You are on page 1of 28

Journal of Fluids and Structures 58 (2015) 291318

Contents lists available at ScienceDirect

Journal of Fluids and Structures


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

Numerical investigations of the vortex interactions for a ow


over a pitching foil at different stages
Chien-Chou Tseng n, Yu-En Cheng
Department of Mechanical and Electro-Mechanical Engineering, National Sun Yat-Sen University, 70 Lienhai Rd., Kaohsiung 80424,
Taiwan ROC

a r t i c l e i n f o

abstract

Article history:
Received 23 January 2015
Accepted 4 August 2015

The uidstructure interaction is investigated numerically for a two-dimensional ow


(Re 2.5  106) over a sinusoid-pitching foil by the SST (Shear Stress Transport) k model.
Although discrepancies in the downstroke phase, which are also documented in other
numerical studies, are observed by comparing with experimental results, our current
numerical results are sufcient to predict the mean features and qualitative tendencies of
the dynamic stall phenomenon. These discrepancies are evaluated carefully from the
numerical and experimental viewpoints.
In this study, we have utilized , which is the normalized second invariant of the
velocity gradient tensor, to present the evolution of the Leading Edge Vortex (LEV) and
Trailing Edge Vortex (TEV). The convective, pressure, and diffusion terms during the
dynamic stall process are discussed based on the transport equation of . It is found that
the pressure term dominates the rate of the change of the rotation strength inside the LEV.
This trend can hardly be observed directly by using the vorticity transport equation due to
the zero baroclinic term for the incompressible ow.
The mechanisms to delay the stall are categorized based on the formation of the LEV.
At the rst stage before the formation of the LEV in the upper surface, the pitching foil
provides extra momentum into the uid ows to resist the ow separation, and hence the
stall is delayed. At the second stage, a low-pressure area travels with the evolution of the
LEV such that the lift still can be maintained. Three short periods at the second stage
corresponds to different ow patterns during the dynamic stall, and these short periods
can be distinguished according to the trend of the pressure variation inside the LEV. The
lift stall occurs when a reverse ow from the lower surface is triggered during the
shedding of the LEV. For a reduced frequency kf 0.15, the formation of the TEV happens
right after the lift stall, and the lift can drop dramatically. With a faster reduced frequency
kf 0.25, the shedding of the LEV is postponed into the downstroke, and the interaction
between the LEV and TEV becomes weaker correspondingly. Thus, the lift drops more
gently after the stall. In order to acquire more reliable numerical results within the
downstroke phase, the Large Eddy Simulation (LES), which is capable of better predictions
for the laminar-to-turbulent transition and ow reattachment process, will be considered
as the future work.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
Dynamic stall
Pitching foil
Turbulence model
Leading Edge Vortex (LEV)
Trailing Edge Vortex (TEV)
Vortex interaction

Corresponding author. Tel.: 886 7 5252000 4238; fax: 886 7 5252000 4299.
E-mail address: tsengch@mail.nsysu.edu.tw (C.-C. Tseng).

http://dx.doi.org/10.1016/j.juidstructs.2015.08.002
0889-9746/& 2015 Elsevier Ltd. All rights reserved.

292

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Nomenclature
Subscript i, j, k, m Einstein notation [1]
C
Chord length [m]
CL
Lift force coefcient [1]
CN
Normal force coefcient [1]
Cp
Pressure coefcient [1]
F1, F2
Function in the turbulent model [1]
k
Turbulent kinetic energy [m2/s2]
kf
Reduced frequency [1]
P
Pressure [Pa]
Pt
Turbulent production term [kg/m/s3]
Q
Second invariant of velocity gradient tensor
[1/s2]

Normalized form of Q [1]


Re
Reynolds number [1]
Ret
Turbulent Reynolds number [1]
S
Strain rate tensor [1/s]
SC, SP, SV Source terms in transport equation of [1/s]

t
u
ui
U
x
1, *, ,

m, mt
, t

Time [s]
x-direction velocity [m/s]
Velocity in i-direction [m/s]
Free stream velocity [m/s]
Position [m]
*, sk, s , s2 Coefcient in turbulent model
[1]
Angle of attack []
Density [kg/m3]
Laminar and turbulent dynamic viscosity [kg/
m s]
Laminar and turbulent kinematic viscosity
[m2/s]
Specic turbulent dissipation rate [1/s]
Angular velocity [1/s]
Azimuthal angle []
Vorticity [1/s]
Reynolds stress [Pa]

1. Introduction
For a uid ow past a foil with rapid motions, such as the pitching, plunging, and apping, the lift force still can be
maintained even when the angle of attack (AoA) exceeds the normal static stall angle (Ekaterinaris and Platzer, 1997; Wang
et al., 2012). This is so-called dynamic stall. The prediction of the dynamic stall is very important in the aerodynamics of
aircraft, helicopter, wind turbine, and turbomachinery.
For maneuverable ghters and helicopter rotors, the vibration, high load, fatigue, and structural failure can be caused due
to the unsteadiness of the dynamic stall phenomenon (Carr, 1988; Gompertz et al., 2011; Mulleners et al., 2012a). As for
insect ights, the Reynolds number is very low due to their sizes, and hence the lift must be generated by the pitching,
plunging, and apping behaviors. The study of the dynamic stall of insects inspires the development of micro air vehicles
(MAV) (Shyy et al., 2013; Kang et al., 2011). Empirical methods are often used in the corresponding industry during 1970s
without knowing the details of the ow physics (Gormont, 1973; Harris et al., 1970; Wang et al., 2012). However, recent
progress in Computational Fluid Dynamics (CFD) makes it possible to predict the dynamic stall process numerically in the
turbomachinery and helicopter industry (Dawes, 2007; Doerffer and Szulc, 2008; Wang et al., 2012).
The formation of the leading edge vortex (LEV) plays an important role for the dynamic stall. During the convection of
the LEV, the low-pressure area of the LEV provides extra lift force to delay the stall. When the shedding of LEV takes place,
the lift force can drop dramatically. The LEV could interact with the surrounding uid ow to induce multiple recirculation
regions, such as the secondary vortex and trailing edge vortex (TEV) (Ekaterinaris and Platzer, 1997; Leishman, 1990;
McAlister et al., 1978; McCroskey et al., 1976; Wang et al., 2010, 2012). The complicated interactions among the LEV, TEV, and
secondary vortex could cause difculties in numerical predictions and experimental measurements. Recently, the forceelement method is applied to decompose the aerodynamic force into several components. The vorticity almost dominates
the contributions of the lift during the entire stroke, which emphasizes the signicance to study the vortex interactions
during the dynamic stall (Niu and Chang, 2013). The dynamic stall characteristics could depend on the reduced frequency,
the freestream ow condition, the Reynolds number, the Mach number, and the foil shape. These effects will be discussed
hereafter in this section.
McAlister et al. (1978) and McCroskey et al. (1976) have experimentally investigated the dynamic stall with different
reduced frequencies (kf 0.05, 0.15 and, 0.25) at Reynolds number 2.5  106 for a NACA0012 foil. As kf 0.05, the shedding of
the LEV and the secondary vortex both occur in the up-stroke, which correspond to two peaks of the lift force during the upstroke. When kf increases to 0.15, the shedding of the secondary vortex can be postponed into the down-stroke. The
dynamic stall due to the shedding of the LEV typically still happens when AoA approaches its maximum amplitude in the
up-stroke. For kf 0.25, even the dynamic stall can be further delayed into the down-stroke. Recent experiments by Lee and
Gerontakos (2004), and Sharma and Poddar (2013) also display the same trends. Similarly, Leishman (1990) has found out
that the increasing reduced frequency could delay the onset of the ow separation and dynamic stall to a higher angle for a
NACA23012 foil at Re1.5  106. Only small values of reduced frequency are required to signicantly delay the dynamic stall
since the ow separation does not have time to develop at this high Reynolds number. Further interactions between the
maximum lift and the reduced frequency will be discussed in Section 8.
Gharali and Johnson (2013) have investigated the phase difference between the freestream velocity oscillation and
pitching pattern oscillation numerically at Reynolds number 105 for a NACA0012 foil. For in-phase oscillations, the lift

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

293

could become 5 times larger than that under the static freestream, and this value could be amplied if the reduced frequency increases. Based on the experimental study of Leu et al. (2012), the unsteady ow elds above NACA0015 foil
(kf 0.09 as Re4  103) pitching with and without upstream turbulence generator have been investigated. They have
observed that higher turbulence intensity (6.9%) of the freestream velocity could delay the dynamic stall signicantly.
The effects of the Reynolds number and foil shape have also been documented for static and dynamic stall conditions.
Loftin and Hamilton (1945) have experimentally studied the aerodynamic characteristics for 10 NACA 6-series foils from
Reynolds number 7  105 to 9  106 under static conditions. The designated minimum drag coefcient increases as the Reynolds number is lowered, and the magnitude of this increase becomes larger with increasing foil thickness. Althaus and
Wortmann (1981), and Leishman (1990) have experimentally investigated the correlation between the maximum static lift
coefcient and Reynolds number for a NACA23012 foil at high Reynolds numbers as O(106). From Re8  105 to 2  106, the
maximum static lift coefcient increases about 10%. Although the stall angle remains around 15, the lift force drops more
rapidly as the Reynolds number goes up, which has also been documented by Loftin and Smith (1949).
For the foil under prescribed motions, the Reynolds number effects of the NREL S809 and S813 foil have been conducted
experimentally at low reduced frequency as kf0.06. For a light-stall motion, the Reynolds number dependency remains
weak for both foils. Under a deep-stall motion, a larger hysteresis and larger values of the peak lift coefcients can be
observed for a smaller Reynolds number (decreases from 1.25 to 0.76  106). The differences are more noticeable for the S813
foil due to its larger maximum thickness (Ramsay et al., 1995; Ramsay and Gregorek 1999). Similar results have also been
observed by Choudhry et al. (2014). They have showed that the maximum lift and stall angle depend on the Reynolds
number (105) as kf above 0.012 for a NACA0021 foil. On the other hand, Leishman (1990) has conducted experiments for
much higher reduced frequency as kf above 0.1 for a NACA23012 foil under high Reynolds numbers (4106). Leishman (1990)
has shown that the loop of entire stroke is more hysteresis when the mean pitching angle and the amplitude of oscillation
become larger. He has also pointed out that the dynamic stall characteristics are only weakly dependent on the Reynolds
number for a NACA23012 foil, which is also similar as results documented by Kang et al. (2012) and McCroskey et al. (1976)
when the Reynolds number goes up. Therefore, one should note that the Reynolds number effect could also depend on the
variation range of the Reynolds number, the conducted reduced frequency, and even the foil geometry (Choudhry et al.,
2014).
Recently, Visbal (2014) has used the Large Eddy Simulation (LES) for a constant pitching rate foil (kf 0.05) at Re5  105.
For a SD7003 foil, the laminar separation bubble (LSB) near the leading edge breaks down as the angle of the attack
increases, and the sudden collapse of the leading edge suction occurs. For a reduced Reynolds number as 2  105, the LSB
becomes larger but breaks down earlier, resulting in a less abrupt leading edge suction collapse. Since the geometry of
SD07003 foil is shaper, the thicker NACA0012 foil exhibits lower suction and a less abrupt collapse.
For a moderate Reynolds number as O(104), Kang et al. (2012) have used SST (Shear Stress Transport) k model to
compare the dynamic stall phenomenon between a SD7003 foil and a at plate as kf 0.25. The leading edge separation of a
SD7003 foil is more susceptible to the Reynolds number due to its more streamlined body. Consequently, the Reynolds
number effect of a SD7003 foil is more important than that of a at plate. However, for a more aggressive prescribed motion,
the ow still could separate near the leading edge for a SD7003 foil, and hence the shape effect becomes less important
under the deep-stall motion. Overall, the effects of the foil shape and Reynolds number decline as the Reynolds number goes
up. As for a low Reynolds number as O(102), the viscosity plays a dominated role such that the ow structures are very
sensitive to the variations in the Reynolds number (Shyy et al. 2007).
Besides, the Mach number effects could also play an important role (Ekaterinaris and Platzer, 1997). McCroskey et al.
(1981) have indicated that the pitching of the airfoil can induce high local Mach number even the freestream Mach number
is very low. For Mach number around 0.2, the sonic conditions already could be observed near the leading edge. Similarly,
Visbal (1990) has utilized CFD and observed that the supersonic ow region starts to grow as the angle of the attack
increases (Re106, Mach number 0.6, kf 0.0225 for a NACA0015 foil). Due to the severe shock-boundary-layer interaction,
the separation near the shock location starts to spread both upstream and downstream. Chandrasekhara and Carr (1990)
have pointed out that the dynamic stall vortex always forms near the leading edge and convects at about 0.3 times the
freestream velocity for all the tested Mach number and reduced frequency. Similar as documented in McCroskey et al.
(1976), the dynamic stall at a given Mach number can be delayed with a higher reduced frequency. On the other hand, it
could occur at a lower angle of attack as the Mach number increases at the same reduced frequency. Recently, Gardner et al.
(2013) have found out that the peak of pitching moment coefcient is changed linearly with reduced frequency and
inversely with the Mach number as Mach number is varied between 0.3 and 0.5.
As for the uncertainty due to the unsteady ow motions, Mulleners et al. (2012b) have investigated the dynamics of the
static stall experimentally. When the AoA exceeds the stall angle, the uctuation of lift could become very signicant
between each realization. This implies the ow is under the transition from the steady to unsteady ow regime. Similarly,
the uctuation of lift can also be very signicant during the dynamic stall process between each pitching cycle (Gardner
et al., 2013; McAlister et al., 1978,1982; McCroskey et al., 1976,1982; Zanotti and Gibertini, 2012). Zanotti et al. (2013) have
also pointed out that the force evaluation for a deep-stall pitching foil is subjected to a large uncertainty. The uctuation and
uncertainty could be related to the severe unsteady and highly nonlinear post-stall ow structures, which will be discussed
later.
The simulation results of this study are mainly based on the two-dimensional ow past a sinusoid-pitching NACA0012
foil at Reynolds number 2.5  106 and kf 0.15 (McAlister et al., 1978; McCroskey et al., 1976). The lift slope rst varies

