You are on page 1of 11

Theor. Comput. Fluid Dyn.

(2008) 22: 305–315


DOI 10.1007/s00162-007-0068-8

O R I G I NA L A RT I C L E

F. Richez · I. Mary · V. Gleize · C. Basdevant

Zonal RANS/LES coupling simulation of a transitional


and separated flow around an airfoil near stall

Received: 3 May 2006 / Accepted: 7 August 2007 / Published online: 1 November 2007
© Springer-Verlag 2007

Abstract The objective of the current study is to examine the course of events leading to stall just before
its occurrence. The stall mechanisms are very sensitive to the transition that the boundary layer undergoes
near the leading edge of the profile by a so-called laminar separation bubble (LSB). In order to provide help-
ful insights into this complex flow, a zonal Reynolds-averaged Navier–Stokes (RANS)/large-eddy simulation
(LES) simulation of the flow around an airfoil near stall has been achieved and its results are presented and
analyzed in this paper. LSB has already been numerically studied by direct numerical simulation (DNS) or
LES, but for a flat plate with an adverse pressure gradient only. We intend, in this paper, to achieve a detailed
analysis of the transition process by a LSB in more realistic conditions. The comparison with a linear instability
analysis has shown that the numerical instability mechanism in the LSB provides the expected frequency of
the perturbations. Furthermore, the right order of magnitude for the turbulence intensities at the reattachment
point is found.
Keywords Stall · Transition · Laminar separation bubble
PACS 47.27.Cn

1 Introduction

Thanks to the Reynolds-averaged Navier–Stokes (RANS) approach, the calculation of flows around a whole
aircraft is nowadays possible. However many studies seem to prove that this approach is not mature for the
prediction of complex flow phenomena. For example, the numerical simulation of the flow around an airfoil
at static and dynamic stall is still a very challenging task for computation fluid dynamics (CFD). The dynamic
stall flow phenomenon concerns the delay in the stalling characteristics of airfoils that are rapidly pitched
beyond the static angle. Dynamic stall has serious implications in terms of achievable performance, which
need to be predicted accurately as soon as possible in the airfoil design cycle. The account of this phenome-
non is important for the design of many industrial domains as helicopter rotor blades (retreating blade) or jet
engines (rotating stall). This phenomenon is characterized by a massive unsteady flow separation and by the
formation of large-scale vortical structures leading to hysteresis effects. As a result, the maximum values of
lift and pitching moment highly exceed their static values. Hence, because of the manufacturing interest, the
Communicated by R.D. Moser
F. Richez (B) · I. Mary · V. Gleize
Department of Computational Fluid Dynamics and Aeroacoustics, ONERA,
29 Avenue de la Division Leclerc, BP 72, 92322 Châtillon, France
E-mail: Francois.Richez@onera.fr

C. Basdevant
Laboratoire d’Analyse, Géométrie et Applications, CNRS, Université Paris-Nord, 93430 Villetaneuse, France
306 F. Richez et al.

Fig. 1 Zonal RANS/LES coupling mesh

unsteady RANS (URANS) equations have been widely used to compute dynamic stall flow, but the results are
far from being totally reliable for predicting the dynamic stall behavior of airfoils [17]; moreover, they have a
large computational cost. In this context, some researchers have chosen to start from scratch by considering
static stall. They have proven that the solution shows grid and turbulent model dependency [4]. And even in
the simpler case of static stall, RANS simulation does not correctly predict the stall angle of attack and the
maximum lift. Other studies [14,18] have shown that this phenomenon is very sensitive to the transition of the
boundary layer at the leading edge. This may be one of the reasons why the RANS approach, which has not
been developed to treat transition, fails to predict stall occurrence. As large-eddy simulation (LES) provides
an effective tool for tackling such flow condition, we have decided, in this study, to use a zonal RANS/LES
coupling method [11] in order to gain a more physical description of the transitional and separated flows.
A first coupling simulation of the flow around an airfoil near stall with an LES domain including the whole
suction-side boundary layer and the wake zone, has been performed, so that it could be almost considered a
full LES. Moreover, we have paid particular attention to the mesh resolution in the transitional zone in order
to obtain an accurate description of the flow in this area. The numerical method, the flow, and the computa-
tion parameters are presented in the two first sections. Then, the results of this calculation are presented and
analyzed in the third section. In particular, a detailed account of the transition, numerically obtained at the
leading edge by an LSB, is given. To prove that the transition scenario observed numerically has a physical
meaning, a linear stability analysis of the mean flow in the transitional zone is presented in the last section. This
study confirms that the flow is highly unstable in this area and that the numerical oscillations observed in the
LES and which lead to transition result, as expected, from a two-dimensional (2D) inviscid Kelvin–Helmholtz
instability mechanism.

