Control of flow separation around an airfoil at low

Reynolds numbers using periodic surface morphing

G. Jonesa,b , M. Santera , M. Debiasic , G. Papadakisa,∗

a Department of Aeronautics, Imperial College London, SW7 2AZ, UK
b Department of Mechanical Engineering, National University of Singapore, 117608,
Singapore
c Experimental AeroScience, Temasek Laboratories, National University of Singapore,

117608, Singapore

Abstract

The paper investigates experimentally the low Reynolds number flow (Rec =
50, 000) around a model that approximates a NACA 4415 airfoil and the control
of separation using periodic surface motion. Actuation is implemented by bond-
ing two Macro Fiber Composite patches to the underside of the suction surface.
Time-resolved measurements reveal that the peak-to-peak displacement of the
surface motion is a function of both the amplitude and frequency of the input
voltage signal but the addition of aerodynamic forces does not cause significant
changes in the surface behavior. The vibration mode is uniform in the spanwise
direction for frequencies below 80 Hz; above this frequency, a secondary vibra-
tion mode is observed. The flow around the unactuated airfoil exhibits a large
recirculation region as a result of laminar separation without reattachment and
consequently produces relatively high drag and low lift forces. Various actuation
frequencies were examined. When actuated at Vf + = 2.0, the spectra in the
vicinity of the trailing edge and near-wake were found to be dominated by the
actuation frequency. Sharp peaks appear in the spectra suggesting the produc-
tion of Large Coherent Structures at this frequency. The increased momentum
entrainment associated with these enabled a significant suppression of the sep-
arated region. The result was a simultaneous increase in CL and decrease in

∗ Corresponding author
Email address: g.papadakis@imperial.ac.uk (G. Papadakis )

Preprint submitted to Journal of Fluids and Structures November 27, 2017

 CD and therefore a large increase in the L/D ratio. In addition, a delay in the
onset of stall results in a significant increase in the maximum achievable lift.
Keywords: separation control, surface morphing, large coherent structures

1. Introduction

Aerodynamics at low Reynolds numbers, Re, is an increasingly active area of

research [1]. Primarily, this is a consequence of the increase in real-world engi-
neering applications, such as wind turbines [2, 3], micro air vehicles (MAV)[1],
5 small unmanned aerial vehicles (UAV) [1, 4] and even a potential system for
exploring Mars [5]. Another reason for the interest in low Re aerodynamics is
the enormously rich flow physics in this flight regime.

1.1. Low Reynolds number aerodynamics

At sufficiently low Re, which according to the classical review of Lissaman

10 [6] can be anything below Re = 500, 000, laminar boundary layers form on
an airfoil’s upper surface which persist beyond the suction peak and into the
pressure recovery region whereupon an adverse pressure gradient (APG) is en-
countered. Viscous effects close to the airfoil’s surface slow down a fluid element
thereby reducing its kinetic energy. Turbulent boundary layers can compensate
15 for this by mixing the low momentum fluid close to the wall with high momen-
tum fluid at the edge of the boundary layer. However, a laminar boundary layer
has no mechanism for re-energizing the near-wall flow, making it incapable of
overcoming even modest APG and therefore highly susceptible to separation
[6, 7].
20 The flow around an airfoil operating at low Re is, to a large extent, con-
trolled by the development of the separated shear layer (SSL) [8, 9]. De-
pending on the angle of attack and Reynolds number, two different scenar-
ios are encountered: development of a Laminar Separation Bubble (LSB) with
turbulent reattachment [7, 10, 11] or massive separation without subsequent
25 reattachment[7, 8, 12, 13]. Flow separation results in low lift, CL , and high

2
 drag CD coefficients, leading to reduced performance of the engineering devices
mentioned above. Control techniques could counter such unfavorable flow con-
ditions and potentially lead to considerable performance improvements.

1.2. Flow control

30 Since separation is associated with significant performance losses, its mitiga-

tion becomes important. Separation can be affected by both passive and active
devices. Passive control has the benefit of requiring no additional power. Two
of the most popular methods are vortex generators (VG) and boundary layer
trips (or Turbulators). Gopalarathnam et al. [14] experimentally investigated
35 the use of the latter and discovered that boundary layer trips can have a positive
net effect on CD . Using a similar concept (surface roughness), Mueller and Batil
[7] found an improvement in the lift-curve slope of a NACA 663 − 018 airfoil
at Re = 40, 000, while Zhou and Wang [15] managed to reduce the presence
of LSB with leading edge bumps and consequently improve L/D. In addition,
40 VG have also been found to have a positive effect of low Re airfoil performance
[16, 17, 18].
The drawback of passive control is that it cannot adapt to changing flow
conditions. Although Gopalarathnam et al. [14] found that trips have a net
drag benefit, the performance of this flow control method was only optimum for
45 a given flight condition. Similarly, while VG were found to improve performance
at certain conditions, they have also been found to exhibit a drag penalty at
others [16, 18]. A similar problem was reported by Mueller and Batil [7] who
found the same surface roughness elements that improved performance at Re =
40, 000, simply increased drag at high Re, where laminar flow exists over a
50 smaller portion of the airfoil surface. Abbas et al. [19] concluded that VG
were most suitable when the separation location was fixed so they could be
optimally positioned upstream of this location. Similarly, for boundary layer
trips Gopalarathnam et al. [14] agree by stating that for a given airfoil, a single
trip location is not optimum for different flight conditions.
55 Therein lies the drawback of passive control; a method that improves per-

3
 formance at one flight condition is likely to degrade performance at another.
On the other hand, active approaches, while requiring additional power, offer
the significant advantage of being aerodynamically innocuous when they are
inactive. Some of the earliest active control devices were based on suction and
60 blowing. In principle these are capable of improving performance, but in prac-
tice complex auxiliary compressors and extensive plumbing systems result in
large power consumption that can often offset the aerodynamic benefits [20].
Seifert et al. [21] discovered that by modulating a steady blowing technique
with a periodic jet, a power saving of 84% could be achieved.
65 The idea of active control based on periodic forcing originates from the
discovery that the formation of Large Coherent Structures (LCS) is accelerated
and regulated by periodic motion [20, 21]. Moreover, LCS have been found
to be the essential building blocks of the mixing layer and are responsible for
transporting momentum across the layer [22]. The production and manipulation
70 of LCS is therefore an efficient way to control mixing in a separated shear
layer. Periodic motion can increase the transfer of high momentum fluid by
enhanced entrainment due to LCS and therefore reduce significantly the size
of recirculation zones. This is the fundamental mechanism that explains the
control authority of periodic actuation. The frequency of actuation is important
75 and an optimum range of reduced frequencies F + = f Xte /U∞ (where Xte is the
distance of the actuator to the trailing edge) between 2-4 has been identified as
optimum to prevent separation [20].

1.3. Scope of this paper

The review paper [23] presents various actuator types used in low-to-moderate
80 speed flows. In recent years, synthetic-jet [22, 24, 25] and plasma-based actu-
ators [26, 27] have emerged as the most popular devices. Their effectiveness is
largely determined by the receptivity of the flow to the imposed disturbances.
These actuators have to be of the right scale and, most importantly, introduced
at the right location (upstream of the separation point). Each type of actua-
85 tor has advantages and disadvantages (refer to table 1 of [23]). For example,

