AIAA SciTech Forum 10.2514/6.

2020-2123
6-10 January 2020, Orlando, FL
AIAA Scitech 2020 Forum

CFD simulations of super/hypersonic

missiles: validation, sensitivity analysis, and
improved design
Vinod Singh Rao1
ANSYS India, Pune, Maharashtra, 411057, India
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Valerio Viti2 and Juan Abanto3

ANSYS Inc., 10 Cavendish Ct., Lebanon, NH 03766, USA

Abstract

The present work focuses on the qualitative and quantitative validation of the computational fluid dynamics
(CFD) solutions using two high-speed vehicle geometries: the high-supersonic NASA Tandem Canard Missile
(TCM) and the aerospike-nosed generic missile geometry (Aerospike). For the NASA TCM case, CFD
simulations were performed for an angle of attack (AOA) sweep from -5deg to +18deg for two Mach numbers,
2.5 and 3.5 for a fully turbulent flow. Aerodynamic coefficients were validated with available test data. A
qualitative analysis of the flow structures is performed for the cases of 0- and 18-degree AOA. For the second
case, the Aerospike, the fully turbulent flow field around the geometry is studied for two AOAs, 0-degree and
+10-degree, both cases at a Mach number of 6.0. The pressure results on the nosecone are validated against
available test data. The Aerospike results are analyzed to understand changes in flow field due to change in
AOA. The results are then utilized to validate the overset approach for the Aerospike flowfield and it is shown
how overset technology can be used to effectively perform a geometric design study of the vehicle. Lastly, a
modified aerospike geometry is proposed based on a sensitivity analysis, to reduce the overall drag on the
missile while maintaining the beneficial effects created by the aerospike geometry.

I. Nomenclature
To = Freestream total temperature (K)
T∞ = Freestream static temperature (K)
P∞ = Freestream static pressure (Pa)
TI = Turbulence intensity
TVR = Turbulence viscosity ratio
AOA = Angle of attack (degree)
CA = Force coefficient in the axial direction (with experimental correction)
CA* = Force coefficient in the axial direction (without any correction)
CN = Force coefficient in the normal direction
CM = Pitching moment coefficient

II. Introduction
A resurgence of interest in the development of high-supersonic and hypersonic vehicles has been witnessed in the past
few years; this interest is driven primarily, but not exclusively, by military applications [1][2]. Hypersonic vehicles

1
Senior Application Engineer
2
Aerospace and Defense Lead, Senior Member AIAA
3
Senior Technical Support Engineer

$copyRight
 offer several advantages over lower-speed counterparts, the main advantage being a significantly reduced response-
or strike-time, depending on the application [3]. Hypersonic vehicles can be roughly classified into three broad
categories: cruise, boost-glide, and re-entry. No sharp demarcation lines exist between these categories, but each one
is associated with diverse and complex flow phenomena that drive unique design challenges. A thorough
understanding of the complex flow structures around these high-speed vehicles, including shocks, shock-boundary-
layer interaction, thin shock-layers, separated flows, etc is critical for the development of a successful design. Vehicle
design has traditionally been performed using empirical/semi-empirical methods for aerothermo-dynamic
characterization. Extensive wind-tunnel testing is used to generate aerodynamic databases and to validate the
theoretical approach. However these test facilities are often limited in the flight regimes they can reproduce, often
being able to match one or two flight parameters, but not all of them. Flight testing is still the golden standard for
realistic data collection, but it is extremely costly, single tests often running in the multi-million dollar budgets with
program development times spanning multiple years. In this scenario, high-fidelity physics-based engineering
simulations play an important role for the development and design of hypersonic vehicles by introducing greater
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

engineering and financial efficiency. Additionally, these methods are also useful to study extreme flow conditions
such as high-speed maneuvers, high angle of attack (AOA) situations and any flight regime close to the limits of the
safety envelope where testing may be riskier.

The present effort builds on previous experimental and simulation work and is aimed at advancing the current
acceptance of CFD methods for the design and analysis of hypersonic vehicles; this is obtained by extending the
already ample body of validation work for two canonical cases, the NASA TCM geometry and the aerospike-nosed
missile using a commercial CFD software.

III. Test cases and numerical approach

The two geometries selected for the study, the NASA Tandem Control Missile (TCM) and the Aerospike, are
representative of vehicles that fly at high-supersonic/low-hypersonic and were chosen because of the complex flow
field that surrounds them at high Mach numbers.

