Harveyca 1
Harveyca 1
Harveyca 1
by
Christina Harvey
Doctoral Committee:
Professor Daniel J. Inman, Chair
Professor Douglas L. Altshuler, University of British Columbia
Professor Carlos E.S. Cesnik
Professor Eric Johnsen
Christina Harvey
harveyca@umich.edu
ORCID iD: 0000-0002-2830-0844
First and foremost, thank you Dr. Inman. Coming to Michigan to work with you is the
reason I was able to continue to develop my passion and propelled me to pursue an academic
position. I am very grateful that you’ve always given me the freedom and support to follow the
science where it led. This thesis never would have happened without the AFOSR BRI “Avian-
Inspired Multifunctional Morphing Vehicles” monitored by Dr. B.L. Lee. That grant was the
reason that I met Dr. Inman and has supported both my masters and PhD studies. Thank you
Doug for your support and advice throughout my career and for taking the chance on hiring an
engineer into a zoology program in the first place.
Although never an official advisor, Vikram you taught me so much about, well, many
things, from evolution to statistics to the fact that I use too many commas (working on it).
Jasmin, thank you for your friendship and willingness to chat over a beer. I am looking for-
ward to many more collaborations with you both. Thank you to everyone in AIMS Lab, past
and present. In particular, thank you Kevin and Piper, you have made this journey a blast.
Shout out to all of my incoming cohort including Leanne, Austin, and Thomas. I would go
through prelims again if it meant getting to regularly spend more time with you all. Thank you
Jana (and Parker and Darcy) for your support and friendship. Thank you to the members of the
graduate and departmental diversity, equity, and inclusion committees. Working with you all
has brought me a greater sense of purpose and joy.
Thank you Mom, Dad, Grandpa, Sam, and James for supporting me, celebrating my achieve-
ments, and checking in on me when I need it most. Finally, to Josh. You are my best friend and
continually bring so much happiness into my life. I am forever grateful that I get to go through
life with you as my partner and I look forward to our many future adventures together.
ii
This work is supported in part by the US Air Force Office of Scientific Research (AFOSR)
under grant number FA9550-16-1-0087, titled “Avian-Inspired Multifunctional Morphing Ve-
hicles” and grant number FA9550-21-1-0325, titled “Towards Neural Control for Fly-by-Feel
Morphing” monitored by Dr. B.L. Lee and in part by the National Science Foundation (NSF)
under grant number 1935216. My studies were further supported by personal fellowships and
scholarships from the National Science and Engineering Research Council of Canada (NSERC
PGS-D, 2018-2021), the Francois-Xavier Bagnoud Fellowship awarded by FXB International
through the University of Michigan Department of Aerospace Engineering (2019-2022) as well
as the Zonta International Amelia Earhart Fellowship (2021-2022).
iii
Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Chapter 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 The rise of the uncrewed aerial vehicle . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Deriving inspiration from birds . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Defining comparable regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Why not discuss efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 How birds morph their wings . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Maneuverability vs. stability . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.7 Dissertation outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
iv
2.3.2 Airfoil properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.3 Numerical lifting-line solution (MachUpX) . . . . . . . . . . . . . . . 26
2.3.4 Wind tunnel study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1 MachUpX validation . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.2 Lift force and pitching moment production . . . . . . . . . . . . . . . . 31
2.4.3 Longitudinal static stability . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.4 Longitudinal balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.1 The role of joint extension trajectories . . . . . . . . . . . . . . . . . . 37
2.5.2 Comparison to existing UAV . . . . . . . . . . . . . . . . . . . . . . . 40
2.5.3 Open loop vs. closed loop . . . . . . . . . . . . . . . . . . . . . . . . 41
2.6 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
v
3.6 Results - Aerodynamic characteristics . . . . . . . . . . . . . . . . . . . . . . 63
3.7 Investigation of the evolutionary implications . . . . . . . . . . . . . . . . . . 64
3.7.1 Evolutionary modeling results . . . . . . . . . . . . . . . . . . . . . . 64
3.7.2 Power analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.7.3 Sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.7.4 Sensitivity of the neutral point approximation . . . . . . . . . . . . . . 69
3.7.5 Effects of including a furled tail . . . . . . . . . . . . . . . . . . . . . 70
3.8 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
vi
