Content uploaded by Tony L. Schmitz
Author content
All content in this area was uploaded by Tony L. Schmitz on Feb 19, 2016
Content may be subject to copyright.
Improving the Fabrication Process of Micro-Air-Vehicle
Flapping Wings
Kelvin Chang,
∗
Anirban Chaudhuri,
†
Jason Rue,
‡
Raphael Haftka,
§
and Peter Ifju
¶
University of Florida, Gainesville, Florida 32611
and
Christopher Tyler,
**
Vasishta Ganguly,
††
and Tony Schmitz
‡‡
University of North Carolina at Charlotte, Charlotte, North Carolina 28223
DOI: 10.2514/1.J053884
The aerodynamic performance of flapping micro air vehicles in hover conditions is dependent on many parameters,
including the wing design. With the goal of optimizing the wing for hover performance, the initial focus was to reduce
the uncertainty in the thrust measurements. This is because lower uncertainty in this metric enables better resolution
in comparing the performance of different designs. Aerodynamic performance variability was deemed to be the fault
of an imprecise manufacturing technique. Therefore, adjustments were made to the fabrication process until a
permissible level of uncertainty was attained for optimization; the goal was less than 5%. This paper chronicles the
progression of the wing fabrication process and details how the uncertainty was evaluated. Four fabrication methods
and two different wing designs are included in this study: a carbon fiber hand layup technique, carbon fiber cured in a
machined mold, and two variations of a machined plastic skeleton reinforced with a carbon fiber rod. The uncertainty
in thrust production, expressed in coefficient of variation, improved from 16.8% for the hand layup method to 2.6%
for the computer numerically controlled plastic skeleton adhered to the nylon membrane with transfer tape.
Additionally, the coefficient of variation for wing weight also reduced (from 11.4 to 2.0%).
Nomenclature
C
4
= bias correction factor
CV
mfg
= manufacturing coefficient of variation
CV
tot
= total coefficient of variation
CV
test
= testing coefficient of variation
n = samples
I. Introduction
Flapping propulsion for micro air vehicles (MAVs) are an
alternative to a fixed wing or rotor-driven configuration. In an
environment where there is continued demand for smaller unmanned
air vehicles (UAVs), small flapping devices can fly like hummingbirds
[1] or even small insects [2,3]. MAVs broadly pertain to flying UAVs
that possess no straight dimension longer than 6 in. They have gained
popularity for their wide range of applications, including inv estigation
of hazardous environments and spy operations, where size and noise
signature are paramount, although designing a UAV in the size range
of a hummingbird comes with many challenges. Realization of a
mass-produced flapping UAV would likely entail a combination of
engineering and affordable materials, where the drive mechanism is
simple, light, and durable [4], and the wings are optimized for
aerodynamic performance. How do we get there? For the wings, it has
been observed that passive deformation from compliance is essential to
many of the mechanisms that produce the necessary aerodynamic
forces for flight [5]. Also, chordwise flexibility assists in thrust
production and efficiency [6]. Thus, the best level of reinforcement for
thrust is a compromise between the extremes. Ultimately, a high-
fidelity model that incorporates this deformation would be required to
optimize the wing skeleton reinforcement design.
A review of different computational models illustrates the breadth
of complexity involved in modeling a hovering wing [7], where the
advance ratio is zero. Progress has been made in developing validated
models [8–11]. However, the complexity of the aeroelastic in-
teraction and highly nonlinear aerodynamics entails exorbitant
computational costs or the use of low-fidelity models [12]. This
hinders their capability to efficiently predict the aerodynamic
performance of a large number of wings, which is important for
optimization. The plunge and pitching motions have been optimized
for thrust in a slow forward flight flapping airfoil, using a gradient-
based optimizer and a two-dimensional model [13]. However,
optimization of a wing in hover mode continues to be a challenge. An
alternative to using a computation model is to drive the optimization
with experiments, using physically measured thrust forces in the
place of a high-fidelity model, even possibly looking at tradeoffs
between thrust and power measurements [14]. An experimental
optimization of the reinforcement structure for hovering flight has
been accomplished for thrust alone [15] and with thrust and power as
objectives [16]. Techniques such as using different surrogates to have
multiple candidate designs every optimization cycle [17] and using
an adaptive sampling technique (efficient global optimization) [18]
were helpful for minimizing the cost of the optimization, but the
main costs pertain to the time spent fabricating and testing wings
in a controlled manner. The first major obstacles to accomplishing
the optimization was both identifying the uncertainty in the thrust
measurement [15] and reducing it to a manageable level by improv-
ing the manufacturing techniques used to make the wings. A
complete analysis of the uncertainty for different manufacturing
methods is provided in this paper. The progression from an incon-
sistent hand-layup approach to a more precise machine controlled
approach is chronicled. It is shown how these improvements reduce
Received 30 August 2014; revision received 16 March 2015; accepted for
publication 3 May 2015; published online 23 July 2015. Copyright © 2015 by
the American Institute of Aeronautics and Astronautics, Inc. All rights
reserved. Copies of this paper may be made for personal or internal use, on
condition that the copier pay the $10.00 per-copy fee to the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include
the code 1533-385X/15 and $10.00 in correspondence with the CCC.
*Graduate Research Assistant, Department of Mechanical and Aerospace
Engineering; kc3635@gmail.com. Student Member AIAA.
†
Graduate Research Assistant, Department of Mechanical and Aerospace
Engineering; anirban.chaudhuri01@gmail.com. Student Member AIAA.
‡
Graduate Research Assistant, Department of Mechanical and Aerospace
Engineering; gatorrue@gmail.com.
§
Distinguished Professor, Department of Mechanical and Aerospace
Engineering; haftka@ufl.edu. Fellow AIAA.
¶
Knox T. Millsaps Professor, Department of Mechanical and Aerospace
Engineering; ifju@ufl.edu.
**Graduate Research Assistant, Department of Mechanical Engineering
and Engineering Science; ctyler6@uncc.edu.
††
Graduate Research Assistant, Department of Mechanical Engineering
and Engineering Science; v.gang84@gmail.com.
‡‡
Professor, Department of Mechanical Engineering and Engineering
Science; tony.schmitz@uncc.edu.
3039
AIAA JOURNAL
Vol. 53, No. 10, October 2015
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
the scatter in weight measurements between nominally identical
wings and their measured aerodynamic force production.
