You are on page 1of 25

Wake transitions and laminar separation

bubble in the flow past an Eppler 61 airfoil


Cite as: Phys. Fluids 31, 114102 (2019); https://doi.org/10.1063/1.5113823
Submitted: 18 June 2019 . Accepted: 16 October 2019 . Published Online: 05 November 2019

Jawahar Sivabharathy Samuthira Pandi , and Sanjay Mittal

COLLECTIONS

This paper was selected as an Editor’s Pick

ARTICLES YOU MAY BE INTERESTED IN

Direct numerical simulations of multi-mode immiscible Rayleigh-Taylor instability with high


Reynolds numbers
Physics of Fluids 31, 112104 (2019); https://doi.org/10.1063/1.5127888

A dynamic wall model for large eddy simulation of turbulent flow over complex/moving
boundaries based on the immersed boundary method
Physics of Fluids 31, 115101 (2019); https://doi.org/10.1063/1.5126853

Effects of stroke deviation on hovering aerodynamic performance of flapping wings


Physics of Fluids 31, 111901 (2019); https://doi.org/10.1063/1.5124916

Phys. Fluids 31, 114102 (2019); https://doi.org/10.1063/1.5113823 31, 114102

© 2019 Author(s).
Physics of Fluids ARTICLE scitation.org/journal/phf

Wake transitions and laminar separation bubble


in the flow past an Eppler 61 airfoil
Cite as: Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823
Submitted: 18 June 2019 • Accepted: 16 October 2019 •
Published Online: 5 November 2019

Jawahar Sivabharathy Samuthira Pandi and Sanjay Mittala)

AFFILIATIONS
Department of Aerospace Engineering, Indian Institute of Technology Kanpur, Kanpur, UP 208016, India

a)
Author to whom correspondence should be addressed: smittal@iitk.ac.in

ABSTRACT
The flow past an Eppler 61 airfoil at 10○ angle to the free-stream is investigated numerically for 100 ≤ Re ≤ 87 000. Vortex shedding is observed
beyond Re ∼ 600 and three-dimensionality sets in at Re ∼ 1280.9 via the mode C instability and hairpin vortex structures that grow with an
increase in Re. At larger Re, the shear layer vortices, arising from the instability of the separated shear layer, interact with the flow close to the
airfoil and cause it to reattach. A Laminar Separation Bubble (LSB) forms at Re ∼ 20 000 and beyond. The airfoil experiences a very significant
increase in the lift and a decrease in the drag at the formation of LSB. Although the flow is three-dimensional, the primary mechanism of the
formation of LSB appears to be two-dimensional. The length of the LSB decreases with an increase in Re. The variation of the shear layer and
primary wake frequency, with Re, is studied. Both follow the power law. However, the variation before and after the formation of LSB is quite
different. Unlike the primary wake frequency, the shear layer frequency suffers a jump at the formation of LSB. Wake formation length (Lf ),
estimated via the spatial distribution of Reynolds stress, exhibits spanwise periodicity at the onset of three-dimensionality.
Published under license by AIP Publishing. https://doi.org/10.1063/1.5113823., s

I. INTRODUCTION airfoil to a longer extent. This leads to a reduction in pressure drag


even though the skin friction drag increases.45 In the present work,
With the advent of Micro-Air-Vehicles (MAVs), there is the transition of the flow including the evolution of the laminar sep-
renewed interest in the low Re (<5 × 105 ) flow past airfoils.1 In gen- aration bubble, for an Eppler 61 airfoil placed at an angle 10○ to the
eral, the maximum lift to drag ratio of airfoils increases with Re and uniform flow, is investigated.
improves considerably only beyond Re ∼ 1 × 105 .2 The range of Re Compared to the transition of the flow past an airfoil, the transi-
for a typical flow past an MAV is 104 –105 . Although there have tion of flow past bluff bodies is relatively well understood. The wake
been numerous efforts3–13 to investigate the flow in this range of of a circular cylinder undergoes transition to three-dimensionality
Reynolds numbers, several physical aspects related to the transition via three distinct instabilities: modes A, B (Refs. 14, 46, and 15–17),
of wake and boundary layer separation are not yet adequately under- and C (Refs. 18, 19, and 42). Table I lists some of the characteristics
stood. The wake of a nominally two-dimensional airfoil becomes of these modes. The mode A instability marks the onset of natural
unstable to three-dimensional instabilities at a relatively low Re.8,9 transition of the flow to three-dimensionality at Re ∼ 200. It is asso-
In certain cases, at high Re, the boundary layer undergoes laminar ciated with spatially periodic streamwise flow structures across the
separation and a subsequent turbulent reattachment on the suction wake with a spanwise wavelength of approximately 4D, where D is
surface forming a laminar separation bubble (LSB).10–13 The LSB the diameter of the cylinder. Mode-A instabilities, for the cylinder,
significantly affects the aerodynamic forces acting on the airfoil.10 are contaminated with vortex dislocations that cause disruption in
Airfoils that are designed for relatively high Re flows may lead to the periodicity of flow structures.14,17 For Re > 270, approximately,
early flow separation at low Re, thereby resulting in less than desir- the flow is marked by a gradual shift of energy from the mode A to
able performance for MAV applications at low Re.45,49 The Eppler 61 mode B instability. The mode B instability is associated with flow
airfoil is especially designed for low Re flows. The overall reduction structures with a spanwise wavelength of 0.8–1.0D. Unlike the mode
in drag is achieved via an early transition of the boundary layer to a A, the transition to mode B instability is not hysteretic. One of
turbulent state so that the flow remains attached to the surface of the the differences between modes A and B is the arrangement of the

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-1


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

TABLE I. Flow past a circular cylinder and an airfoil: the characteristics of modes A, B, and C reported by various researchers. λz is the spanwise wavelength; in the acronym
RT symmetry, R denotes reflection about the wake axis and T denotes translation in time. d is the characteristic length: diameter for the cylinder and c sin α9 for the airfoil. Red ,
for the airfoil, is based on d. The findings of Carmo et al.19 are for a single cylinder for modes A and B, while they are for two cylinders in the staggered arrangement for mode C.

Airfoil
NACA 0012

Circular cylinder α = 12.5 α = 15○
Mode A B C C C
18 19 14 18 14 19 18 19 42 43
Onset Re 180, 190, 200 230, 260, 260.5 170, 178, 165–180 1082 73043
Red 234.243 188.943
λz /d 3.97,19 4.0,18 3.0–4.014 0.825,19 1.014,18 1.57,19 2.018,42 1.38–1.4843 1.35–1.6243
Arrangement of
Staggered14 In-line14 ... ... ...
streamwise vortices
RT symmetry Odd19,42 Even19,42 Neither odd nor even 19,42
Neither odd nor even 43
Neither odd nor even43

streamwise vortices from one braid to the other.14 While they are of Re = 800 and 1200. Re is based on the chord length of the airfoil. They
alternate sign in mode A, they are of the same sign in mode B. proposed that for the airfoil, the characteristic dimension transverse
Beyond Re = 170, mode C instability is observed in the circular to the freestream is given by d = c sin α, where c is the chord of the
cylinder when the flow is excited via a control wire placed parallel to airfoil and α is the angle of attack. Similarly, Red = Re sin α is the
the axis of the cylinder in the wake.18 The mode C instability was also equivalent bluff body Reynolds number for the airfoil. The span-
seen in numerical simulations18 when zero velocity condition was wise wavelength (λz ) of the instability is 0.64d, approximately at
implemented on grid lines at the same locations as in experiments. Red = 273 (Re = 800), and it is similar to the mode B observed for
The spanwise wavelength of the mode C instability is 2.0D.18,42 It has a cylinder.
also been observed for flow past two cylinders in staggered arrange- Bourguet et al.8 investigated the transition of compressible flow
ment, with a spanwise wavelength of 1.57D.19 The mode C instability and, unlike the findings of Hoarau et al. for incompressible flow,
is also different from modes A and B in the sense that it exhibits observed irregular waves along the span of the NACA 0012 airfoil
period-doubling;19,42 the spatial arrangement of secondary stream- α = 20○ and Re = 800. λz was found to vary between 0.74 and 0.83c
wise vortices is the same at time t and t + 2T, where T is the time (2.16–2.42d) and is comparable to that for mode C for a circu-
period of vortex shedding.42 The time evolution of streamwise vor- lar cylinder. For a NACA 0015 airfoil, via stability analysis, Deng
ticity in the wake at cross flow axis for modes A and B exhibit odd-RT et al.43 found that the transition of the wake takes place via mode C
and even-RT symmetry, whereas for mode C it exhibits neither odd- instability at α = 12.5○ for Re = 1082 and at α = 15○ for Re = 730,
RT nor even-RT symmetry.19,42 R denotes reflection about the wake with λz = 0.3c (1.38d) and 0.42c (1.62d), respectively. The character-
axis, and T denotes translation in time. istics of mode C instability reported by Deng et al.43 are summarized
A stability analysis was carried out by Balachandar et al.40 in Table I. At relatively larger α (=17.5○ and 20○ ), several modes
for flow past a square cylinder for 161 ≤ Re ≤ 225 (Re based on were identified in addition to mode C instability. For example, at
the edge length, D). The onset of modes A and B was reported at α = 20○ and Re = 740, two modes are found with λz = 1.02d and
Re = 175 and 205, respectively. In addition, mode S was observed at 2.31d.
Re = 225. The spanwise wavelength (λz ) for the modes A, B, and S The aerodynamics of the airfoils at low Reynolds numbers is
is 5.22D, 1.2D, and 2.8D. The flow structure of mode S is similar to significantly affected by the presence of a laminar separation bubble.
mode C. However, unlike mode C, it appears naturally without any Carmichael7 presented flows past airfoils for Re ranging from 1000
forcing. to 200 000, with special reference to the laminar separation bubble.
In contrast to the transition of flow past a bluff body, Jones It is reported that for Re, based on the chord of the airfoil, between
et al.11 observed, via a computational study, that the transition of 10 000 and 30 000, the boundary layer is completely laminar. There-
the flow past a NACA 0012 airfoil at Re = 50 000 and 5○ angle fore, once it separates, it does not reattach. In the range 30 000 ≤ Re
of attack to the freestream, the three-dimensionality occurs via a ≤ 70 000, the laminar boundary layer separates and the free shear
mode-B like instability. The spanwise wavelength of this instabil- layer can transition to a turbulent state causing the flow to reat-
ity is similar to mode B instability observed in a circular cylinder. tach. Thick airfoils can exhibit a significant amount of hysteresis
The transition was induced by adding three-dimensional distur- in this regime. Most airfoils show laminar separation bubbles for
bances via volume forcing. The forcing was then removed and flow Re ∼ 50 000. There have been several efforts in the past to under-
statistics collected. It was found that the two-dimensional vortex stand the various physical phenomena associated with the laminar
shedding is absolutely unstable to three-dimensional flow perturba- separation bubble. The separated shear layer reattaches to the body,
tions. Hoarau et al.9 investigated the three-dimensional transition of as the flow becomes turbulent due to the merging of the shear layer
incompressible flow on the NACA 0012 airfoil placed at α = 20○ for vortices,5,6 forming a bubble on the suction surface. As a result, a