Chapter 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.1 Key results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.1.1 The elbow and wrist allow adaptive control . . . . . . . . . . . . . . . 106
5.1.2 Birds have the capacity to shift stability states . . . . . . . . . . . . . . 107
5.1.3 Wing joints control the dynamic response . . . . . . . . . . . . . . . . 109
5.2 Next steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.2.1 Translating basic science to applied design . . . . . . . . . . . . . . . . 110
5.2.2 Relaxing flexibility constraints . . . . . . . . . . . . . . . . . . . . . . 111
5.2.3 Incorporating flapping characteristics . . . . . . . . . . . . . . . . . . 111
5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
vii
A.4.3 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.5 Head, neck, legs and tail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
A.5.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
A.5.2 Required measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 123
A.5.3 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
A.6 Torso . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
A.6.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
A.6.2 Required measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 125
A.6.3 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
A.7 Base moment of inertia tensors . . . . . . . . . . . . . . . . . . . . . . . . . . 126
A.8 Base center of gravity vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
A.9 Illustrations of the required measurements . . . . . . . . . . . . . . . . . . . . 129
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
viii
List of Tables
ix
List of Figures
2.1 Gull wings inspired our analyses of how avian joint-driven wing morphing af-
fects longitudinal aerodynamic control, stability and balance. . . . . . . . . . . 23
2.2 MachUpX effectively predicted the aerodynamic forces and moments for the
investigated wing shapes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 All final 3D printed wings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Wings installed in the wind tunnel. . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Joint-driven wing morphing provides a reliable method to control the lift and
pitching moment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.6 Representative example of the relationship between lift and pitching moment
for a statically stable wing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.7 Joint-driven wing morphing provides a reliable method to control static stabil-
ity characteristics, but not the trim state. . . . . . . . . . . . . . . . . . . . . . 36
2.8 Constant joint extension trajectories allow variable longitudinal control strategies. 39
x
3.2 Elbow and wrist joint range of motion informed our analysis. . . . . . . . . . . 49
3.3 Our results agreed with previous estimates of the wing’s moment of inertia. . . 52
3.4 A bird’s center of gravity is minimally affected by elbow and wrist flexion and
extension. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.5 Wing morphing, specifically driven by the elbow, has a strong effect on roll and
yaw inertia components. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.6 Modern birds exhibit highly variable pitch agility characteristics. . . . . . . . . 63
3.7 The elbow and wrist angle configurations that yielded the maximum and mini-
mum static margin for each species. . . . . . . . . . . . . . . . . . . . . . . . 64
3.8 Evolution selects for both pitch stability and instability, but modern birds ex-
hibit highly variable pitch agility and stability characteristics. . . . . . . . . . . 65
3.9 A power analysis confirmed the validity of Ornstein Uhlenbeck (OU) models
for three key traits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.10 Conservative measurement sensitivity analysis revealed a minimal effect on
pitch stability and agility metrics. . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.11 Bootstrapping our results within the conservative center of gravity measure-
ment error supported our results. . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.12 Tail volume coefficients as a function of the body mass for the investigated
specimens. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.1 In the longitudinal plane, gliding flight dynamics are usually dominated by the
short period and phugoid modes. . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.2 Gulls can morph their wings to adjust key flight parameters. . . . . . . . . . . . 77
4.3 Wing morphing allows gulls to switch between statically stable and unstable
configurations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4 Root locus plot of the open-loop system. . . . . . . . . . . . . . . . . . . . . . 96
4.5 The short period and phugoid mode characteristics are significantly affected by
the wing positioning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.6 Short period and phugoid modes were identified from the magnitude of the
eigenvectors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
xi
4.7 Most gull configurations satisfied Level 2 short period requirements for human
pilots per adjusted UAV guidelines. . . . . . . . . . . . . . . . . . . . . . . . . 100
