AIAA JOURNAL

Vol. 51, No. 8, August 2013

Plasma Control of a Turbulent Shock Boundary-Layer Interaction

Nicholas J. Bisek, Donald P. Rizzetta, and Jonathan Poggie
U.S. Air Force Research Laboratory, Wright-Patterson Air Force Base, 45433-7512 Ohio
DOI: 10.2514/1.J052248
The NavierStokes equations were solved using a high-fidelity time-implicit numerical scheme and an implicit
large-eddy simulation approach to investigate plasma-based flow control for supersonic flow over a compression
ramp. The configuration included a flat-plate region to develop an equilibrium turbulent boundary layer at Mach
2.25, which was validated against a set of experimental measurements. The fully turbulent boundary-layer flow
traveled over a 24 deg ramp and produced an unsteady shock-induced separation. A control strategy to suppress the
separation through a magnetically-driven surface-discharge actuator was explored. The size, strength, and placement
of the model actuator were based on recent experiments at the Princeton University Applied Physics Group. Three
control scenarios were examined: steady control, pulsing with a 50% duty cycle, and a case with significant Joule
heating. The control mechanism was very effective at reducing the time-mean separation length for all three cases. The
steady control case was the most effective, with a reduction in the separation length of more than 75%. The controller
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

was also found to significantly reduce the low-frequency content of the turbulent kinetic energy spectra within the
separated region and reduce the total turbulent kinetic energy downstream of reattachment.

Nomenclature = specific heat ratio, 1.4 for air

a, b = nondimensional radii of the ellipsoid for the s = su w , value normalized by the inner length
controller model scale
cf =
u2l uy jw , skin-friction coefficient = 0.99u , boundary-layer height
D = scale parameter for counterflow trip model  = 0
1 u
u dy, compressible boundary-
Et = nondimensional total specific energy layer displacement thickness
FI , GI , HI = inviscid vector fluxes = 0
u
u 
1 uu  dy, compressible
Fv , Gv , Hv = viscous vector fluxes boundary-layer momentum thickness
J = transformation Jacobian = kinematic viscosity
J = scale Joule heating parameter for the controller , , = computational coordinates
model = nondimensional density
lsep = separation length ij = components of the viscous stress tensor
L = scale magnetic body-force vector for the
controller model Subscripts
l = reference or characteristic length
c = center of ellipsoid for the controller model
M = Mach number
m = time-mean value
N = number of points
w = wall value
Pr = Prandtl number, 0.72 for air
= reference or freestream value
p = nondimensional static pressure
pt = p
1 
1M2 2
1 , nondimensional to-
tal pressure Superscripts
Q = conserved variable vector m, n = power of exponents for the controller model
qi = components of the heat flux vector 0 = fluctuating component
Re = u l , Reynolds number
Sc = source vector
fl I. Introduction
St = u , Strouhal number
T = nondimensional static temperature
t
u, v, w
=
=
nondimensional time
nondimensional Cartesian velocity components
W ITH the rapid increase in supercomputer capabilities over the
past two decades, computer modeling is being used to predict
turbulent flows that extend beyond unit-sized problems (in both
in













p the x,
y, z directions geometric and fluid-dynamic complexities). Lately, there has been a
u = w w , wall friction velocity particular emphasis on turbulent-shock boundary-layer interaction
Vd = deposition volume density for controller model (SBLI), which may result in a region of separated flow. This phenom-
x, y, z = Streamwise, normal, and spanwise directions in enon has practical implications because the pressure fluctuations
nondimensional Cartesian coordinates within the unsteady separated region can lead to localized fatigue
loading and the premature failure of the structure. In addition, work
Presented as Paper 2012-2700 at the 42nd AIAA Fluid Dynamics by Plotkin [1], Poggie and Smits [2], and Touber and Sandham [3]
Conference, New Orleans, LA, 2528 June 2012; received 31 July 2012; indicated that the separation acted as an amplifier of the low-
revision received 12 October 2012; accepted for publication 15 October 2012; frequency disturbances. This phenomenon may lead to additional
published online 14 February 2013. This material is declared a work of the problems for a vehicle with SBLI because any disturbances generated
U.S. Government and is not subject to copyright protection in the United upstream of the separation could be amplified in a frequency range
States. Copies of this paper may be made for personal or internal use, on characteristic of the vehicle structure and cause damage or decreased
condition that the copier pay the $10.00 per-copy fee to the Copyright performance to downstream subsystems.
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include
the code 1533-385X/13 and $10.00 in correspondence with the CCC. In addition to contributing to localized fatigue loading, separated
*Research Aerospace Engineer, RQAC. Member AIAA. flow due to SBLI increases the boundary-layer thickness, which

Senior Research Aerospace Engineer, RQAC. Associate Fellow AIAA. decreases the cross-sectional area of the inviscid core for an internal
flow. This effect is detrimental for inlet/isolator performance because
Team Lead, High-Speed Flow Research Group, RQAC. Associate Fellow
AIAA. it decreases the mass-flow rate for air-breathing configurations. The
1789
 1790 BISEK, RIZZETTA, AND POGGIE
unfavorable behavior of the separated flow motivates research to
better understand the physical mechanisms driving the phenomenon
and to identify suitable ways of mitigating or, preferably, eliminating
the region from the flow path. Traditionally, boundary-layer bleed
has been used to remove the unfavorable flow, but this requires
additional plumbing within the vehicle and power to facilitate bleed.
In addition, bleed itself can be detrimental for internal flows because
a portion of the mass flow is removed from the flow path. Micro
vortex generators (VGs) have also been used to control the
unfavorable flow by entraining high-momentum inviscid freestream
flow into the boundary layer. Although the VGs are usually effective
at controlling a supersonic separated flow [4], their existence in the
boundary layer does partially block an internal flow path, which is
undesirable when the flow does not constantly require the control
authority. This deficiency can be mitigated by lowering the VGs into
the surface when they are not needed, but this procedure introduces
mechanical complexity and a relatively slow controller response
time, and it may not always be feasible given the volume constraints
of a specific vehicle configuration. A review of the recent progress
using micro-VGs for flow control is available in [5].
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
It can be difficult and costly to fully study turbulent SBLI in
wind-tunnel experiments due to the extreme environment in which
these conditions exist. In addition, it is complementary to investigate
the phenomenon using computational fluid dynamics, because
computational results can provide additional details and a wider
understanding of the complex interactions existing in the flow. The
ideal approach would be to simulate the flow using direct numerical
simulation (DNS). In the strict definition of DNS, all length and
time scales are fully resolved everywhere in the computational
domain. Unfortunately, for a supersonic vehicle operating at flight
conditions, a DNS of the entire flow path is still well beyond current
computational resources. As such, large-eddy simulation (LES) is a
popular alternative approach for investigating these flows.
The LES approach is predicated on the assumption that turbulent
kinetic energy is transferred from large to small scales in the energy
cascade and that intercepting the energy flow in the inertial subrange
is equivalent to resolving the small scales on which energy
dissipation physically occurs. As such, LES accurately resolves all
length and time scales down to the inertial subrange everywhere in
the computational domain and then relies on a subgrid-scale (SGS)
model or a filtering procedure to model the energy cascade to the
smaller, underresolved scales. It is also possible to dissipate the
energy from the inertial subrange directly from the numerical
method. Because the numerical method does not require an SGS
model or filter to obtain the results, it is sometimes referred to as
direct simulation or DNS in the literature. However, because these
numerical simulations do not resolve all the time and length scales
everywhere in the flow, they are regarded as LES or implicit LES in
this work because the dissipation of energy from the cascade occurs
as a result of the numerical scheme and the grid resolution, which is
applied for scales larger than the Kolmogorov length scale for most of
the computational domain. Regardless of semantics, LES has the
benefit of less-stringent grid and time-scale requirements, while still
providing physically accurate results. The LES approach is widely
accepted and employed in the turbulence community [6] because it
allows for the study of a much wider range of problems [712].
Fig. 1 Flow over a 24 deg compression corner at Mach 2.25.
The present work explores the interaction of a turbulent boundary
layer with a compression corner, as illustrated in Fig. 1. This problem
has been studied numerically by several researchers, including work
by Adams [13], Rizzetta and Visbal [10], Rizzetta et al. [14], Loginov
et al. [15], Wu and Martn [16], Muppidi and Mahesh [17], Priebe and
Pino Martn [18,19], and Grilli et al. [20]. All of these previous
computational studies correspond to a freestream Mach number near
2.9, whereas this study explored a freestream Mach number of 2.25.
Because of the lower Mach number and the large ramp angle of
24 deg, the separation length due to the SBLI is larger than in the
aforementioned studies, and so the loading levels near separation
may be more pronounced.
Based on experimental work, Selig and Smits [21] suggested
that the shock unsteadiness found in Mach 3 turbulent flow over a
24 deg compression corner was partially driven by the large-scale
fluctuations in the separated region. From this observation, Dolling
[22] hypothesized that a significant reduction in the separated-
flow length scales would probably reduce the loading levels found
near separation. This conclusion and previous observations by
other researchers motivated several SBLI control experiments
throughout the 1980s, as outlined in Dollings review [22]. These
early experiments primarily explored passive techniques or mass-
transfer mechanisms. Although these methods successfully reduced
the extent of separation, issues regarding bleed and challenges with
off-design conditions have motivated research in active control
methods. In particular, more-recent experimental work has explored
controllers that operate without mass transfer, using a plasma-based
mechanism [2325]. These plasma-based devices are advantageous
because they have no moving parts, have a minimal aerothermal
penalty when they are not operating, and have a response time short
enough to match characteristic flow time scales.
The main focus of the present work was to reduce the extent of
separation due to an SBLI by using a magnetically-driven surface-
discharge flow controller. The paper will first explore the develop-
ment of a supersonic equilibrium turbulent boundary-layer flow over
a compression corner to establish a baseline flow with a time-mean
separation bubble. A grid study was performed, and the results were
compared to experimental data. The results show that the LES was
accurately capturing the essential flow features. In addition, the flow
spectra upstream of separation were free of disturbances at discrete
frequencies, which was an important detail for this study because it
prevented the peculiarities of the incoming flow from obscuring
the effects of the controller. Finally, a model control device was
introduced into the computations, and its effects on time-mean and
time-dependent statistics were examined for three control strategies.
The results show that the separation length was significantly reduced
by the controller, the lowest frequency content was reduced with
actuation, and the total turbulent kinetic energy of the flow was lower
downstream of reattachment with the application of a magnetically-
driven surface-discharge flow controller.
II. Method
The simulations were carried out with a time-accurate three-
dimensional compressible NavierStokes solver known as FDL3DI
[26], which has been widely used in previous LES for both subsonic
and supersonic flows [2733].
A. Governing Equations
The governing equations were transformed from Cartesian
coordinates into a general time-dependent curvilinear coordinate
system that was recast in strong conservation-law form:
     
