Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

JOURNAL OF AIRCRAFT

Numerical Simulations of the F-16XL at Flight-Test Conditions


Using Delayed Detached-Eddy Simulation

Andrew J. Lofthouse∗ and Russell M. Cummings†


U.S. Air Force Academy, Colorado 80840
DOI: 10.2514/1.C034045
The F-16XL flight-test data have been used for many years in the Cranked-Arrow Wing Aerodynamics Program
International series as a challenging test case for numerical prediction of military aircraft at full-scale flight
conditions. The complexity of the configuration and the variety of flight-test conditions available challenge most
computational fluid dynamics codes and turbulence models. This study documents the use of the Kestrel flow solver
with its solution-based mesh refinement capability and delayed detached-eddy simulation turbulence models for F-
16XL flight condition 25, which is a subsonic high-angle-of-attack flight condition that has proven particularly
challenging for computational researchers. The spectral content of the pressure data and the integrated lift, drag, and
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

pitching moment coefficients are also presented. The more advanced turbulence models, coupled with the more
refined meshes, show very good agreement with flight-test data and reasonable resolution of the unsteady frequency
content.

Nomenclature and transonic conditions and to characterize the stability and control
CD = drag coefficient; positive downwind of the aircraft. The original intent of a flight, wind-tunnel, and CFD
CL = lift coefficient; positive upward correlation experiment would be maintained, albeit for the baseline
Cm = pitch moment coefficient; positive nose up F-16XL configuration only” [1]. CAWAP provided the computa-
Cp = pressure coefficient tional fluid dynamics (CFD) community with an excellent database
for complex aerodynamic validation and evaluation purposes [4].
The Cranked-Arrow Wing Aerodynamics Project International
I. Introduction (CAWAPI), which was initiated by NASA as a follow-on project to the
CAWAP, allowed for a larger international community of researchers
T HE F-16XL was developed in the 1980s by General Dynamics
as a single-place fighter prototype aircraft. Starting with the
F-16A configuration, the F-16XL design included a stretched
to participate in predicting the aerodynamics of the F-16XL.

fuselage and a cranked-arrow wing (as well as modified systems). Along with the Vortex Flow Experiment 2 (VFE-2),
The cranked-arrow wing was a joint design between General CAWAPI was incorporated under the NATO science and
Dynamics and NASA Langley Research Center that provided technology organization Task Group applied vehicle
improved supersonic performance while maintaining the transonic technology-113. The objective of the CAWAPI facet was
performance of the F-16A. Figure 1 shows the cranked-arrow wing to allow a comprehensive validation and evaluation of CFD
design with a leading-edge sweep angle of 70 deg inboard and 50 deg methods against the CAWAP flight database. A number of
outboard of the crank. researchers simulated the flowfield of the F-16XL at a
variety of flight conditions using different numerical
At the juncture of the wing leading edge with the fuselage, an approaches, including structured, block, and unstructured
S-blend curve was placed in the leading edge to alleviate a grids, as well as various turbulence models and numerical
pitch instability that was found to occur at high angles of algorithms. This type of full-scale aircraft configuration
attack in wind-tunnel tests. All flight-test data were collected provides many challenges to state-of-the-art CFD flow
with the air dams (upper-surface fences mounted near the prediction, including the ability to accurately predict
wing leading-edge crank) and wing-tip missiles installed [1]. unsteady flowfields at flight Reynolds numbers [4].

The F-16XL program produced two prototype aircraft that were The success of CAWAPI led to follow-on projects to evaluate
flight tested from 1989 through 1997 at the NASA Dryden Flight improvements to the CFD predictions using Reynolds-averaged
Research Center. Navier–Stokes (RANS) turbulence models (known as CAWAPI-2)
The F-16XL program was cancelled during phase 1 flight testing and with hybrid RANS/large-eddy simulation (LES) turbulence
due to funding limitations, but funding was provided to complete a models (known as CAWAPI-3). The results presented here are part of
modified flight-test program, which became the Cranked-Arrow the CAWAPI-3 program, where the delayed detached-eddy
Wing Aerodynamics Project (CAWAP) program [2,3]. “The revised simulation (DDES) hybrid turbulence model has been used on
objectives were to document upper surface flow physics at high-lift improved computational meshes in order to improve the flowfield
predictions of the F-16XL.
A delta-wing configuration of an aircraft such as the F-16XL,
Presented as Paper 2015-2875 at the 33rd AIAA Applied Aerodynamics
Conference, Dallas, TX, 22–26 June 2015; received 28 May 2016; revision shown in Figs. 1 and 2, behaves significantly differently from a
received 28 March 2017; accepted for publication 7 June 2017; published traditional aircraft configuration with relatively straight wings. At
online 5 July 2017. This material is declared a work of the U.S. Government high angles of attack, the boundary layer separates at the leading edge
and is not subject to copyright protection in the United States. All requests and curls inward to form cores of high vorticity. The size and strength
for copying and permission to reprint should be submitted to CCC at of the vortex cores are dominant features of the flow. The low
www.copyright.com; employ the ISSN 1533-3868 (print) or 0021-8669 pressure of the vortex cores assists in maintaining lift, even at flight
(online) to initiate your request. See also AIAA Rights and Permissions
conditions where traditional aircraft configurations would stall [5].
www.aiaa.org/randp.
*Director, High Performance Computing Research Center. Senior Member The vortex cores are highly unsteady as they travel backward over
AIAA. the wing, eventually breaking down or bursting. These types of
† aerodynamic characteristics are difficult to predict with the
Professor of Aeronautics, Department of Aeronautics. Associate Fellow
AIAA. computational aerodynamic codes available today.
Article in Advance / 1
2 Article in Advance / LOFTHOUSE AND CUMMINGS

