You are on page 1of 16

AIAA JOURNAL

Investigation of the Low-Frequency Oscillations


in the Flowfield About an Airfoil

Eltayeb M. Eljack∗
University of Khartoum, Khartoum 11111, Sudan
and
Julio Soria†
Monash University, Melbourne, Victoria 3800, Australia
https://doi.org/10.2514/1.J058905
The present investigation shows that a triad of three vortices is behind the quasi-periodic self-sustained low-
frequency flow oscillation phenomenon. Large-eddy simulations are carried out at Rec  5 × 104 , M∞  0.4, and
sixteen angles of attack at near-stall conditions. It is shown that the best description of the underlying mechanism is
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

that of a button whirligig. A globally oscillating flow whirls around the airfoil in the clockwise direction and stores
energy in the triad of vortices. When the oscillating flow loses momentum and comes to an equilibrium, the energy
stored in the triad of vortices reaches a threshold. Thus, the triad of vortices reverses the direction of rotation of the
oscillating flow, gives it a pulse of energy to whirl in the anticlockwise direction, and starts a new disequilibrium. When
the oscillating flow rotates in the clockwise direction (in the streamwise direction on the suction surface of the airfoil),
it adds momentum to the boundary layer and helps it to remain attached, and vice versa. Consequently, the
instantaneous flowfield switches between an attached phase and a separated phase in a periodic manner with some
disturbed cycles.

Nomenclature T = temperature
A = correlation matrix T∞ = freestream temperature
hai = spanwise ensemble average of a ui = velocity component in the xi direction
the mth proper orthogonal decomposition U∞ , V ∞ , W ∞ = freestream velocities
am =
coefficient α = angle of attack
a~ = Favre-filtered (a) γ = specific heat ratio
a = filtered (a) Δψ  = time-averaged surplus of ψ
a00 = fluctuation of a Δψ − = time-averaged deficit of ψ
a = phase average of a Δ = filter size
a = time average of a δij = Kronecker delta
CD = drag coefficient Λm = the mth POD eigenvalue
Cf = skin-friction coefficient μ = dynamic viscosity
CL = lift coefficient ν = kinematic viscosity
Cmts = mixed-time-scale model parameter νt = turbulent eddy viscosity
CP = pressure coefficient ξ, η, ζ = curvilinear coordinates
Cτ = time-scale parameter ξ‵ij = transformation metrics tensor
c = airfoil chord ρ = density
E = total energy per unit mass ρ∞ = freestream density
f = nondimensional frequency σ ij = viscous stress tensor
ksgs = velocity scale τij = subgrid-scale stress tensor
L = length of the computational domain τs = time scale
M∞ = freestream Mach number Φ = phase angle
N = number of grid points φm = the mth proper orthogonal decomposition
Pr = Prandtl number eigenfunction
P∞ = freestream pressure Ψ = orthonormal spatial proper orthogonal decom-
p = pressure position modes
q = heat flux ψ^ = high-lift average of ψ
Rec = Reynolds number; U∞ c∕ν ψ̌ = low-lift average of ψ
St = Strouhal number; f  sinα ψ = mean-lift average of ψ
Sij = strain rate tensor ωz = spanwise vorticity

Received 22 August 2019; revision received 20 March 2020; accepted for


publication 26 May 2020; published online 7 September 2020. Copyright © I. Introduction
2020 by the American Institute of Aeronautics and Astronautics, Inc. All
rights reserved. All requests for copying and permission to reprint should be
submitted to CCC at www.copyright.com; employ the eISSN 1533-385X to
T HE laminar separation bubble (LSB), its stability, and its asso-
ciated low-frequency flow oscillation (LFO) have been in the
center of aerodynamics research in the past 30 years. This is because
initiate your request. See also AIAA Rights and Permissions www.aiaa.org/ of the steady increase in applications that operate at low Reynolds
randp.
*Assistant Professor, Mechanical Engineering Department, P.O. Box 321; numbers due to high-altitude flight, where the kinematic viscosity is
emeljack@uofk.edu. relatively higher, and/or having small geometrical dimensions. The

Professor, Department of Mechanical and Aerospace Engineering Labo- investigated configurations are twofold: an LSB on the suction sur-
ratory for Turbulence Research in Aerospace and Combustion, Clayton face of two-dimensional airfoils and on flat plates. The LSB is formed
Campus. Associate Fellow AIAA. naturally in the former type of setup due to the curvature of the
Article in Advance / 1
2 Article in Advance / ELJACK AND SORIA

airfoils, whereas in the latter, the LSB is induced by an imposed experiments showed the growth and bursting of the leading-edge
adverse pressure gradient (APG) [1]. Most of the studies are com- separation bubble and its role in the flow oscillation. Broeren and
pressible; however, several incompressible investigations were car- Bragg [9] investigated the LFO and the intensity of flow oscillation in
ried out recently. The overwhelming majority of previous studies are near-stall conditions. The authors tested 12 airfoils having different
around clean airfoils; however, considerable research work investi- stalling characteristics over a range of angles of attack at a Reynolds
gated the LSB and the LFO around iced airfoils. number of 3 × 105 . They concluded that there is a distinct relationship
The first observations and descriptions of the LSB were reported by between stall type and low-frequency/large-scale unsteady flow. The
Melvill Jones [2]. The author observed a “violent fluctuations” in the authors reported that a combination of thin-airfoil and trailing-edge
lift and the drag at near-airfoil stall conditions. The Strouhal number stall types results in an LFO of magnitudes that are nearly double
of oscillations was estimated later from his data to be in the order of that for pure thin-airfoil stall type. Broeren and Bragg [10] used
magnitude of 10−3 . Melvill Jones also classified the stall of airfoils the airfoil used earlier by Bragg et al. [8] at a chord Reynolds number
into trailing-edge, thin-airfoil, and leading-edge stalls. The author of 3 × 105 and the angle of attack of 15 deg. The authors used
reported that the first two stall types exhibited an LFO. The author conditionally averaged laser Doppler velocimetry to resolve a fully
also observed that for the trailing-edge stall airfoils, the large fluctua- separated boundary layer. The boundary layer is fully separated due to
tions took place at angles of attack a few degrees poststall; whereas it the merging of the leading-edge and the trailing-edge separations
occurred at the inception of stall for the thin-airfoil stall airfoils. when they grow in time. Their data showed the periodic switching
The work of Melvill Jones was continued by Zaman et al. [3], who between stalling and nonstalling behaviors, which involved a leading-
conducted wind–tunnel measurements at NASA Langley Research edge separation bubble and how the bubble interacted with the trail-
Center and reported an unusually low-frequency oscillations in
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