Earlier work on flapping MAV wings used a flexible wing design
[19]. The hand-layup technique used in this study was subsequently
extended to a flapping platform [20], tested experimentally, and
compared with a computational model [21]. This manufacturing
process was also adopted in an experimental investigation of the
deformation during flapping [22], recovering the importance of
flexibility in the production of aerodynamic thrust force. Many
studies use a similar fabrication process, using carbon fiber skeletons
with a thin membrane constructed with a combination of hand-
assembled and commercially available parts [23–30]. In some cases,
extra components like macrofiber composites [31], small check
valves [32], and even strain-rate sensors (polyvinylidene fluoride)
[33,34] are added. Unfortunately, identification of the uncertainty in
the aerodynamic performance from wings made in this manner is
commonly overlooked. For example, the performance of a wing
could change based on a subtle change in the angle of a spar [35],
possibly the result of human error. This paper addresses the
uncertainty in the aerodynamic force that is present due to
manufacturing variability for wings fabricated in a similar manner to
the studies mentioned earlier. A universal approach is detailed for
segregating the contribution of manufacturing variability from the
total uncertainty, opening this analysis technique to novel approaches
in wing fabrication not covered in this paper [36,37]. This paper
provides improvements on these commonly used manufacturing
techniques along with an approach for identifying the contribution of
uncertainty from manufacturing. This can be beneficial to those who
use experimental techniques in flapping.
Tests were completed to identify the performance consistency for a
series of manufacturing methods; each improving upon the pre vious.
Wing replicateswere alsotested to compare thevarious methods for the
repeatability of thrust force production. Experiments included 1) 10
tests of the same wing and 2) tests of multiple (10) wings of the same
design for each manufacturing method. The former was used to
identify the testing repeatability in the absence of manufacturing
uncertainty contributions, whereas the latter was used to explore the
capability of the selected manufacturing method to produce the same
wing in a repeatable manner. Thus, perfect manufacturing repeatability
would present as identical scatter in the measurements made in test sets
1 and 2. Weight measurements were taken for each wing, serving
as a direct metric of manufacturing repeatability. It is shown to
improve with the progression of the manner in which the wings are
manufactured.
II. Experimental Procedure and Setup
Four fabrication techniques were analyzed for two reinforcement
designs (Figs. 1 and 2); this includes two different frame materials.
Because the uncertainty could be driven by the type of wing design,
two different wing designs were investigated. The horizontal spar that
extends along the top of the wing is called the leading edge (LE),
whereas the vertical spar protruding from the triangular corner is
called the root batten; these are present in both wing designs. Batten
supports emanate radially or vertically downward, comprising the
two designs used.
Thrust measurements are the intended metric for aerodynamic
performance for the wings and can be thought of as the value of the
objective function for optimization. To identify this, the wings were
flapped at 20, 25, and 30 Hz with stroke angle amplitude of 96 deg,
and the mean aerodynamic force was recorded. The flapping
mechanism in Fig. 3 secures the wings at the triangular plate in the
base corner of the wing. Every measurement is performed with both
wings secured with a screw attachment. The stroke plane and the
wing plane are perpendicular when the wings are at rest. There is no
hinge in the mechanism to allow for wing rotation at the mount
(supination/pronation), and so the wing frame twists during the
stroke, relying on the torsional stiffness of the wing for rotation. A dc
rotary motor/encoder (Maxon EC16) drives the mechanism, whereas
an EPOS 24∕1 controller was used to define the flapping frequency
via the built-in motor encoder. The mechanism was mounted on a six-
axis force and torque sensor (ATI Industrial Automation Nano17),
which was sampled using a 16 bit data acquisition device (National
Instruments USB-6251). The Nano17 load sensor is calibrated to
resolve down to 0.3 gram force (gf). Both the controller and sensor
were managed using LabVIEW virtual instruments. The sampling
rate was 30 kHz and a second-order Butterworth low-pass filter was
applied with a cutoff frequency of 3 Hz to reduce noise and isolate the
steady-state thrust output. A full measurement for one trial of a
Fig. 1 Four processes are displayed for radial batten wing design along
with dimensions of the wing.
Fig. 2 Four processes are displayed for parallel batten wing design
along with dimensions of the wing.
3040
CHANG ET AL.
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
parallel batten wing design is shown in Fig. 4, comparing the raw
measurement to the filtered data. As seen in Fig. 4a, the noise
overwhelms any prospect of obtaining a time-resolved force
contribution from the wings. This is the reason why the average thrust
was chosen as the metric of interest.
The data was filtered in real time, allowing the user to quickly
identify a wing that had incurred damage. Any wing that was found to
have damage was removed from the study. Each trial consisted of a 6 s
tare reading to zero the force measurements, followed by a 7 s interval
at 20 Hz, and two consecutive 6 s flapping intervals at 25 and 30 Hz.
These are commanded frequencies, and so there is an initial ramping
stage where the flapping mechanism has yet to obtain the desired
frequency. To account for this, the first 2 s of the 20 Hz flapping
interval is ignored, and the first second of the 25 and 30 Hz intervals
are ignored. The mean force in this interval is used in the uncertainty
analysis. Therefore, every wing has one value of thrust for each
flapping frequency. Also, the lift force averages close to zero for
wings in this study. This is due to the symmetric flapping motion.
III. Fabrication Process
Two different wing designs (shown in Fig. 1) were fabricated using
four manufacturing methods. Each of these techniques was used to
make multiple replicates, which were nominally identical. For the
remainder of this manuscript, these manufacturing methods will be
abbreviated as follows:
1) A vacuum-bagged hand layup skeleton bonded to plastic
membrane skin with cyanoacrylate (CA) adhesive is referred to as the
hand layup.
2) A carbon fiber skeleton made from a machined [computer
numerically controlled (CNC)] Teflon® mold and adhered to a
plastic membrane skin with CA is called the machined mold.
3) A CNC machined acetal resin skeleton with a channel milled
along the leading edge to serve as a locating feature for a commercial-
off-the-shelf (COTS) carbon fiber (CF) rod is called a plastic skeleton.
The two are bonded with rubber-toughened CA to make the skeleton
assembly. Then, the assembly is bonded to the membrane with CA.
4) The plastic skeleton with transfer tape is the same skeleton
assembly, differing in that the membrane is adhered with transfer tape
as opposed to CA.
