Professional Documents
Culture Documents
System Identification of An Ornithopter Aerodynamics Model: August 2010
System Identification of An Ornithopter Aerodynamics Model: August 2010
net/publication/268556965
CITATIONS READS
13 171
5 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Jared Grauer on 23 January 2021.
Jared Grauer,∗ Evan Ulrich,† James Hubbard Jr.,‡ Darryll Pines,§ and J. Sean Humbert¶
Department of Aerospace Engineering, University of Maryland, College Park, MD, 20742
Nomenclature
1 of 13
2 of 13
which is connected through a gear box and a four-bar linkage to the wings, thereby governing the flapping
frequency and generation of lift and thrust forces. The longitudinal and lateral inputs control the armature
position of servomotors, which pitch and roll the tail relative to the fuselage in a serial manner, creating
pitching and yawing aerodynamic control torques on the aircraft.
Experience conducting flight tests and laboratory experiments with this ornithopter has illuminated
several characteristics of its flight. Computer aided drawing software was used to model the aircraft geometry,
which resulted in a mass distribution where approximately 76% of the vehicle mass was affixed to the fuselage,
19% comprised wings and 5% comprised the tail. Additionally the center of mass migrates approximately
0.1 m over each wing stroke, and the mass distribution experiences dramatic changes in the primary pitching
and yawing moments of inertia. Constrained flapping experiments performed in a laboratory with a visual
tracking system have shown that for this particular ornithopter the bending moments of the leading edge
spar are relatively mild, whereas smaller ornithopters can have a pronounced bending of the wing.16 These
aircraft have a flapping bandwidth up to about 7 Hz, which results in forcing frequencies that are significantly
closer to the expected rigid body vehicle modes than are seen in conventional aircraft. Flight test experiments
using a custom avionics package have shown that this aircraft experiences heave accelerations of 46.12 m/s2
and pitch rates of 5.62 rad/s.17
3 of 13
CI
4 of 13
Ma v̇ + Ca v + Ka p + Gτ = Bu (8)
where Ma describes rotary inertia; Ca describes linear damping arising from back electro-mechanical forces,
viscous friction, and electrical resistance; Ka is the synthetic stiffness injected by a proportional control law
in the servo motors; G is a matrix of gear ratios; τ are the torques applied to the joints; and B is an input
coupling matrix.
Model parameters in (8) were estimated from the data using the frequency domain equation-error method
and the time domain output-error method.14 Model fits for the DC motor and servo motor are shown for
both methods in Figure 4. In the frequency domain estimation, a frequency grid using intervals of 0.01 Hz
were used on the range where there was appreciable frequency content. The output-error method converged
after 28 iterations for the DC motor, and after 38 iterations for the servo motor. The DC motor results had
model fits of R2 = 0.89 and R2 = 0.98 for the frequency domain equation-error and time domain output-
error analysis, whereas the servo motor results had model first of R2 = 0.96 and R2 = 0.98, respectively.
Estimated model parameters are listed in Table 1 along with standard errors, corrected for colored residuals.
In general there was good agreement between the two methods and overall, the frequency domain equation-
error method produced smaller error bounds. Based on the standard errors of these two methods, a weighted
average was computed and used for the remainder of the work.
The rigid body equations (7) can be substituted for the joint torque in the actuator dynamics (8) to yield
the equation
M(p)v̇ + C(p, v)v + w(p, v) = Bu (9)
where
M(p) = GMb + Ma
C(p, v) = GCb + Ca (10)
w(p, v) = Gg + Ga + Ka p
which couples the rigid body equations with the actuator equations.
5 of 13
Table 1. DC motor and servo motor parameter estimates and standard errors
6 of 13
q̇ (rad/s)
-3000
jω q̃
5000
-4000
0
0 2 4 6 0 5 10 15 20
(a) DC motor
400 1
data
model
q (rad)
200 0
q̃
0 -1
0 4 8 12 0 5 10
Figure 4. Actuator model fits using frequency domain equation-error and time domain output-error
7 of 13
-0.2 0.4
-0.4
0.2
CL CD
-0.6
model model
-0.8 data 0 data
-0.2 -1
-0.4 -0.8
0.8 -0.6 α (rad) -0.8 -0.4 -0.6 α (rad)
0.4 β (rad) 0 -0.8 β (rad) 0 -0.4
-0.4 -1 0.4 -0.2
-0.8 0.8
8 of 13
by algebraic manipulations. The aerodynamic contributions are assumed generated by two wings and a tail
with no interactions. The left hand side of (13) can be subdivided into contributions from the tail, given
by equations (11) and (12), and by the wings. The only remaining unknown term are the aerodynamic
contributions from the wings. By performing a flight test and obtaining state measurements, substitutions
can be made into the right hand side of (13) to yield time histories of the aerodynamic forces and moments
on the ornithopter. It is then a system identification problem to determine the functional dependence of
how those time histories depend on the states of the ornithopter, and the value of any parameters within
the model.