ing-edge separation. Broeren and Bragg [11] studied five different


the flowfield. They measured the lift, the drag, and the wake velocity airfoil configurations (NACA-2414, NACA-64A010, LRN-1007,
under acoustic excitation at chord Reynolds numbers ranging from E374, and Ultra-Sport) by measuring the wake velocity across the
4 × 104 to 1.4 × 105 . The authors observed a sharp spike in the spanwise direction, as well as using minitufts for flow visualization.
velocity spectra at a considerably low frequency: lower than the They found that all the stall types are dependent on the type of the
bluff-body shedding frequency. They noted that the low frequency airfoil. They concluded that the LFO phenomenon always occurs in
varies with the freestream velocity, and the Strouhal number was in the airfoils that exhibit a thin-airfoil stall or the combination of both
the order of 0.02. The work of Zaman et al. [3] was continued at thin-airfoil and trailing-edge stalls.
NASA Lewis Research Center by conducting a series of experiments Tanaka [12] investigated the flowfield around a NACA-0012
on the characteristics, stability, and control of the LSB. Zaman and airfoil near stall conditions. The author performed experimental
McKinzie [4] and Zaman et al. [5] investigated the LFO phenomenon measurements by means of particle image velocimetry. He was
experimentally for the flowfield around two-dimensional airfoil aiming to understand the self-sustained LFO. The flow Reynolds
model at a Reynolds number in the range of 1.5 × 104 –3 × 105 . number was 1.3 × 105 , and the measurements were carried out at
The frequency of oscillation is considered to be low if the Strouhal various angles of attack near stall conditions. In the range of inves-
number is less than 0.2. The authors reported that the phenomenon tigated angles of attack, he found that the frequency of the LFO
could not be reproduced initially in a relatively cleaner wind tunnel increased when the angle of attack was increased. The flow oscil-
at NASA Lewis Research Center. However, the phenomenon was lation had a maximum amplitude at the angle of attack of 11.5 deg.
produced by either raising the turbulence level of the freestream One of the major observations of his work that he identified was that
or exciting the flow by using acoustic waves. The authors studied large vortices play an essential role in the reattachment of the flow.
the fluctuation of the lift coefficient where it has a quasi-periodic However, his observations were overlooked in the literature despite
trend between stalled and nonstalled conditions. In the numerical their importance. The author observed that when the flow is mas-
study, they used a thin layer approximation to solve two-dimensional sively separated, a large vortex that bends the shear layer toward the
Navier–Stokes equations with a Baldwin–Lomax turbulence model. airfoil surface is formed near the leading edge. This vortex introduces
The authors studied the sensitivity of the lift coefficient to the grid the freestream into the separated flow region. He concluded that
size and verified it with experimental data. They concluded that the this vortex might be a part of the mechanism that forces the flow
LFO phenomenon occurs when there is a trailing-edge stall or a thin- to reattach and suggested further investigations on the nature of this
airfoil stall. Acoustic excitation at an appropriate frequency can not vortex. Rinoie and Takemura [13] conducted an experiment very
only diminish the laminar separation but also entirely remove the similar to that carried out by Tanaka [12]. They observed that the flow
LSB. was oscillating between a small bubble formed near the leading
A leading-edge separation bubble induced by a simulated ice was edge of about a 10% of the chord length and a massively separated
investigated experimentally by Bragg et al. [6]. A two-dimensional flow. However, they did not look into the large vortices observed by
NACA-0012 airfoil equipped with an interchangeable leading edge Tanaka [12].
was used in the experiment. The leading-edge portion was replaced by Alam and Sandham [14] investigated an LSB on a flat plate induced
an airfoil with the ice contour attached to it to simulate the iced leading by the action of a suction profile applied as the upper boundary
edge. The authors reported that they measured bubbles with a length condition. They showed that the separation bubble is considered
longer than 35% of the chord. The LFO was observed at a Strouhal absolutely unstable if the separation bubble sustains a maximum
number of about 0.0185, which is comparable to the Strouhal number reverse velocity percentage in the range of 15–20% of the local
of 0.02 observed in previous research. Bragg et al. [7] used a con- freestream velocity. The three-dimensional bubbles with turbulent
figuration similar to that employed by Zaman et al. [5] and measured reattachment have maximum reverse flow of less than 8%, and it is
the frequencies of the LFO with hot-wire measurements in the wake of concluded that, for these bubbles, the primary instability is convective
the airfoil. The measurements were conducted at Reynolds numbers in nature. Sandham [15] modeled the laminar–turbulent transition
of up to 1.3 × 106 . The LFO took place at a Strouhal number of about by an absolute instability. The results are qualitatively comparable to
0.02 and increased slightly with the Reynolds number. Bragg et al. [8] existing experimental data; however, the method needs further devel-
studied an experiment of the flowfield around an LRN(1)-1007 airfoil opment. The author used a time-accurate viscous–inviscid interaction
at a Reynolds number ranging from 3 × 105 to 1.25 × 106 and method for the coupled potential flow and integral boundary-layer
angles of attack in the range of 0–28 deg. The LFO phenomenon equations to study several airfoil models near stall conditions. He
was captured using hot-wire spectra measurements in the airfoil modeled the laminar–turbulent transition by an absolute instability
wake where the LFO phenomenon is observed in the range of angles that sustains the transition of the laminar bubble without using
of attack of 14.4 deg ≤ α ≤ 16.6 deg. The authors reported that upstream perturbations. The Strouhal number of the LFO was pointed
the Strouhal number increased slightly with the Reynolds number out to be dependent on the shape of the airfoil. The author reported
and significantly with the angle of attack. The Strouhal number of that the LFO phenomenon could be captured by a simple model of
the oscillation is an order of magnitude lower than that of bluff- boundary-layer transition and turbulence because the bubble bursting
body shedding. The surface oil flow and laser sheet visualization occurs as a consequence of potential flow/boundary-layer interactions.
Article in Advance / ELJACK AND SORIA 3

Jones et al. [16] performed a linear stability analysis of the flow II. Mathematical Modeling, Computational Setup,
around a NACA-0012 airfoil at the angle of attack of 5 deg. The authors and Validation
observed no evidence of local absolute instability. They confirmed this A. Mathematical Modeling
by performing two-dimensional linear stability analysis via forced
In the present study, the fluid flow is governed by the viscous-
Navier–Stokes simulations. Forced Navier–Stokes simulations did,
compressible Navier–Stokes equations. The nondimensional analysis
however, determine that the time-averaged flowfield is unstable
of these equations is achieved using the following nondimensional
because of an acoustic-feedback instability, in which instability waves
variables:
convecting over the trailing edge of the airfoil generate acoustic waves
that propagate upstream to some location of receptivity and generate uj xj
ρ T μ
further instability waves within the boundary layer. Almutairi et al. [17] uj   ; ρ ; T  ; xj  ; μ ;
used large-eddy simulation (LES) to study a low-Reynolds-number ur ρr Tr c μr
flow about a NACA-0012 airfoil at Rec  5 × 104 and the angle of P t ur
P   2 ; t
attack of 5 deg. The primary objective was to validate their LES code ρr ur c
using the direct numerical simulation data of Jones et al. [18].
Almutairi and Alqadi [19] investigated the phenomenon of natural Here, ρ, T, and μ are the fluid density, temperature, and dynamic
LFO using the LES code developed by Almutairi et al. [17]. They viscosity, respectively. P is the flow pressure, t is time, and c represents
studied the flowfield around a NACA-0012 airfoil at a Reynolds the airfoil chord length. The subscript r denotes the reference variables,
number of 5 × 104 , a Mach number of 0.4, and a range of angles of and the symbol  indicates the dimensional variables.
attack: α  9.25 deg, 9.29 deg, 9.4 deg, and 9.6 deg. The authors
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

The nondimensional Favre-filtered compressible Navier–Stokes


observed a self-sustained natural LFO in all of the investigated angles equations in curvilinear coordinates are given by
of attack. Almutairi et al. [20] performed an LES of the flow around
a NACA-0012 at a Reynolds number of 1.3 × 105 and the angle ∂ρ ∂ ‵
 ξ ρu~   0 (1)
of attack of 11.5 deg. They described the role of transition and how ∂t ∂ξj ij j
the large three-dimensional scales break down to small scales, which
are followed by a turbulent region. They found that the dominant ∂ ∂ ‵
ρu~ i   ξ ρu~ u~  pδij − σ~ ij 
flow mode is a Strouhal number of 0.00826, which is in very good ∂t ∂ξj ij i j
agreement with that of the experiment of Rinoie and Takemura
∂ ∂ ‵
[13]. However, their LES was carried out at a Mach number of 0.4,  − ξ‵ij τij   ξ σ − σ~ ij  (2)
whereas the experiment was carried in an incompressible conditions. ∂ξj ∂ξj ij ij
|{z} |{z}
Dynamic mode decomposition analysis identified another dominant 1 2
frequency that features the trailing-edge shedding. The authors con-
cluded that the LFO is linked with the trailing-edge vortex shedding. ∂ ~  ∂ ξ‵ij ρ E~ pu~ j  q~ j − σ~ ij u~ i 
When the flow is separated, the strong vortex shedding at the trailing ρ E
∂t ∂ξj
edge generates acoustic waves that travel upstream, excite the sepa-  
rated shear layer, force it to undergo transition to turbulence, and ∂ 1
− ξ‵ij Qj  Gj − Dj  (3)
reattach the flow. When the flow is attached, the vortex shedding at ∂ξj 2
|
{z}
the trailing edge dies down, and the feedback is cut off. Consequently, 3
the flow separates again and the process repeats in a periodic manner.
Eljack et al. [21] took up the work of Almutairi et al. [20] and used a where u~ i is the Favre-filtered velocity component in the xi direction,
combination of trailing-edge shedding frequencies to eliminate the p is the filtered pressure, ρ is the filtered density, T~ is the Favre-
low-frequency flow oscillation and improve the airfoil performance. filtered temperature, and E~ is the Favre-filtered total energy per unit
Eljack [22] carried out a careful grid sensitivity study that is based mass. The stress tensor is given by
on the characteristics of the bubble rather than being based on the
 
