, it is important to ensure that the range of
applicability of the turbulence model selected is not violated
through excessive refinement of the mesh. Indeed, these
(b) studies indicate that further refinement of the local feature-
Figure 5. Pressure distribution along the top (blue line) and based refinement regions may lead to more significant
bottom (red line) centerlines of the vehicle. In case (a) the improvements than further refinement of the entire
near vehicle cell size is 8.0 mm and the minimum cell size computational mesh.
resulting from localized refinement is 12.5 percent of the The evaluation of two-equation turbulence models indicate
near vehicle cell size. In case (b) the near vehicle cell size is that good results can be obtained with these models when a
8.0 mm and the minimum cell size resulting from localized well-constructed computational mesh is used in the evaluation.
refinement is 3.125 percent of the near vehicle cell size. Indeed, simulations using the standard high-Reynolds number
k-epsilon model predict the drag coefficient within 0.47 percent
of the measured value from wind tunnel experiments. Slight
TURBULENCE MODEL SELECTION improvements in the drag coefficient prediction are observed
In order to evaluate the sensitivity of the solution to when the more complex k-ω SST model is employed.
the two-equation turbulence model employed, an additional However, the use of the RNG formulation of the two-equation
study was completed using three different turbulence models turbulence model leads to under-prediction of the drag
that are included as standard options in the Star-CD software coefficient as result of the exaggerated separation regions on
package: the underside of the trailer.
(1) the standard high-Reynolds number k-epsilon model Table 3. Results of the evaluation of two-equation
with logarithmic wall function turbulence models for prediction of tractor-trailer truck
(2) the Menter k-omega SST model [5], and drag coefficients.
(3) the renormalization group (RNG) formulation of the k- Turbulence Predicted Drag Percent Error in
epsilon model [6]. Model Coefficient Prediction
The standard k-epsilon model and the SST model are identical Experiment 0.3983 --
in the far field, but the SST model incorporates additional detail High-Reynolds
in the near wall region. The SST model should be more Number k- 0.4002 0.47
sensitive to separation, but the two should show reasonable epsilon Model
agreement. The RNG model is similar to the standard k-epsilon k-ω SST model 0.4001 0.45
model, but includes an additional term to account for the mean RNG model 0.3912 -1.80

5 Copyright © 2003 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org on 11/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 (a) (b) (c)
Figure 6. Pressure distribution along centerline of top surface (blue line) and bottom surface (red line) of vehicle for
simulations using (a) the standard high-Reynolds number k-epsilon, (b) the k-ω SST, and (c) the RNG turbulence models.

ACKNOWLEDGMENTS
This work was completed under the auspices of the U.S. Conference on The Aerodynamics of Heavy Vehicles:
Department of Energy Office of FreedomCAR and Vehicle Trucks, Buses, and Trains, Monterey, CA, Dec 2-6, 2002.
Technologies. The submitted manuscript has been created by 4. J. T. Heineck, Stephen Walker, Dale Satran, “The
the University of Chicago as Operator of Argonne National Measurement of Wake and Gap Flows of a 1/8th Scale
Laboratory (“Argonne”) under Contract No. W-31-109-ENG- Generic Truck Using Three-Component Particle Image
38 with the U.S. Department of Energy. The U.S. Government Velocimetry,” United Engineering Foundation Conference
retains for itself, and others acting on its behalf, a paid-up, on The Aerodynamics of Heavy Vehicles: Trucks, Buses,
nonexclusive, irrevocable worldwide license in said article to and Trains, Monterey, CA, Dec 2-6, 2002.
reproduce, prepare derivative works, distribute copies to the 5. Menter, F.R.,”Zonal Two Equation k-ω Turbulence Models
public, and perform publicly and display publicly, by or on for Aerodynamic Flows”in 24th Fluid Dynamics
behalf of the Government. Conference (Orlando), AIAA paper 93-2906, July 1993.
6. Yakhot, V., Orszag, S.A., Thangam, S., Gatski, T.B., and
REFERENCES Speziale, C.G. ‘Development of turbulence models for shear
1. Star-CD, version 3.150A, CD-Adapco Group, Melville, NY. flows by a double expansion technique’, Phys. Fluids, A4,
2. PowerFLOW, version 3.4, Exa Corporation, Lexington, No. 7, pp. 1510–1520, 1992
MA.
3. Dale Satran, “ An Experimental Study of the Generic
Conventional Model (GCM) in the NASA Ames 7-by-10-
Foot Wind Tunnel,” United Engineering Foundation

6 Copyright © 2003 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org on 11/25/2018 Terms of Use: http://www.asme.org/about-asme/terms-of-use