2 Numerical method

The compressible Navier–Stokes equations are solved in a multiblock structured solver based on a finite volume
method. Overlapping RANS and LES domains are used (see Fig. 1). In the RANS domain, the Spalart–Allmaras
model is used in order to take into account the pressure-side turbulent boundary layer. The selective mixed-scale
model is used to model the subgrid-scale terms [6]. At each time step, the values of the conservative variables
in the RANS domain of the overlapping zone are computed by averaging the LES variables in the cell volume
of the RANS domain. Moreover these conservative variables are used to transport the turbulent viscosity of
the Spalart–Allmaras model. The information transfer between the RANS and the LES domains requires some
boundary conditions. These are obtained by the use of ghost cells and an enrichment procedure [11].
The viscous fluxes are discretized by a second-order-accurate centered scheme. For efficiency reasons, an
implicit time integration is employed to deal with the very small grid size encountered near the wall. Then
a three-level backward differentiation formula is used to approximate the temporal derivative, leading to a
second-order accuracy. An approximate Newton method is used to solve the nonlinear problem. At each iter-
ation of the inner process, the resolution of the linear system relies on the lower–upper symmetric Gauss–Seidel
Zonal RANS/LES coupling simulation 307

(LU-SGS) implicit method. Usually LES requires a high-order centered scheme for the Euler fluxes
discretization in order to minimize dispersive and dissipative numerical errors. However such a scheme cannot
be applied easily in complex geometry. Therefore, a variant of the AUSM+(P) scheme, whose dissipation
is proportional to the local fluid velocity, is employed [9]. The original AUSM+(P) scheme proposed by
Edwards and Liou has been modified to be well adapted to low-Mach-number boundary layer simulations.
These numerical methods, developed at the Office National d’Etudes et Recherches Aérospatiales (ONERA,
The French Aerospace Lab) in the FLU3M code, have been validated in several applied computations [5].

3 Flow and computation parameters

The computation presented in this paper has been achieved for an OA209 profile, which is an helicopter fan
blade profile, with a chord of 0.5 m. The Reynolds number is 1.8 million and the inflow Mach number is
0.16. The angle of attack is 15◦ which is, according to experimental study achieved by the ONERA [12],
just prior to stall occurrence. The 2D RANS domain is a C-grid composed of 1, 153 × 101 points while the
three-dimensional (3D) LES domain is composed of 9.5 million points in the suction-side boundary layer and
by 1.6 million points in the wake zone. Compared to the RANS domain, the LES grid is more highly refined
in the streamwise direction and keeps the same resolution in the wall-normal direction (see Fig. 1). In order to
reduce the computational cost, the LES domain is decomposed into several subdomains, which differ in their
spanwise extent and spanwise resolution. The spanwise extent has been chosen in order to be greater than the
boundary layer thickness along the profile [11]. At the interface between these domains, the shorter domain
in the spanwise direction imposes flow periodicity, since its information is duplicated in the ghost cell of the
larger domain. As the scale of the turbulent vortices varies significantly in the streamwise and the wall-normal
directions, local mesh refinement in the spanwise direction is used and an enrichment procedure is performed
between the LES blocks. Hence, the LES domains are designed in order to satisfy all along the suction side
of the boundary layer these resolution conditions expressed in wall unit in the streamwise, wall normal, and
spanwise directions, respectively: ∆x+ ≤ 50, ∆y+ ≤ 2, and ∆z+ ≤ 15. According to earlier studies, this
grid resolution and the spanwise size should be sufficient to resolve the vortices of the turbulent boundary
layer. The code is parallelized using a domain decomposition technique based on the openMP directive. The
time step used in this simulation is ∆t = 1.3 × 10−7 s. With four inner iterations in the Newton process, larger
time steps have been found to give rise to numerically unstable behavior. The computation has been running at
13 Gflops for 5000 total central processing unit (CPU) hours on a four-processor NEC SX5 to ensure that the
unsteady solution fluctuates around a stationary averaged state. The statistical variables presented thereafter
have been computed during the last six periods of the vortex shedding that occurs at the trailing edge.