Q FI F GI G
 v  v
t J J Re J J Re J
 
HI Hv
  Sc (1)
J Re J
where t is the time; , , and are the computational coordinates; and
J is the transformation Jacobian [34]. The source vector Sc on the
 BISEK, RIZZETTA, AND POGGIE 1791

right side of Eq. (1) was typically set to zero. However, nonzero B. Source Terms
source terms were imposed at specific locations within the domain to The right-hand side of Eq. (1) allowed various flow-control and
transition the flow to turbulence or to apply the model of the flow-trip models to be introduced into the governing equations. This
magnetically-driven surface-discharge controller. The value of Sc paper employed two different source models. The trip model initiated
will be described in greater detail in the following subsection. The transition of the incoming laminar flow to a turbulent boundary layer,
solution vector and vector fluxes (both inviscid and viscous) are while the control model was a phenomenological representation of a
magnetically-driven surface-discharge flow controller.
Q   u v w Et T (2) This work used the counterflow, body-force, bypass-transition
method developed by Mullenix et al. [35] because the method has
been shown to produced a steady, broadband disturbance, which
2 3 2 3
xi ui xi ui transitions supersonic flow. The model specified a nonzero body
6 7 6 7 force over a right-triangle region of the x y computational domain,
6 ux ui  x p 7 6 ux ui  x p 7 which was centered at x  2.5 l and was applied uniformly in the
6 i 7 6 i 7
6 7 6 7 span. The strength of the counterflow body force was controlled by a
FI  6
6 v xi iu  y p 7;
7 GI  6 6 v xi iu  y p 7;
7
6 7 6 7 nondimensional parameter, D  6.2, for the present work. A full
6 wxi ui  z p 7 6 wxi ui  z p 7 derivation of the trip model is available in [35].
4 5 4 5
Control of the separated flow due to SBLI was achieved by the
Et xi ui  xi ui p Et xi ui  xi ui p
2 3 introduction of a streamwise force produced by a gliding surface-
xi ui discharge actuator. In a gliding surface discharge, two flush-mounted
6 7 surface electrodes are separated by an insulator. The electrodes are
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

6 ux ui  x p 7
6 i 7 arranged in the streamwise direction and are slightly diverging from
6 7 each other. If the distance between the electrodes is sufficiently small
HI  6
6 v xi iu  y p 7
7 (3)
6 7 and the voltage between the electrodes is sufficiently high
6 w xi ui  z p 7 (30 kVcm for air [36]), breakdown occurs and a plasma column
4 5
Et xi ui  xi ui p forms between the electrodes at their closest point. The discharge
partially ionizes the surrounding air, which carries the current path
with it.
2 3 2 3 Applying a strong magnetic field perpendicular to the surface
0 0 accelerates the plasma column due to the magnetic body force.
6 7 6 7 Depending on the direction of the magnetic field, the column can be
6 xi i1 7 6 xi i1 7
6 7 6 7 forced to move upstream or downstream at a rate several times faster
6 7 6 7
Fv  6
6 xi i2 7;
7 Gv  6
6 xi i2 7;
7 than the freestream speed [37]. As the column travels down the length
6 7 6 7 of the electrodes, its core area contracts due to the diverging
6
4 xi i3 7
5
6
4 xi i3 7
5 orientation of the electrodes. Contraction of the columns core area
xi
uj ij qi  xi
uj ij qi  leads to instabilities within the plasma column, and it eventually
2 3 extinguishes itself. Once extinction occurs, a new discharge forms at
0 the narrow end of the electrodes and the process repeats.
6 7
6 xi i1 7 Because the magnetically-driven surface-discharge velocity
6 7
6 7 considered in this study was many times faster than the freestream
Hv  6
6 xi i2 7
7 (4) flow, a time-mean representation of the actuators influence was
6 7 appropriate for fluid-dynamic simulations. In addition, the time
6
4 xi i3 7
5 scales of the plasma were orders of magnitude smaller that the
xi
uj ij qi  characteristic flow time of the fluid, and so the actuator could be
incorporated in the fluid through time-mean source terms [38].
using tensor index notation xi , i  1 ::: 3 represents the x, y, and z Figure 2 illustrates the configuration of the magnetically-driven
coordinates, and ui , i  1 ::: 3 represents the u, v, and w Cartesian surface-discharge system applied to control the turbulent SBLI
velocity components. The nondimensional total specific energy is investigated in this paper.
defined as The present work used a phenomenological model of the force and
energy deposition that was similar to that used in the Reynolds-
T 1 averaged calculations of Atkinson et al. [39] and Atkinson [40]. A
Et   u2 (5) hyper-Gaussian function was used to describe the deposition volume-

1M2 2 i
density function, which provided a mathematical representation of
where is the density, p is the pressure, and T is the temperature.
Components of the shear stress tensor and the heat flux vector are Magnetic field
expressed as
 
k ui k uj 2 l uk Flow
nt

Force = J x B
ij   ij ;
rre

plasma
xj k xi k 3 xk l
cu

p
   ram
1 j T flat plate
qi  (6)

1M2 Pr xi j a) Physical configuration y

x
using tensor index notation i , i  1 ::: 3 to represent , , and , and z

ui , i  1 ::: 3 represents the u, v, and w Cartesian velocity

components. All length scales were nondimensionalized by the Flow Joule Heating
Force (L)
reference length l, and all dependent variables were normalized by 2a
( )

p
their respective reference values, expect for pressure, which was 2b
ram
flat plate
nondimensionalized by u2. The Sutherland law for the molecular
viscosity, the perfect gas relationship, and a constant Prandtl number b) Idealized actuator model
Pr  0.72 were employed for the present work. Fig. 2 Illustration of the gliding surface-discharge model.
 1792 BISEK, RIZZETTA, AND POGGIE

the shape of the plasma column and the deposition density within the order spatial resolution of implicit operators was also reduced by
column: performing subiterations. Three applications of the flow solver per
time step were applied throughout this work to preserve second-order
K temporal accuracy.
Vd  f
x; y
ab The compact difference scheme employed on the right-hand side
     
x xc m y yc n of Eq. (9) is based upon the pentadiagonal system of Lele [45] and is
f
x; y  exp (7) capable of attaining spectral-like spatial wave-number resolution.
a b
This was achieved through the use of a centered implicit-difference
where a and b represent the ellipsoidal radii in the x and y directions; operator with a compact stencil, thereby reducing the associated
xc and yc represent the center of the ellipse; and K is a constant such discretization error. For the present computations, a sixth-order
that V d dxdy  1 for the selected values of m and n. For this tridiagonal subset of Leles system was used, which is illustrated here
work, m  n  10, which allowed the deposition rate to be nearly (in one spatial dimension):
uniform within the volume (i.e., f
x; y 1), yet it rapidly decayed to      
zero outside the region. The controller model accounted for both the 1 F F 1 F
 
mean magnetic body force and Joule heating, as indicated in Eq. (8): 3 i1 i 3 i1
   
2 3 14 Fi1 Fi1 1 Fi2 Fi2
0   (10)
9 2 9 4
Lx
V 66
7
7
Sc  d 6 Ly 7 (8) The scheme has been adapted by Visbal and Gaitonde [46] as
J 4 5
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

Lz an implicit iterative time-marching technique. It was used in conjunc-

uLx  vLy  wLz  J tion with a low-pass Pad-type nondispersive spatial filter, which was
incorporated by Gaitonde et al. [47]. Use of the filter has been shown
where V d is the deposition volume density described in Eq. (7). The to be superior to the use of explicitly added artificial dissipation for
components of the nondimensional magnetic body-force vector, Lx , maintaining both stability and accuracy on stretched curvilinear
Ly , and Lz , are proportional to the cross product of the current and meshes [46]. The filter was applied to the solution vector sequentially
the magnetic field produced by the controller. The modification of in each of the three computational directions following each
the total energy equation includes both the reversible work (i.e., subiteration and was implemented in one dimension as
uLx  vLy  wLz ) and Joule heating J, which reflects the direct
thermalization of the electrical power. Joule heating occurs when X
4
an
directed electron motion in the plasma is randomized into heat in f Q^ i1  Q^ i  f Q^ i1 
Qin  Qin  (11)
electron-neutral collisions. An energy exchange occurs as a result of n0
2
these collisions, primarily in vibrational modes of the neutral
nitrogen molecules [37]. Some of the excited vibrational states where Q^ designates the filtered value of Q. The order of the filtering
quickly relax to equilibrium (i.e., direct thermalization), while other operation was permuted and was a postprocessing technique. The
states do not relax before leaving the computational domain. filter was applied to the evolving solution to regularize poorly
resolved features at the grid scale. On uniform grids, the filtering
C. Numerical Method procedures preserve constant functions while completely eliminating
Time-accurate solutions to Eq. (1) were obtained numerically by the oddeven mode decoupling [31,48]. Equation (11) represents a
the implicit approximately-factored finite-difference algorithm of one-parameter family of eighth-order filters. The numerical values
Beam and Warming [41], employing Newton-like subiterations [42], for an are available in [26]. The filter coefficient f is a freely
and may be written as follows: adjustable parameter that may be selected for specific applications,
where jf j < 0.5. The value of f determines sharpness of the filter
    p      p  cutoff and was set to 0.3 for the present simulations.
1 2t FI 1 Fpv 1 2t GI 1 Gpv
 2 J  2 J The spatial filter associated with the high-order compact scheme
J 3 Q Re Q J 3 Q Re Q
    p p  may produce spurious oscillations in the vicinity of shocks, which
1 2t HI 1 Hv
 2 Q can be detrimental to the solvers stability and create numerical error
J 3 Q Re Q
   p  in the solution. To address this issue, a third-order Roe scheme [49]
2t 1 3Q 4Q  Qn1
n
with the van Albada flux limiter [50] was employed near shocks. This

3 2t J hybrid approach was developed and successfully used in previous
      
p 1 p p 1 p 1 p work for a supersonic turbulent compression corner [14]. During
6 FI Fv  6 GI Gv  6 HpI Hv Spc (9)
Re Re Re each subiteration of the solver, the shock location was identified
by the pressure gradient detector developed by Swanson and
where FQ, GQ, and HQ are flux Jacobians; and 2 , 2 , Turkel [51]:
2 , 6 , 6 , and 6 represent the second-order and sixth-order finite-
difference operators in , , and , respectively. Equation (9) jpi1 2 pi  pi1 j
was employed to advance the solution in time, such that Qp1 is the  ;