4
 synthetic jets require no external fluid source, but peak velocities are typically
limited to low to moderate subsonic speeds. The orifices are also potentially
subject to fouling. Plasma-based actuators are easily installed on models, but
require high voltage and have low conversion efficiency from input energy to
90 mechanical output energy [28].
This paper explores an alternative approach that is based on periodic mor-
phing of the suction surface. This method belongs to the moving object/surface
type of actuation, that can take various forms, but the most common are piezo-
electric composite flaps and electro-active dimples [23, 29, 30]. Upon application
95 of a voltage, the dimples produce unsteady surface depressions that interact with
the near-wall structures. In the present work, Macro Fibre Composites (MFC)
actuator patches are employed that provide distributed periodic forcing in the
suction side of the airfoil. These thin, light, and flexible piezoelectric actua-
tors provide distributed forcing so their performance is robust to changes in the
100 angle of attack that affects the location of the separation point. Some studies
have explored the use of such actuators to change or control the shape of aero-
dynamic bodies. Munday & Jacob used THUNDER actuators (a predecessor
of the MFC type) to increase the lift of an airfoil [31]. Their morphing airfoil
model is based on a prototype made by Pinkerton and Moses [32]. Static and
105 dynamic morphing tests were conducted at angles of attack from 0◦ to 9◦ at low
Rec = 25, 000 and 50,000. Static morphing shows modest aerodynamic benefits
with best increase of the L/D ratio of about 2%. However, dynamic morphing
significantly reduces the size of flow separation therefore greatly increasing the
L/D ratio. Bilgen et al. [33] investigated the use of MFCs to change the camber
110 of a symmetric airfoil. The progress of this research led to the fabrication and
flight demonstration of a small aircraft with solid-state control surfaces [34].
Subsequently Bilgen and Friswell [35] developed a novel design for a variable-
camber airfoil employing a continuous inextensible surface with bonded MFC
actuators which can achieve significant change in aerodynamic response. Moli-
115 nari et al. [36] investigated numerically and experimentally the aerodynamic
and structural performance of a wing morphed by MFCs. The wing was tested

5
 in a small unmanned aircraft flight and proved that roll controllability can be
achieved exclusively through morphing. The results show the possibility of re-
placing the conventional ailerons of similarly sized airplanes with the proposed
120 solution, achieving significant efficiency improvements while guaranteeing con-
trollability.
The paper is structured as follows. In Section 2 the model fabrication and
morphing implementation is presented, followed by the characterization of the
response of the skin to different amplitudes and frequencies in Section 3. The
125 experimental methodology for the aerodynamic flow measurements is outlined
in Section 4. Results for the unactuated and actuated airfoil for various angles
of attack are presented in Sections 5 and 6 respectively. The energy efficiency
is considered in Section 7 and the main findings of the paper are summarised in
Section 8.

130 2. Airfoil model fabrication and implementation of morphing

2.1. Model fabrication

A physical model, shown in Fig. 1, capable of performing dynamic surface

morphing was fabricated. The leading edge, trailing edge and pressure surface
were made from PVC and were machined to match a NACA 4415 profile to
135 within ±0.1 mm. The suction surface was fabricated separately to accommodate
the desired surface deformations and was rigidly attached to the body of the
airfoil at 0.07c downstream of the leading edge. At the trailing edge, a thin slide
joint was used to allow for small displacements in the longitudinal direction that
occur when the surface is deformed.

6
 (a) (b)

‫ݕ‬

‫ݔ‬

(c)

(d)

Figure 1: Physical model for wind tunnel testing: a) underside of the suction surface; b)
complete model; c) schematic of actuating motion; d) model compared with true NACA 4415
coordinates (dashed line).

140 To perform the deformation of the suction surface, two MFC patches (Smart
Material M-8557-P1) were bonded to a 0.127 mm (0.005 in) thick sheet of Ti-
tanium with a slow-drying epoxy resin. A vacuum bag was placed over it to
remove any pockets of air between the metal surface and the MFC patches and
to ensure a strong, clean bond. It was then left for twelve hours to set in a mold
145 with curvature equal to that of the suction surface of the airfoil. Electrical
connections were soldered and then insulated using a fast drying epoxy. The
resulting skin can be seen in Fig. 1a. The fully-formed model has a chord, c, of
150 mm and span, s, of 158 mm and can be seen in Fig. 1b. A schematic of the

7
 actuating motion and the origin of the coordinate system can be seen in Fig.
150 1c.
Fig. 1d shows the cross-sectional view of the non-actuated model with true
NACA 4415 coordinates plotted as a dashed line for comparison. Separately
fabricating the suction surface appears to have resulted in a small deviation
compared to a true NACA 4415. The dimensional uncertainty between two
155 points of the model profile is about ±0.2 mm as obtained from an undistorted,
high-resolution, side-view image of the model. The fabricated model had maxi-
mum thickness 14.2% and camber 3.3% (compared to 15% and 4% respectively
for a NACA 4415). The maximum curvature was 47.8 /mm (in the leading
edge) and the angle at the trailing edge was 20.36◦ , values close to 47.21 /mm
160 and 19.56◦ respectively for a NACA 4415.

2.2. Actuation procedure

Surface morphing is caused by the deformation of the MFC patches. These

actuators have three main components: 1) a sheet of aligned fibers of piezoelec-
tric material, 2) a pair of thin polymer films etched with a conductive electrode
165 pattern and 3) an adhesive matrix material [37]. The piezoelectric material cou-
ples mechanical and electrical fields, so the fibers deform when they experience
an electrical field [38]. Within an MFC, the electrical field is provided by the
electrodes and due to the strong bond between the MFC and the Titanium skin,
a voltage supply will cause the airfoil surface to deform.
170 Due to their high electro-mechanical coupling factor and piezoelectric coef-
ficients, piezoceramic materials are commonly employed to form actuators [39].
Specifically, MFC use Lead Zirconate Titanate (PZT) because of its outstand-
ing piezoelectric properties. The drawback of PZT is that apart from being
piezoelectric, it is also ferroelectric. This is a result of requirements on the
175 electrical polarisation for macroscopic piezoelectric effects [40] and means that
MFC exhibit nonlinear behavior such as hysteresis and creep.

8
 (a)

(b)

(c)

Figure 2: Effect of actuator nonlinearities during four cycles through their operational range:
a) surface displacement-vs-Voltage; b) input signal; c) two different examples of the surface
response (note that the different starting positions converge to the same value).

A clear illustration of this can be seen in Fig. 2a. The MFC were driven
with the voltage signal shown in Fig. 2b, which cycled through their operational
range four times. The displacement of the surface throughout these cycles was
180 measured (at 40% from the leading edge and at mid-span, z/s = 0.5) by a
laser displacement sensor (to be introduced below in Section 3). The origin in
Fig. 2a (marked by a triangle) denotes the reference position of the non-actuated
surface before the cycles begin (i.e. at 0 V). In this paper, the non-actuated

9
 surface is considered as the origin and a displacement in the outward direction
185 is considered positive.
It is evident from Fig. 2a that during the actuation cycles the surface location
at 0 V deviates from the reference position, and the actual deviation depends on
how the value of 0 V is approached. It is equal to +0.6 mm (+4.0×10−3 c) when
the voltage grows from negative to positive and -0.6 mm ( −4.0×10−3 c) when the
190 voltage is reduced from positive to negative, giving a total positional change of
1.2 mm (8.0×10−3 c). Furthermore, the final resting place of the surface (marked
by a circle), which follows a more gradual approach to 0 V to avoid an abrupt
halt, is a further -0.5 mm (−3.3×10−3 c) away from the origin. This means the
non-actuated surface deviates by approximately -1.1 mm (7.3×10−3 c) post-cycle
195 compared to pre-cycle.
Another nonlinear effect noticeable in Fig. 2a is creep. The voltage signal
pauses at the maximum positive and negative values for one second during
each cycle. During this time, the surface is observed to move by 0.13 mm
(0.87×10−3 c) and 0.02 mm (0.13×10−3 c) at -500 V and 1500 V respectively.
200 Fig. 2c shows two separate samples of the time-resolved displacement (both
acquired at x/c = 0.4, z/c = 0.5) during the hysteresis cycles. At the beginning
of the cycles, the two samples differ by roughly 0.6 mm (+4.0×10−3 c). Such
an initial displacement can result from a different approach to 0 V in a previ-
ous experiment (for example at a different actuation frequency), or because of
205 surface creep. As the cycles progress, the positions of the measurement points
converge. After four cycles both signals have coalesced and therefore the surface
position can be made independent of the preceding history.
For measurement repeatability, it is essential that the airfoil morphing is the
same between different experiments and therefore such hysteresis cycles must
210 be included in the input signal. This is explained in more detail in the next
section.