4 Numerical results

4.1 Aerodynamic characteristics

Figure 2 shows the lift and drag coefficients calculated by the LES and compared to experimental data [12]
and two RANS simulations using k–ω Wilcox and Spalart–Allmaras models. These RANS simulations have
been computed with the same RANS grid as shown in Fig. 1 and the same modified AUSM+(P) scheme has
been employed. The lift and drag coefficients have also been computed for stall angles of attack, with the
two RANS models, in order to show the inability of the fully turbulent RANS approach to predict the stall
phenomenon accurately. Looking at the values obtained at the 15◦ angle of attack (Table 1), one can see that
the RANS simulations overestimate the experimental lift and underestimate the drag, while the LES has the
quite opposite effect. The history of the LES lift and drag coefficients is shown in Fig. 3. Some oscillations
appear on these curves, due to vortex shedding at the trailing edge. The frequency f of this phenomenon can be
easily calculated and corresponds to a Strouhal number (based on the freestream velocity U0 and the boundary
layer thickness at the trailing edge δ) of St = f δ/U0 = 0.21 as expected. During the last five periods of the
vortex shedding, the lift seems to oscillate around a constant value, which means the solution has statistically
converged. The variation of the lift coefficient is about 6% of its averaged value. The variation of the drag
coefficient is however more important and represents 60% of its averaged value.
LES and RANS pressure distributions are almost identical and in good general agreement with the experi-
mental data (see Fig. 4). Some differences appear at the leading edge. The LES simulation provides a flattened
shape of the pressure distribution, which is characteristic of a separated flow. Let us notice that no such shape is
308 F. Richez et al.

0.05
1.5

0.04 Experiment
Experiment

Drag coefficient
Lift coefficient
1 RANS k-ω Wilcox
0.03 RANS SA
LES
0.5
0.02

0 Experiment
RANS k-ω Wilcox 0.01
RANS SA
LES
-0.5 0
0 5 10 15 0 5 10 15
Angle of attack (degrees) Angle of attack (degrees)

Fig. 2 Experimental and numerical lift and drag coefficients as a function of the angle of attack

Table 1 Experimental and numerical lift and drag coefficients

Experiment LES RANS SA RANS k-ω


Lift 1.416 1.366 1.487 1.490
Drag 0.029 0.039 0.023 0.021

observed for the fully turbulent RANS simulations because the premature increase of the turbulent viscosity at
the leading edge prevents the flow from separating. Looking at the experimental measurements, no conclusion
about the presence, location, and size of the LSB can be drawn. Indeed, the distance between two consecutive
measurement points has the same order of magnitude as the LSB length. Furthermore, the size of the LSB is
very sensitive to the freestream turbulence intensity and to the surface roughness [7,13]. It seems likely that

1.7 0.08
LES
1.6 0.07 RANS k-ω
RANS SA
Drag coefficient

Experiment
Lift coefficient

1.5 0.06

1.4 0.05

1.3 0.04

1.2 LES 0.03


RANS k-ω
1.1 RANS SA 0.02
Experiment
1 0.01
0 0.02 0.04 0.06 0 0.02 0.04 0.06
t (s) t (s)