II. Simulation Procedure


A. Kestrel Flow Solver
Kestrel is a complete virtual flight-test simulation software suite
developed as part of the U.S. Department of Defense (DOD)
Computational Research and Engineering Acquisition Tools and
Environments (CREATE) Program. CREATE was established as a
12 year program in fiscal year 2008 and is managed by the DOD High
Performance Computing Modernization Program. The goal is to
enable improvements in DOD acquisition programs through the use
of scalable, multidisciplinary physics-based computational engineer-
ing software products for use on DOD high-performance computing
resources [13]. CREATE consists of three computational engineering
tool sets for the design of air vehicles, ships, and radio frequency
antennas. The fixed-wing analysis code Kestrel is part of the Air
Fig. 1 F-16XL configuration showing stretched fuselage and cranked-
arrow wing design.
Vehicles Project (known as CREATE-AV) and is a modularized,
multidisciplinary, virtual aircraft simulation tool incorporating
aerodynamics, structural dynamics, kinematics, and kinetics [14].
The geometric features of interest in this study, in addition to the The flow solver component of Kestrel (known as kCFD) is a finite
shape of the main wing, are the air dam and actuator pod toward the
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

volume cell-centered two- or three-dimensional unstructured flow


outer portion of the wingspan, as well as the dummy missile and solver. The unstructured mesh may be composed of arbitrarily shaped
missile rail. These features were present on the aircraft during the cells, with the most common being triangles and quadrilaterals in two
flight test and are included in the computational geometry treated dimensions and tetrahedra, prisms, pyramids, and hexahedra in three
here. Note also that pressure data were taken at several butt line (BL) dimensions. Automatic domain decomposition allows the simulation
locations (where the butt line is measured from the centerline of the to be run on multiple processors in parallel. The method of lines is
aircraft out toward the right wingtip) and fuselage station (FS) employed and allows the temporal and spatial integration schemes to
locations (where the fuselage station locations are measured in differ [15]. The spatial residual is computed with a typical Godunov
inches, back from the nose of the aircraft). scheme [16] with second-order accuracy achieved by the use of linear
This study is a follow-on one to the CAWAPI program and gradients within each cell. Several exact and approximate Riemann
documents the use of the Kestrel flow solver to simulate the F-16XL schemes are implemented to compute the fluxes, with the default
during flight condition 25 (FC25), which was a low-speed high- being a variation of the Harten-Lax-van-Leer-Einfeldt++ scheme
angle-of-attack condition that exhibited some significant challenges [17]. A subiterative point-implicit scheme solved with a symmetric
for numerical simulations due to the flow unsteadiness caused by successive overrelaxation technique provides up to second-order
vortex interactions and the onset of vortex breakdown. To more fully
temporal accuracy [18].
understand the type of unsteadiness and the specific details of this
The Kestrel solver includes several turbulence models, including
complex flowfield, two specific capabilities of Kestrel were
the Spalart–Allmaras (SA) one-equation RANS model [19] and the
exercised: the offline solution-based adaptive mesh refinement, and a
Spalart–Allmaras with rotation correction (SARC) model [20]. The
number of different turbulence models.
solver also includes the delayed detached-eddy simulation hybrid
The turbulence models include Reynolds-averaged Navier–Stokes
RANS/LES turbulence model to resolve the unsteady turbulent
and delayed detached-eddy simulation. RANS models resolve only
fluctuations in separated regions of the flow [7,8]. The turbulence
the time-averaged flow quantities and none of the turbulent
models used in this study included the SA, SARC, SA/DDES, and
fluctuations. They are typically suitable for attached boundary layers,
SARC/DDES models.
and even steady separated flows, with minimal computational cost;
but, they tend to fail to provide accurate results for flows with large
regions of unsteady separated flows [6]. DDES combines RANS B. Mesh Refinement
models in the attached boundary layer with a large-eddy simulation The simulations documented here used several meshes that were
model in the separated regions. LES resolves the larger turbulence generated from a baseline mesh using the offline adaptive mesh
scales numerically but models the smaller, more homogeneous refinement capabilities of Kestrel and the associated postprocessing
scales. LES is much more computationally expensive than RANS tools. The baseline mesh, designated A1, was the original mesh used
models, especially in attached boundary layers [6]. Thus, the during the CAWAPI program [4]. This baseline mesh was a fairly
combination of RANS in attached boundary layers and the LES in coarse mesh of about 12 million cells, with about 1.4 million prisms
separated regions provides a capability to accurately resolve large- near the surface to resolve boundary-layer gradients. Only half of the
scale unsteady, separated flows with a reduced computational cost. full-span geometry was included, with the centerline being a
Nevertheless, hybrid RANS/LES models such as DDES require more symmetry plane. The baseline mesh was refined by subdividing all
refined meshes than RANS-only models in order to resolve the large- cells such that a tetrahedron was split into eight tetrahedral cells and a
scale unsteady turbulent fluctuations. The application of these prism was split into four prisms. This mesh, designated A2, contained
advanced turbulence models requires a balance between minimizing almost 90 million cells, with 5.8 million prisms.
computational cost while maintaining the required level of accuracy. Further meshes were generated from preliminary solutions run on
This balance may be achieved by using solution-based adaptive mesh the baseline mesh during which a refinement variable and tracking
refinement. threshold were employed. Cells in which the solution refinement
Detached-eddy simulation (DES) was first proposed in 1997 [7], variable exceeded the tracking threshold during the simulation were
whereas DDES was proposed in 2006 as an extension to the original flagged and then subdivided (as noted previously, eight tetrahedra
DES formulation that addressed a shortcoming in the sensitivity of resulted from a single tetrahedron that was flagged and four prisms
the model to grid spacing near the wall [8]. Since their development, resulted from a single prism that was flagged).
both models have been used extensively for a variety of unsteady, The computational cost of a DDES solution increases over a
highly separated flows. Some of these studies have included the RANS solution partially because of the increased mesh refinement
F/A-18 to resolve unsteady tail buffet [9] and abrupt wing stall [10]; required to resolve the unsteady turbulent fluctuations in the
a generic fighter configuration at high angle of attack [11]; and the separated flow regions. Each mesh refinement variable will flag
unsteady vortex breakdown of a generic, 70 deg delta wing [12]. different numbers of cells for refinement, depending on the tracking
It is expected that the improvements in turbulence modeling and threshold. The refinement variables used here were selected based on
adaptive mesh refinement that have taken place since [4] was how likely mesh refinement based on those variables would increase
published will increase our understanding of the flowfield for FC25. mesh resolution in the required areas while minimizing increased cell
Article in Advance / LOFTHOUSE AND CUMMINGS 3
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 2 Drawings of the F-16XL aircraft geometry and pressure data locations.

