On the dynamics of axisymmetric turbulent separating/reattaching flows

Pierre-Élie Weiss, Sébastien Deck, Jean-Christophe Robinet, and Pierre Sagaut

Citation: Physics of Fluids 21, 075103 (2009); doi: 10.1063/1.3177352

View online: http://dx.doi.org/10.1063/1.3177352
View Table of Contents: http://aip.scitation.org/toc/phf/21/7
Published by the American Institute of Physics

Articles you may be interested in

Unsteadiness of an axisymmetric separating-reattaching flow: Numerical investigation
Physics of Fluids 19, 065103 (2007); 10.1063/1.2734996

Control of the antisymmetric mode (m = 1) for high Reynolds axisymmetric turbulent separating/reattaching flows
Physics of Fluids 23, 095102 (2011); 10.1063/1.3614481

Spatio-temporal and modal analysis of unsteady fluctuations in a high-subsonic base flow

Physics of Fluids 26, 086101 (2014); 10.1063/1.4891257

Mixing enhancement behind a backward-facing step using tabs

Physics of Fluids 19, 105103 (2007); 10.1063/1.2781597
 PHYSICS OF FLUIDS 21, 075103 共2009兲

On the dynamics of axisymmetric turbulent separating/reattaching flows

Pierre-Élie Weiss,1 Sébastien Deck,1,a兲 Jean-Christophe Robinet,2 and Pierre Sagaut3
1
Department of Applied Aerodynamics, ONERA, Meudon FR-92190, France
2
SINUMEF, Arts et Métiers, ParisTech, 75013 Paris, France
3
Institut Jean Le Rond d’Alembert, Université Pierre et Marie Curie, 75252 Paris Cedex 05, France
共Received 27 October 2008; accepted 30 May 2009; published online 20 July 2009兲
The present work focuses on the intrinsic properties of an axisymmetric separating/reattaching flow.
A numerical simulation of a compressible flow over a cylinder extended by another cylinder of
smaller diameter is performed at a Reynolds number based on free stream velocity U⬁ and on the
diameter of the larger cylinder D of 1.2⫻ 106. Statistical and fluctuating properties are scrutinized
and compared with the available experimental data. Especially, experimental wall pressure
coefficient and rms pressure fluctuations are well-reproduced by the simulation. A large scale
coherent motion associated to fD / U⬁ ⬇ 0.2 rules the organization of the flow. Moreover, these
properties are linked to a linear stability analysis coupled with a two-point correlation analysis
unveiling the major importance of highly coherent antisymmetric modes in the flow behavior.
Finally, the axisymmetric body dynamics seems to be led by a significant absolutely unstable area
suggesting a global instability mechanism. © 2009 American Institute of Physics.
关DOI: 10.1063/1.3177352兴

I. INTRODUCTION works can be found in surveys of Laufer,8 Cantwell,9 and Ho

and Huerre.10 It has been recognized that the harmonic per-
Unsteady flow mechanisms such as wall-bounded turbu-
turbations in turbulent shear flows are governed by an insta-
lent shear phenomena with separation and reattachment have
bility mechanism, which has triggered a number of theoreti-
been studied for decades. Among them, axisymmetric and
cal models within the scope of the linear stability theory of
plane step flows can be distinguished. As emphasized by
turbulent flows 共Refs. 11–15兲. Moreover, Sandberg and
Deck and Thorigny,1 axisymmetric step flows present simi-
Fasel16 investigated numerically the instability mechanisms
larities with plane facing step flows regarding the shear layer
in supersonic axisymmetric wakes behind a bluff body for
instability process. Plane flows featuring separation from a
several Reynolds numbers and they finally related unstable
leading edge have been extensively studied in the past both
modes to coherent structures. Finally, Deck and Thorigny1
numerically and experimentally. Nevertheless, it is observed
conjectured the existence of an absolute unstable area in the
in literature that less attention was paid to the analysis of
middle of the recirculation zone on the present axisymmetric
axisymmetric configurations than to their plane counterpart.
configuration but in the case of a supersonic jet.
Actually, the amount of unsteady analysis and experimental
Considering all that has been said before, it seems of
data about axisymmetric afterbody flows is very limited in
interest to couple the unsteady analysis with the linear sta-
literature. A few authors investigated the disparities between
bility theory after a preliminary validation of the simulation
plane and axisymmetric afterbodies. Roshko2 stressed the
with the available experiments. To achieve this purpose, this
lack of a theory analogous to the one of Von Kàrmàn for the
article is organized as follows: characteristics of the tested
plane vortex street, Calvert3 studied the flow past cones and
configuration and numerical aspects of the simulation are
observed discrepancies regarding the periodic shedding
given in Sec. II. The governing equations and the associated
prominence, and Gai and Patil4 assessed the loss of effi-
numerical method used to perform the linear stability analy-
ciency of a subsonic blunted axisymmetric base controlled
sis are also mentioned. Section III displays results in terms of
by several devices known to yield significant drag reduction
instantaneous and mean fields validating the simulation at
in similar two-dimensional flows. On the experimental side,
the same time with the available experimental data. Then, a
Délery and Siriex5 reviewed results about intricate axisym-
spectral analysis exhibits frequencies of interest with their
metric afterbodies and showed that the wall pressure distri-
spatial distribution supporting, at last, a local linear stability
bution was dramatically influenced by the geometry. More
analysis determining the convective or absolute nature of the
recently, Deprés et al.6 highlighted the features of several
instability in the mixing layer zone.
axisymmetric base flows, depending on the occurrence of a
downstream solid reattachment of the free shear layer or not.
II. METHODS AND CONFIGURATION SETUP
Such flows are known to feature highly coherent struc-
tures 共Ref. 7兲 linked to the Von Kàrmàn shedding process. The axisymmetric afterbody dimensions 共see Fig. 1兲
The organized motions in turbulent flows have a long history, were chosen to dovetail with the experiments of ONERA’s
dating from 1940s, and extensive presentations of earlier S3Ch continuous research wind tunnel described by Deprés
et al.6 and later by Meliga and Reijasse.17 First, the configu-
a兲
Electronic mail: [email protected]. ration is composed of an axisymmetric blunt body with a