Fig. 3 Lift and drag coefficients history compared to RANS and experimental values

Experiment
-10 LES
RANS k-ω
-10 RANS SA
-8
Cp

-8
-6

-6
-4
Cp

-4
-0.01 0.01 0.03 0.05
-2 x/c

0.25 0.75
x/c

Fig. 4 Pressure distribution


Zonal RANS/LES coupling simulation 309

0.06

0.002

0.04

0.001
0.02
Cf

Cf
0
0

-0.02
0.06

-0.001
-0.01 0 0.01 0.02 0.03 0.04 0.05 0.6 0.8 1
x/c LES M2 x/c
0.04 RANS k-ω
RANS SA
Cf

0.02

0 0.2 0.4 0.6 0.8 1


x/c

Fig. 5 Skin friction distribution across the whole upper surface of the airfoil

the LSB size will differ between experiment and LES, as long as these effects are not considered. However,
the sudden decrease of the experimental lift curve (see Fig. 2) leads one to suppose that the LSB is present in
the real flow. Indeed, it is now known that a steep drop of lift at stall is very likely due to the breakdown of the
LSB from the leading edge [3,10,13]. Even if the freestream turbulence intensity and the surface roughness
reduces the LSB length, the instability mechanisms responsible for the transition must be the same in the
experiment and in the LES. Therefore, although the LSB size might be overestimated by the LES, the effects
of this transition type on the downstream turbulent boundary layer should be comparable to the experiment.
Indeed, the presence of the LSB near the leading edge increases the thickness of the downstream turbulent
boundary layer, which might lead to a larger turbulent separation at the trailing edge. This may explain the
large discrepancy in the skin friction between RANS (which does not capture the LSB) and LES all along the
surface (see Fig. 5). The skin friction at the trailing edge shows that both RANS and LES provide a small
separated turbulent flow, which lies on more than 20% of chord for LES and between 12 and 14% for the
RANS simulations. The larger turbulent separated zone of the LES may be a reason for the lower value of
the lift. At the leading edge, the LSB, obtained with LES, is followed by the transition of the boundary layer,
as indicated by the steep slope of the skin friction. Many studies tend to prove that this transition mechanism
is a crucial parameter in stall occurrence [14,18]. Nevertheless, the transition mechanisms in a LSB are very
complex and are not easy to capture in a numerical simulation, all the more so in an applied configuration such
as a pitched airfoil. Then, in order to see the capabilities of LES to give an accurate description of the LSB,
a more detailed analysis of the flow in this area has been carried out. In this following analysis, we make an
attempt to provide helpful insights into the numerical and physical origins of what is observed in this LES.

4.2 Analysis of the transition

The analysis of the mean flow reveals that the laminar boundary layer separates at 0.4% of the chord from the
leading edge. Moreover, the averaged length and height of the LSB are, for this simulation, 1.3 and 0.03% of
the chord, respectively (see Fig. 6). Even if this structure is very small, detailed attention has been paid to the
mesh resolution in this area: within the LSB region, there are about 180 points in the streamwise direction and
40 points in the wall-normal direction. Figure 7 represents, for three different times, the streamwise evolution
of the three components of the instantaneous velocity at a constant distance from the wall. The streamwise
direction is made nondimensional with respect to the length of the LSB. The analysis of these instantaneous
data shows that the boundary layer is laminar and two dimensional at the separation point. The flow becomes
unsteady and three dimensional between 25 and 75% of the bubble length. At the reattachment point the flow
is highly unsteady and three dimensional. Furthermore, the study of the temporal evolution of the skin friction
shows that the instantaneous separation point does not move with time, but the instantaneous reattachment
point fluctuates greatly as the flow transitions and becomes highly unsteady. Its seems that the flapping of the
reattachment point leads to vortex shedding that spreads rearward and is quickly dissipated in the turbulent
310 F. Richez et al.