All the numerical simulations in the present work were run using the commercial CFD code ANSYS FLUENT (Ref.
[1]). FLUENT is a widely used finite-volume code that solves the Reynolds-Averaged Navier-Stokes (RANS)
equations. The FLUENT solver offers the option to solve the RANS using either a pressure-based implicit solver
(PBNS) [6] [7] or a density-based implicit or explicit solver (DBNS) [8] [9], either in steady-state or unsteady mode.
ANSYS FLUENT is capable of handling both structured and unstructured (Tetrahedral, Cutcell, Polyhedral, Poly-
Hexcore, etc.) mesh topology. A full description of the FLUENT solver and its capabilities are provided in Ref [1][5].
In the present work, results are presented for both the PBNS and the DBNS solver for both geometries. All results are
fully converged to a steady-state solution and are second-order (upwind) accurate for the convective fluxes and
second-order (central) accurate for the diffusive fluxes. The diffusion fluxes were computed using a Least-square
formulation.

A. NASA Tandem Canard Missile

The generic NASA Tandem Control Missile (TCM) was extensively studied experimentally by Blair et al. (Ref. [10]).
It represents a typical high-supersonic/low-hypersonic, cruciform, canard-controlled air-to-air missile (AAM)
configuration. TCM is broadly based on the Sidewinder missile with variations in the canard aspect ratio and different
tail fins. Fig. 1 shows the geometric details of the configuration studied in the present work. Various configurations
of TCM have been studied experimentally and using analytical methods over the years (Ref. [10]). Recently, Akgul
et al. (Ref. [13]) did numerical validation of the TCM Model at low supersonic speed (Mach 1.75) using the ANSYS
FLUENT solver. Results show good agreement with the available test data over a range of AOA conditions. The
present study focuses on validating the numerical approach of modeling high supersonic (Mach 2.5 and 3.5) flow
using ANSYS FLUENT. Numerical simulation of the missile installed in a wind tunnel with an AOA sweep is done
to test the predictability of the model, especially at higher AOAs. Fig. 2 shows the computational flow domain with
the missile installed in the wind tunnel test section. Inlet is modeled as pressure inlet with freestream conditions as
outlined in Table 1, and the outlet is modeled a pressure outlet boundary type. Wind tunnel walls and missile surfaces
are modeled as an adiabatic no-slip wall. Mesh is created using FLUENT-Meshing’s watertight workflow with
polyhedral elements. The total mesh count for the domain is about 7.7 million, with a maximum aspect ratio of 200

2
 and a minimum orthogonal quality of 0.1. Fifteen boundary layer wedges are created on the missile surfaces to capture
near-wall effects accurately. The average wall Yplus value on the missile surface is about 5. Fig. 3 shows the surface
mesh resolution at various locations, including the nose, canard, tail, and center section of the flow domain.
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 1 Geometry Details (Ref. [6])

Fig. 2 Flow Domain

3
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 3 Mesh Resolution: Surface, Canard, Tail, Center Section

A time-independent steady-state approach is used to simulate compressible flow over the missile geometry. Pressure
based coupled solver is primarily used to solve the discrete system of equations with second-order spatial discretization
for gradient calculations. Menter’s SST K-Omega model is used to capture the turbulence effects over the flow. Air
is modeled as an ideal gas.

Parameter Value

Mach Number 2.5 and 3.5

Gas Air

T∞ (K) 144.4 and 94.2

To (K) 324.9

p∞ (Pa) 4483.2 and 1196.6

AoA (deg) -5,….+18

Table 1. Test conditions for the NASA TCM geometry

4
 B. Aerospike
The aerospike-nosed missile (“aerospike”) is an interesting aerodynamic concept that makes use of a sting mounted
on the nose of the vehicle to offset the shockwave in front of the main airframe. The aerospike configuration has the
effect i) to reduce the pressure and temperature loads on the vehicle radome as well as ii) reducing the drag of the
vehicle at high-speed. The aerospike concept was first proposed in the 1950s (Ref. [14]-[16]), and over the years, it
has been revisited several times by different authors (Ref.[17]-[21]). The aerospike concept has been successfully used
in the design of the Lockheed Martin Trident I and II missiles [22], as well as the French M51 SLBM. The present
validation focuses on the aerospike geometry studied by Huebner et al. (Ref. [23]) in 1995 and shown in Fig. 4. The
original Huebner paper does not report some of the secondary dimensions of the aerospike, such as the thickness of
the aerodisk or the angle at which the back of the aerodisk intersects the sting; therefore, the current work follows the
assumptions proposed by Roveda [25]. Because of the complexity of its flow field, this geometry has already been
used to validate different commercial CFD codes, including ANSYS FLUENT (Ref. [24]), Metcomp CFD++, and
Cobalt (Ref. [25]). The present work revisits the results presented by Kourbatski and Montanari (Ref. [24]) using more
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

advanced technologies available in ANSYS FLUENT 2019R3, including polyhedral mesh, the MUSCL scheme for
the convective fluxes, and the ANSYS-proprietary High-Speed Numerics algorithm.