1 
jpi1 pi1 j 
pi1  2pi  pi1 
p  1 approximation to Q at the n  1 time level Qn1 , and Q   (12)
Qp1 Qp . For p  1, Qp  Qn , and as p , Qp Qn1 . > 0.05; Roe scheme
Second-order-accurate backward-implicit time differencing was > 0.05 compact scheme
used to obtain temporal derivatives.
The implicit segment of the algorithm, i.e., the left-hand side of where pi is the pressure at grid point i in the specified direction, and
Eq. (9), incorporated second-order-accurate centered differencing for is a constant that can be varied from 0.5 to 1.0 (for this work;
all spatial derivatives and used nonlinear artificial dissipation [43] to  0.5). Once the shock was located, a five-point stencil was
augment stability. For simplicity, the dissipation terms are not shown established around the shock, and the inviscid fluxes from the Roe
in Eq. (9). Efficiency was enhanced by solving this implicit portion of scheme were substituted for the existing compact solutions within the
the factorized equations in diagonalized form [44]. The temporal stencil. Because of the upwind nature of the Roe flux-difference
accuracy can be degraded when the diagonal form is used, and so scheme, the filtering technique was not used in this region. Figure 3
subiterations were employed within each time step to minimize shows instantaneous Mach contours, and the resultant computational
any degradation of the temporal solution. Any deterioration of the grid was modified to show where the Roe scheme replaces the high-
solution caused by the use of artificial dissipation and by lower- order compact scheme.
 BISEK, RIZZETTA, AND POGGIE
Fig. 3 Instantaneous Mach contours with the shock-capturing stencil for Mach 2.25 flow over a 24 deg ramp.
As a result of the hybrid approach, the high-order compact scheme
captures the fine-scale structures in the shock-free regions of the
turbulent flow, while the Roe scheme accurately simulates the flow
near the shock. Because of the high sensitivity of the shock detector,
the Roe scheme was not employed in the boundary layer and was only
applied in the inviscid region of the flow, as shown in Fig. 3b.
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
D. Large-Eddy-Simulation Approach
In the LES approach, physical dissipation at the Kolmogorov scale
is not represented, thereby allowing for less spatial resolution and a
savings in computational resources. For nondissipative numerical
schemes, this leads to an accumulation of energy at high-mesh wave
numbers and ultimately to numerical instability. Traditionally, SGS
models are employed as a means to dissipate this energy. In the
present methodology, the effect of the smallest fluid structures is
accounted for by a high-fidelity implicit large-eddy simulation
(HFILES) technique, which has been successfully used for a number
of turbulent and transitional computations. The present HFILES
approach was first introduced by Visbal and Rizzetta [48] and Visbal
et al. [52] as a formal alternative to conventional methodologies, and
is based upon the high-order compact differencing and low-pass
spatial filtering schemes. This technique is similar to monotonically
integrated large-eddy simulation (MILES) [53] in that it relies upon
the numerical procedure to provide the dissipation that takes place on
the unresolved scales. In contrast to MILES, dissipation is provided
by the Pad-type low-pass filter only at high spatial wave numbers,
where the solution is not resolved. This provides a mechanism for the
turbulence energy to be dissipated at scales that cannot be accurately
represented on a given mesh system, in a fashion similar to SGS
modeling. For purely laminar flows, filtering may be required in
certain circumstances to maintain numerical stability and to preclude
a transfer of energy to high-frequency spatial modes. The HFILES
methodology thereby permits a seamless transition from LES to DNS
as the resolution is increased. In the HFILES approach, the unfiltered
governing equations may be employed, and the computational
expense of evaluating SGS models, which can be substantial, is
avoided. This procedure also enables the unified simulation of
flowfields where laminar, transitional, and turbulent regions coexist
simultaneously.
It should also be noted that the HFILES technique may be
interpreted as an approximate deconvolution SGS model [54], which
is based upon a truncated series expansion of the inverse filter
operator for the unfiltered flowfield equations. Mathew et al. [55]
have shown that filtering provides a mathematically consistent
approximation of unresolved terms arising from any type of
nonlinearity. Filtering regularizes the solution and generates virtual
SGS model terms that are equivalent to those of an approximate
deconvolution.
E. Boundary Conditions
In the present computations, the flow transitioned from laminar to
turbulent on a flat plate situated upstream of the ramp. The flow and
boundary conditions for this portion of the simulation were consistent
with previous studies by Rai et al. [56], Rizzetta and Visbal [57], and
Pirozzoli et al. [58], which investigated supersonic flow on a flat plate
at Mach 2.25. An extrapolated upper boundary condition was used to
1793
accommodate the growing boundary-layer thickness as the flow
transitioned. Grid stretching (i.e., buffer layer [59] or sponge [33])
was used near the upper and exit boundaries to transfer the energy to
higher spatial wave numbers, where the spatial filter removed it from
the computation.
Periodic boundary conditions were applied at the spanwise
boundaries and used a five-point overlap of grid points in the span.
The five-point overlap maintained high-order differencing and
filtering. The inflow boundary was specified using a solution to the
compressible laminar boundary-layer equations [60], with the inflow
boundary-layer height scaled to the reference length l and freestream
conditions applied outside the boundary layer. An extrapolated
boundary condition was used for the exit boundary. Along the wall
surface, a no-slip velocity boundary condition was imposed with an
isothermal wall set to the nominal adiabatic wall temperature. The
surface pressure was computed by enforcing a zero wall-normal
derivative to third-order spatial accuracy.
III. ResultsInflow
The development of a fully turbulent boundary-layer flow was
achieved on a flat plate for Mach 2.25. The supersonic flow over a flat
plate studied in this work was similar to a previous computation [57],
except the grid was significantly more refined, and the counterflow
body-force bypass-transition method was used instead of the
blowing-and-suction method [56], because the latter introduced un-
wanted discrete frequencies into the energy spectra. The reference
conditions for the case are listed in Table 1, which were developed
from a 1955 experiment by Shutts et al. (case 55010501, compiled by
Fernholz and Findley [61]).
Reference conditions for the present computations were l 
6.096 104 m, u  588 ms, M  2.249, and Rel  15; 240.
In the analysis of the results that follow, the solution flow variables
were decomposed into time-mean values and fluctuating components
(i.e., u  um  u 0 , where u 0 is the fluctuating component). Time-
average and time-dependent data were collected for 10 flow-through
times, where one flow-through time is defined as the time for the
freestream flow to traverse the computational domain.
Using the parameters listed in Table 1, a computational domain
was developed to support LES using the guidelines recommended by
Georgiadis et al. [6]. As previously mentioned, the nondimensional
reference length l was set to the incoming boundary-layer height
(i.e., l  x0 ), and a Cartesian coordinate system was established
with its origin corresponding to the upstream location of the
Table 1 Flow conditions for Mach
2.25 air flow over a flat plate
Parameter Value
M 2.249
u 588 ms
T 305R
Tw 580R
p 23,830 Pa
Rem 2.5 107 m1
Re 29305300
 1794 BISEK, RIZZETTA, AND POGGIE

100 normal from the wall to minimize numerical error when computing
surface gradients. An elliptic solver was used on the resulting
grid point distribution to ensure smooth and continuous point
75 distributions throughout the domain, particularly near the ramp
corner. The domain was 5 wide, with constant grid spacing in the
spanwise direction. It was shown in previous work [62] that this
y[ ]

50 width was sufficient to decorrelate the spanwise periodic boundary

Exp. data
conditions.
A spatial-resolution study was conducted using three different
25 grids. Although the concept of grid independence does not exist for
fe r LES [63], it is possible to show that some of the time-mean quantities
Buf
converge, or approach convergence, with adequate resolution. This
0
occurs because grid refinement allows for finer features to be
0 25 50 75 100 125 150 175
captured, thus changing the instantaneous data, while bulk quantities
Trip Constant x[ ]
that are time-averaged and less sensitive to the behavior of the fine
Fig. 4 Schematic of the three regions of the domain. structures should converge. Note that some statistics, such as the
those characteristic of the viscous superlayer, would not be captured
computational domain. The streamwise extent of the domain was expect in true DNS.
approximately 160 l and was split into three regions, as shown in Because the major limitation for LES is computational cost
Fig. 4. Note that only a fraction of the mesh points are shown in Fig. 4 associated with grid resolution, the fine grid was developed first
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

for clarity.
The first region (labeled trip), 0 x 3 l, contained the
counterflow body-force model. The region required a highly refined
grid to accommodate the bypass-transition method without
producing detrimental behavior associated with the strength of the
counterflow trip source term. As such, the recommendations made by
Mullenix et al. [35] were followed, and the grid was monotonically
refined from the leading edge to x  2 l, which corresponds to the
start of the refined portion of the region. The refined section was 1.0 l
long and contained 101 uniformly spaced points.
The second region (labeled constant) contained the rest of the
flat plate and the leading portion of the ramp. The ramp corner was
situated at x  75 l, with the downstream end of the region
corresponding to x 110 l. The grid was monotonically stretched
from the end of the trip region (i.e., x  3 l) to the start of the
constant grid spacing used in the second region (i.e., x  5 l). The
streamwise grid points were uniformly spaced, except for the grid
points immediately adjacent to the ramp corner, which contained a
slight refinement to facilitate a smooth grid distribution of points
through the ramp corner.
The third region (labeled buffer) rapidly stretched the grid along
the ramp to the exit boundary. The grid spacing in the normal
direction was specified at the wall boundary such that y < 1 and
then monotonically stretched using a hyperbolic tangent expansion.
As seen in Fig. 4 and Table 2, this approach resulted in the majority of
the grid points being located within the boundary layer. The upper
boundary of the computational domain was 25 l high for the flat-
plate portion of the domain but increased above the ramp to
accommodate the strong oblique shock. The points were projected
because it corresponded to the most resolution that could be afforded
for the current study. Once the fine grid was constructed, the coarse
grid was created by removing every other point in all three directions,
except for the streamwise point-distribution in the trip region,
because the resolution specified there was required to prevent
potentially anomalous behavior associated with the bypass-transition
method. The medium grid followed the same methodology, except
it had 1.5 times the points of the coarse grid while preserving the grid-
scaling ratio. Table 2 lists the parameters for the grids. The friction
velocity used to determine the inner length scales (i.e., x , y ,
and z ), was slightly different for each grid because transition was
delayed for the coarser grids.
The counterflow trip was centered in the middle of the refined trip
region. In contrast to typical practice for subsonic flows, where the
height of the counterflow trip model is typically on the order of the
incoming boundary layer [64], the trip model for this supersonic
scenario was positioned such that it barely extended outside the
viscous sublayer. The length-to-height ratio of the triangular force
region was 4 1 and had a length of 0.125 l. To achieve transition, the
resultant counterflow body force must be strong enough to generate a
region of reverse flow. For this work, all grids achieved transition
using a value of D  6.2.
Solutions were obtained using a nondimensional time step
t  0.005, which results in t  Rel
uu 2
ul t  0.19. Figure 5
plots the time-mean spanwise-averaged skin-friction coefficient cf
and the boundary-layer momentum thickness for the flow as it
transitions along the flat-plate region.
As seen in Fig. 5, the incoming laminar flow is disturbed by the
counterflow trip (x  2.5 l). The trip was strong enough to produce a