1070-6631/2009/21共7兲/075103/8/$25.00 21, 075103-1 © 2009 American Institute of Physics

075103-2 Weiss et al. Phys. Fluids 21, 075103 共2009兲

FIG. 1. Sketch of the geometry exhibiting the mesh, velocity vectors upstream from the separated area, characteristic dimensions, and the mean organization
of the flow.

diameter D equal to 100 mm. This body has a 120 mm long cation exists between the shear layer and the separating
extension which provides a L / D ratio of 1.2, where L is the boundary layer for laminar separation. In contrast, they ar-
emergence length, a diameter of 40 mm, and a height h for gued that this is not true for high Reynolds number turbulent
the axisymmetric step equal to 30 mm. The configuration is boundary layer. This conclusion is supported by several of
placed into a flow with a free stream Mach number of 0.702 their experiments. Among them, cross-correlation magni-
yielding a Reynolds number based on the forebody diameter tudes between velocities measured in the boundary layer and
ReD ⬇ 1.2⫻ 106. The initial external boundary layer thick- in the entrainment stream is a strong indication that a cou-
ness ␦ was obtained by modeling the necessary length for the pling does not exist between the two regions. One can notice
forebody to have a ␦ / D ratio equal to 0.2. that in these experiments, the separation is directly due to the
The approach used to model the flow is zonal detached abrupt change in geometry and that the Reynolds number is
eddy simulation 共ZDES兲 proposed by Deck18,19 which be- high which is also the case in our study.
longs to the RANS/LES 共Reynolds averaged Navier–Stokes/ The finite-volume solver FLU3M for the compressible
large eddy simulation兲 approaches 共for more details, see the Navier–Stokes equations developed by ONERA has been
discussion of Sagaut et al.20兲. ZDES has been proven to be used to perform the simulations on multiblock structured
efficient in such high Reynolds number configurations to grids. The approximation of time derivatives was carried out
simulate complex turbulent phenomena 共see Ref. 1 or Refs. using the Gear scheme presented by Péchier et al.25 which is
21 and 22兲. ZDES lets the user select RANS and DES areas. backward in time and second-order accurate. The spatial
In our present case, the motivation related to the use of scheme is a modified AUSM+ 共P兲 scheme proposed by Mary
ZDES was to model the area downstream the separated point and Sagaut.26 The accuracy of the solver for DNS, LES, and
of the boundary layer, which is geometrically defined, with hybrid RANS/LES purposes has been assessed in various
LES. Finally, the upstream part is computed using a URANS applications including transitional flows around a two-
共unsteady Reynolds averaged Navier–Stokes兲 model, which dimensional wing profile in near-stall conditions 共Ref. 26兲,
ensures a boundary layer with the expected integral proper- afterbody flows 共Ref. 27兲, cavity flows at high Reynolds
ties. As can be seen in Fig. 4共a兲, instable structures start to number 共Ref. 28兲, and synthetic jets in a cross flow 共Ref. 29兲.
grow at the corner of the larger cylinder 共y / D = 0.5兲, where In these last references, the numerical results are thoroughly
RANS solutions enter the LES domain. In other words, the compared to the available experimental data including spec-
coupling of URANS and LES ignores possible interaction of tral and second-order analysis.
coherent structures from the oncoming boundary layer over The mesh includes almost 12⫻ 106 hexaedric cells, with
the cylinder and the separation bubble. However, this pos- 240 cells in the azimuthal direction 共i.e., 1.5° per plan兲. The
sible interaction is assumed to be negligible which is broadly mesh represented in Fig. 1 is the part of interest of the whole
accepted for high Reynolds number massively separated mesh which is wider in all directions. Downstream the rear
flows. Indeed, on the experimental side, the issue of the re- body, an O–H topology has been set to avoid singularity
action of turbulent flows to a sudden change in wall bound- problem near the axis. Furthermore, as advised by Simon et
ary conditions was investigated by Morris and Foss.23 They al.,21 the early stages of the vorticity thickness development
emphasized that Holmes et al.24 suggested that a communi- are modeled with a 15-point-resolution which then rapidly
075103-3 On the dynamics of axisymmetric turbulent separating/reattaching flows Phys. Fluids 21, 075103 共2009兲

weakly nonparallel, i.e., it only depends on the r direction.