density where it was not needed. The refinement variables used in this are problem specific. The scaled Q criterion was developed to
study included vorticity magnitude, Q criterion, and scaled Q criterion. provide a variable for visualization and refinement tracking that
Vorticity magnitude is a measure of the rotation and shear of a fluid should be universal.
element and is given mathematically as the curl of the velocity vector Mesh B2 was generated by flagging all cells in which the Q
[6]. The values of vorticity magnitude are typically large in regions of criterion exceeded 0.0; mesh C2 was generated using a scaled
high flow rotation, such as separation vortices, as well as regions of Q-criterion threshold value of 0.1; and mesh D2 used a vorticity
high shear, such as boundary layers. The Q criterion is defined as the magnitude of 2001∕s as the refinement value. Each of these values,
rotation rate minus the shearing rate [6]. Values of the Q criterion are except for the scaled Q criterion, may be different for a different flight
typically large in regions of high flow rotation, such as separation condition or application. Therefore, the analyst would be required to
vortices, but are not typically high in regions of high shear, such as generate visualizations in order to determine which specific values
boundary layers. The values of the vorticity magnitude and Q would provide refinement in the regions required. As noted
criterion that are appropriate for a mesh refinement tracking threshold previously, the scaled Q-criterion value is expected to be universal.
4 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 3 Preliminary solutions on mesh A1 used for refinement of the mesh.

Figure 3 shows a sampling of the preliminary solutions on mesh A1 was refined with each method; especially note that the scaled
that were used to flag cells for refinement (note that the Q-criterion Q-criterion (mesh C2) and vorticity magnitude (mesh D2) tracking
isosurface is at a value of 1 × 10−5 instead of 0.0). variables result in refinement that seems to better track the vortices
Details of the resulting meshes are summarized in Table 1. The being shed from the wing.
mesh with the most cells is mesh A2, which refined all cells in the The viscous spacing of each mesh was unchanged from the
original mesh. The next largest was mesh B2, which used the Q original mesh, which was selected to maintain a y less than or equal
criterion of zero as the refinement tracking variable; the size to one for the Reynolds number considered here. A sample solution
difference between mesh A2 and mesh B2 is just over 3M cells. using the SA/DDES turbulence model on mesh B2 is shown in Fig. 9;
Note that three of the four refined meshes (all but mesh C2) the final y value is mostly less than one.
resulted in 5.8M prisms, which is about four times the initial number
of 1.4M, indicating that the entire surface mesh and prism layers C. Reference Conditions
grown from the surface mesh were flagged for refinement. The main The flight conditions for FC25 are low-speed and high angle of
difference between these meshes is the number of cells flagged for attack: the angle of attack is 19.84 deg; altitude is 10,000 ft; Mach
refinement in the flow separation area; more cells were flagged for number is 0.242; and Reynolds number is 32.2 × 106 . The aircraft
refinement using the Q-criterion tracking than with the other two forces and moments are calculated with the following reference
methods. The scaled Q criterion does not flag as many cells in the quantities: reference area is 8.64 × 104 in:2 ; reference length is
prism layer as the other meshes, although some are flagged. 296.4 in.; the moment reference point is located at 326.86, 0.0,
The meshes are visualized in Figs. 4–8, including the surface mesh, 76.06 in.; and the moment reference lengths are 194.42, 296.4, and
the symmetry plane, a slice in the X direction at 350 in. aft of the nose, 194.42 in.
and a slice in the Y direction 75 in. toward the right wing from the A time-step sensitivity study was conducted on mesh B2 to
symmetry plane. Note specifically the areas where the baseline mesh determine the proper time step. The initial time step of 5.0 × 10−4 s

Table 1 Mesh details for F-16XL


Mesh name Refinement variable Threshold value Prisms Tetrahedra Total cells
A1 None (baseline) None 1.4M 10.5M 11.9M
A2 All refined All 5.8M 83.9M 89.7M
B2 Q criterion 0.0 5.8M 80.7M 86.5M
C2 Scaled Q criterion 0.1 2.2M 54.6M 56.8M
D2 Vorticity magnitude 200 1∕s 5.8M 63.0M 68.8M
Article in Advance / LOFTHOUSE AND CUMMINGS 5
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 4 Mesh A1 (baseline).

Fig. 5 Mesh A2: baseline mesh with each cell divided.


6 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 6 Mesh B2: baseline mesh refined in cells where Q criterion exceeded zero.

Fig. 7 Mesh C2: baseline mesh refined in cells where scaled Q criterion exceeded 0.1.
Article in Advance / LOFTHOUSE AND CUMMINGS 7
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 8 Mesh D2: baseline mesh refined in cells where vorticity magnitude exceeded 200 1∕s.

Fig. 9 Contours of y from the SA/DDES solution on mesh B2.

was shown previously to provide good temporal resolution [4]. the surface and above the surface, as well as integrated forces and
Simulations of two additional time steps of 2.5 × 10−4 s and 1.25 × moments, were recorded at every time step. The solution was also
10−4 s showed very little change in the overall forces and moments, averaged over the entire 8000 final time steps.
as shown in Fig. 10. The average lift coefficient varied by about 1%, Standard viscous no-slip wall boundary conditions were used for
and the average drag coefficient varied by about 2% in the results the solid surfaces: a symmetry boundary condition for the symmetry
using the smaller time steps. The average pitching moment plane, and modified Riemann invariants for the far field (modified
coefficient varied by about 6%, although the absolute magnitude of Riemann invariants use fixed inflow values for supersonic flow and
the changes was one order of magnitude smaller than the changes in Riemann invariants for subsonic inflow).
the lift and drag coefficient (the larger percent change is due to the
much lower magnitude of the pitching moment coefficient). Because
of the small differences in average force and moment coefficients, the III. Results
initial time step of 5.0 × 10−4 s was deemed sufficient for this study. This section summarizes the results obtained for the simulations of
The general simulation procedure is as follows. The solver is set to the F-16XL geometry at flight condition 25, using the meshes and
second-order spatial and temporal accuracy. All simulations used turbulence models discussed previously. First, the solutions will be
three Newton subiterations to improve the temporal accuracy. Startup compared qualitatively using volume visualizations. Then, the
transients were flushed with 2000 startup time steps (1 s of simulation computational results will be compared to full-scale flight-test data
time); after which, 8000 time steps (4 s of simulation time) of using plots of pressure coefficient values along slices that correspond
unsteady data were taken. Unsteady data at a series of tap points on to the BL and FS pressure data locations of the full-scale aircraft in
8 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 10 Time-step sensitivity study results.

