Professional Documents
Culture Documents
A Simulation and Design Tool For A Passive Rotation Flapping Wing Mechanism
A Simulation and Design Tool For A Passive Rotation Flapping Wing Mechanism
2, APRIL 2013
787
I. INTRODUCTION
S in natural flyers, miniature flapping wing vehicles are
expected to present a significant advancement in agility
from their fixed and rotary wing counterparts. Inspiration and
basic concepts of flapping flight are taken from studying the
kinematics of dragonflies [1], butterflies [2], hawkmoths [3],
bats [4], and flies [5]. The low Reynolds number operation
regime of small-scale fliers is dominated by nonlinear aerodynamics and temporal wingwake interactions [6], [7] that are
the underlying cause for their, unaccounted by conventional
aerodynamics, large lift force production capacity [8], responsible for the acrobatic maneuverability of these insects. One
major milestone to overcome in creating a robotic flapping
wing platform capable of controlled hover is the generation
of proper and controlled wing motion. Over time, two different approaches to this problem emerged: 1) active control of
both the flapping and rotation angles, and 2) active control of
788
(1)
Fig. 2. Displacement (left axis) and torque transmission curves (right axis)
for 75-Hz resonance single-flapping-wing system. o u t is the torque output of
the four-bar. The lengths of these four-bars links are {L0, L1, L2, L3} = {13.5,
12.1, 10, 5} mm.
(2)
(3)
ARABAGI et al.: SIMULATION AND DESIGN TOOL FOR A PASSIVE ROTATION FLAPPING WING MECHANISM
789
2
(cos sin)rc(r)
dr
4
c(r)
zRA
c(r)2 dr
4
2
dFair =
Fig. 3. Schematics of the passive flapping wing setup. Coordinate sets represent transformations established by (flapping angle) and (rotation angle).
The coordinate systems are shifted for clarity, while in simulation they are all
centered at the point labeled coordinate sets origin. The E coordinate system
is attached to the wing.
(7)
c(r)
cos
(8)
(4)
(9)
(10)
790
2
2
1
1
+
mL i vL i vL i +
JL i L i L i
2
2
1
1
+ me 2 krot 2
(11)
2
2
where m is the wing mass augmented with the added air mass
[as described in (19)], v is the velocity at the wings center
of gravity (CG), J is the wings inertia matrix taken about the
wings center of mass, krot is the torsional spring stiffness at the
rotation axis, mL i , JL i , vL i , and L i are the masses, rotational
inertias, linear and angular velocities of the four-bar links 13,
respectively, me is the effective mass of the actuator, is the
horizontal velocity of the tip of the actuator (scalar), and
is
the angular velocity of the wing defined as
L=T V =
= E
+ sin()E
+ cos()E
+ E
= E
3
1
1
2
3
(12)
expressed in terms of basis vectors of the coordinate frame attached to the wing. Furthermore, the expressions for angular
and linear velocity of the four-bar links are complex in closed
form and, hence, have been substituted by numerically computed trajectories parametrized by the output flapping angle
f
2 (t)
= (t)
(i.e., v2 (t) = f ((t)) and hence dvdt
(t) ). Defining the conventions f f and f df , and substituting
(t)
dt
+ (cos()J
(cos()J
xz +Jxx )+cos()
z z + Jxz )
2
+ 1/2m((2RCG
+ 2 cos(2))2 + 4RCG cos()
+ 2 2 2 ) +
3
mL i 2 xL i 2 +
i=1
2
JL i 2 L2i
i=1
+ JL 3 + me )
(13)
where RCG and CG are the radial and vertical positions, respectively, of the center of mass from the origin of the coordinate
systems, as portrayed in Fig. 4, Jii are the components of the
wings inertia matrix, xL i is the CG position of link i of the
four-bar, and L i is the orientation of the links (note that the
four-bar motion is planar and equations are simplified to reflect
2-D motion). Note that for a thin wing, the inertial components
Jy z and Jxy are extremely small and have been omitted from
the equations. Thus, the Lagrange equations describing the wing
flapping and rotation dynamics are as follows:
d L
L
aero E
d
=M
(14)
1
dt
d L
L
drive + M
aero ) E
3 D (15)
= (M
dt
where d is the rotational damping coefficient, D is the flapping damping coefficient arising from the transmission flexure
drive is the driving flapping torque, and M
aero is the
damping, M
span z dFrot
|rot | =
(18)
span dFrot
where t defines the position of the translational lift center of
pressure as a function of angle of attack for each wing chord
strip c(r) and is obtained from experimental results of Dickson
et al. [28]. rot is the effective moment arm of the rotational
force distribution of Fig. 4(b).