static aerodynamic coefficients. He found that the characteristics of 2μ~ 2μ~ 1 ∂ui ∂ξj ∂uj ∂ξi
the bubble, the angle of attack at which the bubble starts to become σ ij  Sij − δij Skk ; Sij   (4)
Rec 3Rec 2 ∂ξj ∂xi ∂ξi ∂xj
unstable, and the angle of attack at which the LSB becomes an
open bubble are highly sensitive to the grid distribution. The author The heat flux vector is given by
performed LESs over sixteen angles of attack at near-stall conditions
and compared the aerodynamic coefficients to recent and previous μ~ ∂T~ ∂ξj
experimental data. The author showed that, qualitatively, at relatively q~ j  (5)
low angles of attack, the LSB is present on the upper surface of γ − 1Rec Pr M2∞ ∂ξj ∂xi
an airfoil and remains intact. At moderate angles of attack, the LSB
abruptly and intermittently bursts, which causes an oscillation in where γ  1.4 is the specific heat ratio, Rec  5 × 104 is the chord
the flowfield, and consequently the aerodynamic forces. At higher Reynolds number, Pr  0.72 is the Prandtl number, and M∞  0.4 is
angles of attack, the flowfield exhibits a quasi-periodic switching the freestream Mach number. The dynamic viscosity is calculated using
between attached phase and a separated phase, which results in a the dimensionless form of Sutherland’s law; and the relation between
global low-frequency flow oscillation. As the airfoil approaches the temperature, density, and pressure is given by the ideal gas law:
full stall angle, the flow remains separated with intermittent and 3
abrupt random reattachments. The airfoil eventually undergoes a 1C p
full stall as the angle of attack is further increased. Recently, Elawad μ~  T~ 2 ; C  0.3686; T~  γM2∞ (6)
T~  C ρ
and Eljack [23] carried out LESs at a Reynolds number of 9 × 104 , a
Mach number of 0.4, and at near-stall conditions. The oscillation The freestream pressure P∞ estimated using the preceding relation is
behavior of the aerodynamic coefficients reported by the authors is equivalent to 4.464. Also, ξ‵ij is the transformation metrics tensor given
more regular than that observed by Eljack [22] at the Reynolds by
number of 5 × 104 .  
In the present work, a conditional time averaging [the proper 1 ∂ξj  ∂x 
ξ‵ij  ; J  i  (7)
orthogonal decomposition (POD)] and a conditional phase averaging J ∂xi ∂ξj
are used to investigate the underlying mechanism that generates and
sustains the LFO phenomenon in a compressible flow around the The Favre-filtered equations consist of the terms from the Navier–
NACA-0012 airfoil at near-stall conditions. Stokes equations on the left-hand side, in addition to the terms on the
4 Article in Advance / ELJACK AND SORIA

right-hand side resulting from the filtering operation. The under- waves at the high unresolved wave-number range by damping their
bracketed terms represent the effects of the unresolved subgrid-scale amplitude with minimal effect on the resolved waves at lower wave
(SGS) turbulent structures. Also, τij is the SGS stress tensor; Qj is the numbers. The filtering scheme used here is a fourth-order compact
SGS heat flux vector Qj  ρu~ j T~ − u~ j T;
~ and ∂Gj ∕∂xj is the SGS scheme developed by Gaitonde and Visbal [29]. The integral char-
turbulent diffusion term, where Gj  ρe uj uk uk − euj uk uk  is the acteristic boundary condition is applied at the freestream and the far-
SGS viscous diffusion term and Dj  σ ij ui − σ~ ij u~ i . The nonlinear field boundaries of Sandhu and Sandham [30]. The zonal character-
istic boundary condition is applied at the downstream exit boundary
subfilter viscous contribution [term (2) in Eq. (2)] is from the use of T~ of Sandhu and Sandham [31], which is considered a nonreflected
to evaluate μT.
~ This term is very small and negligible compared to
boundary condition to overcome the circulation effects at the boun-
term (1) in the same equation. Similarly, the SGS viscous diffusion daries. The adiabatic and no-slip conditions are applied at the airfoil
term ∂Dj ∕∂xj is also neglected. The SGS heat flux vector Qj and SGS surface. A periodic boundary condition is applied in the spanwise
turbulent diffusion term ∂Gj ∕∂xj are negligible for low-Mach-num- direction for each step of the Runge–Kutta time steps. The internal
ber flows [24]. branch-cut boundary was updated at each step of the Runge–Kutta
The SGS stress tensor τij represents the effect of the small-scale scheme.
turbulent structures and is defined as In terms of the airfoil chord c, the dimensions of the computational
domain were set as follows: Lξ  5.0c in the wake region, from the
τij  ρ ug ~ i u~ j 
i uj − u (8) airfoil trailing edge to the outflow boundary in the streamwise
direction; Lη  7.3c in the wall-normal direction (the C-grid radius);
The SGS stress tensor τij is modeled by using the eddy viscosity and Lζ  0.5c in the spanwise direction, where ξ, η, and ζ are the
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

concept under the low compressibility: curvilinear coordinates along the airfoil surface, in the wall-normal
direction, and in the spanwise direction, respectively.
1 A grid sensitivity study was performed to assess the effect of the
τij − δij τkk  2νt S~ij (9)
3 grid distribution on the characteristics of the LSB and its associated
LFO. Three different grid distributions were used; a coarse grid (grid
where τkk refers to the trace of the SGS Reynolds stress tensor, and νt 1: constructed and used by Almutairi and Alqadi [19]) with the
refers to the turbulent eddy viscosity obtained by the mixed-time- dimensions of N ξ × N η × N ζ  637 × 320 × 86, a fine grid with
scale (mts) model developed by Inagaki et al. [25]. The model is the dimensions of 780 × 320 × 101 (grid 2), and a finer grid with
turned off automatically in the laminar flow region. Thus, it over- the dimensions of 980 × 320 × 151 (grid 3). Table 1 shows a summary
comes the drawbacks of using a wall-damping function. In this model, of the three different grids. Grid 2 is a refined and optimized version of
the eddy viscosity is calculated by using the following definition: grid 1. The grid points in the η direction were redistributed in grid 2
such that around 60% of the total grid points were within one chord
νt  Cmts τs ksgs (10) from the airfoil surface. Grid 2 and grid 3 were generated using
a hyperbolic grid generator to improve the grid orthogonality and
where Cmts refers to the model fixed parameter (in the current study, minimize the grid skewness. The wall-normal grid spacing was
Cmts  0.03). Also, τs refers to the time scale given by reduced to ensure a wall-normal spacing of y ≤ 1. Having inad-
!−1  −1 equate spacing in the ξ direction could affect the development of the
Δ
 Cτ evolving Kelvin–Helmholtz instability, and consequently the transi-
τ−1
s  p  (11) tion process. An equidistant grid spacing with a relatively small Δx
ksgs jSj

was adopted in the range 0 ≤ x∕c ≤ 0.5 for grid 2 and grid 3. The
number of grid points in the spanwise direction was increased from 86
where grid points in grid 1 to 101 grid points in grid 2 and 151 grid points in
q grid 3. This has reduced the grid spacing in viscous wall units to less
jSj
  2Sij Sij ; Δ
  ΔxΔyΔz1∕3 (12) than 15 and significantly improved the solution. An adequate grid
distribution in the spanwise direction is crucial for the breakdown of
the evolving Kelvin–Helmholtz instability. Thus, an adequate grid
Δ
 refers to the filter size, and Cτ is the time-scale parameter. In the
distribution is essential for predicting the correct location of transition.
current study, Cτ  10 [25]. The velocity scale ksgs is defined by
Figure 1 shows the mean pressure and mean skin-friction coefficients
^ 2 for grid 1, grid 2, and grid 3. The time-averaged pressure coefficient
ksgs  u − u (13)
distribution on the airfoil suction side for grid 2 has a much stronger
As long as the flow is fully resolved, the preceding definition gives adverse pressure near the leading edge. For the mean skin-friction
a zero velocity scale in the laminar flow region. Consequently, the coefficient, the bubble size was smaller in the grid 2 case. In contrast,
eddy viscosity νt also approaches zero in this region. it was a bit longer in the coarse grid case (grid 1). The bubble character-
istics are independent of the grid at a grid resolution higher than that of
grid 2, as seen in the figure. It is noted that the differences in the mean
B. Computational Setup pressure and mean skin-friction distributions for grid 1 and grid 2 are
The LES code used in the present simulations is an LES version of huge compared to the small change in the overall grid points. However,
the direct numerical simulation code written and validated by Jones the difference in grid 1 and grid 2 is not only in the number of grid points
et al. [16]. The Navier–Stokes equations were discretized using a but also in the quality of the grid. Grid 1 contains high skewed cells, the
fourth-order explicit central difference scheme for spatial discretiza- grid points are not adequately distributed in the wall-normal direction,
tion in the interior points. The fourth-order boundary scheme of and the grid is stretched in the streamwise direction. On the contrary,
Carpenter et al. [26] was used to treat points near and at the boundary. grid 2 is optimized to have orthogonal cells and minimum skewness.
To preserve the spatial characteristics, the transformation metrics Moreover, the grid points are optimally distributed in the wall-normal
tensor ξ‵ij was evaluated by using the same fourth-order scheme.
Temporal discretization is performed using a low-storage fourth-
order Runge–Kutta scheme. The solution stability was improved Table 1 Computational grid parameters
by implementing an entropy splitting scheme by Sandham and Yee
Grid y Δx Δz Nξ Nη Nζ Total points
[27]. The entropy splitting constant β was set equal to 2.0 [28]. The
combination of a relatively coarse grid and high-order scheme results 1 >1 <50 <50 637 320 86 17,530,240
in the generation of spurious high-frequency waves. The nonphysical 2 <1 <15 <15 780 320 101 25,209,600
waves will contaminate the solution if they are not eliminated at each 3 <1 <10 <10 980 320 151 47,353,600
time step. Thus, the solution must be filtered to eliminate spurious
Article in Advance / ELJACK AND SORIA 5

