International Journal of Heat and Fluid Flow 41 (2013) 8089

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier.com/locate/ijhff

Experimental and computational study of the ow induced by a plasma

actuator
I. Maden a,b,c,, R. Maduta a,b, J. Kriegseis a,b, S. Jakirlic a,b,c,1, C. Schwarz a,b, S. Grundmann a,b,c, C. Tropea a,b,c
a

Department of Mechanical Engineering, Technische Universitt Darmstadt, Germany

Institute of Fluid Mechanics and Aerodynamics (SLA), Petersenstrasse 17, D-64287 Darmstadt, Germany
c
Center of Smart Interfaces (CSI), Petersenstrasse 17, D-64287 Darmstadt, Germany
b

a r t i c l e

i n f o

Article history:
Available online 29 March 2013
Keywords:
Plasma actuator
Wall jet
PIV-based volume-force model
Second-moment closure RANS model
Three-dimensional diffuser

a b s t r a c t
A complementary experimental and computational study of the ow eld evoked by a plasma actuator
mounted on a at plate was in focus of the present work. The main objective of the experimental investigation was the determination of the vector force imparted by the plasma actuator to the uid ow. The
force distribution was presently extracted from the NavierStokes equations directly by feeding them
with the velocity eld measured by a PIV technique. Assuming a steady-in-mean, two-dimensional ow
with zero-pressure gradient, the imbalance between the convective term and the momentum equations
right-hand-side terms reveals the desired resulting force. This force-distribution database was used afterwards as the source term in the momentum equation. Furthermore, an empirical model formulation for
the volume-force determination parameterized by the underlying PIV-based model is derived. The model
is tested within the RANS framework in order to predict a wall jet-like ow induced by a plasma actuator.
The Reynolds equations are closed by a near-wall second-moment closure model based on the homogeneous dissipation rate of the kinetic energy of turbulence. The computationally obtained velocity eld is
analysed along with the experimental data focussing on the wall jet ow region in proximity of the
plasma actuator. For comparison purposes, different existing phenomenological models were applied
to evaluate the new models accuracy. The comparative analysis of all applied models demonstrates
the strength of the new empirical model, particularly within the plasma domain. In addition, the presently formulated empirical model was applied to the ow in a three-dimensional diffuser whose inow
was modulated by a pair of streamwise vortices generated by the present plasma actuator. The direct
comparison with existing experimental data of Grundmann et al. (2011) demonstrated that the specic
decrease of the diffuser pressure corresponding to the continuous forcing was predicted correctly.
2013 Elsevier Inc. All rights reserved.

1. Introduction
The term plasma actuator denotes a ow-control device based
on dielectric barrier discharges (DBDs). A radio-frequency high
voltage applied between a surface-ush mounted electrode and a
buried grounded electrode generates an unsteady electric eld
strong enough to periodically create weakly ionized plasma. The
charges moved during generation and quenching of the plasma
collide with the neutral surrounding gas molecules and transfer
their momentum. By this mechanism a periodic body force is created above the electrodes of the actuator that can be used for
numerous ow control applications. If operated in quiescent air
Corresponding author at. Institute of Fluid Mechanics and Aerodynamics (SLA),
Petersenstrasse 17, D-64287 Darmstadt, Germany.
E-mail addresses: maden@csi.tu-darmstadt.de (I. Maden), s.jakirlic@sla.
tu-darmstadt.de (S. Jakirlic).
1
Principal corresponding author
0142-727X/$ - see front matter 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatuidow.2013.02.013

this body force creates a wall jet (Fig. 1 left). If applied in an existing ow, such as a laminar boundary layer, the body force can be
applied for a benecial modication of the mean ow (Fig. 1 right)
as well as for a modication or creation of velocity uctuations,
see e.g., Grundmann and Tropea (2009) and Benard et al. (2011).
Advantages and disadvantages of discharge based devices are
discussed alongside other actuators for (active) ow control by
Cattafesta and Sheplak (2011). An excellent review of the
actuators working principle is provided by Moreau (2007).
The determination of the magnitude and distribution of the
force imparted from the plasma actuator to the external ow is
of crucial importance for any advanced prediction of dischargebased ow-control scenarios by means of numerical simulations.
The typical spatial and temporal scales of gas-discharge processes
are four to eight orders of magnitude smaller then those of the
resulting ow-control applications. To resolve this discrepancy
for CFD (Computational Fluid Dynamics), several so-called phenomenological plasma actuator models arose in recent years (see

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

81

HV

HV

Fig. 1. Sketch of the airow behavior generated by plasma actuator operation; (left) wall jet formation under quiescent air conditions, and (right) manipulation of an existing
boundary layer.

Fig. 2. Experimental setup for the PIV measurements according to Kriegseis (2011).