The present stability theory is based on the classical small
perturbations technique where the instantaneous flow is the
superimposition of the known mean flow and unknown fluc-
tuations. All the physical quantities q 共velocity, pressure,
etc.兲 are decomposed into a mean value and an infinitesimal
共␧ Ⰶ 1兲 fluctuating one. The streamwise and radial directions
are both inhomogeneous directions for the mean flow 关i.e.,
mean variables can be read Q̄共x , r兲兴. In the following the
classical weakly nonparallel approximation is used, and the
stability of the parallel flow is considered 共the dependency of
the mean flow to the streamwise direction is neglected at the
FIG. 2. Number of points N ML in the vorticity thickness. first order兲. For a cylindrical geometry the perturbation can
thus be written as
q共x,r, ␪兲=
¯ Q共r兲 + ␧关q̂共r兲exp关i共␣x + m␪ − ␻t兲兴 + c.c.兴. 共1兲
increases with the mixing layer growth up to almost a 60-
point-resolution in the region of impingement of the mixing Finally, the linearized Euler equations lead to the well-
layer 共see Fig. 2兲. known compressible Rayleigh equation written for the fluc-
A stability analysis is carried out in the recirculation area tuating pressure, in cylindrical geometry,
共see Fig. 1兲, i.e., for x / D 苸 关0 ; 1.2兴. In this area, the flow is
fully turbulent. However, in the numerical simulation, the
upstream boundary layer is computed by the URANS ap-
d2 p̂
dr2
+ 冋 1 1 d¯␳
− +
2␣
r ¯␳ dr 共␻ − ␣Ū 兲 dr dr
x
册
dŪx dp̂

冋 冉 冊册
proach. As a consequence, no turbulent structure results from
the upstream flow and the coherent structures observed in the m2
+ ¯␳共␣Ūx − ␻兲2M 2 − ␣2 + p̂ = 0 . 共2兲
recirculation zone are directly due to the unstable character r2
of the separated zone and/or of the mixing layer. Besides, the
The boundary condition imposed in the present problem ex-
high Reynolds number flow and the inflectional profile na-
presses that fluctuations do not vary far from the afterbody
ture justify the inviscid approximation for the stability analy-
and the viscosity is not considered at the wall.
sis 共see Refs. 30 and 31 for the justification of the inviscid
The compressible Rayleigh equation is integrated using
approximation兲. An analysis without the inviscid approxima-
a spectral collocation based on the Chebychev Gauss-
tion 共not shown here for the sake of brevity兲 has been per-
Lobatto points. Details of the numerical method used to dis-
formed that yields identical results to the inviscid case. In the
cretize Eq. 共2兲 are specified in Ref. 37. The resulting com-
present configuration, the flow is fully turbulent. The steady
plex nonsymmetric generalized eigenvalue problem is solved
axisymmetric base flow, i.e., the flow that would be observed
by using the direct LAPACK eigenvalue solver, which com-
if all perturbations were damped, is not accessible. In addi-
putes the whole spectrum of eigenvalues of the discretized
tion, continuation methods associated with the Newton algo-
operator. Newton–Raphson algorithm is eventually used to
rithm, widely used at low Reynolds numbers to compute the
follow spatial branches of solutions.
base flow beyond the threshold of instability, cannot be pur-
sued to such large Reynolds numbers. Only the mean flow
can be used to assess the stability properties, keeping in mind III. RESULTS AND DISCUSSION
that this mean flow is not a solution of the steady axisym-
A. Validation
metric Navier–Stokes equations. If the nonlinear feedback is
important, the stability properties differ between base and The unsteady aerodynamic field is time-averaged during
average flows. Some authors such as Hammond and the calculation. The mean streamwise pressure coefficient at
Redekopp,32 Pier,33 Barkley,34 and Barkley and Tuckerman35 the wall Cp = 共P − P⬁兲 / 21 ␳⬁U⬁2 is depicted in Fig. 3共a兲 as a
already studied the stability of average flow for laminar flow function of the location along the emerging body nondimen-
at low Reynolds number. Although the Reynolds number is sioned by three characteristic length scales, i.e., D the diam-
much higher than critical Reynolds number, the stability eter of the larger cylinder, h the step height, and Lr the reat-
analysis of the average flow allows to determine accurately tachment length. This figure allows to differentiate three
the global frequency of the system. However, at this regions described in Ref. 1 on the same axisymmetric con-
Reynolds regime, the stability analysis of base flow does not figuration with a supersonic jet. For x / Lr ⱕ 0, a slow de-
always allow to obtain the right flow dynamics, see Refs. 33 crease in Cp as the streamwise location increases highlights
and 36 for a detailed discussion. the upstream influence of the recirculation region on the
In the present paper, the mean flow 关Fig. 4共b兲兴 is com- main body. The second region shows a small decrease in Cp
puted by a double average both in time t and in azimuth ␪ of in the separation region from x = 0 up to x = Lr / 2 due to the
the ZDES solution. The general equations of motion for the acceleration of the backflow. The last region is characterized
instantaneous flow are the inviscid flow equations, the en- by a strong recompression process where the amplitude
ergy equation, written for the total energy, and the equation Cpmax is reached downstream the mean reattachment point at
of state for a perfect gas. The mean flow is supposed to be x / Lr ⬇ 1.15. The location of the reattachment is xr ⬇ 1.15D
075103-4 Weiss et al. Phys. Fluids 21, 075103 共2009兲

FIG. 4. 共a兲 Snapshot of the coherent structures downstream of the step flow
with their rotation visualized with an isosurface of QU⬁2 / D2 = 50 colored by
the sign of streamwise vorticity 共white for positive and gray for negative兲.
共b兲 time-averaged flow field. 共– –兲 Iso-Cp with ⌬C p = 0.02 between two
lines, 共–兲 isovelocity U / U⬁ = 0.01, and 共䊊兲 velocity profiles at several
streamwise locations.

