See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269050774

CFD Analysis of CUBRC Base Flow Experiments

Conference Paper · January 2010

DOI: 10.2514/6.2010-1250

CITATIONS READS
13 125

3 authors, including:

Michael Barnhardt Graham Candler

33 PUBLICATIONS 587 CITATIONS

University of Minnesota Twin Cities
519 PUBLICATIONS 12,867 CITATIONS
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Development of a Thermal-Vacuum Chamber for Extending Calibration of Optical Particle Detectors to Stratospheric Conditions View project

Characterizing Freestream Tunnel Disturbances View project

All content following this page was uploaded by Michael Barnhardt on 09 April 2021.

The user has requested enhancement of the downloaded file.

 CFD Analysis of CUBRC Base Flow Experiments

Michael Barnhardt∗ and Graham V. Candler†

University of Minnesota, Minneapolis, MN, 55455, USA
Matthew MacLean‡
CUBRC/LENS, Buffalo, NY 14225

This paper presents results from a computational analysis of a series of experiments con-
ducted in the CUBRC 48” reflected shock tunnel for the purpose of studying the transition
of afterbody wake flows. The experiments examined the flow over a spherically-blunted
capsule, roughly chosen to be a scale representation of the new Orion crew module cur-
rently being designed by NASA. In this study, we have focused on three test runs for
analysis, corresponding to low, medium, and high Reynolds number conditions. Each case
has been examined with the intent of understanding the influence of turbulence modeling,
time accuracy, and flux discretization on solution accuracy. Numerically, we find a stan-
dard first order in time, second order in space (with modified Steger-Warming fluxes) to
be adequate for the external, reentry type flows of interest. For the low Reynolds number
case (ReD ≤ 106 ), a laminar Navier-Stokes simulation is sufficient to accurately capture the
statistical character of the flowfield. At the medium Reynolds number (ReD = 6.3×106 ), the
analysis indicates a possibly transitional wake where heat transfer predictions are bounded
by laminar and turbulent simulations. At the high Reynolds number (ReD = 10.8 × 106 ), the
experimental data are best matched by a fully turbulent calculation using the Detached
Eddy Simulation (DES) form of the Spalart-Allmaras turbulence model. Additionally,
comparisons between Reynolds-averaged Navier-Stokes (RANS) and DES calculations re-
veal the inability of RANS models to correctly capture either the rapidly fluctuating flow
transients or mean field. Repeating the DES calculation with an alternative second order
in time, low-dissipation flux scheme shows a substantial improvement in the simulation’s
ability to capture flow transients and may prove crucial for applications in which this is
a priority. It is observed that the improvements seen in moving from RANS to DES and
from Steger-Warming to the low-dissipation scheme are related in that both ultimately
reduce the level of unnecessary dissipation present in the simulation.

I. Introduction
espite nearly 50 years of research, analysis of high-speed, unsteady wake flows remains a challenging and
D unsolved problem. Yet as hypersonic applications become increasingly complex, so too increases the
demand for computational tools capable of accurately predicting unsteady, separated flows. We know from
experience that the effects of turbulent heating can be a significant contributor to overall vehicle heating. In
practice, when any appreciable amount of flow unsteadiness is present, whether laminar or turbulent, it is
most often offset by design margins which try to compensate for the unknown accuracy of the computational
results. The practice of applying design margins is severely limiting, especially in the tight budget climate
of today’s space industry. A prime example is thermal protection system (TPS) design for space vehicles.
The launch cost of a space vehicle is predominantly a function of its mass: more mass requires more fuel
and potentially greater vehicle complexity. In the context of the Space Shuttle, the average $500 Million
launch cost results in a net payload cost of approximately $20,000 per kilogram. It is critically important to
a program budget that payload mass be minimized to the extent possible in order to increase the trade space
when choosing a launch vehicle. One must also consider that, in the event that a particular launch vehicle’s
∗ Research Scientist (currently with ELORET Corp., Sunnyvale, CA), AIAA Member; michael.d.barnhardt@nasa.gov
† Professor,Department of Aerospace Engineering and Mechanics, AIAA Fellow; candler@aem.umn.edu
‡ Senior Research Scientist, AIAA Senior Member; maclean@cubrc.org

1 of 21

American Institute of Aeronautics and Astronautics

payload limit is exceeded, it becomes necessary to trim payload mass. In this scenario, every kilogram
allocated to TPS is one which must be subtracted from the science payload, resulting in the reduction or
elimination of mission objectives. Therefore, by applying margins to the TPS, programs may unnecessarily
reduce risk at the expense of greater mission costs or a reduction in science payload. As of this writing, the
TPS margin policy for the Orion crew module lists the margin due to turbulent afterbody heating second
only to that due to radiative heating.
The analysis of hypersonic wakes was initially stimulated in the late 1950’s by the problems associated
with observing and tracking slender bodies entering the Earth’s atmosphere. Early research naturally concen-
trated on analytical approaches, especially the application of integral methods and the proto-computational
integral-strip method (see Refs. [1,2]). Around this same time, the alternative approach of finding exact solu-
tions to approximate, discretized equations was also being advanced and, as computational power gradually
grew, the integral-strip methods gave way to finite-difference calculations. Reference [3] is an early example
of the finite-difference technique in which the far wake was computed from the boundary layer equations
along with models for turbulent diffusion and an eight-species chemical kinetic mechanism.
In subsequent years, up until the mid 1980’s, the focus of hypersonics shifted toward blunt body anal-
ysis in which the wake problem was overshadowed by the need for accurate modeling of forebody heating.
Eventually, finite-difference was replaced by the conservative finite-volume approach. The Aeroassist Flight
Experiment (AFE), begun in 1988 but which was ultimately canceled, renewed interest in wake flow com-
putations. A series of papers were published which examined the structure of the laminar AFE wake.4–6
The first paper by Gnoffo, in particular, demonstrated by experimental validation that it was possible to
predict laminar aftbody heating with a reasonable degree of accuracy. A similar conclusion was echoed in
Reference [7].
In response to the formation of AGARD Working Group 18, a panel meant to address issues related
to blunt body wake flowfields, two independent test programs were initiated. In Refs. [8, 9], numerical
and experimental results were obtained on a 6 inch diameter 70◦ blunted cone at Mach 6 and 10, respec-
tively. An especially interesting outcome of these studies was confirmation that heating due to shear layer
impingement could actually be greatest when the shear layer was transitional, versus purely laminar or tur-
bulent. Numerical comparisons were somewhat inconclusive however because of the use of an axisymmetric
formulation and lack of a turbulence model, in contradiction to the experimental data which suggested a
transitional/turbulent shear layer. Holden tested an identical configuration using a variety of numerical
techniques, ranging from Direct Simulation Monte-Carlo (DSMC) to continuum Navier-Stokes simulations
and several hybrid combinations thereof.10 Again, the flow was treated axisymmetrically and assumed to
be laminar. Base pressure and heat transfer were generally in poor agreement with experimental data, es-
pecially so for the Navier-Stokes calculations. Expansion of the base flow to a rarefied state was raised as
a possible explanation for the error, although results of hybrid DSMC/Navier-Stokes calculations presented
in the study seem to suggest that this effect is not solely accountable.
Within the past decade, the wake problem has been revisited in light of NASA’s high-profile Mars Science
Laboratory (MSL) and Constellation programs. It is common for these types of missions to employ safety
margins on unsteady/turbulent afterbody heating in excess of 50-100%. Reference [11] performed an in-depth
analysis of available data from the Apollo AS-202 flight test. The flow was uniformly laminar throughout
the trajectory and agreement was generally fair in light of the estimated 20% uncertainty in the flight
data. Still, several of the gauge locations showed gross error and the authors acknowledged unsteadiness
in the computations which was not accounted for in the analysis. A similar attempt was made to examine
aftbody heating of Viking Lander 1 in Reference [12]. This study relied on the same methodology as in
the AS-202 study. Results were generally not very favorable; proposed reasons included underestimation of
conduction effects in the heatshield, improper modeling of surface catalysis, and various limitations of the
computational methods. In fact, these very methods, although they are still widely regarded as the state of
the art in aerothermal modeling today, originated over 20 years ago.
Based on the history we have recited, we may conclude the following about the current state of the art:
• Analysis of high-speed, unsteady wake flows is a very difficult and unsolved problem.
• There are few validation studies by which a researcher may be guided to adopt a consistent strategy
for analysis.
Recently, we have proposed such a methodology for simulations of high-speed, unsteady flows and exam-
ined it in the context of an Earth entry flight environment via the Reentry-F experiment.13 By comparing