Fig. 4 Geometry nomenclature (top) and dimensions (bottom) of the geometry studied by Huebner. The
bottom picture shows the azimuthal locations (φRay) of the pressure probes on the sensor. All dimensions in
inches. Pictures from Roveda21, top, and from Huebner et al.20, bottom.

5
 High Value

Mach Number 6.06

Gas Air

T∞ (K) 58.25

To (K) 486.1

p∞ (Pa) 1951
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

AoA (deg) 0, +10

Table 2. Test conditions for the aerospike geometry

The computational domain used in the CFD study is shown in Fig. 5. The length of the cylindrical body of the
aerospike is not reported by Huebner, and therefore it is assumed to extend six times the body diameter downstream
of the shoulder to keep any downstream effects away from the area of interest (the aerospike and the semi-spherical
sensor). The computational domain is shaped as a truncated cone, with inlet conditions applied at the upstream flat
surface as well as on the curved sides. The back flat surface is set to a pressure outlet. Freestream conditions are
summarized in Table 2, while Fig. 5 also shows the applied conditions on the corresponding surfaces. The surface of
the aerospike is set to the no-slip adiabatic wall. Since no yaw angle is present in the wind tunnel model, only half of
the geometry was solved, with a plane of symmetry boundary condition applied at the midplane.

Fig. 5 Computational domain and geometry of the aerospike used in the present study, including the
definition of the boundary conditions

The computational mesh was generated using FLUENT Meshing, which is an evolution of the TGrid meshing tool
originally used by Roveda in his study. However, differently from Roveda, who used prismatic and hexahedral
elements, the new mesh is a combination of prisms in the inflation layers and polyhedral cells. The overall cell count
is 15.6 million, with a maximum aspect ratio of 214 and an orthogonal quality of 0.59 as defined in the FLUENT
User’s Guide [1]. Twenty inflation layers with an exponential growth rate of 15% in the direction normal to the wall

6
 were used to capture the turbulent boundary layer. Turbulence was modeled using Menter’s SST k-w turbulence
model. The CFD simulations were run using the commercial CFD code ANSYS FLUENT.
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 6 Isoview of mesh on two orthogonal iso-planes and the surface of the aerospike

First proposed and studied by Huebner et al. (Ref. [23]), the configuration proposes the use of an aerospike in order i)
to reduce pressure and temperature loads on the radome and ii) to reduce the drag of a blunt-nosed vehicle at high
speed. Starting from the validation of the numerical approach using Huebner’s data for two AOAs, the complex flow
field around the missile is analyzed. A sensitivity analysis is performed for the aerodisk, and a modified aerospike
geometry is suggested for the reduction of the overall drag while maintaining the beneficial effects of the offset
shockwave. As part of the design workflow, we are showing and validating the use of overset technology to perform
design space analysis.

C. Aerospike with Overset mesh

Overset (or Chimera) is an undergoing technology widely used in many CFD applications, including external aero,
multiphase flows, etc., due to its advantage in body replacement modeling and flexibility with moving mesh
approaches. At least two overlapped grids are required for overset. Hole cutting (overlapped cells turn into dead cells),
overlap minimization (to minimize the overlap between different grids, cells adjacent to dead cells are receptor cells),
and donor search (which attempts to find overlapping solve donor cells for each receptor) are part of overset domain
connectivity. Receptor cells, for which the donor search fails, are called orphan cells. Coupling between grids cannot
be established in the presence of orphan cells and may lead to instabilities and affect the accuracy of the results.
ANSYS FLUENT has numerical orphan treatment based on neighbor cell data. However, it is always advised to avoid
orphan cells (Ref. [1]). Overset method is not conservative by virtue due to interpolations, but conservation errors can
be reduced by keeping grid sizes of receptors and donors similar.