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-2


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

plateau appears in the pressure distribution curve.4 The length of the wake width increases with an increase in α and decreases with
the plateau approximately represents the length of the bubble. Based increasing Re. We study the variation of wake formation length with
on the length, Gaster4 classified the bubbles as short and long. He Re for an Eppler 61 airfoil at α = 10○ . Computation of the Reynolds
showed that at high Reynolds numbers, the bubble is extremely stress at each span location, using the sectional data, enables us to
small, of the order of a percent of the chord. With an increase in study the variation of these length scales across the span.
incidence of the airfoil or a reduction in flow speed, the short bub- All the computations in the present work are for an Eppler 61
ble may either burst to form a long bubble or the flow may fail to airfoil placed at α = 10○ with respect to the incoming uniform flow.
reattach.4 Either way it leads to an increase in drag and reduction in The angle of attack is well below the stall angle while it is still large
lift. Marxen and Rist20 and Marxen and Henningson,21 via experi- enough to induce flow instabilities that offer the possibility of com-
ments on a flat-plate, found that the cause of the bursting of bubble parison with those associated with bluff-body flows. The Reynolds
is insufficient turbulence in the shear layer for the flow to reattach number is varied in the range 100 ≤ Re ≤ 87 000. In addition to three-
to the body up to some far aft location. Subsequently, the separation dimensional computations, simulations are also carried out in two
bubble exhibits the characteristics of a long bubble. In this paper, dimensions. This provides an opportunity to compare the results
we attempt to track the formation of the laminar separation bub- from 2D and 3D computations that help us to understand the role of
ble, its evolution with Re, and its effect on the mean aerodynamic three-dimensionality on the flow and aerodynamic coefficients. The
coefficients for the Eppler 61 airfoil. motivation for this comparison comes from an interesting obser-
The instability of the shear layer and the associated shear layer vation from one of our earlier studies10 where it was found that
vortices are central to the formation of the laminar separation bub- computations carried out in 2D fail to predict the phenomenon of
ble. The shear layer instability in the context of bluff body flows has stall for the Re = 46 000 flow past an Eppler 61 airfoil. On the other
been studied quite extensively.22–26 Prasad and Williamson22 pro- hand, results from 3D computations are in reasonable agreement
posed that, for the flow past a circular cylinder, the variation of the with laboratory measurements.
normalized shear layer frequency (F SL /F K ) is proportional to Re0.67
for Re ≤ 5 × 104 . F SL and F K denote the frequencies associated with
II. GOVERNING EQUATIONS, SOLUTION METHOD,
the shear layer vortices and primary vortex shedding, respectively.
AND PROBLEM SETUP
They utilized data from several studies and sought the best fit over
the entire regime to propose the power law. Thompson and Houri- A. The Navier-Stokes equations
gan23 proposed a different power law using very similar data. They
Assume that Ω ⊂ IRnsd and (0, T) are the spatial and tempo-
suggested that the variation is closer to Re0.5 if one carries out a
ral domains, respectively, where nsd is the number of space dimen-
piecewise fit. They found that the shear layer frequency varies as
sions. Let Γ represent the boundary of Ω, while x and t denote the
Re0.57 for Re ≤ 1 × 104 . At higher Re, the variation is proportional
spatial and temporal coordinates, respectively. The Navier-Stokes
to Re0.52 . Singh and Mittal24 and Behara and Mittal,28 using the data
from their numerical simulations, also showed that the shear layer equations governing incompressible fluid flow are
frequency ratio exhibits a power law variation with Re. Compared ∂u
to the flow past a cylinder, the situation is quite different for an ρ( + u ⋅ ∇u − f) − ∇ ⋅ σ = 0 on Ω × (0, T), (1)
∂t
airfoil.
Yarusevych et al.6 found that the laminar separation bubble ∇⋅u=0 on Ω × (0, T). (2)
greatly affects the variation of the normalized shear layer frequency
Here, ρ, u, f, and σ are the density, velocity, body force, and stress
with Re. They carried out a study for the flow past a NACA 0025
tensor, respectively. The stress tensor consists of its isotropic and
airfoil for 0.55 × 105 ≤ Re ≤ 2.10 × 105 . It was demonstrated that
deviatoric parts,
the frequency ratio, F SL /F K , increases with an increase in Re prior to
the formation of the separation bubble. However, once the bubble 1
σ = −pI + T, T = 2 με(u), ε(u) = ((∇u) + (∇u)T ), (3)
is formed on the suction surface, the frequency ratio decreases with 2
further increase in Re. The difference in the two trends is attributed
where p, I, and μ are the pressure, identity tensor, and dynamic
to the difference in the rates of variation of the primary shedding
viscosity, respectively. The Dirichlet and Neumann-type boundary
frequency with Re in the two regimes.5,6 In the present effort, we
conditions are employed and represented as
investigate the variation of F SL /F K with Re for the flow past an Eppler
61 airfoil. The Re regimes pre- and postformation of the laminar u = g on Γg , n ⋅ σ = h on Γh , (4)
separation bubble are studied. The results are compared to earlier
results for the flow past a NACA 0012 airfoil reported by Huang and where g and h are fixed values. Γg is the subset of the total boundary
Lin.35 Γ on which the Dirichlet boundary conditions are specified, while
Two important characteristics of the wake of a body are the Γh is the part of the boundary, where the Neumann-type boundary
wake formation length (Lf ) and wake width (Lw ). Lf is the loca- condition is set. Here, n is the unit vector normal to the boundary.
tion of the maximum fluctuation of velocity in the wake, while Lw The initial condition on the velocity is provided on Ω at t = 0,
is the transverse separation in the two local peaks of the velocity u(x, 0) = u0 on Ω, (5)
fluctuation. Mahbub Alam et al.44 presented the variation of Lf and
Lw for a wing with the NACA 0012 section for 0○ ≤ α ≤ 90○ and where u0 is divergence free.
5300 ≤ Re ≤ 51 000. The estimates were based on measurements via Pressure and viscous stresses around the surface of the airfoil
Particle Image Velocimetry (PIV) at the midspan. It was found that are integrated for calculating the aerodynamic coefficient, Cx . Its

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-3


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

three components are the lift coefficient (CL ) that is the normal to TABLE II. 3D flow past an Eppler 61 airfoil for Re = 46 000 and α = 10○ : time-
the flow direction, the drag coefficient (CD ) along the flow direction, averaged lift and drag coefficients and the rms values obtained from model free
and the side force coefficient (CS ) along the spanwise direction. Cx is simulation and LES with the σ-model for closure.
given as
Simulation CL CD CLrms CDrms
1
Cx = 1 2 ∫ σ n̂dΓ, (6) Model free 1.5311 0.0686 0.0037 0.000 82
2
ρU∞ cLz Γafl
LES 1.5316 0.0687 0.0037 0.000 97
where n̂ is the unit vector that is normal to the airfoil boundary, Γafl ,
c is the chord, and Lz is the spanwise length of the airfoil.

The time- and span-averaged distribution of pressure- and skin-


B. Solution method friction coefficients on the surface of the airfoil, along with velocity
A stabilized finite element method implemented on a dis- profiles at certain chordwise locations, is presented in Fig. 1. The
tributed memory parallel computing machine is utilized to solve results from the model free simulation and LES are virtually indis-
the flow equations.24,25,34 The stabilized method is based on tinguishable. In view of this observation, all the results presented in
the SUPG29 (Streamline-Upwind/Petrov-Galerkin) and PSPG29 this study are with model free simulations, i.e., without any explicit
(Pressure-Stabilizing/Petrov-Galerkin) stabilization techniques. The turbulence model.
linear algebraic equation system resulting from the finite element
discretization of the governing equations is solved iteratively by C. The problem setup
employing the Generalized Minimal Residual (GMRES) procedure30 An Eppler 61 airfoil, of unit chord, is placed in a hexahedral
in conjunction with diagonal preconditioners. domain. Figure 2 shows the schematic of the computational domain.
Large Eddy Simulation (LES) is often utilized to compute flows The location of the various boundaries along with the boundary con-
at large Re, wherein the unresolved flow structures are modeled ditions is also shown in Fig. 2. A free-stream value is assigned to
using a subgrid scale model. One of the popular models is the the velocity at the upstream boundary, ABCD. At the downstream
Smagorinsky model31 where the molecular viscosity is supplemented boundary, EFGH, a Neumann-type boundary condition for the
with an eddy viscosity. In the present work, no explicit turbulence velocity is specified, which corresponds to zero stress vector. On the
model is used to represent the small scale flow structures. In these lateral walls, symmetry boundary conditions are employed, i.e.,
“model free” computations, the numerical viscosity introduced by the component of velocity normal to and the component of stress
the stabilization terms drains out the energy that the eddy viscos- vector along the boundary are prescribed a zero value. The airfoil
ity is expected to do for the unresolved small scales. Chopra and occupies the entire span, Lz , of the computational domain. No slip
Mittal39 showed that, for flow past a cylinder undergoing bound- condition for the velocity is specified on the surface of the airfoil.
ary layer transition, model-free computations capture most of the As shown in Fig. 2, the upstream and downstream boundaries are
flow features that are observed in measurements from experiments. located at a distance Lu and Ld , respectively, from the leading edge
Singh and Mittal24 and Akin et al.32 showed that in flow past a cylin- of the airfoil. Lh is the separation between the lateral boundaries. The
der in two dimensions, with linear elements, the numerical diffusion Reynolds number is based on the chord of the airfoil. The adequacy
due to the stabilization is much higher than the eddy viscosity gen- of the streamwise and lateral extent of the computational domain to
erated by the Smagorinsky model, except in regions very close to represent unbounded flow is investigated. Two domains, D1 and D2,
the cylinder where it gives unrealistically large values. The same is are utilized to study this. The domain D2 is, approximately, twice the
found to be true in three dimensions as well. Our experience shows size of D1 in each direction. For domain D1, the lengths Lu , Ld , and
that this high unphysical value of eddy viscosity interferes with the Lh are 5c, 11c, and 10c, respectively. These lengths, for domain D2,
transition of the flow including the dynamics of the LSB. The use are 10c, 21c, and 20c, respectively. The spatial resolution is approxi-
of a wall function is expected to alleviate some of these problems. mately the same for both the domains. At Re = 46 000 and α = 10○ ,
Another possibility is to use alternate models that account for wall the difference in time-averaged coefficients of lift and drag for the
effects.33,48 two domains is 0.03% and 2%, respectively, with respect to the val-
Flow at Re = 46 000 is considered to study the difference ues for domain D1. Since the results from the two domains are quite
between the results from a model free simulation and LES. The LES close, the domain D1 is used for all further computations presented
utilizes the σ-model proposed by Nicoud et al.33,48 In this model, in this article.
the turbulent eddy viscosity is defined as μt = (Cσ Δ)2 Dσ , where The adequacy of the spatial resolution of the finite element
Cσ = 1.35 is the model constant, Δ is the element length scale, and domain is tested via a mesh convergence study described in Subsec-
σ3 (σ1 − σ2 )(σ2 − σ3 ) 2 tion III and in more detail in Appendix A. Also tested is the effect of
Dσ = . σ 1 , σ 2 , and σ 3 are the singular values
σ1 Lz , the spanwise extend of the domain. Computations at Re = 1500
of the velocity gradient tensor.33,48 Table II lists the time-averaged and 46 000 are carried out for various Lz . It is found that the aerody-
lift and drag coefficients and their rms values obtained from the namic coefficients show no significant variation with Lz for Lz ⪆ 1c.
model free simulation and LES with the σ-model for closure. The This is described in detail in Appendix B. All the 3D computations
y+ , corresponding to the height of the element lying on the sur- in this study are carried out with Lz = 2c.
face of the airfoil in terms of the inner variables, is less than 1. The The computations have been carried out on a Linux-cluster
results from the two sets of simulations are in excellent agreement. using MPI libraries for interprocessor communication. Each