The preceding list is the order in which they were devised, each
addressing the shortcomings of the previous one. Details of this are
provided in the individual descriptions. Common in all the processes
is the membrane material. A 14-μm-thick nylon film (Honeywell
CAPRAN® 1200 Matte) was used for its minimal weight con-
tribution and strength. Both wing designs (radial and parallel battens)
were produced using each manufacturing method to compare the
thrust and weight repeatability.
A. Hand Layup Method
The initial method for fabricating the wings involved hand layup of
carbon fiber strips on a flat plate, using vacuum pressure to ensure
proper consolidation. The preimpregnated (prepreg) unidirectional
carbon fiber used in this study incorporated an Aldila AR250 resin
system. It was hand cut to approximately 1 mm wide strips, which
were then laid onto a rigid aluminum plate with printed placement
guides. This is shown in Fig. 5. Three layers were used on the leading-
edge spar and one layer was used on the battens. Three layers of
bidirectional carbon weave were also cut in triangular shapes to
function as the rigid plate for attachment to the flapping device. A
similar process was used in the structural deformation study by
Keennon and Klingbiel [1]. The wings were cured in an oven
following the manufacturer’s recommended two-stage curing cycle
with a temperate ramp to 190°F and then to 275°F with a hold
time of 1 h (2.5 h total, including the ramp times). To assist in
consolidation, 30 in Hg of vacuum pressure was applied using a
vacuum bag approach for the entire cure cycle. After cure, the frames
were carefully removed from the plate. The carbon fiber skeletons
were cleared of any cured residual epoxy then adhered to the
membrane plastic via CA. The adhesive was applied to the skeleton
with a small foam brush and then the skeleton was pressed onto the
membrane material. The wing outline was then cut using a knife with
the aid of a metal template in the shape of the wing periphery.
Fig. 3 Flapping mechanism loaded on a six-axis load sensor.
Fig. 4 Comparison of a) raw thrust measurement and b) filtered thrust measurement for one trial.
CHANG ET AL. 3041
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
B. Machined Mold Method
The second manufacturing approach was devised with the aim of
removing the human error involved in placing the battens. To
accomplish this, a CNC-machined polytetraflouroethylene (PTFE,
Teflon) mold was used as a base for the unidirectional carbon fiber
strips. The topology of each wing frame was CNC machined onto the
surface of a 3-mm-thick sheet of PTFE using a 1-mm-diam, four-flute
square end mill on a three-axis Haas® TM-1 milling machine (see
Fig. 6). The speeds and feeds were chosen to avoid melting and/or
plowing of the plastic and to enable proper chip formation. A chip
load of 0.07 mm∕tooth at the maximum attainable spindle speed of
4000 rpm was selected.
The triangular pockets of the mold held three layers of hand-cut
bidirectional carbon fiber; the average thickness of each bidirectional
weave layer was approximately 300 μm. Each layer was cut into the
shape of right triangles, with sides nominally measured at 6 and
12 mm. The linear channels machined into the mold contained the 1-
mm-wide, 130-μm-thick unidirectional carbon fiber strips, the same
type used in the hand layup method. To achieve the necessary
pressure to bond the layers of the wing assembly, the depth of each
mold was undercut in the machining process. The triangular pockets
and channels were machined at commanded depths of 340 and
100 μm, respectively. The widths of the channels were machined to
be 1 mm, or a 100% radial immersion cut.
Controlling the thickness of the battens was another aim of this
manufacturing improvement. This was accomplished with a custom
hand tool. This bladed device was used to consistently cut the
unidirectional strips. It is a hand tool that rigidly holds two cutting
blades separated by 0.8 mm. This enabled uniform strips to be cut with
each pass (see Fig. 7). This device reduced cutting time and could be
configured to cut as many as eightstrips side by side. It servedto further
reduce human error because previous strips were cut using a single
blade tool that would be aligned using a straightedge ruler.
The vacuum bag approach was replaced with a sandwich assembly
to reduce process time. Once the strips were placed in the channels, a
release film (25 μm thick) and a 40A durometer silicone sheet with a
thickness of 1.59 mm were placed on top of the strips to provide
consolidation pressure. The sandwich assembly was produced by
clamping these components between two 6.35-mm-thick aluminum
plates. The plates were secured and clamping pressure was applied
using 16 screw holes around the periphery to tighten down on the
inner layers (see Fig. 8). To reduce warping of the CNC-machined
Teflon sheet during the curing process, it was adhered to the
aluminum plate using spray adhesive. This eliminated shifting of the
PTFE sheet during layup, whereas external pressure supplied from
the tightened nuts was responsible for securing the sheet during the
curing process. The same oven curing profile as the flat plate method
was employed because the same carbon fiber was used. Once cured,
Fig. 5 Parallel batten wings manufactured with hand layup method
assembled on a flat plate before curing cycle.
Fig. 6 PTFE sheet is milled with channels for use in machined mold
method (a subtractive machining process).
Fig. 7 Custom hand tool configured to cut one fixed width strip of
unidirectional prepreg.
Fig. 8 Exploded view of machined mold.
3042
CHANG ET AL.
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
any excess flashing was removed from the skeletons, and they were
adhered to the nylon film in the same manner as the hand layup
method. In the last step, the periphery was trimmed.
C. Plastic Skeleton Methods
This next manufacturing process involves a larger departure, using
a plastic frame as opposed to all carbon fiber. The aim of this
approach was to further reduce human-driven errors while attempting
to preserve or even improve the wing’s thrust production. In this
process, a 250-μm-thick sheet of acetal resin (Delrin®) was CNC
machined in the shape of the wing skeleton and a COTS 0.5-mm-
diam Graphlite® carbon fiber rod was used to reinforce the leading
edge. The pultruded carbon fiber rod is specified to have a 67% fiber
volume. Acetal resin was chosen due to its good fatigue properties,
high stiffness, and machinability. The contour of the wing frame
was machined using a 1-mm-diam, four-flute square end mill on a
three-axis Haas TM-1 milling machine. Each wing frame was
manufactured with a 200-μm-deep, 500-μm-wide channel on the
leading edge to locate the carbon fiber rod (see Fig. 9). This was
accomplished using a 0.40 mm diameter, four-flute square end mill.
The spindle speed (4000 rpm) and feed per tooth (0.07 mm∕tooth)
were selected to avoid polymer melting during machining. Even
though acetal resin has an exceptional natural lubricity, a TiAlN
coating was chosen for each cutter to ensure longer cutting life and
proper chip evacuation under the full immersion cutting conditions.