dient tensor defining vortex tubes. In addition, the rotation is

shown through the coloring of these tubes with the sign of
the streamwise vorticity 共white for positive and gray for
negative兲.
A roll-up 共R兲 of toroidal eddies apparently occurs ap-
FIG. 3. 共a兲 Streamwise distribution of the pressure coefficient. 共〫兲 Ref. 6 parying and rapidly replaced by large three-dimensional
共kulites兲, 共䉭兲 Ref. 6 共steady taps兲, 共쎲兲 Ref. 17, and 共–兲 present case. 共b兲
Streamwise distribution of the coefficient of rms pressure fluctuation. 共〫兲
structures 共T兲. They develop as the shear layer approaches
Ref. 6 共kulites兲, 共쎲兲 Ref. 17, and 共–兲 present case. reattachment and are destabilized by azimuthal instability
modes according to Deck and Thorigny.1 Furthermore, the
occurrence of large scale hairpin vortices 共H兲 in the reattach-
and the dimensions of the recirculation zone is materialized ment zone is shown. After the break up of the eddies the
with a continuous line 关see Fig. 4共b兲兴 as well as some veloc- main structures appear to be longitudinal eddies 共L兲. Some
ity profiles along the configuration and iso-Cp contours in similarities with the axisymmetric bluff body discussed by
the flow field corroborating the values at the wall. Finally, it Sandberg and Fasel16 such as oblique or helical structures,
is worth mentioning that Cp values are in excellent agree- longitudinal structures, and several vortex loops can be no-
ment with the experimental data of Deprés et al.6 and Meliga ticed. However, their scales differ because the convective
and Reijasse17 on analogous configurations. Mach number and the Reynolds number are quite different.
The root-mean-square coefficient of the pressure fluctua- Furthermore, low frequency dynamics of the flow seems to
tion Cprms = P⬘2 / 21 ␳⬁U⬁2 , where P⬘2 is the fluctuating pressure be controlled by the first and not by the second detachment
is shown at the wall in Fig. 3共b兲. The fluctuation level in- of the near wake like in Ref. 16.
creases steadily in the streamwise direction downstream the
B. Spectral analysis
base and reaches a plateau just upstream from the mean shear
layer reattachment location 共Cprms ⬇ 0.035兲 which was con- Regarding the various scales and phenomena qualita-
firmed by the recent experiment of Meliga and Reijasse17 at tively observable in the instantaneous flow field, it appears
x / Lr ⬇ 0.9. Let us be reminded that Hudy et al.38 linked this that the complexity of the flow could be partially uncovered
growth in rms to the organized shear layer structures which through the scrutiny of the unsteady properties of the flow.
get stronger and move closer to the wall. Following this compass, Fig. 5 displays a linear-log axis plot
The behavior of the instantaneous flow field is illustrated of the contribution to the total energy of a peculiar frequency
in Fig. 4共a兲. The turbulent structures are evidenced plotting a band thanks to the estimated power spectral density 共PSD兲
positive isovalue of the second invariant of the velocity gra- function of pressure fluctuations, named G共f兲. This pressure
075103-5 On the dynamics of axisymmetric turbulent separating/reattaching flows Phys. Fluids 21, 075103 共2009兲

FIG. 6. 共Color online兲 Spectral sheet of normalized premultiplied PSD of

the pressure fluctuations at the wall fG p共f兲 / ␴2p for every streamwise and
FIG. 5. Normalized premultiplied PSD of the pressure at the wall for the
azimuthal locations along the emergence and for the particular dimension-
streamwise location x / D = 0.99. 共䊊兲 Ref. 17, 共䉱兲 Ref. 6, and 共–兲 present
less frequency StD = 0.2. 共– –兲 Spatial limits of the absolute area for m = 1.
calculation.