e.g. Jayaraman and Shyy (2008)), each of which providing a temporally constant volume-force distribution f(x, y). Based on the necessarily strong simplication, the spatial distribution of these models
typically result in rather articial and non-physical shapes of the
momentum-transfer domain. Therefore, the computational studies
performed in the past resulted in a moderate success.
A promising alternative to provide an appropriate source term
in the momentum equation within a CFD solution procedure,
therefore, is the retroactive estimation of the volume force from
experimental results. In the present work the PIV measurements
of Kriegseis et al. (2013) in close proximity to dielectric-barrier discharge plasma actuators are utilized to provide such a velocity data
base. This velocity information is then used for the force-estimation purpose and the development of an empirical model for the
prediction of DBD plasma actuator momentum transfer in numerical simulations. The major advantage of the proposed approach is
the successful combination of the accurate spatial resolution of
velocity-information based force-distribution determination with
an efcient straight forward implementation of simple functions
as used for phenomenological models.
2. Experimental setup
To investigate the ow behavior and especially the momentum
transfer to the ow, PIV measurements have been conducted in
close proximity to the actuators discharge region (see Fig. 2 left).
For orientation purposes the wall jet direction as well as the
x-coordinate origin are included in the sketch of the setup (see
Fig. 2right).
A commercial high-speed PIV system comprising a Litron
Nd:YLF (k = 527 nm) dual-cavity laser and two Phantom V12
high-resolution cameras (12 bit, maximum resolution 1280  800
pixels) were utilized, which was operated in single-frame mode
at a repetition rate of 10 k frames per second (fps) and a pulse
duration of 150 ns. This high repetition rate with Dt = 100 ls required the reduction of the spatial resolution down to 800  600
pixels. A maximum number of N = 10 k images per run per camera
was recorded, exploiting the available buffer capacity of 8 GB. The
cameras were mounted facing one another perpendicular to the laser-light sheet as shown in Fig. 2. This arrangement allowed the
simultaneous observation of two different elds of view (FOV).

This choice provided the highest possible spatial resolution

(81.3 pix/mm) in the plasmas immediate vicinity, suitable for
force calculations (FOV #1). In addition, the spatial distribution
of the resulting wall jet downstream of the discharge domain can
be characterized for identical experimental realizations with FOV
#2, which has approximately half the spatial resolution (42.2 pix/
mm) but spans twice the physical domain compared to FOV #1.
A reversely mounted2 120 mm SKR SYMMAR lens and a 105 mm Nikon Nikkor lens were tted to span the observation dimensions
10  7 mm2 (FOV #1) and 19  14 mm2 (FOV #2) respectively.
The test section was enclosed in a closed plexiglass containment (L  W  H = 450  325  345 mm3) with quartz-glass windows to assure best possible quality of optical accesses for laserlight sheet and cameras. Di-Ethyl-Hexyl-Sebacat (DEHS) aerosol
(mean diameter 0.9 lm) was used to seed the containment. The
velocity distribution was calculated from the raw data using commercial software (Dynamic Studio) (see Table 1). Rectangular
interrogation areas (IAs) of 64  16 pixels and 32  16 pixels were
chosen for FOV #1 and FOV #2, respectively. This particular setting
was found to be favorable to calculate a wall jet, i.e. high wall-normal velocity gradients and wall-parallel velocities.
Although the major objective of the PIV investigations was a
force (distribution) quantication and the spatial resolution of
FOV/lens #1 fullled the corresponding requirements, the availability of a second camera (FOV/lens #2) was found to be extremely benecial for other purposes. Based on the calculated DBD
source terms from FOV/lens #1 (see Section 3), these additional
velocity data from FOV/lens #2, comprising a fully developed wall
jet, are available for further validation of the conducted numerical
simulations as shown in the present work. For more details concerning the experimental procedure and the subsequent velocityinformation based force-estimation approaches the reader is referred to Kriegseis (2011).
3. Volume-force estimation approaches
The major objective of the present study is a comparative analysis of advantages and drawbacks of different force estimation ap2
This particular setting was chosen to reduce the observed domain beyond the
lens lower magnication limit of 1:1.

82

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

Table 1
Components of the implemented PIV setup (see Kriegseis et al., 2013): chosen product, properties and corresponding settings.
Component

Description

Settings/properties

Test section
Laser

Perspex containment
Litron Lasers
Model LDY303-PIV
2 Phantom V12
Schneider-Kreuznach SKR
SYMMAR 120/5.6-0.33X (reverse mode!)
Nikon Nikkor 105 mm
f/2.8 AF-Micro
DEHS
Dynamic Studio (Dantec Dynamics)

Dimensions (length/width/height): l1 = 450 mm/w1 = 325 mm/h1 = 345 mm

Laser medium: Nd:YLF 527 nm, 70 W
Pulse duration: 150 ns
800  600 pixels, 10,000 fps
Focal length: f = 120 mm; FOV: 10  7 mm2
Resolution: 81.3 pix/mm
Focal length: f = 105 mm; FOV: 19  14 mm2
Resolution: 42.2 pix/mm
Mean diameter 0.9 lm
Versions 2.1 and 2.3

Cameras
Lens #1
(FOV #1)
Lens #2
(FOV #2)
Seeding
Software

proaches to model a quasi-steady force distribution. On the one

hand any numerical simulation of discharge-based ow control
necessarily involves strong simplication and partly nonphysical
assumptions to achieve phenomenological discharge-force models.
On the other hand limitations of available measurement techniques require similar assumptions for experiment-based determination of the volume-force characteristics. Note that rst efforts to
model the spatio-temporal distribution of the actuator force have
been done by Orlov (2006) on a semi-empirical basis. This
lumped-element based model has been successfully applied e.g.
by Mertz and Corke (2011).
Both, phenomenological and experiment-based approaches as
presented in this study are rst briey recapped in this section.
Subsequently, the new empirical model is introduced. The processing owchart of the present work is summarized in Fig. 3. All approaches considered presume the volume force to act quasisteady, which is an appropriate approximation at least at aerodynamic scales. To assure best possible comparability all model calibrations were conduced such that the integral force magnitude

Z Z

f x; y dA

was identical for all approaches and furthermore veried with

force-balance data of Kriegseis et al. (2011).