Table 2 Parameters for the various grids used for Mach 2.25 turbulent
flow over a 24 deg ramp
Parameter Coarse Medium Fine
XYZ 160 l 25 l 5 l 160 l 25 l 5 l 160 l 25 l 5 l
Nx Ny Nz 721 131 141 1017 196 209 1312 261 277
N (total) 13.3 106 41.7 106 94.9 106
N x
x 2 l 30 30 30
N x
x 3 l 130 130 130
N x
x 5 l 150 160 170
N x
x 110 l 21 31 41
x (refined trip) 0.01 0.01 0.01
x (constant) 0.2 0.15 0.1
x (constant)a 46.1 35.5 23.8
yw 0.004 0.003 0.002
y w (constant)
a
0.92 0.71 0.48
N y
y
x0  70 104 139
N y
y
x60 l  85 127 169
z 0.0362 0.0272 0.0181
z (constant)a 8.4 6.4 4.3
a
These quantities were computed at x  60 l
 BISEK, RIZZETTA, AND POGGIE 1795

0.0025 1 30 + +
Coarse u =y
Medium
Fine
Turbulent Theory, c f
0.002 0.8

20
0.0015 0.6

[ ]
Compare with Exp. data

+
u
cf

+ )+ 5.5
0.001 0.4 (y
10 * log Coarse grid
.5
+ =2 Medium grid
u
Fine grid
0.0005 0.2
Comp. - Rai et al. [56]
Exp. - Shutts et al. [61]
Laminar
0 0 0 0 1 2 3
0 5 10 15 20 25 30 35 40 45 50 55 60 65 10 10 10 10
x [ ] y+
Fig. 5 Spanwise-averaged time-mean skin-friction coefficient cf and Fig. 6 Spanwise-averaged time-mean streamwise velocity profiles using
boundary-layer momentum thickness vs streamwise location. The the Van Driest transform and normalized by friction velocity at x
60.
vertical red line indicates where the computational solutions are
compared to experimental data.
the logarithm region. The coarse grid result is indicative of an
underresolved grid or a solution that has not yet achieved equilibrium
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

small separated region (i.e., the region where cf < 0). The separated
region generated instabilities that grew nonlinearly during the
transition process. The transitioning flow approached the theoretical
curve for equilibrium turbulence near x 50 l, which is consistent
with the domain length required to transition the flow studied by
Mullenix et al. [35]. The theoretical curves in the figure correspond to
the laminar compressible Blasius solution [60] and the turbulent
correlation of White and Christoff [60]. The fine grid transitioned
farther upstream than the coarser grids because the additional
resolution allowed for the development of smaller scale structures.
Although the skin-friction coefficient displayed slight differences,
the boundary-layer momentum thickness was very similar for the
three different grids, and so only the fine grid was included in Fig. 5
for clarity. As the flow transitions, the boundary-layer momentum
thickness grows. At x  60 l, the boundary-layer momentum
thickness is  0.32 l, and so Re 4900, which was within
the range of momentum thicknesses measured in the experiment.
The time-mean spanwise-averaged quantities from the LES were
extracted at this location and compared to the experimental measure-
ments. In addition to the present computations, the computational
results from Rai et al. [56] were also included for comparison.
The streamwise velocity profile was transformed using the van
Driest transformation [65]. The transformed streamwise velocity was
nondimensionalized by the wall-friction velocity u  uvD u. The
profiles were plotted versus the nondimensional inner length scale
y  yu w . Figure 6 plots the van Driest-transformed velocity in
the near-wall region at x  60 l for the various grids and includes
the solutions from Rai et al. [56]. In addition to the experimental
measurements by Shutts et al. (Case 55010501, compiled by
Fernholz and Findley [61]), the experiment data set also included
supersonic flow measurements of Elena and LaCharme [66], which
were collected under similar flow conditions using Laser Doppler
Velocimetry (LDV) and Hot-Wire Anemometry (HWA). The numer-
ical solutions match both the inner layer and logarithmic profiles,
although the coarse grid overpredicted the expected solution within
turbulence.
The root mean square streamwise velocity fluctuations are shown
for both the near-wall and outer region of the boundary layer in Fig. 7.
Since the experimental data by Shutts et al. (Case 55010501,
compiled by Fernholz and Findley [61]), did not include fluctuating
measurements, the near-wall results in Fig. 7a include comparisons to
the incompressible experiment of Karlson and Johansson [67], which
was carried out using LDV techniques for Re  2420. This
comparison is consistent with previous work by Rizzetta and Visbal
[57], who demonstrated that the comparison to incompressible data
was valid because the compressibility effects were not strong for this
flow (i.e., Morkovins hypothesis [68] applies).
For the outer region of the boundary layer, the fluctuating velocity
profile is presented versus distance from the surface, as seen in
Fig. 7b. The distance was normalized by the boundary-layer height
at the location of comparison to the experiments (i.e., x  60 l) to
account for the different boundary-layer heights. The experimental
measurements shown were collected by Elena and LaCharme [66]
using LDV and HWA. The figure shows that the numerical solutions
are in good agreement with the experimental data. Additional valida-
tion of the fully turbulent boundary-layer flow is available in [62].
One-dimensional spectra of the velocity components and instan-
taneous pressure were collected along a line in the homogeneous
spanwise direction at several discrete streamwise locations within the
boundary layer. Each location was located a distance of 1.0 l normal
to the surface, starting at x  55 l and ending on the ramp region
where the grid starts to coarsen due to grid stretching (i.e., x  110).
At x  60 l, the boundary-layer height  1.71 l for the fine grid,
so a distance of 1.0 l normal from the wall corresponds to
approximately 60% of the boundary layer. Data sampling occurred
after every computational time step and was collected for 10 flow-
through lengths.
The power spectral density (PSD) was computed for both
nondimensional frequency and spanwise wave number to confirm
that the grids had sufficient resolution to capture a portion of the

0.5 Coarse grid 0.2 Coarse grid

Medium grid Medium grid
0.4 Fine grid Fine grid
Comp. - Rai et al. [56] 0.15 Comp. - Rai et al. [56]
Exp. - Shutts et al. [61] Exp. - Shutts et al. [61]
0.3
u / um

u2

0.1
2

0.2
0.05
0.1

0 0
0 10 20 30 40 50 0 0.2 0.4 0.6 0.8 1
y+ y/
a) Inner scaling b) Outer scaling
Fig. 7 Spanwise-averaged time-mean streamwise velocity fluctuations vs inner and outer scaling at x
60.
 1796
inertial subrange. The PSD curves were developed using the standard
approach as outlined in [69]. At each spanwise location, a Hanning
window was applied for the entire data set to suppress side-lobe
leakage. The PSD of the turbulent kinetic energy (TKE) was
computed by summing the PSD of each velocity component
and multiplying by half, i.e., PSD
TKE  PSD
u 0   PSD
v 0 
PSD
w 0 2.
The PSD versus spanwise wave number was computed in a similar
manner as the PSD versus nondimensional frequency, except that the
Fourier transform was applied along the homogenous spanwise data
set (excluding the overlap points) for each time step. It is important to
note that the raw signal was not windowed, and the overlap points
were excluded because the solution was periodic in the spanwise
direction. Windowing periodic data would introduce a nonphysical
bias favoring the end points of the Fourier transform. Figure 8 plots
the PSD of the TKE versus normalized frequency and spanwise wave
number for the different grids.
As seen in Fig. 8, all three grids capture a portion of the inertial
subrange, which is indicated by the region with a 53 slope. The
extent of the inertial subrange was limited because the Reynolds
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
number was low for the present study. As expected, the fine grid
resolved more of the inertial subrange before the spatial filter
eliminated the highest, underresolved frequencies.
IV. 24-Degree Ramp
The incoming turbulent boundary-layer flow continued along the
flat plate until it encountered the 24 deg ramp. The flows sharp
turning angle resulted in the formation of an oblique shock above the
boundary layer. Two shock feet form that correspond to the separa-
tion point and reattachment of the separated flow, as illustrated in
Fig. 1. This behavior gives rise to the well-known -shock structure.
Because of the turbulent inflow, the shock front oscillated in the
streamwise direction. The instantaneous separated region is visible in
Fig. 9, which shows an isosurface of the incompressible Q-criterion
[70] (Qcriterion  12 2 S2 ) that was colored by the u velocity. The
Q criterion is the second invariant of the velocity gradient tensor,
which compares the vorticity to the strain rate S and is commonly
used to highlight organized structures in turbulence. The figure also
includes an isosurface of pressure, which is colored to highlight the
shock.
Figure 9 highlights some of the the sub--scale turbulent structures
flowing over the ramp corner. These structures were moving faster
higher in the boundary layer (as indicated by the lighter coloring). As
the flow moved past the shock, it slowed down. The rapid decrease in
velocity and increase in density due to the shock caused the structures
to expand and start to readjust toward a new equilibrium turbulent
boundary layer. The dark-colored structures downstream of the
shock, near the ramp corner, indicate where the flow is reversed. As
shown in the figure, the region of reversed flow was extensive.
Although an instantaneous solution provides a representative idea
of the separated region, the time-mean solution produces a better
understanding of the extent of the shock-induced separation. A plot
0 0
10
Power Spectral Density of TKE
1
10
2 2
10
3 3
10
4 4
10
5 5
10
Grid
6
10 Refinement
7 7
10
0.001 0.01 0.1 1 10
Normalized Frequency,
a) Spanwise-average
Fig. 8 Power spectral density of turbulent kinetic energy at x
60 l, y
1 l.
BISEK, RIZZETTA, AND POGGIE
Fig. 9 Instantaneous isosurfaces of the Q criterion colored by the u
velocity and an isosurface of the shock.
of the normalized skin-friction coefficient in Fig. 10a shows the
extent of the spanwise-averaged time-mean separation. The data were
normalized by the skin-friction coefficient at x  60 l because each
solution had a slightly different incoming boundary-layer profile. As
previously discussed for Fig. 5, grid resolution affected transition,
with more resolution transitioning the flow sooner. As such, the
boundary layer upstream of separation (i.e., x  60 l) was different
for each grid. For the coarse grid, the boundary-layer height was
 1.92 l; for the medium grid, it was  1.74 l; and for the fine
grid, it was  1.71 l. This difference and the separation location
were also accounted for in Fig. 10a, which shows the time-mean
separation length converging with resolution.
Because there are no experimental measurements of the separation
length for this scenario, the separation length from each simulation
was also compared to the experimental correlation by Zheltovodov
and Schlein [71]. The correlation provides bounds for the separation
length versus Reynolds number based on the boundary-layer height.
Fig. 10b shows that the separation lengths for the three grids are
consistent with other LES solutions and the empirical envelope
reproduced from [72].
Figure 10c shows the surface pressure normalized by the
postshock surface pressure. As seen in the figure, the pressure rises
rapidly at the onset of separation, plateaus through the center of the
separated region, and finishes rising as it approaches the inviscid
postshock pressure. The start and end of the pressure plateau
correspond to local maxima of the skin-friction coefficient in the
separated region. This result along with all the previous analysis
implied that the fine grid was sufficiently resolved for LES. As such,
the remaining figures and analysis will only include analysis from the
fine grid.
Although the main focus of this work was to investigate a reduction
of the shock-induced separation, it was also important to explore the
10
Coarse Coarse
Medium 1 Medium
Power Spectral Density of TKE
Fine 10 Fine
10
10
-5/3 slope 10 -5/3 slope
10
6
10
10
100 1 10 100
Wave Number
b) Time-mean
 BISEK, RIZZETTA, AND POGGIE 1797

