ADVANCED COMPUTATIONAL FLUID DYNAMICS SIMULATIONS OF

PROJECTILES WITH FLOW CONTROL

Jubaraj Sahu
Karen R. Heavey
U.S. Army Research Laboratory
Aberdeen Proving Ground, Maryland 21005-5066

ABSTRACT

This paper describes a computational study undertaken as part of a grand challenge project to
determine the aerodynamic effect of flow control in the afterbody regions at subsonic and supersonic
speeds using an advanced scalable unstructured flow solver. High parallel efficiency is achieved for both
steady and time-accurate unsteady flow field simulations using advanced scalable Navier-Stokes
computational techniques on SGI Origin 3000, IBM SP3, and Linux Cluster. Numerical simulations with
the unsteady synthetic jet show the jets to substantially alter the flow field both near the jet and the base
region of the projectile that in turn affects the forces and moments even at zero degree angle of attack.
The results have shown the potential of high performance computing computational fluid dynamics
simulations on parallel machines to provide insight into the jet interaction flow fields leading to improved
projectile designs and accurate prediction of flight trajectories.

INTRODUCTION

As part of a Department of Defense High Performance Computing (HPC) grand challenge project, the
U.S. Army Research Laboratory (ARL) has recently focused on the development and application of state-
of-the art numerical algorithms for large-scale simulations [1,2] to determine both steady and unsteady
aerodynamics of projectiles with flow control. One of the objectives is to exploit computational fluid
dynamics (CFD) techniques on HPC platforms for design and analysis of micro adaptive flow control
(MAFC) systems for steering spinning projectiles for infantry operations. The idea is to determine if the
MAFC using synthetic jets1 can provide the desired control authority for course correction for munitions.
This technology has great potential to reduce the actual testing and development costs by providing the
aerodynamics and flight dynamics of all future low to medium caliber Army weapon systems.

The Army is currently seeking a new generation of autonomous, course-correcting, gun-launched

projectiles for infantry soldiers. Due to small projectile diameter (20 to 40mm), maneuvers by control
surfaces such as canards and fins seem very unlikely. An alternate and new evolving technology is the
micro-adaptive flow control through synthetic jets. These very tiny (of the order of 0.3mm) synthetic
micro-jet actuators have been shown to successfully to modify subsonic flow characteristics and pressure
distributions for simple airfoils and cylinders [3,4]. The synthetic jets (fluid being pumped in and out of
the jet cavity at a high frequency of the order of 1000 Hz) are control devices (Figure 1) with zero net
mass flux and are intended to produce the desired control of the flow field through momentum effects.
Many parameters such as jet location, jet velocity, and jet actuator frequency can affect the flow control
phenomenon. Up to now, the physics of this phenomenon has not been well understood and advanced
numerical predictive capabilities or high fidelity CFD design tools did not exist for simulation of these
unsteady jets. However, the research effort described here has advanced the aerodynamic numerical
capability to accurately predict and provide a crucial understanding of the complex flow physics
associated the unsteady aerodynamics of this new class of tiny synthetic micro-jets for control of modern
projectile configurations. High performance CFD techniques were developed and applied for the design
and analysis of these Micro-Adaptive Flow Control systems for steering a spinning projectile for infantry
operations.

U.S. Government Work Not Protected by U.S. Copyright

 The control of the trajectory of a 40mm spinning projectile is achieved by altering the pressure
distribution on the projectile through forced asymmetric flow separation. Unsteady or time-accurate CFD
modeling capabilities were developed and used to assist in the design of the projectile shape, the
placement of the synthetic actuators and the prediction of the aerodynamic force and moments for these
actuator configurations. Additionally, the advanced CFD capabilities provided a simpler way to explore
various firing sequences of the actuator elements. The CFD simulations were performed at full-scale
conditions with second-order time accuracy and fully viscous turbulent flow. Turbulence was modeled
using both Reyonlds-Averaged Navier-Stokes (RANS) and the hybrid RANS/LES (Large Eddy
Simulation) [5-8] approaches. Calculations for this projectile were performed using a scalable parallel
Navier-Stokes flow solver, CFD++ [9,10]. This flow solver is a multipurpose, unstructured, 3-D,
implicit, Navier-Stokes solver. The method used is scalable on SGI Origin 3000 and IBM SP3. The
method incorporates programming enhancements such as dynamic memory allocation and highly
optimized cache management. It has been used extensively in the parallel high performance computing
numerical simulations of projectile and missile programs of interest to the U.S. Army. The code uses
Riemann solvers to provide proper signal propagation physics, preconditioned forms of the governing
equations for low speed flows, and pointwise turbulence models that do not require knowledge of distance
to walls. Numerical solutions must capture with high fidelity all the flow structures associated the
synthetic jet interactions in a time-dependent fashion. It is imperative that these flow structures near the
“tiny” synthetic jets be modeled accurately with particular attention to turbulence modeling so as not to
dissipate the developing flow structure in the immediate vicinity of the jet due to numerical issues.
Modeling of these synthetic jets requires tremendous grid resolution coupled with highly specialized
boundary conditions for the jet activation and the use of a hybrid RANS/LES approach permitting local
resolution of the unsteady turbulent flow with high fidelity. Accurate determination of the unsteady jet
interaction forces and moments while the projectile is subjected to the pulsating micro-jets is a major
challenge. The Department of Defense high performance computing modernization office selected this
research as a grand challenge project and provided the massive computational resources required by these
unsteady time-accurate simulations. The new capability has been demonstrated and this technology has
recently been successfully applied to the self-correcting projectile for infantry operations program.