3.1. Phenomenological models

Numerical simulations of plasma actuator based ow-control
commonly apply phenomenological models based on Shyy et al.
(2002) or Suzen et al. (2007). The basis of phenomenological models is the Coulomb-force equation

!
~
f E qc

where charged particles of charge density qc are accelerated in an

external electric eld E.
For a given actuator geometry and operating voltage Shyy et al.
(2002) simplify the external electric eld E to a triangle. The
remaining unknown qc is calculated using the rst Maxwells
equation

r E

qc

0

such that

!
!
~
f 0 E r  E

for the calculated distribution of E (cp. Roth et al., 2000; Shyy et al.,
2002). However, this approach is a strong oversimplication, since

Fig. 3. Flowchart of the force determination procedure; PIV-based velocity eld U(x, y) in proximity of the plasma actuator (top-left); resulting volume-force f(x, y) (middleright) and corresponding simulated velocity U(x, y) (middle-left); empirical force model f(x, y) (bottom-right) and corresponding velocity distribution U(x, y) (bottom-left).

83

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

Eq. (4) does not distinguish between the external eld for Eq. (2)
and the local eld created by presence of charges in Eq. (3).
Suzen et al. (2007) overcome the problem by dividing the electric potential U into an external and local component U = / + u.
The external eld E = r/ is determined by a Laplace equation for
the external potential

r  r r/ 0

The local eld ru and the presence of charges are balanced by Eq.
(3), which now includes the two unknowns u and qc. The assumption of equilibrium plasma allows the Boltzmanns relation to be
used, which in combination with the Debye shielding length kd
gives rise to

u

qc k2

0

Replacement of the local potential u as dened by Eq. (6) leads to a

Helmholtz equation for the charge density

r  r rqc

qc

k2d

In the present work, Eqs. (5) and (7) are solved using COMSOL Multiphysics Partial Differential Equation (PDE) solvers. Thus, the required eld strength r/ and charge density qc were available for
solving Eq. (2) for the volume force distribution f(x, y).
3.2. Velocity-information based model
The velocity-information based force-determination approach
relies on the 2-dimensional, steady NavierStokes equation (NSE)
or vorticity equation, in accordance with the proposals of Wilke
(2009) or Albrecht et al. (2011), respectively. Wilke (2009) assumed that the inuence of the unknown pressure gradient was
assumed to be negligible such that the NSE reduces to

!

@u
@u
@2u @2u
l
f x; y q u

v
@x
@y
@x2 @y2

The force distribution according to Wilkes method is shown in

Fig. 3 (middle-right).
Utilization of the vorticity equation eliminates the pressure gradient, but leaves the curl of the volume force, i.e. one equation with
two unknowns. Albrecht et al. (2011) assume the wall normal component of the force to be at least one order of magnitude smaller
than the streamwise one. Subsequent reintegration of the force
gradient leads to

"
u

!#
@x
@x g @2x @2x
dy

v

@x
@y q @x2
@y2

Xx a1 x a2 x2  expa0 xa and
2

Both experiment-based approaches are extensively described by

Kriegseis (2011).
3.3. Present model formulation
The motivation for the development of an empirical model is
twofold. On the one hand, one wants to overcome the enormous efforts of conducting whole-eld-technique measurements and to
facilitate the consequent employment of the large database for
CFD. In other words, one would like to replace it by a straightforward use of applicable equations. On the other hand, in contrast
to the oversimplifying phenomenological models, the developed
equations must represent the spatially accurate volume-force distribution of the data achieved by the experiment-based approaches
(Albrecht et al., 2011; Kriegseis et al., 2013; Wilke, 2009).

Yy b1 y b2 y  expb0 y
a1;2 ; b1;2 2 R;

f x; y q

In the present work the spatial force distribution f(x, y) as determined by the velocity-information based approaches (see Section 3.2) is used as underlying data base to develop the desired
empirical model (see Fig. 3, middle-right). The corresponding integral force value for the electrical operating conditions is F = 25 mN/
m (cp. Kriegseis et al., 2011, 2013). Particularly, the force distribution data base as calculated by means of Eq. (8) was chosen for the
derivation of the new model, so as to assure best possible spatial
accuracy of the underlying data base. Kriegseis et al. (2013)
showed that despite different assumptions and simplications
both experiment-based methods, Eqs. (8) and (9), lead to a similar
spatial distribution of the actuator force. Further to the validity of
either approach, this direct comparison demonstrated that the
respective neglected contributions, i.e. pressure gradients rp and
wall-normal forces fy, are indeed of minor importance to the overall volume-force distribution. However, a force-term estimation
based on the vorticity Eq. (9) leads to up to third derivatives of
experimental data, whereas the NSE-based approach only handles
up to second derivatives. In consequence, the uncertainty of the
resulting force distribution based on Eq. (9) is approx. 50% higher.
Therefore, the force distributions of either approaches were directly implemented to the ow solver, but only the NSE-based results served as an underlying data base for the new model to avoid
any unnecessary accuracy drawbacks (Kriegseis et al., 2013).
As displayed in Fig. 3 (middle-right), the force distribution rst
increases steeply with increasing x values (y respectively) and decreases consequently after reaching a local maximum. This behavior can be mathematically expressed by the product of a
polynomial rise and an exponential decay in x (y respectively).
The governing equations describing the magnitude and distribution of the force components in x-direction and y-direction are as
follows:

10
11

a0 ; b0 > 0;

a; b > 0; x; y P 0
Note that Eqs. (10) and (11) are solely functions of the x- and
y-coordinates, respectively. Therefore, a scalar multiplication of
these independent one-dimensional functions leads to the desired
functional description of the force distribution

f x; y c Xx Yy;

c2R

12

Model Eq. (12) sets an empirical relationship between the force distribution and its spatial domain by determining dependent variables a0,1,2, b0,1,2, c, a and b for Eqs. (10)(12). Based on a leastsquares t (see e.g. Press et al. (1986)) the dependent variables
has been adjusted in order to minimize the difference between
model Eq. (12) and the calculated force distribution based on Eq.
(8), which automatically matches the magnitude of the resulting
force. The force eld obtained by the present model proposal is displayed in Fig. 3 (bottom-right). Some differences between the model force distribution and that obtained by using Eq. (8), Fig. 3
(middle-right), concerning primarily the force intensity, are observed in the immediate vicinity of the upper electrodes trailing
edge (x = 0). Nevertheless, the region inuenced by the plasma actuator determined experimentally and computationally spreads almost identically. Accordingly, the afore-mentioned differences do
not inuence the consequent wall jet evolution. The outcome of
the computational investigation compared directly to the experimentally obtained velocity proles shown in Fig. 7 conrms qualitative similarity.
As it can be seen the model constants in Eqs. (10)(12) are
dependent on the operating voltage of the plasma actuator implying different constant values for different voltage levels. But, the

84

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

determination of the model constants pertinent to the working

conditions of the plasma actuator employed represents integral
part of the entire computational procedure. This could be simply
regarded as a kind of the data input into the computational domain. Accordingly, the presently proposed model formulation in
conjunction with the procedure for determination of the model
constants relevant to a given plasma actuator, as to the one employed presently, can be universally applied to all types of discharge-based momentum-transfer estimations. Nevertheless, the
upcoming work foresees application of the new model to PIV database of varying actuator-force intensities to derive appropriate
equations of the model-coefcients as function of the underlying
actuator force, i.e. a0,1,2, b0,1,2, c, a and b = a0,1,2(F), b0,1,2(F),
c(F), a(F) and b(F) resulting nally in a non-dimensional force eld
distribution. Based on this upgrade a universal application of the
present empirical modeling approach for numerical simulations
of discharge-based ow-control applications will be targeted in
some future works.

The Reynolds number based on y = y1/2 (U = 1/2 Umax) is about four

times lower Rey1=2 U max y1=2 =m  370.
Such a low Reynolds number would possibly point out to a laminar ow conguration. According to Jukes et al. (2006), who comparatively analyzed the wall jet velocity proles originating from a
plasma actuator and their theoretical counterparts discussed in
Glauert (1956), the ow regime is to be regarded as laminar if
y/y1/2(Umax)  0.55 and turbulent if y/y1/2(Umax)  0.25. According
to such a criterion, the present conguration is closer to the turbulent case, see Fig. 5-upper, indicating y/y1/2(Umax)  0.30  0.40.
Furthermore, Chun and Schwarz (1967) showed that the critical
Reynolds number of a laminar wall jet is Rey1=2crit 57. In order to
increase the transparency of such an analysis the analytical velocity proles for both laminar and turbulent regimes proposed by
Glauert are introduced in Fig. 5 (upper diagram). The transitional
nature of the wall jet is indicated by the arrow, i.e. the velocitypeak location y/y1/2 moves to the wall as the wall jet develops. In

4. Computational method
Complementary to the experimental work the ow eld induced by the present plasma actuator was computationally investigated by the Steady RANS (Reynolds-Averaged NavierStokes)
method. The unknown Reynolds stress components are computed
by solving their model differential equations in line with the
homogeneous-dissipation-based (h =   0.5m@ 2k/(@xj@xi)), nearwall second-moment closure due Jakirlic and Hanjalic (2002). Presently the model equation for the Reynolds stress tensor is solved in
conjunction with the equation governing the homogeneous part of
the inverse turbulent time scale xh(=h/k). The latter equation is
derived directly from the h-equation according to the procedure
given in Maduta and Jakirlic (2012) and is completely equivalent
to it. By doing so a number of important advantages pertinent to
the homogeneous-dissipation concept could be retained: proper
near-wall shape of the dissipation rate prole was obtained without introducing any additional term and the correct asymptotic
behavior of the stress dissipation components by approaching
the solid wall is fullled automatically without necessity for any
wall geometry-related parameter. In the case of a standard wall
jet issuing from a slot the Reynolds number denition is based on
the slot height and a relevant bulk velocity. Such a denition is
presently not possible. A relevant Reynolds number based on the
maximum velocity Umax and the thickness yUmin of the wall jet inside the plasma domain is Re U max yUmin =m  1500 (cp. Fig. 4).

Fig. 4. Velocity prole typical of a wall jet conguration and corresponding

nomenclature.

Fig. 5. Computationally, applying present Reynolds stress model, obtained nondimensional velocity proles, U/Umax = f(y/y1/2) (upper) and U+ = y+ (middle), and
streamwise Reynolds stress intensity component (lower) in the developing region,
x = 5, 10, 20, 50, 100, 150 mm of the plasma-induced wall jet.