Comp. data, Zheltovodov and

Schlein [72]
1.5 Coarse grid Exp. envelope, Zheltovodov
Medium grid and Schlein [72]
Fine grid 30 Coarse grid
1 Medium grid
25 Fine grid
c f /c f,x=60
20
0.5

Lsep
separation reattachment 15

10
0
5

-0.5 0 4
-10 0 10 20 10 105 106 107
(x-xsep )/ x=60 Re
a) Skin-friction coefficient b) Separation length as defined in [71] and [72]

1
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

0.8

0.6
p w/p w,i

0.4
Coarse grid
0.2 Medium grid
Fine grid
0
-10 0 10 20
(x-xsep )/ x=60
c) Surface pressure
Fig. 10 Spanwise-averaged time-mean nondimensional skin-friction coefficient, separation length, and surface pressure.

frequency content within and after the separated flow because its ultra-low-frequency content typically observed in experiments
behavior as a frequency-selective amplifier, as suggested by Plotkin in the vicinity of flsep u  0.03 [22]. However, it is important
[1], Poggie and Smits [2], and Touber and Sandham [3], needed to note that the PSD curves may not have sufficient time-history
verification for this scenario. Figure 11 plots the PSD of TKE for the to adequately resolve the ultra-low-frequency content because the
fine grid upstream of separation (  60), above the separated region signal contained eight wavelengths for the frequency f lu
(70 80), and after reattachment (  100), all at  1. Note 0.003. As a result, the PSD was not capturing the entire behavior of
that  60,  1 corresponds to x  60.0 l, y  1.0 l, whereas the signal in this range because it was being masked by noise, which
 100,  1 corresponds to x  97.4 l, y  11.1 l (or is evident in the high scatter visible in the curve.
y yw  1.0 l), due to the ramp shown in Fig. 4. From the figure,
it is clear that the separated region amplified the total TKE, which was
consistent with recent observations by Touber and Sandham [3].
Figure 11b shows the PSD of pressure fluctuations at a few V. Control
locations, including a location just downstream of the time-mean A model magnetically-driven surface-discharge actuator was
separation (  70). As seen in the figure, the pressure fluctuations positioned on the flat plate just upstream of the time-mean spanwise-
have a weaker signal than the TKE (i.e., the magnitude of the PSD is averaged separation point to investigate the reduction of the
lower), and the signal itself has more noise. That said, the flow just separation length caused by the turbulent SBLI. The actuator
downstream of separation (  70) experienced a sudden increase in imparted streamwise momentum into the boundary layer, which
the PSD of pressure fluctuations at the lower end of the spectra. The suppressed separation by increasing the fullness of the velocity
rise occurred near f lu 0.003, which corresponds to the profiles and decreasing the boundary-layer displacement thickness.
0 1
10 10
= 60 = 60
= 70 = 70
Power Spectral Density of Pressure

2
10
Power Spectral Density of TKE

1 = 75 = 100
10
= 80 3
= 100 10

2 4
10 10

5
10
3
10
6
10

4 7
10 10
0.001 0.01 0.1 1 10 0.001 0.01 0.1 1 10
Normalized Frequency, Normalized Frequency,

a) Turbulent kinetic energy b) Pressure

Fig. 11 Spanwise-averaged PSD vs nondimensional frequency at various locations (
1). Note that the station xl
75 corresponds to the corner.
 1798 BISEK, RIZZETTA, AND POGGIE
The actuator model parameters used in the present study were based
on the experimental work of Kalra et al. [23].
Because of the large number of variables present in the
phenomenological actuator model described in Eqs. (7,8), the
parameter space was first explored using two-dimensional k
RANS simulations to determine a set of control parameters that
would reduce the extent of separation. Because the computation
domain is homogenous in the spanwise direction, the plasma
controller was chosen to be uniform in that direction to remove the
additional complexity of spanwise effects. The magnetic body force
and energy deposition was nondimensionalized by a length equal to
the width of the domain
5 l. The controller was 5 l long in the
streamwise direction and was centered near the separation point,
which was consistent with the approach used by Atkinson et al. [39].
The controller volume was centered 0.1 l above the wall and had a
normal extent of 0.04 l. The vertical extent of the controller volume
was consistent with experimental work by Zaidi et al. [73], who
reported that the plasma column was located near the wall at
approximately 10% of the boundary-layer height. Equation (7) set the
volume and distribution of the force and energy deposition from the
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
phenomenological controller, such that xc  69 l; yc  0.1 l;
a  2.5 l; b  0.04 l; m  10, and n  10. The total nondimen-
sional magnetic body force deposited into the plasma column was
L  0.075, which was consistent with the nondimensional total
force produced in the experiment by Kalra et al. [23].
Three control scenarios were explored in the present study. All
three use a nondimensional magnetic body force of L  0.075,
which was confined to the streamwise direction (i.e., Lx  0.075;
Ly  0; Lz  0). It was assumed that the fine grid had sufficient
resolution to capture the small-scale structures present in the flow
when the controller was active since control reduced the separation
length. The simulations were run using a nondimensional time
step t  0.005 and were started from the equilibrium baseline
simulation. Initial transients associated with the introduction of the
controller were allowed to propagate out of the domain by running
the simulation for three flow-through lengths before statistical
information was collected. Time-dependent data were collected
for 10 additional flow-through lengths and the time-mean results
correspond to this time range.
The first scenario allowed for momentum transfer by the mag-
netically driven surface discharge but excluded the reversible work
produced by the magnetic force. Although it is a strong assumption to
ignore the effects of energy deposition, these results provide an upper
bound for the expected performance of a magnetically-driven surface-
discharge controller. This case is referred to as the perfect controller
because it provides all of the advantages of direct momentum
transfer without the unwanted heating of the boundary layer.
The second case was consistent with the perfect controller, except
the actuator was pulsed with a 50% duty cycle. The Strouhal number
of the pulsing frequency was defined based on the reference length
and freestream velocity, St  ufl  0.28. This frequency corre-
sponds to the time required for the inviscid flow to traverse a distance
of 1.8 l, which was approximately equal to the boundary-layer height
at x  60 l. Pulsing the actuator at a 50% duty cycle is advantageous
for practical applications because it reduces the overall power
0.003
0.002
0.001
cf
0 0.4
baseline
-0.001 perfect controller
pulsed controller
real controller
-0.002
60 70 80 90 100 110
controller x [ ]
a) Skin-friction coefficient b) Nondimensional surface pressure
Fig. 12 Spanwise-averaged time-mean skin-friction coefficient and nondimensional surface pressure. Note that the corner is at x
75.
consumption by half. In addition, subsonic simulations of airfoil flow
control by Rizzetta and Visbal [64] have shown that pulsing an
actuator can achieve the same effect as leaving the actuator
continuously on if the pulsing frequency is much higher than the time
needed for the flow to respond to the instantaneous control. This
scenario is referred to as the pulsed controller.
The third case, which was a more-realistic representation of the
actual plasma-based flow controller, included Joule heating effects.
In this scenario, the time-mean total reversible work from the
nondimensional magnetic body force was 0.075, and Joule heating
J  0.025 was also added to the total energy equation via Eq. (8). The
combination of reversible work and Joule heating corresponds to a
gliding surface-discharge controller with a direct thermalization of
2.5%, based on the nondimensional total input power in the electrical
circuit used in the experiment by Kalra et al. [23]. This scenario
assumed the force and energy deposition were constantly on and is
referred to as the real controller. Note that only a fraction of the total
power used in the experiment by Kalra et al. [23] was absorbed by the
flow because most of the power was lost permanently to vibrational
modes [36].
Figure 12 shows the time-mean, spanwise-averaged skin-friction
coefficient and surface pressure normalized by the postshock surface
pressure for all three controllers along with the baseline results. As
seen in the figure, the perfect controller performed the best by
reducing the separation length by more than 75%. The other two
cases attained similar results, each achieving nearly 50% reduction in
the separation length. However, the pulsed case required half the
power, and so it was the most efficient when accounting for both the
reduction in separation length and the power used. All three
controllers show an increase in the skin friction over the region where
the controller was applied (i.e., 66.5 l x 71.5 l) because the
magnetic body force accelerated the near-wall flow over that region.
This increase corresponded with a reduction in surface pressure,
which is visible in Fig. 12b.
The reduction in the size of the separated region was caused by
local streamwise acceleration of the flow near the baseline time-mean
separation. This can be seen in Fig. 13b, which shows the Q criterion
(Qcriterion  1.5) for the perfect control case. Close examination of
the turbulent structures just upstream of separation, compared to the
structures in the baseline flow shown in Fig. 13a, revealed that the
structures were more aligned with the streamwise direction with
control. In addition, these structures have a much higher velocity, as
indicated by their lighter color. Consistent with the baseline case,
once the flow traveled through the shock (and was outside the
influence of the controller), it slowed and readjusted toward its new
equilibrium turbulent profile.
The colored isosurface of pressure seen in Fig. 13 highlights the
shock front and illustrates just how significantly the shock structure
has been changed due to control. As the separation length decreased,
the -shock feet moved closer together. This caused the shock to
become sharper and the shock angle to increase (particularly near the
boundary layer). These changes are easily observed in Fig. 14, which
shows instantaneous planar contours of Mach number for each of the
scenarios. The figure also includes a solid line indicating Mach 1 and
dark (blue online) areas in the separated region where the flow is
1
0.8
0.6
p w/p w,i
baseline
0.2 perfect controller
pulsed controller
real controller
0
60 70 80 90 100 110
x [ ]
controller
 BISEK, RIZZETTA, AND POGGIE 1799