COMPUTATIONAL METHODOLOGY

The complete set of three-dimensional (3-D) time-dependent Navier-Stokes equations [11] is

solved in a time-accurate manner for simulations of unsteady synthetic jet interaction flow field on the
M203 grenade launched projectile with spin. The 3-D time-dependent Reynolds-averaged Navier-Stokes
(RANS) equations are solved using the finite volume method [9]:

∂
WdV + ∫ [F − G]⋅ dA = ∫ HdV
∂t V∫
(1)
V

where W is the vector of conservative variables, F and G are the inviscid and viscous flux vectors,
respectively, H is the vector of source terms, V is the cell volume, and A is the surface area of the cell
face.

Second-order discretization was used for the flow variables and the turbulent viscosity equations.
Two-equation [10] and higher order hybrid RANS/LES [6] turbulence models were used for the
computation of turbulent flows. The hybrid RANS/LES approach based on Limited Numerical Scales
(LNS) is well suited to the simulation of unsteady flows and contains no additional empirical constants
beyond those appearing in the original RANS and LES sub-grid models. With this method a regular
RANS-type grid is used except in isolated flow regions where denser, LES-type mesh is used to resolve
critical unsteady flow features. The hybrid model transitions smoothly between an LES calculation and a
cubic k-e model, depending on grid fineness. A somewhat finer grid was placed around the body, and
near the jet, the rest of the flow field being occupied by a coarser, RANS-like mesh. Dual time-stepping
was used to achieve the desired time-accuracy. In addition, special jet boundary conditions were
developed and used for numerical modeling of synthetic jets. Grid was actually moved to take into
account the spinning motion of the projectile.

Dual time stepping

The “dual-time-stepping mode” of the code was used to perform the transient flow simulations. The term
“dual-time-step” implies the use of two time steps. The first is an “outer” or global (and physical) time
step that corresponds to the time discretization of the physical time variation term. This time step can be
chosen directly by the user and is typically set to a value to represent 1/100 of the period of oscillation
expected or forced in the transient flow. This time step is applied to every cell (not separately varying).
An artificial or “inner” or “local” time variation term is added to the basic physical equations. This time
step and corresponding “inner-iteration” strategy is chosen to help satisfy the physical transient equations
to the desired degree. If the inner iterations converge, then the outer physical transient equations (their
discretization) are satisfied exactly; otherwise approximately. For the inner iterations, the time step is
allowed to vary spatially. Also, relaxation with multigrid (algebraic) acceleration is employed to reduce
the residues of the physical transient equations. It is found that an order of magnitude reduction in the
residues is usually sufficient to produce a good transient iteration. This may require a few internal
iterations to achieve, between 3 and 10 depending on the magnitude of the outer time step, the nature of
the problem, the nature of the boundary conditions and the consistency of the mesh with respect to the
physics at hand.

Unsteady Jet Boundary Condition

A large number of boundary conditions (BC) are available and can be specified at the appropriate
boundaries. Each boundary condition is encoded as a basic form along with a collection of modifiers.
One particular BC used for the simulations presented here is an “oscillating jet” BC. In its basic form, it
is a steady inflow/outflow BC wherein the user supplies the velocity normal to the boundary along with
static temperature and any turbulence quantities. When the velocity provided is negative, it is considered
to be an inflow and when it is positive, it is treated as an outflow. In the case of inflow, the static
temperature and turbulence quantities are utilized along with the inflow velocity. In the case of outflow,
only the velocity is utilized. At inflow, the tangential component of velocity is set to zero and at outflow,
the tangential component is extrapolated from the interior. At outflow, all primitive variables except
normal velocity are extrapolated from the interior. At inflow, the static pressure is taken from the interior.