Fig. 1 Sensitivity of the mean pressure coefficient and the mean skin-friction coefficient to grid refinement at the angle of attack of α  9.25 deg.

direction such that the grid points are intensified near the wall and along samples for flow statistics were collected after 200 nondimensional
the shear layer. Also, an equidistant grid distribution is employed in the time units, which is equivalent to 16 flow-through times. The locally
transition region. Finally, the number of grid points is significantly time-averaged and spanwise ensemble-averaged pressure, velocity
increased in the spanwise direction, which enhanced the prediction of components, and Reynolds stresses were sampled every 50 time steps
the transition location. That is, when the grid is refined, smaller vortices on the x–y plane. A dataset of 20,000 x–y planes was recorded at a
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

are resolved and their evolution is better predicted. Thus, the mean frequency of 204, covering a time period of 100 nondimensional time
pressure coefficient and mean skin-friction coefficient have a huge units (started at flow time of 200 and ended at 300 nondimensional
difference despite the relatively small change in the number of grid time units).
points. A continuous average was performed to show that the flow
The grid sensitivity study shows that the LFO was always captured samples were collected after the flow became statistically stationary
whether a coarse or a fine grid was used. However, when a coarse grid in time. The dataset covers 100 nondimensional time units or four
is used, the LFO is captured at lower angles of attack. As the grid is LFO cycles. The average was performed on four intervals of 25
refined, the LFO is captured at slightly higher angles of attack. nondimensional time units (one LFO cycle) each. Figure 2 shows
profiles of the mean streamwise velocity component (left) and the
C. Validation of the LES Data root mean square of the streamwise velocity fluctuations (right).
Four profiles showing average over the first, second, third, and fourth
A computational grid with the dimensions of N ξ × N η × N ζ  25 nondimensional time units are presented. As seen in the figure, the
780 × 320 × 101 was generated around the NACA-0012 airfoil for mean velocity profiles have collapsed in a single curve. The profiles
each of the sixteen angles of attack (α  9.0–10.1 deg at increments of the root mean square of the velocity show small variation near
of 0.1 deg as well as α  8.5, 8.8, 9.25, and 10.5 deg). A domain the wall. Thus, the flow is statistically stationary in time, and the last
width of Lζ  0.5c was considered sufficient for the current Reyn- 100 nondimensional time units are used to generate statistics of the
olds number. The simulation Reynolds number and Mach number flowfield.
were Rec  5 × 104 and M∞  0.4, respectively. The computa- The time-averaged shape of the LSB at the angles of attack of 9.25
tional domain was initiated with clean freestream conditions and 10.1 deg is shown in Fig. 3. The laminar boundary layer separates
(ρ∞  1, ρ∞ U∞  1, ρ∞ V ∞  0, ρ∞ W ∞  0, and T ∞  1). at point S, undergoes transition to turbulence, and reattaches at point
Hence, the freestream flow direction was parallel to the horizontal R. The flow domain between these two points is split into two
axis. The simulations were performed with a time step of 10−4 non- regions: the free shear layer, and the recirculating bubble. These
dimensional time units. The samples for statistics were collected once two regions are divided by the mean dividing streamline: line S −
the flow became statistically stationary in time after 50 nondimen- T 0 − R on the left-hand side of the figure. The LSB is bounded by the
sional time units. Aerodynamic coefficients were sampled for each mean dividing streamline and the airfoil surface (line S − T − R).
angle of attack at a frequency of 10,000 to generate 2.5 million The free shear layer starts laminar in the vicinity of the airfoil leading
samples over a time period of 250 nondimensional time units (started edge and becomes turbulent downstream of point T 0 [32]. The LSB at
at a flow time of 50 and ended at 300 nondimensional time units). The the angle of attack of 10.1 deg is longer and thicker than the LSB at

Fig. 2 Mean and root mean square of the streamwise velocity obtained from the first, second, third, and fourth 25 nondimensional time units (NTUs) at
x∕c  0.505.

Fig. 3 Time-averaged shape of the LSB at the angles of attack of 9.25 and 10.1 deg.
6 Article in Advance / ELJACK AND SORIA
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

Fig. 4 Profiles of the mean and root mean square (RMS) of the streamwise velocity component: LES data (solid line) and experimental data (circles).

the angle of attack of 9.25 deg. A secondary vortex lies beneath the Near the airfoil leading edge, a laminar boundary layer develops,
LSB and counter-rotates with it. Local vertical axes measured from and the pressure coefficient decreases drastically. The boundary layer
the surface of the airfoil are shown. The distance x∕c of each local remains laminar until the pressure gradient changes from favorable to
axis is measured from the airfoil leading edge. unfavorable. Consequently, the laminar boundary layer detaches and
The mean streamwise velocity and the root mean square of the travels away from the airfoil surface to create a region of separated flow
velocity fluctuations were extracted along the local vertical axes. near the surface. The flow then reattaches to the airfoil surface, and the
Figure 4 shows comparison of the extracted data with the incom- LSB forms as seen in Fig. 3. The pressure gradient is favorable across
pressible experimental data of Ohtake [33] and Ohtake and Motoha- the bubble; however, the pressure gradient becomes unfavorable down-
shi [34]. The experimental data were acquired at Rec  5 × 104 stream of the LSB, but the flow is turbulent and has enough energy to
using hot-wire anemometry. The discrepancy in the velocity near overcome new separation. In numerical simulation, if the velocity field
the wall is due to the fact that the hot wire can measure only positive is well predicted, then one expects the pressure to be very well
velocities, and negative velocities are interpreted as positive veloc- predicted. However, this is not the case in this study. Despite the good
ities with the same magnitude. Also, the hot-wire anemometry does agreement between the predicted and measured velocity fields, the
not recognize which velocity component cools its wire; thus, it LES appears to have underestimated the pressure coefficient in the
overestimate the streamwise velocity component at locations where region occupied by the LSB and the secondary vortex as seen in the
the wall-normal velocity component has significant magnitude, figure. This discrepancy is more significant in the recirculating zone of
hence the discrepancy in the velocity away from the wall. the LSB, upstream of the focus, and the secondary vortex. The height of
Figure 5 compares predicted variations of the pressure coefficient the recirculating zone is very small compared to its length, as seen in
around the airfoil surface with the incompressible experimental data Fig. 3. As this zone is thin, a very small variation in the velocity would
of Ohtake et al. [35] at Rec  5 × 104 transformed to its correspond- result in a considerably large variation in the pressure field. The
ing compressible counterpart. The present study is carried out at a discrepancy in the pressure coefficient is more pronounced at the angle
Mach number of 0.4, whereas the experiment was carried out at of attack of 10.1 deg as the recirculating zone become bigger and the
incompressible conditions. Thus, the Prandtl-–Glauert transforma- flow is massively separated.
tion rule [36] was used to add the effects of compressibility to the The pressure coefficient was duly integrated about the airfoil at
experimental data, both in terms of the shift in the angle of attack and each time step for each angle of attack to obtain time histories of the
the magnitude of the lift coefficient. Reducing the Mach number of lift and the drag coefficients. Time histories are then time averaged
the numerical simulation to match that of the experiment would to obtain the mean lift and drag coefficients. Figure 6 shows plots
considerably increase the computational cost. of the mean lift and drag coefficients versus the angle of attack. The

Fig. 5 Profiles of the mean pressure coefficient plotted versus the distance from the airfoil leading edge x∕c. The experimental data are transformed to
their compressible counterparts.
Article in Advance / ELJACK AND SORIA 7

Fig. 6 Mean lift and drag coefficients plotted versus the angle of attack α. The experimental data of the lift coefficient are transformed to their
compressible counterparts.