10
 2.2.1. Input signal
The input voltage is applied to the MFC via a Smart Material HVA 1500/50-
2 high-voltage amplifier. This unit has a gain of 200 V/V and can accept input
215 signals from DC to 10 kHz AC in the range of -2.5 V to 7.5 V. These are
amplified to -500 V and 1500 V respectively; these values correspond to the
operational range of the actuators.
The input signal is provided to the amplifier as a user-specified text file con-
taining discrete voltage values, via a digital-to-analog converter. This allows for
220 flexibility in the signal’s form, which is necessary if, for example, the hysteresis
cycles are to be included.

Figure 3: Sample of complete input signal that consists of an initial model preparation phase
followed by the main phase (for data acquisition).

An example of the input signal used for dynamic actuation (at Vf = 2 Hz) is
shown in Fig. 3. The signal initially goes through four hysteresis cycles to ensure
a consistent airfoil geometry. This is followed by the main actuation phase in
225 which the signal oscillates at the desired frequency. The gradual increase and
decrease of the amplitude at the beginning and end of the main actuation is
included to make the transition to an oscillating surface as smooth as possible
in an effort to reduce possible damage to the model. The data acquisition

11
 was obtained during the period of the main actuation phase when the surface
230 oscillates with a constant amplitude. This part of the input signal is described
by

V (t) = VA sin(ωt) (1)

where VA is the amplitude, measured in Volts, and the angular frequency, ω, is

defined as 2πVf where Vf is the actuation frequency in Hz.

3. Surface morphing measurements

235 A series of tests were carried out to understand the response of the skin to
different amplitudes, VA , and frequencies, Vf . Two complementary techniques
were used to measure the surface displacement: a laser displacement sensor and
photogrammetry.
A Micro-Epsilon optoNCDT 1710-50 laser displacement sensor was used to
240 measure the deformation of the skin at a point. This unit has a measuring
resolution of 5 µm, an accuracy of 50 µm and it is optimally positioned at a
distance of 550 mm to 600 mm from its target. This range enables the placement
of the sensor outside the wind tunnel, thus allowing measurements of the skin
deformation in still or flowing air (see Fig. 4a). The measurements were acquired
245 at 1.25 kHz, providing time-resolved information of the surface motion.

12
 (a) (b)

Figure 4: Methods of surface displacement measurement: a) laser sensor setup outside the
wind-tunnel test section (not to scale); b) coded targets of the airfoil suction surface for
photogrammetry measurements.

The laser displacement sensor can only provide information at a single point
in one dimension. This limits the amount of analysis that can be done on the
surface motion. In contrast, photographs can provide a view of the whole surface
and, if captured at high speed, information regarding the dynamic behavior
250 can be obtained. To this end, quantitative information was extracted from
photographs (such as that of Fig. 4b) using the method of photogrammetry. A
proprietary software package, ‘Photomodeler,’ published by Eos Systems Inc.
[41] was used for this purpose.
A Photron FASTCAM SA-Z capable of producing mega-pixel (1024×1024)
255 images at up to 20,000 fps, was used to capture snapshots of the airfoil under-
going actuation. In total, 99 binary coded targets (seen in Fig. 4b) were placed
on the surface of the airfoil (9 across the span and 11 in the streamwise direc-
tion). Photomodeler [41] can track the motion of these targets and individually
distinguish them from their binary coding. The four points at the corners of
260 the scene (away from the airfoil) are used in the post-processing to provide a
suitable coordinate system.
It is not possible to obtain 3D information from a single image, since there
is no sense of depth. If a point is captured in more than one image, an algebraic
representation of its corresponding light rays can be found on each image. Solv-

13
265 ing for the intersection of these rays then provides full 3D information. It was
therefore necessary to ensure that each one of the targets in this experimental
setup was imaged multiple times from different angles. For this reason, 8 sepa-
rate sets of images were captured at angles separated by roughly 45◦ from each
other.
270 When a single point in 3D space is captured in two separate images the
intersection of the light rays can be determined by epipolar geometry [42]. The
epipolar geometry depends only on the camera’s internal properties and rela-
tive positions of the camera at each viewing angle. The internal properties are
obtained from a camera calibration process prior to the experiment, leaving rel-
275 ative position as the only unknown in the algebraic representaion of the epipolar
geometry and therefore easily solved for. The mathematics is beyond the scope
of this paper and the reader is referred to the book ‘Multiple View Geometry in
Computer Vision’ by Hartley & Zisserman [42] for details. However, since the
relative position is calculated during the solution procedure, knowledge of the
280 camera’s position at each viewing angle is not required.

3.1. Dynamically actuated surface behavior

The leading and trailing edge of the morphing section of the surface are lo-
cated at 0.07c and 0.93c respectively thus creating a distributed control surface
that can provide periodic forcing over 86% of the chord. The time-dependent
285 displacement of the skin was recorded at three locations: 0.25c, 0.4c and 0.7c.
Fig. 5 shows the peak-to-peak displacements. In order to quantify how aerody-
namic forces due to air flow (mainly pressure distribution) over the surface affect
the response of the skin, measurements were taken in both still and moving air
at Rec = 5 × 104 . It is clear that both amplitude and frequency of the input
290 signal have a significant effect on the dynamic behavior of the skin, whereas the
effect of the aerodynamic forces is one order of magnitude smaller. As a general
rule, the higher the amplitude of sinusoidal voltage, the greater the peak-to-
peak displacement. At frequencies above 30 Hz the amplitude of displacement
drops significantly at all measured locations. This was expected and has been

14
295 observed in previous models [43, 44]. It is attributed to resonance occurring at
around 30 Hz. However, due to a change in the vibration mode, the amplitude
begins to increase again at around 80 Hz. This issue is further examined in the
next section, where the 3D surface measurements are analyzed.

(a) (b)

(c)

Figure 5: Peak-to-peak displacements (as a percentage of c) of the skin at α = 0◦ with moving

(Dashed) and still (Solid) air: a) x/c = 0.25; b) x/c = 0.40; c) x/c = 0.70.

The laser measurements were also used to validate how the photogrammetry
300 captures the dynamic surface motion. Time traces of 10 cycles at 10 Hz, 20 Hz,
40 Hz and 80 Hz are compared with the laser displacement data in still air in
Fig. 6.

15
 (a)

(b)

(c)

(d)

Figure 6: Comparison between Laser Sensor and Photogrammetry displacement measurements

at x/c = 0.4; z/c = 0.5 (Displacements given as a percentage of c): a) 10 Hz; b) 20 Hz; c) 40
Hz; d) 80 Hz.

16
 The results are in very good agreement, which serves to validate the photogram-
metry data. The only small difference appears at the negative peak at 40 Hz.
305 It is possible that the large displacement at 40 Hz has caused a larger than
expected angle between the surface and the sensor’s receiver, causing a small
error in the laser measurement.
In all cases the surface moves more in the negative (inward) direction than
in the positive direction. This is because the maximum positive voltage (1500
310 V) elongating the MFC fibres is larger than the minimum negative voltage (-500
V) contracting them. Thus the inner side of the skin to which the MFC patch is
bonded typically undergoes larger stretching than compression which results in
larger inward than outward bending. Furthermore, the motion appears periodic
in all cases but not perfectly symmetric, particularly at lower frequencies. The
315 peak-to-peak amplitude increases from 10 Hz to 40 Hz and is much lower at
80 Hz, in agreement with Fig. 5.

3.2. 3D surface measurements

Figs. 7-9 show the wall-normal modal behavior obtained by Proper Orthog-
onal Decomposition (POD) of the displacement snapshots from 64 cycles. Only
320 the chord-normal displacement was considered since it was the dominant one
and the most important from the point of view of the application to flow con-
trol. The POD was performed on the snapshots of the y-coordinates and the
resulting modes were projected onto the undeformed x and z-coordinates.
At 10 Hz, almost all of the energy (99.6%) is contained within the first
325 vibration mode, as seen in Fig. 7a. The mode shapes, represented in 2D and 3D
in Figs. 7b and 7c respectively, show that the motion is symmetric about the
center-chord and largely uniform in the spanwise direction. There is a slightly
greater displacement at the far side of the airfoil, towards z = 60 mm, but
Fig. 7c suggests this is minor and can be ignored.