The first modifier available for this BC allows the velocity to oscillate. The base velocity is multiplied
by an amplitude which varies as sin (2πft) where f is the frequency of the oscillation. Thus, the
oscillating velocity can cycle from being positive to being negative and back within each period (or from
being negative to positive and back, based on the sign of the input for the basic BC formulation). A
second modifier permits the steady or oscillating inflow/outflow to be on over certain time intervals and
off during other intervals. During “on” periods, the basic or the basic multiplied by the oscillating
amplitude multiplier (first modifier) is used. The user provides the ranges of time during which the jet is
on. The user also provides a repetition time period (e.g. the time period corresponding to one spin
rotation of the projectile). Within each time period, therefore, there are sets of start and end times which
define when the jet is on. During “off” periods, the amplitude is set to zero. In parts of the cycle when
the jet is off, the boundary condition, thus reverts to the condition of inviscid surface tangency. This
allows slip past the boundary, as would exist (in the form of a shear layer) if the jet was emanating from a
cavity/hole.
 Grid Movement

CFD++ code has many features related to grid movement. Grid velocity can be assigned to each mesh
point. This general capability can be tailored for many specific situations. For example, the grid point
velocities can be specified to correspond to a spinning projectile. In this case, the grid speeds are
assigned as if the grid is attached to the projectile and spinning with it. A proper treatment of grid motion
requires careful attention to the details of the implementation of the algorithm applied to every mesh point
and mesh cell so that no spurious numerical effects are created. For example, a required consistency
condition is that free stream uniform flow be preserved for arbitrary meshes and arbitrary mesh velocities.
This condition is satisfied in the code.

Another important aspect deals with how boundary conditions are affected by grid velocities. Two
significant classes of boundary conditions: slip or no slip at a wall, and far field boundary, are considered
here. Both are treated in a manner which works seamlessly with or without mesh velocities. In both
cases, consider the contravariant velocity which includes the effect of grid motion. Effectively it is the
normal component of the velocity relative to the mesh. At the body surface, it is set to zero. For a no slip
wall, one adds the requirement that the tangential component of the velocity is equal to the mesh velocity,
by default. Effectively, this assures that the velocity of the mesh is equal to the velocity of the flow at the
body. At a far-field boundary, the sign of the contravariant velocity determines inflow (negative sign) or
outflow (positive sign). The magnitude of the contravariant velocity is compared with the local speed of
sound, helping to define one of four possibilities: supersonic inflow, supersonic outflow, subsonic inflow,
subsonic outflow. The characteristics theory is used to determine what and how much information is
applied as the boundary condition for each type. A careful treatment of the grid motion results in its use
in a transparent manner.

Hybrid RANS/LES Model

Currently, the two most popular forms of turbulence closure, namely ensemble-averaged models
(typically based on the RANS equations), and LES with a sub-grid-scale model, both face a number of
unresolved difficulties. Specifically, existing LES models have met with problems related to the accurate
resolution of the near-wall turbulent stresses. In the near-wall region, the foundations of large-eddy
simulation are less secure, since the sizes of the (anisotropic) near-wall eddies approach that of the
Kolmogorov scale, requiring a mesh resolution approaching that of a direct numerical simulation. On the
other hand, existing ensemble-averaged turbulence models are limited by their empirical calibration.
Their representation of small-scale flow physics cannot be improved by refining the mesh and over short
time scales they tend to be overly-dissipative with respect to perturbations around the mean, often
suppressing unsteady motion altogether.

While LES is an increasingly powerful tool for unsteady turbulent flow prediction, it is still
prohibitively expensive. To bring LES closer to becoming a design tool, a hybrid RANS/LES approach
[5-8] has recently been developed and used by many researchers. This approach combines the best
features of Reynolds-averaged Navier-Stokes (RANS) and large-eddy simulation (LES) in a single
modeling framework. One such model based on Limited Numerical Scales (LNS) [6] has been recently
developed by Metacomp Technologies. The LNS model is formulated from an algebraic or differential
Reynolds-stress model, in which the sub-grid stresses are limited by the numerically-computed local
length-scale and velocity-scale products. LNS thus behaves like its parent RANS model on RANS-type
grids, but reverts to an anisotropic LES sub-grid model as the mesh is refined locally, thereby reaching
the correct (DNS) fine-grid limit. Locally-embedded regions of LES may be achieved automatically
through local grid refinement, whilst the superior near-wall stress predictions of the RANS model are
preserved, removing the need of ad hoc, topography-parameter-based wall damping.
 The LNS hybrid formulation is well suited to the simulation of unsteady flows, including mixing
flows, and contains no additional empirical constants beyond those appearing in the original RANS and
LES sub-grid models. With this method a regular RANS-type grid is used except in isolated flow regions
where denser, LES-type mesh is used to resolve critical unsteady flow features. The hybrid model
transitions smoothly between an LES calculation and a cubic k-ε model, depending on grid fineness. A
somewhat finer grid was placed around the body, and near the jet, the rest of the flow field being
occupied by a coarser, RANS-like mesh. To date, the LNS technique has been used successfully on a
number of unsteady flows. Examples include flows over cavities, flows around blunt bodies, flows around
airfoils and wings at high angle-of-attack, separation suppression using synthetic jets, forced and natural
convection flows in a room, and mixing flows in nozzles.