In overset, grid communication is established via interpolation between donor and receptor cells by following the
equation:
𝑁𝑑

∅ = ∑ 𝑤 𝑖 ∅𝑖
ℎ

𝑖=0

Where ϕ is the solution variable, w is the interpolation weights, and Nd, the number of donors. Fig. 7 depicts a receptor
green cell Grid-2 overlapping triangular Grid-1. The number of donors for each receptor is chosen by default from

7
 neighborhood donor (red) cells; this stencil can also be modified by including node connected neighboring (grey)
cells. The central or bounding (blue) donor cell contains the centroid of the receptor. The red and grey are node-
connected donors with the central donor. In ANSYS FLUENT, there are two types of overset interpolation methods:
boundary distance-based and least squares (default).
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 7 Donor-receptor stencil in overlapped zone

In this investigation, the overset method has been validated on the aerospike model and compared with the results
conducted on non-overset (regular) meshes. Density-based solver with SST κ-ω turbulence model has been employed
in all runs with the default least-squares overset interpolation method. Cut cell mesh for the background grid, and
poly-hexcore grid have been used for the component grid. The total number of cells was 8786208 cells generated in
FLUENT Meshing using the Watertight meshing methodology. As the aim of this study was also focused on the
validation of the code for different spike lengths, a collar component grid has also been generated. Fig. 8 shows the
mesh colored by overset mesh type: blue (receptor cells), red (donor cells), and green (solve cells).

Fig. 8 Mesh colored by overset mesh type used on aerospike model

IV. Results
The following sub-sections discuss the results of the simulations for the three cases described above.

8
 A. NASA TCM
Initially, a comparative study between tetrahedral and polyhedral mesh is done for a 2.5 Mach condition at 18 degrees
AOA. Fig. 9 shows the center section plane of the flow domain for both polyhedral and tetrahedral mesh. The
polyhedral mesh has about 8 million elements, whereas the tetrahedral mesh has about 20 million elements. Fig. 10
shows a simulated Schlieren picture for the TCM geometry with overlapped static pressure contours on the surface
for both polyhedral and tetrahedral meshes. The flow field looks similar qualitatively. Shock placements and
downstream flow structures and pressure contours are also similar.
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 9 Polyhedral (8 million) and Tetrahedral (20 million) mesh

Fig. 10 Flow-field with Polyhedral Mesh (left) vs. Tetrahedral Mesh (right)

Coefficient Polyhedral mesh Tetrahedral mesh

CM -106.1 -105.7
coeff.

CA* 1.139 1.157

CN 21.5 21.5

Table 3
Table 3 shows a quantitative comparison of aerodynamic coefficients for an 18-degree case. All aerodynamic
coefficients are well within 1.5 % of the tetrahedral mesh. The results predicted using polyhedral mesh are equally
accurate as tetrahedral but with a significantly lower mesh count. Low mesh count substantially reduces the run time
even though each polyhedral element takes slightly more CPU time per iteration as compared with the tetrahedral
element. All simulations are now done using polyhedral meshes in this test case.

9
 Fig. 11, Fig. 12 and Fig. 13 show comparisons of primary aerodynamic coefficients for flow at Mach 2.5 and 3.5
at various AOAs. The Axial force coefficient (CA) reported in Ref.[10] is adjusted to correspond freestream static
pressure at the base of the missile geometry. In the experimental setup, the base is integrated with the support/sensor
system, and data for the axial force coefficient for the support/sensor system is provided. CFD produced CA values
for the missile are accordingly corrected to compare with the test data by adding CA for the support system and
subtracting effect of freestream static pressure at the base.

The axial force coefficient (CA) data from the experiments show scattering and do not follow an expected smooth
variation with changing AOAs. Such scattering often insinuates towards experimental uncertainties. Prediction from
the CFD model follows a smooth profile as expected. CA prediction from CFD match well with the experimental data
and are within 5% of the test data for 2.5 Mach case. For Mach 3.5 case, CA values are consistently under-predicted
and are within 3% points. CFD predictions deviate more from the experimental data at higher AOA values for the 2.5
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

case. Interestingly, no such deviation is observed for the Mach 3.5 case.

Normal force coefficient (CN) matches very well with the test data across the range of AOAs. Similarly, pitching
moment coefficients (CM) predictions match well and are within 5% points of the test data. There is consistent over-
prediction of pitching moment all along the AOA range for both Mach conditions.

Fig. 11 Axial Force Coefficient (CA)

10
 Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

11
Fig. 13 Moment Coefficient (CM)
Fig. 12 Normal Force Coefficient (CN)
 Fig. 14 shows simulated Schlieren picture obtained by using the first derivative of the density field for the Mach 3.5
condition. Static pressure contours on the missile surface are overlaid. A complex system of oblique shocks is seen.
The shock structures for the 0-degree case are highly symmetrical as expected. Leading shock originating from the
missile nose is only slightly touching the canards’ tips. The flow field for the 18-degree condition is highly
asymmetrical. The windward shock system shows that the oblique shock originating from the missile nose is hitting
the canard at about the mid-span location. The corresponding shock on the leeward side passes without hitting the
canard.

