JOURNAL OF AIRCRAFT

Basic Understanding of Airfoil Characteristics at Low

Reynolds Numbers (104–105)

Justin Winslow,∗ Hikaru Otsuka,† Bharath Govindarajan,‡ and Inderjit Chopra§

University of Maryland, College Park, Maryland 20742
DOI: 10.2514/1.C034415
A computational study has been conducted on various airfoils to simulate flows at Reynolds numbers (Re)
primarily between 104 and 105 to provide understanding and guidance for MAV and other low-Reynolds-number
designs. The computational fluid dynamics tool used in this study is a Reynolds-averaged Navier–Stokes solver with a
Spalart–Allmaras turbulence model and a correlation-based laminar–turbulent boundary-layer transition model.
The airfoils investigated in this study include NACA 0009, NACA 0012 (conventional and reversed configuration),
Clark-Y, flat plate airfoils (1, 3, and 5% thickness), and thin cambered plates (3, 6, and 9% camber). Airfoils were
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

examined for lift and drag performance as well as surface pressure and flow field characteristics. In general, it is
observed that below the Reynolds number of 106 , lift and drag characteristics for most airfoils cannot be assumed to
be constant with the Reynolds number. Below the Reynolds number of 105 , cambered plate airfoils are shown to
have better lift and drag characteristics than thick conventional airfoils with rounded-leading edges. Flat plate
performance is generally invariant to the Reynolds number, but performance improves as thickness is decreased for a
given Reynolds number.

I. Introduction In previous studies, most researchers have generally focused on

105 < Re < 106 as the low-Reynolds-number range. However, this
R ECENT technological advances in compact electronics and
power storage, combined with the growing need for situational
awareness in urban military combat scenarios, have provided the
range excludes further lower Reynolds numbers, that is, less than 105 ,
which is of most interest to MAV, UAV, and low-speed flight
designers. Therefore, in the present work, a chord-based Reynolds
impetus for current micro air vehicle (MAV) interests. Further
number range between 104 and 105 is defined as the low-Reynolds-
motivation behind developing these vehicles is driven largely by
number range. Note that this range is two orders of magnitude lower
potential military and civilian applications, with possible extension than that of large-scale conventional aircraft, which typically operate
into research and commercial sections. These vehicles could be used at a Reynolds number of over 106. Although large-scale airfoil
to investigate dull, dirty, and dangerous environments, such as characteristics, such as high thickness-to-chord ratio (t∕c), rounded
reconnaissance, into hostile territory or scouting missions where the leading edge, and smooth surfaces, have been extensively researched
risk of endangering human life is high: environments that are and optimized for decades, these design characteristics may not be
challenging for conventional aircraft and large-scale unmanned aerial directly applicable to the low-Reynolds-number regime. Figure 1
vehicles (UAVs). Recent interest has also been generated for Mars shows the variation of maximum sectional lift-to-drag ratio with the
exploration UAVs as a fast and mobile alternative to ground-based Reynolds number and it can be seen that the Reynolds number has a
rovers. strong effect on conventional airfoil performance. Although conven-
A common theme across these vehicles is the small geometric tional airfoils perform best at high Reynolds numbers (>106 ), their
scales the rotating wings, that is, rotors, are required to operate, which performance quickly deteriorates at a lower Reynolds number and a
effectively results in very-low-Reynolds-number (Re) flow around geometrically simple flat-plate outperforms the conventional airfoil.
the blades. With respect to the Mars mission, the low atmospheric Therefore, it is evident that simply trying to downsize these airfoils
density and temperature on Mars result in low-Reynolds-number to low-Reynolds-number scales (below 105 ) drastically reduces their
aerodynamics playing a vital role in aerial vehicle design [1]. aerodynamic efficiency in terms of the lift-to-drag ratio [2,3].
Low-Reynolds-number flows are characterized by the increasing Consequently, aerodynamic bodies operating at low Reynolds numbers
importance of viscous forces within the fluid compared with with conventional airfoils incur high power requirements because of
inertial forces. Consequently, boundary-layer physics such as flow excessive aerodynamic drag. Therefore, there is a need to systematically
separation, re-attachment zones, and the amount of laminar/turbulent understand the performance of airfoils at these low Reynolds numbers
flow on the airfoil varies. Therefore, caution must be exercised when and their sensitivities to geometric variations, such as camber and
using geometrically scaled airfoils from full-scale to small-scale thickness.
regimes as they might not provide the best possible aerodynamic
performance.
II. Flow Separation Physics
Before investigating the effect of different airfoil parameters have on
Received 12 February 2017; revision received 30 August 2017; accepted
for publication 13 September 2017; published online 13 December 2017.
their aerodynamic performance, a brief description of the separation
Copyright © 2017 by Winslow, Otsuka, Govindarajan, and Chopra. Published physics is presented with respect to variation in Reynolds numbers.
by the American Institute of Aeronautics and Astronautics, Inc., with Flow field studies, such as particle image velocimetry (PIV), smoke
permission. All requests for copying and permission to reprint should be flow visualization, and hot-wire turbulence measurement, have been
submitted to CCC at www.copyright.com; employ the ISSN 0021-8669 conducted by previous researchers to provide a physical insight into
(print) or 1533-3868 (online) to initiate your request. See also AIAA Rights the effect of the Reynolds number, highlighting the difference between
and Permissions www.aiaa.org/randp. low- and high-Reynolds-number flows. It is known that the sensitivity
*Graduate Research Assistant, Department of Aerospace Engineering, of airfoil performance at different Reynolds number ranges is strongly
Alfred Gessow Rotorcraft Center; jwinslow@umd.edu. related to its boundary-layer characteristics.
†
Visiting Student, Alfred Gessow Rotorcraft Center; h-otsuka@frl.mech.
tohoku.ac.hp. Figure 2 shows a schematic of the streamlines over a representative
‡
Assistant Research Professor, Alfred Gessow Rotorcraft Center; bharath@ airfoil at various Reynolds numbers for a moderate angle of attack. At
umd.edu. higher, more conventional Reynolds numbers (>500;000), represented
§
Distinguised University Professor, Alfred Gessow Professor and Director, in Fig. 2c, laminar flow over the upper airfoil surface is subjected to a
Alfred Gessow Rotorcraft Center; chopra@umd.edu. Fellow AIAA. large adverse pressure gradient close to the leading edge. Because of the
Article in Advance / 1
 2 Article in Advance / WINSLOW ET AL.

in size as the Reynolds number is further decreased for a given angle

of attack, resulting in even lower airfoil efficiencies [6].