294

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

linearly at small angle of attacks. The linear regime could somehow exceed the static angle of attack. Then the lift curve
slope starts to decrease and a slight curvature is observed. As the LEV forms, the lift slope suddenly increases due to the
extra lift during the shedding of the LEV. Finally, the LEV reaches the trailing edge slightly before the maximum angle of
attack, and then the dynamic stall occurs (Leishman, 1990; McAlister et al., 1978; McCroskey et al. 1976). The pre- and poststalls regimes can be dened based on this dynamic stall angle (Choudhry et al., 2014; McAlister et al., 1978; Mulleners et al.,
2012b).
In this study, instead of using the pre- and post-stalls denition, the mechanisms to delay the stall will be categorized
into two stages depending on the formation of the LEV. The objective of this study is to utilize SST k model to analyze the
different mechanisms to delay the stall at these two stages. Our numerical results are compared with several different
experimental measurements, and it shows that the current CFD results are sufcient to predict the mean features and
qualitative tendencies of the dynamic stall phenomenon in the upstroke phase and slightly after the stall during the
downstroke process. Therefore, we have conned our analysis within this range of AoA mostly. After the validation, the rst
stage before the formation of the LEV will be compared with the static condition, and the vortex interaction will be
emphasized in the second stage. Furthermore, the ow structures are evaluated carefully at different angles and reduced
frequencies to study their interactions with the lift force.

2. Numerical approaches
Reynolds-Averaged Navier-Stokes (RANS) model, including k model (Bardina et al., 1997; Jones and Launder, 1972;
Launder and Spalding, 1974; Shyy et al., 2005) and k model (Menter, 1992, 1994; Shyy et al., 2005), is used widely due to
its balance between the accuracy and computational cost. In RANS methodology, the ow variables are considered as
summations of its ensemble average value and a uctuation value (Launder and Spalding, 1974; Shyy et al., 2005). For
convenience sake, the overbar which stands for average will be dropped hereafter.
The continuity and momentum equations are listed below:

Continuity:

(ui )
=0
xi

(1)

u
(ui uj )
P

=
+
(ij + 2Sij )
Momentum: i +
xj
xi
xj
t
Sij =

uj
1 ui

, ij = ui uj
+
xi
2 xj

(2)

The ow conditions in this study are all incompressible with given density . In Eq. (1), x is the coordinate, t is time, u is
the velocity, and subscripts i and j stand for Einstein index. In Eq. (2), P is the pressure, is the uid viscosity, Sij is the strain
rate, and ij is the so-called Reynolds stress. This nonlinear term in the current study is modeled by the SST k turbulence
model (ANSYS, 2015; Menter, 1992, 1994; Menter et al., 2003):

SST k turbulence model



(uj k )

k
k +
= min (Pt , 10*k) *k +
( + k t )

t
x
xj
xj

production
dissipation

diffusion


(uj )

= 1 Pt 2 +
( + t )
+ 2 (1 F1) 2
+

xj xj
t
x
x
x

j
j
j
t

dissipation

production

cross diffusion
diffusion

Pt = ij ui , ij = 2t Sij 2 kij
3
xj

1
= k
1 (Sij Sij )1/2 F2
t

max * ,

0.31

(3)

In order to close the turbulence modeling of the Reynolds stress, for SST k model, the turbulent kinetic energy (TKE) k
and the specic turbulent dissipation rate are modeled as transport equations with production, dissipation, and diffusion
terms. In Eq. (3), Pt is the production term of k, which comes from interactions between mean and Reynolds stress. By
Boussinesqs hypothesis, the Reynolds stress ij can be expressed by the mean ow strain rate Sij and turbulent eddy
viscosity t. The production term of k equation and the Boussinesqs hypothesis for t are modied to limit eddy viscosity t
in high stress regions. The concept of SST k model is to reduce the excess eddy viscosity such that the large periodic
motion still can be captured.

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

295

The blending function F1 in equation depends on the local turbulent quantities and switches SST k model into k
model at regions far away from the solid wall boundary (F1 0). The purpose is to reduce the sensitivity issue of inlet
turbulent quantities from k model. On the other hand, in order to resolve the ow eld inside the boundary layer, SST k
model preserves the essence of k model in the near wall region (F1 1). The wall function treatment in the current study
is inactivated since the local Reynolds number (y ) of the rst grid away from the wall is smaller than 5 (Menter, 1992,
1994; Wang et al., 2010). The other blending function F2 in Boussinesq's hypothesis for t also has similar characteristic.
Furthermore, s2 is a constant 1.168, and the values of other coefcients in Eq. (3), such as 1, *, , *, sk, and s, depend
on the local turbulence strength. The low Reynolds number modication of the coefcient * limits the excess eddy viscosity
and predicts the laminar-to-turbulent transition based on the turbulent Reynolds number Ret (Menter et al., 2003,2006;
Wang et al., 2010):

0.024 + Ret
k
* =
, Ret =
1 + Ret

(4)

For a fully-turbulent SST model, * is a constant equal to 1.


The algorithm to solve this set of governing equations in this study is PISO (Pressure Implicit with Split Operator) (Issa,
1982). When the compressibility becomes important as Mach number is greater than 0.3, the AUSMD (Advection Upwind
Splitting Method based on ux Difference) scheme is suggested to modify the convective nonlinear terms by the local Mach
number (Niu and Liou, 1999). Since the ow conditions of current problems are all incompressible, QUICK (Quadratic
Upstream Interpolation for Convective Kinetics) (Patankar, 1980) is sufcient to discretize convective nonlinear terms, and
while the central difference is used for diffusion terms.

3. Simulation setups, computational domains, and boundary conditions


Fig. 1(a) shows the representative two-dimensional computational domains. As for the boundary conditions, the velocity
is given while pressure is extrapolated at the inlet. The turbulent quantities are also specied at inlet based on the eddy-tolaminar viscosity ratio. For the outlet, pressure is given, and other ow variables are extrapolated. On the walls, pressure is
extrapolated, k is zero due to no-slip boundary condition, and is specied based on near wall region treatments (Menter,
1992, 1994; Shyy et al., 2005; Wang et al., 2010,2012).
The computational domain is separated by an interface between a sub-domain near the foil and a dynamic mesh domain
away from the foil. When the pitching foil oscillates as the prescribed sine motion, the mesh will be reconstructed at each
numerical time step. The structured meshes inside the sub-domain move with the foil like rigid body motion without
deformation while the unstructured dynamic mesh away from the foil could deform.
In this study, an incompressible (Mach number 0.1) two-dimensional ow with Reynolds number 2.5  106 over an
oscillatory NACA0012 foil is simulated based on experiment setups (McAlister et al., 1978; McCroskey et al., 1976). The

Fig. 1. The schematic shows (a) the representative two-dimensional computational domains, boundary conditions, and the grid layout inside the sub-grid
region, and (b) the grid sensitivity test based on the experimental studies by McAlister et al. (1978) and McCroskey et al. (1976) (NACA0012 foil, Re2  106,
and kf 0.15).

296

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Reynolds number is dened as UC/ where and are the density and viscosity of air, respectively, C is the chord length of
the foil (0.31 m), and U is the free stream velocity. The prescribed pitching motion is given as a time-varying sine curve
15 10sin(f  t) where is the AoA, and f is the angular velocity. In this study, the angular velocity is obtained
according to the reduced frequency kf, which is dened as (fC)/2U. Two reduced frequencies are setup as 0.15 and 0.25
respectively. All the ow conditions and geometries are identical to those described in the experiment (McAlister et al.,
1978; McCroskey et al., 1976). The representative structured mesh near the pitching foil is highlighted in Fig. 1(a). The
numerical time step is given according to the period of a pitching cycle divided by 200. Fig. 1(b) highlights the grid sensitivity test by the lift curves over the entire stroke as kf 0.15 (NACA0012 foil, Re 2.5  106). The number of grid points of
G1, G2, G3, and G4 corresponds to 105, 1.2  105, 1.6  105, and 1.8  105. Basically, the increase of the number of grid points
concentrates inside the sub-domain near the foil. From Fig. 1(b), the results between G3 and G4 are very consistent with
only about 0.10.2 phase difference in the downstroke. As a result, the grid layout of G3 will be utilized in this study
hereafter. More detail validations between numerical and experimental results will be discussed in next section.