A flight test was performed where the ornithopter was flown indoors while trimmed for straight and level
mean flight. A Vicon visual tracking system,22 consisting of eight cameras, was used to estimate at 500 Hz
the spatial position of several retro-reflective markers placed on the ornithopter. From the marker positions,
position and velocity state vectors were computed by employing a nonlinear least squares estimator to fit
rigid bodies to the marker locations and extracting out the states. Figure 6(a) displays a photograph taken
of the capture volume and Figure 6(b) shows the rigid body estimation results for a single capture frame.
The data from this experiment was substituted into (13) to produce aerodynamic forces on the wings.
Model structure determination, using a modified step-wise regression algorithm,14 was used to determine the
functional dependence of the lift and drag forces on the ornithopter states. In the past orthogonal regressors23
and a quasi-steady aerodynamics model24 have been used to model the lift coefficient. The method of using
orthogonal regressors resulted in a relatively large model structure, and the quasi-steady model resulted in a
model needing acceleration terms, which in turn requires additional states in the dynamics model.25, 26 In the
current study, only the ornithopter state variables were considered as model regressors. Additionally, step-
wise regression was performed in the frequency domain instead of the time domain to explore the spectral
content of the data.
The model structure for lift and drag were chosen as
2 2
CD = CD0 + CDθ̇ θ̇11 + CDθ2 θ11 + CDθ̇2 θ̇11 (15)
11 11 11
which can be found from the ornithopter state vector and requires no additional states in the model. Param-
eter estimates were determined in both the time domain and the frequency domain using the equation-error
method, and are shown with residuals in Figure 7. As the frequency domain method subtracts out the bias
and trend of the data before the estimation, a second time-domain estimation was required to find the lift
and drag bias coefficients. The lift estimation had a fit of R2 = 0.84 in the time domain and R2 = 0.61 in
the frequency domain, whereas the drag estimation had a fit of R2 = 0.88 in the time domain and R2 = 0.64
in the frequency domain. These model fits are more accurate than those previously reported for system
identification results with this ornithopter,23, 24 and significantly better than the analytical models27 and
computational fluid dynamics work12 attempted previously. Estimation model parameters with standard
error bounds corrected for colored residuals are displayed in Table 3. With the exception of the coefficient
CLνx , the two methods provide consistent estimates of the model parameters. Generally it is the frequency
domain method that produced the estimates with lower standard errors.
V. Conclusions
Ornithopters are flapping-wing vehicle which have the potential to not only accomplish standard missions
for unmanned aircraft such as way-point navigation, but also multi-mission capabilities such as robust
outdoor flight, agile indoor flight, and precision perching at a variety of speeds and sizes. To realize an
autonomous ornithopter that could fulfill this potential, a simple and accurate model for the aerodynamic
forces must be developed to program into autopilots uniquely tailored for flapping-wing flight dynamics. In
9 of 13
-1.1
-1
fuselage
-0.9
right wing
-0.8 left wing
tail
3.6
3.8 -0.6
4 -0.3
4.2 0
0.3
4.4
0.6
10 of 13
model
0.07
0
-1 0
0.07
1
residual
0
0
-1 -0.07
0 0.2 0.4 0.6 0 10 20 30 40 50
0.4 0.04
data
drag coefficient
model
0
0.02
-0.4
-0.8 0
0.4 0.02
0
residual
0
-0.4
-0.8 -0.02
0 0.2 0.4 0.6 0 10 20 30 40 50
Figure 7. Wing aerodynamic model fits using equation-error in the time and frequency domains
11 of 13
the past this task has been accomplished using first principles modeling, numerical simulations, and time-
averaged wind tunnel data. Each of these methods introduces either drastic simplifications of aerodynamics,
or changes the flow environment experienced by the ornithopter during testing.