experimental data of the lift coefficient of Ohtake et al. [35] were average of an instantaneous variable of the flowfield hψi defined on
transformed to their compressible counterparts using the Prandtl– three different levels a mean-lift average, a high-lift average, and a
Glauert transformation rule. The drag force is not defined for an low-lift average. The mean-lift average of the variable hψi;  is simply
inviscid flow; therefore, the Prandtl–Glauert transformation rule the time average of all data samples of the variable. The high-lift
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

cannot be used to add effects of compressibility to the incompressible average of the variable hψi; ^ is the time average of all data samples that
experimental data. The lift coefficient compares with acceptable have values higher than the mean value. The low-lift average of the
accuracy to the experimental data. However, the lift coefficient values variable 〈ψ̌〉; is the average of all data samples that have values less
estimated from the LES data are less than their corresponding exper- than the mean value. The lift coefficient signal is used to identify the
imental values. The discrepancy in the lift coefficient is due to the data samples that are above or below the mean. It is implemented by
differences between the predicted and measured pressure coeffi- taking the mean of the lift coefficient at each angle of attack. The
cients, as discussed earlier in this paper. The numerical simulation indices of data points of the time history of the lift coefficient that are
data of the drag mimic that of the experiment. The discrepancy is due above/below the mean of the lift coefficient are then stored in high-
to the effect of compressibility and the difference between the mea- lift/low-lift data files, respectively. The indices are then used to locate
sured and calculated pressures as discussed earlier in this paper. The the data of other flow variables that are above/below their corre-
black stars display the LES data of Almutairi and Alqadi [19] using sponding mean, and consequently used to estimate the low-lift and
grid 1. It is presented here to demonstrate how the grid distribution high-lift time averages for all of the flow variables. Thus, the condi-
affects the LSB and the LFO, and consequently the variation of the lift tional time averaging is synchronized for all variables so that at a
coefficient with the angle of attack at near-stall conditions. given time step, all variables are allocated to either the high-lift or
low-lift regimes. Figure 7 illustrates the concept of the conditional
III. Results and Discussion time averaging that is based on the lift coefficient.
When the LFO is fully developed, the flowfield switches between
A. Conditional Time Average of the Oscillating Flow an attached phase and a separated phase, and the lift coefficient signal
The signal of the lift coefficient at each angle of attack was used as is Gaussian. Consequently, the time-averaged surplus and deficit of
a reference for conditionally time averaging the flowfield. The time the flow above and below its mean feature the time-averaged attached
phase and separated phase of the oscillating flow, respectively. In
this sense, one can define a time-averaged surplus Δψ  and deficit
Δψ − of the flow variable above and below that of the mean flow,
respectively. Thus, the time-averaged surplus of the flowfield is
estimated from the high-lift flowfield minus the mean-lift flowfield
(Δψ   hψi ^ − hψi).
 The time-averaged deficit of the flowfield is
estimated from the low-lift flowfield minus the mean-lift flowfield
(Δψ −  hψ̌i − hψi).
Figure 8 shows streamlines patterns of the time-averaged attached
phase of the oscillating flow (ΔU 
1 and ΔU 2 ) and the time-averaged
separated phase of the oscillating flow (ΔU−1 and ΔU−2 ) superim-
posed on color maps of the time-averaged attached phase of the
oscillating pressure ΔP and the time-averaged separated phase of
Fig. 7 Time history of the lift coefficient at the angle of attack of 9.80 deg. the oscillating pressure ΔP− , respectively, at the angle of attack of
The dashed, solid, and dashed–dotted horizontal lines show the high-lift, 9.8 deg. As seen in the figure, the time-averaged attached and
mean-lift, and low-lift averages, respectively. separated phases of the oscillating flow are rotating in a closed loop

Fig. 8 Streamlines patterns of the time-averaged attached and separated phases of the oscillating flow superimposed on color maps of the time-averaged
oscillating pressure at α  9.8 deg.
8 Article in Advance / ELJACK AND SORIA

around the airfoil in the clockwise and the anticlockwise directions, location of the maximum TKE, indicated by the black filled circle,
respectively. On the suction surface of the airfoil, the time-averaged moves downstream and away from the airfoil surface as the angle of
attached and separated phases of the oscillating flow are positive and attack increases.
negative, as well as flow in the streamwise and antistreamwise Figure 11a shows a sketch of the triad of vortices. As seen in the
directions, respectively. The oscillating flow is observed in all of figure, the triad of vortices consists of two corotating vortices (TCVs;
the investigated angles of attack, including at the zero angle of attack. an upstream vortex and a downstream vortex), and a secondary vortex
However, the frequency and magnitude of the oscillations vary with lies beneath them and counter-rotates with them. The triad of vortices is
the angle of attack. At a zero angle of attack, the frequency of formed in the vicinity of the leading edge on the suction surface of the
oscillations is the same as that of the von Kármán alternating vortices airfoil in a region almost coinciding with that of the LSB. The upstream
at the trailing edge. However, the frequency of oscillations is inverse vortex of the TCVs front tip forms a half-saddle (HS1) on the airfoil
proportional to the angle of attack, and it shifts to a low frequency at surface just upstream of the separation point of the shear layer, and its
the inception of stall. The magnitude of the oscillating flow is very rear end forms a full saddle. The downstream vortex of the TCVs is
small and negligible compared to other flow modes at the zero angle submerged entirely inside the region of the LSB and aligned such that
of attack, increases until it becomes significant at the inception of its front tip forms a full saddle with the upstream vortex, whereas its
stall, and dominates all flow modes at the angle of attack of 9.8 deg. rear tip forms a half-saddle (HS2) on the airfoil surface upstream of the
On the suction surface of the airfoil, when the oscillating flow is in the reattachment point, as seen in the figure. During the attached and
streamwise direction, it adds momentum to the boundary layer and separated phases of the time-averaged oscillating flow, the TCVs rotate
helps it to remain attached against the APG. On the contrary, when in the clockwise and anticlockwise directions, respectively.
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

the oscillating flow is in the antistreamwise direction, it subtracts During the attached phase, the time-averaged oscillating flow
momentum from the boundary layer and forces it to separate, despite along the outer and inner halves of the laminar portion of the
the favorable pressure gradient. separated shear layer flows in the streamwise and antistreamwise
The left-hand side of Fig. 9 shows the time-averaged attached directions, respectively. Consequently, a clockwise twister is created,
phase (top) and the time-averaged separated phase (bottom) of the as seen in the vector plot in Fig. 11b. Thus, the upstream vortex of the
flowfield at α  9.6 deg. The black, filled circle displays the loca- TCVs is driven by the gradient of the oscillating velocity across the
tion of the maximum turbulent kinetic energy (TKE). It is noted that laminar portion of the separated shear layer and is faithfully aligned
the location of the maximum TKE moves downstream and away from with it. The direction of rotation is reversed in the separated phase.
the airfoil surface in the time-averaged separated phase of the flow- That is, the time-averaged oscillating flow flows in the antistream-
field, and vice versa. The right-hand side of the figure displays the wise direction, and the upstream vortex rotates in the anticlockwise
time-averaged attached phase (top) and the time-averaged separated direction. The downstream vortex of the TCVs rotates in the same
phase (bottom) of the oscillating flow at α  9.6 deg. It is also noted direction as that of the upstream vortex, as seen in Fig. 11c. The focus
that the triad of vortices is entirely submerged in the recirculating of the downstream vortex is located just upstream and below that of
region of the LSB. the LSB, in the mean sense. The secondary vortex acts as a roller
Figure 10 shows the time-averaged attached phase (left) and support that facilitates the rotation and orientation of the TCVs. Thus,
separated phase (right) of the oscillating flow for the angles of attack the triad of vortices is a self-sustained system of vortices that feeds on
of 9.4, 9.8, and 10.5 deg. The dashed line displays the mean dividing the mean flow.
streamline. The figures are zoomed in in such a way that the strength Most of the energy fed into the triad of vortices is extracted by the
and orientation of the triad of vortices and how they vary with the upstream vortex of the TCVs. The upstream vortex extracts energy
angle of attack can be examined. The color maps indicate the time- from the mean flow until it saturates. Then, the upstream vortex slides
averaged oscillating spanwise vorticity along the laminar portion of over the downstream vortex of the TCVs, merges with it, expands,
the separated shear layer. The strength of the triad of vortices can be and advects downstream. When the upstream vortex expands, it
examined using the magnitude of oscillating spanwise vorticity. The changes the direction of the oscillating flow from streamwise to the
strength of the triad of vortices increases consistently as the angle of antistreamwise direction, and vice versa. Hence, if the instantaneous
attack increases, until it peaks at the angle of attack of 9.8 deg. After flowfield is attached, it will separate as a consequence of the expan-
that, the strength decreases with the angle of attack. It is noted that the sion of the upstream vortex, and vice versa.

Fig. 9 Streamline patterns of the flowfield and the oscillating flow superimposed on color maps of the streamwise velocity component and the oscillating
streamwise velocity, respectively, at α  9.6 deg.
Article in Advance / ELJACK AND SORIA 9
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

Fig. 10 Streamline patterns of time-averaged attached and separated phases of the oscillating flow superimposed on color maps of time-averaged
attached and separated phases of oscillating spanwise vorticity.