C共f,r,x,⌬␾兲 = 共Cr + jCi兲共f,r,x,⌬␾兲

spectrum, located at the station x / D = 0.99 among the rear S12共f,r,⌬␾,x兲
= 共3兲
body is represented with a solid line and compared to the 冑S1共f,r, ␾1,x兲S2共f,r, ␾2,x兲 ,
available experiments of Deprés et al.6 and Meliga and
Reijasse17 symbolized with a triangle and an open circle,
where j = 冑−1, and Cr and Ci are the real and imaginary part
respectively. One can notice that the broadband spectrum is
of the cross-spectral density function S12 and ⌬␾ = ␾1 − ␾2,
well reproduced by the simulation.
respectively. Assuming that the disturbances do not
This broadband high frequency contribution prevails exhibit any particular direction of propagation, one has
upon the spectrum since the organized shear layer structures, S12共⌬␾兲 = S12共−⌬␾兲. Besides, the hypothesis of isotropy
which grow in strength, move closer to the wall before crash- yields Ci = 0. Consequently, the Cr function is 2␲ periodic
ing against it and breaking into smaller scales. Note also that with respect to ⌬␾ and can be expressed as, thanks to a
this broadband contribution is centered around fLr / U⬁ Fourier transform in azimuthal modes,
⬇ 0.6, which agrees with the earlier conclusions of Mabey.39
This is the characteristic frequency of vortical structures seen ⬁
in free shear layers. In addition, this spectrum displays a Cr共f,⌬␾兲 = 兺 Cr,m共f兲cos共m⌬␾兲. 共4兲
peak at fLr / U⬁ ⬇ 0.1 close to former observations in two- m=0
dimensional separating/reattaching flows 共see Ref. 40 for in-
stance兲. This low frequency peak is often attributed 共see Ref. Cr,m represents the percentage of the fluctuating energy at
41兲 to the “flapping” of the shear layer and reflects the over- frequency f relative to the azimuthal mode m since
all separation-bubble growth/decay dynamics. Finally, the 兺mCr,m = 1. Let us be reminded that m = 0 and m = 1 modes
presence of the well-known “shedding” appears through a are characterized by an in phase and antiphase relationship of
peak near fD / U⬁ ⬇ 0.2. signals, respectively, recorded simultaneously at two dia-
In order to evidence the relationship between the fre- metrically opposed locations. Figures 7共a兲–7共h兲 show the
quencies of the absolutely unstable waves and those given by Cr,m spectrum for the first four azimuthal modes
the eigenvalue problem, PSDs of the pressure fluctuations at 共m 苸 关兩0 – 3兩兴兲 at the streamwise locations x / D = 1.15 and
the wall have been plotted for the whole flow field on the x / D = 0.72 in the separated area. One can notice the very
emergence. In addition, values of the energy at the particular good agreement with the experimental data by Deprés
et al.6共black triangles兲 and Meliga and Reijasse17 共black bul-
dimensionless frequency StD = 0.2 have been selected and re-
lets兲. The m = 0 contribution is confined at low frequencies
ported on the map G共x / D , ␪兲 shown in Fig. 6. It appears that
and decreases continuously with increasing frequencies
the spatial location of the Strouhal number StD = 0.2 matches
which is due to the low flapping frequency of the shear layer
well the absolute instability location provided by the eigen-
共see Ref. 1兲. The spectrum Cr,1 at 0.72D displays a peak near
value problem which constitutes a first step toward the iden- fD / U⬁ ⬇ 0.2 and shows that experimentally more than 80%
tification of the antisymmetric mode 共m = 1兲. of the pressure fluctuations at this specific frequency are due
To get deeper insight into the spatial organization of the to this antisymmetric mode. Figures 7共e兲 and 7共f兲 clearly
flow at these frequencies, the azimuthal coherence of two illustrate the fact that the energy related to the dimensionless
pressure sensors p1共r , x , ␾1兲 and p2共r , x , ␾2兲 located in a frequency fD / U⬁ ⬇ 0.2 has mainly been redistributed from
plane normal to the inflow at a constant position x and a the axisymmetric m = 0 mode to the antisymmetric m = 1
constant radius r can be considered. Assuming the hypothesis mode. Then, as represented in Figs. 7共c兲, 7共d兲, 7共g兲, and 7共h兲,
of an homogeneous flow, i.e., without any preferred angle of for higher modes 共m ⬎ 1兲 no particular frequency contributes
reference ␾1, the complex coherence function may be ex- for a significant amount of energy in the fluctuations. Then, a
pressed as Cr,m spectrum has been plotted for every streamwise location
075103-6 Weiss et al. Phys. Fluids 21, 075103 共2009兲

FIG. 8. 共a兲 Spectral sheet of the azimuthal pressure mode Cr,1 for every
streamwise location along the emergence in function of nondimensional
frequencies; 共– –兲 spatial limits of the absolute area for m = 1. 共b兲 m = 0 and
m = 1 modes for the axisymmetric case; 共– –兲 Im共␻兲m=0 = f共x / D , x / Lr , x / h兲,
共–兲 Im共␻兲m=1 = f共x / D , x / Lr , x / h兲, 共䊊兲 StDm=0 = f共x / D , x / Lr , x / h兲, and 共〫兲
StDm=1 = f共x / D , x / Lr , x / h兲.