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-4


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 1. 3D flow past an Eppler 61 airfoil at Re = 46 000 and α = 10○ : comparison of time- and span-averaged (a) pressure distribution, (b) skin-friction coefficient, and (c)
velocity profiles at various chordwise locations on the surface of the airfoil for model free simulation and LES with the σ- model for closure.

compute node consists of two Intel Xeon processors with a clock implementation of the finite element formulation and its speed-up
speed of 2.5 GHz. The nodes are interconnected via an Infini- study can be found in our earlier work.34
Band network with a rated speed of 56 Gbps. The peak computing
performance of the cluster is 359.6 TF. Details of the parallel
III. THE FINITE ELEMENT MESH
The finite element mesh for the three-dimensional computa-
tions is constructed by stacking, along the span, several copies of
the two-dimensional mesh around an airfoil. The two-dimensional
mesh consists of a structured part close to the airfoil and an unstruc-
tured part, created via Delaunay’s triangulation, in the rest of the
domain. The structured mesh enables good resolution of the bound-
ary layer. The unstructured mesh allows one to generate grids
for complex geometries while still minimizing the number of grid
points.
To adequately capture the flow physics for the range of Re
considered while still using reasonable computational resources,
the mesh utilized for Re > 4000 is different from that used for
Re ≤ 4000. A mesh convergence study is conducted to investigate
the adequacy of the spatial resolution in 2D and 3D. We con-
sider flow at two representative Re (=1500 and 46 000). The flow at
Re = 1500 is associated with a three-dimensional instability with
a fairly regular spanwise structure. On the other hand, the flow is
FIG. 2. Flow past an Eppler 61 airfoil: problem setup and boundary conditions.
highly aperiodic along the span at Re = 46 000. In addition, a laminar

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-5


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 3. Flow past an Eppler 61 airfoil at α = 10○ : the evolution of flow states with increasing Re. The onset of three-dimensionality and transition of the wake followed by the
formation of laminar separation bubble is highlighted.

separation bubble (LSB) exists on the upper surface of the the separated shear layer, turbulent reattachment of the boundary
airfoil. layer, and the formation of a laminar separation bubble. With fur-
First, a mesh convergence study is carried out to resolve the 2D ther increase in Re, the separation bubble moves upstream and its
flow. Five meshes with varying resolution are considered. A mesh length decreases. Beyond a certain Re, the boundary layer achieves
with 26 530 nodes and 52 806 elements is found to be adequate for a turbulent state right at the leading edge of the airfoil. The laminar
the Re = 1500 flow, while a mesh with 46 168 nodes and 92 031 ele- separation bubble ceases to exist in this state.
ments is required to resolve the flow structures associated with the Figure 4 shows the variation of mean aerodynamic coefficients
Re = 46 000 flow. Next, a convergence study was undertaken to study with Re obtained with two- and three-dimensional simulations. Data
the spanwise resolution. Four values of N z , the number of elements for at least 30 time units, for the fully developed unsteady flow, are
along the span, were considered (N z = 100, 200, 250, and 300). The used to compute the time-average and the rms of the various quan-
3D mesh is formed by stacking the respective converged 2D mesh tities presented in this work. Time histories of lift and drag coef-
for the Re = 1500 and 46 000 flow. It is found that N z = 100 and ficients for the 3D computations, at certain representative Re, are
N z = 200 are enough to resolve the flow at the respective Re. The shown in Fig. 5. Most of the flow phenomena referred to in Fig. 3
details are presented in Appendix A. are imprinted in the variations shown in Figs. 4 and 5. The drag and
lift are constant-in-time for Re ≤ 500. The Re = 600 flow is asso-
IV. WAKE TRANSITION ciated with low amplitude unsteady oscillations. The transition of
the wake from a two- to three-dimensional state can be observed
A. Overview of the evolution of the flow with Re
from the divergence of the aerodynamic coefficients from 2D and
Computations for the Eppler 61 airfoil have been carried out for 3D computations shown in the insets of Fig. 4. The critical Re for the
100 ≤ Re ≤ 87 000. The angle of attack is α = 10○ . Figure 3 depicts the transition of flow from a two- to three-dimensional state is estimated
successive states of the flow with an increase in Re observed in the from the time histories of the spanwise component of velocity, w, in
present work. The flow is steady and two-dimensional for low Re. the wake at x/c = 1.25, y/c = −0.05, and z/Lz = 0.75.
Beyond Re ∼ 600, the primary instability of the wake leads to vor- For each Re, a 3D calculation is initiated from the fully devel-
tex shedding. Three-dimensionality in the flow sets in at Re ∼ 1280.9 oped 2D unsteady solution. The three-dimensionality is allowed to
via the appearance of hairpin vortex structures and mode C instabil- grow spontaneously from the base state. As an example, the time his-
ity. The size of the hairpin vortices increases with an increase in Re. tory of the spanwise component of velocity for Re = 1500 is shown
Their spanwise distribution is regular up to a certain Re and become in Fig. 6(a). The growth rate (λr ) of w, when the disturbance is small
irregular at Re = 1700 and beyond. The spanwise instability becomes and the flow is still in the linear regime, is estimated from the slope
stronger with an increase in Re. At higher Re, the separated shear of the envelope of the curve for the time history. The method to
layer becomes unstable and rolls into shear layer vortices. These estimate λr for Re = 1500 is shown in Fig. 6(a). A similar analysis
vortices enhance the mixing of the flow leading to the transition of is carried out at each Re. The variation of λr with Re is shown in

FIG. 4. Flow past an Eppler 61 airfoil at


α = 10○ : variation of time-averaged (a)
lift and (b) drag coefficients with Re. The
red triangle shows the critical Re for the
onset of three-dimensionality.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-6


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 5. 3D flow past an Eppler 61 airfoil


at α = 10○ : time histories of (a) lift and
(b) drag coefficients for various Re.

Fig. 6(b). The critical Re, corresponding to zero growth rate of w, is shows an additional frequency corresponding to 12 St for 1300 ≤ Re
estimated from this curve by extrapolating the growth rate to lower ≤ 1850 (225.73 ≤ Red ≤ 321.23). The power spectrum for Re = 1850
Re. It is estimated to be ∼1280.9 and marked in Fig. 6(b). Based on that brings this out is shown in Fig. 7(b). This additional frequency,
the length scale9 d = c sin α, the equivalent critical Re corresponds as will be shown later in this paper, reflects the period-doubling19,42
to Red ∼ 222.42. The critical Re for transition of flow from steady to nature of the three-dimensional instability.
unsteady and from the two- to three-dimensional state can also be Figure 8 shows the isosurfaces of the spanwise and streamwise
determined from linear stability analysis.47 This, however, has not components of instantaneous vorticity field of the fully developed
been undertaken in the present work. unsteady flow at various Re. The Re = 1300 flow is associated with
The Strouhal number (St) corresponding to the vortex shed- extremely weak three-dimensionality [Fig. 8(a)]. The insets in Fig. 4
ding frequency is also investigated for 2D and 3D calculations. Its show that both the 2D and 3D computations predict virtually the
variation with Red is shown in Fig. 7(a). The St is estimated from the same values of CL and CD for Re ≤ 1280. The three-dimensionality
time history of the transverse component of velocity, v, recorded by is weak at its onset and the spanwise undulation of the primary vor-
a probe located at x/c = 2.25 and y/c = 0.05. The length scale for both tices increases with an increase in Re as also the divergence between
St and Red is d = c sin α.9 In the 3D simulations, the probe is located the time-averaged coefficients obtained with 2D and 3D computa-
at midspan. St increases with Red for both 2D and 3D flows. How- tions (see Fig. 4). This transition in the wake is also noted in the
ever, beyond the onset of three-dimensionality, the 2D computations loss of periodicity in the time histories of the aerodynamic coeffi-
overpredict St. A similar trend is observed in the variations in CL and cients, as shown in Fig. 5. With an increase in Re, the wake of the
CD (Fig. 4). The time history of the v component of velocity, in 3D, airfoil is accompanied with small scale structures. At Re = 20 000

FIG. 6. 3D flow past an Eppler 61 air-


foil at α = 10○ : (a) time history of the
spanwise velocity component (w) for a
probe located at x/c = 1.25, y/c = −0.05,
and z/Lz = 0.75 for Re = 1500 flow and
(b) variation of growth rate (λr ) with Re.
The inset in (b) shows the critical Re,
determined by extrapolating the growth
rate.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-7


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 7. Flow past an Eppler 61 airfoil at


α = 10○ : (a) variation of St with Red
(lower axis) and Re (upper axis) from
2D and 3D calculations and (b) power
spectrum for Re = 1850. St is obtained
from the time-history of the transverse
component of velocity at (x/c, y/c)
= (2.25, 0.05). For the 3D computations,
the probe is located at the midspan.

[Fig. 8(g)], these small scale flow structures are observed close to laminar separation bubble. The critical Re for the formation of the
the leading edge of the airfoil. The abrupt increase in the mean lift LSB is different for 2D and 3D simulations. It is Re ∼ 7000 for the
coefficient, and decrease in the drag coefficient (Fig. 4), corresponds 2D computations, while it is Re ∼ 20 000 for the 3D simulations.
to the reattachment of the boundary layer and the formation of the McArthur,36 in his experiments with the Eppler 387 airfoil, found an

FIG. 8. Flow past an Eppler 61 airfoil at α = 10○ : isosurfaces of spanwise vorticity (ωz ) and streamwise vorticity (ωx ) for the instantaneous flow when the lift coefficient
achieves its peak value. The (Re, ωz , ωx ) for various cases are (a) (1300, ±2, ±0.01), (b) (2000, ±2, ±2), (c) (5000, ±2, ±2), (d) (8000, ±2, ±2), (e) (10 000, ±2, ±2), (f)
(19 000, ±2, ±2), (g) (20 000, ±2, ±2), and (h) (87 000, ±2, ±2).