4. Validations and analysis of the discrepancies between CFD and experimental results
Before further analysis according to experimental setups of McAlister et al. (1978) and McCroskey et al. (1976), we
provide two more cases for further validations between our current numerical framework and the experimental results.
The rst case is the experimental study of Leu et al. (2012). The prescribed pitching motion of a NACA0015 foil is
30tan  1 (sin (f t ) / cos (f t + )), f 0.82 rad/s, kf 0.09, the tip speed ratio 2, and the Reynolds number is 4.5  103.
The number of grid points is 67 000. Fig. 2(a) shows validations between the current simulation results (solid lines) and
experimental measurements (dash lines) based on the x-velocity (u/U) at cross-section locations as x/C 25%, 50%, 75%, and
100%. The qualitative trends can be captured by our numerical predictions while the reverse ow sometimes seems to be
over-predicted by our current simulation results, especially at x/C 100%. Fig. 2(b) shows the comparisons of vorticity. The
LEV is divided into two parts by the TEV at 23.3, and this phenomenon can also be detected in the experimental results

Fig. 2. The schematic shows validations between the current simulation results and experimental measurements (NACA0012 foil, Re 4.5  103, and
kf 0.09). (a) Validations of the x-velocity u/U (current simulation results: solid lines; experimental measurements: dash lines) (b) Validations of the
vorticity n.

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

297

Fig. 3. The schematic shows validations between the current simulation results (solid lines) and experimental measurements (McCroskey, 1981) based on
Cp along the entire foil surface (NACA0012 foil and Re 4  106).

as 22in Fig. 2(b). Although approximate 1 phase difference can be observed in Fig. 2(b), the evolutions of the LEV and
TEV are mostly consistent (Tseng and Hu, 2015).
The second case is the experimental report of McCroskey (1981). A NACA0012 foil is prescribed a sinusoidal-pitching
motion as 12 2sin(0.2t) at Re 4  106, and Cp along entire foil surface is compared in Fig. 3 (the number of grid points
is about 105). The pressure coefcient Cp is dened as the difference between local and freestream pressure normalized by
the freestream dynamic pressure. One can see that the numerical prediction can almost match the experimental data during
the upstroke process. However, certain discrepancy occurs during the downstroke process. The suction peak near the
leading edge is overestimated by our numerical predictions. Nevertheless, our results are qualitatively consistent with the
experimental measurements of Leu et al. (2012) and McCroskey (1981) in terms of the velocity, evolution of vortex, and
surface pressure.
Fig. 4(a) and (b) compares the normal force coefcient CN between our current numerical results and experimental
measurements (McAlister et al., 1978; McCroskey et al., 1976), which is the case of interest of this study. Similar as the lift
coefcient CL normalizes the force perpendicular to the streamwise direction by the freestream dynamic pressure, the
normal force coefcient CN normalizes the force perpendicular to the chord of the foil. The red arrows and intervals
highlight the stall and evolution of the secondary vortex predicted by our current simulation. The wording evolution
includes the formation, convection, and shedding. In Fig. 4(a) for the static condition, our numerical results can match the
experimental results exactly at 5, 10, 15, and 17. As for the dynamic condition with kf 0.15 and kf 0.25 illustrated in
Fig. 4(a) and (b) respectively, the numerical result and experimental measurement are comparable with consistent trends.
The maximum lift and its corresponding dynamic stall angle are well captured by our numerical prediction as listed in
Table 1. As reported in the experiment, the stall is delayed into the down-stroke as kf 0.25, which is also presented in our
simulation. The instantaneous lift is predicted well by the current numerical results in the up-stroke. However, for kf 0.15
in Fig. 4(a), the numerical result of the down-stroke from 24 to 15 deviates from the experimental measurement signicantly. Similar trends also occur for kf 0.25 in Fig. 4(b) from 22 to 10. As depicted in Fig. 4, the discrepancy is highly
correlated to the evolution of the secondary vortex. Because the secondary vortex of kf 0.25 is further delayed than that of

298

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Fig. 4. Validations between numerical and experimental results. The experimental curves in (a) and (b) are based on the experimental studies by
McCroskey et al. (1976) and McAlister et al. (1978). (c) It shows signicant error bar during the down-stroke, which is adapted from Zanotti and Gibertini
(2012). (For interpretation of the references to color in this gure, the reader is referred to the web version of this article.) (a) Static and dynamic condition
as kf =0.15 (NACA0012 foil, Re =2.5106) (b) Dynamic condition as kf =0.25 (NACA0012 foil, Re =2.5106) (c) Dynamic condition as kf =0.1 from Ref. of Zanotti
and Gibertini (2012).

Table 1
Comparisons between numerical prediction and experimental data with respect to the maximum CN and stall angle.

kf 0.15
kf 0.25

Maximum CN (CFD/exp.)

Stall angle (CFD/exp.)

2.91/3.05
2.83/2.92

24.7/24.9
24.5/23.8

kf 0.15, it can be seen that the discrepancies in the down-stroke of kf 0.25 in Fig. 4(b) occur later than those of kf 0.15 in
Fig. 4(a).
As discussed in Eq. (4), it deals with the laminar-to-turbulent transition and reduces the accumulation of the eddy
viscosity near the stagnation point. For * 1, it can be regarded as the fully-turbulent SST model. For the fully-turbulent SST
model, Kang et al. (2012) have utilized this model to analyze the light- and deep-stall motion as Re104. It is found that the
numerical predictions are in good qualitative agreement with the experimental PIV data in terms of mean ow elds.
However, the computations are unable to predict the reattachment of the separated ow correctly. Martinat et al. (2008)
have used the fully-turbulent SST model, two types of k- models, and SpalartAllmaras (eddy viscosity) model to predict
ows over a deep-stall pitching motion under Re105 and 106. Furthermore, Ahmad et al. (2010) have compared the performances of the fully-turbulent SST model, standard k model, standard k model, realizable k model, and RNG k
model on a deep-stall pitching motion as Re106. Both Ahmad et al. (2010) and Martinat et al. (2008) have concluded that
SST k model can capture the behaviors of the dynamic stall the best among all the selected models. Similar analysis can
also be found in references of Sohail and Ullah (2011) and Velkova et al. (2012).
Ekaterinaris and Menter (1994) have considered the laminar-to-turbulent transition and indicated that the results of
turbulence models could become better after the implement of the transition model, and the SST model performs better
than standard k and k model under the light-stall condition. Wang et al. (2010) have utilized Eq. (4) to account for the
transition in their SST model, and the results can match the experimental measurements better than the standard k
model for a deep-stall pitching motion as Re 3.75  105. Recently, Wang et al. (2012) have used a more complicated Re
transition model. Besides transport equations of k and , the other two transport equations for intermittency and

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

299

momentum thickness Reynolds number Re are added (Menter et al., 2006). For a deep-stall pitching motion as
Re1.35  105, the Re based SST model performs better than that of RNG k model.
Therefore, it is usually accepted that SST k model can capture the behaviors of the dynamic stall the best among all the
categories of RANS models. However, for the deep-stall condition, the discrepancies between numerical results by SST
model and experimental measurements still could be very obvious during the downstroke. During the dynamic stall process, the shedding of LEV can interact strongly with the evolution of the trailing edge vortex (TEV) and other recirculation
regions. Then the so-called secondary vortex is generated under by the mergence and interaction of these multiple recirculation region. As a matter of course, this strong nonlinear nature can cause the numerical difculties in prediction of the
post-stall ow structures, which could result in the discrepancies highly correlated to the secondary vortex. The evidence is
already highlighted in Fig. 4(a) and (b). Similar discrepancies in the downstroke have also been reported in different
numerical studies (Ahmad et al., 2010; Gharali and Johnson, 2013; Liggett, 2012; Martinat et al., 2008; Sohail and Ullah,
2011; Velkona et al., 2012; Wang et al., 2010,,2012).
The other factor causing numerical difculties could be the laminar-to-turbulent transition. Leading-edge transition
could affect the boundary layer development and the ensuing ow separation and reattachment. As a result, the use of fullyturbulent model could limit the occurrence of the laminar separation at the leading edge, leading to an inaccurate turbulent
ow development, as well as the prediction of the evolution of the LEV (Wang et al., 2010). The transition model and its
implement into turbulence models have already been documented in references of Ekaterinaris and Platzer (1997), Hill
et al., 2004, and Menter et al. (2006). Although Martinat et al. (2008) have claimed that the transition effects could be
avoided as Re106, Ekaterinaris and Menter (1994) and Ekaterinaris and Platzer (1997) have shown that this effect still could
be important as Re106. Ekaterinaris and Menter (1994) have assumed that the transition onset occurs immediately
downstream of the suction peak location, and the ow from the stagnation point until the transition onset is computed as
laminar with an eddy viscosity of almost zero. The eddy viscosity computed by SST model after the transition point increases
rapidly. Comparing with the experimental measurements under the light-stall conditions, they have demonstrated that the
numerical prediction by this transition consideration can predict a lower and more reasonable suction peak near the leading
edge than the fully-turbulent SST model. However, the quantitative agreement still could become poor during the downstroke with a much deeper stall predicted (Ekaterinaris and Menter, 1994; Ekaterinaris and Platzer, 1997; Hill et al., 2004;
Wang et al., 2012).
As pointed out by Kang et al. (2012), the SST k model and other RANS categories could underpredict the Reynolds
stresses in the detached shear layer. Consequently, it affects the prediction of the laminar-to-turbulent transition and leads
to the delayed recovery of the ow reattachment compared with experimental results. Visbal (2014) has also made similar
statements. RANS may display signicant deciencies regarding predictions for the laminar-to-turbulent transition, reattachment, and LSB, which could be partially corrected by the empirical transition model. Although the development of the
transition model will be continued due to the design purpose, the complex ow physics near the leading edge cannot be
truly predicted by this methodology. Visbal (2014) has shown that the LES computation with suitable high order numerical
scheme can capture the dynamics of the laminar-to-turbulent transition, ow reattachment, and formation of the LSB. The
discrepancies of our numerical results in Figs. 24 could also come from the decient predictions of the RANS model
regarding these details.
Nevertheless, the SST k model is still capable of qualitative prediction for the dynamic stall behaviors, and hence we
use SST k model in the current study. Moreover, we utilize the low Reynolds number modication as shown in Eq. (4) to
account for certain laminar-to-turbulent transition effect. As one can see in Fig. 4(a) and (b), our numerical predictions are in
good qualitative agreement with the experimental results in the upstroke phase and slightly after the stall during the
downstroke, and the signicant oscillations and discrepancies occur within the large portion of the downstroke process.
Consequently, at the current stage, we mostly conne our ow structure analysis in this study to the upstroke phase and
slightly after the stall during the downstroke process. As discussed in the last paragraph, in order to extend our results into
downstroke phase in the future, the LES computation will be more suitable than the RANS model.
The discrepancies may also come from the three-dimensional effect (Choudhry et al., 2014; Ekaterinaris and Platzer,
1997). The three-dimensional effect is investigated numerically by Martinat et al. (2008) and Wang et al. (2012), and this
effect can come into play even in two-dimensional based experiments (Raffel et al., 1995; Wernert et al., 1996). For deepstall conditions, when AoA is close to its maximum amplitude or during the evolution of the secondary vortex in the downstroke process, the three-dimensional effect can become important (Martinat et al., 2008; Wernert et al., 1996). Martinat
et al. (2008) have found out that the coherent structure predicted numerically by three-dimensional computation is weaker
than that by two-dimensional computation. The uctuation of lift in the down-stroke becomes less signicant in threedimensional numerical studies. However, the discrepancy between CFD and experimental data in down-stroke is only
improved slightly (Martinat et al., 2008; Wang et al., 2012). Besides, the turbulence is essentially three-dimensional phenomenon. Therefore, the three-dimensional effect and its interactions with turbulence are worth further investigation.
However, two-dimensional studies are able to capture a signicant part of dynamic stall process and remain mainstream in
this led.
Experimentally, the measured normal force coefcient CN is calculated by averaging over the total number of cycles. At
any given cycle, the differences from the average value are most likely to occur, which is quantied as the standard
deviation. The high standard deviation in the experimental measurements can be attributed to the strong unsteadiness of
the secondary vortex after the lift stall (Gardner et al., 2013; McAlister et al., 1978,1982; McCroskey et al., 1982; Zanotti and