2 of 21

American Institute of Aeronautics and Astronautics

our simulations to the experimental data, we concluded that the methodology could indeed predict vehicle
heating in a turbulent environment within the limits of our ability to accurately specify the flight conditions.
Yet the uncertainty of the experimental specification remains and, with it, a small amount of uncertainty
in the analysis itself is unavoidable: there is simply no mechanism by which we can unequivocally identify
the relative contributions of the various sources of error. A second consideration is the relative simplicity
of the Reentry-F configuration. The slender shape of the cone, with its sharply terminating base, reduces
the complexity of the near wake to its basic elements: a shear layer and recirculating region. While this is
helpful as a means to understand the essential structure and behavior of a turbulent base flow, the Reentry-F
test is insufficient to answer all questions pertaining to validation of our methodology for a practical flight
configuration like the spherical capsule proposed for NASA’s Orion crew module.
In this paper, we present further validation of our methods against spherical capsule experiments con-
ducted in the high-speed wind tunnels at CUBRC. The controlled nature of the CUBRC tests and the
use of modern data acquisition techniques provides a tremendous wealth of reliable data. Improvements
in instrument design allow modern models to be densely instrumented with gauges capable of accurately
resolving transient phenomena within the kilohertz frequency range (recall for comparison that Reentry-F’s
heat gauges operated at a 10 Hz sampling rate). Moreover, the capsule model which we have selected for
this study is essentially a stand-in for an Apollo-style configuration and hence the Orion crew module as
well. The tests are therefore relevant toward understanding the nature of separated flows in a practical flight
configuration.

II. Description of Experiment

A. CUBRC Facilities
The CUBRC facilities are comprised of four distinct tunnels: the LENS-I, LENS-II, and 48” reflected shock
tunnels, and the LENS-XX expansion tunnel.14, 15 Each tunnel is designed to replicate a particular range of
velocity and altitude conditions. Taken together, the CUBRC tunnels are able to reproduce almost entirely
the Apollo and Space Shuttle entry trajectories in the continuum regime.
The reflected shock tunnels work by heating and pressurizing a stagnant test gas which is then expanded
through a converging-diverging nozzle to achieve the desired test condition. They were designed primarily
to investigate flow physics of interceptor and scramjet configurations at full-scale. LENS-I is thus capable of
duplicating flight conditions from Mach 7 to 15 while matching Reynolds number up to Mach 22. LENS-II
is a complementary tunnel with an operational capability from Mach 3.5 to 10. The LENS-XX expansion
tunnel is designed to produce a high enthalpy flow in such a way that the flow is largely free of frozen,
dissociated contaminants. This is accomplished in two stages. First, the test gas is compressed and heated
by a shock without being stagnated by a reflected shock. Next, additional energy is added to the flow via an
unsteady expansion through an acceleration segment. The test gas is then expanded isentropically through
the test chamber to reach its final high-enthalpy state. While this process typically results in shorter test
times than a reflected shock tunnel (up to 4 ms, versus approximately 25 ms in LENS-I and 100 ms in
LENS-II),16 it is necessary in order to cleanly duplicate a flight condition in excess of 10 MJ/kg.17
The driving factor in selecting a suitable set of tests for this study was to maximize the test duration
time such that the turbulent base flow could fully establish and still have sufficient time to acquire data
over many flow times (defined by the freestream velocity and capsule diameter). A secondary consideration
was to maximize the peak Reynolds number to ensure turbulent behavior. A trade study was performed in
Ref. [18] in which they concluded that the 48” tunnel operating at a nominal Mach number of 6 offered the
most balanced solution. At this condition, the tunnel can reach Reynolds numbers as high as 8.5 × 107 per
meter, or a full order of magnitude greater than what was seen to produce transitional wake behavior in
previous studies. Preliminary runs indicated that reliable heat transfer measurements typically took around
9 ms to establish, followed by an additional 10-12 ms of usable test time.
Further details of the CUBRC facilities’ design and operation may be found in the previously cited works.