The milling process and resulting plastic wing frames are illustrated
in Fig. 10. The carbon fiber rod leading edge was adhered to the
plastic frame with CA, the skeleton was adhered to the film with CA,
and then the periphery nylon film material was trimmed.
Unfortunately, there were adhesive failures in the bond between
the nylon film and plastic skeleton; the final manufacturing
improvement addresses this. The last manufacturing process is a
variation of the plastic frame approach described earlier. As opposed
to using CA, a high-strength acrylic adhesive transfer tape (Scotch®
3M 9471LE 2.3 mil∕58 μm thick) was used to adhere the plastic
frame to the nylon film. This was incorporated to add compliance and
toughness to the bond between the plastic frame and CAPRAN film,
making it less prone to failures. As with the previous evolutions, the
aim was to sustain thrust force production as well. Extra steps were
required for this process, although the overall time was reduced due to
the elimination of adhesive curing time. The frame was first placed
onto the transfer tape, which consists of a thin film of adhesive
(mastic) on a wax paper. The mastic is designed to separate from the
wax paper easily. A sheet was used to sandwich the frame (e.g., white
copy paper). This step was used to remove the adhesive that was not
needed, breaking the edge of the adhesive film (Fig. 11c). At this
point, the adhesive was on the frame and there was no adhesive
between the battens. The wax paper backing was peeled away to leave
the frame with a complete layer of adhesive. Subsequently, the frame
was removed and carefully laid onto the nylon film, and then hand
pressure was applied to insure uniform adhesion (Fig. 11d). The
process of applying the adhesive to the backside of the frame is
provided in Figs. 11a–11d as illustrations.
IV. Manufacturing Uncertainty Quantification
The uncertainty quantification for wing thrust measurements
outlined by Chaudhuri et al. [15] was used in this work. To quantify
the manufacturing uncertainty, the testing uncertainty must be
separated from the total uncertainty [Eq. (1)]. To quantify testing
uncertainty, the coefficient of variation (CV) of 10 measurements of
Fig. 9 Plastic skeleton wing construction with cross-sectional geometry
of leading edge shown (A-A).
Fig. 10 CNC process: a) black plastic sheet is milled; b) root area;
c) milled channel for carbon fiber rod.
Fig. 11 Adhesive application process: a) skeleton placed on transfer
tape; b) copy paper sheet applied, adhering to exposed mastic; c) paper
peeled from assembly; and d) skeleton lifted from transfer tape backing,
retaining the mastic for adhesion to plastic membrane.
CHANG ET AL. 3043
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
each wing was found. The CV is simply the ratio of the standard
deviation to the mean. In this experiment, 10 measurements from
each wing were used to determine the mean thrust. This reduced the
testing uncertainty in the mean thrust by a factor of
p
10. The root
mean square of the reduced coefficients of variation [38,39] was
calculated to find the testing uncertainty (CV
test
).
The total uncertainty (CV
tot
), which combined the testing and
manufacturing uncertainties of the wings, was computed by first
taking the mean of 10 measurements, obtaining a mean thrust value
for each replicate (sampled manufactured means). Then, the CV of
the sampled manufactured means for each design (one for radial and
one for parallel) was calculated. These two coefficients of variation
were then combined using the root mean square to quantify the CV
tot
.
The manufacturing uncertainty can be computed by subtracting
the testing uncertainty in the mean thrust for each wing from the total
uncertainty. However, for small sample sizes, the process of taking
the square root of the variance to obtain the standard deviation
introduces a bias in the standard deviation estimate [40]. The same
applies to the CV because it is the standard deviation over the mean. A
simple bias correction can be completed for normal distributions
(which was assumed here) using a correction factor [c
4
n, where n is
the number of samples] that depends on the sample size [40]. Because
of there being 10 replicates available for each design, the corrected
coefficient of variation for manufacturing uncertainty (CV
mfg
) was
obtained by dividing by the correction factor for a sample size of 10
c4100.9727, which accounts for a 3% increase of the
calculated CV. The affect for this correction is minimal for this case,
though it can be close to 9% for a sample size of 4:
CV
mfg
CV
2
tot
− CV
2
test
p
c
4
10
(1)
V. Results and Discussion
Finding the proper manufacturing process for the experimental
optimization was a matter of identifying a procedure that possesses
the smallest uncertainty and is not time intensive, because it is
expected that many wings would need to be made for an optimization.
Detailed numerical results for each manufacturing process and wing
design are provided along with their shortcomings.
A. Thrust Force Uncertainty
For this study, the analysis is limited to 30 Hz flapping frequency,
because it represented the upper limit of frequency that the wings
could handle without incurring damage on a regular basis. It also
introduced the largest testing-based variability and demand on the
controller. It also produced a feasible thrust to propel a freestanding
hovering device, making it most relevant to a possible application.
Table 1 provides a dissected view of CV
mfg
for each manufacturing
process. The values were calculated using Eq. (1) for a 30 Hz
commanded flapping frequency. The testing uncertainty is the CV
test
value divided by
p
10, accounting for the 10 trials for each design.
For clarification, the CV
test
and CV
tot
values were formulated from
biased standard deviations. As mentioned earlier, they were corrected
using c4n. The manufacturing CV is provided, as a percentage, for
each manufacturing process in Fig. 12.
The uncertainty in these results was calculated with a bootstrapping
resampling technique, where 5000 datasets were sampled from the
original dataset with replacement. This approach provides the sam-
pling distribution for statistics like CV
tot
. It entails making N datasets
from an original set, then calculating the same metric for each all N
datasets. This gives a distribution for the metric and was chosen for its
simplicity. It assumes that the samples are independent and that they
comefrom the same distribution;there is no normality assumption. The
uncertainty value provided in the tables (Tables 1–3) and plots(Figs. 12
and 13) pertain to one standard deviation of the bootstrap distribution
of the statistic.
Table 1 shows different mean thrust for each manufacturing meth-
od. This is investigated more through observations of the deformation
during flapping. The details of this are provided in the Appendix. The
differences in mean thrust between the methods can possibly be
attributed to the differences in stiffness of the constructions. In
subsequent work, the stiffness [41] is characterized for wings made in
the plastic construction.