III. Previous Studies

The effects of decreasing the Reynolds number on airfoil
performance were initially investigated by researchers, such as
Schmitz [7], as early as 1930. Schmitz conducted experiments on three
airfoils, a thin flat plate, thin cambered plate, and a conventional N60
airfoil (12.4% t∕c, 4% camber), within a Reynolds number range of
2 × 104 to 2 × 105 . This study resulted in two key conclusions: 1) thin
plate airfoils consistently performed better below Re 105, and 2) the
flow is more susceptible to separation below a Reynolds number of
Fig. 1 Effect of Reynolds number on airfoil maximum sectional 105 , especially for the thicker N60. In the 1980s, interest in low-Re
lift-to-drag ratio. (Adapted from McMasters and Henderson [2] and aerodynamics was renewed by researchers, such as Mueller, Selig, and
Mueller [3].) Hoerner, whose studies [8] also agreed with those of Schmitz’s
experiments. Specifically, thin flat and cambered plates are more
efficient than conventional airfoils below the Reynolds number of 105 .
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

inherently lower kinetic energy of the laminar boundary layer, the flow Another observation was that cambered plates show small variations in
separates as a shear layer. This shear layer gains momentum from the the lift coefficient (cl ) and drag coefficient (cd ) with the Reynolds
free-stream and reattaches as a turbulent boundary layer, creating a number, whereas flat plates remain virtually unchanged, which is also
laminar separation bubble (LSB) [4]. Because the turbulent boundary evident from Fig. 1. Hoerner showed that between Reynolds numbers
layer is energized from the free-stream, it is much less prone to of 4 × 104 and 1.2 × 105 , the maximum lift coefficient (clmax ) of the
separation and usually remains attached till the trailing edge. However, cambered plate rose by 4%, whereas that of the N60 airfoil rose by over
at sufficiently high angles of attack, the turbulent boundary layer will 180% [8]. Further low-Reynolds-number experiments by Selig et al.
begin to separate close to the trailing edge, resulting in increased [9] and Mueller and Batill [10] on a range of airfoils also concluded that
pressure drag and decreased lift. As the angle of attack is further clmax consistently increases with higher Reynolds numbers and that the
increased, the turbulent separation point propagates along the airfoil minimum drag coefficient significantly decreases above a Reynolds
surface toward the leading edge and in doing so, the drag rises number of 105 . Selig further observed that for the 60 sailplane-type
dramatically along with an abrupt drop in lift. The airfoil is considered airfoils tested, the drag polar was visually similar and was nearly
stalled. This form of separation is known as trailing edge separation and insensitive to Reynolds number variations once above 105 . But below
is typically associated with thick, rounded leading edge airfoils. this threshold, there was a large degree of nonlinearity in the drag polar
At lower Reynolds numbers (50;000 < Re < 100;000), the at different Reynolds numbers [9]. The initial results from each of these
separation bubble and turbulent boundary-layer thickness both increase studies indicated that low-Reynolds-number airfoil performance was
highly sensitive to the airfoil geometry and the specific Reynolds
in size (a consequence of the higher contribution of the viscous forces),
number at which they operate.
resulting in increased parasitic drag. As shown in Fig. 2b, the separated
Although the experimental measurements of previous studies are
shear layer still gains enough momentum from the free-stream to
useful for understanding trends in low-Reynolds-number aerodynamics,
reattach to the airfoil surface as a turbulent boundary layer. However, in they also have their limitations. Previous studies on Reynolds numbers
this Reynolds number range, the reattachment point is relatively far back below 105 typically contain lift and drag data only for a single Reynolds
on the airfoil. As the angle of attack increases, the reattachment point number [11] for each airfoil, which is not sufficient to fully characterize
moves toward the trailing edge, creating relatively large separation its performance across the spectrum. Furthermore, the data sets that are
bubbles (15–40% chord) [5]. When the airfoil is close to stall, the flow below 105 contain uncertainties and discrepancies. For instance, drag
fluctuates between reattachment and complete separation, resulting in measurements taken on the E387 airfoil (9.1% t∕c, 3.2% camber) at a
hysteresis, which can result in an unsteady aerodynamic phenomenon Reynolds number of 6 × 104 have been shown to vary by 28–68%
with large variations in pitching moment as well. From an experimental between independent measurements, which have been taken at different
standpoint, lift and drag measurements are difficult to quantify, resulting facilities [12], whereas at a higher Reynolds number of 2 × 105 , the drag
in large uncertainty bands. measurements were much more consistent and varied by less than
At much lower Reynolds numbers, that is, 104 < Re < 5 × 104 , 13% [12], which is still a significant margin around the mean. These
shown in Fig. 2a, the laminar separation point is delayed until close to discrepancies are due to factors such as the drag measurement method
the trailing edge of the airfoil, even at very low angles of attack, used (wake-momentum deficit [13] or load measurement), inadequate
because of the increased stability of the boundary layer, which is force measurement sensitivity [14], and wind tunnel turbulence levels
more resistant to flow transition [5]. As the angle of attack increases, [15]. Additionally, below 105 small changes in a Reynolds number have
the separation point moves toward the leading edge, resulting in a a much more prominent effect on airfoil performance than they would at
pronounced shear layer. Unlike at higher Reynolds numbers, the a higher Reynolds number. It has been shown that for the NACA 0012
shear layer is unable to transition and reattach on to the airfoil as a airfoil, the lift curve is highly nonlinear [16], particularly below 5 × 104,
turbulent boundary layer, resulting in high-pressure drag and poor lift resulting in the lift coefficient at the Reynolds number 5 × 104 being
generation [5]. The airfoil is effectively in trailing edge stall for most three times higher than that at 104 for a given angle of attack [16]. As
or all of its operational range. The trailing edge separation increases noted previously, many of the airfoils tested below a Reynolds number