PARALLEL COMPUTATIONAL ASPECTS

The CFD++ computational fluid dynamics simulations software was designed from the outset to
include three unification themes: unified physics, unified grid, and unified computing. The “unified
computing” capability includes the ability to perform scalar and parallel simulations with great ease.
Some of the relevant attributes are explained here. The parallel processing capability in CFD++ was
designed in the beginning to be able to run on a wide variety of hardware platforms and communications
libraries including MPI, PVM, and proprietary libraries of nCUBE, Intel Paragon etc. Currently MPI is a
de facto standard and CFD++ uses public domain and vendor-specific versions of MPI depending on the
hardware platform. The code is compatible with and can provide good performance on standard Ethernet
(e.g. 100Mbit, 1Gbit, 10Gbit) as well as high performance communications channels of Myrinet and
Infiniband etc.

It is very easy to use CFD++ on any number of CPUs in parallel. The mesh files, restart and plot files)
that are needed/generated for single CPU runs are identical to those associated with multi CPU runs (for
any number of CPUs). One can switch the use of an arbitrary number of CPUs at any time. The only
extra file required is a domain-decomposition file, which defines the association between cell number and
which CPU should consider that cell as its “native” cell. Depending on the number of CPUs being
employed, the corresponding domain decomposition is utilized. The software suite includes several
domain decomposition tools and it is also fully compatible with the METIS tool developed at the
University of Minnesota. The code runs in parallel on many parallel computers including those from
Silicon Graphics, IBM, Compaq (DEC and HP), as well as on PC workstation clusters. Excellent
performance (see Figure 1 for the timings on a 4-million mesh) has been observed up to 64 processors on
a Silicon Graphics O3K (400 MHz) and an IBM SP3 (375 MHz). Preliminary results (see Figure 2) on a
new Linux PC cluster (3.06 GHz, 128 node, 256 Processors) seem to show 2 to 3-fold reduction in CPU
time for number of processors larger than 16. The reason for somewhat poor performance with 8
processors is most likely due to memory requirements. Work is in progress with runs for a larger 12-
million mesh and on larger number of processors. All operating systems are supported including vendor
specific Unix versions, Linux versions and Windows NT/2000/XP.

The computational algorithms implemented in CFD++ synergistically help achieve all three
unification goals. In particular, the parallel processing capability is fully compatible with all types of
meshes including structured and unstructured, steady and moving and deforming, single and multi block,
patched and overset types. Similarly, the parallel processing capability is fully compatible with inviscid
laminar and turbulent flows, steady and transient flows, incompressible and compressible flows, etc. The
basic solution algorithm includes a successive cell-by-cell relaxation procedure coupled to an algebraic
multigrid model. Inter-CPU communications are included at the fine grid level as well as all the multigrid
levels to help ensure high degree of robustness consistently observed in using CFD++, independent of the
number of CPUs being employed.
 4 Million Point Grid using CFD++

1400

1200
4 Million Point Grid using CFD++

Time Steps per Hour

1000

800 SGI O3K

Time Steps per Hour

350
300 IBM SP3
250 600 Cluster
200 O3K
150 SP3 400
100
50 200
0
0 20 40 60 80 0
0 40 80 120 160
Number of Processors
Number of Processors

Figure 1. Parallel Speedups on the SGI and Figure 2. Parallel Speedups on larger number of
IBM computers. processors (includes Linux Cluster).

PROJECTILE GEOMETRY AND COMPUTATIONAL GRID

The projectile used in this study is a 1.8-caliber ogive-cylinder configuration (see Figure 3).
Here, the primary interest is in the development and application of CFD techniques for accurate
simulation of projectile flow field in the presence of unsteady synthetic jets (see Figure 4). The first step
here was to obtain converged solution for the projectile without the jet. Converged jet-off solution was
then used as the starting condition for the computation of time-accurate unsteady flow field for the
projectile with synthetic jets. The jet locations on the projectile are shown in Figure 5. The jet
conditions were specified at the exit of the jet for the unsteady (sinusoidal variation in jet velocity) jets.
The jet conditions specified include the jet pressure, density and velocity components. Numerical
computations have been made for these jet cases at subsonic Mach numbers, M = 0.11 and 0.24 and at
angles of attack, α = 0o to 4o. The jet width was 0.32 mm, the jet slot half-angle was 18o, and the peak jet
velocities used were 31 and 69 m/s operating at a frequency of 1000 Hz.