The static pressure distribution on the missile surface is again symmetrical for the 0-degree AOA case. Notice the
pressure rise on the leading surface of the canards due to the shock wave when compared with the cylindrical missile
surface just upstream. The rear surface of the canards sees a reduction in pressure due to the expansion fan on the
centerline of the diamond-shaped canards. The 18-degree case shows similar variations in pressure, but there are
significant differences between the windward and leeward sides due to high AOA. The windward side canard sees
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

much higher pressure compared with the leeward side.

Fig. 14 Simulated Schlieren picture for the NASA TCM with contours of static pressure on missile surface at
Mach 3.5 with AOA of a) 0-degree b) 18-degree

B. Aerospike
The aerospike case was run at 2 AOAs, 0deg, and 10deg, both with the pressure-based and density-based solver to
compare the agreement of the two solution methods. All the solutions were iteratively fully converged by monitoring
the residuals and a selection of flow parameters to make sure that the flow field had reached a steady-state solution.

Both Huebner (Ref. [23]) and Roveda (Ref.[25]) presented an in-depth analysis of the flow field created by the
aerospike; in the present work, we reference the features identified by these two works to verify the numerical
approach qualitatively. The terminology used in the description of the aerospike geometry is the same as that defined
by Roveda (Ref.[25]), and it is here reprinted in Fig. 4 for reference. Fig. 15 shows a qualitative comparison of the
numerical results of the experimental flowfield for 0deg and 10deg AOA. The experimental Schlieren pictures are
compared with the simulated Schlieren pictures obtained by computing the 1st spatial derivative of the density. For the
0deg AOA, the computed locations of the bow shock ahead of the aerodisk and of the separation-induced shock are
correctly predicted, as well as the recompression shock downstream of the aerodisk. The right side of Fig. 15 shows
the comparison of the surface oil flows, with the experimental results at the top and the numerical solution at the
bottom; again, a general agreement is noticeable, with the CFD solution also correctly predicting the extent of the
low-shear region around the sting originally described by Huebner. Note that this low-shear region on the surface of
the sensor represents the counter-rotating portion of the recirculation region at the base section of the sting.

The flow field becomes highly asymmetrical when the aerospike is at 10deg AOA, Fig. 15(b). The two Schlieren
pictures on the left side of the picture show qualitatively similar flow features, with a large separation region on the
lee-side of the sting and a smaller flow recirculation region on the windward side. These two regions are identifiable
in the surface oil flow on the sensor surface (right side of the picture); on the windward side, the oil flow highlights
the separation region stretching downstream from the lee-side of the sting up to the shoulder. Also, the oil flow shows

12
 the impingement of the shock generated by the windward separation onto the sensor surface: line “A” marked in the
picture by Huebner and correctly predicted by the numerical solution.
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 15 Qualitative validation of aerospike CFD simulations at (a) 0deg and (b) 10deg AOA. From left to
right, top row experiments: Schlieren picture on the plane of symmetry, oil streamlines on sensor and
pressure distribution on radome, experimental vs. CFD. CFD solution obtained using the FLUENT PBNS
solver
Huebner measured the pressure distribution on the sensor of the aerospike using pressure orifices; this pressure data
is used for the quantitative validation of the ANSYS FLUENT solution. The comparison between the experimental
and the numerical data for the 0deg and the 10deg AOA cases is plotted in Fig. 16 and Fig. 17, respectively. As
expected, the numerical pressure distribution for the 0deg AOA case is symmetric for the three azimuthal locations

13
 (φ=0, 90, and 180deg) around the sensor. Some minor differences between the three lines representing the CFD
solution are caused by numerical noise, primarily the non-symmetric mesh. The agreement between the experimental
data and the numerical solution is remarkable, with some minor discrepancy around the location where the flow re-
attaches to the sensor and where the flow most likely experiences transition from laminar to turbulent, 0.5 inch< s <
0.7 inch.

A positive agreement between experimental and CFD solutions is also observed for the 10deg AOA case, Fig. 17. The
experimental pressure distribution is closely matched by the CFD for all the three azimuthal locations; the line for
φ=180deg, i.e., on the windward side, shows a sharp pressure increase at s=0.5inch which corresponds to the
impingement of the separation-induced shock on the sensor as discussed above in Fig. 15. The same sharp increase in
pressure is displayed by the line for φ=90deg at s=0.80inch, which corresponds positively to the impingement of the
separation-induced shock on the sensor. Note that for φ=0deg the pressure is relatively constant since the leeward side
of the sensor is occupied by the low-shear and separation region of the flow. Also, separation-induced shock on the
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

leeward side does not impinge on the surface of the sensor or the vehicle at all, as shown in the simulated Schlieren
of Fig. 15(a), bottom right.