Subtle changes in CV
tot
were seen for the transfer tape, although
adhesive failure of the wings was observed much less frequently with
this method. The transfer film was also a faster/easier method for
applying the nylon film to the skeleton. The manufacturing improve-
ments assist in realizing wing-specific differences that were previ-
ously overwhelmed by larger uncertainties. Therefore, there is more
resolution in comparing different designs, which is also important
for subsequent optimization [42]. The two design coefficients of
variation presented in Table 1 can be combined, using the root mean
Table 1 Thrust CV results for each manufacturing process and design with one standard
deviation intervals
Hand layup Machined mold Plastic skeleton Plastic skeleton with transfer tape
Radial batten
Mean thrust, g 5.77 5.49 7.52 7.68
CV
tot
,% 12.7 1.89 7.19 1.42 3.15 0.50 2.40 0.39
CV
test
,% 0.67 0.10 0.82 0.07 0.82 0.09 0.74 0.18
CV
mfg
,% 13.0 1.94 7.3 1.48 3.1 0.54 2.3 0.44
Parallel batten
Mean thrust, g 5.73 5.62 7.37 7.70
CV
tot
,% 19.41 4.11 10.51 2.49 2.52 0.36 2.74 0.72
CV
test
,% 0.78 0.08 0.79 0.09 0.55 0.05 0.45 0.03
CV
mfg
,% 19.9 4.2 10.8 2.6 2.5 0.4 2.8 0.8
Note: The calibration error is 0.3 g. This is an additional possible bias.
Fig. 12 CV
mfg
is provided for different manufacturing processes as a
percentage.
3044
CHANG ET AL.
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
square to compare the different manufacturing techniques. This is
presented in Table 2. The testing uncertainty, contributing less to the
total uncertainty, was found to vary little between manufacturing
processes.
This is expected because no intended modifications to the process
of testing the wings were made between any of the tests, although the
subtle changes could be attributed to diversity in the mounting or
fatigue of the wing between trials, all of which could be specific to the
manufacturing process used to make the wing. The results show a
sixfold improvement of manufacturing uncertainty when the radial
and parallel batten statistics are combined.
B. Weight Uncertainty
The weight of each of the 160 wings (80 wing pairs) varied with
both the manufacturing method and design. Each individual wing
was measured with a Gemini-20 portable milligram scale and less
scatter was observed with each manufacturing progression (Table 3).
The average pair discrepancy pertains to the weight difference
between the left and right wings. A reduction in weight discrepancy is
present for the parallel design, though it does not fluctuate for the
radial batten wing design. The weight of the plastic skeleton wings
compare with those used on the Saturn prototype, a long-endurance
variation of the Nano Hummingbird, which possesses 0.26 g for a
pair of wings, making each wing weigh approximately 0.13 g; both
wings pertain to 1% of the flier’s total weight [1].
The CV for both designs and all four manufacturing processes is
presented in Fig. 14. Each bar entails 20 independent wing measure-
ments (10 pairs); the plot covers every combination of fabrication
process and wing design.
C. Cross-Sectional Geometry
From visual inspection, the hand layup approach exhibited the
most cross-sectional variation. Distortion of the cross section was due
to external pressure applied from the vacuum bag, resulting in a
curved geometry; this is visible in Fig. 14. This deformation was
consistently present and is likely a key contributor to the variability
seen in the thrust performance. Though the widened base of the cross
section provides more surface area to promote adhesion, it reduces
the second moment of area and thus the bending stiffness in an
uncontrolled manner; this makes for varying levels of deformation
during flapping, accompanied by differences in the thrust perfor-
mance in nominally identical wings.
The next progression was the machined mold approach which,
from visual inspection, successfully controlled batten thicknesses
and improved the precision of the batten placement. The machined
mold provided constraining channel walls that deterred batten
movement during the curing cycle. The wing frame, fabricated from
the machined mold process, possessed a defined leading edge and
more consistent geometric shape. These improvements are reflected
in both the weight CV and thrust force CV compared with the hand
layup process. By using machined channels, consistency in the batten
Table 2 Combined manufacturing uncertainty with one standard deviation intervals
Method Total uncertainty, CV
tot
% Testing uncertainty, CV
test
% Manufacturing uncertainty, CV
mfg
%
Hand layup 16.38 2.41 0.73 0.06 16.8 2.5
Machined mold 9.01 1.54 0.80 0.06 9.2 1.6
Plastic skeleton 2.85 0.31 0.70 0.06 2.8 0.3
Plastic skeleton with transfer tape 2.58 0.41 0.61 0.11 2.6 0.4
Note: The calibration error is 0.3 g. This is an additional possible bias.
Table 3 Weight results with one standard deviation intervals
Hand layup Machined mold Plastic skeleton Plastic skeleton with transfer tape
Radial batten
Average pair discrepancy, g 0.006 0.005 0.006 0.006
Mean weight, g 0.155 0.004 0.158 0.003 0.150 0.002 0.157 0.001
Standard deviation, g 0.012 0.002 0.009 0.002 0.006 0.001 0.004 0.001
CV, % 7.8 1.2 5.7 1.1 4.2 0.8 2.5 0.448
Parallel batten
Average pair discrepancy, g 0.011 0.006 0.005 0.002
Mean weight, g 0.153 0.007 0.142 0.001 0.132 0.001 0.135 0.001
Standard deviation, g 0.022 0.004 0.003 0.001 0.003 0.001 0.002 0.000
CV, % 14.2 2.5 1.9 0.5 2.4 0.6 1.4 0.2
Combined
CV, % 11.4 1.5 4.3 0.7 3.5 0.5 2.0 0.3
Fig. 13 Weight CV categorized by manufacturing process and design.
Fig. 14 Uncontrolled geometric changes of LE spar due to vacuum bag
curing process. Progression shows view of LE cross section.
CHANG ET AL. 3045
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
placement was achieved, and reduced distortion of the cross-
sectional geometry was visible; this is shown in Fig. 15. Another
contributor to the batten width consistency is the comb cutter hand
tool. It prepared strips of unidirectional carbon fiber with uniform
thickness, a visual improvement over a single blade hand-cutting tool
used with a straightedge.