Fig. 2. Finally, a comparison of the integrated loads (lift, drag, and capture as much unsteadiness as the largest meshes: mesh A2 and
pitching moment coefficients) will be made. mesh B2 (all-refined and refinement based on Q criterion). This is
particularly apparent in the top view of Fig. 12. Between the two
A. Volume Visualizations largest meshes, it appears that the solution on mesh B2 resolves
Snapshots of the unsteady solution are visualized using several slightly more unsteadiness than the solution on mesh A2.
different views in Figs. 11–16. Visualizations of isosurfaces of the Q The difference between solutions with different turbulence models
criterion at 1.0 × 10−3 , colored by Cp, on the four refined meshes is more striking. It is not surprising that the two RANS solutions
are found in Figs. 11–13. Using mesh B2, the same comparison is using the SA and SARC turbulence models do not resolve as much
made between the solutions using different turbulence models in unsteadiness as the DDES solutions, although it should be noted that
Figs. 14–16. they both resolve the large-scale vortex motions and that the SARC
There are several flow features of note that are sources of the flow solution exhibits slightly greater resolution of the vortices than
unsteadiness. The first is a strong vortex shed from the main leading- the SA solution. The differences between the SA∕DDES and
edge apex of the wing that travels aft and leads to a large area of SARC∕DDES solutions are not as apparent, although the top view in
unstable and unsteady flow above the air dam and actuator pod. The Fig. 15 seems to indicate that the SA∕DDES solution resolves
second is a vortex that is formed from the crossflow above the wing slightly more unsteadiness than the SARC/DDES solution. Note that
over the air dam. This vortex also travels downstream and interacts the improved delayed detached-eddy simulation was not available at
with the leading-edge apex vortex. The third is a vortex shed from the the time the earlier work was accomplished [4].
crank in the wing, just outside of the air dam. This vortex travels aft The final volume visualizations to be discussed are presented in
and outboard, and it interacts with the vortex system and wake from Fig. 17. These are snapshots of the instantaneous solution visualized by
the missile and wingtip rail. Note that the flowfield behind the air dam isosurfaces of the Q criterion at three different levels [1.0 × 10−3 (as
is highly unsteady when compared with the relatively steady apex before), 1.0 × 10−4 , and 1.0 × 10−5 ], as well as an isosurface of the
and crank vortices. vorticity magnitude at 200 1∕s. The high level of unsteadiness in the
Although all of the simulations captured the gross effects of these flowfield above the aft portion of the wing is apparent. Thus, it appears
vortex systems, the degree of unsteadiness captured is dependent on that the apex and crank vortices break down over the aft portion of the
the mesh refinement and turbulence model used. The smallest wing. Note, though, that although this solution using SARC∕DDES on
meshes, including mesh C2 and mesh D2 (refinement tracking mesh B2 does resolve the unsteadiness in the immediate vicinity of the
variables of scaled Q criterion and vorticity magnitude) did not surface mesh, the amount of unsteadiness behind the aircraft is not well
Article in Advance / LOFTHOUSE AND CUMMINGS 9
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 11 Q criterion  1.0 × 10−3 . SA∕DDES turbulence model on various meshes: isoview.

Fig. 12 Q criterion  1.0 × 10−3 . SA∕DDES turbulence model on various meshes: top view.

resolved as compared to solutions using an offbody Cartesian mesh 1. Fuselage Station 300
capability of Kestrel, as shown in other work [15]. FS300 is a slice in the X direction that is 300 in. from the nose of
the aircraft. Figure 18 shows a slice of the average solution using
B. Tap Data Comparison SARC∕DDES on mesh B2, as well as comparisons of the
A quantitative comparison of the pressure coefficient data obtained computational solutions using all meshes and turbulence models.
from the computational simulations will now be made with flight-test Finally, a more detailed comparison of the amount of unsteadiness
data of the full-scale F-16XL. Slices in the X direction (fuselage predicted by the computations is obtained by plotting the average Cp ,
stations) and Y direction (butt lines) are compared and discussed in the standard deviation (or rms), and the minima and maxima of the
each subsection that follows. computed solution using SARC∕DDES on mesh B2.
10 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 13 Q criterion  1.0 × 10−3 . SA∕DDES turbulence model on various meshes: side view.

Fig. 14 Q criterion  1.0 × 10−3 . Various turbulence models on mesh B2: isoview.

Pressure coefficient data are shown with respect to the location the flight-test data, it generally falls within the standard deviation of
along the cutline, in the spanwise direction, and normalized by the the unsteady data.
half-span distance of b∕2. Data are shown from the centerline
(2y∕b  0) to the edge of the wing (2y∕b  1). 2. Fuselage Station 337.5
This location is characterized by a strong apex vortex, with some The computed solution and flight-test data for FS337.5 is seen in
smaller vortex structures toward the outer wingspan. The computed Fig. 19. The Cp slice shows that the main apex vortex is still present,
solutions do not exhibit a large amount of disagreement, and they with a smaller vortical structure coming from the outboard edge
agree fairly well with the experimental data provided. There are some interacting with it. The majority of the computational solutions are
slight differences between the solution on mesh A1 (baseline mesh) in good agreement with each other, with the exception of SA/DDES
and the refined meshes at the outer location where the computed on the baseline mesh, mesh A1, and the SA turbulence model on
solution overpredicts the suction peak of the smaller vortices. mesh B2. Again, there is some significant unsteadiness in the
It is evident from the minima, maxima, and rms of the computed pressure data, as exhibited by the minima, maxima, and rms; but,
Cp data that there is significant unsteadiness at this location; although the experimental data fall very closely to the rms limits of the
the average of the computed Cp does not always agree perfectly with computational solution.
Article in Advance / LOFTHOUSE AND CUMMINGS 11
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 15 Q criterion  1.0 × 10−3 . Various turbulence models on mesh B2: top view.