In this paper, aerodynamic models are determined using standard system identification techniques applied
to free flight data. Past experience and analysis with the ornithopter has led to the development of a
multiple rigid body model of the flight dynamics. System identification methods were used to identify
models for actuator dynamics in static testing environments, and aerodynamics of the tail in a wind tunnel.
Incorporating these models facilitated the system identification of the wing aerodynamics from a free flight
test. The flight test was conducted indoors using a Vicon visual tracking system to obtain spatial positions
of retro-reflective markers placed on the ornithopter, from which state measurements were extracted. System
identification was then performed to determine the model structure of the lift and drag forces on the wing,
and to estimate the model parameters.
In contrast to previous system identification work with this ornithopter, only the state vector was con-
sidered for model regressors, which led to a simpler model structure and ultimately more accurate results,
particularly for the drag model, than previously reported. Good models fits were obtained and parameter
estimates were generally consistent between the two methods of parameter estimation used. An accurate, low
order model of the wing aerodynamics was presented, suitable for stability analysis and control development
and hardware implementation.
VI. Acknowledgments
This work was funded in part by the Army MAST CTA and the NASA Langley NIA Langley Professor
Program. Flight tests were conducted at the University of Maryland. Conversations with Eugene Morelli
at the NASA Langley Research Center are acknowledged and appreciated. The authors would like to also
acknowledge support from the NASA Langley Research Center, the National Institute of Aerospace, and the
University of Maryland in conducting this research, and would like to thank the members of the Morpheus
Laboratory and the Autonomous Vehicle Laboratory for their continued teamwork and motivation.
References
1 Seshadri, P., Benedict, M., and Chopra, I., “Experimental Investigation of an Insect-based Flapping-Wing Hovering
Micro Air Vehicle,” Aeromechanicanics Specialists’ Conference, American Helicopter Society, 2010.
2 Shyy, W., Lian, Y., Tang, J., Viieru, D., and Liu, H., Aerodynamics of Low Reynolds Number Flyers, Cambridge
namics for Micro Air Vehicle Applications, edited by T. Mueller, Vol. 195 of Progress in Aeronautics and Astronautics, 2001,
pp. 1–9.
4 Larijani, R. and DeLaurier, J., “A Nonlinear Aeroelastic Model for the Study of Flapping Wing Flight,” Fixed and
Flapping Wing Aerodynamics for Micro Air Vehicle Applications, edited by T. Mueller, Vol. 195 of Progress in Aeronautics
and Astronautics, 2001, pp. 399–428.
12 of 13
2009-6158, 2009.
10 Dietl, J. and Garcia, E., “Stability in Ornithopter Longitudinal Flight Dynamics,” Journal of Guidance, Control, and
Wing Ornithopter Flight,” Journal of Aircraft, Vol. 46, No. 6, Dec 2009, pp. 2016–2031.
13 Sane, S. and Dickinson, M., “The Aerodynamic Effects of Wing Rotation and a Revised Quasi-Steady Model of Flapping
Flight,” Journal of Experimental Biology, Vol. 205, No. 8, April 2002, pp. 1087–1096.
14 Klein, V. and Morelli, E., Aircraft System Idenfication: Theory and Practice, AIAA Education Series, AIAA, 2006.
15 Kinkade, S., “Hobby Technik,” www.flappingflight.com, 2008.
16 Harmon, R., Aerodynamic Modeling of a Flapping Membrane Using Motion Tracking Experiments, Master’s thesis,
Guidance, Control, and Dynamics, Vol. 32, No. 1, Jan.-Feb. 2009, pp. 326–331.
18 Slotine, J. and Li, W., Applied Nonlinear Control, Prentice Hall, 1991.
19 Ortega, R. and Loria, A., Passivity-Based Control of Euler-Lagrange Systems, Springer, 1998.
20 Grauer, J. and Hubbard, J., “Identification and Integration of Ornithopter Actuator Models,” No. AIAA-2009-5937,
2009.
21 Grauer, J. and Hubbard, J., “Experimental Determination of Ornithopter Control Derivatives,” Region 1 Mid-Atlantic
2005.
27 Harmon, R., Grauer, J., Hubbard, J., and Humbert, S., “Experimental Determination of Ornithopter Membrane Wing
Shapes Used for Simple Aerodynamic Modeling,” No. AIAA Paper 2008-6397, August 2008.
13 of 13