B. Capsule Configuration
As stated earlier, the nominal model configuration used in this study is an Apollo-style spherical capsule
chosen to closely resemble the proposed Orion spacecraft being designed by NASA as a replacement for the
soon-to-retire Space Shuttle fleet. A detailed schematic is shown in Figure 1. All dimensions shown in the

3 of 21

American Institute of Aeronautics and Astronautics

 (a) Capsule (b) Round Sting Mount

Figure 1. Schematic of the spherical capsule geometry.

figure are given in inches. The scale of the model was chosen after carefully considering the test core diameter
of the 48” tunnel as well as the time necessary for the model’s wake to fully establish. The model is fixed with
a 28◦ angle of attack, which is near the peak L/D for this configuration. Four sting mount configurations
were used during the course of testing: (1) round center mounted sting, (2) diamond cross-section center
mounted sting, (3) a split sting attached to the capsule at two lateral points, and (4) an aerodynamic blade
sting attached to the windward backshell. Of these, only the round center mounted sting was considered in
this study.
The topologies and sizing used in this work are developed from experience with the simulations presented
in Ref. [13]. While not shown here, the sizing was further refined through a preliminary grid resolution
study of the most turbulent case and a single grid was used for all simulations. To the extent possible, grid
refinement was localized to the wake region of the leeside aftbody so that unnecessary refinement of the far
field was minimized. The final result is a mesh comprised of 9.7 million cells. The cell width in the wake is
nominally 2.5 mm, or approximately 1% of the model diameter. Figure 2 depicts the surface and symmetry
planes of the mesh.
The capsule model is instrumented with a total of 85 sensors, including all inserts and optional mea-
surements. Of these, 38 are piezoelectric pressure gauges and 47 are heat transfer gauges. With the round
center-mounted sting in place, the number of gauges was reduced to 35 and 39, respectively. The complete
instrumentation map is shown in Figure 3. The forebody instrumentation is relatively sparse and is meant
only for basic confirmation of test conditions. Four coaxial thermocouple gauges are placed on the capsule
centerline, with two additional thin-film gauges at each shoulder. The pressure gauges are located along two
orthogonal rays. It is important to note that the limited amount of forebody instrumentation also limits
our knowledge of boundary layer transition, a fact which affects some of our subsequent modeling choices,
as will be seen.
The remaining gauges are distributed in rays along the aftbody. In this study, because the number of
gauges is quite large, proper identification of each component is a critical task in order to coherently break
down the presentation of results. To this end, we have chosen to label the instrument rays according to a
clockwise angular naming convention, where the angle is denoted by φ. At 28◦ angle of attack, the φ = 0◦
ray corresponds to the most leeward side of the capsule while φ = 180◦ is most windward. The individual
gauges are identified using a consecutive numbering scheme. Thus, accounting for missing and inoperable
gauges, the pressure gauges are simply labeled [1 - 35] and the heat transfer gauges [36 - 74]. Note that this
numbering scheme differs somewhat from that seen in Ref. 18. For detailed comparison to that study, refer
to Table 1 for a complete transposition of the two labeling schemes. Throughout the remaining text, we
have made an effort to further identify the gauges descriptively so that the spatial relationships among the

4 of 21

American Institute of Aeronautics and Astronautics

 (a) Symmetry Plane (b) Perspective View

Figure 2. Grid topology used in the spherical capsule simulations.

gauges are clear.

C. Test Conditions
Numerous datasets are available from this series of experiments, representing runs using the various sting
configurations across a range of Reynolds numbers and freestream enthalpies. For simplicity, we have chosen
to limit our analysis to the runs which used the round center mounted sting. Five runs were made in this
configuration, spanning a range of Reynolds numbers from 2.6 × 106 to 8.5 × 107 per meter. The relevant
freestream conditions are given in Table 2. Of these cases, Run 5 exhibited peculiarities during the data
reduction process and so it was rejected for inclusion in this study. We also determined that Runs 1 and 3
were redundant because their respective Reynolds numbers placed them both in the transitional regime and
our primary interest is to assess turbulence modeling, not enthalpy effects. Our final analysis is therefore
concerned with Runs 4, 3, and 2, representing nominally low, medium, and high Reynolds number wake
flows, respectively.

III. Computational Approach

Our computational approach has been documented in several prior publications. Here we restrict our
description to a brief summary only; we refer the reader to Refs. [13, 19, 20] for more detailed descriptions of
the methods applied in this study.
We utilize two CFD codes: the DPLR code developed at NASA Ames and its unstructured analog, US3D,
developed at the University of Minnesota. The two codes share a great deal of heritage. Both are parallel,
finite-volume codes which solve the Navier-Stokes equations extended to include finite-rate chemistry and
thermal nonequilibrium.19–21 For the purposes of this study, US3D and DPLR may be considered nearly
identical, save for a handful of modeling options and differences associated with the unstructured formulation
of US3D, which will be made clear below.
Nearly all cases presented herein are simulated using a modified form of the Steger-Warming flux-vector
splitting scheme for the inviscid fluxes with second order spatial accuracy via MUSCL extrapolation. Sim-
ilarly, the viscous fluxes are computed to second order. A difference arises, however, in the formulation
of gradients: DPLR uses a central differencing approach while US3D computes gradients from a weighted
least-squares reconstruction. A first order backward Euler scheme is used for time integration, with time
accuracy maintained by intentionally limiting time steps to be less than the smallest observed convective
time scale. An exception to this approach is presented in the Results section for the high Reynolds number
case, Run 2. For this case, we have also implemented the kinetic energy consistent, low dissipation flux

5 of 21

American Institute of Aeronautics and Astronautics

 Table 1. Transposition of gauge numbering schemes referred to in this study.

Current Ref. 18
Pressure 1 - 35 P1 - P7, P10 - P37
Heat Transfer 36 - 74 T1 - T11, T13 - T17, T21 - T43
Forebody
1 - 3, 12 - 15 P1 - P3, P14 - P17
36 - 41 T1 - T6
φ = 0◦
4-7 P4 - P7
42 - 46 T7 - T11
φ = 13◦
16 - 18 P18 - P20
52 - 55 T21 - T24
φ = 26◦
19 - 21 P21 - P23
56 - 59 T25 - T28
φ = 38◦
22 - 24 P24 - P26
60 - 63 T29 - T32
φ = 64◦
25 - 27 P27 - P29
64 - 66 T33 - T35
φ = 90◦
28 - 30 P30 - P32
67 - 69 T36 - T38
φ = 135◦
31 - 33 P33 - P35
70 - 72 T39 - T41
φ = 150◦
34, 35 P36, P37
73, 74 T42, T43
φ = 180◦
8 - 11 P10 - P13
47 - 51 T13 - T17

Table 2. Test conditions for the spherical capsule with round center mounted sting.