300

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Gibertini, 2012). The detail of standard deviation during the dynamic stall process is reported quantitatively in the recent
experiment by Gardner et al. (2013) and Zanotti and Gibertini (2012). The ow conditions in the experiment done by Zanotti
and Gibertini (2012) are more similar to our current case, and this experimental study also reports similar experimental
average CN curve as shown in Fig. 4(a) by McCroskey et al (1976) and McAlister et al (1978). The experimental result of
Zanotti and Gibertini (2012) is illustrated in Fig. 4(c). The ow condition is based on Mach number 0.15 and Reynolds
number1.5  106 over an oscillatory NACA23012 foil by a time-varying sine curve 10 10sin(f  t) with reduced
frequency kf 0.1. The error bar over entire cycle shows low standard deviation in the up-stroke. The error bar over entire
cycle shows low standard deviation in the up-stroke. In down-stroke process from 20 to 8, the high error bar indicates
signicant standard deviation during this period. For instance, at 15.4, the average lift is 1.11 and varies between 1.36 and
0.87. Although the tested Mach number is higher (0.30.5) by Gardner et al. (2013), the high standard deviation is still
observed during the down-stroke process.
To sum up, the post-stall ow is complicated with high unsteadiness, which results in the discrepancy in the downstroke
process. In the future, more efforts should be added by utilizing the LES model, which can better capture the complex ow
physics near the leading edge. Nevertheless, by comparing our CFD results with other experimental measurements as shown
in Figs. 24, our results are sufcient to predict the mean features and qualitative tendencies of the dynamic stall phenomenon. Due to the limitation of the RANS model in highly separated ows, we will conne our ow structure analysis
mostly in upstroke phase and slightly after the stall during the downstroke process.

5. The ow structure and its mechanism to delay the stall at the rst stage for the dynamic condition (kf 0.15)
In Fig. 4(a), for the dynamic condition as kf 0.15, CN is still increasing after static stall angle 15. At this point, the ow
remains attached, and the LEV has not formed until 22. This period from 15 to 22will be categorized as the rst stage
in this study.
The characteristic of the lift curve has been well documented in terms of the angle of attack under the static condition
(Althaus and Wortmann, 1981; Choudhry et al., 2014; Leishman, 1990; Loftin and Smith, 1949; Mulleners et al., 2012b).
However, the detail ow structures after the static stall have received relatively little attention. At 17, The lift coefcient of
dynamic condition (CN 1.77 as kf 0.15) is much higher than that of static condition (CN 1.18). The goal of this section is
to investigate and compare the ow structures at 17under static and dynamic conditions, and hence the mechanism to
delay the stall before the formation of the LEV can be understood.
5.1. Flow structures at the rst stage for the dynamic condition at 17
Figs. 5 and 6 illustrate the elds of the pressure coefcient Cp and representative streamlines near the foil at 17for both
static and dynamic condition (kf 0.15) respectively, and Fig. 7 depicts Cp along the entire foil surface for both cases. In
Figs. 5(a) and 6(a) around the leading edge, the black dots in the lower surface represent the stagnation point that separates
ow toward upper and lower surface. The maximum pressure in the ow eld occurs at this stagnation point, which
corresponds to the location of the maximum Cp in the lower surface as shown in Fig. 7. In Fig. 7, not only the value of
maximum Cp but also their corresponding locations are identical for both the static and dynamic conditions. As for the
minimum Cp in the upper surface, their locations are still identical as shown in Fig. 7 However, the minimum Cp can almost
reach 12 for the dynamic condition, and while it is only 8 for the static condition. Thus the pressure in the upper surface
near the leading edge is apparently lower for the dynamic condition as shown in Fig. 6(a) than that of static condition in
Fig. 5(a).

Fig. 5. Cp contours and representative streamlines for the static condition at 17. (a) Near the leading edge (b) The entire foil

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

301

Fig. 6. Cp contours and representative streamlines for the dynamic condition as kf 0.15 at 17. (a) Near the leading edge (b) The entire foil

Fig. 7. Cp along the entire foil surface at 17for both static and dynamic condition (kf 0.15).

Fig. 8. The locations of cross sections to sample the streamwise velocity.

Therefore, the distance of the acceleration path (from maximum to minimum Cp location around the leading edge shown
in Fig. 8) and the value of maximum Cp in the lower surface are the same. However, the value of the minimum Cp in the
upper surface is lower for the dynamic condition. This phenomenon can induce stronger ow acceleration near the leading
edge for the dynamic condition. Since the leading edge is rotating clockwise during the up-stroke, a movement component
same as the direction of the steamwise velocity can be decoupled and result in additional momentum into the uid ow.
Thus the stronger acceleration is acquired for the dynamic condition.
The streamwise velocities are sampled along 12 different vertical cross sections as shown in Fig. 8 and the results are
illustrated in Fig. 9. In Fig. 7, from the slope of Cp of the upper surface, the favorite pressure gradient only occurs very near
the leading edge (from the leading edge to the minimum Cp location), and the adverse pressure gradient (from the minimum
Cp location to the trailing edge) takes place at a signicant part of the entire upper surface.

302

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Fig. 9. The streamwise velocity u/U at 17along the sampled cross sections. The solid line represents velocity for the static condition, and the dash lines
represents for the dynamic condition. In each gure, the upper proles correspond to the velocity in the upper surface, and vice versa for the lower velocity
prole. The vertical distribution in between indicates the interior domain of the foil (a) Cross section 1 (b) Cross section 2 (c) Cross section 3 (d) Cross
section 4 (e) Cross section 5 (f) Cross section 6 (g) Cross section 7 (h) Cross section 8 (i) Cross section 9 (j) Cross section 10 (k) Cross section 11..

By considering the effective AoA, the dynamic condition actually experiences a higher AoA and has faster incident
velocity than those under the static condition. Thus the adverse pressure gradient in the upper surface is stronger for the
dynamic condition. This can be seen from the steeper slope of Cp in the upper surface for the dynamic condition in Fig. 7.
Since the acceleration near the leading edge is stronger for the dynamic condition, its velocity distribution is faster as
shown in Figs. 9(a) of cross section 1 and (b) of cross section 2. The resulting stronger inertial inside the boundary layer
provides better resistance against the adverse pressure gradient in the upper surface for the dynamic condition. Although its
adverse pressure gradient is also stronger as described in the previous paragraph, the stronger inertial overwhelms such
that the streamwise velocity in the upper surface is always positive in Fig. 9. Accordingly, the streamlines remain attached as
shown in Fig. 6(b) for the dynamic condition. On the contrary, for the static condition in the upper surface, the streamwise
velocity near the solid boundary becomes negative from Fig. 9(f)(k), indicating that the ow separation has already taken
place as shown in Fig. 5(b).
As discussed in Fig. 7, the adverse pressure gradient of the static condition in the upper surface is weaker. Therefore, the
free stream of the static condition encounters a weaker deceleration. The free stream velocity of the static condition in the
upper surface is eventually faster than that of the dynamic condition after cross Section 7 in Fig. 9(g). In Fig. 9(k) at the
trailing edge, the free stream velocity in the upper surface of the static condition is about 20% faster than that of the dynamic
condition. In order to meet the faster free stream velocity at the trailing edge, the lower surface of the static condition has to
experience a stronger favorite pressure gradient to accelerate the ow, which can be seen from its steeper slope of Cp of the
lower surface in Fig. 7. Thus, the velocity in the lower surface for the static condition can be faster than that for the dynamic
condition, which is shown in Fig. 9 through entire cross sections.

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

303

5.2. The mechanism to delay the stall at the rst stage


As discussed in Section 5.1, in the upper surface, the minimum Cp value of the dynamic condition is lower. Although the
steeper slope of its stronger adverse pressure gradient grants the dynamic condition a more signicant increasing trend, it
still ends up with an overall lower pressure distribution in the upper surface for the dynamic condition in Fig. 7. In the lower
surface, the maximum Cp values for both conditions are identical. Then the gentler decreasing trend of the weaker favorite
pressure gradient grants an overall higher pressure distribution in the lower surface for the dynamic condition in Fig. 7.
Consequently, the lift coefcient of dynamic condition (CN 1.77 as kf 0.15) is much higher than that of static condition
(CN 1.18) such that the stall is delayed even before the formation of LEV.
For the static condition, the stall occurs because the ow is separated as shown in Fig. 5(b). Usually, a low-pressure
region will exist to maintain the centrifugal force if the vorticity is high enough inside the separation region. However, from
Fig. 9(f) to (k), the streamwise velocity inside the recirculation region is around 0.1to  0.2U, and only a very low vorticity
value can be obtained. Consequently, in Fig. 5(a), there is no low-pressure structure inside this closed streamline region to
provide extra lift.
In this stage before the LEV appears, the pressure distribution in the upper surface drops as AoA increases from 14.7to
20.5in Fig. 10. The lower minimum Cp value near the leading edge is observed to provide enough acceleration to maintain
the ow attached at an even higher AoA, and hence the lift force remains increasing correspondingly in Fig. 4(a). In Fig. 11, it
can be seen that the size of the low-pressure area near the leading edge will be enlarged during this short period. The
further increment in size will activate the formation of the LEV, which can be categorized as the second stage in the next
section.

6. The ow structure and its mechanism to delay the stall at the second stage (kf 0.15)
During the rst stage, the ow remains attached and the lift is increasing due to the mechanisms shown in Figs. 10 and
11. At the second stage, the LEV appears around 22as kf 0.15 such that the mechanism to maintain the lift will be totally
different. The representative streamlines in Figs. 12 and 13 indicate the evolution of the LEV and TEV. Besides, vorticity and
(Davidson, 2004; Haller, 2005; Jeong and Hussain, 1995) dened in Eqs. (5) and (6) are used to describe the rotation
strengths inside the recirculation regions:

Fig. 10. Cp along the solid boundary for the dynamic condition (kf 0.15) as 14.720.5.

Fig. 11. Cp contours near the leading edge for the dynamic condition (kf 0.15) at 14.7and 20.5.

304

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Fig. 12. The ow structures by streamlines and vorticity for the dynamic condition (kf 0.15) from 20.5to 24.7.