0.02
Hb/c=0.03%

0.015

y/c

Lb/c=1.3%
0.01

0.005 0.01 0.015


x/c
Fig. 6 Mean flow streamlines in the LSB

(a) at t1
2
t2
1.5 t3
u/U0

0.5

-0.5
0 0.5 1
x/Lb

(b) 2.5 at t1
2 t2
t3
1.5
v/U0

1
0.5
0
-0.5
0 0.5 1
x/Lb

(c) 1 at t1
t2
0.5 t3
w/U0

-0.5

0 0.5 1
x/Lb

Fig. 7 Instantaneous streamwise (a), wall normal (b), and spanwise (c) velocity components at 3% of the bubble length from
boundary, as a function of the streamwise coordinate, at three different times

boundary layer. Looking at the instantaneous Q criteria isosurface (see Fig. 8), one can see two-dimensional
vortex structures that appear in the second half of the bubble. It seems that a specific wavenumber is selected
before the flow becomes three dimensional at the reattachment point. A more detailed study presented hereafter
will confirm this assumption.
Zonal RANS/LES coupling simulation 311

Fig. 8 Instantaneous Q criteria isosurface near the transitional bubble colored by the streamwise velocity component

(a) (b) (c) (d) 0.015 (e)


0.06

0.0016
0.003 0.003
0.01
0.04
d/c

d/c

d/c

d/c
d/c

0.0008
0.0015 0.0015 0.005
0.02

0 0 0 0 0
0.025 0.05 0.075 0.1 0.25 0.5 0.15 0.3 0.45 0.6 0.05 0.1 0.15 0.2 0.025 0.05 0.075 0.1 0.125

urms/U 0 urms/U 0 urms/U 0 urms/U 0 urms/U 0

Fig. 9 Root-mean-square streamwise velocity profiles u rms at the separation point (a), in the middle of the LSB (b), at the
reattachment point (c), at 10% of the chord (d), and at 50% of the chord (e)

Figures 9 and 10 represent the root-mean-square (rms) streamwise and wall-normal velocity fluctuation
profiles in the local frame, in the LSB, and much further into the turbulent boundary layer. This confirms that
the flow is still laminar at the separation point. The maximum fluctuations occur at the reattachment point where
u rms attains 70% of the freestream velocity as can be seen from Fig. 9c. According to previous works [19] for
the LSB obtained on a flat plate with a semicircular leading edge, these maximum values seem to be strongly
overestimated. This is due to the fact that the velocity fluctuations are made nondimensional with respect to the
freestream velocity U0 while the flow is accelerated up to 3 × U0 in this area (see Fig. 11). Hence, considering
the edge velocity as a reference, one can find a maximum streamwise rms fluctuation of 23%, which is very
close to the value found by Yang et al. [19]. Alam and Sandham [1] found a maximum streamwise fluctuation
intensity of 16%. However, in their DNS, the transition results from the growth and stretching of -vortices,
which is a different mechanism from the 2D mode observed in the present LES. The experimental studies of
LSB usually give a maximum of streamwise fluctuations between 15 and 20%, depending on the flow parame-
ters. Hence, in this LES, the steep variation of the streamwise turbulence intensity near the reattachment point
is indicative of a quick transition, which is expected because of the highly unstable inflexional mean velocity
profile. Moreover, the turbulence intensity peak value at the reattachment point of the LSB seems to have the
right order of magnitude. Then, further downstream, the turbulence intensity reduces and reaches a typical
turbulent boundary layer level between 10 and 15% (see Figs. 9e and 10e). Note that the double-peaked shape
of the u rms and vrms profiles in the bubble (Figs. 9b, 10b) resembles a Tollmien–Schichting wave in a transi-
tional separated flow [2], and is different from the single-peaked profile described by Spalart and Strelets [16],
which might result from an absolute instability mechanism as suggested by Alam and Sandham [1].
To study the instability mechanism of the flow in the LSB more accurately, a Fourier transform of the
pressure signals at different locations along the bubble has been calculated. Figure 11 shows the mean velocity
312 F. Richez et al.