Run 1 Run 2 Run 3 Run 4 Run 5

ReL 13.50 42.40 24.80 2.57 84.90 (106 /m)
M∞ 6.37 6.41 6.43 6.46 6.45
3
ρ∞ 0.05261 0.19252 0.11229 0.01371 0.33424 (kg/m )
U∞ 964.63 1102.46 1106.85 1292.38 994.11 (m/s)
T∞ 57.08 73.63 73.76 99.62 59.13 (K)
TWall 295.56 295.56 292.78 292.78 292.78 (K)

6 of 21

American Institute of Aeronautics and Astronautics

 Figure 3. Instrumentation map for spherical capsule experiments. (l) aftbody, (m) side view, (r) forebody.

scheme of Subbareddy and Candler.22 Time integration of the scheme is second order by way of a dual-time
sub-iterative technique.
As noted in Table 2, the flows of interest all have a Mach number near 6.5. At this Mach number,
there is insufficient energy to cause dissociation of the gas - a perfect gas model is therefore employed for
all simulations. Similarly, simple transport models are assumed to be adequate: Blottner curve fits and
an Eucken relation are used for species viscosity and thermal conductivity, respectively, along with Wilke’s
mixing rule to obtain mixture values, and a constant Lewis number assumption (equal to 1.4) is used to
calculate diffusion coefficients.
A variety of turbulence models are tested, although our primary focus is on the application of DES with
the Spalart-Allmaras (S-A) one-equation model.23, 24 The DES model available in US3D was previously
shown in Reference [13] to accurately model the turbulent Reentry-F base flow (ReD > 107 ). To understand
how the DES formulation functions in a separated flow, consider first the Spalart-Allmaras RANS model
(including the compressibility corrections of Catris and Aupoix25 ):
   2
Dρν̃ 1 √ √ √ ν̃
= ∇ · (µ∇ν̃ + ρν̃∇ ρν̃) + cb2 (∇ ρν̃)2 + cb1 ρS̃ ν̃ − cw1 fw ρ (1)
Dt σ d

where ρ is the local fluid density, ν̃ is related to the eddy viscosity, S̃ is a measure of vorticity, and d is
the distance to the nearest wall. Definitions for the remaining variables can be found in the previously
cited references. Note that the destruction term scales as d−2 . Therefore, with the model active far from
solid surfaces, there is very little destruction of turbulent viscosity, and high levels of ν̃ persist. The DES
formulation replaces the length scale of a RANS model with a new one, such that the model behaves as
a Smagorinsky-like subgrid scale model in regions where the flow is separated. Thus the new length scale
acts as a filter, allowing the simulation to resolve eddies down to the grid scale while modeling those that
are smaller. For the particular case of the S-A model, this is accomplished in a straightforward manner by
replacing d with a new variable that depends on some measure of the local grid cell size, ∆:

d˜ = min (d, CDES ∆) (2)

With this definition, the modified equation is equivalent to the unchanged S-A RANS model near walls.
Away from walls, if the filter width (i.e. cell size) is chosen to lie in the inertial subrange, then turbulent
production and destruction should roughly balance each other. Under this assumption, it is easy to show

7 of 21

American Institute of Aeronautics and Astronautics

that the S-A model reduces to the familiar Smagorinsky form
cb1 2
ν̃ = (CDES ∆) S̃ (S-A DES)
cw1 fw
2
νt = (Cs ∆) S (Smagorinsky)
CDES is an adjustable parameter, conventionally set to 0.65. It should be noted that the DES approach is
not limited to the S-A model, but has been extended to other popular two-equation formulations as well (see
Reference [26]). In this paper, we also examine for comparison the traditional RANS formulation of the S-A
model represented by Equation (1) (available in US3D and DPLR), the algebraic Baldwin-Lomax model27
(DPLR), and a variant of the commonly used two-equation SST model (DPLR).

IV. Numerical Results

A. Flow Structure
Flow visualizations are shown in Figures 4 and 5: the former giving a three-dimensional view of the flow-
field with streamlines and spanwise temperature contours; the latter showing temperature contours in the
symmetry plane. Aside from the higher freestream enthalpy apparent in the forebody contours of Run 4,
all of these runs share essentially similar flow features. The flow along the windside aftbody is aligned with
the freestream and therefore remains steady and attached. Depending on the freestream Reynolds number,
the boundary layer in this region may transition to turbulence. The boundary layer then separates at the
base of the capsule, creating a small recirculating region that interacts with the sting mount. As we move
circumferentially about the capsule toward the leeside, the flow gradually begins to separate until we are in
the wake region. At this point, the flow is fully separated and we see a marked increase in unsteadiness. All
cases are characterized by the shedding of vortices from the aftbody, although the range of apparent length
scales is greater in the turbulent cases as expected. Furthermore, a secondary interaction exists between the
larger separated capsule flow and that which emanates from behind the sting.

B. Low Reynolds Number, Run 4

We first examine the capabilities of our method in a nominally laminar flow as a means of isolating purely
numerical errors from those due to turbulence modeling. Reference [11] demonstrated a reasonable ability
to predict laminar afterbody heating on the Apollo AS-202 command module using DPLR without resolving
the temporal evolution of the flow. All trajectory points considered in that study had Reynolds numbers
based on body diameter less than 8.0 × 105 . From Table 2, the Reynolds number of Run 4 based on a
10 inch (0.254 m) model diameter is 6.5 × 105 ; thus, at a minimum, we should be able to reproduce their
demonstrated accuracy.
Figure 6 depicts comparisons between experimental and numerical measurements of time-averaged pres-
sure and heat transfer rate for Run 4 at all gauge locations. We remark here that not all gauges were operable
during each run. Gauges P8, P9, T12, and T18 - T20 (as referenced in [18]) were all removed in order to
accommodate mounting of the capsule to the round center-mounted sting. Thus, as stated earlier, the gauges
have been renumbered in a consecutive manner, ignoring the absent gauges for ease of presentation.
In this numbering scheme, the forebody instrumentation is represented by pressure gauges [1 - 3, 12
- 15] and heat transfer gauges [36 - 41]. Interestingly, agreement between the laminar computation and
experiment is uniformly excellent except for the forebody pressure. In fact, the experimental measurements
for these gauges were independently deemed anomalous prior to conducting the computational phase of this
study. The reasons for the discrepancy remain unclear although it has been speculated that it is perhaps due
to calibration error.a That the computational results are unable to adequately predict what is typically a
very routine measurement supports this conjecture. Nevertheless, all of the remaining gauges compare quite
favorably, always within a few percent of the experimental value.
Also seen in Figure 6 are the results of two separate turbulent simulations which were run as a check
against the laminar flow assumption at this Reynolds number. While the pressure field is largely unaffected,
as expected, we see that the predicted heat flux is uniformly over-predicted by both Baldwin-Lomax and S-A
RANS models. Hence our results corroborate the conclusion of Reference [11] that a laminar flow assumption
is valid at the Reynolds numbers considered.
a MacLean, private communication

8 of 21

American Institute of Aeronautics and Astronautics

Figure 4. Three-dimensional flow visualization (Run 4). Spanwise contour planes are colored by temperature.

(a) Run 2 (b) Run 3 (c) Run 4

Figure 5. Flow visualization of temperature contours in the symmetry plane.

9 of 21

American Institute of Aeronautics and Astronautics

 (a) (b)

Figure 6. Comparison of experimental and numerical measurements for Run 4. (a) Pressure, (b) Heat transfer
rate.