Fig. 13. The ow structures by streamlines and for the dynamic condition (kf 0.15) from 20.5to 24.7. Values above  0.2 is cutoff after gure for
22.7.

uj
ui
ijk
k =

xj
xi

(5)

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

1 1
Q k = k k Sij Sij

22

1
S
S

ij ij

2 k k
k =
1
S
S
+

ij
ij

2 k k

305

(6)

The magnitude of indicates the angular velocity of the uid, which is often used to describe the strength of rotation
motion. The positive value of corresponds to the counter clockwise rotation, and vice versa for the negative value. The
notation ijk in Eq. (5) is the Levi-Civita symbol. In Eq. (6), the nominator of is  2Q. Q is the second invariant of the velocity
gradient tensor (Davidson, 2004; Haller, 2005; Jeong and Hussain, 1995). It is further normalized by the sum of the magnitude of the vorticity and strain rate, which is dened as . The value of can only vary between  1 and 1 (Jeong and
Hussain, 1995). The more negative value of reveals that the ow is more dominated by vorticity rather than strain. Under
the two-dimensional computational at the current study, only z component of and can exist.

6.1. The strength of the rotation motion


From the representative streamlines in Figs. 12 and 13, the ow remains attached at 20.5. At 22.7, the streamlines
have already formed a closed recirculation region, which corresponds to the formation of the LEV. The LEV is then detached
from the leading edge, and the multiple recirculation regions are generated near the leading edge at 23.6. As LEV is near

Fig. 14. Flow structures of Cp, u/U, , and streamline from 20.5to 22.7(kf 0.15). The legend for each contour is shown as below.

306

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

the trailing edge at 24.7, the lift reaches its maximum as shown in Fig. 4(a). From 24.9to 24.7, the TEV has already
formed, and then the LEV is dissipated gradually.
However, in Fig. 12, is very high at 20.5even though the ow is still attached. Near the boundary layer of the upper
surface, the gradient of x-velocity u is very high along the y direction, and this large amount of strain rate results in a high
value of vorticity at region without really having a vortex. From 22.7to 24.7, in spite of the fact that the outlines of the
closed streamlines of the LEV become less slender in streamwise direction to favor the rotation motion, the magnitude of
the vorticity within LEV keeps decreasing during this period.
In Fig. 13, is used instead. At 20.5, we have found that is between 0 and  0.2 near the boundary layer as denoted by
the arrow. Since no vortex is supposed to be identied at this instant,  0.2 is dened as a threshold, and all the values
above  0.2 will be cutoff hereafter. This cutoff approach will fail in Fig. 12 since the value of near the boundary at 20.5is
even stronger than that inside the LEV. From 22.7to 24.7, the outlines of the closed streamlines imply that the ow
structures become favorable to the rotation motion. This phenomenon can be captured by since its value inside the LEV
region drops from  0.3 to 0.9, revealing that the strength of the rotation motion becomes more dominated. During
24.9to 24.7, the TEV has already formed, and the streamline structure near the LEV region becomes more slender, which
disfavors the rotation motion. Consequently, inside the LEV region increases to a less negative value, which corresponds to
the dissipation of the LEV.
In Fig. 12, a region without really having a vortex could possibly identied by vorticity , and the value of is more
consistent to the strength of the rotation motion. Therefore, will be used to identify the rotation strength hereafter.
6.2. Details of the ow structures and the mechanism to delay the stall at the second stage for the dynamic condition (kf 0.15)
As shown in Fig. 13 for kf 0.15, the ow structures change rapidly after the formation of the LEV. In this second stage,
details will be analyzed according to three short periods (i) 21.722.7, (ii) 22.724.7, and (iii) 24.724.7. Each period
has different ow patterns. In the rst short period, the pressure inside the LEV keeps decreasing and remains at the same

Fig. 15. Cp along the solid boundary from 20.5to 22.7(kf 0.15). The left gure (a) presents Cp for entire upper and lower surface, and the right
gure (b) highlights Cp near LEV in the upper surface.

Fig. 16. The representative streamlines and value for (a) 22.7and (b) 23.6(kf 0.15). The velocity vectors for 23.6inside the red box in (b) are
highlighted in (c). (For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

307

level in the second short period. Finally, when the LEV travels close to the trailing edge in the third short period, the
pressure suddenly increases to result in the lift stall. Similar observations can be found in references documented by
Ekaterinaris and Platzer (1997), McAlister et al. (1978), Visbal (2014), and Zanotti and Gibertini (2012). Due to the limitation
of the RANS model in highly separated ows, we conne our ow structure analysis mostly in upstroke phase and slightly
after the stall during the downstroke process.
6.2.1. AoA 21.722.7
The Cp distributions at 20.5in Figs. 14(a) and 15(a) are still similar as those presented in Figs. 10 and 11 at the rst stage.
The minimum Cp near the leading edge in Fig. 15(b) is still low enough such that the acceleration still can maintain the ow
to attach to the foil surface, and hence the u/U contours in Fig. 14(a) reveals no reverse ow. However, as the AoA increases
further, the minimum Cp in the upper surface increases instead as shown by the arrow in Fig. 15(b). As the consequence, the
ow acceleration near the leading edge in the upper surface becomes weaker and fails to resist the stronger adverse
pressure gradient. In Fig. 14(b) at 21.7, the velocity near the solid boundary becomes negative in the u/U contour, and
detached streamlines from the leading edge are observed. The positive u/U value of the free stream and the negative value of
the reverse ow in Fig. 14(b) generate a recirculation region of the LEV, and the negative value inside the LEV is detected.
This rotation motion transfers the low-pressure area from the leading edge at 20.5in Fig. 14(a) to the center of the
recirculation region at 21.7in Fig. 14(b). As 22.7in Fig. 14(c), from its u/U contour, the even stronger adverse pressure
gradient results in a more signicant reverse ow near the boundary layer. The greater relative motion between the

Fig. 17. Flow structures of Cp, u/U, , and streamline from 23.5to 24.7(kf 0.15). The legend for each contour is shown as below.

308

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

incoming free stream and the reverse ow requires a stronger centrifugal force to maintain the rotation motion. Consequently, from the corresponding Cp contours, the pressure within the LEV as 22.7in Fig. 14(c) becomes even lower than
that of 21.7in Fig. 14(b). These tendencies of the ow structures from AoA21.7to 22.7favor the rotation motion such
that the value inside the LEV is decreasing, and the size of the LEV is also enlarged.
From 20.5to 22.7in Fig. 15(a), the Cp distributions along the lower surface are very consistent. However, the Cp
distributions are quite different in the upper surface. The concave Cp distribution in the upper surface corresponds to the
size of the LEV as denoted in Fig. 15(b). From 21.7to 22.7, the LEV is growing in size such that the concave Cp distribution
becomes wider as shown in Fig. 15(b), and hence the lift is still increasing accordingly.

Fig. 18. Cp along the upper surface during the detachment of the LEV from 23.6to 24.7(kf 0.15). The left gure (a) highlights Cp of 24.7, and the right
gure (b) compares Cp in the upper surface from 23.6to 24.7.

Fig. 19. Flow structures of and streamline as 24.7to 24.3(kf 0.15). For interpretation of the references to color in this gure the reader is referred to
the web version of this article.)

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

309

6.2.2. AoA 22.724.7


As discussed in Fig. 14, the ow structure is changing to favor the rotation motion. When AoA increases from 22.7to
23.6, this tendency leads to an even more circular shape of the recirculation region. Therefore, the curvature of the
streamlines near the leading edge becomes higher, which results in the detachment of the LEV from the leading edge as
shown from Fig. 16(a) to (b).
As the black arrows depict in Fig. 16(c) for 23.6, the free stream and the front part of LEV have opposite ow directions
after the detachment of the LEV. In order to satisfy the continuity, two new recirculation regions with different rotation
directions are created, namely Vortices A and B. For Vortex A, the ow near the boundary layer is positive with counter
clockwise rotation, and vice versa for the Vortex B. New local minimums of Cp in the upper surface can be observed within
the recirculation region of Vortices A and B, which is similar as the trend shown in Fig. 15(b) for the LEV.
If the low-pressure area of the LEV is too close the solid boundary, the centrifugal force may hardly overcome the viscous
force, and hence this kind of ow structure will not benet the rotation strength. In the Cp contours of Fig. 17, after the
detachment, the low-pressure area of the LEV is transported away from the solid boundary. Therefore, the value inside the
LEV also indicates that its rotation strength is increasing, and the recirculation region of the LEV also becomes larger from
Fig. 17(a)(c).
In Fig. 14(b) and (c), from 20.5to 22.7before the detachment of the LEV, the value of Cp within the LEV keep
decreasing from 4 to 7. In contrast to this decreasing trend, in the Cp contours of Fig. 17, the minimum Cp inside the LEV
after its detachment maintains around  7.5. Therefore, instead of utilizing an even lower pressure within the LEV as shown
in Fig. 14, the mechanism to favor the rotation motion at this short period is due to the fact that the rotation center of the
LEV is transported away from the solid boundary, which is shown in Fig. 17.
The negative value of is also observed in Fig. 17 within Vortex A and B previously dened in Fig. 16(c). The values
become even more negative as the corresponding recirculation regions become larger in size from 23.6to 24.7. Since the
LEV is closer to Vortex A, the value shows that the rotation strength of Vortex A is overall stronger than that of Vortex B.
During this period, in the u/U contours of Fig. 17, the speed of the reverse ow within the LEV reaches its maximum
about u/U 1.7 in Fig. 17(a) as 23.6. Since the low-pressure area is moving away from the upper surface, the speed of the
reverse ow becomes slower as shown in Fig. 17(b) and (c), but its region becomes larger instead.

Fig. 20. Cp contours and streamline from as 24.7to 24.3(kf 0.15).