(a) (b) (c) (d) 0.015 (e)


0.06

0.0016
0.003 0.003
0.01
0.04
d/c

d/c

d/c

d/c
d/c
0.0008
0.0015 0.0015 0.005 0.02

0 0 0 0 0
0.025 0.05 0.075 0.1 0.25 0.5 0.15 0.3 0.45 0.6 0.05 0.1 0.15 0.2 0.025 0.05 0.075 0.1 0.125
vrms/U 0 vrms/U 0 vrms/U 0 vrms/U 0 vrms/U 0

Fig. 10 Root-mean-square wall-normal velocity profiles vrms at the separation point (a), in the middle of the LSB (b), at the
reattachment point (c), at 10% of the chord (d), and at 50% of the chord (e)

x/Lb=0 x/Lb=0.24 x/Lb=0.42 x/Lb=0.59 x/Lb=077 x/Lb=1

0.0012
y/c

0.0004

0 5 10 15
U/U 0

Fig. 11 Mean velocity profiles in the local frame at six stages in the LSB

120
x/Lb=1

100
0<x/Lb<1

80
SPL (dB)

x/Lb=0
60

40
f = 90000 Hz

20

4 5 6
10 10 10
f (Hz)
Fig. 12 Sound pressure level (SPL) at six stages in the LSB

profiles at six different locations in the LSB. The pressure spectrum at each location is represented in Fig. 12.
The amplitude increases as the section location moves downstream. Because of the inflection point of the mean
velocity profile, numerical truncation errors trigger the transition of the boundary layer. This type of mean
velocity profile is sufficiently unstable to force the flow transition without introducing artificial perturbations.
Figure 12 shows that these oscillations, essentially composed of the frequency f =90,000 Hz, appear in the
LSB and quickly grow when convected along the bubble until nonlinear effects lead to a large bandwidth spec-
trum. Furthermore, one can observe a very high-frequency peak in the pressure spectrum at around 600,000 Hz,
which seems to have no physical meaning. This peak only appears in the pressure spectrum and has not been
observed in the velocity spectra. This phenomenon might result from acoustic reflection at the edges of the LES
domains, but no proof of this hypothesis has been sought in this present study. Therefore, we have considered
these high-frequency waves to originate from numerical effects, and they are ignored hereafter.
Zonal RANS/LES coupling simulation 313

5 Temporal instability analysis of the transition

To ensure that this computation gives a physical description of the transition, a local linear inviscid stabil-
ity analysis [15] was carried out. This study should allow one to compare the calculated frequency of the
most highly amplified waves in the LSB simulation with the theoretical frequency of the most linearly unstable
waves that the mean flow undergoes. To perform this analysis, the continuity and incompressible Navier–Stokes
equations are considered as follows:
 
div u = 0 (1)
  1 1  
∂t u + div u ⊗ u = − grad p + div τ (2)
ρ ρ
 
with τ = µ grad u + t grad u .
The flow solution variables are then expressed as a superposition of the basic flow Q = (U0 , V0 , W0 , p0 )
 
and the disturbance quantities q p = u p , v p , w p , p p . The basic flow is assumed to be parallel:

Q = (U0 (y), 0, 0, p0 (y)) .


This assumption may appear to be very strong in the case of separated flow, but the wavelength is small enough
compared to the bubble length to consider the flow as locally parallel. Furthermore, this hypothesis greatly
simplifies the formulation and the resolution of this theoretical analysis.
Then, considering this basic flow Q solution of Eqs. (1) and (2), linearizing the Navier–Stokes system at
Q, and omitting the viscous terms, we can obtained the disturbances equations:
∂x u p + ∂ y v p = 0 (3)
∂t u p + U0 ∂x u p + v p ∂ y U0 = − ∂x p p (4)
∂t v p + U0 ∂x v p = − ∂ y p p (5)
According to Eq. (3), a stream function of the disturbances ψ can be introduced as follows:
u p = ∂y ψ
v p = −∂x ψ
Then, the system (3–5) can be reduced to the following equation:
(∂t + U0 ∂x )∆ψ − U0 ∂x ψ = 0 (6)
where a prime (’) denotes a y-derivative.
Using normal modes:

ψ(x, y) = φ(y)ei(kx−ωt)
we obtain an equation known as the Rayleigh equation:

(U0 − c)(φ  − k 2 φ) − U0 φ = 0 (7)


ω
where c = is the phase velocity.
k
The modes are discretized using the Chebyshev collocation. The matrix-eigenvalue problem is then solved
numerically for the dispersion relation D(ω, k) = 0.
The three mean velocity profiles, represented with squares, triangles, and circles in Fig. 11, have been
considered as basic flows. Then, a temporal analysis has been performed for each mean velocity profile. For
this type of analysis, a spatial disturbance characterized by its wavenumber is introduced into the flow, and
the temporal growth rate α = I maginar y(ω) and corresponding frequency f = Real(ω)/2π are computed.
The results obtained for different values of the real wavenumber k are plotted in Fig. 13. We can see that, as
expected, the inflectional mean velocity profile in the bubble is unstable via an inviscid Kelvin–Helmholtz
instability mechanism, because of the positive value of the temporal growth rate. Moreover, the most unstable
frequency for the three basic flows is around 90,000 Hz, which is in good agreement with the LES. Therefore,
the scenario of the numerically observed transition seems to be realistic. Indeed, it has already been shown
314 F. Richez et al.

30000

α (s-1)

10000
f = 90000 Hz

50000 150000
f (Hz)
Fig. 13 Temporal growth rate of a disturbance as a function of its frequency for three different mean velocities

that the transition in a LSB is dominated by two-dimensional Kelvin–Helmholtz waves even if three-dimen-
sional disturbances can exist and play an important role [2]. Let us also notice that the initial amplitude of the
perturbations is not controlled here, although Augustin et al. have shown this has a strong effect on the overall
size of the LSB.
This linear stability analysis has revealed that the wavelength of the most unstable mode has an order of
magnitude of 1/8 × L b . This shows the LSB is long enough to consider the parallel assumption of the linear
stability analysis as satisfactory. Furthermore, the grid resolution in the LSB is fine enough to have 20 grid
points by wavelength of the unstable mode. Hence, the grid resolution seems to be sufficient to resolve the
growth of this unstable mode accurately.
The question of the convective or absolute nature of this instability mechanism is not treated here, but
no low-frequency fluctuations coexisting with the high-frequency oscillations studied above have been found
despite the fact that they have been observed in other LSB simulations [8]. The reason may be that the reversed
flow in the bubble does not exceed 10%, which should make the flow only convectively unstable.

6 Conclusion

An LES of the flow around an airfoil profile at a high angle of attack and for a realistic Reynolds number has
been achieved. This simulation shows that the transition of the suction-side boundary layer via an LSB near
the leading edge can be captured by LES. Moreover, knowing that this structure plays an important role in the
stall phenomenon, particular attention can be paid to the resolution of the mesh in this area. Good agreement
of the calculated aerodynamic characteristics with available experimental and RANS data has been obtained.
Furthermore, we have also successfully compared the numerical results of the transition with a linear inviscid
instability theory. Hence, this simulation will be considered as a reference solution for ongoing simulations.
Subsequent studies will consist of reducing the LES domain to the transitional zone, first investigating
the influence of the LES domain size on the transitional flow and then studying the influence of the turbu-
lence model in the RANS domain in order to find out whether some turbulence models are more suitable
than the Spalart–Allmaras model for the coupling. Once the most well adapted turbulence model is found, the
RANS/LES coupling method will be applied to a stalled configuration.