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-8


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

abrupt increase in the lift coefficient beyond Re = 20 000. A narrow for Re ≥ 1700. Furthermore, even though ωx (t) is identical to ωx (t
region of hysteresis was also observed at the jump when the mea- + 2T), ωx (t + T) is no longer equal to −ωx (t).
surements were carried out with increasing and decreasing angles of The variation of the spanwise wavelength (λz /d) of the three-
attack, in increments of 0.1○ . dimensional instability with Red ,9 determined from the spanwise
variation of ωx shown in Fig. 10, is presented in Fig. 11. The Re
B. Mode C instability, its evolution, and variation based on the chord of the Eppler 61 airfoil is also marked on the
of spanwise wavelength with Re top of the plot. λz /d from the present study is compared with earlier
results in Fig. 11. Also shown in Fig. 11 is λz /d for other airfoils9,43 as
We attempt to identify the mode via which the 3D instabil-
well as λz /d for the wake of a circular cylinder,18 for mode C, at the
ity sets in. For the same, λz , RT symmetry (R denotes reflection
onset of transition. The numbers appear to compare well across dif-
about the wake axis and T denotes translation in time), and the pres-
ferent flows. λz /d, for the Eppler 61 airfoil, varies between 1.12 and
ence/absence of period doubling are studied for the Re = 1350 flow.
1.45d. For the dataset considered, the maximum value is attained for
Some of these features that have been studied in the past for certain
Red = 316.9 (Re = 1825).
other flows are listed in Table I. Figure 9 shows the instantaneous
vorticity for the Re = 1350 flow at three time instants corresponding
to t, t + T, and t + 2T, where T represents the time period of pri- C. Hairpin vortex structure
mary vortex shedding. We note that ωx (t) = −ωx (t + T) and ωx (t) The Q criterion defined by Hunt et al.41 is used to identify
= ωx (t + 2T), i.e., the flow has a time period of 2T. Figure 9(d) the vortical structures in the flow. The Q isosurface for Re = 1400
shows the space-time variation of streamwise vorticity in the wake at and rendered with spanwise (ωz = ±0.5) and streamwise
x/c = 2.5 and midspan for few cycles of primary vortex shed- (ωx = ±0.5) components of vorticity are shown in Figs. 12(a) and
ding. It is seen that the flow exhibits neither odd-RT nor even- 12(b), respectively, for the instantaneous flow at a time instant when
RT symmetry. The spanwise variation of ωx at three time instants CL achieves its maximum value. The spanwise primary vortices show
(t, t + T, and t + 2T) in the wake at x/c = 1.97 and y/c = 0.075 is shown undulations as well as hairpin structures. Each wave of the spanwise
in Fig. 10(a). From these variations, the spanwise wavelength, λz , of undulation is associated with a hairpin structure.
the three-dimensional instabilities in the flow is estimated to be λz /c To investigate the time evolution of the hairpin structures,
= 0.233 (λz /d = 1.34). On comparing the attributes of the 3D flow the instantaneous flow at various time instants during the develop-
instability for Re = 1350 with those reported by other researchers ment of three-dimensionality is studied. Figure 13 shows the time
for different flows (listed in Table I), we conclude that this flow is history of the spanwise component of velocity in the wake of the
associated with the mode C instability. airfoil at x/c = 1.25, y/c = −0.05, and z/Lz = 0.75 for Re = 1400.
We now investigate the evolution of mode C instability with Re The regime of linear growth of three-dimensionality is identified
by examining the spanwise variation of ωx shown in Fig. 10 for three in Fig. 13 via shading. Figure 14 shows the flow at various time
time instants at various Re. For Re ≤ 1600, the spanwise variation of instants that are marked in Fig. 13. The flow is devoid of hairpin
ωx at all these time instants is very regular and ωx (t) = −ωx (t + T) structures in most of the linear regime. They appear in the near
= ωx (t + 2T). The spanwise regularity reduces with an increase in Re wake at close to the end of the linear regime, as shown in Figs. 14(a)

FIG. 9. Flow past an Eppler 61 airfoil at Re = 1350 and α = 10○ : isosurfaces of streamwise component of vorticity (ωx = ±0.1) at (a) t, (b) t + T, and (c) t + 2T to illustrate the
mode C instability. The airfoil is also shown in the pictures. (d) Time evolution of streamwise vorticity along the y-axis at (x/c, z/Lz ) = (2.5, 0.5).

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-9


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 10. 3D flow past an Eppler 61 air-


foil at α = 10○ : spanwise variation of ωx
in the wake (x/c = 1.97, y/c = 0.075) for
various Re (a) 1350, (b) 1400, (c) 1600,
(d) 1700, (e) 1800, and (f) 1850 at t,
t + T, and t + 2T.

and 14(b). Once the flow achieves a limit cycle, the undulation An instantaneous picture of the flow in the near wake for the
of the primary vortices as well as the hairpin structures becomes fully developed unsteady flow at various Re is shown in Fig. 15. The
strong, as seen from the flow [Figs. 14(c) and 14(d)] at t = t c and t d , time instant, for all the cases, is the one at which the CL achieves
respectively. its peak value. The Q isosurface, colored by the pressure coefficient
(Cp ), is used to show the flow. The hairpin structures are barely vis-
ible for Re = 1300 [Fig. 15(a)]. They are fairly well developed and
almost uniformly distributed along the span for 1350 ≤ Re ≤ 1600
[Figs. 15(b)–15(d)]. Their size increases with an increase in Re. For
Re ≤ 1700, the distribution of hairpins along the span is no longer
uniform. Interestingly, the distribution for Re = 1700 is symmet-
ric about the midspan. This spatial symmetry is lost for Re > 1700
[Figs. 15(e)–15(i)]. In fact, for Re = 15 000 and 38 000 [Figs. 15(h)–
15(i)], the hairpin structures appear to be randomly interspersed in
the wake.
We attempt to study the evolution of the geometry of the hair-
pin structures with Re. The spanwise spacing between the centerlines
of two hairpins (hp ) and the distance between the two limbs of each
hairpin (hs ) are utilized to characterize the geometry of these struc-
tures. These dimensions are marked in Figs. 15(d)–15(f) for various
Re. For Re ≤ 1600, the distribution of hairpins along the span is uni-
form. Therefore, a single set of values of hs , and hp each, describe
FIG. 11. Variation of the normalized mean spanwise wavelength ( λdz ) of the three-
the geometry. However, for Re > 1600, the distribution of hairpins
dimensional instability with Red . Also shown is ( λdz ) for the NACA 001543 airfoil at is not uniform across the span. In such cases, we seek a spanwise
α = 20○ , α = 17.5○ , α = 15○ , and α = 12.5○ , the NACA 00129 airfoil at α = 20○ , average of hp . Figures 15(e) and 15(f) show the different spacings
and a circular cylinder.18 The Re shown on the upper edge, along the x-axis, is
between two consecutive hairpins along the span marked as hp1 ,
based on the chord-length of the Eppler 61 airfoil.
hp2 , and hp3 . The spanwise Cp distribution along a line that passes

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-10


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 12. Flow past an Eppler 61 airfoil


at Re = 1400 and α = 10○ : Q isosurface
(Q = 0.01) colored with (a) spanwise and
(b) streamwise components of vorticity.
Also shown are the close-up views of
hairpin vortices.

FIG. 13. 3D flow past an Eppler 61 airfoil at Re = 1400


and α = 10○ : time history of the spanwise component of
w velocity in the wake of the airfoil at (x/c, y/c, z/Lz ) = (1.25,
−0.05, 0.75). The regime of linear growth of the instability
is shaded.

through the hairpins is utilized to identify their spatial location. The span. In this regime, spacing between successive hairpins is the same
(x, y) coordinates for this line are suitably chosen for each Re. The Cp as the spanwise wavelength of the mode C instability, i.e., hp = λz . In
distribution is shown in Fig. 16. Consider, for example, the case of addition, in this regime, hs ∼ 0.5hp . In general, for all Re, the span-
Re = 1350 [Figs. 15(b) and 16(a)]. As shown in Fig. 15(b), the Cp on wise wavelength of the mode C instability is roughly the same as the
the prongs of the hairpin is larger than that on the primary vortices. average spacing between the hairpin structures and twice the spacing
We utilize this variation to estimate hs and hp . They are marked in between the two limbs of the hairpin.
Fig. 16. In those cases, where hp is not uniform along the span, the
various possible values of hp , as shown in Figs. 15(e) and 15(f), are
also marked. Figure 17 shows the variation of mean hp and hs with V. SPATIOTEMPORAL STRUCTURE OF THE FLOW
Red . For comparison, λz is marked as well. For 1300 ≤ Re ≤ 1600, The sectional lift coefficient Cl (z, t), at any spanwise location,
the hairpins are periodic and their distribution is uniform along the embodies the integrated signature of the flow past the airfoil at that

FIG. 14. 3D flow past an Eppler 61 airfoil at Re = 1400 and α = 10○ : instantaneous Q isosurface (Q = 0.01) at different time instants (a) t a , (b) t b , (c) t c , and (d) t d marked in
Fig. 13. The trailing edge of the airfoil is shown by dashed lines. The flow is from left to right.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-11


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 15. 3D flow past an Eppler 61 airfoil at α = 10○ : instantaneous Q isosurface colored with the coefficient of pressure for various Re. The range of the color map
(Cpmin = −C̃, Cpmax = C̃) and q the value of Q for the isosurface is different for each Re. (Re, ±C̃, q) for various cases are (a) (1300, ±0.05, 0.01), (b) (1350, ±0.05, 0.01),
(c) (1500, ±0.05, 0.01), (d) (1600, ±0.05, 0.01), (e) (1700, ±0.05, 0.01), (f) (1850, ±0.05, 0.01), (g) (4000, ±0.05, 0.01), (h) (15 000, −1, 0.2, 0.1), and (i) (38 000, −1.5, 0.2,
0.1). For each case, the time instant is when CL achieves its maximum value for the fully developed unsteady flow. Also marked is the spacing between the hairpins (hp ) and
their limbs (hs ).

spanwise section. The span-averaged lift coefficient CL (t) is related It is very interesting to observe the spatiotemporal evolution
L /2 ̃l (z, t) ̃l with Re. The flow exhibits periodicity both in space and
to Cl (z, t) as CL (t) = L1z ∫−Lz z /2 Cl (z, t)dz. We further define C of C
̃l (z, t) = Cl (z, t) − CL (t). It represents the spanwise fluctuation time at the onset of three-dimensionality and up to a certain Re.
as C
For example, for Re = 1300 and 1600 [Figs. 18(a) and 18(b)], the
in the sectional lift coefficient with respect to the span-averaged lift
̃l (z, t) at various Re flow is associated with a very regular spatiotemporal pattern. At
coefficient at any time instant. The variation of C
is shown in Fig. 18 and is utilized to understand the spatiotempo- Re = 1700, it retains temporal periodicity and a spatial symmetry
ral structure of the flow. The flow is two-dimensional and C ̃l (z, t) is about midspan [Fig. 18(c)]. The spanwise variations for Re = 1800
zero for Re < 1280.9. For Re ≥ 1280.9, the flow is three-dimensional and 1850 [Figs. 18(d) and 18(e)] are not periodic but still retain a
and associated with mode C instability and hairpins. Figures 18(a)– certain pattern while the flow is still associated with a temporal peri-
18(e) show C ̃l is periodic in time with a period of 2T, where T is the odicity. This pattern completely disappears at Re = 4000 and 12 000
time period for primary vortex shedding. As for the spanwise varia- [Figs. 18(f) and 18(g)], and the flow appears to be dominated by
tion of C ̃l , it is very regular for Re ≤ 1600 and becomes irregular for some large scale spanwise variations. The flow at Re = 46 000, on the
Re ≥ 1700. other hand, is primarily associated with small scale variation, both