4.8 Gull wrist gains allows substantial control over gust response. . . . . . . . . . . 102
A.1 Dorsal view of entire bird modeled as a composite of simplified geometric shapes.114
A.2 Simplified bone diagram including the referenced frames of reference. . . . . . 116
A.3 Simplified feather diagram including the referenced frames of reference. . . . . 119
A.4 Entire bird measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
A.5 Wing specific measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
A.6 Body specific measurements. . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
xii
List of Symbols
xiii
ycen y-axis position of the wing’s centroid m
yCG y-axis position of the center of gravity m
z Position along the z-axis in the body frame m
zc Position along the z-axis in the stability frame m
zCG z-axis position of the center of gravity m
A System matrix varies
Ai Coefficients from linear model fits varies
B Input coupling matrix varies
Bi Coefficients from linear model fits varies
CD Drag coefficient –
CDmorph Drag coefficient relative to max. wing characteristics –
CL Lift coefficient –
CLmorph Lift coefficient relative to max. wing characteristics –
CLαt Lift-slope of the tail –
CLα Lift-slope of the wing-body configuration –
CM Pitch coefficient –
CMmorph Pitch coefficient relative to max. wing characteristics –
D Drag N
Di Coefficients from linear model fits varies
Ei Coefficients from linear model fits varies
F Force vector N
Gi Coefficients from linear model fits varies
I Moment of inertia tensor kg · m−2
Ixx Moment of inertia component about the x-axis kg · m−2
Ixz Moment of inertia component in the x-z plane kg · m−2
Iyy Moment of inertia component about the y-axis kg · m−2
Izz Moment of inertia component about the z-axis kg · m−2
L Lift N
M Pitching moment N ·m
M Moment vector N ·m
S Characteristic area m2
St Tail area m2
Sw Wing area m2
Smax Maximum wing area for all morphed configurations m2
U x-axis velocity component m · s−1
U∞ Freestream scalar velocity m · s−1
xiv
U Input vector varies
V y-axis velocity component m · s−1
W z-axis velocity component m · s−1
VH Tail volume coefficient –
W Weight N
X State vector varies
α Angle of attack rad or ◦
αOU Selection strength of the OU model –
β Sideslip angle rad
γ Flight path angle rad
Downwash angle due to the wing rad
ζ Damping ratio –
η Effect size component (i.e., partial η 2 is the effect size) –
ηt Dynamic pressure ratio at the horizontal tail –
θ Pitch angle in the body axis rad
θCG Phenotypic optimum of the center of gravity varies
θSM Phenotypic optimum of the static margin varies
λ Eigenvalue varies
ν Inverse Wishart scalar degrees of freedom –
µ Dynamic viscosity kg·m−1 ·s−1
ρ Fluid density kg · m−3
σ Variance varies
φ Roll angle in the body axis rad
ψ Yaw angle in the body axis rad
ω Frequency rad/s
ω Angular velocity vector rad/s
◦
Γ Sweep angle at the shoulder joint
◦
Λ Dihedral angle at the shoulder joint
*Note: Overdot signifies a time derivative of the variable. Subscript of 0 indicates the trim condition.
xv
List of Abbreviations
AIC Akaike Information Criterion
AICc Akaike Information Criterion with correction for small sample sizes
BM Brownian Motion
CG Center of Gravity
OU Ornstein Uhlenbeck
MAC Mean Aerodynamic Center
MCMC Markov-Chain Monte Carlo
NP Neutral Point
PGLMM Phylogenetic Generalized Linear Mixed Models
ROM Range Of Motion
UAV Uncrewed Aerial Vehicle
VRP Vehicle Reference Point
3D Three (3) Dimensional
xvi
Abstract
Uncrewed aerial vehicle (UAV) design has advanced substantially over the past century;
however, there are still scenarios where birds outperform UAVs. Birds regularly maneuver
through cluttered environments or adapt to sudden changes in flight conditions, tasks that chal-
lenge even the most advanced UAVs. Thus, there remains a gap in our general knowledge of
flight maneuverability and adaptability that can be filled by improving our understanding of
how birds achieve these desirable flight characteristics. Although maneuverability is difficult
to quantify, one approach is to leverage an expected trade-off between stability and maneu-
verability, wherein a stable flyer must generate larger moments to maneuver than an unstable
flyer. Bird’s stability, and adaptability, has previously been associated with their ability to
morph their wing shape in flight. Birds morph their wings by actuating their musculoskeletal
system, including the shoulder, elbow and wrist joints. Thus, to take an important step towards
deciphering avian flight stability and adaptability, I investigated how the manipulating avian
wing joints affect longitudinal stability and control characteristics.
First, I used an open-source low fidelity model to calculate the lift and pitching moment of
a gull wing and body across the full range of flexion and extension of the elbow and wrist. To
validate the model, I measured the forces and moments on nine 3D printed equivalent wing-
body models mounted in a wind tunnel. With the validated numerical results, I identified that
extending the wing using different combinations of elbow and wrist angles would provide a
method for adaptive control of loads and static stability. However, I also found that gulls were
unable to trim for the tested shoulder angle.