17
 (a)

(b) (c)

Figure 7: Vibration modes when Vf = 10 Hz: a) energy content of the first 20 modes; b) 2D
representation of first mode shape; c) 3D representation of first mode shape.

330 When Vf is increased to 40 Hz (Fig. 8), the motion is still uniform in the
spanwise direction and symmetric about the mid-chord. The energy in the first
mode decreases very slightly as Vf increases but still remains very high, at
99.2%.
However, doubling Vf to 80 Hz (Fig. 9) causes a large change in the modal
335 behavior. Firstly, the shape of the first mode is quite different. It is no longer
symmetric since the location of maximum displacement has moved towards the
slide joint at the trailing edge. This is consistent with the increase in peak-to-
peak displacements at x/c = 0.7 at high frequencies observed in Fig. 5. Fur-
thermore, the motion appears more three-dimensional in the spanwise direction.
340 Again it is the z = 60 mm side of the airfoil that has the greater displacement.
This suggests that the MFC are either not perfectly aligned or do not behave
in exactly the same way. Animation of the surface motion at this frequency
reveals that the three-dimensionality is still small. Aside from the change in the
shape of mode 1, there has also been a change in the energy content, which has

18
345 dropped to 88.3%. This is still a large portion of the total energy. However, a
second vibration mode has been triggered at this frequency that contains more
than 10% of the energy. This mode shape exhibits two peaks and is skewed in
the z-direction since the peaks have a maximum at opposite sides of the airfoil
in the spanwise direction.

(a)

(b) (c)

Figure 8: Vibration modes when Vf = 40 Hz: a) energy content of the first 20 modes; b) 2D
representation of first mode shape; c) 3D representation of first mode shape.
]

19
 (a)

(b) (c)

(d) (e)

Figure 9: Vibration modes when Vf = 80 Hz: a) energy content of the first 20 modes;
b) 2D representation of first mode shape; c) 3D representation of first mode shape; d) 2D
representation of second mode shape; e) 3D representation of second mode shape.

350 Having quantified the effect of actuation on surface motion, the following
section will present the experimental methodology.

4. Experimental methodology

The aerodynamic characteristics of the model were measured in the closed-

loop, subsonic wind tunnel, shown in Fig. 10a, at the Temasek Laboratories of
355 the National University of Singapare (NUS). The tunnel has a 600 mm square

20
 test section with a length of 2 m. The test section is connected to the exit of the
wind tunnel nozzle with 12:1 contraction ratio. The turbulence intensity level
of the wind tunnel freestream is less than 0.5% at velocities of approximately
5 m/s, which corresponds to Rec = 50, 000. The model was mounted vertically
360 on a turntable in the test section floor and a splitter plate was installed 160 mm
above it. The turntable allows precise positioning (within ±0.05◦ ) of the angle
of attack of the model, α. A boundary layer ingestion slot with a sharp leading
edge spanning 72% of the test section width was utilized to maintain a floor
boundary layer roughly as thin as the one on the surface of the splitter plate.
365 The leading edge of the model was positioned 400 mm downstream of the leading
edge of the ingestion slot, at which location the boundary layer thickness of the
empty test section with U∞ = 5 m/s has been measured to be less than 1 cm.
The mounting can be seen in Fig. 10b.

(a) (b)

Figure 10: (a) Closed-loop, subsonic wind tunnel at the Temasek Laboratories, NUS, and (b)
the vertical mounting of the airfoil model beneath a splitter plate (flow direction is normal to
the page).

The coordinate system used in this paper are the Cartesian coordinates of
370 the wind tunnel test section. Refering to the Fig. 10b, the flow direction is into
the page and used as the reference x-direction, measured from the airfoil leading
edge (see also figure 1c). As the model is vertically mounted, the vertical test
section direction, z, is the spanwise coordinate measured from mid-chord.
The turntable on which the model was mounted incorporated a force balance
375 consisting of a ATI Gamma SI-32-2.5 piezoelectric gauge capable of measuring

21
 the forces and moments along three perpendicular axes. Two axes, aligned with
the axial (chordwise) and the normal coordinates of the airfoil, were used to
measure the axial and the normal forces generated by the model. The third axis,
coinciding with the axis of rotation of the turntable and aligned in the vertical
380 direction of the tunnel, passed through the airfoil mid-chord point (c/2) and
was used to measure the pitching moment. The balance was factory calibrated
and the corresponding conversion factors were stored in the acquisition unit, so
the values of the forces and moments were already corrected. The range (and
resolution) of the measured forces and moments are 32 (±6 × 10−3 ) N and 2.5
385 (±5×10−4 ) Nm, respectively. For each measurement, samples were acquired for
10 seconds, at a minimum of 16 samples per actuation cycle (1.25 kHz in cases
without actuation). The lift and the drag coefficients, CL and CD respectively,
were calculated at each α from the corresponding values of the chordwise and
normal forces.
390 Flow-field velocity measurements were obtained using a two-velocity-component
PIV system. The flow was uniformly seeded with water-based particles from a
SAFEX fog generator. Droplets were produced with an average Sauter mean
diameter (SMD) of 1 µm. Their reflections correspond to no more than 3 pixels
in the captured images, allowing a good resolution of the particle displacement
395 when cross-correlating images. A dual-head Litron DualPower 200-15 Nd:YAG
laser operating at the second harmonic (532 nm) at approximately 150 mJ per
pulse was used in conjunction with sheet-forming optics to form a thin sheet
(≈1 mm) on the x-y plane at 70% along the airfoil span. Two images cor-
responding to the pulses from the laser were acquired by a 2048×2048 pixels
400 HiSense 620 camera which viewed the streamwise laser sheet orthogonally over
the entire field of view.
A computer with dual Intel Core processors was used for data acquisition.
The images were processed using Dynamic Studio from Dantec Dynamics. For
each image, subregions were adaptively cross-correlated using multi-pass pro-
405 cessing with a final 50% overlap. The final interrogation window size was 32×32
pixels giving a final spatial resolution of 3.52 mm. The minimum seeding den-

22
 sity was 3-5 particles per interrogation area. The time separation between the
two laser flashes was varied between 100 and 250 µs such that the maximum
displacement of a particle in the region of interest was no more than 25% of the
410 interrogation window. This was optimum for the PIV processing software to
calculate accurately the particle velocity. The resulting vector fields were post-
processed to remove remaining spurious vectors. A Zeiss 50 mm f/2.0 macro
lens provided a 225 mm×225 mm field of view. This equates to a velocity vector
grid of 127×127 points with resolution of approximately 110 µm per pixel. For
415 each acquisition 300 images were captured at 5 Hz for statistical averaging.
Profiles of the wake velocity were also obtained at the mid-span location,
1.25c, 1.50c and 1.70c downstream of the leading edge using a single, miniature-
wire hotwire probe. 217 samples at 6 kHz were acquired at 49 different cross-
stream locations between ±0.43c in the cross stream direction. The velocity
420 was calibrated with a 1% error using a dedicated Dantec Dynamics velocity
calibrator.

5. Unactuated airfoil

5.1. Time-averaged results

Iso-contours of the streamwise velocity for α = 0◦ are presented in Fig. 11a,

425 while the time-averaged, spanwise vorticity, Ωz , is shown in Fig. 11b. The vor-
ticity contours show Ωz is concentrated exclusively in a thin, separated shear
layer between x/c = 0.65 and x/c = 0.8 with a very slow moving, quiescent flow
occurring beneath. From descriptions by Tani [45] and Horton [46], this region
exhibits the characteristics of a laminar separated shear layer. Such a region
430 results in a pressure plateau on the airfoil suction surface and a reduction of the
suction peak [11] and CL . Further downstream, the magnitude of Ωz decreases
and the region of higher vorticity becomes more dispersed. This indicates that
unsteadiness sets in beyond x/c = 0.8 resulting from the breakdown of the
shear layer. Fig. 11a shows the magnitude of the reverse flow increases at this
435 point and the local velocity vectors reveal a recirculating vortex is present in –

23
 and confined to – this region. This recirculation zone was also observed in the
experiments by Yarusevych et al. [11] as upstream entrainment of smoke, ema-
nating from a smoke wire at the trailing edge. The recirculating flow structure
formed at the termination of a laminar shear layer resembles the well-known
440 reverse-flow vortex found in the turbulent portion of an LSB [46]. This suggests
the breakdown and dispersion of high vorticity in the shear layer is a result of
turbulent transition and a pressure recovery would occur in the region of the
recirculating fluid. More details about the transition are provided below when
the velocity spectra are examined.