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-12


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 16. 3D flow past an Eppler 61 air-


foil at α = 10○ : spanwise Cp distribution
for the fully developed unsteady flow for
various Re at a time instant when CL is
maximum. The (x/c, y/c) location is cho-
sen at each Re to bring out the spanwise
variation across the hairpins. The (Re,
x/c, y/c) for various cases are (a) (1350,
2.084, 0.084), (b) (1600, 2.046, 0.094),
(c) (1700, 1.898, 0.062), and (d) (1850,
1.905, 0.072).

along the span and in time. There is no discernible pattern along the 2D simulations show that at Re ∼ 7000, the shear layer vortices
̃l exhibits a virtually homogeneous distribution.
span, rather the C in the separated shear layer cause the flow to reattach to the sur-
face of the airfoil forming the laminar separation bubble. The same
occurs for 3D simulations at a larger Re (∼20 000). The forma-
VI. LAMINAR SEPARATION BUBBLE tion of the bubble has a significant influence on the aerodynamic
Figure 19 shows the spanwise component of vorticity, ωz , for coefficients: an increase in lift and a reduction in drag coefficient
the fully developed unsteady flow at a time instant when the lift (see Fig. 4).
coefficient is at its peak value. The vorticity field and the stream- Figure 20 shows the time- and span-averaged flow for the 3D
lines for the span- and time-averaged flow, for the 3D-computations, computations along with the u′ v ′ component of the Reynolds stress.
are shown in Fig. 20. Figures 19 and 20 show the role of the The Reynolds stress represents the intensity of unsteadiness in the
shear layer vortices in the formation of the laminar separation bub- flow. As shown in Fig. 19(e), for the Re = 10 000 flow, the sepa-
ble as well as its evolution with an increase in Re. The separated rated shear layer rolls into small vortices toward the aft section of
shear layer becomes unstable beyond a certain Re and rolls up into the airfoil. The u′ v ′ component of Reynolds stress is observed to be
shear layer vortices. The location of the formation of these vortices relatively large in this region, as shown in Fig. 20(a). At Re = 19 000,
moves upstream, toward the leading edge, with an increase in Re. the location of the onset of shear layer instability moves upstream
and the shear layer vortices become stronger [Fig. 19(f)]. The same
is observed from Fig. 20(b). As Re increases further, the shear layer
vortices begin to form further upstream, thereby getting closer to the
upper wall of the airfoil. This facilitates their interactions with the
flow close to the surface of the airfoil. At a certain Re, this interaction
energizes the flow close to the wall of the airfoil causing it to reattach.
The time-averaged vorticity field and streamlines in Fig. 20(c) clearly
show the appearance of laminar separation bubble at Re = 20 000.
On further increase in Re, the bubble moves upstream as seen from
Figs. 19(h) and 20(d).
Figures 19(g) and 20(c) show the mechanism of the formation
of the LSB. The boundary layer is associated with very low u′ v ′ at
the point of separation. However, the unsteadiness in the rolled up
vortices formed due to the instability of the separated shear layer
leads to large u′ v ′ causing a reattachment of the flow. Figure 20(c)
FIG. 17. 3D flow past an Eppler 61 airfoil at α = 10○ : variation of mean hp /d, hs /d clearly shows that the u′ v ′ is large toward the aft end of the LSB. The
for the hairpin structures and λz /d for the mode C instability, with Red . Re based
same is observed in Fig. 20(d) for higher Re (=46 000). Figure 21(a)
on the chord of the airfoil is shown on the upper edge along the x-axis.
shows the close-up view of the time- and span-averaged streamlines

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-13


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

̃l for various Re. The range of the color map (C


FIG. 18. 3D flow past an Eppler 61 airfoil at α = 10○ : space time contour of C ̃lmin = −C̃, C
̃lmax = +C̃) is different for each Re.
(Re, ±C̃) for each case is (a) (1300, ±5 × 10−5 ), (b) (1600, ±1 × 10−4 ), (c) (1700, ±5 × 10−4 ), (d) (1800, ±1 × 10−3 ), (e) (1850, ±1 × 10−3 ), (f) (4000, ±1 × 10−2 ), (g)
(12 000, ±2 × 10−2 ), (h) (20 000, ±2 × 10−2 ), and (i) (46 000, ±1 × 10−2 ).

at this Re along with the mean velocity profiles at various chordwise onset of reattachment is associated with a sudden upstream move-
locations. The region of reverse flow, on the upper surface, is clearly ment of the separation point by 10% of chord, approximately. As
seen. The velocity profiles in terms of the inner variables (u+ and y+ ) already seen from the flow pictures, compared to the 3D computa-
are shown at x/c = 0.32 and 0.52 in Fig. 21(b). Both these locations tions, the onset of reattachment of the flow occurs at a lower Re for
are downstream of the LSB after the flow has undergone a turbu- the 2D computations. This is related to the relatively stronger shear
u yuτ
lent reattachment. u+ and y+ are defined as u+ = and y+ = , layer vortices in the 2D computations. With an increase in Re, both
√ uτ ν separation and reattachment points move toward the leading edge.
where uτ (= C̄f /2U∞ ) is the friction velocity and ν is the kine- It is interesting to note that beyond Re ∼ 20 000, both the two- and
matic viscosity of the fluid. Both the velocity profiles shown in three-dimensional simulations predict the same location of separa-
Fig. 21(b) exhibit log-law beyond the viscous sublayer, indicating a tion and reattachment points and the length of the bubble. This is
turbulent boundary layer. in line with the suggestion by Braza et al.27 and Singh and Mittal24
To further investigate the LSB, we track the locations for flow that the formation of the laminar separation bubble is primarily a
separation and reattachment with Re. Figure 21(c) shows the vari- two-dimensional phenomenon. Figure 21(d) shows that the length
ation, with Re, of flow separation and reattachment points on the of the LSB is largest at its onset and decreases with an increase in Re.
upper surface of the airfoil. The length of the separation bubble, Lb , Figure 21(c) indicates that the separation point is located very close
is estimated via the difference between the streamwise location of the to the leading edge in this entire regime of Re. The upstream move-
reattachment and separation points. Its variation with Re is shown ment of the reattachment point causes the shortening of the LSB
in Fig. 21(d). We note that both in the 2D and 3D simulations, the with an increase in Re.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-14


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 19. Flow past an Eppler 61 air-


foil at α = 10○ : spanwise vorticity (ωz )
for the instantaneous flow at various Re
(a) 5000, (b) 7000, (c) 10 000, (d) 19 000,
(e) 10 000, (f) 19 000, (g) 20 000, and
(h) 46 000. The pictures are taken at
a time instant when the lift coefficient
achieves its peak for the fully evolved
unsteady flow. The pictures for the 3D
flow are at midspan.

FIG. 20. Flow past an Eppler 61 airfoil


at α = 10○ : spanwise component of vor-
ticity for the time- and span-averaged
flow (left) and u′ v ′ component of the
Reynolds stress superimposed with the
time- and span-averaged streamlines
(right) for various Re (a) 10 000, (b)
19 000, (c) 20 000, and (d) 46 000.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-15


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 21. Flow past an Eppler 61 airfoil at α = 10○ : (a) close-up view of time- and span-averaged streamlines for Re = 46 000 flow showing the laminar separation bubble
(LSB) along with the mean velocity profile at various chordwise locations. Also marked are the separation (S) and reattachment (R) points on the upper surface of the airfoil.
The velocity profiles in terms of the inner variables are shown for x/c = 0.32 and 0.52 in (b). Dashed lines are the ideal profiles for linear sublayer and log-law for a zero
pressure-gradient turbulent boundary on a flat plate. The variation of the location of (S) and (R) with Re, for the time-averaged 2D and time- and span-averaged 3D flows, is
shown in (c). (d) shows the variation of the length (Lb ) of the LSB with Re.

The bubble significantly affects the pressure distribution on the increase in Re. This confirms the association of the plateau with the
airfoil. Shown in Fig. 22 is the chordwise distribution of the time- LSB. Arena and Mueller,12 via experiments, showed the appearance
and span-averaged pressure coefficient (Cp ) for various Re. For ref- of a slight adverse pressure gradient as the turbulent boundary layer
erence, also shown are the x/c locations of separation and reattach- redevelops on the suction surface beyond the location of the reat-
ment points. The impact of the LSB on the pressure distribution on tachment. In line with their observation, Fig. 22 shows an adverse
the upper as well as lower surface of the airfoil is clearly observed. pressure gradient just downstream of the reattachment point. An
For the curves corresponding to Re = 20 000 and above, a plateau interesting effect of the formation of the LSB is the upstream move-
exists in the Cp distribution on the upper surface of the airfoil. It lies ment of the separation point on the upper surface and downstream
between the separation and reattachment points. The plateau does movement of the stagnation point on the lower surface of the air-
not exist for the Re = 19 000 flow, which is devoid of the LSB. It foil. This is observed from the close-up view of the time- and span-
is longest at the onset of the LSB and decreases in length with an averaged streamlines in Fig. 23. The downstream movement of the

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-16


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 22. Flow past an Eppler 61 airfoil at α = 10○ : chord-


wise distribution of the time- and span-averaged pressure
coefficient (Cp ) for various Re. The hollow circles and solid
squares show the x/c location of the separation and reat-
tachment points, respectively, on the suction surface of the
airfoil. For reference, the airfoil is also shown in solid thick
line.

stagnation point causes a slight increase in the pressure on the lower suggested two techniques to determine the shear layer frequency.
surface of the airfoil, as shown in Fig. 22. In the first technique, F SL corresponds to the maximum in the
broadband peak of the long-time-averaged velocity spectra. Since
the shear-layer instability occurs intermittently, long-time-averaged
VII. FREQUENCY OF THE PRIMARY AND SHEAR velocity spectra appear reduced in amplitude. In the second method,
LAYER INSTABILITIES F SL is measured from the period of shear-layer fluctuations using
The time histories of the cross flow component of velocity time traces. A statistically significant sample is chosen to estimate a
at certain spatial locations are utilized to estimate the frequencies reasonable value. A histogram between the percentage of total num-
associated with the primary vortex shedding and shear layer vor- ber of shear-layer cycles and frequency is utilized to select F SL . It
tices. We denote the nondimensional dominant frequencies asso- corresponds to that frequency which has the most prominent bin
ciated with the vortex shedding by F K and shear layer by F SL . The in the histogram. Both techniques result in very similar results. The
length and velocity scales used for the nondimensionalization are the “histogram” technique was used later by Singh and Mittal24 and
characteristic vertical dimension (c sin α) defined by Hoarau et al.9 Behara and Mittal28 to determine F SL from their computational data
and free-stream speed, respectively. A probe located at (x/c = 1.2, for flow past a cylinder. In the present work, the same technique is
y/c = −0.05, z = 0) is used to estimate F K . F SL is estimated via a probe utilized to estimate F SL .
located at (x/c = 0.8, y/c = 0.05, z = 0) for Re ≤ 19 000. Subsequent to Figures 24(a) and 24(b) show the variation of F K and F SL with
the formation of the LSB, the shear layer instability moves upstream; Re from the present study. Also shown is the best fit for these data
the time histories of the velocity at (x/c = 0.25, y/c = 0.05, z = 0) along with measurements from an earlier study by Huang and Lin.35
are used to estimate F SL for Re ≥ 20 000. It is well accepted that the Both F K and F SL exhibit a power law. However, the variation prior to
vortex shedding is a consequence of an absolute instability of the and after the formation of LSB is different. F K varies as Re0.22 before
wake while the shear layer instability is convective in nature.25,26 As the appearance of LSB. This result is in very good agreement with the
a result, F K is associated with a relatively sharp peak in the velocity one from Huang and Lin35 for a finite wing with a NACA 0012 sec-
spectra while the shear layer instability leads to broadband spec- tion at α = 9○ . After the LSB is formed, F K is found to vary as ∼Re1.03 .
tra. Prasad and Williamson22 studied the shear layer instability, via The rate of increase in F K , with Re, is significantly larger after the
laboratory experiments, in the flow past a circular cylinder. They formation of LSB than the corresponding rate of increase before the