C. Medium Reynolds Number, Run 3

We next turn our attention toward modeling of transitional and turbulent flows. For a transitional flow, we’ve
simply selected Run 3 by virtue of it having the next highest Reynolds number, one which also coincides
with previous experience simulating the transitional/turbulent flow of the Reentry-F vehicle. For this case,
turbulence is modeled exclusively using the S-A DES model.
Because the flow is transitional, we can anticipate that it will be difficult to accurately capture the
experiment in the absence of some relatively sophisticated transition model. We therefore adopted the
approach to simulate the experiment as both a laminar and fully turbulent flow with the hope that the
results will bound the data. This is reflected in Figure 7, which shows time-averaged results in the same
manner as for Run 4.
Pressure again is mostly unaffected by the turbulence model. In both simulations, we obtain uniformly
excellent agreement with the experiments. There is some slight variation for pressure gauges located in the
separated flow, however the variation is only a fraction of a percent relative to the forebody pressure. The
variation can therefore be expected to be a very small contributor to the aerodynamic forces. This will
be discussed further in a later section comparing RANS and DES simulations. Heat transfer, on the other
hand, is considerably more erratic. First, we note the large differences between the experimentally measured
forebody values and those predicted by the simulations. The laminar model here is clearly inappropriate,
underpredicting the peak heating value by nearly 60%. Still, one cannot say that the turbulent simulation
performs perfectly either, particularly in the stagnation region between the nose and windside shoulder.
This kind of behavior has been noted in several prior studies dealing with high Reynolds number forebody
flows. An explanation for the disagreement is not currently available, however it is apparent in all of our
simulations for which the forebody exhibited characteristics of turbulent behavior.
Otherwise, the aftbody simulations perform as expected. That is, the laminar solution uniformly un-
derpredicts the experimental heat transfer values while the fully turbulent solution uniformly overpredicts
them. This is best exemplified by the windside gauges [47 - 51] along the φ = 180◦ ray. Here we see that
the laminar profile maintains an essentially flat character while the turbulent profile immediately peaks near
the capsule shoulder (gauge 51) and then gradually decreases toward the base (gauge 47). In contrast, we
see a slow ramping up of turbulent heating from the shoulder to the base in the experiment. By the time
the flow reaches gauge 47, the comparison between experiment and the fully turbulent simulation is actually
quite good. This suggests that the flow is indeed transitional and that an accurate simulation of this flow
would be possible with the incorporation of a suitable transition model.

10 of 21

American Institute of Aeronautics and Astronautics

 140000 25
Experiment Experiment
Laminar Sim Laminar Sim
120000 Turbulent Sim Turbulent Sim
20

Heat Transfer Rate (W/cm )

2
100000
Pressure (Pa)

80000 15

60000
10
40000

20000 5

0
0
0 10 20 30 40 50 60 70
Gauge Number Gauge Number

(a) (b)

Figure 7. Comparison of experimental and numerical measurements for Run 3. (a) Pressure, (b) Heat transfer
rate.

D. High Reynolds Number, Run 2

Let us now consider the high Reynolds number case of Run 2. All simulations at this condition were assumed
to be fully turbulent - in this section we consider only the S-A DES model. As with Runs 3 and 4, the time-
averaged results are presented in Figure 8 for the entire range of gauges. Again, agreement of the pressure
profiles is uniformly excellent. The apparent disagreement at gauges [8 - 11] is due to these gauges being
inoperable during the test. With regard to heat transfer rate, we see much improved agreement over the
transitional results at nearly every gauge, often within the measured standard deviation of the flowfield.
Yet there are a couple of problematic regions. The difficulty of predicting stagnation point heating at high
Reynolds number is once again evident. The windside boundary layer also has a slower transition relative
to the computation. Nevertheless, these errors seem to be relatively localized and the aftbody gauges are
unaffected.
It is instructive to break down the presentation into the constituent rays of gauges in order to gain some
insight into the spatial dependence of our computational accuracy. This is shown in Figures 9 and 10, which
depict pressure and heat transfer comparisons for each ray of instrumentation. Here the computational
results are denoted with a solid line and the experiment by symbols. The bars reflect the measured standard
deviation of the flow at each location and therefore provide an indication of the variation and unsteadiness
of the local flowfield. As the x-coordinate increases, we are moving from the shoulder toward the base;
corresponding gauge numbers are shown for the first and final gauges in each ray.
Principally, as is evident from Figure 8 as well, we see very good agreement for both pressure and heat
transfer. For the most part, the computation lies within one standard deviation of the experiment, and when
even that fails the disagreement is often within a few percent. There are however a couple of points which
do stand out. First, with pressure we see significant disagreement at gauges 25, 31, and 35 (corresponding to
P27, P33, and P37). Each of these gauges is spatially surrounded by other gauges that perform well in the
comparison, which raises the possibility that they are outliers, yet there is no indication from the experiment
of malfunction. One can see however that both gauges 25 and 31 are located near the junction of the conical
afterbody and flat base. The inability to accurately predict these locations may well be due to the sharp
flow separation and interaction between the aftbody flow and the flow behind the sting. It is worth noting
that, irrespective of the cause, the effects are localized to these gauges. Gauge 35 has indications of being an
outlier due to its rather benign location in the windside acreage and the fact that it is surrounded by three
other gauges which compare favorably with the experiment.
The most notable aspect of the heat transfer measurements is that the computation tends to overpredict
results near separation points. This is particularly true for locations in which the flow is transitioning

11 of 21

American Institute of Aeronautics and Astronautics

 35
Experiment Experiment
200000 Simulation Simulation
30

Heat Transfer Rate (W/cm )

2
25
150000
Pressure (Pa)

20

100000
15

10
50000

0
0
0 10 20 30 40 50 60 70
Gauge Number Gauge Number

(a) (b)

Figure 8. Comparison of experimental and numerical measurements for Run 2. (a) Pressure, (b) Heat transfer
rate.

from the windside to the leeside, see for instance the φ = 90◦ ray. This seems to indicate an inability
of the computational models to accommodate a rapid expansion of the flowfield, not unlike the frequently
seen heat transfer spikes at the shoulders of capsule geometries. It is also indicative of the grid possibly
being under-resolved near the separation points. Again, we note that this phenomenon is almost completely
localized such that it has little or no impact on the gas subsequent to separation.
Finally, we may gain further insight into our numerical performance for this particular case by looking
closer at the detailed transients of a few selected gauges. While we cannot expect to perfectly replicate
transient phenomena in our simulations, we can reasonably compare frequencies and amplitudes of oscillation.
For this purpose, we have run our simulations over a time period comparable to the flow time achieved in
the experiment and extracted transient data along the φ = 0◦ ray.
The resulting comparisons for gauges [42 - 46] are shown in Figure 11. Note that increasing gauge number
indicates movement from the shoulder toward the base of the capsule. Gauge 42 therefore reflects the state of
the flow just subsequent to the strong expansion off the leeside shoulder. Here we see that the ability of the
computation to correctly capture the amplitude and frequency of the flow is severely degraded. Interestingly,
this does not detrimentally affect the time-averaged result, although this is perhaps only a fortuitous result.
As we move away from the shoulder toward the base, we see a steady improvement in the simulation’s
prediction of oscillation frequencies and amplitudes. There is however a slight bias toward overprediction
near the base, a fact which is apparent in the time-averaged results of Figure 10. These results suggest
that an additional local refinement of the computational mesh near the shoulder separation and wake/sting
interaction may be necessary in order to accurately capture transient behavior in these regions.