A computational grid expanded near the vicinity of the projectile is shown in Figure 6. Grid
points are clustered near the jet as well as the boundary layer regions to capture the high gradients flow
regions. The computational grid has 211 points in the streamwise direction, 241 in the circumferential
direction, and 80 in the normal direction. The unsteady simulation took thousands of hours of CPU time
on Silicon Graphics Origin and IBM SP3 parallel computers running with 16–32 processors.

Lift

Synthetic
Jet

Pulsating
Synthetic Jet
Diaphragm

Figure 3. Projectile geometry. Figure 4. Schematic of a synthetic jet.

 Jet

Jet

Figure 5. Aft-end geometry showing the jet location. Figure 6. Computational grid near the projectile.
RESULTS

Non-spinning Projectile Case

Time-accurate unsteady numerical computations using advanced viscous Navier-Stokes methods were
performed to predict the flow field and aerodynamic coefficients on both a non-spinning and a spinning
projectile. Limited experimental data [12,13] exists only for the non-spinning case and was used to
validate the unsteady CFD results. Unsteady time-accurate CFD computations require huge computer
resources. Based on the results obtained so far from unsteady computations with a single synthetic jet on
a 40mm subsonic grenade (Figure 1), it is clear that time-accurate unsteady simulation even with a single
synthetic jet requires a large amount of resources. The unsteady CFD modeling technique required 50-
100 time steps to resolve a full jet cycle (each unsteady synthetic jet operates at a high frequency of the
order of 1000 Hz). Typically, time-accurate CFD modeling of each jet cycle required about 400
processor-hours on a SGI Origin 2000 for a mesh size about four million grid points. Multiple cycles
(over a few hundreds) of synthetic jet operation and thus, tens of thousands time steps are required to
reach periodic time-accurate unsteady results [14].

3-D unsteady CFD results were obtained at a subsonic Mach number of 0.11 and several angles of
attack from 0° to 4° using both RANS and the hybrid RANS/LES approaches. These 3-D unsteady CFD
results have provided fundamental understanding of fluid dynamics mechanisms associated with the
interaction of the unsteady synthetic jets and the projectile flow fields. The unsteady jets were discovered
to break up the shear layer coming over the step in front of the base of the projectile. It is this insight that
was found to substantially alter the flow field (making it unsteady) both near the jet and in the wake
region that in turn produced the required forces and moments even at zero degree angle-of-attack (level
flight). Time-accurate velocity magnitude contours (Figure 7) confirm the unsteady wake flow fields
arising from the interaction of the synthetic jet with the incoming free stream flow at Mach = .11. Figure
8 shows the particles emanating from the jet and interacting with the wake flow making it highly
unsteady. More importantly, the break up of the shear layer is clearly evidenced by the particles clustered
in regions of flow gradients or vorticity. Verification of this conclusion is provided by the excellent
agreement between the predicted (solid line) and measured [12] (solid symbols) values of the net lift force
due to the jet (Figure 9). The net lift force (Fy) was determined from the actual time histories of the
highly unsteady lift force (an example shown in Figure 10 for various angles of attack) resulting from the
jet interaction at zero degree angle of attack and computed with the new hybrid RANS/LES turbulence
approach.
 Shear Layer

Jet
Jet

Figure 7. Velocity magnitudes, M=0.11, α = 0°. Figure 8. Particle Traces, M = 0.11, α = 0°.

0.2
1.2
1
Change in Lift Force

0.15
EXPERIMENT 0.8
0.6
CFD (Hybrid RANS/LES)
0.1 Fy 0.4
0.2
0.05 0 LNS - A0
LNS - A2
-0.2
LNS - A4
0 -0.4
0 1 2 3 4 5 6 0 10 20 30 40 50
Time (ms)
Angle of Attack

Figure 9. Computed change in lift force due to jet Figure 10. Computed lift force for various angles
at various angles of attack, M = 0.11. of attack, M = 0.11.

Spinning Projectile Case

In this case, the projectile (40 mm grenade) spins clockwise at a rate of 67 Hz looking from the front
(see Figure 11). The jet actuation corresponds to one-fourth of the spin cycle from -45° to +45° with zero
degree being the positive y-axis. The jet is off during the remaining three-fourths of the spin cycle. The
unsteady CFD modeling required about 600 time steps to resolve a full spin cycle. For the part of the spin
cycle when the jet is on, the 1000 Hz jet operated for approximately for four cycles. The actual
computing time for one full spin cycle of the projectile was about 50 hours using 16 processors (i.e. 800
processor-hours) on an IBM SP3 system for a mesh size about four million grid points. Multiple spin
cycles and, hence, a large number of synthetic jet operations were required to reach the desired periodic
time-accurate unsteady result. Some cases were run for as many as 60 spin cycles requiring over 48,000
processor hours of computer time. Computed particle traces (colored by velocity) emanating from the jet
into the wake are shown in Figure 12 at a given instant in time for M = 0.24 and α = 0°. The particle
traces emanating from the jet interact with the wake flow making it highly unsteady. It shows the flow in
the base region to be asymmetric due to the interaction of the unsteady jet.
 Y

Jet-on
t=3.73 ms
t=0
Z

Figure 11. Schematic showing the jet actuation in Figure 12. Computed particle traces colored by
One spin cycle (view from the front or the nose). velocity, jet-on, M = 0.24, α = 0°.