mode m = 1兲 and of the absolute growth rate Im共␻兲 共dashed

line is for m = 0 and solid line is for m = 1兲 along the exten-
sion. The solid horizontal line denotes the threshold of mar-
FIG. 7. 关共a兲–共d兲兴 Spectra of the first four azimuthal pressure modes Cr,0, ginal absolute instability Im共␻兲 = 0. The mean flow appears
Cr,1, Cr,2, and Cr,3 for the streamwise location x / D = 1.15. 共䉱兲 Ref. 6, 共쎲兲
Ref. 17, and 共–兲 present calculation. 关共e兲–共h兲兴 Spectra of the first four azi- to be linearly absolutely stable for an axisymmetric distur-
muthal pressure modes Cr,0, Cr,1, Cr,2, and Cr,3 for the streamwise location bance 共m = 0兲 for all streamwise locations and frequencies.
x / D = 0.72. 共䉱兲 Ref. 6, 共쎲兲 Ref. 17, and 共–兲 present calculation. Concerning a helical disturbance 共m = 1兲, well represented in
Ref. 44, mean flow is linearly absolutely unstable for a broad
range of abscissa x / D 苸 关0.35; 0.75兴 close to StD ⬇ 0.2. The
providing a spectral layer 关Fig. 8共a兲兴. It exhibits the highest coexistence of absolute helicoidal 共m = 1兲 unstable global
percentages of the fluctuating energy for Strouhal numbers modes within the recirculation region and convectively un-
StD based on the forebody diameter close to 0.2 and which stable shear layer modes corroborates the results of Sandberg
are localized between x / D ⬇ 0.35 and x / D ⬇ 0.75. Let us and Fasel.16 To be able to compare with the previous ZDES
now consider a perturbation approach of this flow. results, it is necessary to determine if the flow is linearly
In the context of spatiotemporal stability analysis, the globally unstable because being absolutely unstable is only
singularities of the resulting dispersion relation are deter- necessary but not a sufficient condition 共Ref. 45兲. Different
mined and the ones satisfying the Briggs–Bers 共Refs. 42 and selection criteria for the global frequency, ␻g, whose success
43兲 pinching criterion have been selected. Frequency- depends on the type of considered flow, have been proposed
wavenumber pair 共␻0 , ␣0兲 solutions to the dispersion relation in the past, i.e., Pierrehumbert46 for the maximum growth
关i.e., ␻0共␣0兲 = 0兴 and corresponding to the zero group veloc- criterion, Koch47 for the hydrodynamic resonance criterion,
ity 关i.e., ⳵␻ / ⳵␣共␣0兲 = 0兴 are prospected by an iterative and Monkewitz and Nguyen48 for the initial growth criterion.
Newton–Raphson algorithm. Results are summarized in Fig. In the 1990s, the study of linear global stability of weakly
8共b兲. Figure 8共b兲 shows the streamwise evolution of the nonparallel shear flows provided the first criterion based on
Strouhal number StD based on the diameter of the larger solid theoretical grounds, Ref. 45. More recently, in the
cylinder 共circles stand for mode m = 0 and diamonds are for framework of the fully nonlinear Navier–Stokes equations,
075103-7 On the dynamics of axisymmetric turbulent separating/reattaching flows Phys. Fluids 21, 075103 共2009兲

the frequency selection criterion was obtained by Pier and

Huerre49 which generalized the initial growth criterion.
In this present compressible turbulent afterbody flow the
linear frequency selection criterion 共Ref. 45兲 is given by

⳵␻0
␻g = ␻0共Xs兲 and = 0, 共5兲
⳵ Xs

where Xs represents a saddle point in the complex X-plane.

However, since derivatives of ␻共X兲 are only known along the
real x-axis, the location of the saddle point Xs of ␻共X兲 was
found through use of the Cauchy–Riemann equations and
analytic continuation to complex values of X = xr + ixi. Details
of the method used are described in Ref. 32. Finally, the
location of the saddle point for this flow is Xs / D = 0.502
+ 6.7⫻ 10−3i nearly on the real x-axis and close to the loca-
tion of the maximum absolute growth rate. The linear selec-
tion criterion provides the best accuracy regarding the Strou- FIG. 9. 共Color online兲 共a兲 Streamwise evolution 共䉮兲 of the absolute wave-
hal number observed in numerical simulations. Indeed, this length ␭0 = f共x / D , x / Lr , x / h兲 and 共–兲 of the absolute growth rate Im共␻兲m=1
linear criterion points out that the dominant global mode is = f共x / D , x / Lr , x / h兲. The dotted lines stand for the wavelength which corre-
sponds to the absolute growth rate given by the linear frequency selection
close to StD = 0.227 which is in perfect agreement with ZDES criterion. 共b兲 Instantaneous spatial structure visualized for the m = 1 compo-
results 共StD ⯝ 0.2 on Fig. 5兲 while the nonlinear criterion nent of the real part of the pressure coefficient Re共p兲 with the absolute
gives StD = 0.325. The fact that stability analysis uses the dimensionless wavelength ␭0 / D.
average flow and not the base flow is a possible explanation
of this result. Pier33 showed for the cylinder wake at low
Reynolds numbers, when the average flow is used, that the
criterion of Chomaz et al.45 gave an excellent result. Appar-
IV. CONCLUSIONS
ently, although the Reynolds number is high, the presence of
the wall significantly reduces the absolute instability leading
Axisymmetric turbulent separating/reattaching flow has
to the global mode frequency selection by the linear criterion
been investigated at ReD ⬇ 1.2⫻ 106. First the simulation is
and not by the nonlinear criterion. On the other hand, the
well validated considering the very good agreement between
spectrum of the azimuthal antisymmetric pressure mode, de-
the Reynolds averaged data as well as the spectral content
picted in Fig. 8共a兲, finds a maximum of fluctuating energy
and the experiments.
precisely in the absolutely unstable area of the mean flow.
Figure 9共a兲 illustrates the streamwise variations of the It is shown that a large scale coherent motion at fre-
absolute wavelength 共symbolized by inverted triangles兲 de- quency fD / U⬁ ⬇ 0.2 dominates an ordered structure in this
fined as ␭0 = 2␲ / Re共k兲, where Re共k兲 is the real part of the separating/reattaching flow. These results corroborate the ex-
absolute wavenumber, and of the absolute growth rate Im共w兲 periments of Deprés et al.6 and Meliga and Reijasse.17 The
共solid line兲. The absolute wavelength at the location of the stability analysis has allowed us to show that axisymmetric
absolute frequency predicted with the linear frequency crite- turbulent afterbody dynamics is clearly dominated by a heli-
rion is equal to 2.05D. cal absolute instability while other modes 共m = 0 and m ⱖ 2兲
One can now consider an instantaneous pressure remain of convective nature. These results are in excellent
field p共r , ␪ , z兲 which is decomposed into azimuthal agreement with the azimuthal antisymmetric pressure mode
Fourier series according to the following expression: spectra. This last point is of major importance since the
p共r , ␪ , z兲 = 兺m pm共r , z兲eim␪. Figure 9共b兲 represents the real part analysis of the flow dynamics and the perturbation technique
of the pressure coefficient Re共p兲 for the m = 1 component. converge on the same spatial information.
Near the separation region, small scale structures are visible In future work, it would be interesting to relax the
becoming alternated large scale structures downstream the weakly nonparallel hypothesis of the actual stability analysis
emergence. by studying directly the global stability of such flow. An
The absolute wavelength shown in Fig. 9共a兲 compares example of such a possible approach is described in Ref. 51.
well with the spatial distribution observed in Fig. 9共b兲. Such The knowledge of the main mechanisms governing the flow
an agreement between the decomposition into azimuthal dynamics opens interesting perspectives of flow control. In-
Fourier series and the absolute wavelength resulting from the deed, considering the localized absolute area as a receptive
local stability analysis was observed by Meliga50 for a blunt- zone, the use of any passive or active device in this area
based bluff body featuring a fluidic reattachment. These clear could be the most efficient in modifying the flow physics.
agreements show that the dynamics of the axisymmetric af- The suppression of the absolute zone might modify the in-
terbody is dominated by a global instability phenomenon at stability nature of the flow from absolute to convective and
least for this regime. hence eliminate the low frequency dynamics.
075103-8 Weiss et al. Phys. Fluids 21, 075103 共2009兲