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

this context it is worth noting that an attempt to compute the present ow without introducing turbulence model resulted in a nonconverged solution. Accordingly, a transitional regime would be a
more accurate description. Indeed, the computational results displayed in Fig. 5, following closely the ndings of Eriksson et al.
(1998), conrm it.
In general, a wall jet conguration is a combination of a wall
boundary layer and a free jet. At an appropriately high Reynolds
number the velocity prole (relevant to the wall boundary layer region) obeys the logarithmic law, see e.g. Eriksson et al. (1998). In
this transitional ow the region of initial development of the velocity prole resembles that of a laminar-like ow developing consequently towards a prole corresponding to a fully-turbulent ow.
The turbulence evolution exhibiting initially an appropriately
low intensity is consistent with such a mean ow development.
Indeed the streamwise Reynolds stress intensity (Fig. 5-lower)
exhibits a very low level at the very beginning of the wall jet development (x = 5  10 mm). The location x = 5 mm relates to the area
of the plasma domain itself. Here, the viscous sublayer (the thickness of which corresponds to y+  2) starts its development. The
turbulence intensity peak at x = 20 mm coincides with the high
negative velocity gradient in the outer part of the wall jet at
y+  35. The presently obtained maximum value u+  3.5 is completely in accordance with the measurements of Eriksson et al.
(1998). Furthermore, the relevant turbulent kinetic energy eld
and its production rate evaluated computationally (not shown
here) agrees well with the ndings of Dejoan and Leschziner
(2006), whose LES study resulted in its maximum in the shear layer
corresponding to  8%. According to Eriksson et al. (1998) the
transition onset of the initially laminar wall jet takes place exactly
in its outer part. Afterwards (x P 50 mm) the prole shape exhibits
two peaks, one near-wall peak originating from the appropriately
strengthened positive velocity gradient in the buffer layer
(y+  15) and the afore-mentioned outer-part peak. In between
the velocity proles reach their maximum corresponding approximately to the logarithmic region, Fig. 5-middle. Due to a somewhat
lower Reynolds number of the presently investigated wall jet conguration the peak of the velocity proles does not reach completely the log-law. Nevertheless, the streamwise Reynolds stress
component evolution indicates clearly a developed turbulent ow.
Keeping in mind a transitional and non-equilibrium character of
the ow, necessity to employ a near-wall turbulence model becomes obvious.
All computations were performed using the code OpenFOAM,
an open source Computational Fluid Dynamics toolbox, utilizing
a cell-center-based nite volume method on an unstructured
numerical grid and employing the solution procedure based on
the SIMPLE procedure for coupling the pressure and velocity elds.
The convective transport in momentum equations was discretized
by the 2nd order central differencing scheme implemented in the
deferred-correction manner.
The dimensions of the ow area (measurement window)
accommodating the experimentally determined velocity eld, i.e.
the resulting volume force was Lx  Ly = 8.87  4.38 mm2, Fig. 6.
This area was experimentally resolved by a grid comprising Nx  Ny = 90  46 cells; the velocity components values have been
provided at each experimental node. The corresponding numerical resolution is Nx  Ny = 28  36. The experimentally obtained
force eld database was linearly interpolated to the numerical grid.
The dimensions of the experimentally considered box-like ow domain, conned by the solid walls, in the vertical x-y plane are
450  345 mm2 (spanwise dimension is 325 mm).
The 2D ow domain, equivalent to the 3D domain, was adopted
numerically, with the boundaries of the vertical x-y plane being set
appropriately to avoid effects of the secondary ow currents (in the
spanwise direction) to the volume force, as it was observed in the

85

Fig. 6. Sketch of the domains for PIV experiment (subscript 1, cp. Fig. 2), 2D
simulation (subscript 2, ) and 3D simulation (subscript 3, - - - - -, w1 = w3).

experimental work. The consequent 3D computations accounting

for the experimentally investigated ow domain in its entirety, justied the present 2D discretization (see. Fig. 6).
The size of the 3D domain is 450  170  325 mm3 in the
streamwise, wall normal and spanwise direction, respectively.
The computational grid contains 293  123  121 (4.36 million)
nodes. The grid was squeezed towards the solid walls to provide
the dimensionless height of the wall-next computational cells
being Dy+ < 1. The cavities presented in the experimental setup
(on the left and right sides of the stand, Fig. 6) were treated numerically by applying the symmetry-plane boundary condition.
5. Results and discussion
The present wall jet ow represents a transitional, non-equilibrium ow affected strongly by a near-wall Reynolds stress anisotropy. This is also the main reason for employing the Reynolds
stress model, and not an anisotropy-blind k- model; some preliminary computations using the latter model have also been performed. Therefore, the following comparative discussion about
the varying actuator-model approaches considers only the Reynolds stress model computations. The wall jet-like mean velocity
proles obtained by employing different actuator models in conjunction with a near-wall Reynolds stress model are depicted at
several downstream locations of the discharge domain in Fig. 7.
The direct comparison of the two experiment-based force estimation approaches conrms the validity of the respective independent assumptions. Eq. (8) neglects the pressure gradient, whereas
Eq. (9) implies no simplications concerning the pressure term.
Both proles agree well with one another for all streamwise
locations.
The insufcient spatial accuracy of the phenomenological models inside the plasma region (0 6 x 6 6 mm) is obvious. Downstream of the discharge domain (x P 6 mm) the fully developed
wall jet can be predicted by Shyys model, since the correct overall
momentum is imparted to the ow. Although Suzens model results in the shape-accurate wall jet proles inside the discharge
domain, the velocity magnitude in the fully-developed wall jet region is underpredicted. This is assumed to happen due to the wallnormal distribution of the force density, which is too much concentrated to the immediate wall vicinity (for the Suzens model),
whereas a constantly decreasing force density is characteristic for
the Shyys model.
In contrast, application of the new empirical model results in an
accurate prediction of the wall jet proles inside as well as downstream of the discharge domain. Good agreement of the wall jet
proles inside the plasma demonstrate the improved spatial accuracy of the empirical model as compared to the phenomenological
approach. Consideration of the correct force magnitude F assures
furthermore the accurate prediction of the maximum wall jet