The case for 0deg AOA was also run using the ANSYS FLUENT density-based Navier-Stokes solver (DBNS) for
validating further the results obtained using the pressure-based Navier-Stokes solver (PBNS) discussed above. Fig.
18 shows the comparison of the pressure distributions obtained using the DBNS solver and the PBNS solver as well
as the Huebner experimental results. The 6 CFD curves are almost identical, with minor discrepancies due to the
numerical error caused by a non-symmetric mesh. The Huebner experimental data is also included to emphasize the
agreement of the numerical solutions to the measured data.

Fig. 16 Pressure distribution on the sensor, comparison of Huebner’s experimental data, and ANSYS
FLUENT pressure-based solution for the three azimuthal locations at 0deg AOA.

14
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 17 Pressure distribution on the sensor, comparison of Huebner’s experimental data and ANSYS
FLUENT pressure-based solution for the three azimuthal locations at 10deg AOA

Fig. 18 Comparison of the pressure distribution results for the FLUENT Pressure-Based Navier-Stokes solver
(PBNS) vs. the FLUENT Density-Based Navier-Stokes solver (DBNS) for the three azimuthal locations at
0deg AOA. Huebner's experimental data included for completeness.

15
 Huebner did not measure the steady-state temperature distribution on the sensor, so it is not possible to perform a
validation of the energy field. However, it is still possible to do a qualitative analysis of the energy field as well as a
quantitative comparison of the temperature distribution obtained by ANSYS FLUENT with that computed by Roveda
(Ref. [25]). Fig. 19 shows the temperature distribution on the midplane of the flow field for 0deg and 10 deg AOA.
The figure for the 0deg AOA case (left) exhibits a perfectly symmetric distribution about the centerline of the
aerospike. Also, the main features observed in the Schlieren pictures are visible. Of interest is the relatively constant
high temperatures observed in the recirculation region that envelops the sensor and the shoulder of the vehicle. The
relative uniformity of the temperature distribution for the 0deg AOA is confirmed by the plots along the three
azimuthal angles in Fig. 20. The plot also compares the current results obtained using FLUENT PBNS solver with the
temperature results obtained by Roveda using CFD++ and Cobalt; the temperature distribution obtained by Cobalt is
slightly under-predicted as compared with the ANSYS FLUENT and the CFD++ results, which are virtually identical.

The right side of Fig. 19 shows the temperature distribution for the 10deg AOA case. Again, the main flow features
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

observed in this figure correspond to those observed in the Schlieren picture of Fig. 15; notice the high temperatures
close to the sensor on the windward side, indicating higher thermal loads than on the leeward side.

Fig. 19 Temperature distribution on the plane of symmetry for 0deg (left) and 10deg (right) AOA obtained
with FLUENT PBNS

16
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 20 Temperature distribution on the sensor, comparison of the ANSYS FLUENT pressure-based solution
for the three azimuthal locations versus the numerical results reported by Roveda [25] for the 0deg angle-of-
attack case. The Roveda results have been averaged over the three azimuthal locations for clarity for each of
the two solvers used.

C. Aerospike geometry optimization using sensitivity analysis

A feasibility study was performed on a coarser computational mesh to demonstrate the potential improvement of the
current aerodisk geometry. The aerodisk in the baseline case studied by Huebner is a flat disk with a low-angle
supporting shoulder in the back (see Fig. 4). The sensitivity field was extracted from the flow solution with the
observable set to the overall drag of the aerospike geometry as well as the area-weighted pressure on the wall of the
sensor. Only the shape of the aerodisk was allowed to be modified, with all of the other aerospike surfaces kept fixed.
The two optimization goals specified are i) the overall drag reduction while ii) maintaining an unchanged pressure on
the sensor surface. This latter goal is practically a design constraint, and it was specified to prevent the adjoint solver
from identifying minimum-drag solutions that admit the impingement of the bow shockwave on the sensor walls. Such
impingement would negate one of the two significant benefits of the aerospike design, i.e., keeping the high-pressure
and high-thermal loads created by the shock impingement away from the radome/sensor.

Fig. 21 shows the contours of the magnitude of the shape sensitivity. The ring of high values around the edge of the
aerodisk indicates the strong influence that this sharp angle has on the drag. This effect is validated by the changes in
the geometry performed by geometry optimization. Fig. 22 shows the comparison of the original (left side of figure)
vs. the optimized (right side of the picture) geometry after one cycle of shape optimization for a drag reduction goal
of 2% while maintaining constant pressure on the sensor of 15,771Pa. The original drag force on the entire vehicle
was 95.7N, while for the optimized shape, it was reduced to 94.1N.

17
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 21 Shape sensitivity magnitude. The red ring at the edge of the aerodisk indicates the strong influence of
this edge on the drag.