ACKNOWLEDGMENTS Dynamical Systems and Turbulence 共Cambridge University Press, Cam-

bridge, 1992兲.
25
The authors thank the Centre National d’Etudes Spa- M. Péchier, P. Guillen, and R. Caysac, “Magnus effect over finned projec-
tiales 共CNES兲 for financial support within the framework of tiles,” J. Spacec. Rockets 38, 542 共2001兲.
26
I. Mary and P. Sagaut, “Large eddy simulation of flow around an airfoil
the research and technology program ATAC 共aerodynamics near stall,” AIAA J. 40, 1139 共2002兲.
of nozzles and afterbodies兲. The Ph.D. work of P. E. Weiss 27
S. Deck, E. Garnier, and P. Guillen, “Turbulence modelling applied to
was funded by CNES and ONERA. space launcher configurations,” J. Turbul. 3, 57 共2002兲.
28
L. Larchevêque, P. Sagaut, I. Mary, and O. Labbé, “Large-eddy simulation
1
S. Deck and P. Thorigny, “Unsteadiness of an axisymmetric separating- of a compressible flow past a deep cavity,” Phys. Fluids 15, 193
reattaching flow,” Phys. Fluids 19, 065103 共2007兲. 共2003兲.
29
2
A. Roshko, “On the wake and drag of bluff bodies,” J. Aeronaut. Sci. 22, J. Dandois, E. Garnier, and P. Sagaut, “Numerical simulation of active
124 共1955兲. separation control by a synthetic jet,” J. Fluid Mech. 574, 25 共2007兲.
30
3
J. R. Calvert, “Experiments on low speed flow past cones,” J. Fluid Mech. W. Blumen, P. G. Drazin, and D. F. Billings, “Shear layer instability of an
27, 273 共1967兲. inviscid compressible fluid. Part 2,” J. Fluid Mech. 71, 305 共1975兲.
4 31
S. L. Gai and S. R. Patil, “Subsonic axisymmetric base flow experiments P. Huerre and P. A. Monkewitz, “Absolute and convective instabilities in
with base modifications,” J. Spacecr. Rockets 17, 42 共1980兲. free shear layer,” J. Fluid Mech. 159, 151 共1985兲.
5 32
J. Délery and M. Siriex, “Base flows behind missiles,” ONERA Technical D. Hammond and L. Redekopp, “Global dynamics of symmetric and
Report No. AGARD LS-98, 1979. asymmetric wakes,” J. Fluid Mech. 331, 231 共1997兲.
6 33
D. Deprés, P. Reijasse, and J. P. Dussauge, “Analysis of unsteadiness in B. Pier, “On the frequency selection of finite-amplitude vortex shedding in
afterbody transonic flows,” AIAA J. 42, 2541 共2004兲. the cylinder wake,” J. Fluid Mech. 458, 407 共2002兲.
7 34
A. Roshko, “Structure of turbulent shear flows: A new look,” AIAA J. 14, D. Barkley, “Linear analysis of the cylinder wake mean flow,” Europhys.
1349 共1976兲. Lett. 75, 750 共2006兲.
8 35
J. Laufer, “New trend in experimental turbulence research,” Annu. Rev. D. Barkley and L. S. Tuckerman, “Mean flow of turbulent-laminar pat-
Fluid Mech. 7, 307 共1975兲. terns in plane Couette flow,” J. Fluid Mech. 576, 109 共2007兲.
9 36
B. J. Cantwell, “Organized motion in turbulent flow,” Annu. Rev. Fluid D. Sipp and A. Lebedev, “Global stability of base and mean flows: A
Mech. 13, 457 共1981兲. general approach and its applications to cylinder and open cavity flows,” J.
10
C. M. Ho and P. Huerre, “Perturbated free shear layer,” Annu. Rev. Fluid Fluid Mech. 593, 333 共2007兲.
37
Mech. 16, 365 共1984兲. J.-Ch. Robinet, J.-P. Dussauge, and G. Casalis, “Wall effect on the
11
A. K. M. F. Hussain and W. C. Reynolds, “The mechanics of an organized convective-absolute boundary for the compressible shear layer,” Theor.
wave in turbulent shear flow,” J. Fluid Mech. 41, 241 共1970兲. Comput. Fluid Dyn. 15, 143 共2001兲.
12 38
A. K. M. F. Hussain and W. C. Reynolds, “The mechanics of an organized L. M. Hudy, A. M. Naguib, and W. M. Humphreys, “Wall-pressure-array
wave in turbulent shear flow. Part 2. Experimental results,” J. Fluid Mech. measurements beneath a separating/reattaching flow region,” Phys. Fluids
54, 241 共1972兲. 15, 706 共2003兲.
13 39