Fig. 2 Illustration highlighting conventional airfoil separation characteristics at different Reynolds number regimes below 106 .
 Article in Advance / WINSLOW ET AL. 3

of 105 were done so for very specific conditions, resulting in sporadic A C-mesh was used as the grid type around the airfoil with an outer
test data, which is insufficient to capture the performance sensitivity of boundary at 15 chord lengths away from the airfoil surface. Mesh sizes
the airfoils to the Reynolds number. typically contained 267 grid points in the surface-wrap direction and
On the one hand, it may be possible to perform a large number of 123 grid points in the wall-normal direction. Grid spacing within the
experiments to assess the performance characteristics of various airfoils boundary layer in the normal direction was 0.001% of the chord length.
at different Reynolds numbers over a wide range of angles of attack and The freestream Mach number was fixed at 0.1 because the majority of
Mach numbers. However, these tests can be prohibitively expensive studies did not report an exact Mach number, and the flows were
and time-consuming, making them practically infeasible. On the other assumed to be within the incompressible range. TURNS2D contains a
hand, computational fluid dynamics (CFD) has been significantly low-Mach preconditioner to aid in the convergence of flows at these
refined over the past three decades and has now been routinely used low Mach numbers. The above parameters were found to be sufficient
by academia and industry to the point of accurately predicting both for solution convergence at the range of low Reynolds numbers
internal and external aerodynamic flows past complex geometries. investigated.
CFD is being extensively used to design and predict the performance of
a wide range of single- and multi-element airfoils. Therefore, the goal
of the present work is to use an in-house computational tool capable of V. Results and Discussion
generating reliable lift, drag, and moment data for an arbitrary two- A. TURNS2D Validation
dimensional (2D) airfoil at different Reynolds numbers to extract To demonstrate the capability of TURNS2D, validation cases were
useful trends from the data to 1) understand the behavior of airfoils at carried out by comparing previous experimental results to current
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

these low Reynolds numbers, and 2) provide insight and guidance for computational results. Figures 3a and 3b show the pressure distribution
low-Reynolds-number designers. on the upper surface of a NACA 0009 airfoil at various angles of attack
at a Reynolds number of 5 × 104 . Experimental results for lift, drag,
and pressure distribution were reported by Lutz et al. [19]. The open
IV. Computational Method circles represent experimental data and the solid lines represent
The CFD solver used in this study to investigate low-Reynolds- numerical predictions. It is evident that at low angles of attack
number aerodynamics is the Transonic Unsteady Rotor Navier–Stokes (α ≤ 5 deg ), the pressure distribution predicted by TURNS2D agrees
2D (TURNS2D) fluid dynamics solver developed in-house [17]. well with experimental data and the location and magnitude of the
TURNS2D has been widely used in the past for flows past airfoil and pressure peak captured quite accurately. At higher angles of attack, as
rotor blades at high Reynolds numbers with confidence. TURNS2D shown in Fig. 3b, the agreement is less satisfactory but still follows the
uses a dual-volume formulation to solve for the Reynolds-averaged trend as the angle of attack increases. It should be noted that at these
Navier–Stokes (RANS) equations. The inviscid fluxes are computed very low Reynolds numbers, significant separation is observed on the
using a third-order MUSCL reconstruction scheme in conjunction with upper surface of the airfoil even at moderately high angles of attack,
Roe’s scheme, whereas the viscous fluxes are evaluated using a and is captured to some extent in the current prediction, but not in its
second-order central difference. For steady problems, an Euler implicit entirety. The incorrect prediction of the plateau in the pressure curve
formulation was used for time marching, with the matrix inversion could be attributed to modification required in the transition model
performed using a lower–upper symmetric line–Gauss Seidel employed within the CFD formulation for low-Re flows. Figure 4a
approach. It has been previously recognized that at low Reynolds shows the Mach contours of the flow along with the streamlines for an
numbers, the flow is laminar, which can then possibly transition to the angle of attack of 6 deg and a Reynolds number of 50,000. The
turbulent regime. Therefore, it is imperative that the solver not assume streamlines on the upper surface show the process of flow separation
a fully laminar or a fully turbulent flow over the airfoil. A key feature in and reattachment at this moderate angle of attack. Furthermore, Fig. 4b
TURNS2D is the inclusion of a laminar–turbulent transition model shows the intermittency contours around the airfoil, which varies from
coupled with the Spalart–Allmaras (SA) turbulence model [18], which 0 (fully laminar) to 1 (fully turbulent). The flow on the upper surface is
improves the numerical solution by better predicting the flow transition laminar, which through a separation bubble, reattaches and transitions
physics. The formulation keeps track of the “intermittency” in the flow, to a turbulent flow. The lower surface remains laminar for much of its
whose value varies between 0 and 1, indicating a fully laminar or fully chord length. However, as noted previously, there are discrepancies in
turbulent flow, respectively. RANS equations were solved using the pressure distribution between the experiments and numerical
TURNS2D to generate sectional coefficients of surface pressure (cp ), predictions.
skin friction (cf ), lift (cl ), drag (cd ), and pitching moment (cm ) under Figures 5a and 5b show the lift and drag comparison, respectively,
given set of flow conditions. for the NACA 0009 airfoil at a Reynolds number of 5 × 104 between

Fig. 3 Nondimensional pressure over the upper surface of a NACA0009 airfoil for low (a) and high (b) angles of attack at an Re of 5 × 104 . (Experimental
data from Lutz et al. [19].)
 4 Article in Advance / WINSLOW ET AL.
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

Fig. 4 Mach number and intermittency contours at 6 deg angle of attack for the NACA 0009 highlighting the presence of flow transition on the upper
surface.