Fig. 13 Instantaneous isosurface of the Q-criterion colored by the u velocity and an isosurface of the shock.

reversed. The figures show that the controller decreased the suggest that the flow did not have sufficient time to fully recover to
boundary-layer height, which generated a weak expansion in the the baseline state during the period when the controller was inactive.
inviscid flow above the controller. The expansion slightly accelerated As such, it may be possible to pulse the controller with more power
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

the inviscid preshock flow near the boundary layer.
Figure 15 shows spanwise-averaged time-mean contours of
temperature for each scenario. The temperature within the perfect
controller region dropped as the flow was accelerated because
the model did not modify the total energy equation. As the tempera-
ture fell, so did the boundary-layer height, which is visible in the
temperature contours. The addition of streamwise momentum im-
parted by the controller also caused the boundary-layer displacement
thickness to decrease, which made the flow less susceptible to
separation.
The pulsed case in Fig. 15c demonstrated similar characteristics as
those exhibited by the perfect controller in Fig. 15b, except the 50%
duty cycle exerted less control (on average), and so the resultant
temperature drop was not as apparent but was still present. These
results along with the instantaneous planar Mach contours in Fig. 14c
at the same frequency to achieve the same result as the perfect
controller. This scenario may be explored in future work.
The real scenario always added energy in the region of the
controller through reversible work and Joule heating, as seen in
Fig. 15d. The additional heat accumulated over the length of the
controller, and so the boundary-layer height did not decrease as
dramatically as the other control cases, even though the magnetic
body force was accelerating the near-wall flow.
Figure 15 also includes streamlines that highlight the mean flow
behavior near the compression corner. As seen in Fig. 15b, the perfect
case has essentially eliminated the spanwise-averaged time-mean
recirculating core, even though the mean skin-friction coefficient was
negative over part of the range. The pulsed and real controllers in
Figs. 15c, 15d both exhibit a significant reduction in the size and
strength of the recirculating bubble, though the recirculating core still

Fig. 14 Instantaneous midspan Mach contours with the solid line indicating Mach 1.
 1800 BISEK, RIZZETTA, AND POGGIE
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

Fig. 15 Spanwise-averaged time-mean temperature contours and mean-flow velocity streamlines. The streamlines have been terminated upstream of
x
71.

exists near the ramp corner. The slight cooling of the inviscid flow just
upstream of the shock for the control cases was due to the weak
expansion fan discussed previously.
The temperature and Mach contours are consistent with the results
in Fig. 16, which plots the boundary-layer height for each case. The
figure also includes streamwise velocity profiles at various locations.
As seen in the figure, the controller significantly changed the velocity
profiles, especially near x  70 l, which showed the baseline flow
separated while each controller delayed separation by dramatically
increasing the fullness of the profile. Downstream of reattachment
(i.e., x  85 l), all the solutions have recovered to similar profiles,
though the cases with control were fuller. The maximum streamwise
velocity was slightly lower with control because the increased shock
angle corresponds to a greater increase in entropy as the flow
propagates through the shock. As a result, the postshock temperature
and density were higher and the Mach number was lower. These
changes correspond to a higher streamwise mass-flow rate near the
surface with control, whereas the mass-flow rate was slightly lower
throughout most of the postshock boundary layer.
Using the spanwise-averaged time-mean solutions, the total
TKE at a given streamwise location was computed by integrating
the TKE along a line normal to the surface, total TKE  0
u
02
v
02  w
02 2 d, where corresponds to the top boundary of the
computational domain. The same procedure was also completed to
determine the average total pressure, pt  p
1 
1M2
2
1 . The average total pressure was limited to the 5 l from the the baseline PSD of TKE curve increasing in magnitude with lower
surface, average pt  15 l 50 l pt d, because the controller only
affects the first few boundary-layer thicknesses. The total pressure was
evaluated because it is proportional to the work available to the
combustor, which is a critical parameter when designing an internal
flow path of an air-breathing configuration. The total TKE and the
average total pressure were plotted versus distance along the wall in
Fig. 17. As seen in the figure, the total TKE decreased monotonically
with control, while the average total pressure was significantly
higher throughout the ramp corner, but this improvement was less
impressive downstream of reattachment. The perfect controller
exhibits the largest change in total pressure near the corner, but its
steeper shock angle, which is visible in Fig. 16, produced a greater
entropy rise through the shock and, consequently, a lower stagnation
pressure loss downstream of reattachment. This stagnation pressure
loss allowed the baseline case to achieve a slightly higher average
total pressure by the end of the computational domain compared to the
perfect controller. The two other control scenarios also mitigated the
stagnation pressure loss through the ramp corner but maintained a
shallower shock angle, which yielded a moderate improvement to the
average total pressure, even far downstream of reattachment.
Although the main purpose of the controller was to reduce
the extent of the time-mean separation, it was also of interest to
investigate the effect that control had on the low-frequency content in
the frequency spectra. Figure 18, shows the PSD of TKE for the
baseline case and the three controllers at various locations: upstream
of separation (  60), over the separated region (70 80), and
far downstream of reattachment (  100), all at  1. The most
obvious difference in the curves can be seen in Fig. 18b, which shows
frequency, while the control cases appear consistent with the
upstream flow. This was expected because each of the controllers
delayed the time-mean separation beyond x  70 l and the separated
 BISEK, RIZZETTA, AND POGGIE 1801

6
baseline

y [ ]
4 perfect
pulsed
2 real

0
65 70 75 80 85
x [ ]

x = 65 x = 70 x = 75 x = 80 x = 85
2 2 2 2 2

1.5 1.5 1.5 1.5 1.5

(y-yw)/ (x=60)

(y-yw)/ (x=60)

(y-yw)/ (x=60)

(y-yw)/ (x=60)

(y-yw)/ (x=60)
1 1 1 1 1
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248

0.5 0.5 0.5 0.5 0.5

0 0 0 0 0
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
um um um um um
Fig. 16 Spanwise-averaged time-mean boundary-layer height near the compression corner (top) and mean-flow streamwise velocity profiles at various
locations (bottom).

flow would add low-frequency content. However, the result also controller. However, looking at the PSD of the pressure fluctuations
implies that the unsteady shock front must not have spent much time in Fig. 19, it is clear that the pulsing frequency dominates the curve at
upstream of the location during the simulation because the movement the sampling location above the controller (  70). Note that the
of the shock front upstream of x  70 l would have added some level of noise in the PSD of pressure fluctuations in Fig. 19 has been
large-scale, low-frequency content to the spectra. reduced by using a shorter signal. This was accomplished by
In the pulsed and real cases, the extent of the time-mean separation applying the Hanning window to a segment length of two flow-
was still fairly extensive (see Fig. 12), and so each of their PSD through lengths (200 lu ) with a 50% overlap of the raw signal.
of TKE curves had a higher value than the perfect controller The nine PSD curves were then averaged to produce Fig. 19. All other
downstream of controllers deposition volume. For the perfect PSD analyses presented in the paper were generated using a single
controller, the additional magnetic force was strong enough to signal with a segment length of 10 flow-through lengths.
significantly reduce the amplification of the TKE due to the Figure 19 shows that, although the pulsing frequency dominated
separation, thus eliminating much of the low-frequency amplification the curve where the controller was applied, the discrete frequency
of the incoming boundary-layer PSD. Accordingly, the majority of disappeared at stations further downstream. This was the result of
the remaining amplification of the TKE spectra in this scenario can be the energy from the pulsing frequency cascading into the sur-
attributed to the shock. That said, all three controller scenarios have rounding frequencies and its remaining signal being masked by the
lower PSD magnitudes compared to the baseline for all locations amplification of the PSD curve due to the shock. Because the driving
downstream of control. In addition, the profiles for the controlled frequency of the pulsed scenario did not persist far downstream, its
cases were flatter for the lower-frequency range (i.e., flu < 0.1). potential influence on the postshock flow can be assumed to be small.
A close examination of Fig. 18b shows that the pulsing frequency As seen in Figs. 1215, the magnetically-driven surface-discharge
(i.e., St  flu  0.28) is not readily apparent as a distinct controller was capable of reducing the time-mean separation by
spectral peak for the pulsed control, even though this location locally accelerating the flow near the nominal separation point.
corresponds to the models deposition volume. A detailed investi- Contraction of the separation length reduced the size of the largest
gation revealed that the magnitude of the incoming TKE profile was length-scale structures found in that region and, ultimately, reduced
sufficiently large to mask the discrete signal introduced by the the total TKE in the flow (both in the time-mean separated region