Given that there is no analytical solution for the added air
mass force on a wing planform moving in 3-D fashion, dFair
of (7) does not accurately describe the moments exerted by
the added air mass on the flapping and rotation angles. As an
approximation, the effect of added air mass forces was implemented in the form of virtual mass that augmented the physical
mass of the wing, similar to the inertial implementation in [26]
and [29]. Stemming from the concept that the added mass effect
can be estimated as dm = 4 c(r)2 dr [30], the expression for
m emerges as follows:
c(r)2 dr
m = mw +
(19)
span 4
where mw is the physical mass of the wing. This approximation
allows a simplified treatment of the added air mass effects that
is sufficient for the development of a simulation tool able to
predict general dynamics and lift force trends. Although this
mass augmentation approach is used in the differential equations
defining the system dynamics, the expression for added air mass
force of (7) is used to estimate the aerodynamic lift generated
by this effect in all lift plots presented in this paper.
The previously presented Lagrange equations (14) and (15)
are fully general, governing the dynamics of a passive rotation
drive around the flapping
flapping wing driven by a torque M
axis. In our case of the driving mechanism consisting of a piezo drive is
electric actuator coupled to a four-bar transmission, M
given by (3). Thus, substituting all the known quantities into (14)
and (15) and simplifying, we obtain the nonlinear differential
equations governing the flapping and rotation angles that can be
found in the Appendix, due to their length and complexity.
III. MANUFACTURING AND EXPERIMENTAL
PARAMETER DETERMINATION
In order to best quantify the theoretical simulations performance, two experimental systems consisting of an actuator, a
four-bar and wing were manufactured, having different resonant frequencies of 30 and 75 Hz. Both systems feature the
ARABAGI et al.: SIMULATION AND DESIGN TOOL FOR A PASSIVE ROTATION FLAPPING WING MECHANISM
791
Fig. 5. Overall dimensions of the manufactured wing and actuator for the
(a) 30-Hz resonant system and (b) 75-Hz system. The actuator mount is 3-D
printed from VisiJet HR 200 material, and assembled with Cyanoacrylate glue.
Given that the slider crank and four-bar flexures are sources
of unwanted compliance, the stiffness and damping properties
of the transmission are measured experimentally. Compliance
of the slider crank is measured via an indentation test at the tip
of the first transmission link, with the actuator rigidly fixed. The
recorded tip stiffness is transformed via the transmission ratio
to stiffness of the slider crank and modeled in simulation in
series with the actuator tip stiffness kact . Throughout our experiments, slider crank stiffness ranges as low as ksc = 300 N/m
and as high as ksc = 1650 N/m. The large span of stiffness values is attributed to the easy misalignment of the slider crank
links, resulting in flexure buckling. For all simulations in this
paper, the 30-Hz resonance flapping system was simulated with
a nominal slider crank stiffness of ksc = 1400 N/m, while the
75-Hz system had a value of ksc = 300 N/m, per experimental
measurements.
The damping coefficient of the four-bar was measured by imparting an impulse on wing-driving mechanism, with the wing
rotated 90 , i.e., = 0 , so as to eliminate aerodynamic drag
from the measurements. The simulated impulse response was
then matched to experimentally recorded vibrations via altering
the flapping damping parameter D from (15), converging to a
final value of D = 100 Nmms, employed in simulation.
IV. EXPERIMENTAL SETUP AND WING KINEMATICS
Experimental lift was measured by mounting the single-wing
assembly to a load cell through a mechanical amplification linkage to increase the sensed lift forces, and the low-friction, singledegree-of-freedom (DOF) pivot isolates the vertical forces generated by the wing [see Fig. 6(b)]. A 25-g load cell (Transducer
Techniques) for the 75-Hz resonant prototype (30-g load cell
for the 30-Hz system) was attached to the main shaft via a
contact connection, such that no torques in the shaft would be
transmitted into the load cell. The actuator, wing, and four-bar
mechanism were mounted directly to a 3-D printed (Invision
HR 3D Printer) plastic mount that was bolted onto an aluminum
shaft, portrayed in Fig. 6(a). Given the large inertia of the balance system, its resonant frequency was determined at 13 Hz.
This is sufficiently distant from the operating frequencies of
30 and 75 Hz to ensure lift measurements were unexaggerated;
however, functioning as a low-pass filter, measurements are restricted to mean lift force only.