Z
B. Application of the POD and the Dynamics of the Triad of Vortices
am t  VX; tΨm X dX (17)
The POD method was proposed by Lumley to objectively recover Domain
the most energetic structures of a turbulent flowfield [37]. There are
various methods to implement the POD. The reader is referred to the The fluctuating flow variables V are then reconstructed from
works of Lumley [37,38], Sirovich [39], and Holmes et al. [40] for
more details on the theory of the POD method and its various X
M
implementations to numerical and experimental data. The snapshot VX; t  am tΨm X (18)
POD method [39], used in the present work, is formulated as follows: m1

Aφ  Λφ (14) The POD method is used to shed light on how the upstream vortex
of the TCVs expands, advects, and eventually changes the direction
A is the correlation matrix, φm are the POD eigenfunctions, and of the oscillating flow. The locally time-averaged and spanwise
ensemble-averaged fluctuating streamwise velocity, wall-normal
Λm are the POD eigenvalues. The spanwise ensemble averaged
velocity, and pressure are used. The two-point correlations of the
fluctuating streamwise velocity, wall-normal velocity, and pressure
three variables, in time, are duly estimated using 20,000 data points
(hu10 0 i; hu20 0 i and hp 0 0 i) are used to formulate the correlation matrix
that span 100 nondimensional time units, or four low-frequency
hAi as follows:
cycles. The eigenvalue problem was solved, and the POD eigenval-
1 ues and eigenvectors were obtained.
Aij  VX; ti ; VX; tj  (15) Figure 12 shows the two most dominant low-frequency POD
M
modes (LFO mode 1 and LFO mode 2) for the angle of attack of
2 00 3 9.8 deg. The two low-frequency modes were constructed by multi-
hu1 i
6 00 7 plying their corresponding orthonormal spatial POD modes by the
V6 hu
4 2 i5
7 average amplitude of their corresponding POD coefficients: ja1 tj
hp 0 0 i and ja2 tj. LFO mode 1 features the globally oscillating flow
around the airfoil, the oscillating pressure along the airfoil chord,
where ⋅; ⋅ represents the inner product process, and M is the number and the process that generates and sustains the triad of vortices. LFO
of snapshots. The eigenvalues and eigenfunctions are used to con- mode 2 features the expansion and advection of the upstream vortex
struct the spatial POD modes Ψ ~ m X as follows: of the TCVs. The third dominant POD mode is a High Frequency
Oscillation (HFO) mode featuring the oscillating flow mode along
1 X
M the wake of the airfoil. The pressure coefficient was estimated using
~ m X 
Ψ p φm tk VX; tk  (16) the pressure field reconstructed using LFO mode 1 and LFO mode 2.
M Λm k1 The obtained pressure coefficient was then duly integrated around the
airfoil to obtain the reconstructed lift coefficient. Figure 13 compares
The obtained spatial POD modes Ψ~ m X are orthogonal but not the lift coefficient reconstructed using LFO mode 1 and LFO mode 2
normalized. The spatial POD modes are duly normalized to obtain the to that estimated from the LES data at the angle of attack of 9.8 deg.
orthonormal spatial POD modes Ψ. The POD coefficients am t are LFO mode 1 and LFO mode 2 reconstruct the lift favorably, as seen in
then determined using the figure.
10 Article in Advance / ELJACK AND SORIA

a) The triad of vortices


Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

b) The upstream vortex of the two corotating vortices

c) The downstream vortex of the two corotating vortices


Fig. 11 Sketch showing the triad of vortices, and vector plots of the two corotating vortices.

Fig. 12 Streamlines patterns superimposed on color maps of the pressure field for the most dominant low-frequency POD modes at α  9.8 deg.

LFO mode 1 and LFO mode 2 govern the evolution and dynamic of
the triad of vortices; however, the link between these two modes is
missing. What is the significance of each mode compared to other
flow modes at each instant in time? Scaling the temporal POD modes
with their corresponding eigenvalues would give an estimate of the
percentage of the scaled amplitude of each of the temporal POD
modes at any instant in time. The percentage of the energy content in
POD mode m at any instant in time can be estimated as follows:

Λm φm t


Fig. 13 POD reconstruction of the lift coefficient using LFO mode 1 and ζm t  PM m φm t
(19)
LFO mode 2 at α  9.8 deg. m1 Λ
Article in Advance / ELJACK AND SORIA 11

Fig. 14 Percentage of energy content of the fluctuating flow in each POD mode: LFO mode 1 (solid line), and LFO mode 2 (dashed–dotted line).

Figure 14 shows the percentage of the energy content of the   hψ̌ i  hψ 0 i


hψi  hψi (20)
fluctuating flow in each POD mode as a function of time at the angles |{z}
hψ 0 0 i
of attack of 9.7 and 9.8 deg. The scaled temporal POD modes make
more sense since they display the temporal evolution of the percent-
age of energy content of each flow mode. The phase angles of one
where hψi is the spanwise ensemble-averaged mean of the variable;
LFO cycle are shown in the figure. At these angles of attack, the LFO
hψ̌i is the spanwise ensemble-averaged and conditionally phase-
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

cycle becomes more regular and repeats periodically with some


averaged flow variable; and hψ 0 i is the spanwise ensemble-averaged
disturbed cycles. LFO mode 1 and LFO mode 2 are interlinked. That
is, if LFO mode 1 peaks at a relatively high amplitude, so will LFO fluctuation of the variable “turbulent fluctuations.”
Figure 14 shows the temporal evolution of the energy content of
mode 2, and vice versa. Also, when the periodic cycle of LFO mode 1
is disturbed, so is that of LFO mode 2, as seen in the figure. LFO mode LFO mode 1 and LFO mode 2. The flow is fully separated at the phase
1 peaks at an energy content of 75%, and LFO mode 2 peaks at about angle of Φ  0 deg, and the lift coefficient is at its minimum value,
25%. The negative sign of the percentage of the energy content of the as seen in Fig. 16. Figures 17 and 18 show streamline patterns of the
two modes indicates that the oscillating flow is rotating in the phase-averaged flow superimposed on color maps of the phase-
anticlockwise direction rather than the percentage of energy content averaged pressure, as well as color maps of the phase-averaged
being negative. streamwise velocity, respectively, for the angle of attack of 9.7 deg.
Figure 15 shows the reconstructed fluctuating flow using LFO mode The events illustrated in these four figures can be pieced together to
1, LFO mode 2, and the HFO mode at the angle of attack of 9.8 deg. The show that the self-sustained separation and reattachment of the flow
starting point in time for the reconstructed flow at a flow time of 206.0 repeats periodically at low frequency in the following sequence of
is shown in Fig. 13 and indicated by the black circle and Φ  180 deg. events:
The evolution of the oscillating flow is visualized in step-by-step 1) At Φ  0 deg, the percentage of the energy content of the LFO
snapshots that reveal how the triad of vortices switches the direction mode 1 is at its maximum amplitude, and the negative sign indicates
of the oscillating flow and separates the instantaneous flowfield. When that the oscillating flow on the suction surface of the airfoil is in the
the cycle of the LFO proceeds to phase angles Φ > 0 deg, the process antistreamwise direction. Hence, the oscillating flow subtracts
reverses its direction and the fully separated instantaneous flow starts to momentum from the boundary layer and forces it to separate. Thus,
attach. Thus, a similar sequence of events takes place. the instantaneous flowfield is fully separated.
2) At Φ  90 deg, the percentage of the energy content of LFO
mode 1 is zero, which means that the instantaneous flowfield reaches
C. Conditional Phase Average of the Flowfield an equilibrium state. The percentage of the energy content of LFO
The conditional time average revealed a triad of vortices and an mode 2 reaches the threshold of −25%. Thus, the upstream vortex
oscillating flow that sustains the LFO phenomenon. Furthermore, the of the TCVs expands, flips the direction of the oscillating flow to the
POD method unveiled two low-frequency modes, LFO mode 1 and clockwise direction, and starts a new imbalance.
LFO mode 2, that govern the evolution and dynamic of the triad of 3) At Φ  180 deg, the percentage of the energy content of LFO
vortices and switch the direction of rotation of the oscillating flow. In mode 1 is at its maximum amplitude, and the positive sign indicates
this section, the previously defined conditional time average on three that the oscillating flow on the suction surface is in the streamwise
levels (high-lift, mean-lift, and low-lift) was expanded to a series of direction. Hence, the oscillating flow adds momentum to the boun-
37 intervals that gives continuous phase information of one complete dary layer and helps it to attach. Thus, the flow is fully attached
cycle. Figure 16 illustrates the conditional phase-averaging process. downstream of the LSB.
The right-hand side of the figure shows time history of the lift 4) At Φ  270 deg, the percentage of the energy content of LFO
coefficient at the angle of attack of 9.80 deg. The solid line denotes mode 1 is zero, which means that the instantaneous flowfield reaches
the lift coefficient signal, and the dotted line displays the low-pass- a new equilibrium state. The percentage of the energy content of LFO
filtered lift coefficient. The left-hand side of the figure shows one mode 2 reaches the threshold of 25%. Thus, the upstream vortex
sinusoidal cycle with a magnitude range of that of the low-pass- expands, flips the direction of the oscillating flow to the anticlock-
filtered lift coefficient divided into 36 equally distributed phases. wise direction, and starts a new imbalance.
The phase-averaging process is implemented in the same manner 5) At Φ  360 deg, The percentage of the energy content of LFO
described and used in Sec. III.A for the conditional time averaging. mode 1 is at its maximum amplitude, and the negative sign indicates
The primary objective is to examine the evolution of the low- that the oscillating flow on the suction surface is in the antistreamwise
frequency flow modes; thus, filtering out the high frequencies does direction and subtracting momentum from the boundary layer. Thus,
not affect the resulting phase-averaged flowfield. The low-pass- the instantaneous flowfield is fully separated, again.
filtered lift coefficient signal was split into an ascending part (Φ  In summary, the sustainability of the LFO phenomenon, in a
0 deg to Φ  180 deg), in which the lift is increasing with time, and naturally evolving flow, is governed by LFO mode 1 and LFO
a descending part (180 deg < Φ < 360 deg). The phase-averaging mode 2 as shown in Figs. 14 and 15. These two low-frequency modes
process was applied to the ascending and the descending low-pass- are interlinked; they govern and sustain the LFO phenomenon. The
filtered lift coefficients and their corresponding low-pass-filtered process is self-sustained and quasi periodic in a manner that once
data points. The flowfield studied in the present work involves the oscillating flow reaches an equilibrium, the upstream vortex of
exceptionally low frequency, and one cycle spans over 25 nondimen- the TCVs expands and creates a new disequilibrium. However, for
sional time units. Collecting data that cover 100 cycles is not feasible reasons that are not clear now, sometimes the triad of vortices reverses
in numerical simulation. For a spanwise ensemble-averaged instan- the process before the cycle is completed and the periodic cycle is
taneous flow variable hψi, the decomposition is as follows: disturbed.
12 Article in Advance / ELJACK AND SORIA
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