The computed surface pressures from the unsteady flow fields were integrated to obtain the
aerodynamic forces and moments [15] from both unsteady RANS (URANS) as well as the hybrid
RANS/LES approach referred here as the LNS. The jet-off unsteady RANS calculations were first
obtained and the jets were activated beginning at time, t = 28 ms. Computed normal or lift force (FY) and
side force (FZ) were obtained for two different jet velocities, Vj = 31 and 69 m/s and are shown here in
Figure 13 for the bigger jet as a function of time. These computed results clearly indicate the unsteady
nature of the flow field. When the jet is turned off, the levels of these forces drop to the same levels prior
to the jet activation corresponding to the jet-off wake flow. Figure 14 shows the comparison of the
predicted lift force using URANS and LNS models. The URANS result clearly shows when the jet is on
and when it is off during the spin cycle.

As described earlier, the comparisons for the non-spinning cases showed that the level of lift
force predicted by LNS closely matched the data. Here, the addition of spin as well as the jet actuation
for part of the spin cycle, further complicates the analysis of the CFD results with LNS. The level of
oscillations seen is quite large and the effect of the jet cannot be easily seen in the instantaneous time
histories of the unsteady forces and moments. To get the net effect of the jet, unsteady computations were
run for many spin cycles of the projectile with the synthetic jets. The CFD results are plotted over only
one spin cycle, each subsequent spin cycle was superimposed and a time-averaged result was then
obtained over one spin cycle. In all these cases, the jet is on for one-fourth of the spin cycle (time, t=0 to
3.73 ms) and is off for the remainder (three-fourths) of the spin cycle. Figures 15 through 16 show the
time-averaged results over a spin cycle. Figure 15 shows the computed lift force again averaged over
many spin cycles for the peak jet velocity of 69 m/s. The jet effect can clearly be seen when the jet is on
(t=0 to 3.73 ms) even after 5 or 10 spin cycles. The net lift is about 0.17 Newton due to the jet actuation
and seems to have converged after 20 spin cycles. For the remainder of the spin cycle, the jet is off;
however, the effect of the jet on the wake still persists and this figure shows that lift force (mean value .07
Newton) is still available. Figure 16 shows the computed averaged lift force after 50 and 60 spin cycles
for jet velocities 31 and 69 m/s, respectively. It clearly shows that the larger jet producing larger lift force
than the smaller jet when the jet is activated.

The computed lift force along with other aerodynamic forces and moments, directly resulting
from the pulsating jet, were then used in a trajectory analysis [16] and as shown in Figure 17, the
synthetic micro-jet produced a substantial change in the cross range and thus, provided the desired course
correction for the projectile to hit its target.
 0.6
0.1
Jet on
0.4
0.06
Jet off 0.2
Forces

0.02
Fy 0
-0.02
-0.2
LNS
-0.06 Fy Fz -0.4
URANS

-0.1 -0.6
20 30 40 50 60 70 80 20 30 40 50
Time (ms) Time (ms)
Figure 13. Computed lift and side forces, URANS, Figure 14. Computed lift forces, URANS and LNS
M = 0.24, Vj = 69 m/s, α = 0°. M = 0.24, Vj = 69 m/s, α = 0°.
0.4
0.4
Fy (2)
Fy (5)
Fy (10) 0.3
0.3 Fy (20) Fy (50), Vj= 31 m/s
Fy (30)

Forces (Newton)
Fy (40) Fy (60), Vj = 69 m/s
Forces (Newton)

0.2
0.2

0.1
0.1

0
0

-0.1
-0.1
0 5 10 15
0 5 10 15
Time (ms)
Time (ms)
Figure 15. Computed lift force over many spin cycles, Figure 16. Computed lift force over many spin
LNS, Vj = 69 m/s, M = 0.24, α = 0°, Spin = 67 Hz. cycles for different jet velocities, LNS,
M = 0.24, α = 0°, Spin = 67 Hz.
Target
3.00

2.00

1.00

0.00
Force, Fy

40.0 41.0 42.0 43.0 44.0 45.0 46.0 47.0 48.0 49.0 50.0

synthetic No synthetic -1.00

micro-jet micro-jet
-2.00 Fy - Jetoff
Fy _ Jeton (RANS)
-3.00 Fy - Jeton (LNS)

-4.00
Range, m

Figure 17. Synthetic jet control on the flight trajectory. Figure 18. Computed side force, URANS and LNS.