During each test, the input voltage to the actuator, a sinusoidal
wave specified by the peak-to-peak voltage Vpp and a constant
dc offset Vdc . Wing kinematics were captured using a strobe light
and a high-definition camcorder (Sony HDR-SR11). Video was
taken from the top view of the wing, with the camera placed
792
Fig. 6. Photographs of the experimental lift acquisition setup with main components labeled. (a) Actuator-four-bar-wing experimental prototype mounted
onto a 3-D printed plastic part. (b) Lift balance mechanical amplification
device.
ARABAGI et al.: SIMULATION AND DESIGN TOOL FOR A PASSIVE ROTATION FLAPPING WING MECHANISM
793
Fig. 9. Frequency sweep obtained with the complete system simulation, including the actuator drive and wing dynamics. System parameters: V p p = 180 V,
V d c = 30 V, k sc = 1400 N/m, k ro t = 20 mNmm, and d = 10 Nmms. Curves
labeled with a slider crank stiffness are obtained with simulated systems having
critical values k sc = 300 and 1650 N/m.
Fig. 8. Frequency sweep of the 30-Hz resonance system simulating only wing
rotation, with flapping angle set by a sinusoid wave with experimentally obtained
amplitudes. (a) Experimental and simulated wing kinematics, and (b) lift forces.
System parameters: V p p = 180 V, V d c = 30 V, k ro t = 20 mNmm, and d =
10 Nmms.
Fig. 10. Frequency sweep of the 75-Hz resonant system simulating only wing
rotation, with flapping angle set by a sinusoid wave with an experimentally
obtained amplitude. (a) Experimental and simulated wing kinematics, and
(b) lift forces. System parameters: V p p = 150 V, V d c = 50 V, k ro t =
37.5 mNmm, and d = 13 Nmms.
794
Fig. 12. Frequency spectrum of the system impulse response. The motion of
the first four-bar link was recorded after an impulse has been applied. The first
and second system resonant peaks are indicated by arrows. Note that the 50and 100-Hz peaks of the experimental system are due to electrical noise in the
environment and do not reflect physical resonant frequencies.
Fig. 11. Frequency sweep obtained with the complete simulation. System
parameters: V p p = 150 V, V d c = 50 V, k sc = 300 N/m, k ro t = 37.5 mNmm,
and d = 13 Nmms. The lift bounding curve corresponding to a slider crank
stiffness of k sc = 300 N/m coincides with the nominal frequency response,
while the other bounding curve corresponds to k sc = 1650 N/m.
The presented comparison between experimental measurements and both wing rotation only and the complete system
simulation effectively illustrates the capabilities and limitations
of the theoretical model of the single-flapping-wing system.
The overall good agreement to experimental results of both
kinematics and lift forces produced by the simulation of wing
rotation only suggests that wing inertial properties, flexure stiffness, and damping properties, as well as the rough, quasi-steady
aerodynamic models are largely correct. The complete simulation, encompassing the actuator drive, yields less acceptable
results, overpredicting lift force, especially for the 75-Hz resonant frequency system. Given that most of this discrepancy is
attributed to the underprediction of flapping amplitude and that
wing rotation dynamics are correct, as established earlier, the
underlying cause of the problem is traced to the flapping motion of the wing. The decrease in wing flapping displacement
could arise from actuator depolarization, inconsistencies in the
manual alignment and assembly of the four-bar transmission, as
well as from spanwise aerodynamic moments not encompassed
by quasi-static 2-D approximations.
One of the limitations of the developed theoretical simulation
is its inability to model accurately the behavior of the system for
frequencies higher than the first resonance. These discrepancies
are thought to be products of nonlinear aerodynamic effects
caused by large stroke and rotation amplitudes. As a test to
this hypothesis, the 75-Hz resonant systems impulse response
is recorded, featuring small flapping and rotation stroke amplitudes (<5 ) and thus rendering the aforementioned dynamic
effects not important. Comparing the frequency components
of the experimental and simulated impulse responses obtained
with the fast Fourier transform algorithm, shown in Fig. 12, we
observe overlapping peaks at 78 and 129 Hz, suggesting that
the inertial, force transmission, and stiffness properties of system components were modeled adequately in the numerical
ARABAGI et al.: SIMULATION AND DESIGN TOOL FOR A PASSIVE ROTATION FLAPPING WING MECHANISM
simulation. Furthermore, the second resonance peak is decreased from 129 to 95 Hz in actuated flapping experiments,
suggesting a large effect of added air mass and potentially other
unmodeled forces on wing dynamics. The two resonance peaks
should not be viewed as the flapping and rotation resonances
of the wing, but rather as different modes of vibration of the
system. Given that these resonances are reflected as peaks in
flapping amplitude, they portray an optimally balanced state
between internal actuator stresses and aerodynamic forces imparted on the wing. The first resonance achieves this balance
via approximately symmetric wing rotation, with wing pitch
reversal occurring roughly at the time of stroke reversal (the
exact timing of rotation is a complex function of many different parameters), thus resulting in a considerable amount of lift
force. The second resonant peak maximizes flapping amplitude
because the drag on the wing significantly decreases, thus, allowing larger deflections of the actuator. The reduction of drag
occurs because the wing begins to rotate out of phase with flapping, such that the flapping and rotation curves of Fig. 7 are 180
offset in phase. This onset of second resonance, when the center
of mass of the wing performs very little translational motion, is
the cause for the sharp decrease in experimental lift in the two
systems shortly after the first resonant frequency. The reason
for the simulations overprediction of the second resonant frequency is attributed to unmodeled transient aerodynamic forces,
such as wake capture [6], [25]. These dynamic effects generated
by the wings past trajectory create impulse-like loading on it
during each stroke, thus, exciting the second resonance mode
much earlier than predicted.