FIG. 23. Flow past an Eppler 61 airfoil


at α = 10○ : time- and span-averaged
streamlines close to the leading edge for
(a) Re = 19 000 and (b) Re = 20 000.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-17


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 24. Flow past an Eppler 61 air-


foil at α = 10○ : variation of normalized
frequencies associated with (a) primary
shedding (F K ) and (b) shear layer (F SL )
with Re. The variation of the ratio of the
two frequencies is shown in (c). Also
shown in (a) is the F K for the NACA 0012
airfoil.35

formation of the LSB. This result is in line with the observation by by Yarusevych et al.6 in their experiments for the NACA0025
Yarusevych et al.6 for their experiments for a NACA 0025 airfoil for airfoil.
5.5 × 104 ≤ Re ≤ 2.10 × 105 . Yarusevych et al.6 found that F K does
not follow a power law in the large Re regime; rather, it exhibits a
VIII. WAKE FORMATION LENGTH
linear relationship with Re. However, for low Re, the power law vari-
ation holds and was found to be Re1.04 for α = 10○ . Interestingly, The wake formation length (Lf ) is determined from the stream-
this is quite close to the variation in the post reattachment regime wise location of the maximum value of u′ u′ component of Reynolds
for the present study. The variation of F SL , before and after the for- stress.44 Figure 25 shows the span-averaged u′ u′ for Re = 1300. Two
mation of LSB, is ∼Re1.02 and ∼Re0.80 , respectively. Unlike F K , F SL local maxima are identified: one in the upper-wake and the other in
undergoes a sudden increase at the formation of LSB Re at which the the lower-wake. The streamwise distance of each of these maxima,
LSB is formed. Williamson14 showed that the shear layer frequency from the leading edge of the airfoil, is the wake formation length: Lfu
is inversely proportional to the shear layer momentum thickness at and Lfl . The transverse distance, from the leading edge, is referred to
the flow separation (θSEP ). It is seen from Fig. 21(c) that the for- as Lyu and Lyl . The subscripts “u” and “l” refer to the local maxima
mation of the LSB, at Re ∼ 20 000, is accompanied with a sudden in the upper-wake and lower-wake, respectively.
upstream movement of the separation point. A relatively early flow u′ u′ estimated from the two-dimensional spanwise section
separation leads to a reduction in θSEP and a consequent increase (x–y) of the flow, for Re beyond the onset of three-dimensionality,
in F SL . On the other hand, the primary vortex shedding is a wake is expected to exhibit variation across the span. Figure 26 shows
instability, and therefore, F K does not experience a jump on the for- the spanwise variation of the local maxima of u′ u′ , estimated from
mation of LSB. For the flow past a cylinder, it is well known that a spanwise section of the flow, for the upper and lower wake at
the ratio of the shear layer to primary vortex shedding frequency two values of Re. The variation exhibits spanwise periodicity for
(F SL /F K ) exhibits a power-law variation with Re.22,23,28 The variation Re = 1600 while it is fairly irregular for Re = 2000. Lfu , Lfl , Lyu , and
of F SL /F K with Re for the Eppler airfoil at α = 10○ from the present Lyl (not shown here) exhibit a similar variation. The span-averaged
study is shown in Fig. 24(c). The variation is Re0.80 before the appear- length scale, for various Re, is shown in Fig. 27. Lfu and Lfl decrease
ance of LSB and Re−0.23 beyond the formation of LSB. Although with an increase in Re for up to Re ∼ 2000. The onset of shear
both F K and F SL increase with an increase in Re, the ratio decreases layer instability leads to a gradual increase in the length scales with
with Re beyond the formation of LSB. This was also observed an increase in Re. As expected, the upper wake is relatively more

FIG. 25. 3D flow past an Eppler 61 air-


foil at α = 10○ and Re = 1300: span-
averaged u′ u′ component of Reynolds
stress. The wake formation length, Lfu
and Lfl , are defined from the leading
edge of the airfoil for the upper and lower
wake.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-18


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 26. 3D flow past an Eppler 61 airfoil


at α = 10○ : maximum u′ u′ across the
span for the upper and lower wake for
(a) Re = 1600 and (b) Re = 2000.

FIG. 27. Flow past an Eppler 61 airfoil at α = 10○ : variation


of the wake formation length (Lfu , Lfl ) with Re.

affected. Therefore, the rise in Lfu is larger than that in Lfl . Lfl is primary vortex shedding time period. This is consistent with the
largely unaffected by the formation of LSB. observations from earlier studies for the flows with mode C insta-
bility and is confirmed from the spatiotemporal distribution of the
streamwise vorticity. The flow exhibits neither odd-RT nor even-
IX. CONCLUSIONS RT symmetry. Here, R denotes reflection about the wake axis and
The flow past an Eppler 61 airfoil at an angle of attack α = 10○ T denotes translation in time. The flow is very regular and periodic
has been investigated for 100 ≤ Re ≤ 87 000. Both 2D and 3D com- along the span for Re ≤ 1600. At Re = 1700, there are additional
putations have been carried out. A parallel implementation of the variations along the span although the flow retains symmetry about
stabilized finite element method has been deployed for the computa- midspan. At larger Re, the flow loses symmetry as well as spanwise
tions. The sequence of flow states, with an increase in Re, that leads periodicity.
to transition of the flow is identified. The flow is steady at low Re The separated shear layer rolls up into shear layer vortices.
and becomes unsteady beyond Re ∼ 600 via the primary instability The location of the instability of the separated shear layer moves
of the wake leading to vortex shedding. Three-dimensionality sets upstream with an increase in Re. At Re ∼ 20 000 and beyond, the
in at Re ∼ 1280.9 via mode C instability and hairpin vortices. The shear layer vortices interact with the flow close to the surface of
hairpin vortex structures appear in the near wake as a consequence the airfoil leading to its reattachment and formation of the Lami-
of the undulations in the primary vortices. The spanwise wavelength nar Separation Bubble (LSB). The reattachment of the flow leads to
of the instability varies between 0.194 and 0.251c for 1280.9 < Re an increased suction and a plateau of relatively large suction, in the
≤ 1850. The data are in good agreement with those from earlier region of LSB, on the upper surface of the airfoil. The airfoil experi-
studies for bluff body flows and other airfoils. The evolution of the ences a very significant increase in the lift and a decrease in the drag
hairpin vortices with Re is also studied in terms of their width and at the formation of LSB. The unsteady activity in the formation of
spacing along the span. They are uniformly distributed along the the bubble is evidenced by a relatively large Reynolds stress in the
span at the onset of three-dimensionality and interspersed rather region occupied by LSB. The length of the bubble is estimated via
randomly at relatively large Re. The spanwise variation of the lift tracking the separation and reattachment point of the time-averaged
coefficient, with respect to the span-averaged value, is utilized to flow. The formation of LSB is associated with a sudden upstream
study the spatiotemporal structure of the flow. Period doubling is movement of the separation point. The length of the LSB is largest at
observed for the flow in the regime 1300 ≤ Re ≤ 1850. The flow its onset. With an increase in Re, both the separation and reattach-
is periodic with a time period of 2T, where T corresponds to the ment points move upstream. However, since the separation point

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-19


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

is already close to the leading edge, the upstream movement of the by the particular airfoil and the angle of attack, this study provides a
reattachment point with an increase in Re is more significant. Thus, qualitative map of the rich behavior of the flow for a general airfoil,
the LSB shortens with an increase in Re. The sudden upstream move- with increasing Re.
ment of the separation point at the onset of formation of LSB causes
a downstream movement of the stagnation point on the lower sur-
face of the airfoil. This, in turn, affects the pressure on the lower ACKNOWLEDGMENTS
surface. Most of the computations were carried out at the High Perfor-
Although the flow is already three-dimensional in the flow mance Computing facility at the Computer Center, Indian Institute
regime when the LSB forms, the primary mechanism of the forma- of Technology Kanpur, India, and funded primarily by the Depart-
tion of LSB appears to be two-dimensional. The shear layer vortices ment of Science and Technology (DST), India. The authors are
are relatively stronger in the 2D computations compared to those grateful to Dr. Suresh Behara and C. S. Narain K. for their help with
in 3D. Consequently, the LSB forms at a relatively lower Re in the some of the calculations and the related discussions at the start of the
2D computations (Re ∼ 7000). Beyond Re ∼ 20 000, the variation of work.
the length of the bubble, with Re, from 2D and 3D computations is
in good agreement. The effect of the LSB on the force coefficients
APPENDIX A: MESH CONVERGENCE STUDY
is also quite similar for the 2D and 3D computations. Barring the
location of the jump, the 2D computations show a reasonable agree- 1. 2D mesh
ment with the variation of lift coefficient, with Re, to the predictions
a. Re = 1500, α = 10 ○
from 3D computations. The drag is relatively more sensitive; 2D
computations overpredict drag beyond the formation of LSB. The uncertainty in the results from the computations is esti-
The variation of the shear layer and primary wake frequency, mated using a procedure proposed by Celik et al.37 The method
with Re, is studied. Both follow the power law. However, the vari- utilizes computations on a set of grids with varying spatial resolution
ation before and after the formation of LSB is quite different. F K and Richardson extrapolation to arrive at an estimate of the numer-
varies as Re0.22 prior to and Re1.03 after the formation of LSB. The ical uncertainty in the fine-grid solution. The time-averaged aero-
corresponding variation of F SL in the two regimes is Re1.02 and Re0.80 . dynamic force coefficients are chosen as the key variables for deter-
1 2
Unlike F K , F SL suffers a jump at the formation of LSB. This is related mining the uncertainty. To this extent, three meshes (M2D , M2D , and
3
to the upstream movement of the separation point and a consequent M2D ) are used in the present study. The number of grid points on the
1 2
reduction in the momentum thickness of the boundary layer at the surface of the airfoil for mesh M2D and M2D are 200 while the mesh
3
separation point. The variation of the frequency ratio, F SL /F K , also M2D consists of 300 points on the surface of the airfoil. Table III sum-
exhibits a power law: Re0.80 and Re−0.23 pre- and postformation of marizes the details of the three meshes and the time-averaged force
LSB, respectively. coefficients obtained from them. The three sets of results are in very
Wake formation length (Lf ) is estimated by locating the local good agreement. Using these values and applying the extrapolation
maxima of u′ u′ . Lf has been estimated in experimental studies in procedure proposed by Celik et al.,37 the time-averaged lift and drag
the past by making measurements at a certain span location of the coefficients corrected for spatial discretization errors for Re = 1500
model. It is shown in this study that for a three-dimensional flow, are found out to be 1.0503 and 0.1657, respectively. The error in the
estimation of u′ u′ from a 2D section shows spanwise variation. time-averaged lift and drag coefficient, relative to the extrapolated
Span-averaged values of Lf are presented for various Re. Two val- value, for the three meshes is listed in Table III. The three meshes
ues of Lf are reported: Lfu and Lfl , for the upper- and lower-wake, are associated with a relatively low error and, therefore, provide ade-
2
respectively. quate spatial resolution. Hence, mesh M2D is sufficient to capture the
The LSB plays a very significant role in the performance of flow for Re ≤ 4000.
the airfoil at low Re. The present study investigates the flow on an
airfoil that has been specifically designed for low Reynolds num- b. Re = 46 000, α = 10 ○
ber flows. Furthermore, the study is restricted to a certain angle of Next, the convergence study is undertaken for Re = 46 000 flow
attack. The study brings out the evolution of the flow and the transi- to investigate the adequacy of the 2D mesh. Similar to the conver-
tion of the boundary layer to a turbulent state with an increase in Re. gence study carried out for Re = 1500 flow, we consider three finite
2 4 5
Although the transition to various states are expected to be affected element meshes, namely, M2D , M2D , and M2D . A close-up view of the

TABLE III. 2D flow past an Eppler 61 airfoil at Re = 1500: airfoil is placed at α = 10○ . Time-averaged lift and drag coefficients
for three different finite element meshes. The fluctuation in lift and drag coefficient is also listed. Also listed is the percentage
error in time-averaged lift and drag coefficient with respect to the extrapolated value.