Some preliminary results have also been obtained where a part of the trajectory is computed using
a coupled multi-disciplinary computational fluid dynamics (CFD)/rigid body dynamics (RBD) technique.
When the jet is on, it is applied in the positive y-direction (z is positive up, in this case). Figure 18 shows
the side force (Fy) and the lift or normal force (Fz) obtained with the LNS model as a function of the
range. The effect of the jet is seen to increase the side force; however, the effect of the jet on the lift
force is almost negligible. Figure 19 shows the computed y-distance as a function of x or the range. Even
though the jet was turned on at x = 41 m, the effect of the jet is not immediately felt for about 1.5 m.
Both jet-on results obtained with the RANS and the hybrid RANS/LES or LNS approaches are compared
with the jet-off results. The RANS model predicts a small increase in y with increase in range. With the
LNS model however, y decreases a little at first until x = 46.5 m and then increases more rapidly. One
can clearly see the effect of the jet in the y-distance that increases with an increase in the range. These
computed results strongly indicate that applying the jet in the positive y-direction moves the projectile in
the same positive direction with little or no effect on the other aerodynamic forces.
0.290

0.289

0.288

0.287

0.286

Y-Coord, m 0.285

0.284 CFD - Jetoff

0.283 CFD - Jeton (RANS)

CFD - Jeton (LNS)

0.282

0.281

0.280
40.0 41.0 42.0 43.0 44.0 45.0 46.0 47.0 48.0 49.0 50.0
Range, m

Figure 19. Computed Y distance, URANS and LNS.

The ARL is also interested in the use of flow control to maneuver projectiles at supersonic
speeds. To what extent the synthetic unsteady jets can be effectively used for supersonic flow
applications is not quite clear. Therefore, research has focused on alternate flow control mechanisms at
supersonic speeds and advanced CFD has successfully been applied to perform computational
aerodynamic analyses of a supersonic finned projectile (see Figure 20). Advanced CFD capabilities were
applied to a number of design configurations to determine the feasibility of alternate flow control
mechanisms (e.g. steps and plates placed between a set of fins) to provide control maneuverability of the
projectile at supersonic speeds. This work was instrumental in providing the physical insight into the fluid
dynamics interactions and led to an optimized location of a divert mechanism (plate) for maximum effect
on the shock position and pressure distribution. Figure 21 shows the computed pressure contours for
this projectile (without any control plate) at Mach = 2.5 and zero degree angle of attack. One can clearly
see the bow shock in front of the nose of the projectile as well the shock wave emanating from the leading
edge fin body junction. Also evident is the intersection of the shocks in between the adjacent fins. The
pressure distribution as expected is symmetric between each set of fins.

Figure 20. Finned Configuration Figure 21. Computed pressure contours,

Mach = 2.5 and angle of attack, α = 0o.

Computed results obtained for this projectile with a control plate are shown in the next three figures.
Figure 22 shows the computed longitudinal pressure contours at the same Mach = 2.5 and zero degree
angle of attack. The addition of the control plate changes the flow field in the fin and the wake areas. In
addition, it makes the flow field asymmetric which can be clearly seen in the pressure contours in the
wake. Figures 23 and 24 show the computed surface pressure contours in the vicinity of the fin region
 and a cross-sectional view of the computed pressure contours, respectively. The control plate blocks the
flow resulting in low velocity or high pressure region in front (see Figure 23). The high pressure region
covers almost the entire area in between the fins where the plate is located. The surface pressure
distribution in between the other fins is similar to that of the no plate case. The high pressure region
shown in red is a result of the plate being located in that area. As seen in Figure 24, the effect of the
control plate extends beyond the two fins i.e. beyond the quadrant of the plate’s location. Also, the
control plate makes the flow field asymmetric with respect to the vertical and horizontal axes. This
asymmetry in the pressure distributions results in aerodynamic forces and moments that can be used to
control or maneuver the flight of the projectile.

Figure 22. Computed longitudinal pressure contours

with a control plate, Mach = 2.5 and angle of attack, α = 0o.

Figure 23. Computed surface pressure contours with a Figure 24. Computed cross-sectional pressure contours
control plate, Mach = 2.5 and angle of attack, α = 0o. with a control plate, Mach = 2.5 and angle of attack, α = 0o.