(a) Iso-contours of u/U∞ (b) Iso-contours of Ωz .

Figure 11: Iso-contours of time-average streamwise velocity and spanwise vorticity obtained
from a field of view focused on the separated region of the non-actuated airfoil (α q
= 0◦ ). In
(b), the arrows indicate local velocity vectors (with vector length proportional to |U |/U∞
so that small velocities can be more easily seen).

445 5.2. Spectral results

The frequency spectra in the wake were obtained from the hotwire signal
at the y/c location corresponding to half the velocity deficit at a particular
streamwise location. This signal of 217 samples, sampled at 6 kHz, was divided
into signals of 213 samples using a Hanning window with 50% overlap. The
450 average of the spectra at three streamwise locations x/c = 1.25, 1.50 and 1.70
are plotted in Fig. 12a. The spectra are displaced by a factor of 10 for clarity
of presentation, with the x/c = 1.25 spectrum as the reference. The frequencies
are normalized using the chord length, c, and the approaching velocity, U∞ ,
as Stc = f c/U∞ . A very strong peak is observed, centered at Stc = 2.4
455 (corresponding to fw = 85Hz). This frequency is considered to be the vortex

24
 shedding frequency in the wake. The peak is very prominent close to the airfoil,
but further downstream loses its distinct character.

(a) (b)

Figure 12: Frequency spectra of the non-actuated airfoil at α = 0◦ : a) 3 locations in the near
wake; b) 3 locations close to the trailing edge (the dotted line signify the first subharmonic).

LCS have also been identified in the shear layer [11, 47, 48]. The frequency
spectra at the y/c location with largest r.m.s. velocity are plotted in Fig. 12b
460 at 3 points close to the trailing edge. Owing to the length of the hotwire probe
support, location x/c = 0.8 is as far upstream as measurements could be taken;
unfortunately it was not possible to obtain measurements in the purely laminar
portion of the separated region.
A distinct peak at approximately Stc = 4.8 (corresponding to fs = 170Hz),
465 marked with a solid line in Fig. 12b, appears at x/c = 0.8, 0.9 and 1.0. Such
distinct peaks are attributed to the instability of the shear layer leading to
vortex roll up [49, 11]. The appearance of a subharmonic and higher harmonics
(the former shown in Fig. 12b by dashed lines) indicate nonlinear interactions

25
 that eventually lead to transition and full turbulence [50]. As the shear layer
470 transitions, the roll-up vortices break up [11, 51], the spectral peak at Stc = 4.8
is eroded (compare spectra at x/c = 0.90 and 1.0), and a typical turbulent
power-law spectrum with a slope of −5/3 appears.
The evolution of the spectral peaks from x/c = 0.8 to x/c = 1.70 reveals an
interesting result. The peak at Stc = 2.4 found in the wake, and attributed to
475 the shedding of coherent vortex structures, is actually the first subharmonic of
the shear layer frequency, which results from shear layer roll-up. Yarusevych et
al. [11] observed similar behavior and attributed the subharmonic to the merging
of the shear layer vortices.
The spectra in Fig. 12b tie in well with the observations in Fig. 11b. The
480 termination of the region of strong, concentrated vorticity of the shear layer (at
x/c = 0.8) coincides with the appearance of higher harmonics in the spectra,
which indicates nonlinear interactions and the onset of transition. The typi-
cal turbulent spectra at the trailing edge are consistent with the location of
a reverse-flow vortex in Fig. 11a which, according to Horton [46], should be
485 confined to the turbulent portion of the flow.

6. Actuated airfoil results

It was shown in Fig. 5 that the peak-to-peak displacement of the surface was
a function of both Vf and VA . Since only the effect of frequency is investigated,
VA was modified for each Vf examined to maintain a constant peak-to-peak
490 displacement of 0.27% of the chord at x/c = 0.4. This is comparable to the
value of Munday & Jacob [31] of 0.2% of the chord. Furthermore, it was shown
that a second vibration mode takes effect when Vf > 80 Hz so the actuation fre-
quencies were kept below this value. More specifically, the actuation frequencies
investigated are Vf = 10 Hz, 40 Hz and 70 Hz. Throughout the discussion of
495 results, these frequencies will be referred by their normalised form, Vf + , defined

26
 as:

Vf c
Vf + = (2)
U∞

The normalised frequencies are therefore Vf + = 0.3, 1.1 and 2.0.

6.1. Time-averaged results

6.1.1. Effect of actuation on time-averaged flow fields

500 The flow fields in Fig. 13 show time-averaged contours of u/U∞ when α = 0◦
for four frequencies. The non-actuated case in Fig. 13a is the same as the
baseline cases in Fig. 11a. This should be used as reference when comparing
with the actuated cases from (b), (c) and (d), which display the flow fields when
the surface was actuated at Vf + = 0.3, 1.1 and 2.0 respectively.

(a) (b)

(c) (d)

Figure 13: Iso-contours of u/U∞ around the airfoil when α = 0◦ with field of view focused on
separation region: a) Vf + = 0.0; b) Vf + = 0.3; c) Vf + = 1.1; d) Vf + = 2.0.

505 Comparing the baseline with the actuated cases for the two higher frequen-
cies, Vf + = 1.1 and 2.0, it is clear that the separating shear layer is closer to

27
 the wall and there is a noticeable reduction in the size of the reverse flow region.
This reduction is expected to result in a significant performance improvement.
There is, however, little difference in the size of this region when Vf + = 0.3
510 (although there are some localised differences with respect to the baseline case).
It is expected that since the size of the separated region is almost the same,
actuation at Vf + = 0.3 will have little effect on the performance of the airfoil.
Note that such differences on the location of the separating shear layer and the
size of the recirculation zone cannot be obtained using hotwire anemometry. For
515 this reason, we compare results obtained by PIV with a field of view focused on
the separated region in the aft part of the airfoil.

(a) (b) (c)

Figure 14: Profiles of Ue /U∞ in the wake of the airfoil at 4 actuation frequencies: a) 1.25c,
b) 1.50c, c) 1.70c.

Profiles of the time-averaged velocity in the wake were taken at three stream-
wise locations x/c = 1.25, 1.50 and 1.70. Since this is a single-wire hotwire
probe, no sense of the flow direction is provided. Therefore the profiles in
520 Fig. 14 correspond to the velocity flowing perpendicular to the wire, Ue . The
profiles reveal the cross stream extent of the wake and the magnitude of the
velocity deficit. The latter is related to the profile drag. For the unactuated
case, the width of the wake is comparable to the airfoil thickness (0.13c). This
is qualitatively the same as the wide wake observed by Yarusevych et al. [52] in

28
525 their case of separation without reattachment. The profiles are unchanged when
Vf + = 0.3, while significant reductions in the velocity deficit at all streamwise
locations occur when Vf + = 1.1 and 2.0. This adds weight to the hypothesis
that Vf + = 1.1 and 2.0 are capable of significant performance improvements
while Vf + = 0.3 has little effect. The reduction in the velocity deficit and
530 wake width for Vf + = 1.1 is not as large as when Vf + = 2.0. It would
therefore be expected that the performance recovery for the latter Vf + will be
more substantial than for the former.