Mesh Nodes Elements Cl Cd Clrms Cdrms error(Cl ) (%) error(Cd ) (%)


1
M2D 15 045 29 836 1.0549 0.1691 0.0364 0.0079 0.44 2.05
2
M2D 26 530 52 806 1.0540 0.1710 0.0360 0.0075 0.35 3.19
3
M2D 47 173 93 981 1.0522 0.1696 0.0362 0.0075 0.18 2.35

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-20


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 28. A close-up view of the finite element mesh.

TABLE IV. 2D flow past an Eppler 61 airfoil at Re = 46 000: time-averaged and rms values of lift and drag coefficients for
three different finite element meshes. Also listed is the percentage error in time-averaged lift and drag coefficient with respect
to the extrapolated value.

Mesh Nodes Elements Cl Cd Clrms Cdrms error(Cl ) (%) error(Cd ) (%)


2
M2D 26 530 52 806 1.4647 0.1132 0.1221 0.0308 5.55 7.09
4
M2D 46 168 92 031 1.5348 0.1051 0.1454 0.0379 0.73 0.57
5
M2D 89 898 179 075 1.5451 0.1057 0.1506 0.0379 0.06 0

4
mesh M2D , near the airfoil, is shown in Fig. 28. The structured mesh coefficients. The difference in the lift and drag coefficients for mesh
1 2
extends up to 0.85c from the surface of the airfoil. The thickness of M3D and M3D is 0.09% and 0.06%, respectively. The computed aero-
the first layer of elements lying on the surface of the airfoil is 0.0002c. dynamic force coefficients and their fluctuations are identical for
2 3
The time-averaged lift and drag coefficients for the three meshes mesh M3D and the more refined mesh M3D . In addition, the iso-
are shown in Table IV. Applying the extrapolation procedure, the surface of the instantaneous streamwise vorticity (ωx ), for the fully
lift and drag coefficient corrected for spatial discretization error is developed unsteady solution at a time instant when the lift coeffi-
found out to be 1.5461 and 0.1057, respectively. Of the three meshes, cient achieves its maximum value, obtained for all three meshes is
4 5
M2D and M2D are associated with a relatively low error and, there- shown in Figs. 29(a)–29(c). Each mesh accommodates about nine
fore, provide adequate spatial resolution for these range of Reynolds and a half cycles of streamwise vorticity (ωx ) in the wake of the
number. Therefore, for Re > 4000, mesh M2D 4
is good enough to airfoil. The wavelength (λz ) present in the flow computed using all
resolve the flow. For more details on the convergence study includ- three meshes is ∼0.21c. All three meshes provide adequate spatial
2
ing the effect of reducing the first element thickness, the interested resolution. Mesh M3D is, therefore, utilized for all computations for
reader is referred to our earlier article.10 Re ≤ 4000.

b. Re = 46 000, α = 10 ○
2. 3D mesh convergence study
Next, a mesh convergence study is undertaken for three-
a. Re = 1500, α = 10 ○ dimensional computations at Re = 46 000. At this Re, a laminar sepa-
The flow at Re = 1500 is associated with a three-dimensional ration bubble exists on the upper surface of the airfoil. Three meshes
4 5 6
instability with a fairly regular spanwise structure. The adequacy of are utilized: M3D , M3D , and M3D . The spanwise extent of the domain
4 5 6
the spatial resolution of the mesh along the span, used for com- is Lz = 2.0c. Meshes M3D , M3D , and M3D are created by stacking 200,
4
putations, is investigated. Hoarau et al.9 showed that the spanwise 250, and 300 sections of the more refined 2D mesh M2D uniformly
wavelength of the three-dimensional instabilities in the wake of
an airfoil scales with c sin α, where α is the angle of attack. With
TABLE V. 3D flow past an Eppler 61 airfoil for Re = 1500 and α = 10○ : time-averaged
this scaling, the spanwise wavelength of the three-dimensional flow lift and drag coefficients for three different finite element meshes. Nz is the number of
structures is expected to be ∼0.174c for α = 10○ . To include suffi- sections of the 2D mesh (M2D 2 ) that are uniformly stacked along the span, L = 2.0c.
z
cient number of wavelengths of the instability, the spanwise extent The approximate number of nodes and elements, in millions, in the 3D mesh is listed.
of the domain is chosen to be Lz = 2.0c for the present computa-
1 2 3
tions. Three meshes are chosen. Meshes M3D , M3D , and M3D are Mesh Nz Nodes Elements CL CD CLrms CDrms
generated by stacking 100, 250, and 300 sections of the 2D mesh
2 1
M2D (see Table III) uniformly along the span. While the 2D mesh M3D 100 2.65M 5.22M 1.0188 0.1679 0.0270 0.0057
2
consists of 3-noded triangular elements, the elements in the 3D M3D 250 6.63M 13.14M 1.0179 0.1678 0.0265 0.0056
3
mesh are 6-noded right prisms with a triangular base. Table V M3D 300 7.95M 15.78M 1.0179 0.1678 0.0265 0.0056
shows the three finite element meshes and their time-averaged force

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-21


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 29. Re = 1500 and α = 10○ : isosur-


face of streamwise vorticity (ωx = ±0.1)
for the fully developed unsteady solution
at a time instant when the lift coefficient
achieves its maximum value for the three
1 2 3
meshes: (a) M3D , (b) M3D , and (c) M3D .

along the span. Table VI summarizes the details of the three meshes various span considered. It can be seen that the results do not change
and the time-averaged force coefficients obtained from them. The significantly beyond Lz = 0.6c. The wavelength present in the flow for
results obtained are in good agreement. The difference in the lift and various spanwise extent of the domain is also listed in Table VII. It
4 5
drag coefficients between mesh M3D and M3D is 0.20% and 0.29%. does not change much for Lz = 0.8c and beyond. Lz = 2.0c is used for
5 6
Similarly, the corresponding value between mesh M3D and M3D is all computations in this work.
0.01% and 0.29%. The difference for all three meshes is quite small.
Also the difference in the fluctuations of lift and drag coefficients 2. Re = 46 000, α = 10 ○
The effect of the span on the flow, at Re = 46 000 and α = 10○ , is
is fairly small for the meshes considered. A convergence study was
also carried out for Re = 87 000. The difference in the time-averaged
5 6 studied by performing simulations for three values of span: 0.5c, 1c,
lift and drag coefficients between mesh M3D and M3D is 0.34% and
5 6 and 2c. The mesh is uniform along the span. The number of elements
0.35%. We conclude that both meshes M3D and M3D provide ade-
along the span is such that the spatial resolution is identical for the
quate spatial resolution for the range of Re computed in this work.
three cases. The time-averaged lift and drag coefficients and their
All the computations in this paper, for Re > 4000, are carried out with
6 fluctuations obtained for three spans are listed in Table VIII. The
mesh M3D . A time step of Δt = 0.002 is utilized for time integration
values obtained for span Lz = 1.0c and 2.0c are in good agreement.
of the governing equations.
The chordwise distribution of the time-averaged pressure coefficient
(Cp ), at two different locations along the span, is shown in Figs. 30(a)
APPENDIX B: EFFECT OF SPAN
1. Re = 1500, α = 10 ○ TABLE VII. 3D flow past an Eppler 61 airfoil at Re = 1500 and α = 10○ : time-averaged
lift and drag coefficients and their rms values obtained for various span lengths. Also
Figures 29(a)–29(c) show that the fully developed unsteady listed is the spanwise wavelength present in the wake of the airfoil.
flow at Re = 1500 is associated with 9.5 cycles of streamwise vor-
ticity (ωx ) of the three-dimensional instability along the span. With Lz /c CL CD CLrms CDrms λz /c
Lz = 2.0c, the corresponding wavelength of the spanwise instability is
λz = 2/9.5c (=0.21c), approximately. To investigate the effect of span 0.25 1.0198 0.1680 0.0273 0.0058 0.25
of the computational domain on λz , computations are carried out 0.40 1.0181 0.1678 0.0264 0.0056 0.19
2
for various values of Lz . The 2D section of the mesh, M2D , as well as 0.50 1.0199 0.1680 0.0273 0.0058 0.25
the spanwise resolution is identical in all the cases. For each span, a 0.60 1.0183 0.1679 0.0264 0.0056 0.19
fully developed 2D solution is stacked along the span and 3D com- 0.80 1.0182 0.1678 0.0264 0.0056 0.20
putations are carried out sufficiently to allow the solution to reach 1.00 1.0178 0.1678 0.0265 0.0056 0.22
a fully developed unsteady state. This flow is utilized to estimate the 1.20 1.0183 0.1679 0.0264 0.0056 0.20
wavelength of the three-dimensional instability. Table VII shows the 2.00 1.0179 0.1678 0.0265 0.0056 0.21
time-averaged lift and drag coefficients along with the rms values for 3.00 1.0180 0.1678 0.0267 0.0056 0.22

TABLE VI. 3D flow past an Eppler 61 airfoil for Re = 46 000 and α = 10○ : time-
averaged lift and drag coefficients for three different finite element meshes. Nz is the TABLE VIII. Re = 46 000 and α = 10○ : time-averaged lift and drag coefficients and
number of sections of the 2D mesh (M2D 4 ) that are uniformly stacked along the span.
their rms values obtained for various span lengths. The 2D sectional mesh used is
The approximate number of nodes and elements, in millions, in the 3D mesh is listed. 4 , and the spanwise resolution of the mesh is identical in all the cases.
M2D

Mesh N z Nodes Elements CL CD CLrms CDrms Lz /c CL CD CLrms CDrms


4
M3D 200 9.23M 18.31M 1.5343 0.0688 0.0036 0.000 84 0.5 1.5344 0.0686 0.0069 0.001 41
5
M3D 250 11.54M 22.91M 1.5311 0.0686 0.0037 0.000 82 1.0 1.5307 0.0684 0.0045 0.000 90
6
M3D 300 13.85M 27.51M 1.5309 0.0684 0.0035 0.000 81 2.0 1.5309 0.0684 0.0035 0.000 81

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-22


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 30. Flow past an Eppler 61 airfoil at Re = 46 000 and α = 10○ : chordwise variation of the time-averaged coefficient of pressure Cp on the surface of airfoil at two different
span locations (a) z/Lz = 0.25 and (b) z/Lz = 0.75. Also shown in (c) is the location of separation and reattachment points for the time-averaged flow across the span for
various lengths of the span.