0.07 baseline 1.4

0.06 perfect
pulsed 1.3
0.05 real
Average pt
Total TKE

0.04 1.2

0.03 1.1 baseline

0.02 perfect
1 pulsed
0.01
real
0 0.9
60 70 80 90 100 110 60 70 80 90 100 110
controller X [ ] controller X [ ]
a) Total turbulent kinetic energy b) Average total pressure
Fig. 17 Integrated quantities vs distance along the geometry. The corner is at x
75.
 1802 BISEK, RIZZETTA, AND POGGIE
a) = 60
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
c) = 80
Fig. 18 Power spectral density of the turbulent kinetic energy vs frequency (
1). Note that the corner is at
75.
and downstream of reattachment). In addition, all three controllers
removed the lowest frequency content from the PSD spectra, thus
mitigating localized fatigue loading. Although the perfect controller
was the most effective at reducing the time-mean separation length,
the pulsed strategy accomplished adequate control while using half
the power, thus making it more efficient than the other scenarios
considered in this study. In addition, the driving frequency of the
pulsed controller did not persist downstream, and the flow did not
appear to have sufficient time to recover to its baseline state between
pulses. Thus, it may be practical to consider modulated control
systems due to beneficial power supply tradeoffs.
Although the three control simulations yielded a significant
reduction in the separation length, the frequency spectra still
exhibited a shift toward lower frequencies and an increase in the
TKE, particularly over the remaining separated region. Although a
large portion of this shift can be attributed to the shock, the controller
model has many parameters that could be investigated in future
Fig. 19 Power spectral density of pressure fluctuations for Mach 2.25
flow over a 24 deg ramp with the Pulse controller.
b) = 70
d) = 100
studies to optimize the control strategy for reducing the detrimental
effects of turbulent-shock boundary-layer interaction. This includes
changing the strength, location, and size of the controller and
exploring the use of multiple controllers that are pulsed at different
phases with each other. In addition, it may prove beneficial to explore
finite-span controllers that would develop a pair of vortical structures
to bring high-momentum inviscid freestream flow into the boundary
layer in a manner similar to micro vortex generators. It may also
be useful to consider strategies where the discharge is driven by a
nonuniform magnetic field.
VI. Conclusions
A high-order compact difference scheme and third-order Roe
scheme were used in a hybrid approach to perform large-eddy
simulations for a Mach 2.25 turbulent flow over a 24 deg ramp to
investigate the effect of plasma-based flow control for turbulent-
shock boundary-layer interaction. The computational domain
included a large flat plate upstream of the ramp corner where a
laminar boundary layer was perturbed to transition to turbulence
through a counterflow body-force bypass-transition method.
The fully turbulent boundary layer was analyzed and was found to
agree well with experiments and other computational results. A grid-
resolution study was performed, and it showed that each grid
captured a portion of the inertial subrange of the turbulent kinetic-
energy frequency spectra. Analysis of the frequency spectra showed a
major increase in lower-frequency content as the flow moved over the
separated region, although the spectra did not clearly exhibit the ultra-
low-frequency content that has been observed experimentally in
some turbulent-shock boundary-layer interaction flows.
A model of a magnetically-driven surface-discharge actuator,
based on recent experiments at the Princeton University Applied
Physics Group, was applied to the flow. Several control cases were
examined. The first case assumed zero net energy deposition by the
actuator. In the second scenario, the same actuator was pulsed with a
50% duty cycle and a Strouhal number of 0.28. The third case allowed
 BISEK, RIZZETTA, AND POGGIE
for Joule heating and reversible work such that direct thermalization
by the actuator was 2.5%. For each scenario, the controller was
centered near the time-mean separation point and applied uniformly
across the spanwise-periodic computational domain. All three
controllers effectively reduced the separation length, based on the
skin-friction coefficient, with the first case reducing the separation
length by over 75%. The pulsed and 2.5% thermalization controllers
each reduced the separation length by over 45%. The reduction in
separated length coincided with the elimination of the lowest-
frequency energy content and a lower total turbulent kinetic energy in
the flow.
Acknowledgments
The first author would like to thank D. Garmann for his useful
discussions regarding FDL3DI, signal analysis, and Matlab scripts.
In addition, the authors would like to thank M. Visbal and D.
Gaitonde for their valuable conversations and advice. This work
was sponsored in part by the U.S. Air Force Office of Scientific
Research (AFOSR) under grant LRIR 12RB01COR, monitored
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
by J. Schmisseur, AFOSR/RSA. The computational resources
were supported by a grant of supercomputer time from the U.S.
Department of Defense Supercomputing Resource Center at the U.S.
Army Engineer Research and Development Center (Vicksburg, MS)
and the U.S. Air Force Research Laboratory (Wright-Patterson Air
Force Base, OH).
References
[1] Plotkin, K. J., Shock Wave Oscillation Driven by Turbulent Boundary-
Layer Fluctuations, AIAA Journal, Vol. 13, No. 8, 1975, pp. 1036
1040.
doi:10.2514/3.60501
[2] Poggie, J., and Smits, A. J., Experimental Evidence for Plotkin Model
of Shock Unsteadiness in Separated Flow, Physics of Fluids, Vol. 17,
No. 1, 2005, p. 018107.
doi:10.1063/1.1833405
[3] Touber, E., and Sandham, N. D., Low-Order Stochastic Modeling of
Low-Frequency Motions in Reflected Shock-Wave/Boundary-Layer
Interactions, Journal of Fluid Mechanics, Vol. 671, March 2011,
pp. 417465.
doi:10.1017/S0022112010005811
[4] Babinsky, H., and Ogawa, H., SBLI Control for Wings and Inlets,
Shock Waves, Vol. 18, No. 2, 2008, pp. 8996.
doi:10.1007/s00193-008-0149-7
[5] Lin, J. C., Review of Research on Low-Profile Vortex Generators to
Control Boundary-Layer Separation, Progress in Aerospace Sciences,
Vol. 38, Nos. 45, 2002, pp. 389420.
doi:10.1016/S0376-0421(02)00010-6
[6] Georgiadis, N. J., Rizzetta, D. P., and Fureby, C., Large-Eddy
Simulation: Current Capabilities, Recommended Practices, and Future
Research, AIAA Journal, Vol. 48, No. 8, 2010, pp. 17721784.
doi:10.2514/1.J050232
[7] Shan, H., Jiang, L., Zhao, W., and Liu, C., Large Eddy Simulation of
Flow Transition in a Supersonic Flat-Plate Boundary Layer, 37th AIAA
Aerospace Sciences Meeting and Exhibit, AIAA Paper 1999-0425,
Jan. 1999.
[8] Urbin, G., and Knight, D., Large-Eddy Simulation of a Supersonic
Boundary Layer Using an Unstructured Grid, AIAA Journal, Vol. 39,
No. 7, 2001, pp. 12881295.
doi:10.2514/2.1471
[9] Yan, H., Knight, D., and Zheltovodov, A. A., Large Eddy Simulation of
Supersonic Flat Plate Boundary Layer Part 1, 40th AIAA Aerospace
Sciences Meeting and Exhibit, AIAA Paper 2002-0132, Jan. 2002.
[10] Rizzetta, D. P., and Visbal, M. R., Application of Large-Eddy
Simulation to Supersonic Compression Ramps, AIAA Journal, Vol. 40,
No. 8, 2002, pp. 15741581.
doi:10.2514/2.1826
[11] Garnier, E., and Sagaut, P., Large Eddy Simulation of Shock/
Boundary-Layer Interaction, AIAA Journal, Vol. 40, No. 10, Oct. 2002,
pp. 19351944.
doi:10.2514/2.1552
[12] Touber, E., and Sandham, N. D., Comparison of Three Large-Eddy
Simulations of Shock-Induced Turbulent Separation Bubbles, Shock
Waves, Vol. 19, No. 6, 2009, pp. 469478.
doi:10.1007/s00193-009-0222-x
1803
[13] Adams, N. A., Direct Simulation of the Turbulent Boundary Layer
Along a Compression Ramp at M  3 and Re  1685, Journal of
Fluid Mechanics, Vol. 420, No. 1, 2000, pp. 4783.
doi:10.1017/S0022112000001257
[14] Rizzetta, D. P., Visbal, M. R., and Gaitonde, D. V., Large-Eddy
Simulation of Supersonic Compression-Ramp Flow by a High-Order
Method, AIAA Journal, Vol. 39, No. 12, 2001, pp. 22832292.
doi:10.2514/2.1266
[15] Loginov, M. S., Adams, N. A., and Zheltovodov, A. A., Large-Eddy
Simulation of Shock-Wave/Turbulent-Boundary-Layer Interaction,
Journal of Fluid Mechanics, Vol. 565, Oct. 2006, pp. 135169.
doi:10.1017/S0022112006000930
[16] Wu, M., and Martin, M. P., Direct Numerical Simulation of Supersonic
Turbulent Boundary Layer Over a Compression Ramp, AIAA Journal,
Vol. 45, No. 4, April 2007, pp. 879889.
doi:10.2514/1.27021
[17] Muppidi, S., and Mahesh, K.,DNS of Unsteady Shock Boundary Layer
Interaction, 49th AIAA Aerospace Sciences Meeting, AIAA Paper
2011-724, Jan. 2011.
[18] Priebe, S., and Pino Martn, M., Low-Frequency Unsteadiness in the
DNS of a Compression Ramp Shockwave and Turbulent Boundary
Layer Interaction, 48th AIAA Aerospace Sciences Meeting, AIAA
Paper 2010-108, Jan. 2010.
[19] Priebe, S., and Pino Martn, M., Low-Frequency Unsteadiness in
Shock Wave/Turbulent Boundary Layer Interaction, Journal of Fluid
Mechanics, Vol. 699, No. 1, May 2012, pp. 149.
doi:10.1017/jfm.2011.560
[20] Grilli, M., Schmid, P. J., Hickel, S., and Adams, N. A., Analysis
of Unsteady Behavior in Shockwave Turbulent Boundary Layer
Interaction, Journal of Fluid Mechanics, Vol. 700, June 2012, pp. 1628.
doi:10.1017/jfm.2012.37
[21] Selig, M. S., and Smits, A. J., Effect of Periodic Blowing on Attached
and Separated Supersonic Turbulent Boundary Layers, AIAA Journal,
Vol. 29, No. 10, Oct. 1991, pp. 16511658.
doi:10.2514/3.10787
[22] Dolling, D. S., Unsteady Phenomena in Shock Wave/Boundary Layer
Interaction, AGARD TR R-792, May 1993, pp. 4-14-46.