6.2. Spectral results

The spectra shown in Fig. 12 provided insight into the presence of LCS in the
535 shear layer and near wake. Figs. 15 and 16 show the spectra for the actuated
airfoil. When Vf + = 0.3, a strong peak is present at Stc = 0.3 in the
shear layer (Fig. 15b) and in the near wake (Fig. 16b) meaning that the flow is
responding to the surface actuation and flow structures, centred the actuation
frequency, are being produced. Spectral peaks at the naturally occurring shear
540 layer and vortex shedding frequencies remain as clear as before, indicating that
the flow still has a strong tendency to develop in its natural (i.e. non-actuated)
way. This agrees with the observation that the flow field and wake profiles are
almost the same when Vf + = 0.0 and 0.3.
When Vf + = 1.1 (Fig. 15c and Fig. 16c), strong peaks at Stc = 1.1
545 (and its harmonics) are present at all locations. Evidence of activity at the
naturally occurring frequencies remains, with corresponding peaks in the shear
layer upstream of turbulent transition. The harmonics are particularly amplified
around these natural frequencies. Such a phenomenon was also observed by Sato
et al. [26] and attributed to the presence of linear unstable modes. It can be
550 considered that actuation at Vf + = 1.1 promotes the development of LCS at
the actuation frequency and therefore modifies the flow development. Natural
frequencies are still strongly amplified in the shear layer but the wake is clearly
dominated by the actuation frequency. This agrees with the reduction in the
size of the wake when Vf + = 1.1 in Fig. 14 but only modest changes in the

29
555 flow field around the airfoil in Fig. 13c.
Actuation at Vf + = 2.0 (Figs. 15d and 16d) causes the spectra at all
locations to be entirely dominated by the actuation frequency and its harmon-
ics. This means that the flow has locked-on to the surface motion, which is
generating LCS at the actuation frequency. These LCS then dominate the sub-
560 sequent flow development. Such structures are known to enhance momentum
entrainment to the near-wall region and therefore suppress separation [26, 21].
This explains the significant reduction in separation and velocity deficit when
Vf + = 2.0 (refer to Fig. 14). It is also worth pointing out that this phenomenon
occurs when Vf + is close to the wake frequency of Stc = 2.4, which is a sub-
565 harmonic of the shear layer frequency. Lock-on to the naturally occurring wake
shedding frequency has been also observed by Zhang and Samtaney [24] when
forcing a separated flow with synthetic jets.

30
 (a) (b)

(c) (d)

Figure 15: Frequency spectra of the actuated airfoil when α = 0◦ at 3 locations close to the
trailing edge: a) Vf + = 0.0; b) Vf + = 0.3; c) Vf + = 1.1; d) Vf + = 2.0.

31
 (a) (b)

(c) (d)

Figure 16: Frequency spectra of the actuated airfoil when α = 0◦ at 3 locations in the wake:
a) Vf + = 0.0; b) Vf + = 0.3; c) Vf + = 1.1; d) Vf + = 2.0.

32
 6.3. Flow Fields at 5◦ , 10◦ and 15◦ .
Flow fields for the actuated airfoil when α = 5◦ , 10◦ and 15◦ are shown
570 in Figs. 17-19. When α = 5◦ (Fig. 17) broadly the same observations can be
made as those when α = 0◦ (Fig. 13). There is no significant reduction in the
size of the separated region when Vf + = 0.3 when compared with the baseline
case while the shear layer seems to reattach when Vf + = 2.0. The case of
Vf + = 1.1 exhibits a modest reduction in separation, more significant than
575 it was when α = 0◦ , but not nearly as large as when Vf + = 2.0. Again, it
is expected that the airfoil performance when Vf + = 0.3 will be similar to
that of the non-actuated case at this angle of attack but improvements would
be expected for Vf + = 1.1 and, more significantly, for Vf + = 2.0.

(a) (b)

(c) (d)

Figure 17: Iso-contours of u/U∞ around the airfoil when α = 5◦ : a) Vf + = 0.0; b) Vf + = 0.3;
c) Vf + = 1.1; d) Vf + = 2.0.

When α = 10◦ (Fig. 18), Vf + = 2.0 again causes a large reduction in the
580 size of the separated region. However, at this angle of attack, Vf + = 1.1 is
also observed to exhibit significant control authority. Actuating at Vf + = 0.3
leaves the velocity field largely unaffected.

33
 (a) (b)

(c) (d)

Figure 18: Iso-contours of u/U∞ around the airfoil when α = 10◦ : a) Vf + = 0.0; b) Vf + = 0.3;
c) Vf + = 1.1; d) Vf + = 2.0.

At the post-stall angle of 15◦ (Fig. 19), the massively separated flow has been
supressed to a large extent when Vf + = 2.0 i.e. the stall has been delayed.
585 Actuation with Vf + = 1.1 does not entirely remove the separated region but it
is noticeably reduced when compared to the baseline flow field. Actuation can
therefore be expected to greatly improve airfoil performance at high angles of
attack.

34
 (a) (b)

(c) (d)

Figure 19: Iso-contours of u/U∞ around the airfoil when α = 5◦ : a) Vf + = 0.0; b) Vf + = 0.3;
c) Vf + = 1.1; d) Vf + = 2.0.

6.4. Actuated airfoil performance

590 The above discussion regarding the flow development around the airfoil high-
lighted reductions in the size of the separated region and velocity deficit in the
wake when Vf + = 1.1 and Vf + = 2.0. In contrast, when Vf + = 0.3 very
little changed compared with the baseline case. It is therefore expected that
improvements in airfoil performance would result from actuation at Vf + = 1.1
595 and, more significantly, at Vf + = 2.0, whereas Vf + = 0.3 would have little
effect. Fig. 20 shows the force coefficients as a function of α and corroborate
this. The values are compared with those of a NACA 4415 at high Rec [53]. For
the unactuated airfoil, a substantial drop in CL is observed for all α compared
to the high Rec values, which is in agreement with the literature [12]. In addi-
600 tion, Fig. 20 shows very large increases in CD for all α when Rec = 50, 000. A
simultaneous increase in CD and decrease in CL causes an order of magnitude
decrease in the L/D ratio.

35
 (a) (b)

(c)

Figure 20: Aerodynamic force coefficients obtained from the actuated airfoil compared to the
baseline case and the high Rec reference case from Jacobs & Pinkerton [53] : a) CL ; b) CD ;
c) L/D.

Also worth noting is the onset of stall, which occurs at α = 10◦ and limits CLmax
to 0.91, 73% lower compared to Rec = 3.11×106 . The CL curves for the non-
605 actuated and Vf + = 0.3 cases are very similar and result from the inability of
actuation to affect the natural flow development, which was shown in Figs. 15b
and 16b. At higher values of Vf , the previously reported reduction in the size
of the separated regions and velocity deficit have translated into significant
improvements in performance. For example, when Vf + = 2.0 the regions of

36
610 separation were reduced significantly at all angles of attack and correspondingly
a significant increase in CL and decrease in CD is observed for all values of α.
Furthermore, Figs. 20a-b show that when Vf + = 1.1 the effect on CL and CD
increases with α which would support the greater reduction in separation at
α = 10◦ and 15◦ when compared with the lower angles. In addition, Vf + = 2.0
615 is capable of delaying the onset of stall to beyond 16◦ . As a consequence CLmax
is increased by roughly 30%, from 0.9 to 1.2.
The L/D curve can be seen in Fig. 20c. There are a few outliers in the CD
curves, most noticeably at α = 4◦ and 7◦ when Vf + = 2.0, but also at α = 4◦
when Vf + = 1.1. These outliers cause discontinuities in the L/D curve. For
620 this reason, a 4th -order polynomial was fitted through the remaining points; this
is also plotted in the figure as solid line. A four-fold increase in L/D at 0◦ and
L/Dmax is achieved when Vf + = 2.0 although this is still some way short of
the high Rec performance. This is expected since the improvements in both CL
and CD shown in Fig. 20, while significant, did not completely recover the high
625 Rec coefficients.

7. Control Efficiency

In the previous sections, it was established that actuation at Vf + = 2.0

can reduce considerably the recirculation zone, which results in drag reduction.
The examined control method therefore is capable of saving energy. There are
630 various ways to evaluate the performance of actuators [28]. In the present paper,
the efficiency is defined as the ratio of the energy saved due to drag reduction
to the electric energy supplied to the actuator. This same definition of energy
efficiency for plasma actuators is employed in [27]. The energy input can be
obtained from the measured current and voltage waveforms. An example of 10
635 cycles of the pre-amplified voltage and current signals is shown in Fig. 21 below.