The shortcomings of the machined mold approach were based on
inconsistency at the unconstrained surface, where the wing’s frame
does not touch the mold’s walls (interfaces with release film). Under-
and overfilling of the channels of the carbon was observed. A
combination of insufficient channel depth, inconsistent pressure, and
variations in the amount of carbon in the channel were responsible for
the geometries displayed in Fig. 16. The flatness of this face is
important for an even bond to the nylon film, and so adhesive failures
were seen more frequently as compared with the hand layup wings.
This top face would often cure to a curved shape as shown in Fig. 16
(underfilled) causing voids/bubbles in the bond line where cohesive
failure would emanate. The sensitivity of the interplay between the
applied pressure, channel depth, and amount of carbon fiber placed
into the mold was found to produce an inconsistent top surface
geometry. Increasing the machining accuracy would possibly
improve this, though it would add considerably to machining time
and costs.
The plastic skeleton wings, machined with a CNC, provided the
least scatter in the thrust and weight. The introduction of a transfer
tape adhesive to bond the nylon film to the frame introduced minimal
improvements to the scatter, although it was the most durable
construction, reducing fabrication time and complexity. The failures
seen in wings using CA were likely due to the low surface energy of
the Delrin, which hinders adhesion, and the brittle nature of CA,
resulting in cohesive failure. The adhesion of the leading-edge carbon
rod is shown in Fig. 17, where the CA adhesive fills the rectangular
channel and interfaces with the sides as well. This is the result of
wicking, where the excess CA is sometimes visible on the top surface
of the plastic skeleton; it provides more surface area for the adhesive
to bond to.
VI. Conclusions
With reduced uncertainty in average thrust, smaller differences in
thrust due to design variations can be resolved. This reduction is
especially important for application to experimental optimization,
where surrogates of the thrust performance (with respect to design
variables) take into consideration the level of uncertainty in the
measurements when fitting the data. The manufacturing process used
to fabricate the flapping wings will continuously evolve, though it has
developed to a level, based on uncertainty analysis, that is worthy
of an experimental optimization. The weight, measured for all of
the wings tested, revealed lower uncertainty with the progressive
improvements of the manufacturing method. Through extensive
testing of 80 wing pairs, it has been shown that the CV was reduced
by more than 6 times with manufacturing process improvements that
incorporate more precise techniques and materials. This includes
testing of two different wing designs, showing that the uncertainty
measures vary less with improved manufacturing methods. A CNC-
machined acetal resin frame attached to a nylon film with transfer
film adhesive was found to have the least manufacturing uncertainty.
This approach was the product of successive fabrication process
improvements, some of which were compared here. The repeatability
obtained through these methods was the result of carefully followed
procedures. Therefore, small details of the process were provided.
There is hope that the improvements and this analysis can be used
as a framework for others to interrogate the suitability of specific
manufacturing methods to produce repeatable performance in
flapping wing tests of all types, including those not covered in this
work. As it stands, this is the most comprehensive prospective study
of manufacturing uncertainty in flapping wings and how various
techniques affect aerodynamic force measurements. In subsequent
work, the authors implemented a power consumption metric along
with an uncertainty analysis so that efficiency can be used as an
objective in an optimization. Also, the stiffness was characterized for
different designs in the optimizations. In future work, deformation
throughout the flap cycle will be measured to recover the
characteristics of high-thrust performance wings.
Appendix: Wing Deformation Study
A separate study was conducted to observe the pattern of
deformation slightly after stroke reversal, where deformation was
maximum. This was done to help understand the increase in mean
thrust with the introduction of a plastic frame construction. The full-
field deformation results seen in Fig. A1 were captured using digital
image correlation at approximately 5 deg from the stroke reversal,
where the out-of-plane deformation was found to be greatest. A
pair of Point Grey Research Flea2® cameras were used along
with Correlated Solution’s VIC-3D™ software. The wing was
painted black and speckled with white enamel paint in a thin coat.
Measurements were taken on wings flapped at 25 Hz, because 30 Hz
flapping frequency made for excessive deformation, where the
membrane would orient out of the cameras’ view, making for poor
correlation.
Figure A1 shows the deflection of parallel batten wings in three
different constructions. Here, the plastic wing is one where the frame
was attached to the membrane with transfer tape. The CA-adhered
Fig. 15 Comparison of wings made with hand layup and machined
mold manufacturing methods.
Fig. 16 Typical defects occur when either the mold is under- or
overfilled. Microscope still shows a leading edge that was sectioned.
Fig. 17 Magnified view of cross section where leading edge carbon fiber
spar is adhered to plastic skeleton.
3046
CHANG ET AL.
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
variation was neglected because its construction is similar. The plot
shows the out-of-plane deflection of the wings, and the white
segments on the trailing edge pertain to unmeasured regions (due to
glare and folding of the membrane during flapping). The hand layup
wing had the least leading-edge tip deflection and the machine mold
wing had the most. The plastic distinguishes itself with deflection at
the root of the wing, indicating more twist throughout the body. This
suggests that twisting throughout the length of the wing body has
positive effects on the average thrust.
Acknowledgments
This work was supported by the U.S. Air Force Office of Scientific
Research grant FA9550-11-1-0066 from David Stargel, Grant
Monitor. Also, thanks to Jordan Van Hall for his manufacturing and
testing contributions.
References
[1] Keennon, M., and Klingebiel, K., “Development of the Nano
Hummingbird: A Tailless Flapping Wing Micro Air Vehicle,” AIAA
Aerospace Sciences Meeting, AIAA, Reston, VA, 2012, pp. 1–24.
[2] Wood, R. J., “First Takeoff of a Biologically Inspired At-Scale Robotic
Insect,” IEEE Transactions on Robotics, Vol. 24, No. 2, 2008, pp. 341–
347.
doi:10.1109/TRO.2008.916997
[3] Ma, K. Y., and Felton, R. J., “Wood, Design, Fabrication, and Modeling
of the Split Actuator Microrobotic Bee,” 2012 IEEE/RSJ International
Conference on Intelligent Robots and Systems, IEEE Publ., Piscataway,
NJ, 2012, pp. 1133–1140.
doi:10.1109/IROS.2012.6386192
[4] Bejgerowski, W., Ananthanarayanan, A., Mueller, D., and Gupta, S. K.,
“Integrated Product and Process Design for a Flapping Wing Drive
Mechanism,” Journal of Mechanical Design, Vol. 131, No. 6, 2009,
Paper 061006.
doi:10.1115/1.3116258
[5] Mountcastle, A. M., and Daniel, T. L., “Aerodynamic and Functional
Consequences of Wing Compliance,” Experiments in Fluids, Vol. 46,
No. 5, 2009, pp. 873–882.