Fig. 16 Q criterion  1.0 × 10−3 . Various turbulence models on mesh B2: side view.

3. Fuselage Station 375 vortices are still very coherent. Smaller vortical structures are also
Further aft at FS375 (375 in. from the nose), the smaller outboard seen in the vicinity of the main vortex on the inward side, and upward
vortex has merged with the apex vortex, as shown in Fig. 20. A larger from the surface. There are some fairly significantly differences in the
negative Cp peak is seen in the computational data toward the peak suction pressure predicted in the air dam and crank vortices in
outboard section of the wing that corresponds to the flow over the air the different computational solutions; but, without experimental data
dam. There are no flight-test data available at this location, but the with which to compare, it is not clear which is the most correct. The
other data points fall close to the limits of the rms of the SA and mesh A1 solutions can be discounted as to their accuracy
computational pressure data. As before, the SA turbulence model and based on previous cut lines considered.
mesh A1 show the most deviation from the experimental data. It is significant to note the increased level of unsteadiness shown by
the pressure minima and maxima from the SARC/DDES solution on
4. Fuselage Station 407.5 mesh B2, especially in the outboard location. This is the area of the
At FS407.5, the apex vortex starts to lose some coherence, whereas flow where the air dam and crank vortices interact with each other.
the air dam and wing crank vortices are starting to form, as shown in Note that the data slice extends only to the outboard edge of the
Fig. 21. The computational data compare well with experiments, with main wing (2y∕b  1) and does not include the missile body or
the exception of a single point inboard of the apex vortex. The missile rail.
negative peak of the air dam vortex is captured fairly well (within
the rms limits). The SA and mesh A1 solutions both overpredict the 6. Butt Line 55
negative suction of the crank vortex, as compared with the other Slices of the computed solutions taken along butt lines (constant Y
computed solutions. values) will now be considered. The first is FS55 (55 in. from the
aircraft centerline). As seen in Fig. 23, The main flowfield structure is
5. Fuselage Station 450 the main apex vortex. It appears that the solution on mesh A1
FS450 is the location in the X direction furthest aft at which (baseline) is the only computed solution that compares well with the
solution data are compared. As seen in Fig. 22, the apex vortex is suction peak of the experimental data. Given the poor agreement seen
losing significant strength, whereas the air dam and wing crank at other locations, as well as the artifact near the trailing edge, this can
12 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 17 Isosurfaces of various solution variables, colored by Cp , for SARC∕DDES on mesh B2.

only be a coincidence. It is apparent, however, that the experimental There is also some significant unsteadiness above the wing surface.
suction peak falls within the rms of the computed SARC/DDES This position has historically been a challenging location for
solution on mesh B2. computational prediction due to the high levels of unsteadiness
There is no significant difference between the different turbulence caused by the multiple vortices.
models on mesh B2 for this particular case. It does not appear that any of the computed solutions compare
particularly well with the experimental data, except for the two that
7. Butt Line 70 have not shown themselves to be accurate thus far: the SA solution on
The solution at butt line 70 (70 in. from the centerline), seen in mesh B2 and the SA/DDES solution on mesh A1. All other solutions
Fig. 24, is characterized again by the apex vortex. None of the do not appear to agree very well. However, the experimental data are
computed pressure values quite resolve the suction peak, but the again within the rms pressure data of the SARC/DDES solution on
experimental data fall well within the pressure maxima and minima, mesh B2, with the exception of one point near the leading edge. It
and close to the rms limits. The rms limits demonstrate the significant appears that this particular point could be on the lower surface, but it
is not clear where it is located.
unsteadiness in the flowfield near the apex vortex.
11. Butt Line 184.5
8. Butt Line 80
The final slice of data to be considered here is taken along butt
All computed solutions agree very well with the experimental data
line 184.5, near the wingtip, as shown in Fig. 28. The main flow
at butt line 80 (Fig. 25), even capturing the suction peak at the leading
feature seen in the slice plane is the crank vortex, but its effect is not
edge. Although there is significant unsteadiness near the leading
seen in the computational or experimental data, which is fairly flat
edge, the average computed pressure values come very close to the
along the entire wing surface from leading edge to trailing edge.
experimental data. However, there is some slight disagreement at the
Although the pressure averages do not necessarily agree
trailing edge of the aircraft. completely, the experimental data are within the rms limits of the
computed unsteady data.
9. Butt Line 95
Butt line 95 continues outboard from BL80, and the solution 12. Spectral Data of Select Points
shown in Fig. 26 continues to show the apex vortex as it progresses In addition to the surface pressure data discussed previously, data
downstream and outboard. Again, most of the computed solutions were taken at several locations above the F-16XL wing during the
agree fairly well with each other and the experimental data, with the simulations. The spectral content of the data was analyzed using a
exception of a region somewhat downstream of the apex vortex. windowed fast Fourier transform with a Hanning filter applied to
each window. A window size of 2048 time steps with 50% overlap
10. Butt Line 153.5 between adjacent windows was used. The data were taken at each
Although BL95 was inboard of the air dam and actuator pod, butt time step of 5 × 10−4 s for 8000 time steps (4 s of simulation time).
line 153.5 is just outboard of them both. The solution is now The locations of the pressure tap data taken is summarized in
characterized by the air dam and crank vortices, as seen in Fig. 27. Table 2 and shown graphically in Fig. 29. Taps 1–9 of Table 2 are
Article in Advance / LOFTHOUSE AND CUMMINGS 13
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 18 Pressure tap data for fuselage station 300.

Fig. 19 Pressure tap data for fuselage station 337.5.


14 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 20 Pressure tap data for fuselage station 375.

Fig. 21 Pressure tap data for fuselage station 407.5.


Article in Advance / LOFTHOUSE AND CUMMINGS 15
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 22 Pressure tap data for fuselage station 450.

Fig. 23 Pressure tap data for butt line 55.


16 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 24 Pressure tap data for butt line 70.

Fig. 25 Pressure tap data for butt line 80.


Article in Advance / LOFTHOUSE AND CUMMINGS 17
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 26 Pressure tap data for butt line 95.

Fig. 27 Pressure tap data for butt line 153.5.