Fig. 15 Streamline patterns superimposed on color maps of the spanwise vorticity ωz for the POD reconstruction of the fluctuating flow at
α  9.8 deg.

At the angle of attack of 9.25 deg, the mean flowfield is attached angle of attack of 9.8 deg. At higher angles of attack, the mean
downstream of the LSB, and the upstream vortex of the TCVs expands flowfield is fully separated and the upstream vortex expands intermit-
intermittently. Therefore, the instantaneous flowfield separates inter- tently. Thus, the instantaneous flowfield attaches intermittently. At the
mittently. As the angle of attack increases, the upstream vortex expands angle of attack of 10.5 deg, the upstream vortex expands less fre-
more frequently, as illustrated in Fig. 14. Hence, the instantaneous quently, and the flowfield remains separated for longer time intervals.
flowfield separates more frequently until the LFO process becomes The airfoil eventually undergoes a full stall when the angle of attack is
regular and more pronounced in the magnitude of oscillation at the further increased, as shown qualitatively by Eljack [22].
Article in Advance / ELJACK AND SORIA 13

Fig. 16 The conditional phase-averaging process: one sinusoidal cycle (left), and time history of the lift coefficient at α  9.80 deg (right).
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

Fig. 17 Streamline patterns of the phase-averaged flow superimposed on color maps of the phase-averaged pressure at α  9.7 deg.
14 Article in Advance / ELJACK AND SORIA
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

Fig. 18 Color maps of the phase-averaged streamwise velocity at α  9.7 deg.

IV. Conclusions momentum to the boundary layer and helps it to remain attached
Large-eddy simulations of the flowfield around a NACA-0012 air- against an adverse pressure gradient, and vice versa.
foil were performed at a Reynolds number of 5 × 104 , a Mach number The spanwise width of the computational domain was set to 0.5
of 0.4, and sixteen angles of attack near stall conditions to investigate chords. This is a limitation of this study, and the findings of the present
the underlying mechanism behind the low-frequency flow oscillation. work need to be confirmed in a study with sufficiently large computa-
It is shown that a triad of three vortices, two corotating vortices and a tional domain. The presented mechanism also implies that the con-
secondary vortex counter-rotating with them, is behind the quasi- ditions for the LFO are met in both compressible and incompressible
periodic self-sustained LFO phenomenon. A global oscillation in the flows. However, it is not clear yet whether the LFO phenomenon can be
flowfield around the airfoil is observed in all of the investigated angles generated and sustained in an incompressible flow. Therefore, an
of attack. When the oscillating flow rotates in the clockwise direction, experimental work in a water tunnel or solving the incompressible
the upstream vortex of the TCVs whirls in the clockwise direction, Navier–Stokes equations using direct numerical simulation is merited.
stores energy until it saturates, expands, and switches the direction of The present investigation paves the way for formulating a time-
rotation of the oscillating flow; then, the process reverses its direction. accurate physics-based model of the LFO and stall prediction, and it
When the direction of the oscillating flow is clockwise, it adds opens the door for control of their undesirable effects. The model
Article in Advance / ELJACK AND SORIA 15

presented in this study has profound implications to the constantly [15] Sandham, N. D., “Transitional Separation Bubbles and Unsteady
increasing applications that operate at low-Reynolds-number con- Aspects of Aerofoil Stall,” Aeronautical Journal, Vol. 112, No. 1133,
ditions such as unmanned aerial vehicles, micro unmanned aerial July 2008, pp. 395–404.
vehicles, low-pressure compressors and turbines blades, helicopter https://doi.org/10.1017/S0001924000002359
[16] Jones, L. E., Sandberg, R. D., and Sandham, N. D., “Stability and
rotor blades, and vertical axis wind turbines. Finally, the present work Receptivity Characteristics of a Laminar Separation Bubble on an Aero-
can be extended to shed light on dynamic stall of airfoils operating at foil,” Journal of Fluid Mechanics, Vol. 648, April 2010, pp. 257–296.
low Reynolds numbers. https://doi.org/10.1017/S0022112009993089
[17] Almutairi, J. H., Jones, L. E., and Sandham, N. D., “Intermittent Burst-
ing of a Laminar Separation Bubble on an Airfoil,” AIAA Journal,
Acknowledgments Vol. 48, No. 2, Feb. 2010, pp. 414–426.
All computations were performed on the Aziz Supercomputer at https://doi.org/10.2514/1.44298
King Abdulaziz University’s High Performance Computing Center. [18] Jones, L. E., Sandberg, R. D., and Sandham, N. D., “Direct Numerical
The authors would like to acknowledge the computer time and Simulations of Forced and Unforced Separation Bubbles on an Airfoil at
technical support provided by the center. The authors would like to Incidence,” Journal of Fluid Mechanics, Vol. 602, May 2008, pp. 175–
207.
thank Tomohisa Ohtake and Yuta Yamaguchi of the Department of https://doi.org/10.1017/S0022112008000864
Aerospace Engineering at Nihon University for sharing their exper- [19] Almutairi, J., and Alqadi, I., “Large-eddy Simulation of Natural Low-
imental data, which made the validation of the large-eddy simulation Frequency Oscillations of Separating-Reattaching Flow near Stall Con-
data possible. The authors would also like to thank the anonymous ditions,” AIAA Journal, Vol. 51, No. 4, April 2013, pp. 981–991.
reviewers for their constructive comments, which helped improve a https://doi.org/10.2514/1.J052165
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