CONCLUDING REMARKS

This paper describes a computational study undertaken to determine the aerodynamic effect of
flow control in the afterbody regions at subsonic and supersonic speeds using a scalable unstructured flow
solver on various parallel computers such as SGI, IBM, and Linux Cluster. Advanced scalable Navier-
Stokes computational techniques were used to obtain numerical solutions for both steady and unsteady
aerodynamic flow fields. High parallel efficiency is achieved for both steady and time-accurate unsteady
cases. Numerical results show the effect of the jet on the flow field and on the aerodynamic coefficients
for the unsteady jet interaction flow fields at subsonic speeds. The unsteady jet is shown to substantially
alter the flow field both near the jet and the base region of the projectile that in turn affects the forces and
moments even at zero degree angle of attack. For the spinning projectile cases, the net time-averaged
results obtained over the time period corresponding to one spin cycle clearly showed the effect of the
synthetic jets on the lift as well as the side forces. The jet interaction effect is clearly seen when the jet is
on during the spin cycle. However, these results show that there is an effect on the lift force (although
reduced) for the remainder of the spin cycle even when the jet is off. This is a result of the wake effects
that persist from one spin cycle to another. The impulse obtained from the predicted forces for two jets
seem to asymptote after 30 spin cycles.

The results have shown the potential of high performance computing computational fluid
dynamics simulations on parallel machines to provide insight into the jet interaction flow fields leading to
improve designs. The same flow solver has been applied to a different type of flow control problem for a
finned configuration at a supersonic speed and the results look promising. Also, preliminary results
obtained from multidisciplinary coupled CFD and rigid body dynamics calculations show great promise
for accurate prediction of the aerodynamics and the actual flight trajectories simultaneously.

REFERENCES

1. J. Sahu, K. R. Heavey, and E. N. Ferry, "Computational Fluid Dynamics for Multiple Projectile
Configurations", Proceedings of the 3rd Overset Composite Grid and Solution Technology Symposium, Los
Alamos, NM, October 1996
2. J. Sahu, K. R. Heavey, and C. J. Nietubicz, "Time-Dependent Navier-Stokes Computations for Submunitions
in Relative Motion", 6th International Symposium on Computational Fluid Dynamics, Lake Tahoe, NV,
September 1995
3. B. L. Smith and A. Glezer, “The Formation and Evolution of Synthetic Jets.” Journal of Physics of Fluids,
vol. 10, No. 9, September 1998
4. M. Amitay, V. Kibens, D. Parekh, and A. Glezer, “The Dynamics of Flow Reattachment over a Thick Airfoil
Controlled by Synthetic Jet Actuators”, AIAA Paper No. 99-1001, January 1999
5. S. Arunajatesan and N. Sinha, “Towards Hybrid LES-RANS Computations of Cavity Flowfields”, AIAA
Paper No. 2000-0401, January 2000
6. P. Batten, U. Goldberg and S. Chakravarthy, "Sub-grid Turbulence Modeling for Unsteady Flow with
Acoustic Resonance", AIAA Paper 00-0473, 38th AIAA Aerospace Sciences Meeting, Reno, NV, January
2000
7. R.D. Sandberg and H. F. Fasel, “Application of a New Flow Simulation Methodology for Supersonic
Axisymmmetric Wakes”, AIAA Paper No. 2004-0067.
8. S. Kawai and K. Fujii, “Computational Study of Supersonic Base Flow using LES/RANS Hybrid
Methodology”, AIAA Paper No. 2004-68.
9. O. Peroomian, S. Chakravarthy, S. Palaniswamy, and U. Goldberg, “Convergence Acceleration for Unified-
Grid Formulation Using Preconditioned Implicit Relaxation.” AIAA Paper 98-0116, 1998
10. U. Goldberg, O. Peroomian, and S. Chakravarthy, “A Wall-Distance-Free K-E Model With Enhanced Near-
Wall Treatment” ASME Journal of Fluids Engineering, Vol. 120, 457-462, 1998
11. T. H. Pulliam and J. L. Steger, “98dOn Implicit Finite-Difference Simulations of Three- Dimensional Flow”
AIAA Journal, vol. 18, no. 2, pp. 159–167, February 1982
12. C. Rinehart, J. M. McMichael, and A. Glezer, “Synthetic Jet-Based Lift Generation and Circulation Control
on Axisymmetric Bodies.” AIAA Paper No. 2002-3168
13. McMichael, J., GTRI, Private Communications.
14. J. Sahu, “Unsteady Numerical Simulations of Subsonic Flow over a Projectile with Jet Interaction” AIAA
Paper 2003-1352, Reno, NV, 6-9 January 2003
15. Sahu, J., “Unsteady CFD Modeling of Aerodynamic Flow Control over a Spinning Body with Synthetic Jet.”
AIAA Paper 2004-0747, Reno, NV, 5-8 January 2004.
16. Costello, M., Oregon State University. Private Communications.