37
 Figure 21: Signals of instantaneous current (A) and voltage (V) for Vf + = 2.0.

The average power consumption of the actuators is given by [54]

Z T
1
Pavg = V (t)I(t) dt (3)
T 0

and was calculated numerically using the trapezoidal rule to be 1.08 Watts.
Drag reduction by actuation can be easily expressed in terms of power saving
by multiplying the measured drag of the airfoil model by the freestream velocity.
Fig. 22a shows the power saving of the airfoil model at Rec = 50×103 against
640 the angle of attack. As can be seen, the power saving, and consequently the
efficiency, increases with α, which is expected since the effect of separation, and
consequently its control, is greater at higher angles of attack. The maximum
power saving occurs when the actuation suppresses stall.

38
 (a) Power saving (b) Efficiency

Figure 22: Power saving and efficiency against Angle of Attack.

The energy efficiency is defined as

Power Saving
Efficiency = (4)
Power Input

645 and is plotted in Fig. 22b. The energy consumption (power input) is a function of
the Vf and VA only i.e. independent of α. However, energy saving increases and
the system becomes more efficient at large values of α, reaching 40% efficiency
when α = 16◦ . In [27], an energy efficiency of 51% is reported when the electrical
power applied to the plasma actuator is used.
650 It is noted that there is considerably scope for the presented design to be
optimized to improve the efficiency. The efficiency of the present technique could
in principle be improved by varying the skin compliance, for example by using
different materials or skin thickness. In addition the MFC placement has not
currently been optimized. Note also that the employed definition of efficiency
655 takes into account only drag reduction, and does not incorporate the associated
increase in CL and CLmax that can achieved when Vf + = 2.0. The latter
benefits can be more important than drag reduction for certain applications.
Finally, in addition to the current application, the presented design can be
used for optimal aerodynamic shaping of a wing at other flight conditions [43,

39
660 44]. In fact, suitable input signals would allow periodic surface morphing to be
superimposed to aerodynamic shaping thus simultaneously providing two types
of aerodynamic control with one actuator.

8. Conclusion

An airfoil model has been designed and manufactured with an upper skin
665 that can be dynamically actuated by applying a voltage signal to two MFC
patches bonded to its underside. These piezoelectric actuators are very thin,
light and robust, and have low power consumption. The feasibility of applying
these actuators for dynamic surface morphing was first investigated with a series
of surface displacement measurements. It was found that the displacement
670 depends on both frequency and voltage but the presence of flow was not found
to change the skin’s behavior significantly. A variety of voltage amplitudes
and frequencies were tested and the MFC proved to be capable of accepting
frequencies of at least 120 Hz at voltages of at least 400 V. However, at all
values of VA the peak-to-peak displacement of the current setup was found to
675 drop significantly at values of Vf > 30 Hz. Furthermore, the nature of the
surface motion appears to change when Vf > 80 Hz owing to the increased
influence of a second vibration mode. For purposes of this paper the frequency
of actuation was kept below 80 Hz.
The effect of actuation on aerodynamic performance was investigated with
680 corresponding force balance, PIV and hotwire measurements. The flow de-
velopment around the non-actuated airfoil is similar to the separation without
reattachment flow mode described in the literature. Large separated regions and
a wide wake were observed which are responsible for the significant performance
losses when compared with the same airfoil geometry at higher Re. Analysis of
685 the spanwise vorticity and frequency spectra suggested that the initially lami-
nar shear layer experienced a ‘roll-up’ and subsequent vortex shedding before
transition to turbulence occurred towards the trailing edge. Vortex shedding
was then set up in the wake.

40
 When the surface was actuated, the frequency spectra revealed peaks at the
690 actuation frequency. When Vf + = 0.3 (Vf = 10 Hz) the effect was not strong
enough to dominate the flow development and the naturally occurring frequen-
cies remained. This resulted in little change in the separated region and wake
and consequently only a modest effect on the airfoil performance was observed.
In contrast, when Vf + = 2.0 (Vf = 70 Hz) the spectra were dominated by
695 the actuation frequency, with the naturally occurring frequencies absent. The
strong, defined peaks in the spectra are the result of the periodic surface motion
producing large coherent structures. These structures add momentum entrain-
ment and cause significant reduction in separation and velocity deficit in the
wake. This was found to translate into a large improvement in performance,
700 with reductions in CD and increases in CL at all values of α.

Acknowledgments

The first author received an International Joint-PhD scholarship from the

International Office of Imperial College London and generous support from
Temasek Laboratories, National University of Singapore, Singapore.

705 References

[1] T. J. Mueller, J. D. DeLaurier, Aerodynamics of small vehicles, Annual

Review of Fluid Mechanics 35 (2003) 89–111.

[2] A. Wright, D. Wood, The starting and low wind speed behaviour of a small
horizontal axis wind turbine, Journal of Wind Engineering and Industrial
710 Aerodynamics 92 (14-15) (2004) 1265–1279.

[3] S. McTavish, D. Feszty, F. Nitzsche, Evaluating Reynolds number effects

in small-scale wind turbine experiments, Journal of Wind Engineering and
Industrial Aerodynamics 120 (2013) 81 – 90.

[4] M. S. Francis, Unmanned air systems: Challenge and opportunity, Journal

715 of Aircraft 49 (2012) 1652–1665.

41
 [5] M. Anyoji, T. Nonomura, H. A. A. Oyama, K. Fujii, H. Nagai, K. Asai,
Computational and experimental analysis of a high-performance airfoil un-
der low-reynolds-number flow condition, Journal of Aircraft 51 (2014) 1864–
1872.

720 [6] P. B. S. Lissaman, Low-Reynolds-number airfoils, Annual Review of Fluid

Mechanics 15 (1983) 223–239.

[7] T. J. Mueller, S. M. Batil, Experimental studies of separation on a two-

dimensional airfoil at low Reynolds numbers, AIAA Journal 20 (1982) 457–
463.

725 [8] S. Yarusevych, J. G. Kawall, P. E. Sullivan, Separated-shear-layer devel-

opment on an airfoil at low Reynolds numbers, AIAA Journal 46 (2008)
3060–3069.

[9] S. Yarusevych, M. S. H. Boutilier, Vortex shedding of an airfoil at low

Reynolds numbers, AIAA Journal 49 (2011) 2221–2227.

730 [10] F.-B. Hsiao, C.-F. Liu, Z. Tang, Aerodynamic performance and flow struc-
ture studies of a low Reynolds number airfoil, AIAA Journal 27 (2) (1989)
129–137.

[11] S. Yarusevych, P. E. Sullivan, J. G. Kawall, On vortex shedding from an air-

foil in low-Reynolds-number flows, Journal of Fluid Mechanics 632 (2009)
735 245–271.

[12] S. Wang, Y. Zhou, M. M. Alam, H. Yang, Turbulent intensity and Reynolds

number effects on an airfoil at low Reynolds numbers, Physics of Fluids
26 (11).

[13] M. M. Alam, Y. Zhou, H. X. Yang, H. Guo, J. Mi, The ultra-low Reynolds

740 number airfoil wake, Experiments in Fluids 48 (1) (2010) 81–103.

[14] B. D. M. A. Gopalarathnam, B. A. Broughton, M. S. Selig, Design of low

Reynolds number airfoils with trips, Journal of Aircraft 40 (2003) 768–777.

42
 [15] Y. Zhou, Z. J. Wang, Effects of surface roughness on separated and transi-
tional flows over a wing, AIAA Journal 50 (2012) 593–609.

745 [16] J. C. Lin, Review of research on low-profile vortex generators to control

boundary-layer separation, Progress in Aerospace Sciences 38 (4) (2002)
389–420.

[17] M. Kerho, S. Hutcherson, R. Blackwelder, R. Liebeck, Vortex generators

used to control laminar separation bubbles, Journal of Aircraft 30 (3) (1993)
750 315–319.