3
and 30(b) for various span lengths. It is seen that the Cp distribu- D. E. Gault, “Boundary-layer and stalling characteristics of NACA63-009 airfoil
tion is virtually the same for all span lengths. The variation of the section,” NACA TN-1894, 1949.
4
separation and reattachment points along the span of the airfoil for M. Gaster, The Structure and Behavior of Laminar Separation Bubbles, Aeronau-
different span lengths is shown in Fig. 30(c). It can be observed that tical Research Council, R&M No. 3595 (London: Her Majesty’s Stationery Office,
1969).
there is no significant spanwise variation in the location of separa- 5
S. Yarusevych and P. E. Sullivan, “Coherent structures in an airfoil boundary
tion and reattachment point on the airfoil surface. Zhang and Sam- layer and wake at low Reynolds numbers,” Phys. Fluids 18, 044101 (2006).
taney38 reached a similar conclusion in their work. They investigated 6
S. Yarusevych, P. E. Sullivan, and J. G. Kawall, “On vortex shedding from an
the effect of the spanwise extent of the computational domain on the airfoil in low-Reynolds-number flows,” J. Fluid Mech. 632, 245 (2009).
flow past a NACA 0012 airfoil for Re = 50 000 placed at 5○ angle of 7
B. H. Carmichael, “Low-Reynolds number airfoil survey,” NASA CR-165803,
attack. Various values of span length in the range 0.1–0.8c, where c is 1981.
8
the chord length of the airfoil, were considered. It was found that the R. Bourguet, M. Braza, A. Sevrain, and A. Bouhadji, “Capturing transition fea-
time-averaged flow quantities are virtually independent of the aspect tures around a wing by reduced-order modeling based on compressible Navier-
Stokes equations,” Phys. Fluids 21(9), 094104 (2009).
ratio, for the range considered. The mean separation and reattach- 9
Y. Hoarau, M. Braza, Y. Ventikos, D. Faghani, and G. Tzabiras, “Organized
ment locations are also independent of the aspect ratio, although the modes and the three-dimensional transition to turbulence in the incompressible
mean velocity field within the separation bubble is slightly affected. flow around a NACA0012 wing,” J. Fluid Mech. 496, 63 (2003).
The effect of aspect ratio is more notable on the turbulent statistics. 10
S. B. Savaliya, S. P. Kumar, and S. Mittal, “Laminar separation bubble on an
In the present study, to capture the flow dynamics for the Reynolds Eppler 61 airfoil,” Int. J. Numer. Methods Fluids 64(6), 627 (2010).
11
number regime considered, simulations are carried out with L. E. Jones, R. D. Sandberg, and D. Sandham, “Direct numerical simulations of
Lz = 2.0c. forced and unforced separation bubbles on an airfoil at incidence,” J. Fluid Mech.
602, 175 (2008).
12
A. V. Arena and T. J. Mueller, “Laminar separation, transition, and turbulent
reattachment near the leading edge of airfoils,” AIAA J. 18(7), 747 (1980).
REFERENCES 13
M. Alam and D. Sandham, “Direct numerical simulations of short lami-
1 nar separation bubbles with turbulent reattachment,” J. Fluid Mech. 410, 1
S. V. Shkarayev, P. G. Ifju, J. C. Kellogg, and T. J. Mueller, Introduction
to the Design of Fixed-Wing Micro Air Vehicles Including Three Case Stud- (2000).
14
ies, AIAA Education Series (AIAA, 2007), eISBN: 978-1-60086-210-6, URL: C. H. K. Williamson, “Three-dimensional wake transition,” J. Fluid Mech. 328,
http://arc.aiaa.org/doi/book/10.2514/4.862106. 345 (1996).
2 15
J. H. McMasters and M. L. Henderson, “Low speed single element airfoil synthe- H. Persillon and M. Braza, “Physical analysis of the transition to turbulence in
sis,” in NASA conference publication 2085, Science and technology of low speed and the wake of a circular cylinder by three-dimensional Navier-Stokes simulation,”
motorless flight, Part 1 (1979), p. 1–31. J. Fluid Mech. 365, 23 (1998).

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-23


Published under license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

16 33
M. Braza, D. Faghani, and H. Persillon, “Successive stages and the role of natural F. Nicoud and F. Ducros, “Subgrid-scale stress modelling based on the square
vortex dislocations in three-dimensional wake transition,” J. Fluid Mech. 439, 1 of the velocity gradient tensor,” Flow, Turbul. Combust. 62(3), 183–200 (1999).
(2001). 34
S. Behara and S. Mittal, “Parallel finite element computation of incompressible
17
S. Behara and S. Mittal, “Wake transition in flow past a circular cylinder,” Phys. flows,” Parallel Comput. 35, 195 (2009).
Fluids 22, 114104 (2010). 35
R. F. Huang and C. L. Lin, “Vortex shedding and shear-layer instability of wing
18
H. Q. Zhang, U. Fey, B. R. Noack, M. Konig, and H. Eckelmann, “On the at low-Reynolds numbers,” AIAA J. 33(8), 1398 (1995).
transition of the cylinder wake,” Phys. Fluids 7, 779 (1995). 36
J. McArthur, “Aerodynamics of wings at low Reynolds numbers: Boundary layer
19
B. S. Carmo, S. J. Sherwin, P. W. Bearman, and R. H. J. Willden, “Wake transi- separation and reattachment,” Ph.D. thesis, University of Southern California,
tion in the flow around two circular cylinders in staggered arrangements,” J. Fluid USA, 2008.
Mech. 597, 1–29 (2008). 37
I. B. Celik, U. Ghia, P. J. Roache, C. J. Freitas, H. Coleman, and P. E. Raad,
20
O. Marxen and U. Rist, “Mean flow deformation in a laminar separation bubble: “Procedure for estimation and reporting of uncertainty due to discretization in
Separation and stability characteristics,” J. Fluid Mech. 660, 37 (2010). CFD applications,” J. Fluids Eng. 130(7), 078001 (2008).
21 38
O. Marxen and D. S. Henningson, “The effect of small-amplitude convective W. Zhang and R. Samtaney, “Assessment of spanwise domain size effect on the
disturbances on the size and bursting of a laminar separation bubble,” J. Fluid transitional flow past an airfoil,” Comput. Fluids 124, 39–53 (2016).
Mech. 671, 1 (2011). 39
G. Chopra and S. Mittal, “The intermittent nature of the laminar separation
22
A. Prasad and C. H. K. Williamson, “The instability of the shear layer separating bubble on a cylinder in uniform flow,” Comput. Fluids 142, 118–127 (2016).
from a bluff body,” J. Fluid Mech. 333, 375 (1997). 40
J. Robichaux, S. Balachandar, and S. P. Vanka, “Three-dimensional Floquet
23
M. C. Thompson and K. Hourigan, “The shear-layer instability of a circular instability of the wake of square cylinder,” Phys. Fluids 11, 560 (1999).
cylinder wake,” Phys. Fluids 17, 021702 (2005). 41
J. C. R. Hunt, A. Wray, and P. Moin, “Eddies, stream, and convergence zones in
24
S. P. Singh and S. Mittal, “Flow past a cylinder: Shear layer instability and drag turbulent flows,” Center for Turbulence Research Report No. S88, 1988.
crisis,” Int. J. Numer. Methods Fluids 47, 75 (2005). 42
I. Yildirim, C. C. M. Rindt, and A. A. van Steenhoven, “Mode C flow transition
25
S. Mittal, J. J. Kottaram, and B. Kumar, “Onset of shear layer instability in flow behind a circular cylinder with a near-wake wire disturbance,” J. Fluid Mech. 727,
past a cylinder,” Phys. Fluids 20, 054102 (2008). 30–55 (2013).
26 43
B. Kumar, J. J. Kottaram, A. Singh, and S. Mittal, “Global stability of flow past a J. Deng, L. Sun, and X. Shao, “Floquet stability analysis in the wake of a
cylinder with centerline symmetry,” J. Fluid Mech. 632, 273–300 (2009). NACA0015 airfoil at post-stall angles of attack,” Phys. Fluids 29, 094104 (2017).
27 44
M. Braza, P. Chassaing, and H. Ha Minh, “Prediction of large-scale transition Md. Mahbub Alam, Y. Zhou, H. X. Yang, H. Guo, and J. Mi, “The ultra-low
features in the wake of circular cylinder,” Phys. Fluids A 2, 1461–1471 (1990). Reynolds number airfoil wake,” Exp. Fluids 48, 81–103 (2010).
28 45
S. Behara and S. Mittal, “Transition of the boundary layer on a circular cylinder T. J. Mueller and T. F. Burns, “Experimental studies of the Eppler 61 airfoil at
in the presence of a trip,” J. Fluids Struct. 27, 702–715 (2011). low Reynolds numbers,” AIAA J. 82, 0345 (1982).
29 46
T. E. Tezduyar, S. Mittal, S. E. Ray, and R. Shih, “Incompressible flow computa- C. H. K. Williamson, “Vortex dynamics in the cylinder wake,” Annu. Rev. Fluid
tions with stabilized bilinear and linear equal order interpolation velocity pressure Mech. 28, 477 (1996).
elements,” Comput. Methods Appl. Mech. Eng. 95, 221 (1992). 47
W. Zhang and R. Samtaney, “BiGlobal linear stability analysis on low-Re flow
30
Y. Saad and M. Schultz, “GMRES: A generalized minimal residual algorithm past an airfoil at high angle of attack,” Phys. Fluids 28, 044105 (2016).
for solving nonsymmetric linear systems,” SIAM J. Sci. Stat. Comput. 7, 856 48
F. Nicoud, H. B. Toda, O. Cabrit, S. Bose, and J. Lee, “Using singular values to
(1986). build a subgrid-scale model for large eddy simulations,” Phys. Fluids 23(8), 085106
31
H. Johari and K. Stein, “Near wake of an impulsively started disk,” Phys. Fluids (2011).
14(10), 3459–3474 (2002). 49
R. Eppler and D. M. Somers, “Low speed airfoil design and analysis,” in
32
J. Akin, T. Tezduyar, M. Ungor, and S. Mittal, “Stabilization parameters and Advanced Airfoil Technology Research, Vol. I, NASA Conference Publication 2045,
Smagorinsky turbulence model,” J. Appl. Mech. 70(1), 2–9 (2003). Part 1 (NASA, 1978), pp. 73–99.

Phys. Fluids 31, 114102 (2019); doi: 10.1063/1.5113823 31, 114102-24


Published under license by AIP Publishing

You might also like