310

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Fig. 18(a) highlights the Cp along the upper surface as 24.7. Three local minimums can be found, and from left to right,
each concave distribution corresponds to the recirculation region of Vortices B, A, and LEV. In Fig. 18(a), it is clear that the
low-pressure area of the LEV is the largest and strongest, indicating that the LEV plays the most important role to maintain
the lift.
After the detachment of the LEV, the low-pressure area of the LEV is transported away from the upper surface, and hence
the values of Cp corresponding to the LEV in the upper surface increase as shown in Fig. 18(b). However, the region of each
recirculation region become larger in size as shown in Fig. 17, which still causes an increasing lift force from 23.6to 24.7.
In fact, the lift force reaches its maximum at 24.7in Fig. 4(a).
6.2.3. AoA24.724.3
In Fig. 4(a), the normal lift coefcient reaches its maximum CN 2.90 at 24.7. As AoA increases to 24.9, CN then drops
to 2.81. As the pitching motion begins the down-stroke process from 24.7to 24.3, CN drops rapidly from 1.7 to 1.25.
In Fig. 19(a) as 24.7, the LEV is very close to the trailing edge. At this instant, the streamline near the trailing edge and
the red arrow shown in Fig. 19(a) indicate the uid ow from the lower surface cannot enter the upper surface. As 24.9in
Fig. 19(b), the LEV are even closer to the lower surface, and hence curvature of streamlines from the lower surface are
changed more signicantly. The uid ow from the lower surface now will be attracted into the upper surface by the suction
force of the LEV. The red arrows in Fig. 19(b) as 24.9highlight this phenomenon. The lift is found to be stall when the
reverse ow from the lower surface is triggered.
From 24.7to 24.9, instead of keeping constant as shown in Fig. 17, the pressure of the center of the LEV starts to
increase as shown from Fig. 20(a) and (b). However, the reverse ow from the lower surface brings additional momentum to
compensate the rotation strength of the LEV, and hence its value still keeps decreasing from Fig. 19(a) and (b) even after
the lift stall. Since the strong rotation strength of the LEV still can be maintained as 24.9, the drop of CN is not very
signicant (from 2.9 to 2.81). At this instant, the LEV is still right above the upper surface, and hence the reverse ow has to
make a sharp turn into the upper surface as shown in Fig. 19(b).
As 24.9, since the LEV is leaving the upper surface, the pressure gradient around the LEV will become even weaker as
shown in Fig. 20(c), leading to a weaker rotation motion. Therefore, the recirculation region of the LEV is less circular in
shape, and the value of becomes less negative in Fig. 19(c). At this instant, the space between the LEV and the trailing edge
becomes larger enough, allowing the reverse ow from the lower surface to make a more smooth turn in Fig. 19(c) than that
of Fig. 19(b). Consequently, a new recirculation region is created right above the trailing edge, and the negative value
inside this region identies the so-called Trailing Edge Vortex (TEV). A low-pressure area corresponding to this TEV can be
found in Fig. 20(c).
In Figs. 19(d) and 20(d) for 24.7, the pressure inside the TEV drops even lower than that of 24.9. Therefore, the
inside the TEV becomes more negative values to represent a stronger rotation. At 24.3, although the pressure inside the
TEV in Fig. 20(e) is somehow higher than that of Fig. 20(d), the ow structure of the streamlines still favor the rotation
motion, which results in a lower inside TEV in Fig. 19(e) than that of Fig. 19(d). From 24.7 to 24.3, the pressure within
the LEV is rising to a level very close to that of the surrounding uid in Fig. 20(d) and (e). Thus the recirculation region of the
LEV is disappearing, which corresponds to the increasing value inside the LEV in Fig. 19(d) and (e). Therefore, when the
TEV develops its own strength, its more circular recirculation region will break down the recirculation region of the LEV and
enhances its dissipation.
Once the reverse ow occurs at 24.7in Fig. 19(a), the high pressure in the lower surface around the trailing edge will
decrease as shown in Fig. 20(a) and (b), which is the evidence to shows that the reverse ow is responsible for the lift stall.
After the formation of the TEV, the pressure difference between the high-pressure area in the lower surface and the lowpressure area inside the TEV will lead to a signicant acceleration for the reverse ow from the lower surface. Therefore,

Fig. 21. u/U contours and streamline from as 24.7to 24.3(kf 0.15).

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

311

Fig. 22. Cp along the solid surface from 24.7to 24.3(kf 0.15).

from 24.7to 24.9, the velocity near the trailing edge in the lower surface can be accelerated around twice of the free
stream velocity as shown in Fig. 21. Then the high velocity in the lower surface will cause a more signicant decreasing
trend of high pressure in the lower surface as shown in Fig. 20(c) than that before the formation of the TEV illustrated in
Fig. 20(b).
As discussed in the previous two paragraphs, when the TEV has formed right after the stall, the development of the TEV
can further suppress the LEV, which can increase the pressure in the upper surface. Besides, the high pressure in the lower
surface will decrease more signicantly. Therefore, the evolution of the TEV can reduce the lift dramatically if it happens
right after the stall.
From 24.7to 24.9, the pressure along the entire solid boundary is shown in Fig. 22(a). Before the reverse ow from
the lower surface is triggered, the pressure in the lower surface changes very slightly as shown in Fig. 15(a). Once the reverse
ow occurs as illustrated in Figs. 19 and 20, the pressure of the lower surface can vary more obviously as shown in Fig. 22(a).
Comparing 24.9to 24.7in Fig. 22(a), the pressure in the upper surface goes up since the low-pressure area of the LEV is
leaving the foil region, and the pressure in the lower surface decreases due to the reverse ow as discussed above. In the
consequence, CN drops from 2.90 to 2.81. From 24.7to AoA24.3, the TEV has already formed a new low-pressure area
as shown in Fig. 20(d), and hence Cp around the trailing edge in the upper surface drops again as shown in Fig. 22(b) as
24.7. But except region near the TEV, the pressure in the upper surface becomes much higher than that shown in Fig. 22
(a) as 24.7and 24.9. Therefore, CN drops to 1.7 at 24.7. As 24.3, the low-pressure area of the TEV is moving away from
the upper surface as shown in Fig. 20(e). Then CN drops rapidly as illustrated in Fig. 22(b) from 1.7 to 1.25. From
AoA 24.7to 24.3, CN drops signicantly from 2.9 to 1.25 with only 1 variation. This rapid drop of lift is due to the strong
interactions between LEV and TEV.
To sum up, before the formation of the reverse ow from the lower surface, the interaction between high-pressure ow
in the lower surface and low-pressure ow in the upper surface is almost absent. When the reverse ow is observed as
shown at 24.9in Fig. 19(b), the momentum exchange between the high- and low-pressure uid ows tends to balance the
pressure difference between lower and upper surface, and hence the stall occurs. If the TEV develops right after the stall, the
lift force can drop more dramatically.
Furthermore, in Section 6, three short periods, 21.722.7, 22.724.7, and 24.724.7, are analyzed at the second
stage. For 21.722.7, as shown in Fig. 14, the LEV remains attached to the leading edge, at this period, the pressure of the
rotation center of the LEV keeps decreasing to strengthen the LEV and enlarge its size. For 21.722.7, the LEV is detached
from the leading edge. The pressure of the rotation center is almost constant as shown in Fig. 17. The mechanism to
strengthen and enlarge the LEV is due to the fact that the rotation center of the LEV is transported away from the solid
boundary. Finally, for 24.724.7, the LEV is leaving the foil region. The LEV is then dissipated due to the increasing pressure
within the LEV as shown in Fig. 20. Each short period at the second stage corresponds to different ow patterns during the
dynamic stall as discussed through Sections 6.2.1 to 6.2.3, and these short periods can be distinguished according to the
trend of the pressure variation inside the LEV, which is decreasing, unchanged, and then increasing respectively.

7. Transport equation of
The transport equation of will be derived such that its convective, pressure, and diffusion terms can be further
investigated. For two-dimensional incompressible and laminar ows, the transport equation of 1/2SijSij and 1/2kk
(enstropy) can be derived (Chen, 1997; Davidson, 2004):

312

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Fig. 23. Sum of source terms /t SC SP SV. The unit is 1/s, and the dark lines represent the contour level as  0.2 (kf 0.15). For interpretation of the
references to color in this gure the reader is referred to the web version of this article.)

Fig. 24. The viscous terms SV. The unit is 1/s, and the dark lines represent the contour level as  0.2 (kf 0.15). (a) 23.6 (b) 24.3.

1
2
D 2 Sij Sij = Sij 2P + S Sij
ij
Dt
xm xm
xm xm

1
2
D1

2 k k = + k 2 k k
xm xm
Dt

(7)

In Eq. (7), three-dimensional effects such as the vortex stretching have already been neglected. D/Dt represents the
material derivative, and the notation of k emphasizes that the vorticity can only exist in the k direction when the twodimensional ow is conned to the i and j direction, and m represents as the repeated notation. Therefore, from Eq. (6), the

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

313

corresponding transport equation of can be derived by the chain rule:

)(

D Sij Sij 2 k k / Sij Sij + 2 k k


Dk
=
Dt
Dt

) = 2

k k

D Sij Sij
2

Dt

2Sij Sij
1

Sij Sij + 2 k k

D k k
2
Dt

(8)

Then insert Eq. (7) into (8):


2k2

= um k +
t
xm

Convective

(S S

terms

Sji x x
i

+ 2 k k

Pressure

ij ij

terms

= SC +

SP +

SV

2Sij
2
2k Sji k x x Sij x kx
m m
m m

(S S

+ 2 k k

Diffusion

ij ij

terms

(9)

In order to include the turbulent effect of the vorticity equation, Senocak (2002), Huang et al. (2014), and Ji et al. (2014)
have simply replace the laminar kinematic viscosity in the vorticity transport equation by sum of the laminar and turbulent kinematic viscosity (t). This treatment will drop the terms related to the spatial gradients of the eddy viscosity t
and the turbulent kinetic energy k in the diffusion terms, which are usually assumed to be small. Similarly, we also replace
by t in the diffusion terms of in Eq. (9). The convective, pressure, and diffusion terms are denoted as SC, SP, and SV,
respectively.
Fig. 23 illustrates the sum of source terms SC SP SV, which is the local time rate of change, /t. In Fig. 13, we have
dened o  0.2 is a threshold to identify a rotation motion. The dark line in Fig. 23 represents the contour level as
 0.2, and the enclosed area of this contour level indicates the size of the LEV. One can see that /t is largely negative
inside the LEV, and the region of negative /t becomes larger From Fig. 23(a) as 21.7to Fig. 23(d) as 24.3. Correspondingly, as shown from Fig. 14(b) to Fig. 17(a)(c) at the same angles of Fig. 23, inside the LEV decreases to favor the
rotation strength, and the LEV grows in size during the evolution. Moreover, the most negative value of /t is always

Fig. 25. The convective and pressure terms as 22.7. The unit is 1/s, and the dark lines represent the contour level as  0.2 (kf 0.15). (a) SP
(b) SC

Fig. 26. The convective and pressure terms as 23.6. The unit is 1/s, and the dark lines represent the contour level as  0.2 (kf 0.15). (a) SP
(b) SC

314

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Fig. 27. u/U contours and streamlines for 20.5at different frequencies (a) kf=0.15 (b) kf=0.25 (b) kf=0.25.

Fig. 28. The ow structures by streamlines and as kf 0.25 from 23.6to 20.5.

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

315

located in the downstream part of the LEV or slightly outside the LEV (as shown by the red arrow in Fig. 23(b)), and hence
the LEV develops toward downstream. On the contrary, the most positive value of /t is slightly outside the LEV in the
upstream direction (as shown by the black arrow in Fig. 23(b)), which tends to detach the LEV.
The diffusion term SV is also illustrated in Fig. 24. Comparing the level of the legend of Fig. 24 to that of Fig. 23, the
diffusion term SV only contributes a little inside the LEV. The strong diffusion term seems to surround the edge of the LEV,
but the strength is usually 3 to 4 times weaker than that of the convective and pressure term in Figs. 25 and 26.
Therefore, the convective and pressure term play more important roles than the diffusion term. Figs. 25 and 26 show the
pressure term SP and convective term SC as 22.7and 23.6, respectively. It is clear that the pressure term dominates inside
the LEV as shown in Figs. 25(a) and 26(a). As discussed in Section 6.2.2 based on the concept of centrifugal force, the ow
structure becomes more circular in shape and favor the rotation strength, which reects the larger area of negative SP from
Figs. 25(a) to 26(a) as the angle increases. In other words, the pressure term dominates and enhances the rotation strength
inside the LEV. As a result, when the LEV leaves the foil, the rotation strength could become weaker due to the sudden
increase of the pressure and ensuing increase of the pressure term SP. As for the convective term SC, by comparing
Fig. 23(b) and (c) to Figs. 25(b) and 26(b) at the same angles, one can see that the most negative/positive /t slightly
outside LEV in the upstream/downstream direction mainly comes from the convective term SC. Therefore, it is the convective
effect to detach the LEV and move it downstream. Overall, the value of the convective term SC inside the LEV is mostly
positive, which cancels certain contribution of the pressure term SP.
As discussed in Figs. 12 and 13 of Section 6.1, a region without really having a vortex could possibly identied by vorticity
, and the value of is more consistent to the strength of the rotation motion. Moreover, for an incompressible ow, the
baroclinic term in the transport equation of vorticity is zero, and hence the pressure effect can hardly be evaluated directly
through the vorticity equation. However, the transport equation of retains the pressure term, which is very pronounced
during the evolution of the LEV as shown in this section.