[23] Kalra, C., Shneider, M. N., and Miles, R., Numerical Study of
Boundary Layer Separation Control using Magnetogasdynamic Plasma
Actuators, Physics of Fluids, Vol. 21, No. 10, Oct. 2009, p. 106101.
doi:10.1063/1.3233658
[24] Im, S., Do, H., and Cappelli, M. A., Dielectric Barrier Discharge
Control of a Turbulent Boundary Layer in a Supersonic Flow, Applied
Physics Letters, Vol. 97, No. 4, July 2010, p. 041503.
doi:10.1063/1.3473820
[25] Webb, N., Clifford, C., Porter, A., and Samimy, M., Control of
Oblique Shock Wave-Boundary Layer Interactions Using Plasma
Actuators, 6th AIAA Flow Control Conference, AIAA Paper 2012-
2810, June 2012.
[26] Gaitonde, D., and Visbal, M. R., High-Order Schemes for Navier
Stokes Equations: Algorithm and Implementation into FDL3DI, U.S.
Air Force Research Lab. TR 1998-3060, Wright-Patterson AFB, Ohio,
Aug. 1998.
[27] Visbal, M. R., Computational Study of Vortex Breakdown on a
Pitching Delta Wing, AIAA 24th Fluid Dynamics Conference, AIAA
Paper 1993-2974, June 1993.
[28] Gordnier, R. E., and Visbal, M. R., Numerical Simulation of Delta-
Wing Roll, Aerospace Science and Technology, Vol. 2, No. 6, 1998,
pp. 347357.
doi:10.1016/S1270-9638(99)80023-6
[29] Visbal, M. R., Gaitonde, D. V., and Gogineni, S., Direct Numerical
Simulation of a Forced Transitional Plane Wall Jet, 29th AIAA Fluid
Dynamics Conference, AIAA Paper 1998-2643, 1998.
[30] Rizzetta, D. P., Visbal, M. R., and Blaisdell, G. A., Application of a
High-Order Compact Difference Scheme to Large-Eddy and Direct-
Numerical Simulation, 30th AIAA Fluid Dynamics Conference, AIAA
Paper 1999-3714, JuneJuly 1999.
[31] Morgan, P. E., Rizzetta, D. P., and Visbal, M. R., High-Order
Numerical Simulation of Turbulent Flow over a Wall-Mounted Hump,
AIAA Journal, Vol. 44, No. 2, 2006, pp. 239251.
doi:10.2514/1.13597
[32] Garmann, D. J., and Visbal, M. R., Numerical Investigation of
Transitional Flow Over a Rapidly Pitching Plate,Physics of Fluids,
Vol. 23, No. 9, 2010, p. 094106.
doi:10.1063/1.3626407
[33] White, M. D., and Visbal, M. R., High Fidelity Analysis of Aero-
Optical Interaction with Compressible Boundary Layers, 41st
Plasmadynamics and Lasers Conference, AIAA Paper 2010-4496,
JuneJuly 2010.
 1804 BISEK, RIZZETTA, AND POGGIE
[34] Anderson, D., Tannehill, J., and Pletcher, R., Computational Fluid
Mechanics and Heat Transfer, McGrawHill, New York, 1984.
[35] Mullenix, N. J., Gaitonde, D. V., and Visbal, M. R., A Plasma-Actuator-
Based Method to Generate a Supersonic Turbulent Boundary
Layer Inflow Condition for Numerical Simulation, 20th AIAA
Computational Fluid Dynamics Conference, AIAA Paper 2011-3556,
June 2011.
[36] Raizer, Y. P., Gas Discharge Physics, SpringerVerlag, Berlin, 1991.
[37] Macheret, S., Physics of Magnetically Accelerated Nonequilibrium
Surface Discharges in High-Speed Flow, 44th AIAA Aerospace
Sciences Meeting and Exhibit, AIAA Paper 2006-1005, Jan. 2006.
[38] MacCormack, R. W., Flow Calculations with Strong Magnetic Fields,
42nd AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper
2003-3623, Jan. 2003.
[39] Atkinson, M. D., Poggie, J., and Camberos, J. A., Control of Separated
Flow in a Reflected Shock Interaction Using a Magnetically-
Accelerated Surface Discharge, Physics of Fluids, 2012 (accepted for
publication).
doi:10.1063/1.4772197
[40] Atkinson, M. D., Hypersonic Flow and Flight Control Using Plasma
Actuators, Ph.D. Thesis, Dept. of Aerospace Engineering, Univ. of
Dayton, Dayton, OH, 2012.
[41] Beam, R., and Warming, R., An Implicit Factored Scheme for the
Downloaded by NAGOYA UNIVERSITY on July 2, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.J052248
Compressible NavierStokes Equations, AIAA Journal, Vol. 16, No. 4,
1978, pp. 393402.
doi:10.2514/3.60901
[42] Gordnier, R. E., and Visbal, M. R., Numerical Simulation of Delta-
Wing Roll, 31st Aerospace Sciences Meeting and Exhibit, AIAA Paper
1993-0544, Jan. 1993.
[43] Jameson, A., Schmidt, W., and Turkel, E., Numerical Solutions of the
Euler Equations by Finite Volume Methods Using RungeKutta Time
Stepping Schemes, AIAA 14th Fluid and Plasma Dynamics
Conference, AIAA Paper 1981-1259, June 1981.
[44] Pulliam, T. H., and Chaussee, D. S., A Diagonal Form of an Implicit
Approximate-Factorization Algorithm, Journal of Computational
Physics, Vol. 39, No. 2, 1981, pp. 347363.
doi:10.1016/0021-9991(81)90156-X
[45] Lele, S., Compact Finite Difference Schemes with Spectral-like
Resolution, Journal of Computational Physics, Vol. 103, No. 1, 1992,
pp. 1642.
doi:10.1016/0021-9991(92)90324-R
[46] Visbal, M. R., and Gaitonde, D. V., High-Order-Accurate Methods for
Complex Unsteady Subsonic Flows, AIAA Journal, Vol. 37, No. 10,
1999, pp. 12311239.
doi:10.2514/2.591
[47] Gaitonde, D., Shang, J. S., and Young, J. L., Practical Aspects of High-
Order Accurate Finite-Volume Schemes for Electromagnetics, 35th
AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper 1997-
0363, Jan. 1997.
[48] Visbal, M. R., and Rizzetta, D. P., Large-Eddy Simulation on
Curvilinear Grids Using Compact Differencing and Filtering Schemes,
Journal of Fluids Engineering, Vol. 124, No. 4, 2002, pp. 836847.
doi:10.1115/1.1517564
[49] Roe, P. L., Approximate Riemann Solvers, Parameter Vectors, and
Difference Schemes, Journal of Computational Physics, Vol. 43, No. 2,
1981, pp. 357372.
doi:10.1016/0021-9991(81)90128-5
[50] van Albada, G. D., van Leer, B., and Roberts, W. W. Jr., A Comparative
Study of Computational Methods in Cosmic Gas Dynamics,
Astronomy and Astrophysics, Vol. 108, No. 1, April 1982, pp. 7684.
[51] Swanson, R. C., and Turkel, E., On Central-Difference and Upwind
Schemes, Journal of Computational Physics, Vol. 101, No. 2, 1992,
pp. 292306.
doi:10.1016/0021-9991(92)90007-L
[52] Visbal, M. R., Morgan, P. E., and Rizzetta, D. P., An Implicit LES
Approach Based on High-Order Compact Differencing and Filtering
Schemes, 16th AIAA Computational Fluid Dynamics Conference,
AIAA Paper 2003-4098, June 2003.
[53] Fureby, C., and Grinstein, F. F., Monotonically Integrated Large Eddy
Simulation, AIAA Journal, Vol. 37, No. 5, 1999, pp. 544556.
doi:10.2514/2.772
[54] Stoltz, S., and Adams, N. A., An Approximate Deconvolution
Procedure for Large-Eddy Simulation, Physics of Fluids, Vol. 11,
No. 7, July 1999, pp. 16991701.
doi:10.1063/1.869867
[55] Mathew, J., Lechner, R., Foysi, H., Sesterhenn, J., and Friedrich, R.,
An Explicit Filtering Method for Large Eddy Simulation of
Compressible Flows, Physics of Fluids, Vol. 15, No. 8, Aug. 2003,
pp. 22792289.
doi:10.1063/1.1586271
[56] Rai, M. M., Gatski, T. B., and Erlebacher, G., Direct Simulation of
Spatially Evolving Compressible Turbulent Boundary Layers, 33rd
Aerospace Sciences Meeting and Exhibit, AIAA Paper 1995-0583,
Jan. 1995.
[57] Rizzetta, D. P., and Visbal, M. R., Large-Eddy Simulation of
Supersonic Boundary-Layer Flow by a High-Order Method,
International Journal of Computational Fluid Dynamics, Vol. 18,
No. 1, 2004, pp. 1527.
doi:10.1080/10618560310001614926
[58] Pirozzoli, S., Grasso, F., and Gatski, T. B., Direct Numerical
Simulation and Analysis of a Spatially Evolving Supersonic Turbulent
Boundary Layer at M  2.25, Physics of Fluids, Vol. 16, No. 3,
March 2004, pp. 530545.
doi:10.1063/1.1637604
[59] Pirozzoli, S., and Grasso, F., Direct Numerical Simulation of
Impinging Shock Wave/Turbulent Boundary Layer Interaction at
M  2.25, Physics of Fluids, Vol. 18, No. 6, 2006, p. 065113.
doi:10.1063/1.2216989
[60] White, F. M., Viscous Fluid Flow, 3rd ed., McGrawHill, New York,
2006.
[61] Fernholz, H. H., and Finley, P. J., A Critical Compilation of
Compressible Turbulent Boundary Layer Data, AGARD Rept. 223,
1977.
[62] Bisek, N. J., Rizzetta, D. P., and Poggie, J., Exploration of Plasma
Control for Supersonic Turbulent Flow over a Compression Ramp,
42nd AIAA Fluid Dynamics Conference and Exhibit, AIAA Paper
2012-2700, June 2012.
[63] Ghosal, S., Mathematical and Physical Constraints on Large-Eddy
Simulation of Turbulence, AIAA Journal, Vol. 37, No. 4, 1999,
pp. 425433.
doi:10.2514/2.752
[64] Rizzetta, D. P., and Visbal, M. R., Numerical Investigation of Plasma-
Based Control for Low-Reynolds-Number Airfoil Flows, AIAA
Journal, Vol. 49, No. 2, Feb. 2011, pp. 411425.
doi:10.2514/1.J050755
[65] van Driest, E. R., On the Turbulent Flow Near a Wall, Journal of the
Aeronautical Sciences, Vol. 23, 1956, pp. 10071011.
[66] Elena, M., and LaCharme, J. P., Experimental Study of a Supersonic
Turbulent Boundary Layer Using Laser Doppler Anemometer,
Journal de Mcanique Thorique et Applique, Vol. 7, 1988,
pp. 175190.
[67] Karlson, R. I., and Johansson, T. G., Experimental Study of a
Supersonic Turbulent Boundary Layer Using a Laser Doppler
Anemometer, Journal of Theoretical and Applied Mechanics, Vol. 7,
No. 2, 1988, pp. 175190.
[68] Smits, A. J., and Dussauge, J.-P., Turbulent Shear Layers in Supersonic
Flow, 2nd ed., Springer, New York, 2006.
[69] Bendat, J. S., and Piersol, A. G., Random Data, 2nd ed., Wiley, New
York, 1986.
[70] Jeong, J., and Hussain, F., On the Identification of a Vortex, Journal of
Fluid Mechanics, Vol. 285, Feb. 1995, pp. 6994.
doi:10.1017/S0022112095000462
[71] Zheltovodov, A. A., and Schlein, E., The Peculiarities of Turbulent
Separation Development in Disturbed Boundary Layers, Modeling in
Mechanics, Vol. 2, No. 19, 1998, pp. 5358 (in Russian).
[72] Zheltovodov, A. A., Some Advances in Research of Shock Wave
Turbulent Boundary Layer Interactions, 44th AIAA Aerospace Sciences
Meeting and Exhibit, AIAA Paper 2006-496, Jan. 2006.
[73] Zaidi, S. H., Smith, T., Macheret, S., and Miles, R. B., Snowplow
Surface Discharge in Magnetic Field for High Speed Boundary Layer
Control, 44th AIAA Aerospace Sciences Meeting and Exhibit, AIAA
Paper 2006-1006, Jan. 2006.
P. Tucker
Associate Editor