86

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

Fig. 7. Experimentally and computationally obtained mean velocity proles U(y) at selected streamwise locations x; implemented models: Eq. (12) new model, Eq. (8)
according to Wilke (2009), Eq. (9) according to Albrecht et al. (2011), Eq. (4) according to Shyy et al. (2002), Eqs. (2) + 5 and 7 according to Suzen et al. (2007); PIV data
according to Kriegseis et al. (2013): x = 0, 3, 6 mm FOV/lens #1, x = 10 mm FOV/lens #2 (cp. Table 1). The corresponding integral force value for the electrical operating
conditions is F = 25 mN/m (cp. Kriegseis et al., 2011, 2013).

velocity downstream of the momentum-transfer domain for

x P 6 mm.
6. Flow in an asymmetric 3D diffuser with inow modied by a
plasma actuator
The newly formulated model can only be plausible if tested
thoroughly. The ultimate goal is the application of the model proposed to ows of arbitrary geometrical complexity. Therefore, both
the PIV-based model and the present empirical plasma actuator
model have been applied to the ow in a three-dimensional diffuser, whose inow was appropriately manipulated by a streamwise-orientated (spanwise-forcing) arrangement of the present
plasma actuator mounted on the upper duct wall, Fig. 10. The Reynolds number based on the height of the inow duct (h = 1 cm) and
relevant mean ow velocity is Reh = 10,000. The corresponding
experimental investigation was performed by Grundmann et al.
(2011), whereas the baseline conguration without actuation
was measured by Cherry et al. (2008, 2009). Fig. 8 depicts the
geometry and dimensions of the 3D diffuser.

Accordingly, the fully-developed duct ow (width B = 3.33 cm)

expands into a diffuser whose upper wall and one side wall are
appropriately deected. The duct ow is characterized by secondary currents induced by the Reynolds stress anisotropy. Correspondingly, only an anisotropy-resolving turbulence model, such
as the presently employed Reynolds-stress model, is capable of
capturing such a ow structure. This is an important prerequisite
for capturing correctly the ow within the diffuser section. The adverse pressure gradient effect imposed onto the duct ow causes a
three-dimensional (corner) separation at smooth walls (characterized by non-xed separation region) occupying gradually the
upper diffuser wall in its entirety. Unlike in a number of ow congurations with a plasma actuator acting in a unidirectional manner with respect to the mean ow (with streamwise forcing
resulting in a momentum superimposed to the mean ow one),
the plasma actuator arrangement in the present 3D diffuser acts
perpendicular (spanwise forcing) to the mean ow through a pair
of counter-rotating streamwise vortices affecting directly the
secondary motion intensity and consequently the separated ow
within diffuser.

Fig. 8. Schematic of the asymmetric 3D Diffuser.

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

The solution domain consisting of a part of the inow duct

(length = 5 h), diffuser section (L = 15 h) and a fairly long straight
outlet duct (length = 30 h) was meshed by Nx  Ny  Nz = 150  91  121. The solution domain (it relates in particular to the outlet duct) as well as the grid size adopted presently are result of a
very intensive study performed in the framework of the ERCOFTAC
Workshop Series on Rened Turbulence Modeling (see e.g., Jakirlic
et al. (2010)). The proles of mean velocities and turbulence quantities at the inow plane (x =  5 h) are obtained by a separate
computation of the duct ow utilizing periodic inlet/outlet conditions underlying the fully-developed ow characteristics. The standard zero-gradient boundary conditions for all dependant variables
have been applied at the outlet.
Fig. 9 displays the mean velocity eld in a cross-section within
the inow duct of the non-actuated baseline conguration showing a good agreement between experiment and computation. The
velocity vector plot illustrates the secondary ow structure characterized by jets directed towards duct corners generating a pair of
counter-rotating vortices at each jets side. This phenomenon associated with rectangular ducts is the consequence of the Reynolds
stress anisotropy. Accordingly, the models based on the isotropic
eddy-viscosity concept are incapable of capturing these anisotropy-induced secondary currents. Correct representation of the
duct inow is important prerequisite for a successful diffuser computation. Similar as in the wall jet case the employment of the Reynolds stress model is here of decisive importance.

0.5
0.5

2.5

U/Ubulk

y/h

z/h
1

0.5
0.5

2.5

U/Ubulk

y/h

87

The plasma actuator arrangement mounted on the upper wall of

the inow duct is displayed in Fig. 10. It consists of one exposed
electrode situated in the middle of the upper wall and two
grounded (covered) electrodes placed at its both sides. The actuator conguration considered in the present work corresponds to a
continuously operating case implying steady forcing. The corresponding integral force value for the electrical operating conditions
is F = 12 mN/m (cp. Kriegseis et al., 2011, 2013).
Such a streamwise-oriented constellation enables a wall jet-like
ow directed towards upper duct corners. Fig. 11 illustrates the
corresponding velocity prole shape being typical of the wall jet
ow conguration, as discussed in the previous sections. Both plasma actuator models associated with the background Reynolds
stress model of turbulence resulted in a very good agreement with
the experimentally determined velocity evolution. The location
y = 0 coincides with the top duct wall. The thickness of the ow region directed towards duct corners is about 1.7  1.8 mm, which
corresponds closely to a sixth of the duct height. In the reminder
of the cross-section a low-intensity ow towards central duct
plane (z = B/2 = 1.665 cm) was generated, see also Fig. 12.
The wall jets induced by the plasma actuators at both ends of
the exposed electrode resemble a pair of counter-rotating streamwise vortices, see Fig. 12.
These jets oppose the secondary ow encountered in the baseline conguration. A certain strengthening of the secondary motion
intensity in the area of plasma actuator with respect to the mean
ow straining and its weakening in the duct corners and the lower
half of the inow duct relative to the case without actuation (Fig. 9)
is obvious. The immediate consequence of such an action is a more
uniform axial velocity eld in a larger portion of the duct modifying appropriately the turbulence structure whereas a certain increase of the turbulence activity is documented in the plasma
actuator locality an appropriate decrease of the turbulence intensity takes place in the ow regions characterized by a attened
velocity prole. It causes consequently the alteration of the separated ow topology within diffuser implying the shift of the recirculation zone from the upper wall (corresponding to the baseline
case) to the deected side wall.
The latter discussion is based on the computational results. The
nal outcome of the continuous steady forcing imparted to the
uid ow through afore-described plasma actuator arrangement
is represented through a respective pressure decrease compared
to the case without actuation. The corresponding pressure distribution at the bottom diffuser wall is shown in Fig. 13 illustrating
a steep rise of the pressure level immediately after entering the diffuser section until reaching a region characterized by a monotonic
increase. The lower pressure level associated with the actuated dif-