doi:10.1007/s00348-008-0607-0
[6] Heathcote, S., and Gursul, I., “Flexible Flapping Airfoil Propulsion
at Low Reynolds Numbers,” AIAA Journal, Vol. 45, No. 5, 2007,
pp. 1066–1079.
doi:10.2514/1.25431
[7] Shyy, W., Aono, H., Chimakurthi, S. K., Trizila, P., Kang, C.-K., Cesnik,
C. E. S., and Liu, H., “Recent Progress in Flapping Wing Aerodynamics
and Aeroelasticity,” Progress in Aerospace Sciences, Vol. 46, No. 7,
2010, pp. 284–327.
doi:10.1016/j.paerosci.2010.01.001
[8] Gordnier, R. E., Kumar Chimakurthi, S., Cesnik, C. E. S., and Attar, P. J.,
“High-Fidelity Aeroelastic Computations of a Flapping Wing with
Spanwise Flexibility,” Journal of Fluids and Structures, Vol. 40,
July 2013, pp. 86–104.
doi:10.1016/j.jfluidstructs.2013.03.009
[9] Ansari, S. A., Żbikowski, R., and Knowles, K., “Non-Linear Unsteady
Aerodynamic Model for Insect-Like Flapping Wings in the Hover. Part
1: Methodology and Analysis,” Journal of Aerospace Engineering,
Vol. 220, No. 2, 2006, pp. 61–83.
doi:10.1243/09544100JAERO49
[10] Gogulapati, A., Friedmann, P. P., Kheng, E., and Shyy, W.,
“Approximate Aeroelastic Modeling of Flapping Wings in Hover,”
AIAA Journal, Vol. 51, No. 3, 2013, pp. 567–583.
doi:10.2514/1.J051801
[11] Pourtakdoust, S. H., and Aliabadi, S. K., “Evaluation of Flapping Wing
Propulsion Based on a New Experimentally Validated Aeroelastic
Mode,” Scientia Iranica, Vol. 19, No. 3, 2012, pp. 472–482.
doi:10.1016/j.scient.2012.03.004
[12] Stanford, B., Kurdi, M., Beran, P., and McClung, A., “Shape, Structure,
and Kinematic Parameterization of a Power-Optimal Hovering Wing,”
Journal of Aircraft, Vol. 49, No. 6, 2012, pp. 1687–1699.
doi:10.2514/1.C031094
[13] Tuncer, I., and Kaya, M., “Optimization of Flapping Airfoils for
Maximum Thrust,” AIAA Journal, Vol. 43, No. 11, 2005, pp. 2329–
2336.
doi:10.2514/1.816
[14] Chaudhuri, A., Haftka, R., Chang, K., Van Hall, J., and Ifju, P., “Thrust-
Power Pareto Fronts Based on Experiments of a Small Flapping Wing,”
10th AIAA Multidisciplinary Design Optimization Conference, AIAA,
Reston, VA, 2014, pp. 1–9.
[15] Chaudhuri, A., Haftka, R. T., Ifju, P., Chang, K., Tyler, C., and Schmitz,
T., “Experimental Flapping Wing Optimization and Uncertainty
Quantification Using Limited Samples,” Structural and Multidiscipli-
nary Optimization, Vol. 51, No. 4, 2015, pp. 957–970.
doi:10.1007/s00158-014-1184-x
[16] Chaudhuri, A., Chang, K., Haftka, R., and Ifju, P., “Multi-Objective
Experimental Optimization with Multiple Simultaneous Sampling for
Flapping Wings,” 56th AIAA/ASCE/AHS/ASC Structures, Structural
Dynamics, and Materials Conference, AIAA, Reston, VA, 2014,
pp. 1–17.
doi: 10.2514/6.2015-1586
[17] Viana, F., and Haftka, R., “Efficient Global Optimization with
Experimental Data: Revisiting the Paper Helicopter Design,” 52nd
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and
Materials Conference, AIAA, Reston, VA, 2011, pp. 1–21.
[18] Jones, D., Schonlau, M., and Welch, W., “Efficient Global Optimization
of Expensive Black-Box Functions,” Journal of Global Optimization,
Vol. 13, No. 4, 1998, pp. 455–492.
doi:10.1023/A:1008306431147
[19] Ifju, P., Jenkins, D., Ettinger, S., and Lian, Y., “Flexible-Wing-Based
Micro Air Vehicles,” AIAA Paper 2002-0705, 2002.
[20] Wu, P., Ifju, P., and Stanford, B., “Flapping Wing Structural
Deformation and Thrust Correlation Study with Flexible Membrane
Wings,” AIAA Journal, Vol. 48, No. 9, 2010, pp. 2111–2122.
doi:10.2514/1.J050310
[21] Wu, P., Sällström, E., Ukeiley, L., Ifju, P., Chimakurthi, S., Aono, H.,
Cesnik, C. E. S., and Shyy, W., “Integrated Experimental and
Computational Approach to Analyze Flexible Flapping Wings in
Hover,” Conference Proceedings of the Society of Experimental
Mechanics, Soc. for Experimental Mechanics, Bethel, CT, 2011,
pp. 1441–1451.
doi:10.1007/978-1-4419-9834-7
[22] Wu, P., Stanford, B. K., Sällström, E., Ukeiley, L., and Ifju, P. G.,
“Structural Dynamics and Aerodynamics Measurements of Biologi-
cally Inspired Flexible Flapping Wings,” Bioinspiration and
Biomimetics, Vol. 6, No. 1, 2011, Paper 016009.
doi:10.1088/1748-3182/6/1/016009
[23] Nakata, T., Liu, H., Tanaka, Y., Nishihashi, N., Wang, X., and Sato, A.,
“Aerodynamics of a Bio-Inspired Flexible Flapping-Wing Micro
Air Vehicle,” Bioinspiration & Biomimetics, Vol. 6, No. 1, 2011,
Paper 045002.
doi:10.1088/1748-3182/6/4/045002
[24] Mueller, D., Bruck, H. A., and Gupta, S. K., “Measurement of Thrust
and Lift Forces Associated with Drag of Compliant Flapping Wing for
Micro Air Vehicles Using a New Test Stand Design,” Experimental
Mechanics, Vol. 50, No. 6, 2009, pp. 725–735.
doi:10.1007/s11340-009-9270-5
[25] Fenelon, M., and Furukawa, T., “Design of an Active Flapping
Wing Mechanism and a Micro Aerial Vehicle Using a Rotary
Actuator,” Mechanism and Machine Theory, Vol. 45, No. 2, 2010,
pp. 137–146.
doi:10.1016/j.mechmachtheory.2009.01.007
[26] Shkarayev, S., Maniar, G., and Shekhovtsov, A. V., “Experimental and
Computational Modeling of the Kinematics and Aerodynamics of
Fig. A1 Displacement plots for three wing constructions.