Fig. 5 Lift and drag comparison of TURNS2D with experiments for NACA 0009 at an Re of 5 × 104 . (Experimental data from Lutz et al. [19].)

experiments [19] and CFD predictions for a range of angles of attack. peak lift coefficient was observed in decreasing the Reynolds number
It can be seen that the lift and drag curve predictions from TURNS2D from 106 to 3 × 105 . The drag also rises at a lower angle of attack as the
compare very well with experiments until close to stall. The nonlinear Reynolds number is decreased, indicating early onset of flow separation.
behavior in lift is well-captured by TURNS2D. Near stall conditions, Whereas a linear lift-curve slope was observed at the higher Reynolds
the SA model and the transition model may be limited because of the number of 1 million, a nonlinear behavior was observed at the lower
large separated regions of flow. The SA turbulence model within Reynolds number.
RANS is known to overpredict the eddy viscosity, which tends to lead The reason for this nonlinearity was mentioned by Ohtake et al. [16],
to high rates of dissipation in the flow. To better predict the onset of which was linked to the presence of flow separation on the upper surface
stall and the immediate poststall characteristics, either a large-Eddy of the airfoil at different angles of attack. A comparison was made
simulation or a detached Eddy simulation may be required, which is between various angles of attack between 0 and 14 deg at a Reynolds
beyond the scope of the present study. number of 3 × 105 for the NACA0012 airfoil. The Mach contour and
streamlines close to the surface are shown in Fig. 7. The nonlinear
B. Effect of Reynolds Number on NACA 0012 phenomenon of the lift curve slope can be explained by the existence of
The NACA 0012 is a widely popular airfoil, which has been two separate phenomena that occur for the low-Reynolds-number flows.
experimentally tested for a wide range of Reynolds numbers. Therefore, From the angle of attack of 0 to 8 deg, there is a separation bubble on
these experimental results serve a dual purpose of validating the the upper surface of the airfoil that travels from the trailing edge to the
numerical simulation at various Reynolds numbers, while providing leading edge, effectively changing the point of flow transition from
insight into the effects of the Reynolds number on airfoil performance. laminar to turbulent. This separation bubble travel was not observed in
Figures 6a and 6b show the comparison of lift and drag, respectively, the higher-Reynolds-number case of 1 × 106 . Second, after 8 deg, there
between TURNS2D and experimental measurements taken at Sandia is the presence of a conventional trailing edge separation bubble formed
National Laboratories [20] at two Reynolds numbers of 3 × 105 and as the angle of attack increases till the point of complete separation on the
1.8 × 106 . TURNS2D follows the experimental observation quite well upper surface of the airfoil, that is, stall, at around an angle of attack of
except for some deviation in the magnitude of lift and drag in the stall 12–14 deg. These two phenomena could explain the two slopes
region, which is expected from RANS-based models. A 23% drop in observed in the lift curve diagram at lower Reynolds numbers.
 Article in Advance / WINSLOW ET AL. 5

Fig. 6 High-Reynolds-number lift and drag comparison of CFD predictions for NACA0012 at Reynolds numbers of 3 × 105 and 1 × 106 . (Experimental
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

data from Sheldahl and Klimas [20].)

Fig. 7 Process of separation bubble travel at a lower Reynolds number of 3 × 105 . Red arrow highlights the separation bubble.

Fig. 8 Lift, drag, and pitching moment comparison of CFD predictions for NACA0012 between Reynolds numbers of 104 and 105 . (Experimental data
from Ohtake et al. [16].)

Figure 8 shows the comparison of lift, drag, and pitching moment for Reynolds number (see Fig. 6), the lift curves are highly nonlinear,
the NACA 0012 airfoil at various low Reynolds numbers between 104 particularly for α < 5 deg , and consequently, a single lift-curve-slope
and 105 . The results show an extreme sensitivity of airfoil performance value cannot be assumed in this range. It should also be noted that the
to changing Reynolds numbers below 105. As the Reynolds number drag bucket is significantly smaller at a low Reynolds number and the
decreases, there is an increase in drag (again, possibly because of minimum drag increases compared with the minimum drag in Fig. 6b.
premature separation of the laminar boundary layer). This separation can Although TURNS2D has difficulty in predicting some of the exact
be inferred from a large decrease in lift and the shrinking of the drag values as noted by experiments, the general trend is well captured. The
bucket, as shown in Figs. 8a and 8b, respectively. The peak lift moment curves are shown to vary significantly as the Reynolds number
coefficient decreases by approximately 46% with an order drop in changes and a clear trend is not identifiable, but these are still predicted
magnitude of the Reynolds number from 105 to 104 . Unlike the higher moderately well.
 6 Article in Advance / WINSLOW ET AL.

Fig. 9 Effect of flat plate thickness on lift and drag at a Reynolds number of 104 . (Experimental data from Okamoto et al. [21].)
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