Acknowledgments The authors would like to acknowledge Michel Costes from the ONERA Applied Aerodynamics depart-
ment for his advice and knowledge concerning stall phenomenon. We would also like to thank Lutz Lesshafft from the ONERA
CFD Department and Daniel Arnal and his team for enlightening discussions about instability mechanisms. We wish to thank
Guillaume Desquesnes from the ONERA CFD department, who has been helpful in the development of the instability analysis
code.
Zonal RANS/LES coupling simulation 315

References

1. Alam, M., Sandham, N.D.: Direct numerical simulation of ‘short’ laminar separation bubbles with turbulent reattachment.
J. Fluid Mech. 410, 1–28 (2000)
2. Augustin, K., Rist, U., Wagner, S.: Investigation of 2D and 3D boundary-layer disturbances for active control of laminar
separation bubbles. AIAA Paper 2003-613, Reno (2003)
3. Gaster, M.: The structure and the behaviour of laminair separation bubbles. NPL Aero Report 1181 ARC 28.226 (1966)
4. Gleize, V., Szydlowski, J., Costes, M.: Numerical and physical analysis of turbulent viscous flow around a NACA0015 profile
at stall. European Congress on Computational Methods in Applied Sciences and Engineering. Jyväskylä, Finland (2004)
5. Larchevêque, L., Sagaut, P., Mary, I., Labbé, O., Comte, P.: Large-eddy simulation of a compressible flow past a deep
cavity. Phys. Fluids 15(1), 193–210 (2003)
6. Lenormand, E., Sagaut, P., Ta Phuoc, L., Comte, P.: Subgrid-scale models for large-eddy simulations of Compressible Wall
Bounded Flows. AIAA J. 38(8), 1340–1350 (2000)
7. Mack, L.M. Transition prediction and linear stability theory. AGARD CP-224, pp 1-1–1-22 (1977)
8. Marquillie, M., Ehrenstein, U.: On the onset of nonlinear oscillations in a separating boundary-layer flow. J. Fluid Mech.
490, 169–188 (2003)
9. Mary, I., Sagaut, P.: LES of a flow around an airfoil near stall. AIAA J. 40(6), 1139–1145 (2002)
10. Mc Cullough, G.B., Gault, D.E.: Examples of three representative types of airfoil-section stall at low speed. NACA TN2502
(1951)
11. Nolin, G., Mary, I., Ta-Phuoc, L.: RANS eddy viscosity reconstruction from LES flow field for turbulent boundary layers.
AIAA Paper 2005-0143. Toronto, Canada (2005)
12. Pailhas G., Houdeville R., Barricau P., Le Pape A., Faubert A., Loiret P., David F. Experimental investigation of dynamic
stall. 31th European Rotorcraft Forum. Florence, Italy (2005)
13. Roberts, W.B.: Calculation of laminar separation bubbles and their effect on airfoil performance. AIAA J. 18(1), 25–30 (1980)
14. Sankar, L., Zibi-Bailly, J., Le Balleur, J., Blaise, D., Rouzaud, O., Rhee, M.: A comparative study of three methodologies for
modeling dynamic stall. In: 28th European Rotorcraft Forum, Bristol (2002)
15. Schmid, P.J., Henningson, D.S.: Stability and Transition in Shear Flows. Springer, Heidelberg (2001)
16. Spalart, P., Stelets, M.K.: Mechanisms of transition and heat transfer in a separation bubble. J. Fluid Mech. 403, 229–
249 (2000)
17. Szydlowski, J., Costes, M.: Simulation of flow around a static and oscillating in pitch NACA0015 airfoil using URANS and
DES. ASME Transfer/Fluids Engineering Summer Conference, Charlotte (2004)
18. Wilder, M.C., Chandrasekhara, M.S., Carr, L.W.: Transition effects on compressible dynamic stall of transiently pitching
airfoils. AIAA Paper 1993-2978, Orlando (1993)
19. Yang, Z., Voke, P.R.: Large-eddy simulation of boundary-layer separation and transition at a change of surface curvature.
J. Fluid Mech. 439, 305–333 (2001)

You might also like