CHANG ET AL. 3047
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
Flapping Wing,” Journal of Aircraft, Vol. 50, No. 6, 2013, pp. 1734–
1747.
doi:10.2514/1.C032053
[27] Mazaheri, K., and Ebrahimi, A., “Experimental Investigation of the
Effect of Chordwise Flexibility on the Aerodynamics of Flapping Wings
in Hovering Flight,” Journal of Fluids and Structures, Vol. 26, No. 4,
2010, pp. 544–558.
doi:10.1016/j.jfluidstructs.2010.03.004
[28] Hu, H., Kumar, A. G., Abate, G., and Albertani, R., “Experimental
Investigation on the Aerodynamic Performances of Flexible Membrane
Wings in Flapping Flight,” Aerospace Science and Technology, Vol. 14,
No. 8, 2010, pp. 575–586.
doi:10.1016/j.ast.2010.05.003
[29] Yoon, S., Kang, L., and Jo, S., “Development of Air Vehicle with Active
Flapping and Twisting of Wing,” Journal of Bionic Engineering, Vol. 8,
No. 1, 2011, pp. 1–9.
doi:10.1016/S1672-6529(11)60007-3
[30] Nguyen, Q. V., Truong, Q. T., Park, H. C., Goo, N. S., and Byun, D.,
“Measurement of Force Produced by an Insect-Mimicking Flapping-
Wing System,” Journal of Bionic Engineering, Vol. 7, Sept. 2010,
pp. S94–S102.
doi:10.1016/S1672-6529(09)60222-5
[31] Kim, D.-K., Kim, H.-I., Han, J.-H., and Kwon, K.-J., “Experimental
Investigation on the Aerodynamic Characteristics of a Bio-Mimetic
Flapping Wing with Macro-Fiber Composites,” Journal of Intelligent
Material Systems and Structures, Vol. 19, No. 3, 2007, pp. 423–431.
doi:10.1177/1045389X07083618
[32] Ho, S., Nassef, H., Pornsinsirirak, N., Tai, Y.-C., and Ho, C.-M.,
“Unsteady Aerodynamics and Flow Control for Flapping Wing Flyers,”
Progress in Aerospace Sciences, Vol. 39, No. 8, 2003, pp. 635–681.
doi:10.1016/j.paerosci.2003.04.001
[33] Raney, D. L., and Slominski, E. C., “Mechanization and Control
Concepts for Biologically Inspired Micro Air Vehicles,” Journal of
Aircraft, Vol. 41, No. 6, 2004, pp. 1257–1265.
doi:10.2514/1.5514
[34] Yang, L.-J., Hsu, C.-K., Ho, J.-Y., and Feng, C.-K., “Flapping Wings
with PVDF Sensors to Modify the Aerodynamic Forces of a Micro
Aerial Vehicle,” Sensors and Actuators A: Physical, Vol. 139, Nos. 1–2,
2007, pp. 95–103.
doi:10.1016/j.sna.2007.03.026
[35] Gerdes, J. W., Cellon, K. C., Bruck, H. A., and Gupta, S. K.,
“Characterization of the Mechanics of Compliant Wing Designs for
Flapping-Wing Miniature Air Vehicles,” Experimental Mechanics,
Vol. 53, No. 9, 2013, pp. 1561–1571.
doi:10.1007/s11340-013-9779-5
[36] Pornsin-Sirirak, T. N., Tai, Y. C., Nassef, H., and Ho, C. M., “Titanium-
Alloy MEMS Wing Technology for a Micro Aerial Vehicle
Application,” Sensors and Actuators A: Physical, Vol. 89, Nos. 1–2,
2001, pp. 95–103.
doi:10.1016/S0924-4247(00)00527-6
[37] Laliberté, J. F., Kraemer, K. L., Dawson, J. W., and Miyata, D., “Design
and Manufacturing of Biologically Inspired Micro Aerial Vehicle Wings
Using Rapid Prototyping,” International Journal of Micro Air Vehicles,
Vol. 5, No. 1, 2013, pp. 15–38.
doi:10.1260/1756-8293.5.1.15
[38] Gluer, C., Blake, G., Lu, Y., Blunt, B. A., Jergas, M. I., and H. K.,
Genant, “Accurate Assessment of Precision Errors: How to Measure the
Reproducibility of Bone Densitometry Techniques,” Osteoporosis
International, Vol. 5, No. 4, 1995, pp. 262–270.
doi:10.1007/BF01774016
[39] Dransfield, R. D., and Brightwell, R., “How to Get on Top of Statistics:
Design and Analysis for Biologists, with R. InfluentialPoints,” 2012,
http://www.influentialpoints.com/.
[40] Trietsch, D., Statistical Quality Control: A Loss Minimization
Approach, World Scientific, River Edge, NJ, 1999.
[41] Chang, K., Chaudhuri, A., Tang, J., Van Hall, J., Ifju, P., Haftka, R.,
Tyler, C., and Schmitz, T., “Stiffness Investigation of Synthetic Flapping
Wings for Hovering Flight,” Advancement of Optical Methods in
Experimental Mechanics, Springer International, New York, 2015,
pp. 249–256.
doi:10.1007/978-3-319-06986-9_28
[42] Chang, K., Rue, J., Ifju, P., Haftka, R., Schmitz, T., Tyler, C., Chaudhuri,
A., and Ganguly, V., Analysis of Thrust Production in Small Synthetic
Flapping Wings, Springer International, New York, 2014, pp. 1–8.
doi:10.1007/978-3-319-00876-9
N. Wereley
Associate Editor
3048 CHANG ET AL.
Downloaded by UNIV OF NORTH CAROLINA-CHARLOTTE on February 19, 2016 | http://arc.aiaa.org | DOI: 10.2514/1.J053884