Fig. 23 is a comparison of the Mach number contours on the plane of symmetry of the geometry. Note the subtle
changes in the shock right in front of the stagnation point, as well as a slightly reduced separation region in the lee
side of the aerodisk. It is expected, albeit not verified in the present study, that further design iterations would
accentuate the sharpness of the aerodisk while maintaining the same cross-sectional area at its back end to maintain
the bow shock away from the sensor.

Original Optimized

Fig. 22 Original aerodisk geometry (left) and modified aerodisk geometry (right) after one design change
iteration. The isometric view at the bottom right also shows the area of the aerospike surface that was allowed
to be morphed; this area corresponds to the aerodisk. All other surfaces were unchanged.

18
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Original Optimized

Fig. 23 Mach contours on the plane of symmetry for the original geometry (left) and the optimized geometry
(right) of the aerodisk.

D. Aerospike design permutations using overset mesh

The results obtained from the overset mesh approach were validated using the non-overset solution discussed above,
and the comparison is shown in Fig. 24. This quantitative comparison of the pressure and temperature ratios on the
middle section of the sensor shows good agreement between the overset solution and the Huebner experimental data
(Ref. [23]) and the non-overset mesh results for the pressure ratio. Fig. 24 also shows good agreement for the
temperature ratio computed using the overset technology, the non-overset solution (from Fig. 20), and the results from
Roveda (Ref. [25]). The qualitative comparison of temperature distribution on the symmetry plane on regular (non-
overset) and overset mesh is shown in Fig. 25. Again, overall good agreement is observed between the two solutions.

With the above validation in mind, the overset technology can be used to perform parametric design studies for the
geometry. In the present work, we are showing the results obtained by varying the length of the sting using different
component meshes and keeping the same background mesh. Fig. 26 shows these results; the Mach number contours
are plotted on the symmetry plane for the three limiting cases of the geometry i) without a spike, ii) with short spike,
and iii) with a full-length spike (baseline). Of critical importance for these simulations was the placement of the collar
grid at the intersection between the sensor and the spike walls, to minimize the orphan cells and improve the accuracy
of the solution. The overset technology proved to be an effective and efficient design tool to study the effect of
geometric parameters without the need for time-consuming remeshing procedures.

19
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

Fig. 24 Distribution of the pressure (left) and temperature (right) ratios on the sensor wall obtained on non-
overset (regular) and overset meshes

Fig. 25 Temperature contours on the symmetry plane on non-overset (regular) mesh (left) and overset mesh
(right)

Fig. 26 Mach number distribution on the symmetry plane on the sensor without spike (left), with a short spike
(center) and with a full spike (right)

Summary
Two high-speed vehicle configurations, namely the NASA TCM at high supersonic and the aerospiked missile at
low-hypersonic speed, are studied using commercial CFD code ANSYS FLUENT. The results obtained are validated
against available experimental data. Predictions match well, both qualitatively and quantitatively. The aerospiked

20
 missile case is further modeled with a shorter spike using the overset mesh technology, which allows convenient
partial replacement modeling for design studies. The results obtained using overset mesh match well with non-overset
(regular) mesh as well as with the test data. Further, a feasibility study for the optimization of the aerodisk geometry
was performed using the sensitivity solution of an adjoint solver. This study demonstrates the aerodynamic benefits
that can be obtained by using an optimized aerodisk. The authors plan to extend the study to calculate aerodynamic
coefficients on a flexible model using fluid-structure interaction modules to assess deviation in aerodynamic
performance due to structural flexure.

Acknowledgments
The authors are thankful to Harshrajsinh Jadeja (ANSYS India) for geometry creation and assistance meshing of
the NASA TCM model. The authors greatly appreciate his expertise and valuable help in this study. Also, the authors
wish to acknowledge the support they received by Dr. Min Xu (ANSYS, Inc., Lebanon, NH, USA), for his assistance
with the running of the FLUENT Adjoint solver.
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