18 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 28 Pressure tap data for butt line 184.5.

shown in Fig. 29, as well as taps 10–15. Tap 5 is the horizontal and
vertical center of the cluster of nine tap points.
The sound pressure level (SPL) of each of the apex tap points for
the computational solutions on each mesh is shown in Fig. 30. The
dominant frequencies in the unsteady data should appear as peaks in
the SPL plots. There is one peak at about 45 Hz that appears in the
solutions on all of the meshes, but it appears that the two largest
meshes (mesh A2 and mesh B2) seem to have some lower-frequency
content that does not appear in the solutions on the two smaller
meshes (mesh C2 and mesh D2). Recall that the smaller meshes
tended to refine cells in the area of the main wing vortices but not
elsewhere (although mesh D2 does have a refined prism layer near the
surface, whereas mesh C2 does not).
The data with the largest SPL value come from tap 5, which is the
center data point of the array of nine points near the apex. The data at
Fig. 29 Data extraction points. Main leading-edge apex points and wing
apex points are shown.

Table 2 Tap locations

Tap no. Location X, in. Y, in. Z, in. this point were plotted for each of the SA/DDES solutions on all
1 Main apex 196.8505 49.2126 94.4882 meshes, as well as for the different turbulence models on mesh B2, in
2 Main apex 196.8505 49.2126 98.4253 Fig. 31. Note that each DDES solution on the larger meshes has a
3 Main apex 196.8505 49.2126 100.3938 peak at about the same location. The RANS solutions and the DDES
4 Main apex 196.8505 55.1181 94.4882 solution on the smallest mesh (mesh A1) do not show peaks at the
5 Main apex 196.8505 55.1181 98.4253
6 Main apex 196.8505 55.1181 100.3938 same point.
7 Main apex 196.8505 62.9922 94.4882 The decibel value of the maximum SPL of each solution, along
8 Main apex 196.8505 62.9922 98.4253 with the frequency value, is summarized in Table 3. Note that all of
9 Main apex 196.8505 62.9922 100.3938 the frequencies are at about 45 Hz, with SPL values at around 145 dB.
10 Wing apex 452.7562 173.2284 94.4882 Although the solutions on the largest meshes have the same peaks, it
11 Wing apex 452.7562 173.2284 102.3623 is much easier to distinguish these peaks on the smaller meshes that
12 Wing apex 452.7562 173.2284 110.2363 do not have the lower-frequency content.
13 Wing apex 452.7562 181.1025 94.4882
The spectral content of the data taken at taps 10–15 for the
14 Wing apex 452.7562 181.1025 102.3623
15 Wing apex 452.7562 181.1025 110.2363 solutions on different meshes is shown in Fig. 32. There are no
apparent peaks in the spectral content at these locations.
Article in Advance / LOFTHOUSE AND CUMMINGS 19
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 30 Apex tap spectral analysis for SA/DDES on various meshes.

Fig. 31 Apex pressure tap spectral analysis for various turbulence models and meshes; tap 5.

Table 3 Maximum SPL for tap 5


C. Integrated Loads
The integrated aerodynamic loads in the form of lift, drag, and Turbulence model Mesh SPL, dB Frequency, Hz
pitching moment coefficients were recorded at each time step SA∕DDES A2 146 46
throughout the unsteady portion of the simulations. The values of SA∕DDES B2 146 43
SA∕DDES C2 145 46
each during the final 8000 time steps (4 s of simulation time) were
SA∕DDES D2 145 44
analyzed, and the average, minima, maxima, and rms values are SARC∕DDES B2 149 43
summarized in Tables 4–6.
20 Article in Advance / LOFTHOUSE AND CUMMINGS
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 32 Wing pressure tap spectral analysis for SA/DDES on various meshes.

Notably, the lift values for each solution method, regardless of Table 5 Drag coefficient: average, minimum, maximum, and rms
mesh and turbulence model, agree fairly well with respect to the
average values. Additional scatter is seen in the minima and maxima, Turbulence
model Mesh CD min CD avg CD max CD rms
but they also agree fairly well. The unsteadiness of the solution is
measured by the rms values. Here, the RANS solutions showed the SA∕DDES A1 4.414E − 1 4.518E − 1 4.636E − 1 3.874E − 3
SA∕DDES A2 4.474E − 1 4.602E − 1 4.764E − 1 3.980E − 3
lowest amount of unsteadiness, with the SA solution’s CL rms value SA∕DDES B2 4.480E − 1 4.616E − 1 4.750E − 1 4.390E − 3
nearly one magnitude lower than the others. The drag predictions SA∕DDES C2 4.424E − 1 4.556E − 1 4.670E − 1 4.028E − 3
followed similar trends. The rms values for lift and drag were one to SA∕DDES D2 4.452E − 1 4.610E − 1 4.786E − 1 4.270E − 3
two orders of magnitude less than their actual values. SA B2 4.588E − 1 4.606E − 1 4.682E − 1 1.339E − 3
The pitching moment coefficient, on the other hand, showed a SARC B2 4.560E − 1 4.622E − 1 4.706E − 1 2.128E − 3
significant amount of sensitivity to the flow unsteadiness, with the SARC∕DDES B2 4.468E − 1 4.600E − 1 4.738E − 1 2.886E − 3
rms values only one order of magnitude lower than the average
values. The minimum pitching moment coefficient of many of the
solutions was negative, whereas the maximum of them all was