C. Effect of Thickness on Flat Plates

Specific airfoil characteristics, such as camber and thickness, heavily
affect airfoil lift and drag qualities. Figure 9 shows the variation in lift
and drag against angle of attack for conventional flat plates of varying
thickness (t∕c of 1, 3, and 5%) at a Reynolds number of 104 . Results are
shown from both TURNS2D and experimental measurements. It is
evident that as flat plate thickness-to-chord (t∕c) decreases, lift and
drag characteristics improved, with the 1% flat plate being the most
efficient. Thicker flat plates tend to increase the pressure drag and,
therefore, results in poorer aerodynamic performance compared with
thin plates. For each flat plate, the lift-curve slope is generally constant
up to approximately an angle of attack of 7 deg, after which the slope
abruptly changes to a much shallower value, as shown in Fig. 9a.
However, the high lift-curve slope is maintained to higher angles of
attack as the t∕c ratio is decreased. Furthermore, flat plate drag
decreases as t∕c decreases, as shown in Fig. 9b. The drag coefficient
appears to decrease with lower t∕c between 0 deg < α < 10 deg and
then converges to similar values at higher angles of attack (≈20 deg ), a
trend that is not particularly well captured by TURNS2D. Therefore,
minimum t∕c was observed to be a key characteristic for increased
aerodynamic efficiency at low Reynolds numbers. Fig. 10 Mach contours and surface pressure over flat plates of different
thickness at 5 deg angle of attack for a Reynolds number of 104 .
An explanation for the decreased performance as the flat plate t∕c
increases is the increased separation of the flow around the airfoil. The
thicker the flat plate, the more exacerbated the sharp corner is at the
leading edge. This increased “bluntness” of the airfoil leading edge maximum lift coefficient was observed as the camber increased from
increases the strength of the adverse pressure gradient, causing the 3 to 9%. However, there was also an increase in drag as the camber
laminar flow to separate very close to the leading edge, resulting in was increased—most pronounced for the 9% camber.
higher form drag. Furthermore, the increased separation observed for Although the general trend in lift and drag is captured by
thicker flat plates reduces the pressure differential between the upper TURNS2D, the magnitude of lift and drag is not accurately predicted.
and lower surfaces, thus reducing its lifting force. Figure 10 shows the The discrepancy was linked to the transition model used in TURNS2D
surface pressure distribution and Mach contours (with streamlines) on and perhaps its limitation in assessing the transition characteristics
flat plates of various thickness-to-chord ratios (1, 3, and 5%) at an angle of low-Reynolds-number flows. The transition model employed
of attack of 5 deg and a Reynolds number of 104 . The large suction within TURNS2D was calibrated such that the source terms of the
peak for the lower-thickness plates is the primary source of higher lift. intermittency transport equation were tuned to match the transition
As in all three cases, the flow is almost entirely separated from the onset locations from T3-series zero-pressure gradient flat experiments
upper surface. Furthermore, a plot of nondimensional pressure (Cp ) [18]. It is not entirely clear how this transition model will work for
versus z∕c was examined to verify that the pressure-drag is indeed low-Reynolds-number flows and highly cambered flat plates, which
encounter signification adverse pressure gradients. The other variable
higher for the thicker flat plates. However, as the plots are rectangles
is the free-stream turbulence intensity (FSTI), which was set to 0.03%
(z∕c is constant across the upper surface and lower surface,
in the present study. The variation of lift and drag to FSTI is not entirely
respectively), they are not presented in the paper.
clear and will be studied in the future.
Figure 11c shows the lift-to-drag ratio of each cambered plate as
D. Effect of Camber on Thin Plates predicted by TURNS2D, a combination of the data presented in
To understand the performance trends at low Reynolds numbers Figs. 11a and 11b. It is evident that despite the higher lift coefficient
arising from camber, thin plates with a varying camber were of the 9% camber when compared with the other camber ratios, a
investigated. Figure 11 shows the lift and drag at various angles of moderate camber of 6% yields the highest lift-to-drag of 15. While
attack for 1% thick flat plates with cambers of 3, 6, and 9% at a the 3% camber airfoil also has a similar value of peak lift-to-drag, the
Reynolds number of 104 . Results are shown from TURNS2D and 6% camber flat plate was able to maintain a higher lift-to-drag ratio
compared against experimental measurements of Okamoto et al. over a wider range of angles of attack, which would be preferred from
[21]. The results show that as camber is increased, so does the lift a design standpoint, that is, robustness in design. Clearly, camber has
curve and peak lift by a significant margin. An increase of 47% in the a strong effect on low-Reynolds-number aerodynamic performance.
 Article in Advance / WINSLOW ET AL. 7
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

Fig. 11 Effect of thin plate (t∕c1%) camber on lift, drag, and lift-to-drag at a Reynolds number of 104 . (Experimental data from Okamoto et al. [21].)

The effect of increasing the airfoil camber causes a greater is also accompanied with a drag penalty that is most noticeable via the
differential change in momentum of the flow around the airfoil, which increase in the base drag coefficient, as shown in Fig. 12b, which is
causes differences in the pressure difference, thus increasing lift. more pronounced for the lower Reynolds number. Consequently, the
However, the trade-off is that an increase in the camber of the flat plate lift-to-drag curves of the cambered airfoils are similar to each other at
also increases drag (form drag). When a high camber is introduced, a the Reynolds number of 2 × 104 , but are slightly higher compared with
large separation region is created on the aft end the airfoil. This the symmetric airfoil’s lift-to-drag, as shown in Fig. 12c.
separation region effectively increases the size of the wake behind the However, at a higher Reynolds number of 1 × 105 , increasing
thin plate, which increases the momentum deficit and consequently camber has a more pronounced effect on performance. It is seen in
the drag. At a low Reynolds number (104 ), the camber value at which Fig. 12a that increasing the Reynolds number increases the lift
the drag increase begins to outweigh the lift gains appears to be at 6%. coefficient by a larger amount for higher-camber airfoils. Additionally,
Based on Fig. 11, a low-Reynolds-number wing design for high-lift the base drag decreases as the Reynolds number increases for higher-
application can use a 9% camber, whereas a 6% camber should be used cambered airfoils, particularly within the angle-of-attack range of
if efficiency is the desired design factor. 4–12 deg. Therefore, cumulative performance of the airfoil (reflected in
Prior results have compared numerical predictions from TURNS2D the lift-to-drag curves) is more differentiated than that at the lower
against various experimental observation of lift, drag, and some Reynolds number and can achieve a maximum lift-to-drag ratio of
pressure distribution for different airfoils at multiple low Reynolds around 40 compared with a maximum of 15 at a lower Reynolds
numbers. TURNS2D was observed to capture the magnitudes and number of 104 . The performance of the symmetric airfoil changes
trends in the aerodynamic data fairly well, with certain mispredictions marginally because of a change in the Reynolds number. The above
close to stall, which in turn could serve as motivation for further result indicates the degree to which an airfoil performance characteristic
developing RANS-based numerical tools and transition models at is dependent on the operating Reynolds number.
these low Reynolds numbers. The rest of the Results section is devoted
to further understanding the behavior of airfoils based on aerodynamic F. Flat Plate and NACA 0012
prediction from TURNS2D. It was seen in Fig. 1 that simple flat plates outperform conventional
airfoils for Reynolds numbers lower than 5 × 104 . To obtain a deeper
E. Effect of Reynolds Number on Thin Cambered Airfoils understanding of this rationale as to why certain airfoil characteristics
The effect of camber on thin (3% t∕c) NACA airfoils was examined are better suited to low-Reynolds-number flows than others, the flow
with the use of computational results from TURNS2D. Figure 12 shows field solution from TURNS2D was examined. Figure 13 shows the
the performance results for the symmetric NACA0003 airfoil, 2% comparisons in Mach contours between the NACA 0012 airfoil and a
cambered NACA2403, 4% cambered NACA4403, and 6% cambered 2% t∕c flat plate at a Reynolds number of 2 × 104 at three different
NACA6403 at two Reynolds numbers of 2 × 104 (indicated by the angles of attack. The key distinguishing factor between the thick
dashed lines) and 1 × 105 (indicated by the solid lines). At both these NACA 0012 and the thinner flat plate is the difference in boundary-
Reynolds numbers, increasing the camber increases the lift coefficient layer separation, inferred from the streamlines shown in Fig. 13. For
over the given range of angles of attack. However, the increase in camber the NACA 0012, the upper surface boundary layer stays attached for a
 8 Article in Advance / WINSLOW ET AL.
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

Fig. 12 Effect of camber on lift, drag, and lift-to-drag ratio for NACA airfoils of 3% t∕c ratio at Reynolds numbers of 2 × 104 and 1 × 105 .