References
[1] Woolf, A. F., Conventional Prompt Global Strike and Long-Range Ballistic Missiles: Background and Issues,
Congressional Research Service, Report R41464, January 2019.
[2] Sayler, K. M., Hypersonic Weapons: Background and Issues for Congress, Congressional Research Service, Report
R45811, July 2019.
[3] Cummings, A., Hypesonic weapons: tactical uses and strategic goals, War on the Rocks, November 2019.
[4] ANSYS, Inc., ANSYS Fluent User’s Guide, Release 2019R3, 2019.
[5] ANSYS, Inc., ANSYS Fluent Theory, Release 2019R3, 2019.
[6] Kim, S.E., Mathur, S., Murthy, J., Choudhury, D., A Reynolds-averaged Navier-Stokes solver using unstructured mesh-
based finite-volume scheme, 36th AIAA Aerospace Sciences Meeting and Exhibit, 1998.
[7] Kim, S.E., Makarov, B., An implicit fractional-step method for efficient transient simulation of incompressible flows,
17th AIAA Computational Fluid Dynamics Conference, 2005.
[8] Weiss, J. W., Maruszewski, J. P., Smith, W. A., Implicit Solution of Preconditioned Navier-Stokes Equations Using
Algebraic Multigrid, AIAA Journal, Volume 37, Number 1, January 1999.
[9] Weiss, J. W., Smith, W. A., Preconditioning Applied to Variable and Constant Density Flows, AIAA Journal, Volume
33, Number 11, January 1995.
[10] Blair, Jr., A. B., Allen, J. M., Hernandez, G., Effect of tail-fin span on stability and control characteristics of a canard-
controlled missile at supersonic Mach number, NASA Technical Paper 2157, June 1983.
[11] Allen, J. M., Blair, Jr., A. B., Comparison of Analytical and Experimental Supersonic Aerodynamic Characteristics of a
Forward Control Missile. Journal of Spacecraft and Rockets 1982 19:2, 155-159
[12] Blair, Jr. A. B., , Rapp, G., Experimental and analytical comparison of aerodynamic characteristics of a forward-control
missile, 18th Aerospace Sciences Meeting.
[13] Akgül, A. & Akargun, Hayri Yigit & Atak, B. & Çetiner, A.E. & Göker, O.. (2012). Numerical Investigation of NASA
Tandem Control Missile and Experimental Comparison. Scientific Technical Review. 62. 3-9.
[14] Stalder, R, J., Nielsen, V.,H., Heat Transfer from a Hemisphere-cylinder Equipped with Flow-separation Spikes, NACA-
TN-3287, 1954.
[15] Crawford, D. H., Investigation of the Flow Over a Spiked-Nose Hemisphere-Cylinder at a Mach Number of 6.8, NASA
TN D-118, 1959.
[16] Seymour M. B., Vas I. E., Preliminary Investigations of Spiked Bodies at Hypersonic Speeds, Journal of the Aerospace
Sciences, February, Vol. 26, No. 2, 1959.
[17] Menezes, V., Saravanan, S., Jagadeesh, G., and Reddy K. P. J., “Experimental Investigations of Hypersonic Flow over
Highly Blunted Cones with Aerospikes, AIAA Journal, Vol. 41, No. 10, October 2003, pp. 1955-1966.
[18] Reding J. P., Guenther, R. A., and Richter, B. J., Unsteady Aerodynamic Considerations in the Design of a Drag-
Reduction Spike, Journal of Spacecraft and Rockets, Vol. 14, No. 1, January 1977.
[19] Srulijes, J., Gnemmi, P., Seiler, F., and Runne, K., Shock tunnel high-speed photography and CFD calculations on spike
tipped bodies, SPIE 4948, 25th International Congress on High-Speed Photography and Photonics, 2003.
[20] Yamauchi M., Fujii, K., and Higashino, F. Numerical Investigation of Supersonic Flows Around a Spiked Blunt Body,
Journal of Spacecraft and Rockets, Vol. 32, No. 1, 1995.
[21] Gnemmi, P., Srulijes, J. and Roussel, K. Flowfield Around Spike-Tipped Bodies for High Attack Angles at Mach 4.5,
Journal of Spacecraft and Rockets, Vol. 40, No. 5, September-October 2003, pp. 622-631.
[22] Waterman, M. D., Richter, B. J., Development of the Trident 1 aerodynamic spike mechanism, 13th Aerospace Mech.
Symposium, NASA Johnson Space Center, 1979.
[23] Huebner, L.D., Mitchel, A.M., and Boudreaux, E.J., “Experimental Results on the Feasibility of an Aerospike for
Hypersonic Missiles”, AIAA Paper 95-0737.

21
 [24] Kurbatskii, K., Montanari, F., Application of Pressure-Based Coupled Solver to the Problem of Hypersonic Missiles with
Aerospikes, 45th AIAA Aerospace Sciences Meeting and Exhibit 8 - 11 January 2007, Reno, Nevada, AIAA Paper 2007-
462.
[25] Roveda, R., Benchmark CFD Study of Spiked Blunt Body Configurations, 47th AIAA Aerospace Sciences Meeting,
January 2009.
Downloaded by UNIVERSITY OF TEXAS AT AUSTIN on January 9, 2020 | http://arc.aiaa.org | DOI: 10.2514/6.2020-2123

22