Table 4 Lift coefficient: average (avg), minimum (min), maximum Table 6 Pitching moment coefficient: average, minimum, maximum,
(max), and rms and rms
Turbulence Turbulence
model Mesh CL min CL avg CL max CL rms model Mesh CM min CM avg CM max CM rms
SA∕DDES A1 7.404E − 1 7.772E − 1 8.066E − 1 1.038E − 2 SA∕DDES A1 4.774E − 3 1.394E − 2 2.446E − 2 4.018E − 3
SA∕DDES A2 7.460E − 1 7.866E − 1 8.278E − 1 1.171E − 2 SA∕DDES A2 −6.412E − 3 1.129E − 2 2.836E − 2 4.928E − 3
SA∕DDES B2 7.504E − 1 7.918E − 1 8.254E − 1 1.287E − 2 SA∕DDES B2 −4.744E − 3 9.994E − 3 2.212E − 2 4.700E − 3
SA∕DDES C2 7.492E − 1 7.836E − 1 8.136E − 1 1.138E − 2 SA∕DDES C2 1.536E − 3 1.308E − 2 2.504E − 2 3.832E − 3
SA∕DDES D2 7.386E − 1 7.848E − 1 8.318E − 1 1.238E − 2 SA∕DDES D2 −2.072E − 3 1.094E − 3 2.396E − 2 4.154E − 3
SA B2 7.836E − 1 7.926E − 1 8.088E − 1 2.626E − 3 SA B2 3.076E − 3 1.030E − 2 1.416E − 2 1.086E − 3
SARC B2 7.738E − 1 7.920E − 1 8.144E − 1 5.806E − 3 SARC B2 −3.848E − 3 4.770E − 3 1.196E − 2 2.096E − 3
SARC∕DDES B2 7.492E − 1 7.864E − 1 8.244E − 1 1.366E − 2 SARC∕DDES B2 −4.868E − 3 1.163E − 2 2.648E − 2 5.222E − 3
Article in Advance / LOFTHOUSE AND CUMMINGS 21
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

Fig. 33 Convergence plots and spectral content for lift, drag, and pitching moment.

positive, indicating a large amount of sensitivity to the details of forces and moments (although SA/DDES on mesh B2 is slightly off
the flow. with the pitching moment). SA/DDES on mesh B2 produces nearly
A spectral analysis was conducted on the integrated forces and the same frequency value, but with a slightly lower peak. The smaller
moments, using the same methodology as described previously for the refined meshes, mesh C2 and mesh D2, do not produce the same
tap points in order to more fully understand the frequency content. The frequency content for the integrated loads.
convergence data and the spectral content for each of the solutions are
shown in Fig. 33, and the peak loads and frequencies are summarized D. Computational Cost
in Table 7. Note that the spectral content for the integrated loads is The computational costs of the simulations reported herein are
given in terms of the force or moment coefficient power, which is summarized in Table 8. The simulations were run on an IBM
equivalent to sound pressure level for pressure spectra. iDataPlex with Intel Sandy Bridge processors and an Infiniband
Interestingly, SARC/DDES on mesh B2 and SA/DDES on mesh interconnect. Table 8 shows the number of computational cores used
A2 give nearly the same spectral peak at about 12.7 Hz for all three for each simulation; the total wall-clock time (in hours:minutes:
22 Article in Advance / LOFTHOUSE AND CUMMINGS

Table 7 Integrated loads spectral analysis (peak power and The spectral content of pressure data taken at specific points, as
frequency of peak) well as of the integrated forces and moments, was also presented.
Turbulence Again, the largest meshes, coupled with the DDES turbulence
model Mesh C2L ∕Hz C2D ∕Hz C2M ∕Hz models, showed the best resolution of frequency content.
SARC∕DDES B2 0.00588∕12.7 0.00201∕12.7 0.00152∕12.7
SA∕DDES A2 0.00545∕12.7 0.00186∕12.7 0.00153∕12.7
SA∕DDES B2 0.00469∕12.8 0.00159∕12.7 0.00141∕13.5 Acknowledgments
SA∕DDES C2 0.00344∕13.6 0.00117∕13.6 — —/— — The authors gratefully acknowledge the support of the U.S. Air
SA∕DDES D2 0.00284∕13.8 0.000969∕13.6 — —/— — Force Academy and the U.S. Air Force Office of Scientific
Research. Pre- and postprocessing were accomplished using the
High-Performance Computing (HPC) Portal to remotely connect to
the U.S. Department of Defense (DOD) HPC machines. The
generous use of DOD high-performance computing resources was
Table 8 Computational cost of each simulation
indispensable to this investigation and was greatly appreciated.
Total Turbulence No. of Wall- CPU Relative Special thanks go to David McDaniel and Scott Morton of the DOD
Mesh cells model cores clock time hours cost Computational Research and Engineering Acquisition Tools and
A1 11.9M SA∕DDES 1024 04:01:19 4118 1.0 Environments Team, and Principle Developers of Kestrel, for their
A2 89.7M SA∕DDES 2048 15:04:47 30883 7.5 assistance in pre- and postprocessing the simulations. The views
B2 86.5M SA∕DDES 2048 14:38:55 30000 7.3 expressed in this paper are those of the authors and do not reflect the
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