It is generally desirable to engineer the system such as to space
the two resonance modes far apart in the frequency domain.
One may attempt to do so by utilizing a stiff wing rotational
joint, such as not to excite the second mode; however, this is
suboptimal from the performance perspective, since the wing
will rotate insufficiently in its first resonance mode, which is the
main operating point. Furthermore, stiffening either the wing
rotation flexure or the actuator has the effect of shifting both
peaks of Fig. 12 to higher frequencies, as well as increasing
their separation. The proximity of the two resonance peaks in
an optimized lift system is inherent in the passive wing rotation
system at hand, where the inertia of the wing is significant and
plays a major role in pitch reversal. By reducing the overall
inertia/mass of the wing and employing a correspondingly stiff
wing rotation joint, it is possible to increase the frequency of
the second resonance much higher, thus, reducing interference
and broadening the frequency range of maximum lift.
Finally, although the model generally overpredicts total lift
and overestimates the systems second resonance frequency,
its usefulness lies in capturing trends in wing kinematics as a
function of frequency, amplitude, spring stiffness, etc. Thus,
the model can be used as a design tool to obtain dimensions
and configuration of the actuator, four-bar transmission, and
wing that would ensure wing lift is maximized at a particular
frequency. The first resonance peak is predicted accurately in
simulation, but, since it is sensitive to slider crank stiffness,
it should be designed in accordance with the bounding values
of that parameter. Although the predicted magnitude of max-
795
796
Fig. 13. Simulated and experimental system response to a sweep in the input
waveform amplitude. Simulation parameters: V d c = 50 V, frequency = 70 Hz,
k ro t = 50 mNmm, and d = 5 Nmms. Experimental lift data are averaged
over three datapoints.
Fig. 14. Simulated system response to a sweep in the rotational spring stiffness. Simulation parameters: V p p = 150 V, V d c = 50 V and frequency = 70 Hz,
and d = 5 Nmms. Experimental lift data are averaged over three datapoints.
affecting the lift force. The simulated system response to changing peak-to-peak driving amplitude, overlayed with experimental data, is portrayed in Fig. 13. A good correlation between
experimental and simulated data is observed, although simulated lift is consistently higher than in experiment. The linear
behavior of the overall lift force and large span of produced
forces renders this control scenario very promising. Intuitively,
an increase in the lift force could always be achieved by increasing the flapping amplitude; yet, this approach is limited by
electrical depoling and stress induced crack propagation of the
piezoelectric actuator.
The third envisioned method of controlling lift is modulating
the stiffness of the rotational flexure that suspends the wing.
Unlike the high-bandwidth control architecture required for the
Micromechanical Flying Insect [34], the inherently stable na-
Fig. 15. Experimental lift surface. Represents variation of produced lift with
input voltage and torsional spring stiffness. Experimental parameters: V d c =
50 V, frequency = 70 Hz, and d = 5 Nmms.
ARABAGI et al.: SIMULATION AND DESIGN TOOL FOR A PASSIVE ROTATION FLAPPING WING MECHANISM
VII. CONCLUSION
It is of great importance for any design project to have a simulation tool to predict the performance of the final system. Given
that our goal is the development of a miniature, controlled, flapping flight robot, a full simulation of its subsystema singlewing designcould greatly speed up the design process, as
well as enable optimization of its components. The numerical simulation presented in this paper includes a model of the
bending piezoelectric actuator, transmission kinematics, and the
passively rotating wing aerodynamics. Two experimental prototypes featuring different resonance frequencies were manufactured from carbon fiber composite material, enabling the
comparison of simulation to experimental results. Overall, there
was broad agreement between the flapping and rotation trajectory trends between the numerical simulation and experiment.