Next, I developed an open-source, mechanics-based method (AvInertia) to calculate the
inertial characteristics of 22 bird species across the full range of flexion and extension of the
elbow and wrist. This method allowed a detailed investigation of how manipulating the elbow
and wrist angle changed the center of gravity and moment of inertia tensor. Leveraging the pre-
vious aerodynamic results, I developed a method to estimate the neutral point of any bird wing
xvii
configuration and derived a novel metric for pitch agility. With the neutral point and center of
gravity, I found that the majority of investigated species had the ability to shift between stable
and unstable flight. Further, I implemented an evolutionary analysis that revealed evidence of
evolutionary pressures maintaining this capacity to shift, which transforms our understanding
of avian flight evolution.
Finally, I combined the aerodynamic and inertial results to investigate the dynamic stability
of a gull across a range of shoulder, elbow, and wrist angles. This analysis revealed that a
positive dihedral and forward-swept wing allowed a trimmed flight condition. For trimmed
configurations, I found that high wrist angles were statically unstable and exhibited a non-
oscillating, divergent response to disturbances. Lower wrist angles were both statically and
dynamically stable and exhibited a short period and phugoid mode like traditional aircraft. I
found that most trimmed configurations exhibited short period characteristics that would be
flyable by a human pilot, although with a heavily damped phugoid mode.
In summary, I found that the avian elbow and wrist joints can act as adaptive controls
and permit birds to shift between stable and unstable flight. Identifying these characteristics
provides a starting point for future UAV designs that hope to incorporate avian-like maneuver-
ability and adaptability.
xviii
Chapter 1
Introduction
I saw a billboard with a bird sitting on the rim of a nest, nurturing her young
fledglings into the flying world. It read, “Birds Learn to Fly. Why Can’t You?”
That did it.
– Janet Harmon Bragg
1
Since these early days, UAV designs have evolved drastically, ranging across multiple or-
ders of magnitude in scale and weight. Designs have also branched out from a conventional
fixed-wing design into rotorcraft and aircraft that morph the shape of their wings in flight [6].
In addition to their “morphological” diversity, modern UAVs are also no longer solely used
for warfare or combat [3, 7], having become indispensable tools for many different missions.
Modern uses for UAVs include environmental and climate monitoring [8, 9], assessing in-
frastructure [10, 11], agriculture [12, 13], humanitarian missons such as disaster management
[14, 15] and public health support [16, 17], and possibly package delivery [18].
Despite the extraordinary advancements in the field, UAVs still exhibit a few key weak-
nesses. UAVs are often constrained by their range, endurance, and/or payload capacity. Fur-
ther, UAVs struggle to operate effectively between distinct missions or disparate flight con-
ditions. For example, a surveillance drone is not effective for maneuvering through cluttered
environments and maneuverable quadcopters can be destabilized by gusty conditions. Thus,
it would be beneficial for a UAV to be able to adapt to different missions and to sudden flight
disturbances. It is this challenge, the concept of adaptability, that is a focus for this thesis.
UAV adaptability is limited, in part, because of an expected trade-off between aircraft ma-
neuverability and efficiency. Since maneuverability can encompass a broad range of attributes,
for the purposes of this work, I define maneuverability as the ability to change the velocity
vector direction and/or magnitude [19, 20]. Conventionally, a surveillance drone with longer
wings is more aerodynamically efficient and can cover a larger range than smaller designs.
However, aircraft with large wingspans also have a larger moment of inertia and mass, which
reduce their capacity for translational and rotational accelerations. This trade-off is one of the
reasons that there is such a diversity of aircraft designs, as each aircraft needs to be inten-
tionally designed to satisfy its particular mission constraints and goals. Therefore, with the
intention of advancing adaptability and maneuverability in mind, we can now look for novel
ways to improve and advance the design of adaptable and maneuverable UAVs.
2
1.2 Deriving inspiration from birds
There are over 10,000 species of birds [21] with body masses spanning four-orders of mag-
nitude, from 0.002 kg to 20 kg. Avian morphology is extremely diverse with much variation
in the size and shape of wings, tails, and bodies. Understanding avian diversity as well as the
evolutionary pressures that have led to modern species is a major driver for some ornithological
studies. To survive and procreate, different species rely on different strategies that may, or may
not, require specific flight characteristics for success. In fact, flightlessness has independently
evolved across and within multiple lineages of birds [22]. Thus, it cannot be assumed that birds
have been evolving towards an optimal flight configuration [23].