previous version of the paper. [20] Almutairi, J., Alqadi, I., and Eljack, E., “Large Eddy Simulation of a
NACA-0012 Airfoil near Stall,” Direct and Large-Eddy Simulation IX,
edited by J. Fröhlich, H. Kuerten, B. Geurts, and V. Armenio, Vol. 20,
References ERCOFTAC Series, Springer, Cham, Jan. 2015, pp. 389–395.
[1] Gaster, M., “The Structure and Behaviour of Laminar Separation Bub- https://doi.org/10.1007/978-3-319-14448-1_49
bles,” Aeronautical Research Council RM 3595, Her Majesty’s Station- [21] Eljack, E., Alqadi, I., and Almutairi, J., “Influence of Periodic Forcing
ery Office, March 1967. on Laminar Separation Bubble,” Direct and Large-Eddy Simulation X,
[2] Jones, B. M., “Stalling,” Journal of the Royal Aeronautical Society, edited by D. Grigoriadis, B. Geurts, H. Kuerten, J. Fröhlich, and V.
Vol. 38, No. 285, Aug. 1934, pp. 753–770. Armenio, ERCOFTAC Series, Vol. 24, Springer, Cham, Jan. 2018,
https://doi.org/10.1017/S0368393100109782 pp. 199–204.
[3] Zaman, K. B. M. Q., Bar-Sever, A., and Mangalam, S. M., “Effect of https://doi.org/10.1007/978-3-319-63212-4_24
Acoustic Excitation on the Flow over a Low-Re Airfoil,” Journal of [22] Eljack, E., “High-Fidelity Numerical Simulation of the Flow Field
Fluid Mechanics, Vol. 182, Sept. 1987, pp. 127–148. Around a NACA-0012 Aerofoil from the Laminar Separation Bubble
https://doi.org/10.1017/S0022112087002271 to a Full Stall,” International Journal of Computational Fluid Dynam-
[4] Zaman, K. B. M. Q., and McKinzie, D. J., “A Natural Low Frequency ics, Vol. 31, Nos. 4–5, May 2017, pp. 230–245.
Oscillation in the Wake of an Airfoil near Stalling Conditions,” NASA https://doi.org/10.1080/10618562.2017.1330953
Lewis Research Center TM-100213, Cleveland, OH, 1988. [23] Elawad, Y. A., and Eljack, E. M., “Numerical Investigation of the Low-
[5] Zaman, K. B. M. Q., McKinzie, D. J., and Rumsey, C. L., “A Natural Frequency Flow Oscillation over a NACA-0012 Aerofoil at the Incep-
Low-frequency Oscillation of the Flow over an Airfoil Near Stalling tion of Stall,” International Journal of Micro Air Vehicles, Vol. 11, April
Conditions,” Journal of Fluid Mechanics, Vol. 202, May 1989, pp. 403– 2019, pp. 1–17.
442. https://doi.org/10.1177/1756829319833687
https://doi.org/10.1017/S0022112089001230 [24] Vreman, A. W., Direct and Large-Eddy Simulation of the Compressible
[6] Bragg, M. B., Khodadoust, A., and Spring, S. A., “Measurements in a Turbulent Mixing Layer, Univ. of Twente, Enschede, The Netherlands,
Leading-Edge Separation Bubble due to a Simulated Airfoil Ice Accre- 1995.
tion,” AIAA Journal, Vol. 30, No. 6, June 1992, pp. 1462–1467. [25] Inagaki, M., Kondoh, T., and Nagano, Y., “A Mixed-Time-Scale SGS
https://doi.org/10.2514/3.11087 Model with Fixed Model-Parameters for Practical LES,” Journal of
[7] Bragg, M. B., Heinrich, D. C., and Khodadoust, A., “Low-Frequency Fluids Engineering, Vol. 127, No. 1, Jan. 2005, pp. 1–13.
Flow Oscillation over Airfoils near Stall,” AIAA Journal, Vol. 31, No. 7, https://doi.org/10.1115/1.1852479
July 1993, pp. 1341–1343. [26] Carpenter, M. H., Nordström, J., and Gottlieb, D., “A Stable and
https://doi.org/10.2514/3.49069 Conservative Interface Treatment of Arbitrary Spatial Accuracy,” Jour-
[8] Bragg, M. B., Heinrich, D. C., Balow, F. A., and Zaman, K. B. M. Q., nal of Computational Physics, Vol. 148, No. 2, Jan. 1999, pp. 341–365.
“Flow Oscillation over an Airfoil near Stall,” AIAA Journal, Vol. 34, https://doi.org/10.1006/jcph.1998.6114
No. 1, Jan. 1996, pp. 199–201. [27] Sandham, N. D., Li, Q., and Yee, H. C., “Entropy Splitting for High
https://doi.org/10.2514/3.13045 Order Numerical Simulation of Compressible Turbulence,” Computa-
[9] Broeren, A. P., and Bragg, M. B., “Low-Frequency Flowfield Unsteadi- tional Fluid Dynamics 2000, Vol. 178, No. 2, May 2002, pp. 307–322.
ness During Airfoil Stall and the Influence of Stall Type,” 16th AIAA https://doi.org/10.1006/jcph.2002.7022
Applied Aerodynamics Conference, AIAA Paper 1998-2517, June 1998. [28] Jones, L. E., “Numerical Studies of the Flow Around an Airfoil at Low
https://doi.org/10.2514/6.1998-2517 Reynolds Number,” Ph.D. Thesis, Univ. of Southampton, Southampton,
[10] Broeren, A. P., and Bragg, M. B., “Flowfield Measurements over an England, U.K., 2008.
Airfoil During Natural Low-Frequency Oscillations near Stall,” AIAA [29] Gaitonde, D. V., and Visbal, M. R., “High-Order Schemes for Navier–
Journal, Vol. 37, No. 1, Jan. 1999, pp. 130–132. Stokes Equations: Algorithm and Implementation into FDL3DI,” U.S.
https://doi.org/10.2514/2.678 Air Force Research Lab., Air Vehicles Directorate AFRL-VA-WP-TR-
[11] Broeren, A. P., and Bragg, M. B., “Spanwise Variation in the Unsteady 1998-3060, Wright–Patterson AFB, OH, 1998.
Stalling Flowfields of Two-Dimensional Airfoil Models,” AIAA Jour- [30] Sandhu, H. S., and Sandham, N. D., “Boundary Conditions for Spatially
nal, Vol. 39, No. 9, Sept. 2001, pp. 1641–1651. Growing Compressible Shear Layers,” Queen Mary and Westfield
https://doi.org/10.2514/2.1501 College, Univ. of London Rept. QMW-EP-1100, Faculty of Engineer-
[12] Tanaka, H., “Flow Visualization and PIV Measurements of Laminar ing, London, 1994.
Separation Bubble Oscillating at Low Frequency on an Airfoil Near Stall,” [31] Sandberg, R. D., and Sandham, N. D., “Nonreflecting Zonal Character-
International Congress of the Aeronautical Sciences, International Coun- istic Boundary Condition for Direct Numerical Simulation of Aerody-
cil of the Aeronautical Sciences, Yokohama, Japan, 2004, pp. 1–15. namic Sound,” AIAA Journal, Vol. 44, No. 2, Feb. 2006, pp. 402–405.
[13] Rinoie, K., and Takemura, N., “Oscillating Behaviour of Laminar https://doi.org/10.2514/1.19169
Separation Bubble Formed on an Aerofoil Near Stall,” Aeronautical [32] Roberts, W. B., “Calculation of Laminar Separation Bubbles and Their
Journal, Vol. 108, No. 1081, March 2004, pp. 153–163. Effect on Aerofoil Performance,” AIAA Journal, Vol. 18, No. 1,
https://doi.org/10.1017/S0001924000151607 Jan. 1980, pp. 25–31.
[14] Alam, M., and Sandham, N. D., “Direct Numerical Simulation of https://doi.org/10.2514/3.50726
“Short” Laminar Separation Bubbles with Turbulent Reattachment,” [33] Ohtake, T., “Flow Field Around NACA-0012 Airfoil in Low Reynolds
Journal of Fluid Mechanics, Vol. 410, May 2000, pp. 1–28. Numbers Part 2 Characteristics of Boundary Layer,” Journal of the
https://doi.org/10.1017/S0022112099008976 Japan Society for Aeronautical and Space Sciences, Vol. 64, No. 2,
16 Article in Advance / ELJACK AND SORIA

2016, pp. 123–130. [37] Lumley, J. L., “The Structure of Inhomogeneous Turbulent Flows,”
https://doi.org/10.2322/jjsass.64.123 Atmospheric Turbulence and Radio Propagation, edited by A. M.
[34] Ohtake, T., and Motohashi, T., “Flow Field Around NACA-0012 Airfoil Yaglom, and V. I. Tatarski, Nauka, Moscow, Jan. 1967, pp. 166–178.
at Low Reynolds Numbers Part 1 Characteristics of Airfoil Wake,” [38] Lumley, J. L., “Coherent Structures in Turbulence,” Transition and
Journal of the Japan Society for Aeronautical and Space Sciences, Turbulence, edited by R. E. Meyer, Academic Press, New York, 1981,
Vol. 57, No. 669, 2009, pp. 397–404. pp. 215–242.
https://doi.org/10.2322/jjsass.57.397 [39] Sirovich, L., “Turbulence and the Dynamics of Coherent Structures.
[35] Ohtake, T., Nakae, Y., and Motohashi, T., “Nonlinearity of the Aerody- Parts I–III,” Quarterly of Applied Mathematics, Vol. 45, No. 3, 1987,
namic Characteristics of NACA-0012 Aerofoil at Low Reynolds Num- pp. 561–590.
bers,” Journal of the Japan Society for Aeronautical and Space https://doi.org/10.1090/qam/910462
Sciences, Vol. 55, No. 644, 2007, pp. 439–445. [40] Holmes, P., Lumley, J. L., and Berkooz, G., Turbulence, Coherent
https://doi.org/10.2322/jjsass.55.439 Structures, Dynamical Systems and Symmetry, Cambridge Univ. Press,
[36] Glauert, H., “The Effect of Compressibility on the Lift of an Aerofoil,” New York, 1996.
Proceedings of the Royal Society of London. Series A, Containing
Papers of a Mathematical and Physical Character, Vol. 118, No. 779, P. G. Tucker
March 1928, pp. 113–119. Associate Editor
Downloaded by CARLETON UNIVERSITY on September 12, 2020 | http://arc.aiaa.org | DOI: 10.2514/1.J058905

You might also like