E. DES v. RANS
Comparisons are shown in Figure 12 of DES and RANS simulations for the Run 2 condition. We’ve restricted
the plots to show only afterbody gauges in order to draw out differences between models more clearly. As
noted earlier, there is relatively small variation in computed pressure profiles - for this case, the largest
difference occurs near gauges 26 and 27 and amounts to approximately 0.5% of the forebody stagnation
pressure. SST-RANS and SA-DES produce the best results, while Baldwin-Lomax and SA-RANS tend
to over- and under-predict the experiment, respectively. Nevertheless, relative to the dominant forebody
pressure, each model produces fair or better results. Differences in predicted heat transfer between the
models are more pronounced, with variations as large as 7% (gauge 45) of the experimentally measured
stagnation value. One can actually deduce from Figure 12(b) whether the gauges are located in attached
or separated flow regions by comparing the S-A DES and S-A RANS curves. In attached regions, S-A DES

12 of 21

American Institute of Aeronautics and Astronautics

 WINDWARD(180O)
LEEWARD(0O)

Figure 9. Comparison of experimental and numerical pressure measurements along instrument rays for Run
2. Symbols denote experiment, solid lines denote simulation (SA-DES).

13 of 21

American Institute of Aeronautics and Astronautics

 WINDWARD(180O)
LEEWARD(0O)

Figure 10. Comparison of experimental and numerical heat transfer rates along instrument rays for Run 2.
Symbols denote experiment, solid lines denote simulation (SA-DES).

14 of 21

American Institute of Aeronautics and Astronautics

 U
43

42
44

46
45

Figure 11. Comparison of CFD to experimental transient heat transfer along φ = 0◦ ray for Run 2.

15 of 21

American Institute of Aeronautics and Astronautics

 (a) (b)

Figure 12. Comparison of numerical measurements along the afterbody for Run 2. (a) Pressure, (b) Heat
flux.

and S-A RANS give nearly identical results, a consequence of the RANS formulation being a limiting case
of DES under such conditions. For gauges located in separation regions, however, the two models diverge.
Significantly, we see that the predictive accuracy of the S-A RANS model is degraded relative to S-A DES,
even on an identical, highly-refined mesh.
On a per-gauge basis, there is no discernible pattern of over- or under-prediction in the S-A RANS
model. On the other hand, SST-RANS almost uniformly under-predicts the heat transfer rate. In both
cases, the excessive dissipation introduced by the RANS model essentially creates an ‘alternative’ flow
structure. Separation and reattachment lines are moved, and convection streamlines are dramatically altered
to accommodate the more ‘viscous’ flow. For the capsule case (and indeed for all cases investigated during the
course of our research), the result is that the RANS models tend to produce a much smoother flowfield with
little local fluctuation, while the DES model exhibits a much greater degree of disorder and unsteadiness.
Analytically, this is precisely what we expect to occur because, for a sufficiently refined grid, the DES
simulation essentially amounts to transferring the bulk of modeling work from the approximate turbulence
model to the exact Navier-Stokes equations. In other words, the success of DES in our simulations stems
not so much from the model producing a more ‘correct’ value of eddy viscosity, but merely by the fact that
it interferes minimally with the fidelity that Navier-Stokes already provides. (This is not to suggest that the
particulars of the subgrid model are trivial; they certainly are not.)
Interestingly, the heat transfer results from the Baldwin-Lomax model would appear to contradict our
generalization of RANS models. With the obvious exception of gauge 45, the Baldwin-Lomax model does
a surprisingly good job of reproducing the experiment. This is puzzling because the model is generally
regarded as appropriate only for attached boundary layers. To understand our result, in Figure 13 we have
plotted the ratio of eddy viscosity to molecular viscosity found in the symmetry planes of the SST and
Baldwin-Lomax simulations. Here we see that the SST model is working as described: an enormous amount
of dissipation is being added throughout the wake. In contrast, the dissipation of the Baldwin-Lomax model
is rather small - more importantly, dissipation is clearly confined to the grid blocks attached to the surface.
In fact, Figure 13 demonstrates that the seemingly good performance of the Baldwin-Lomax model is simply
a result of the DPLR implementation which, for this specific block topology, prevents the model from being
active except in very limited regions near the body. A different block topology in DPLR would conceivably
result in a very different answer. The model is therefore artificially behaving as a sort of wall model, not
unlike the function of the DES formulation, which reinforces our previous assertion that excessive values of
eddy viscosity are the largest obstacle to overcome in a simulation when the grid is sufficiently refined. We
emphasize again that this behavior should not be considered a general feature of the Baldwin-Lomax model,
but is particular to its implementation in DPLR.

16 of 21

American Institute of Aeronautics and Astronautics

Figure 13. Contour plot of eddy-to-molecular viscosity ratio in the symmetry plane for SST and Baldwin-
Lomax simulations.

Thus, the capsule surface is subjected to different levels and distribution of heating depending on which
model one chooses. This observation is quantitatively verified in Figure 14 which depicts the simulated
transient heat fluxes at gauge 45. In this figure, we have started the SA-DES and SA-RANS simulations
from an identical initial condition and allowed them to evolve over a period of time while keeping the CFL
number fixed. Note that the time scale has been truncated relative to Figure 11 in order to draw out
distinctions in the resolved heat flux measurements.
First, after the initial ∼0.5 ms transient has passed, during which time the flow is adjusting to the models
being switched on, it is apparent that the SA-RANS simulation has a greatly diminished ability to resolve
flow unsteadiness. Looking at the standard deviation (using the full time trace data as shown in Figure 11),
we find that the magnitude relative to the experimentally measured mean is 52% for SA-DES and only
0.64% for SA-RANS. Thus the amplitude of fluctuation is reduced by nearly two orders of magnitude. The
SA-DES simulation, on the other hand, compares quite favorably to the experiment.
Second, the SA-RANS simulation greatly overpredicts the experiment across nearly the entire flow time
and so it fails to achieve even the correct mean heat flux. We compute the relative error of the mean heat
flux for this case to be 30% and 156% for SA-DES and SA-RANS, respectively. The computed statistics for
the experiment and both simulations at gauge 45 are summarized in Table 3. By integrating over all of the
gauges located in the wake, we obtain total error estimates of 36% and 57% for SA-DES and SA-RANS,
respectively. The reduction in total error for the SA-RANS case is chiefly due to the inclusion of gauges
located near the shoulder separation (like gauges 42 and 43 in Figure 11) where flow unsteadiness is fairly
weak. In light of RANS being the most favored modeling approach employed in today’s production CFD
codes, we can surmise that the use of more sophisticated methods like DES, if applied during the design
phase of an actual flight vehicle, could crucially impact development of an afterbody thermal protection
system.