The major discrepancy between the measured and predicted lift
forces is attributed to manufacturing/material imperfections of
the actuator and/or four-bar transmission, and unmodeled aerodynamic forces. However, the model is able to predict the main
resonance peak accurately along with general trends of wing
kinematics and lift behavior as functions of frequency. Thus,
given its limitations in predicting the magnitude of produced
lift force, the simulation can be used as a design tool for optimization of wing shape, actuator and four-bar geometry, as well
as a future dual-wing flapping robotic platform. Although the
presented experimental prototype features a lift-to-weight ratio
of 1/6 (measured on a robot employing the same construction), scaled down variants of this design have demonstrated
liftoff [14]. Thus, given the direct scalability of the SCM manufacturing technique and similarity of Reynolds number, the
presented here theoretical model and analysis is valid and useful for scaled down robotic platforms capable of liftoff.
Compared to other wing rotation ideologies, a completely
passive rotation design benefits from mechanical simplicity,
passive stability of wing trajectory, and low controller bandwidth. However, the high dependence of wing dynamics on
aerodynamic forces renders the system as lacking in robustness
and consistent performance, requiring an adequate controller to
compensate for these design issues. Control over the amplitude
of the actuator driving waveform, as well as, the potential use of
smart materials in the wing rotation joint, is envisioned to enable
indirect control over the trajectory of the wing. The proposed
passive design is by no means seen as a replacement for active
wing control mechanisms, whose ability to generate arbitrary
forces and torques are essential for execution of their tasks, but
merely as an alternative flapping platform that may be realizable
on a shorter time scale for specific applications.
APPENDIX
FULL EQUATIONS OF MOTION FOR THE
PASSIVE ROTATION WING
Given the definition for the Lagrangian in (13) and performing the differentiations in (14) and (15), the full equations of
motion of wing motion are obtained. The equation governing
the rotation angle is the simplest of the two as it does not
797
E
CG + Jxx ) + krot = Maero E1 d
1
(20)
where all the constant definitions are presented in the main body
of this paper.
The equation governing the motion of is more complex as it
involves numerous inertial terms and parameterizations in terms
of . We need to remind ourselves that according to the notation
discussed in the text, we have the following:
f
f
,
(t)
f
2 f
,
(t)2
df
f ,
dt
d2 f
f 2 .
dt
(21)
2
JL i L2i + JL 3
+ cos()(mR
CG CG + Jxz )
i=1
2
+ Jy y Jz z ) +
3
i=1
2
JL i L i L i 2 + me 2
i=1
drive + M
aero ) E
3 D.
= (M
(22)
The coupled equations for (t) and (t) are solved numerically with MATLABs ode15s solver. The obtained results are
presented in the main body of the paper.
ACKNOWLEDGMENT
The authors would like to thank R. Smith for his work on the
mean lift measuring setup, as well as prototype manufacturing
and testing.
REFERENCES
[1] A. Bergou, S. Xu, and Z. Wang, Passive wing pitch reversal in insect
flight, J. Fluid Mech., vol. 591, pp. 321337, 2007.
[2] R. B. Srygley, A. L. R. Thomas, Unconventional lift-generating mechanism in free-flying butterflies, Nature, vol. 420, pp. 660664, 2002.
[3] T. L. Hedrick and T. L. Daniel, Flight control in the hawkmoth manduca
sexta: The inverse problem of hovering, J. Exp. Biol., vol. 209, pp. 3114
3130, 2006.
[4] F. T. Muijres, L. C. Johansson, R. Barfield, M. Wolf, G. R. Spedding, and
A. Hedenstrom, Leading-edge vortex improves lift in slow-flying bats,
Science, vol. 319, pp. 12501253, 2008.
[5] S. Fry, R. Sayaman, and M. H. Dickinson, The aerodynamics of hovering
flight in drosophila, J. Exp. Biol., vol. 208, pp. 23032318, 2005.
798
Lindsey Hines received the B.Sc. degree in mechanical engineering and the B.A. degree in mathematics
from the University of St. Thomas, Saint Paul, MN, in
2008, and the M.Sc in robotics from Carnegie Mellon
University, Pittsburgh, PA, in 2011, where she is currently working toward the Ph.D. degree in robotics.
Her interests include flapping flight and robust
control.
Ms. Hines has been awarded both the National Science Foundation and National Defense Science and
Engineering Graduate Fellowship, and placed first in
the 2011 Graduate Robot Design Competition.