W. C. Reynolds and A. K. M. F. Hussain, “The mechanics of an organized D. G. Mabey, “Pressure fluctuations caused by separated bubble flows at
wave in turbulent shear flow. Part 3. Theoretical models and comparisons subsonic speeds,” Royal Aircraft Establishment Technical Report No.
with experiments,” J. Fluid Mech. 54, 263 共1972兲. RAE-TR-71160, 1971.
14 40
D. Wee, T. Yi, A. Annaswamy, and A. F. Ghoniem, “Self-sustained oscil- I. Lee and H. J. Sung, “Characteristics of wall pressure fluctuations in
lations and vortex shedding in backward-facing step flows: Simulation and separated and reattaching flows over a backward-facing step: Part 1. Time-
linear instability analysis,” Phys. Fluids 16, 3361 共2004兲. mean statistics and cross-spectral analyses,” Exp. Fluids 30, 262 共2001兲.
15 41
Y. Lifshitz, D. Degani, and A. Tumin, “Coherent perturbations in a turbu- D. M. Driver, H. L. Seegmiller, and J. Marvin, “Time-dependent behavior
lent boundary layer subjected to a highly adverse pressure gradient,” of a reattaching shear layer,” AIAA J. 25, 914 共1987兲.
42
AIAA Paper No. 2005-4809, June 2005. R. J. Briggs, Electron-Stream Interaction With Plasmas 共MIT, Cambridge,
16
R. D. Sandberg and H. F. Fasel, “Numerical investigation of transitional 1964兲.
supersonic axisymmetric wakes,” J. Fluid Mech. 563, 1 共2006兲. 43
A. Bers, in Physique des Plasmas, edited by C. Dewitt and J. Peyraud
17
P. Meliga and P. Reijasse, “Unsteady transonic flow behind an axisymmet- 共Gordon and Breach, New York, 1975兲, p. 117.
44
ric afterbody with two boosters,” Proceedings of the 25th AIAA Applied A. Weickgenannt and P. A. Monkewitz, “Control of vortex shedding in an
Aerodynamics Conference, Miami, June 2007, Paper No. 2007-4564. axisymmetric bluff body wake,” Eur. J. Mech. B/Fluids 19, 789 共2000兲.
18 45
S. Deck, “Zonal-detached-eddy simulation of the flow around a high-lift J.-M. Chomaz, P. Huerre, and L. M. Redekopp, “A frequency selection
configuration,” AIAA J. 43, 2372 共2005兲. criterion in spatially developing flows,” Stud. Appl. Math. 84, 119 共1991兲.
19 46
S. Deck, “Numerical simulation of transonic buffet over a supercritical R. T. Pierrehumbert, “Local and global baroclinic instability of zonally
airfoil,” AIAA J. 43, 1556 共2005兲. varying flow,” J. Atmos. Sci. 41, 2141 共1984兲.
20 47
P. Sagaut, S. Deck, and M. Terracol, Multiscale and Multiresolution Ap- W. Koch, “Local instability characteristics and frequency determination of
proaches in Turbulence 共Imperial College Press, London, 2006兲. self-excited wake flows,” J. Sound Vib. 99, 53 共1985兲.
21 48
F. Simon, S. Deck, P. Guillen, P. Sagaut, and A. Merlen, “Numerical P. A. Monkewitz and L. N. Nguyen, “Absolute instability in the near-wake
simulation of the compressible mixing layer past an axisymmetric trailing of two-dimensional bluff bodies,” J. Fluids Struct. 1, 165 共1987兲.
edge,” J. Fluid Mech. 591, 215 共2007兲. 49
B. Pier and P. Huerre, “Nonlinear self-sustained structures and fronts in
22
F. Simon, S. Deck, Ph. Guillen, and P. Sagaut, “Reynolds averaged Navier spatially developing wake flows,” J. Fluid Mech. 435, 145 共2001兲.
50
Stokes/large eddy simulations of supersonic base flow,” AIAA J. 44, 2578 P. Meliga, “A theoretical approach for the onset and control of unsteadi-
共2006兲. ness in compressible afterbody flows,” Ph.D. thesis, Ecole Polytechnique,
23
S. C. Morris and J. F. Foss, “Turbulent boundary layer to single-stream 2008.
shear layer: The transition region,” J. Fluid Mech. 494, 187 共2003兲. 51
J.-Ch. Robinet, “Bifurcations in shock wave/laminar boundary layer inter-
24
P. Holmes, J. L. Lumley, and G. Berkooz, Coherent Structures, Symmetry: action: Global instability approach,” J. Fluid Mech. 579, 85 共2007兲.