8. Comparisons between kf 0.15 and 0.25


The analysis of the ow structures in Section 6 mainly focuses on the dynamic condition as kf 0.15. In order to gain more
insights of the uidstructure interaction during the dynamic stall, the ow structures are compared between kf 0.15 and
0.25 in this section. Since more momentum can be transferred from the solid boundary into the uid ow at a higher
oscillation frequency, the streamwise velocity near the upper surface of kf 0.25 is higher than that of kf 0.15, which is
illustrated in Fig. 27 as 20.7. Accordingly, for kf 0.25, the ow separation is further delayed than that of kf 0.15, and so is
the evolution of the LEV. This phenomenon can be observed from Figs. 14(b) for kf 0.15 and 28(a) for kf 0.25. Their sizes of
the recirculation regions of the LEV are comparable, but the AoA of the latter is apparently higher than the former. Then the
stall is postponed into the down-stroke as shown in the CN curve of Fig. 4(b) as kf 0.25.
Therefore, even the stall occurs at 24.7for kf 0.15, the LEV of kf 0.25 is still under development from 24.7to 24.9in
Fig. 28(b)(c), and hence CN still increases slightly from 2.7 to 2.75 even to the downstroke process. The stall angle as
kf 0.25 happens at 24.5with maximum CN 2.84. Fig. 28(d) shows ow structures as stall just occurs for kf 0.25
(CN 2.81). The LEV at 24.3in Fig. 28(d) is very close to the trailing edge, and the low-pressure area of the LEV is sufcient
to induce reverse ow from the lower surface. Same as the mechanisms shown in Figs. 1921 for kf 0.15, once the ow is
attracted from the lower to upper surface, CN will drop to cause the stall.
Table 2
The reduced frequency, stall angle, and the corresponding maximum lift summarized from the selected references.
Reference

Pitching pattern and ow conditions

kf

Stall angle

Maximum lift

Leishman (1990)

10 10sin(f  t) NACA23012 foil , Re 1.5  106

Lee and Gerontakos (2004)

10 5sin(f  t) NACA0012 foil , Re 1.35  105

0.1
0.16
0.2
0.025
0.05
0.1
0.025
0.05
0.1
0.05
0.1
0.15
0.25
0.01
0.1
0.2
0.4

20.3
19
18.9
14
14.8
14.9
18.6
21
24
21.3
23.7
24.9
23.8
17.6
22.6
25
24.69

CL 2.01
CL 1.78
CL 1.73
CL 1.3
CL 1.33
CL 1.40
CL 1.31
CL 1.66
CL 1.97
CN 2.43
CN 2.89
CN 3.05
CN 2.92
CN 1.17
CN 1.86
CN 2.13
CN 1.92

15 10sin(f  t) NACA0012 foil , Re 1.35  105

McAlister et al. (1978); McCroskey et al. (1976)

15 10sin(f  t) NACA0012 foil , Re 2.5106

Sharma and Poddar (2013)

10 15sin(f  t) NACA0015 foil , Re 2  105

316

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

Since the LEV occurs later as kf 0.25, when it is close to the trailing edge as shown in Fig. 28(d) and (e), the oscillation
period has already entered downstroke process. In other words, the trailing edge now is tilting upward to approach the LEV.
Then the space between the LEV and the trailing edge is not large enough such that the reverse ow still has to make a sharp
turn into the upper surface as highlighted by the arrows in Fig. 28(d) and (e). Consequently, the TEV is further delayed. The
TEV is under development during 21.7to 20.5in Fig. 28(f) and (g), displaying a delay at least by 2.8 between the stall
(24.5) and the formation of the TEV. As discussed from Fig. 19 to 21 for kf 0.15, the pressure in the lower surface increases
rapidly when the TEV is formed right after the stall (0.4 phase difference as shown in Fig. 19(a) and (c)). This mechanism
will be absent for kf 0.25 since the phase between stall and formation of the TEV is more pronounced (2.8). Different
oscillation frequencies could activate different interactions between the LEV and TEV, especially when the lift stall occurs
around the maximum angle of attack. In other words, the lift curve slope after stall strongly depends on the position of the
stall angle. If the stall angle remains within the upstroke close to the maximum angle, the slope drops very rapidly such as
kf 0.15 in Fig. 4(a) by a strong interaction between LEV and TEV. On the contrary, if the stall occurs within the downstroke,
the drop of lift will be more gentle as shown in Fig. 4(b) for kf 0.25 due to the delay of the TEV. This kind of trend can also
be observed in the references by McAlister et al. (1978), McCroskey et al. (1976), and Sharma and Poddar (2013).
As shown in our simulation results, it is no doubt that increasing reduced frequency could delay the onset of the ow
separation and dynamic stall to a higher angle, which has already been reported in references by Lee and Gerontakos
(2004), Leishman (1990), McAlister et al. (1978), McCroskey et al. (1976), and Sharma and Poddar (2013). However, an
increasing reduced frequency is not always guaranteed to a higher maximum lift coefcient. Table 2 lists the reduced
frequency, stall angle, and its corresponding maximum lift coefcient from selected references. As shown in Table 2,
Leishman (1990) has reported that the maximum lift force decreases as the reduced frequency increases while Lee and
Gerontakos (2004) have observed an opposite trend. Moreover, the trend is not even monotonic in references by McAlister
et al. (1978), McCroskey et al. (1976), and Sharma and Poddar (2013). It seems that the relations between the reduced
frequency and the maximum lift are implicit. However, their connections still can be discovered from Table 2.
From the reference of Lee and Gerontakos (2004), the reduced frequency only has little inuence on the stall angle and
its maximum lift under the light-stall pitching while the inuence becomes more pronounced for the deep-stall pitching.
Futhermore, one can see from Table 2 that the maximum lift depends on the stall angle strongly under the same Reynolds
number. If the stall angle is delayed by a higher reduced frequency and remains in the upstroke, the maximum lift will
increase. On the contrary, if the stall angle is already postponed into the downstroke, further increase of the reduced
frequency can lower the maximum lift. This trend is well captured in our numerical prediction and the corresponding
measured results as shown in Table 1 (McAlister et al., 1978; McCroskey et al. 1976). In Table 1, both stall angles are close to
the maximum angle, and hence the differences of the maximum CN under kf 0.15 and 0.25 are very little. Moreover, the
maximum CN of kf 0.15 is slightly higher than that of kf 0.25 because the stall angle for kf 0.25 is already within the
downstroke process.

9. Summary and conclusions of current work


In this study, the uidstructure interaction is investigated for a two-dimensional ow (Re2.5  106) over a sinusoidpitching foil by the Shear Stress Transport (SST k) turbulence model. Although discrepancies in downstroke phase, which
are also documented in other numerical studies, are observed between our CFD and experimental results, our current CFD
results are sufcient to predict the mean features and qualitative tendencies of the dynamic stall phenomenon. Due to the
limitation of the RANS model in highly separated ows, we conne our ow structure analysis mostly in upstroke phase and
slightly after the stall during the downstroke process.
The mechanisms to delay of stall are categorized based on the formation of the Leading Edge Vortex (LEV). At the rst
stage, the ow structures are carefully investigated by comparisons between the static and the dynamic condition. The ow
has already separated as 17,and the free stream velocity near the trailing edge in the upper surface is faster for the static
condition. In order to meet the continuity, the uid ow in the lower surface for the static condition has to experience a
higher favorite pressure gradient, and hence the pressure in the lower surface is overall lower. Accordingly, the lift stall
occurs, and while the lift maintains increasing for the dynamic condition.
At the second stage for kf 0.15, the value is used to quantify the rotation strength inside the vortex. During the
evolution of the LEV, the low-pressure area of the LEV is transported toward the trailing edge, and multiple recirculation
regions are created. Before 24.7, the rotation strengths of the LEV and those recirculation regions are increasing, and so are
their sizes. The lift force still can be maintained accordingly. After 24.7, the lift is reduced once the LEV is close enough to
the trailing edge to induce the reverse ow. When the TEV is triggered, the acceleration of the reverse ow becomes
signicant, and the pressure in the lower surface can increase rapidly. Furthermore, the development of the TEV can break
down the recirculation region of the LEV. Therefore, the drop of the lift is very signicant for kf 0.15 since the TEV is formed
right after the stall. The transport equation of has been derived as the convective, pressure, and diffusion terms. It is found
that the viscous term is the least important while the pressure term dominates the rate of the change of the rotation
strength inside the LEV. This trend can hardly be observed directly by using the vorticity transport equation due to the zero
baroclinic term for the incompressible ow.

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

317

For kf 0.25, the stall is further delayed into the down-stroke. As the shedding of the LEV takes place, the TEV is absent.
Therefore, the lift drops gently after the stall for kf 0.25. As for the maximum lift, it depends on the stall angle strongly. If
the stall angle is delayed by a higher reduced frequency and remains in the upstroke, the maximum lift will increase. On the
contrary, if the stall angle is already postponed into the downstroke, further increase of the reduced frequency can lower the
maximum lift.
Due to the limitations of the RANS model, the laminar-to-turbulent transition, reattachment, and formation of the LSB
are very difcult to be predicted properly near the leading edge, which could result in the discrepancies during the
downstroke phase in this study. However, the LES computation is believed to be capable of dealing with these details.
Therefore, the LES computation will be considered as the future work such that more reliable numerical results during the
downstroke phase could be acquired.

Acknowledgments
The present efforts are partially supported by the Ministry of Science and Technology in Taiwan with Project number
102-3113-P-006-011.