z/h

y/h

0.5

2.5

z/h
Fig. 9. Experimentally (upper) and computationally (middle and lower), utilizing
the present model formulation, obtained axial velocity contours and velocity
vectors at the very beginning of expansion (at x/h = 0) of the baseline conguration;
only one half of the inow duct is shown here.

Fig. 10. Schematic of the actuator placement and geometry for the spanwise
forcing according to Grundmann et al. (2011).

88

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

Fig. 11. Experimentally and computationally obtained mean velocity proles at

position 3 mm downstream of the junction of the two electrodes. The corresponding integral force value for the electrical operating conditions is F = 12 mN/m (cp.
Kriegseis et al., 2011, 2013).

tonically. This behavior is qualitatively correct captured by the

present model formulations.

0.5
0.5

2.5

U/Ubulk

y/h

z/h
1

bulk

0.5
0.5

2.5

U/U

y/h

z/h
1

y/h

Fig. 13. Pressure recovery along the diffuser using steady spanwise forcing;
implemented models: Eq. (12) new model, Eq. (8) PIV-based model; experimental
data: Grundmann et al., 2011.

0.5

2.5

z/h
Fig. 12. Experi mentally (upper) and computationally (middle and lower), utilizing
the present model formulation, obtained axial velocity contours and velocity
vectors at the very beginning of the duct expansion (at x/h = 0) of the actuated
conguration; only one half of the inow duct is shown here.

fuser conguration is predicted by both plasma actuator models in

a good agreement with experimental response. While the pressure
coefcient development pertinent to the baseline ow exhibits a
continuous monotonic increase, the actuated conguration is characterized by a certain plateau at location x/L = 0.4  0.7 implying
approximately a zero streamwise pressure gradient in this area
at the bottom wall. Afterwards the pressure rise continues mono-

7. Conclusions
In the present work a new model formulation is proposed to
simulate discharge-based ow manipulation. The model development relies on the velocity eld determined by a complementary
experiment performed by using the PIV technique in the area of
the plasma actuator. The major objective behind this new modeling approach is to provide a simple functional description of the
plasma actuator force distribution, while assuring accurate spatial
distribution of the force density. The model sets an empirical relationship between the force distribution and its spatial domain. The
ow eld induced by the present model is computationally investigated by the RANS method solving the homogeneous-dissipationbased, near-wall second-moment closure model for the unknown
Reynolds stress components due Jakirlic and Hanjalic (2002). The
new empirical plasma actuator model has been calibrated according to quiescent-air PIV experiments by Kriegseis et al. (2013). Furthermore, a comparative analysis of different volume-force
estimation strategies is provided, where particularly phenomenological and experiment-based approaches are distinguished. The
present work shows that the numerical result of the wall jet ow
inside the plasma region (0 6 x 6 6 mm) is extremely sensitive to
different discharge-force models.
Depending on the simulated ow-control scenario, this sensitivity is a key factor dening success or failure of a numerical simulation. Since any counteraction of ow instabilities (e.g. Tollmien
Schlichting waves, Grundmann and Tropea (2009)) requires a precisely adjusted force distribution to achieve successful ow-control
results, an accurate prediction of this distribution is of utmost
importance for a computational simulation of the ow behavior.
The present model, therefore, represents a contribution towards
such an accurate force-distribution prediction, which is based on
simple and straight forward applicable functional relationships.
As a rst step, the new model was applied to predict the wall jet
velocity prole evolution for the considered integral force value of
F = 25 mN/m for which a complementary experiment has been performed. The new empirical model as well as the PIV-based model
have been assessed in comparison with the relevant experimental
results as well as some most-widely used phenomenological models. Unlike the latter model formulations, both present models, the
PIV-based one and the empirical one, returned the wall jet velocity
development in very good agreement with experimental database.
Afterwards both models have been applied to a separated ow
in a three-dimensional diffuser whose inow structure was