Fig. 13 Difference in boundary-layer characteristics between NACA 0012 and 2% t∕c flat plate at a Reynolds number of 2 × 104 . Contours represent
Mach number.

majority of the airfoil until it separates close to the trailing edge. This This low-Reynolds-number separation effect is particularly influential
separation characteristic is similar to flow-field results from previous on the lift curve for angles of attack of 0 and 2 deg for a Reynolds
studies [5,6] on thick, conventional airfoils at Reynolds numbers less number of 2 × 104 . In this range of angle of attack, the separation
than 5 × 104 , that is, a classic representation of trailing edge stall. For region can be large enough such that the upper-surface pressure is
the range of angles of attack shown, the flow separates close to the higher than the lower-surface pressure for most of the chord. This
trailing edge to gain momentum from the free-stream and reattach as a effect is evident in Fig. 14b at α
2 deg where the upper- and lower-
turbulent boundary layer. The resulting separation region causes high surface pressure curves for the NACA 0012 cross at 30% chord. At
form drag and prevents the formation of a beneficial drag bucket at Reynolds numbers lower than 5 × 104 , this pressure difference causes
low angles of attack. the lift curve slope close to α
0 deg to be shallower than at other
The large trailing edge separation on the NACA 0012 results in low angles, and can cause slightly negative lift values as seen in Fig. 8a for
lift because of the decreased suction pressure on the upper surface. a Reynolds number of 4 × 104 .
 Article in Advance / WINSLOW ET AL. 9

Fig. 14 Pressure distribution on the upper and lower surfaces of a NACA 0012 and 2% t∕c flat plate at a Reynolds number of 2 × 104 .
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

Fig. 15 Characteristics of a NACA 0012 airfoil in conventional and reversed configurations at a Reynolds number of 2 × 104 at 7 deg angle of attack.
Contours represent nondimensional density. (Experimental data from Laitone [14] and Ohtake et al. [16].)

The pressure distribution over the upper surface of the flat plate in Figure 15 shows the characteristics of a NACA 0012 in its
Fig. 14 also indicates the presence of the LSB. At an angle of attack of conventional configuration (rounded leading edge) compared with its
5 deg, there is a flattening of the pressure distribution on the upper reversed configuration. Figure 15a compares the lift variation with
surface of the thin plate, which is typical of a separation bubble. The angle of attack between experimental data [14,16] and TURNS2D
reattachment of the upper boundary layer preserves the pressure predictions at a Reynolds number of 2 × 104 . The reversed configured
difference and, therefore, enables the flat plate to provide more lift results in a much higher maximum lift compared with the baseline
than the NACA 0012 at a Reynolds number lower than 5 × 104 . configuration. In particular, there is a very strong effect on peak lift
The 2% t∕c flat plate airfoil exhibits significantly different flow between angles of attack of 7 and 10 deg. The computational flow
characteristics than the NACA 0012 in the same flow regime. As seen fields around the NACA 0012 at α
7 deg provides insight into the
in Fig. 13, the flat plate does not exhibit trailing edge separation. difference in lift values. Figure 15b shows that the NACA 0012 in its
Instead, the laminar flow is tripped early along the surface by the sharp conventional configuration creates a large separation region over 80%
leading edge. The separated laminar shear layer gains momentum of the upper surface. However, when the NACA 0012 is reversed under
from the free-stream and transitions to a turbulent flow. Because this the same flow condition, the airfoil behaves more like a flat plate; see
transition occurs near the leading edge, there is sufficient chord length Fig. 13. Most notably, the now-sharp leading edge creates a similar
remaining for the turbulent boundary layer to reattach, forming a separation bubble as seen on the 2% t∕c flat plate, and this separation of
leading edge separation bubble. This bubble is most visible at α
the boundary layer on the reversed NACA 0012 allows it to reattach as
5 deg in Fig. 13 for the flat plate. a turbulent boundary layer further downstream. The flow reattachment
maintains lower pressure on the upper surface and subsequently results
G. Conventional and Reversed NACA 0012 in increased lift compared with the conventional configuration.
The effect of reversing airfoils at a low Reynolds number may Despite the fact that the NACA 0012 has the same maximum thickness
actually be beneficial depending on the airfoil. The rationale behind in either configuration, these results indicate that the distribution
the reversing the airfoil is that the sharp trailing edge is now the of the thickness along the chord, particularly at the leading edge, has an
aerodynamic leading edge, which will promote early separation of effect on airfoil performance at a low Reynolds number that is
the laminar flow and can possibly result in the reattachment of the counterintuitive to expectations at higher Reynolds numbers.
boundary layer through a separation bubble. In addition to the
maximum thickness of the airfoil, its leading edge geometry, that is, H. Overall Reynolds Number Effects
sharp or rounded, has a strong effect on its performance. By reversing As a concluding study, TURNS2D has been used to determine
a symmetric airfoil, the leading edge geometry is essentially changed, trends in aerodynamic performance of different airfoils across a range
but the maximum thickness of the airfoil remains the same. of Reynolds numbers. Figure 16 shows the effect of the Reynolds
 10 Article in Advance / WINSLOW ET AL.
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

Fig. 16 Effect of Reynolds number on multiple airfoils in terms of the drag polar.

number (ranging for 2 × 104 to 1 × 106 ) on multiple types of airfoils: NACA 0012 and Clark-Y are much more sensitive to the Reynolds
a 2% thick flat plate, 6% cambered plate (2% t∕c), NACA 0012, and number than plate-type airfoils. As seen in Fig. 16, the peak lift
Clark-Y. This comparison was conducted to obtain an understanding coefficient of the 6% cambered plate increases from approximately
of Reynolds number effects on vastly different airfoils. The major 1.15 at a Reynolds number of 2 × 104 to approximately 1.32 at a
distinguishing factor between the airfoils is that the more conventional Reynolds number of 106 , and the peak lift coefficient of the flat plate is