[18] M. Manolesos, S. G. Voutsinas, Experimental investigation of the flow past

passive vortex generators on an airfoil experiencing three-dimensional sep-
aration, Journal of Wind Engineering and Industrial Aerodynamics 142
(2015) 130–148.

755 [19] A. Abbas, J. de Vicente, E. Valero, Aerodynamic technologies to improve

aircraft performance, Aerospace Science and Technology 28 (1) (2013) 100
– 132.

[20] D. Greenblatt, I. J. Wygnanski, The control of flow separation by periodic

excitation, Progress in Aerospace Sciences 36 (7) (2000) 487–545.

760 [21] A. Seifert, A. Darabi, I. Wygnanski, Delay of airfoil stall by periodic exci-
tation, Journal of Aircraft 33 (4) (1996) 691–698.

[22] R. Kotapati, R. Mittal, O. Marxen, F. Ham, D. You, L. C. III, Nonlin-

ear dynamics and sythetic-jet-based control of a canonical separated flow,
Journal of Fluid Mechanics 654 (2010) 65–97.

765 [23] L. N. Cattafesta, M. Sheplak, Actuators for active flow control, Annual
Review of Fluid Mechanics 43 (2011) 247–272.

[24] W. Zhang, R. Samtaney, A direct numerical simulation investigation of the

synthetic jet frequency effects on separation control of low-re flow past an
airfoil, Physics of Fluids 27(5):055101.

43
770 [25] N. A. Buchmann, C. Atkinson, J. Soria, Influence of znmf jet flow control
on the spatio-temporal flow structure over a naca-0015 airfoil, Experiments
in Fluids 54 (3) (2013) 1–14.

[26] M. Sato, T. Nonomura, K. Okada, K. Asada, H. Aono, A. Yakeno, Y. Abe,

K. Fujii, Mechanisms for laminar separated-flow control using dielectric-
775 barrier-discharge plasma actuator at low reynolds number, Physics of Fluids
27(11):117101.

[27] T. N. Jukes, K.-S. Choi, Long lasting modifications to vortex shedding using
a short plasma excitation, Physical Review Letters 102:254501 (2009) 1–4.

[28] A. Seifert, Evaluation criteria and performance comparison of actuators,

780 in: V. Theofilis, J. Soria (Eds.), Proceedings of the International Confer-
ence on Instability and Control of Massively Separated Flows, Springer
International Publishing, Prato, Italy, 2013, pp. 59–64.

[29] E. P. DeMauro, H. DellOrso, S. Zaremski, C. M. Leong, M. Amitay, Con-

trol of laminar separation bubble on naca 0009 airfoil using electroactive
785 polymers, AIAA Journal 53 (8) (2015) 2270–2279.

[30] S. Dearing, S. Lambert, J. Morrison, Flow control with active dimples,

Aeronautical Journal 11 (1125) (2007) 705–714.

[31] D. Munday, J. Jacob, Active control of separation on a wing with oscillating

camber, Journal of Aircraft 39 (2002) 187–189.

790 [32] J. L. Pinkerton, R. W. Moses, A feasibility study to control airfoil shape

using thunder, Tech. rep., NASA Technical Memorandum 4767 (1997).

[33] O. Bilgen, K. B. Kochersberger, D. J. Inman, O. J. I. Ohanian, Novel,

bidirectional, variable-camber airfoil via macro-fiber composite actuators,
Journal of Aircraft 47 (1) (2010) 303–314.

795 [34] O. Bilgen, L. M. Butt, S. R. Day, C. A. Sossi, J. P. Weaver, A. Wolek,

W. H. Mason, D. J. Inman, A novel unmanned aircraft with solid-state

44
 control surfaces: Analysis and flight demonstration, aiaa paper 2011-2071,
in: 52nd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynam-
ics, and Materials Conference, Denver, Colorado, 2011.

800 [35] O. Bilgen, M. I. Friswell, Implementation of a continuous-inextensible-

surface piezocomposite airfoil, Journal of Aircraft 50 (2) (2013) 508–518.

[36] G. Molinari, A. F. Arrieta, M. Guillaume, P. Ermanni, Aerostructural

performance of distributed compliance morphing wings: Wind tunnel and
flight testing, AIAA Journal 54 (12) (2016) 3859–3871.

805 [37] W. K. Wilkie, R. G. Bryant, J. W. High, R. L. Fox, R. F. Hellbaum,

J. A. Jalink, B. D. Little, P. H.Mirick, Low-cost piezocomposite actuator
for structural control applications, Proc. SPIE 3991 (2000) 323–334.

[38] C. K. Lee, Y. H. Hsu, W. W. Hsiao, J. W. J. Wu, Electrical and mechanical

field interactions of piezoelectric systems: foundation of smart-structure-
810 based piezoelectric sensors and actuators, piezotransformers, and free-fall
sensors, in: Industrial and Commercial Applications of Smart Structures
Technologies, Vol. 5054, Smart Structures and Materials, San Diego, Cali-
fornia, 2003, p. 1.

[39] L. Ma, Y. Shen, J. Li, H. Zheng, T. Zou, Modeling hysteresis for piezo-
815 electric actuators, Journal of Intelligent Material Systems and Structures
(2015) 1–8.

[40] D. A. Hall, Review: Nonlinearity in piezoelectric ceramics, Journal of Ma-

terials Science 36 (2001) 4575–4601.

[41] Eos Systems Inc., Vancouver, Canada, Photomodeler User guide (2014).

820 [42] R. Hartley, A. Zisserman, Multiple View Geometry in Computer Vision,

Cambridge University Press, 2011.

[43] M. Debiasi, Y. Bouremel, H. H. Khoo, S. C. Luo, E. T. Zhiwei, Shape

change of the upper surface of an airfoil by macro fiber composite actuators,

45
 in: 29th AIAA Applied Aerodynamics Conference, Honolulu, Hawaii, 2011,
825 p. 3809.

[44] M. Debiasi, Y. Bouremel, H. H. Khoo, S. C. Luo, Deformation of the upper

surface of an airfoil by macro fiber composite actuators, in: 30th AIAA
Applied Aerodynamics Conference, New Orleans, Louisiana, 2012, p. 2405.

[45] I. Tani, Low-speed flows involving bubble separations, Progress in

830 Aerospace Sciences 5 (1964) 70–103.

[46] H. Horton, A semi-empirical theory for the growth and bursting of lami-
nar separation bubbles, Ph.D. thesis, University of London, Queen Mary
(1969).

[47] S. Burgmann, C. Brücker, W. Schröder, Scanning piv measurements of a

835 laminar separation bubble, Experiments in Fluids 41 (2006) 319–326.

[48] S. Burgmann, W. Schröder, Investigation of the vortex induced unsteadi-

ness of a separation bubble via time-resolved and scanning piv measure-
ments, Experiments in Fluids 45 (2008) 675–691.

[49] S. Yarusevych, P. E. Sullivan, J. G. Kawall, Coherent structures in an

840 airfoil boundary layer and wake at low Reynolds numbers, Physics of Fluids
18(4):044101.

[50] A.Dovgal, V. Kozlov, A. Michalke, Laminar boundary layer separation: In-

stability and associated phenomena, Progress in Aerospace Sciences 30 (1)
(1994) 61 – 94.

845 [51] M. Lang, U. Rist, S. Wagner, Investigations on controlled transition devel-

opment in a laminar separation bubble by means of lda and piv, Experi-
ments in Fluids 36 (2004) 43–52.

[52] S. Yarusevych, P. E. Sullivan, J. G. Kawall, On vortex shedding from an

airfoil in low-reynolds-number flows, Journal of Fluid Mechanics 632 (2009)
850 245–271.

46
[53] E. N. Jacobs, R. Pinkerton, Tests of n.a.c.a. airfoils in the variable density
wind tunnel series 44 and 64, Tech. rep., NASA Technical Note 401 (1931).

[54] C. K. Alexander, M. N. O. Sadiku, Fundamentals of Electric Circuits, 5th

Edition, McGraw Hill, 2012.

47