17 of 21

American Institute of Aeronautics and Astronautics

 4
Experiment
RANS
DES
Heat Transfer Rate (W/cm )
2
3

0 0.5 1 1.5 2 2.5

Time (ms)

Figure 14. Heat flux transients from gauge 45 of Run 2 simulated with SA-RANS and SA-DES.

Table 3. Computed statistics for gauge 45 of Run 2.

µ σ µ/µexp σ/µexp
2
(W/cm )
Experiment 1.1178 0.5962 1.0000 0.5334
SA-DES 1.4509 0.5855 1.2980 0.5238
SA-RANS 2.8601 0.0072 2.5587 0.0064

18 of 21

American Institute of Aeronautics and Astronautics

 2
Experiment
DES
Low-Dissipation

Heat Transfer Rate (W/cm )

2
1.5

0.5

0
0 0.5 1 1.5 2 2.5
Time (ms)

Figure 15. Heat flux transients from gauge 42 of Run 2 simulated with the original Steger-Warming scheme
and the low-dissipation scheme of Subbareddy and Candler.

F. Flux Dissipation and Time-Accuracy

Finally, let us emphasize that nearly all of the preceding analysis has been predicated on comparisons with
statistical data, i.e. mean, RMS, and standard deviation. By doing so, we necessarily circumvent notions of
time-accuracy. It is clear, however, that accuracy in a statistical sense does not imply transient accuracy.
As an example, one could model the function f (t) = sin t as f˜(t) = sin 10t or f˜(t) = 10 sin t and still obtain
the correct mean value of 0, while incorrectly predicting by factors of 10 the transient period and amplitude,
respectively. We noted this phenomenon in Figure 11, where gauge 42 of the simulation clearly does not
capture the same amplitude of fluctuation as seen in the experiment, yet it ultimately yields a reasonable
estimate of mean heat flux.
Based on the results presented in the preceding sections of this paper as well as Ref. [13], it can be argued
that, with regard to our stated applications of interest (external, reentry-type flows), no further improvement
on the numerical scheme’s accuracy is necessary in order to produce accurate statistical results. That is to
say, for these cases, a traditional first order in time, second order in space discretization is sufficient when
coupled with an adequate turbulence model and good CFD practices (grid refinement studies, time-step
limiting). However, this may not hold true for cases which require that the transient processes be resolved.
To illustrate how the apparent limitations displayed in Figure 11 might be mitigated, we ran the same case
incorporating the kinetic energy conistent, low-dissipation flux scheme discussed in Section III. The results
are depicted in Figure 15. Recall for this case that the Steger-Warming scheme failed to capture correctly
the amplitude and frequency of the transient heat flux. By comparison, Subbareddy’s low-dissipation scheme
does a remarkable job of recovering amplitudes comparable to those seen in the experiment. By eliminating
unnecessary numerical dissipation, the scheme also allows us to take advantage of the simulation’s temporal
resolution. It is thus capable of resolving transient flow features on a smaller scale than even that captured
in the experiment. While this does not conclusively demonstrate the performance benefits of the scheme, it
does show how simulations are critically impacted by the choice of numerics, notwithstanding their level of
physical modeling fidelity.

V. Conclusions
In this paper, we examined the spherical capsule experiments of Ref. [18] in order to (1) obviate concerns
that were raised concerning ambiguity of the Reentry-F flight data, and (2) extend our analysis to realistic

19 of 21

American Institute of Aeronautics and Astronautics

reentry vehicle geometries. Three test runs were chosen for analysis, corresponding to nominally laminar,
transitional, and turbulent conditions.
Assuming laminar flow, the low Reynolds number case (Run 4) displayed uniformly excellent agreement
with the experiment, with the exception of the forebody pressure gauges. This was attributed to a possible
calibration error during testing. The medium Reynolds number case (Run 3) posed some difficulty for our
simulation procedure as it appeared transitional and there was inadequate knowledge of the boundary layer
transition front available from the experiment. The case was instead run twice, fully laminar and fully
turbulent. The two simulations bounded the experiment nicely, suggesting that inclusion of an appropriate
transition model would be sufficient to adequately simulate the flow. In the high Reynolds number case (Run
2), good agreement was obtained using the SA-DES model. A more detailed analysis was then presented
to shed light on the relationship between flow structure and DES performance. In particular, the model’s
performance was found to deteriorate in the vicinity of strong expansions, such as those seen near the vehicle
shoulder and base. This limitation was also found to be fairly localized and had little discernible impact
on gauges located downstream of the separation. Furthermore, the DES simulation showed exceptional
improvement over the traditional RANS models. Total error integrated over the separated flow gauges was
reduced from 57% for SA-RANS to 36% for SA-DES. Inspection of the heat flux transients showed that the
SA-RANS model nearly eliminated the inherent unsteadiness of the flow while the SA-DES model compared
favorably with experiment. Finally, we examined the effects of numerics on the solutions by incorporating
a low dissipation flux scheme and second order time accuracy. Their combined effect was to qualitatively
enhance the simulation’s ability to resolve transient characteristics of the flow.
This study demonstrated that currently available computational methods are capable of accurately simu-
lating unsteady, separated flows at high Mach numbers. Specifically, for the external, reentry flows considered
here, a standard first order in time, second order in space Steger-Warming scheme was numerically sufficient
to accurately model each of the test cases. In general, the most critical barrier to overcome in a simulation is
dissipation, whether physical or numerical. From a physical standpoint, it is easily demonstrated (and fairly
well-documented in the literature) that classic RANS models produce excessive turbulent dissipation which
then result in unrealistic wake structure and, therefore, unrealistic predictions of base pressure and heat flux.
The DES model provides a clear improvement in this regard; additionally, more recent versions of DES have
been proposed (see Ref. [28]) which may improve agreement further. The effects of numerical dissipation,
on the other hand, are generally felt at the discretization scale and act to damp out small-scale fluctuations
which could adversely affect applications in which transient resolution of the flow is crucial. It is therefore
recommended that future studies on problems outside the scope of interest here continue to investigate the
influence of time accuracy and flux dissipation. Improvements in these areas may prove critical for more
demanding applications. We believe the dual-time and low-dissipation schemes discussed in Section III to
be very promising candidates for enhancing the effectiveness of our approach.