C2 56.8M SA∕DDES 2048 07:33:44 15487 3.8 official policy or position of the U.S. Air Force, the U.S. Department
D2 68.8M SA∕DDES 2048 12:03:07 24682 6.0 of Defense, or the U.S. Government.
B2 86.5M SA 2048 14:26:56 29591 7.2
B2 86.5M SARC 2048 14:25:52 29554 7.2
B2 86.5M SARC∕DDES 2048 14:43:43 30164 7.3
References
[1] Obara, C. J., and Lamar, J. E., “Overview of the Cranked-Arrow Wing
Aerodynamics Project International,” Journal of Aircraft, Vol. 46, No. 2,
seconds); the total CPU hours consumed (wall clock multiplied by 2009, pp. 355–368.
number of cores); and the relative cost, with the baseline cost being [2] Lamar, J., Obara, C., Fisher, B., and Fisher, D., “Flight, Wind-Tunnel,
that of the simulation on mesh A1. and Computational Fluid Dynamics Comparison for Cranked Arrow
Most of the simulations required approximately 30,000 CPU hours Wing (F-16XL-1) at Subsonic and Transonic Speeds,” NASA TP-2001-
of time, or about seven times the cost of the simulation on the baseline 210629, 2001.
[3] Lamar, J. E., and Obara, C. J., “Review of Cranked-Arrow Wing
mesh. Note that the computational cost is directly related to the size of Aerodynamics Project: Its International Aeronautical Community
the mesh, and it only varies weakly with the different turbulence Role,” 45th AIAA Aerospace Sciences Meeting and Exhibit, AIAA
models. It is evident that the larger meshes provide better results and Paper 2007-0487, Jan. 2007.
that there is minimal additional cost for a DDES turbulence model. [4] Morton, S., McDaniel, D., and Cummings, R., “F-16XL Unsteady
Simulations for the CAWAPI Facet of RTO Task Group AVT-113,” 45th
AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper 2007-
IV. Conclusions 0493, Jan. 2007.
Computational simulations of a complex, nonlinear aerodynamic [5] Bertin, J. J., and Cummings, R. M., Aerodynamics for Engineers,
Pearson, Upper Saddle River, NJ, 2014, pp. 404–414.
flow about the F-16XL aircraft at low-speed high-angle-of-attack [6] Cummings, R. M., Mason, W. H., Morton, S. A., and McDaniel, D. R.,
flight conditions were completed using the Kestrel flow solver on a Applied Computational Aerodynamics: A Modern Engineering
variety of meshes with several turbulence models. Two Reynolds- Approach, Cambridge Univ. Press, New York, 2015, pp. 596–602.
averaged Navier–Stokes (RANS) models, the Spalart–Allmaras and [7] Spalart, P. R., Jou, W.-H., Strelets, M., and Allmaras, S. R., “Comments
the Spalart–Allmaras with rotation correction, were coupled with the on the Feasibility of LES for Wings, and on a Hybrid RANS/LES
delayed detached-eddy simulation and used. Several meshes were Approach,” 1st AFOSR International Conference on DNS/LES,
obtained by refining a baseline mesh using solution-based adaptive Greyden Press, Centerville, OH, 1997, pp. 137–147.
mesh refinement. The solution variables used were the Q criterion, [8] Spalart, P. R., Deck, S., Shur, M. L., Squires, K. D., Strelets, M. K., and
scaled Q criterion, and vorticity magnitude. The meshes that gave the Travin, A., “A New Version of Detached-Eddy Simulation, Resistant to
Ambiguous Grid Densities,” Theoretical and Computational Fluid
best results, as judged by comparison with the surface pressure Dynamics, Vol. 20, No. 3, July 2006, pp. 181–195.
coefficient data from flight tests as well as the unsteady content, were doi:10.1007/s00162-006-0015-0
the largest meshes. These were mesh A2, generated by refining all [9] Morton, S. A., Cummings, R. M., and Cholodar, D. B., “High
cells of the original baseline mesh, and mesh B2, generated by Resolution Turbulence Treatment of F/A-18 Tail Buffet,” Journal of
refining cells where the Q criterion exceeded a value of 0.0. Aircraft, Vol. 44, No. 6, 2007, pp. 1769–1775.
Generation of mesh A2 did not require any additional solutions, [10] Forythe, J. R., and Woodson, S. H., “Unsteady Computations of Abrupt
whereas mesh B2 required an initial solution on mesh A1, but the Wing Stall Using Detached-Eddy Simulation,” Journal of Aircraft,
additional cost of computing an unsteady delayed detached-eddy Vol. 42, No. 3, 2005, pp. 606–616.
[11] Jeans, T. L., McDaniel, D. R., Cummings, R. M., and Mason, W. H.,
simulation (DDES) solution on mesh A2 was minimal compared to
“Aerodynamic Analysis of a Generic Fighter Using Delayed Detached-
the solution on mesh B2 plus the initial solution on mesh A1. Eddy Simulation,” Journal of Aircraft, Vol. 46, No. 4, 2009, pp. 1326–
Therefore, the total turnaround time could be quickest by simply 1339.
refining all the cells of an initial, relatively coarse mesh. [12] Morton, S. A., “Detached-Eddy Simulations of Vortex Breakdown over
As expected, the RANS turbulence models and the smallest a 70-Degree Delta Wing,” Journal of Aircraft, Vol. 46, No. 3, 2009,
meshes showed the least amount of unsteadiness in the computed pp. 746–755.
solutions. The RANS turbulence models did not resolve any of the [13] Post, D. E., “Highlights of the CREATE Program,” 2010 DoD High
turbulent fluctuations but only the mean flow properties, and so they Performance Computing Modernization Program Users Group
could not be expected to resolve the level of unsteadiness of a hybrid Conference, IEEE Publ., Piscataway, NJ, 2010, pp. 430–437.
RANS/large-eddy simulation model such as DDES. The additional doi:10.1109/HPCMP-UGC.2010.55
[14] Roth, G. L., Morton, S. A., and Brooks, G. P., “Integrating CREATE-
cost of the DDES turbulence models was minimal (on the same AV Products DaVinci and Kestrel: Experiences and Lessons Learned,”
mesh), and so should always be used for flow conditions with a large 50th AIAA Aerospace Sciences Meeting, AIAA Paper 2012-1063,
amount of unsteady separation. The largest meshes, coupled with the Jan. 2012.
DDES turbulence models, showed very good agreement with the [15] Morton, S., and McDaniel, D., “Numerical Simulation of the F-16XL at
flight-test data, especially when considering the unsteady rms limits. Full-Scale Flight Test Conditions Using a Near-Body Off-Body CFD
Article in Advance / LOFTHOUSE AND CUMMINGS 23

Approach,” 33rd AIAA Applied Aerodynamic Conference, AIAA Paper [18] Tomaro, R. F., Strang, W. Z., and Sankar, L. N., “An Implicit Algorithm
2015-2873, June 2015. for Solving Time Dependent Flows on Unstructured Grids,” 35th AIAA
[16] Godunov, S. K., “A Difference Scheme for Numerical Computation of Aerospace Sciences Meeting, AIAA Paper 1997-0333, Jan. 1997.
Discontinuous Solution of Hydrodynamic Equations,” Sbornik [19] Spalart, P. A., and Allmaras, S. R., “A One-Equation Turbulence Model
Mathematics, Vol. 47, No. 3, 1959, pp. 271–306. for Aerodynamic Flows,” 30th AIAA Aerospace Sciences Meeting and
[17] Tramel, R. W., Nichols, R. H., and Buning, P. G., “Addition of Exhibit, AIAA Paper 1992-0439, Jan. 1992.
Improved Shock-Capturing Schemes to OVERFLOW 2.1,” 19th AIAA [20] Spalart, P. R., and Shur, M. L., “On the Sensitization of Turbulence
Computational Fluid Dynamics Conference, AIAA Paper 2009-3988, Models to Rotation and Curvature,” Aerospace Science and Technology,
June 2009. Vol. 1, No. 5, 1997, pp. 297–302.
Downloaded by MONASH UNIVERSITY on September 25, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034045

You might also like