Fig. 17 Effect of Reynolds number on multiple airfoils in terms of lift-to-drag against angle of attack.
 Article in Advance / WINSLOW ET AL. 11

virtually unchanged. Over the same Reynolds number range, the peak form drag. An increase in camber (from 3 to 9%) on thin plate airfoils
lift coefficient more than doubles for the NACA 0012 and Clark-Y, (t∕c of 1%) resulted in an increase in both the lift and
increasing from 0.55 to 1.3 for the NACA 0012 and from 0.72 to 1.63 the lift-to-drag ratio compared with flat plate airfoils of the same
for the Clark-Y. thickness. For a low-Reynolds-number wing design to achieve high
The largest change in performance between the four airfoils occurs lift-to-drag, a 6% camber appeared optimum. However, a 9% camber
higher than a Reynolds number of 105 . At Reynolds numbers could be used if a maximum lift is a stronger design factor.
of 2 × 104 and 4 × 104 , the thin flat and cambered plates clearly 3) The effect of camber on thin NACA airfoils (NACA 0003, 2403,
outperform the conventional airfoils in terms of lift and drag. But at a 4403, 6403) was examined at Reynolds numbers of 2 × 104 and 105 ,
Reynolds number of 1.2 × 105 , the lift generated by the Clark-Y is and high-camber airfoils (NACA 4403 and 6403) were shown to have
comparable to that of the 6% cambered plate, and the lift of the NACA the highest lift-to-drag at both Reynolds numbers. Additionally, the
0012 is greater than that of the flat plate for a given drag coefficient. lift-to-drag curves were similar at the lower Reynolds number and
These performance margins increase further at a higher Reynolds more differentiated at the higher Reynolds number. The symmetric
number of 106 . It should also be noted that the Reynolds number NACA 0003 airfoil performance changes very little with an increase
appears to have an uniform effect on the drag coefficient of each of the in the Reynolds number and behaves similar to a flat plate. As camber
four airfoils. Figure 16 shows that as the Reynolds number increases, is increased, however, the Reynolds number has a profound effect on
the drag polars gradually shift to the left, indicating a decrease in drag. the lift-to-drag, which improves by a greater margin.
The minimum drag appears to decrease by approximately the same 4) Examination of the boundary layer at a low Reynolds number
amount for all four airfoils. Because the geometric characteristics of the shows significant differences between thin plate airfoils and thicker
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

four airfoils are vastly different, the similar decrease in their minimum airfoils. The NACA 0012 is more susceptible to trailing edge
drag is likely attributed to decreasing viscous drag effects as the separation at a low Reynolds number, whereas the flat plate trips the
Reynolds number increases. laminar boundary layer at the leading edge, allowing it to reattach
Figure 17 shows the effect of the Reynolds number on each airfoil further downstream. The reattachment of the upper boundary layer
(2% t∕c flat plate, 6% cambered plate, NACA 0012, and Clark-Y) preserves lower surface pressure and, therefore, allows the flat plate
separately in terms of lift-to-drag (cl ∕cd ) as a function of the angle of to provide more lift than the NNACA 0012 at Reynolds numbers
attack. Similarly to the results observed in Fig. 16, the conventional lower than 5 × 104 .
NACA 0012 and Clark-Y performance is shown to be more sensitive 5) When the NACA 0012 is placed in a reversed configuration at a
to the Reynolds number compared with the plate-type airfoils. Reynolds number of 2 × 104 , its lift qualities improve. This effect is
Surprisingly, the 6% cambered plate appears to achieve the same peak because of the sharp leading edge, creating an early LSB similar to the
lift-to-drag as the Clark-Y at a Reynolds number of 106 , as seen in flat plate. Despite the fact that the NACA 0012 has the same
Fig. 17b. However, the Clark-Y maintains a larger envelope of a higher maximum thickness in either configuration, the result indicates that
lift-to-drag ratio for a broader range of angles of attack. The primary the distribution of the thickness along the chord, particularly at the
implications for the design drawn from these results are as follows: leading edge, has an effect on airfoil performance at a low Reynolds
1) below the Reynolds number of 105 , cambered plate airfoils number that is counter to what is expected at a high Reynolds number.
demonstrate better lift-to-drag characteristics than conventional
airfoils; 2) above the Reynolds number of 105 , thicker airfoils have
Acknowledgments
better lift-to-drag characteristics than thin flat or cambered plates;
3) below the Reynolds number of 106 , lift-to-drag characteristics for This work is supported by the U.S. Army/U.S. Navy/NASAVertical
most airfoils cannot be assumed to be constant with the Reynolds Lift Research Center of Excellence Grant Number W911W6-11-2-
number; 4) zero-cambered flat plate performance is generally invariant 0012 with Mahendra Bhagwat as technical monitor. The authors would
to the Reynolds number. like to thank Alex Moodie, AMRDEC, for useful discussions.