Acknowledgments
This research is supported by the NASA Fundamental Aeronautics Program under grant NNX07AC28A
with Dr. Seokkwan Yoon as Program Manager. This research is also supported by the Air Force Office
of Scientific Research (AFOSR) under Grant No. FA9550-04-1-0341. The views and conclusions contained
herein are those of the authors and should not be interpreted as necessarily representing the official policies
or endorsements, either expressed or implied, of the AFOSR or the U.S. Government. Computer time was
provided by the University of Minnesota Supercomputing Institute.

References
1 Pallone, A., Erdos, J., Eckerman, J., and McKay, W., “Hypersonic Laminar Wakes and Transition Studies,” AIAA Paper

1963-171 , 1963.
2 Lien, H., Erdos, J., and Pallone, A., “Nonequilibrium Wakes with Laminar and Turbulent Transport,” AIAA Paper

1963-447 , 1963.
3 Zeiberg, S. L. and Bleich, G. D., “Finite-Difference Calculation of Hypersonic Wakes,” AIAA Journal, Vol. 2, No. 8,

1964, pp. 1396–1402.

4 Gnoffo, P. A., “A Code Calibration Program in Support of the Aeroassist Flight Experiment,” Journal of Spacecraft and

Rockets, Vol. 27, No. 2, 1990, pp. 131–142.

5 Gnoffo, P. A., Price, J. M., and Braun, R. D., “On the Computation of Near Wake, Aerobrake Flowfields,” AIAA Paper

1991-1371 , 1991.

20 of 21

American Institute of Aeronautics and Astronautics

 6 Venkatapathy,
E., Palmer, G., and Prabhu, D. K., “AFE Base Flow Computations,” AIAA Paper 1991-1372 , 1991.
7 Conti,
R. J. and MacCormack, R. W., “Navier-Stokes Computation of Hypersonic Near Wakes with Foreign Gas Injec-
tion,” AIAA Paper 1992-0838 , 1992.
8 Horvath, T. J., McGinley, C. B., and Hannemann, K., “Blunt Body Near-Wake Flow Field at Mach 6,” AIAA Paper

1996-1935 , 1996.
9 Horvath, T. J. and Hannemann, K., “Blunt Body Near-Wake Flow Field at Mach 10,” AIAA Paper 1997-0986 , 1997.
10 Holden, M. S., Harvey, J. K., Boyd, I. D., and Horvath, T. J., “Experimental and Computational Studies of the Flow

Over a Sting Mounted Planetary Probe Configuration,” AIAA Paper 1997-0768 , 1997.
11 Wright, M. J., Prabhu, D. K., and Martinez, E. R., “Analysis of Afterbody Heating Rates on the Apollo Command

Modules, Part 1: AS-202,” Journal of Thermophysics and Heat Transfer , Vol. 20, No. 1, 2006, pp. 16–30.
12 Edquist, K. T., Wright, M. J., and Allen, G. A., “Viking Afterbody Heating Computations and Comparisons to Flight

Data,” AIAA Paper 2006-0386 , 2006.

13 Barnhardt, M. D. and Candler, G. V., “Detached Eddy Simulation of the Reentry-F Flight Experiment,” AIAA Journal,

under review.
14 Holden, M. S. and Parker, R. P., “LENS Hypervelocity Tunnels and Application to Vehicle Testing at Duplicated Flight

Conditions,” Advanced Hypersonic Test Facilities, edited by F. K. Lu and D. E. Marren, Vol. 198, chap. 4, American Institute
of Aeronautics and Astronautics, Reston, VA, 2002.
15 Holden, M. S., Wadhams, T. P., and Candler, G. V., “Experimental Studies in the LENS Shock Tunnel and Expansion

Tunnel to Examine Real-Gas Effects in Hypervelocity Flows,” AIAA Paper 2004-0916 , 2004.
16 Holden, M., Wadhams, T., and MacLean, M., “Experimental Studies in the LENS Supersonic and Hypersonic Tunnels

for Hypervelocity Vehicle Performance and Code Validation,” AIAA Paper 2008-2505 , 2008.
17 MacLean, M., Mundy, E., Wadhams, T., Holden, M., Barnhardt, M. D., and Candler, G. V., “Experimental and Numerical

Study of Laminar and Turbulent Base Flow on a Spherical Capsule,” AIAA Paper 2009-0783 , 2009.
18 MacLean, M., Mundy, E., and Parker, R., “Year 1 Summary Report for Study of Wake Flows on Capsule Bodies,” Tech.

rep., CUBRC, 2008.

19 Nompelis, I. N., Drayna, T., and Candler, G. V., “Development of a Hybrid Unstructured Implicit Solver for the

Simulation of Reacting Flows Over Complex Geometries,” AIAA Paper 2004-2227 , 2004.
20 Nompelis, I. N., Drayna, T., and Candler, G. V., “A Parallel Unstructured Implicit Solver for Hypersonic Reacting Flow

Simulations,” AIAA Paper 2005-4867 , 2005.

21 Wright, M. J., White, T. R., and Mangini, N., “Data Parallel Line Relaxation (DPLR) Code User Manual: Acadia -

Version 4.01.1,” Tech. rep., NASA TM 2009-215388, 2009.

22 Subbareddy, P. and Candler, G. V., “A Fully Discrete, Kinetic Energy Consistent Finite-Volume Scheme for Compressible

Flows,” Journal of Computational Physics, Vol. 228, 2009, pp. 1347–1364.

23 Spalart, P. R. and Allmaras, S. R., “A One-Equation Turbulence Model for Aerodynamic Flows,” AIAA Paper 1992-0439 ,

1992.
24 Spalart, P. R., Jou, W.-H., Strelets, M., and Allmaras, S. R., “Comments on the Feasibility of LES for Wings and on a

Hybrid RANS/LES Approach,” AFOSR International Conference on DNS/LES , 1997.

25 Catris, S. and Aupoix, B., “Density Corrections for Turbulence Models,” Aerospace Science and Technology, Vol. 4, 2000,

pp. 1–11.
26 Strelets, M., “Detached Eddy Simulation of Massively Separated Flows,” AIAA Paper 2001-0879 , 2001.
27 Baldwin, B. S. and Lomax, H., “Thin Layer Approximation and Algebraic Model for Separated Turbulent Flows,” AIAA

Paper 1978-0257 , 1978.

28 Spalart, P. R., Deck, S., Shur, M. L., Squires, K. D., Strelets, M., and Travin, A., “A New Version of Detached-Eddy

Simulation, Resistant to Ambiguous Grid Densities,” Theoretical and Computational Fluid Dynamics, Vol. 20, No. 3, 2006,
pp. 181–195.

21 of 21

American Institute of Aeronautics and Astronautics

View publication stats