I. Maden et al. / International Journal of Heat and Fluid Flow 41 (2013) 8089

manipulated by a pair of streamwise vortices issued by a spanwise-forcing plasma actuator arrangement operating continuously
at the top wall. The corresponding pressure decrease relative to the
non-actuated diffuser conguration was correctly captured agreeing well with relevant experiment of Grundmann et al. (2011).
Acknowledgment
The authors gratefully acknowledge nancial support by the
German Research Foundation (Deutsche Forschungsgemeinschaft,
DFG) under grant EXC 259.
References
Albrecht, T., Weier, T., Gerbeth, G., Metzkes, H., Stiller, J., 2011. A method to estimate
the planar, instantaneous body force distribution from velocity eld
measurements. Physics of Fluids 23, 021702-+.
Benard, N., Cattafesta, L.N., Moreau, E., Grifn, J., Bonnet, J.P., 2011. On the benets
of hysteresis effects for closed-loop separation control using plasma actuation.
Physics of Fluids 23, 083601-+.
Cattafesta, L.N., Sheplak, M., 2011. Actuators for Active Flow Control. Annual Review
of Fluid Mechanics 43, 247272.
Cherry, E.M., Elkins, C.J., Eaton, J.K., 2008. Geometric sensitivity of threedimensional separated ows. International Journal of Heat and Fluid Flow 29
(3), 803811.
Cherry, E.M., Elkins, C.J., Eaton, J.K., 2009. Pressure measurements in a threedimensional separated diffuser. International Journal of Heat and Fluid Flow 30
(1), 12.
Chun, D.H., Schwarz, W.H., 1967. Stability of the Plane Incompressible Viscous Wall
Jet Subjected to Small Disturbances. Physics of Fluids 10 (5), 911915.
Dejoan, A., Leschziner, M.A., 2006. Separating the effects of wall blocking and nearwall shear in the interaction between the wall and the free shear layer in a wall
jet. Physics of Fluids 18, 065110.
Eriksson, J.G., Karlsson, R.I., Persson, J., 1998. An experimental study of a twodimensional plane turbulent wall jet. Experiments in Fluids 25, 5060.
Glauert, M.B., 1956. The wall jet. Journal of Fluid Mechanics 1 (06), 625643.
Grundmann, S., Sayles, E.L., Eaton, J.K., 2011. Sensitivity of an asymmetric 3D
diffuser to plasma actuator induced inlet condition perturbations. Experiments
in Fluids 50, 217231.
Grundmann, S., Tropea, C., 2009. Experimental Damping of Boundary-Layer
Oscillations using DBD Plasma Actuators. International Journal of Heat and
Fluid Flow 30, 394402.

89

Jakirlic, S., Hanjalic, K., 2002. A new approach to modelling near-wall turbulence
energy and stress dissipation. Journal of Fluid Mechanics 439, 139166.
S. Jakirlic, G. Kadavelil, E. Sirbubalo, D. von Terzi, M. Breuer and D. Borello: 14th
ERCOFTAC SIG15 Workshop on Turbulence Modelling: Turbulent Flow
Separation in a 3-D Diffuser. Sapienza University of Rome, September 18,
2009, ERCOFTAC Bulletin, December Issue, No. 85, pp. 5-13, 2010.
Jayaraman, B., Shyy, W., 2008. Modeling of dielectric barrier discharge-induced uid
dynamics and heat transfer. Progress in Aerospace Sciences 44, 139191.
Jukes, T., Choi, K., Johnson, G., Scott, S., 2006. Characterization of Surface PlasmaInduced Wall Flows Through Velocity and Temperature Measurements. AIAA
Journal 44, 764771.
J. Kriegseis: Performance Characterization and Quantication of Dielectric Barrier
Discharge Plasma Actuators. PhD Thesis, Technische Universitt Darmstadt,
Germany, 2011.
Kriegseis, J., Grundmann, S., Tropea, C., 2011. Power Consumption, Discharge
Capacitance and Light Emission as Measures for Thrust Production of Dielectric
Barrier Discharge Plasma Actuators. Journal of Applied Physics 110, pp. 013305+.
Kriegseis, J., Schwarz, C., Tropea, C., Grundmann, S., 2013. Velocity-informationbased force-term estimation of dielectric barrier discharge plasma actuators.
Journal of Physics D: Applied Physics 46, 055202.
Maduta, R., Jakirlic, S., 2012. An eddy-resolving Reynolds stress transport model for
unsteady ow computations. In: Fu, S., Haase, W., Peng, S.-H., Schwamborn, D.
(Eds.), A dvances in Hybrid RANS-LES Modelling 4, Notes on Numerical Fluid
Mechanics and Multidisciplinary Design, vol. 117. Springer Verlag, pp. 7789
(ISBN 978-3-642-31817-7).
Mertz, B.E., Corke, T.C., 2011. Single-dielectric barrier discharge plasma actuator
modelling and validation. Journal of Fluid Mechanics 669, 557583.
Moreau, E., 2007. Airow Control by Non-Thermal Plasma Actuators. Journal of
Physics D: Applied Physics 40, 605636.
D.M. Orlov: Modelling and simulation of single dielectric barrier discharge plasma
actuators. PhD Thesis, University of Notre Dame, Indiana, USA, 2006.
Press, W.H., Teukolsky, S.A., Vetterling, W.T., Flannery, B.P., 1986. Numerical
Recipes: The Art of Scientic Computing. Cambridge University Press.
Roth, J.R., Sherman, D.M., Wilkinson, S.P., 2000. Electrohydrodynamic Flow Control
with a Glow-Discharge Surface Plasma. AIAA Journal 38 (7), 11661172.
Shyy, W., Jayaraman, B., Andersson, A., 2002. Modeling of glow discharge-induced
uid dynamics. Journal of Applied Physics 92, 6434-+.
Y. Suzen, P. Huang and D. Ashpis: Numerical Simulations of Flow Separation Control
in Low-Pressure Turbines using Plasma Actuators. AIAA-Paper 2007-937, 2007.
B.
Wilke:
Aerodynamische
Strmungssteuerung
mittels
dielektrischen
Barriereentladungs-Plasmaaktuatoren. PhD Thesis, DLR Gttingen, Germany,
2009.