References
VI. Conclusions [1] Shrestha, R., Benedict, M., Hrishikeshavan, V., and Chopra, I., “Hover
The goal of the present work was to use TURNS2D (a structured Performance of a Small-Scale Helicopter Rotor for Flying on Mars,”
RANS solver with laminar-turbulent transition model) as a means to Journal of Aircraft, Vol. 53, No. 4, 2016, pp. 1160–1167.
understand the low-Reynolds-number aerodynamics and, therefore, to doi:10.2514/1.C033621
provide guidelines for MAV and low-Reynolds-number designs. The [2] McMasters, J. H., and Henderson, M. L., “Low Speed Single-Element
results from TURNS2D were compared with available experimental Airfoil Synthesis,” Technical Soaring, Vol. 6, No. 2, 1980, pp. 1–21.
[3] Mueller, T. J., “Aerodynamic Measurements at Low Reynolds Numbers
lift, drag, pitching moment, and surface pressure data as means of for Fixed Wing Micro-Air Vehicles,” RTO AVT/VKI Special Course on
validation. Overall, the calculated values from TURNS2D agree quite Development and Operation of UAVs for Military and Civil
well with the experimental measurements at low Reynolds numbers. It Applications, Von-Karman Inst., Brussels, Sept. 1999, http://www.dtic
was also observed from previous works that the uncertainty and error .mil/docs/citations/ADP010760.
bounds in experimental results were higher for low-Reynolds-number [4] Abbott, I. H., and von Doenhoff, A., Theory of Wing Sections,
flows, which could also contribute to certain misalignments between McGraw–Hill, New York, 1959, Chap. 5.
experimental observations and numerical predictions. The current [5] Mueller, T. J., and DeLaurier, J. D., “Aerodynamics of Small Vehicles,”
work also identified the need for further assessment of turbulence Annual Review of Fluid Mechanics, Vol. 35, No. 1, 2003, pp. 89–111.
doi:10.1146/annurev.fluid.35.101101.161102
models for low-Reynolds-number flows. The following specific
[6] Tsuchiya, T., Numata, D., Suwa, T., and Asai, K., “Influence of
conclusions were drawn from this study: Turbulence Intensity on Aerodynamic Characteristics of an NACA 0012
1) The results from TURNS2D show the extreme sensitivity of at Low Reynolds Numbers,” 51st AIAA Aerospace Sciences Meeting,
airfoil performance to changing the Reynolds number below 105, AIAA Paper 2013-65, Jan. 2013.
particularly for conventional airfoils such as the NACA 0012. As the doi:10.2514/6.2013-65
Reynolds number decreases there is an increase in drag, particularly [7] Schmitz, F. W., “Aerodynamics of the Model Airfoil Part I,” Airfoil
because of premature flow separation and failure to reattach, resulting Measurements, Translated from German, Redstone Scientific
in a reduced drag bucket and a large decrease in lift with the Information Center RSIC-721, Nov. 1967.
maximum lift coefficient decreases by approximately 46% between [8] Hoerner, S. F., Fluid-Dynamic Lift, edited by L. A. Hoerner, 2nd ed.,
the Reynolds number of 105 and 104 . Below a Reynolds number of 1985, pp. 42–44.
[9] Selig, M. S., Donovan, J. F., and Fraser, D. B., Airfoils at Low Speeds, H. A.
105 , the lift curves are highly nonlinear, particularly for α < 5°, and a Stokely, Virginia Beach, VA, 1989, Chap. 12.
single lift-curve-slope value may not be appropriate. [10] Mueller, T. J., and Batill, S. M., “Experimental Studies of Separation on
2) As flat plate thickness-to-chord (t∕c) decreases (from 5 to 1%), a Two-Dimensional Airfoil at Low Reynolds Numbers,” AIAA Journal,
lift and drag characteristics improved, with the 1% flat plate being the Vol. 20, No. 4, 1982, pp. 457–463.
most efficient at a Reynolds number of 104 , a consequence of lower doi:10.2514/3.51095
 12 Article in Advance / WINSLOW ET AL.

[11] Williamson, G. A., McGranahan, B. D., Broughton, B. A., Deters, [17] Srinivasan, G. R., and Baeder, J. D., “TURNS: A Free-Wake Euler/
R. W., Brandt, J. B., and Selig, M. S., Summary of Low-Speed Airfoil Navier-Stokes Numerical Method for Helicopter Rotors,” AIAA Journal,
Data, Vol. 5, Univ. of Illinois at Urbana-Champaign, Champaign, IL, Vol. 31, No. 5, 1992, pp. 959–962.
2012, Chap. 4. doi:10.2514/3.49036
[12] Lyon, C. A., Broeren, A. P., Giguere, P., Gopalarathnam, A., and Selig, [18] Medida, S., and Baeder, J., “Application of the Correlation-Based γ-Re θt
M. S., Summary of Low-Speed Airfoil Data, Vol. 3, Univ. of Illinois at Transition Model to the Spalart-Allmaras Turbulence Model,” 20th AIAA
Urbana-Champaign, Champaign, IL, 1997, Chap. 5. Computational Fluid Dynamics Conference, AIAA Paper 2011-3979,
[13] Selig, M. S., Lyon, C. A., Giguere, P., Ninham, C. P., and Guglielmo, J. J., June 2011.
Summary of Low-Speed Airfoil Data, Vol. 2, SoarTech Publ., Virginia doi:10.2514/6.2011-3979
Beach, VA, 1996, Chap. 5. [19] Lutz, T., Wurz, W., and Wagner, S., “Numerical Optimization and Wind-
[14] Laitone, E. V., “Wind Tunnel Tests of Wings at Reynolds Numbers Below Tunnel Testing of Low Reynolds-Number Airfoils,” Fixed and Flapping
70000,” Experiments in Fluids, Vol. 23, No. 5, 1997, pp. 405–409. Wing Aerodynamics for Micro Air Vehicles Applications, edited by
doi:10.1007/s003480050128 T. J. Mueller, Vol. 195, AIAA, Reston, VA, 2001, pp. 169–190.
[15] McAurthur, J., “Aerodynamics of Wing at Low Reynolds Numbers: doi:10.2514/5.9781600866654.0169.0190
Boundary Layer Separation and Reattachment,” Doctor of Philosophy [20] Sheldahl, R. E., and Klimas, P. C., “Aerodynamic Characteristics of Seven
Dissertation, Dept. of Aerospace and Mechanical Engineering, Univ. of Symmetrical Airfoil Sections Through 180-Degree Angle of Attack for
Southern California, Los Angeles, CA, 2008. Use in Aerodynamic Analysis of Vertical Axis Wind Turbines,”
[16] Ohtake, T., Nakae, Y., and Motohashi, T., “Nonlinearity of the Sandia National Laboratories SAND-80-2114, Albuquerque, NM,
Aerodynamic Characteristics of NACA 0012 Aerofoil at Low Reynolds March 1981, https://www.osti.gov/scitech/biblio/6548367.
Numbers,” Japanese Society for Aeronautical and Space Science [21] Okamoto, M., Yasuda, K., and Azuma, A., “Aerodynamic Characteristics
Downloaded by UNIVERSITY OF FLORIDA on December 23, 2017 | http://arc.aiaa.org | DOI: 10.2514/1.C034415

Papers, Vol. 55, No. 644, 2007, pp. 439–445. of the Wings and Body of a Dragonfly,” The Journal of Experimental
doi:10.2322/jjsass.55.439 Biology, Vol. 199, No. 2, 1996, pp. 281–294.