References
ANSYS, 2015. www.uent.com.
Ahmad, K.A., Abdullah, M.Z., Watterson, J.K., 2010. Numerical modeling of a pitching foil. J. Mek. 30, 3747.
Althaus, D., Wortmann, F.X., Stuttgarter Prolkatalog: Messergebnisse aus dem Laminarwindkanal des Instituts fr Aerodynamik und Gasdynamik der
Universitt Stuttgart 1981 F. Vieweg.
Bardina, J.E., Huang, P.G., Coakley, T.J., 1997. Turbulence modeling validation, testing, and development. NASA Technical Memorandum 110446.
Carr, L.W., 1988. Progress in analysis and prediction of dynamic stall. J. Aircr. 25 (1), 617.
Chandrasekhara, M.S., Carr, L.W., 1990. Flow visualization studies of the Mach number effects on dynamic stall of an oscillating airfoil. J. Aircr. 27 (6),
516522.
Chen, C.J., 1997. Fundamentals of Turbulence Modelling, Taylor & Francis Washington, DC.
Choudhry, A., Leknys, R., Arjomandi, M., Kelso, R., 2014. An insight into the dynamic stall lift characteristics. Exp. Therm. Fluid Sci. 58, 188208.
Davidson, P.A., 2004. Turbulence: An Introduction for Scientists and Engineers. Oxford University Press, New York.
Dawes, W.N., 2007. Turbomachinery computational uid dynamics: asymptotes and paradigm shifts. Philos. Trans. R. Soc. A 365, 25532585.
Doerffer, P., Szulc, O., 2008. Numerical simulation of model helicopter rotor in hover. Task Q. 12 (3), 227236.
Ekaterinaris, J.A., Menter, F.R., 1994. Computation of oscillating airfoil ows with one-and two-equation turbulence models. AIAA J. 32 (12), 23592365.
Ekaterinaris, J.A., Platzer, M.F., 1997. Computational prediction of airfoil dynamic stall. Prog. Aerosp. Sci. 33, 759846.
Gardner, A.D., Richter, K., Mai, H., Altmikus, A.R.M., Klein, A., Rohardt, C.H., 2013. Experimental investigation of dynamic stall performance for the EDI-M109
and EDI-M112 airfoils. J. Am. Helicopter Soc. 58, 012005.
Gharali, K., Johnson, D.A., 2013. Dynamic stall simulation of a pitching airfoil under unsteady freestream velocity. J. Fluids Struct. 42, 228244.
Gompertz, K., Kumar, P., Jensen, C.D., Peng, D., Gregory, J.W., Bons, J.P., 2011. Modication of a transonic blowdown wind tunnel to produce oscillating
freestream Mach number. AIAA J. 49 (11), 25552563.
Gormont, R.E., 1973. A mathematical model of unsteady aerodynamics and radial ow for application to helicopter rotors. U.S. Army AMRDL Eustis
Directorate Report TR 72-67.
Haller, G., 2005. An objective denition of a vortex. J. Fluid Mech. 525, 126.
Harris, F.D., Tarzanin, F.J., Fisher, R.K., 1970. Rotor high-speed performance; theory vs. test. J. Am. Helicopter Soc. 15 (3), 3544.
Hill, J.L., Shaw, S.T., and Qin, N., 2004. Investigation of Transition Modelling for Aerofoil Dynamic Stall. In: Proceedings of the 24th International Congress of
the Aerospace Sciences.
Huang, B., Zhao, Y., Wang, G., 2014. Large eddy simulation of turbulent vortex-cavitation interactions in transient sheet/cloud cavitating ows. Comput.
Fluids 92, 113124.
Issa, R.I., 1982. Solution of the implicit discretized uid ow equations by operator splitting. J. Comput. Phys. 62, 4065.
Jeong, J., Hussain, F., 1995. On the identication of a vortex. J. Fluid Mech. 285, 6994.
Ji, B., Luo, X., Arndt, R.E., Wu, Y., 2014. Numerical simulation of three dimensional cavitation shedding dynamics with special emphasis on cavitationvortex
interaction. Ocean Eng. 87, 6477.
Jones, W.P., Launder, B.E., 1972. The prediction of laminarization with a two-equation model of turbulence. Int. J. Heat Mass Transf. 15, 301304.
Kang, C.K., Aono, H., Cesnik, C.E., Shyy, W., 2011. Effects of exibility on the aerodynamic performance of apping wings. J. Fluid Mech. 689, 3274.
Kang, C.K., Aono, H., Sik Baik, Y., Bernal, L.P., Shyy, W., 2012. Fluid dynamics of pitching and plunging at plate at intermediate Reynolds numbers. AIAA J. 51
(2), 315329.
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent ow. Comput. Methods Appl. Mech. Eng. 3, 269289.
Lee, T., Gerontakos, P., 2004. Investigation of ow over an oscillating airfoil. J. Fluid Mech. 512, 313341.
Leishman, J.G., 1990. Dynamic stall experiments on the NACA 23012 aerofoil. Exp. Fluids 9, 4958.
Leu, T.S., Yu, J.M., Hu, C.C., Miau, J.J., Liang, S.Y., Li, J.Y., Cheng, J.C., Chen, S.J., 2012. Experimental study of free stream turbulence effects on dynamic stall of
pitching airfoil by using particle image velocimetry. Appl. Mech. Mater. 225, 103108.
Liggett, N.D., 2012. Numerical Investigation of static and dynamic stall of single and apped airfoils. PhD dissertation in the School of Aerospace Engineering of Georgia Institute of Technology, Atlanta.
Loftin, L. K., Smith, H. A., 1949. Aerodynamic Characteristics of 15 NACA Airfoil Sections at Seven Reynolds Numbers from 0.7106 to 9.0106. National
Advisory Committee for Aeronautics, Technical Note, Washington, DC.
Martinat, G., Braza, M., Hoarau, Y., Harran, G., 2008. Turbulence modeling of the ow past a pitching NACA0012 airfoil at 105 and 106 Reynolds numbers. J.
Fluids Struct. 24, 12941303.
McAlister, K.W., Carr, L.W.McCroskey, W.J., 1978. Dynamic Stall Experiments on the NACA 0012 airfoil. NASA Technical Paper 1100.
McAlister, K.W., Pucci, S.L., McCroskey, W.J., Carr, L.W., 1982. An Experimental Study of Dynamic Stall on Advanced Airfoil Sections Volume 2. Pressure and
Force Data. NASA Technical Memorandum; 84245.
McCroskey, W.J., Carr, L.W., McAlister, K.W., 1976. Dynamic stall experiments on oscillating airfoils. AIAA J. 14 (1).
McCroskey, W.J., McAlister, K.W., Carr, L.W., Pucci, S.L., Lambert, O., Indergrand, R.F., 1981. Dynamic stall on advanced airfoil sections. J. Am. Helicopter Soc.
26 (3), 4050.

318

C.-C. Tseng, Y.-E. Cheng / Journal of Fluids and Structures 58 (2015) 291318

McCroskey, W.J., 1981. The phenomenon of dynamic stall. NASA TM-81264.


McCroskey, W.J., McAlister, K.W., Carr, L.W., Pucci, S.L., 1982. An Experimental Study of Dynamic Stall on Advanced Airfoil Sections Volume 1. Summary of
the Experiment. NASA Technical Memorandum; 84245.
Menter, F.R., 1992. Improved two-equation k turbulence model for aerodynamic ows. NASA Technical Memorandum; 103975.
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA J. 32 (8), 15981605.
Menter, F.R., Kuntz, M., Langtry, R., 2003. Ten years of industrial experience with the sst turbulence moDel. In: Hanjalic, K., Nagano, Y., Tummers, M. (Eds.),
Turbulence, Heat and Mass Transfer, 4. , Begell House, Redding, CT, pp. 625632.
Menter, F.R., Langtry, R., V olKer, S., 2006. Transition modelling for general purpose CFD codes. Flow Turbul. Combust. 77, 277303.
Mulleners, K., Kindler, K., Raffel, M., 2012a. Dynamic stall on a fully equipped helicopter model. Aerosp. Sci. Technol. 19 (1), 7276.
Mulleners, K., Pape, A.L., Heine, B., Raffel, M., 2012b. The dynamics of static stall. In: Proceedings of the16th Int Symposium on Applications of Laser
Techniques to Fluid Mechanics.
Niu, Y.Y., Chang, C.C., 2013. How do aerodynamic forces of the pitching rigid and exible airfoils evolve? AIAA J. 51 (12), 29462952.
Niu, Y.Y., Liou, M.S., 1999. Numerical simulation of dynamic stall using an improved advection upwind splitting method. AIAA J. 37 (11), 13861392.
Patankar, S.V., 1980. Numerical Heat Transfer and Fluid Flow. Hemisphere Inc., McGraw-Hill, New York, NY, USA.
Raffel, M., Kompenhans, J., Wernert, P., 1995. Investigation of the unsteady ow velocity eld above an airfoil pitching under deep dynamic stall conditions.
Exp. Fluids 19 (2), 103111.
Ramsay, R.R., Hoffman, M.J., and Gregorek, G.M., 1995. Effects of grit roughness and pitch oscillations on the S809 airfoil. Airfoil performance report.
Ramsey, R.R., Gregorek, G.M., 1999. Effects of grit roughness and pitch oscillations on the S813 airfoil. Airfoil performance report.
Senocak, I., 2002. Computational methodology for the simulation of turbulent cavitating ows. Department of Mechanical Engineering, University of
Florida, Gainesville, Florida Ph.D. thesis.
Sharma, D.M., Poddar, K., 2013. Investigation of dynamic stall characteristics for ow past an oscillating at various reduced frequencies by simultaneous PIV
and surface pressure measurements. 10th International Symposium On Particle Image Velocimetry, Delft, The Netherlands.
Shyy, W., Aono, H., Kang, C.K., 2013. An Introduction to Flapping Wing Aerodynamics. Cambridge University Press, New York.
Shyy, W., Lian, Y., Tang, J., Viieru, D., Liu, H., 2007. Aerodynamics of Low Reynolds Number Flyers. Cambridge Univ. Press, NewYork1158.
Shyy, W., Thakur, S.S., Ouyang, H., Liu, J., Blosch, E., 2005. Computational Techniques for Complex Transport Phenomena. Cambridge University Press, New
York.
Sohail, M.A., Ullah, R., 2011. CFD of oscillating airfoil pitch cycle by using PISO algorithm. World Acad. Sci. Eng. Technol. 60, 13551359.
Tseng, C.C., HuH.A., 2015. Dynamic Behaviors of the Flow past a Pitching Foil based on Eulerian and Lagrangian Viewpoints. Under the review for the1st
revision.
Velkova, C., Todorov, M., Dobrev, I., Massouh, F., 2012. Approach for numerical modeling of airfoil dynamic stall. Bul. Trans. Proceedings.
Visbal, M.R., 1990. Dynamic stall of a constant-rate pitching airfoil. J. Aircr. 27 (5), 400407.
Visbal, M.R., 2014. Analysis of the Onset of Dynamic Stall Using High-Fidelity Large-Eddy Simulations. In: Proceedings of AIAA 52nd Aerospace Sciences
Meeting.
Wang, S., Indham, D.B., Ma, L., Pourkashanian, M., Tao, Z., 2010. Numerical investigations on dynamic stall of low Reynolds number ow around oscillating
airfoils. Comput. Fluids 39, 15291541.
Wang, S., Indham, D.B., Ma, L., Pourkashanian, M., Tao, Z., 2012. Turbulence modeling of deep dynamic stall at relatively low Re. J. Fluids Struct. 33, 191209.
Wernert, P., Geissler, W., Raffel, M., Kompenhans, J., 1996. Experimental and numerical investigations of dynamic stall on a pitching airfoil. AIAA J. 34, 5.
Zanotti, A., Gibertini, G., 2012. Experimental investigation of the dynamic stall phenomenon on a NACA 23012 oscillating airfoil. Proc. Inst. Mech. Eng. Part
G, 13751388.
Zanotti, A., Gibertini, G., Grassi, D., Spreaco, D., 2013. Wake measurements behind an oscillating Airfoil in Dynamic Stall Conditions. ISRN Aerospace
Engineering 2013, 111.