Nevertheless, some birds exhibit flight styles that are more maneuverable and adaptable
than comparatively-sized aircraft [24] such as navigating through crowded cities and forests
[25] as well as performing evasive maneuvers to escape from predators [26]. For example,
gulls (colloquially seagulls) are flight generalists that can perform thermal or dynamic soaring
[27] as well as cleverly maneuver to steal food from other unsuspecting birds (or tourists) [28].
These flight generalists provide a promising model for adaptable UAV designs as a single bird
can effectively operate at conditions that would satisfy many disparate mission profiles.
3
(a)
10−1
Passeriformes (P)
RQ−4 Global Hawk
Australaves (Au)
Coraciimorphae (Cor)
Insects
Strigiformes (Strig)
Birds
10−2 Accipitriformes (Ac)
UAVs
Mach number
Aequorlitornithes (Ae)
Aircraft
Gruiformes (G)
Columbaves (Col)
Strisores (Stris)
10−3
Ac
Ae
3
Clade
Col
Strig
Au
P
Cor
Stris
1
104 3 × 104 7 × 104 2 × 105 7 × 105 104 3 × 104 7 × 104 2 × 105 7 × 105
Reynolds number Reynolds number
Figure 1.1: Gliding bird Reynolds and Mach number regimes. (a) Reynolds (Re) and Mach
(M) number flight regimes. (b) Flight regime for gliding birds colored by clade. (c) Reynolds
number regime for each avian clade, the transparent gray line represents approximately the
critical Reynolds number ≈ 104 per smooth airfoil classifications.
available for birds, I surveyed the literature to identify studies that reported both non-flapping
airspeeds and wing shape parameters. With the outputs of this survey, I calculated the range
of Mach and Reynolds numbers used by birds, insects, UAVs, and crewed aircraft (Fig. 1.1)
[8, 25, 33, 35–114]. For birds, the ranges were computed from live measurements where birds
were observed to hold their wings extended during some portion of the flight and does not in-
clude any flapping-only flights. Airspeed measurements were reported using varied techniques
including global positioning system (GPS) loggers, rangefinders, radar, and wind tunnel stud-
ies.
4
The first important flow similarity parameter is the Mach number, which quantifies the
compressibility of the flow. Mach number is the ratio of the freestream velocity to the speed
of sound. Defining the Mach number regime is important when traveling close to or above the
speed of sound. Gliding birds do not approach this barrier with a Mach number range of 0.01
to 0.08 (Fig. 1.1a and b). Since this range of gliding bird flight occurs well below 0.3, it can
be modeled as incompressible flow [115].
The second important flow similarity parameter is the Reynolds number, which quantifies
the ratio of inertial to viscous properties in a flow as defined by:
ρcU∞
Reynolds number = (1.1)
µ
where c represents the characteristic length scale, U∞ represents the incoming freestream ve-
locity, and ρ and µ represent the air density and dynamic viscosity, respectively.
Unlike large aircraft or small insects, birds and UAVs operate at an intermediate Reynolds
number where neither viscous nor inertial effects can be neglected (Fig. 1.1a and b). Specifi-
cally, I found that gliding bird flight occurs between a Reynolds number of 1.5×104 to 5.2×105 ,
if the characteristic length (c) is selected to be the mean projected wing chord (defined as the
projected wing area divided by projected wingspan). In the aeronautical literature, the charac-
teristic length is usually chosen as the mean aerodynamic chord, defined as a weighted average
of the wing chord along the span. However, most avian studies do not publish the wing chord
as a function of span and instead often publish a mean projected chord calculated from a fully
extended wing. Because birds have been observed to both increase or decrease their mean
projected chord as wind speeds increase, there is increased uncertainty on the upper Reynolds
number bound [39, 85].
It is informative to separate the collected avian data into major clades defined by the phylo-
genetic classification established by Prum et. al (Fig. 1.1c and Table 1.1) [116]. A phylogeny
captures how closely related different species are, in a manner similar to a family tree. Clades
with large species such as seabirds (Aequorlitornithes) and raptors (Accipitriformes) tend to
glide near the upper bound of the Reynolds number range whereas small birds such as swifts
(Strisores) and passerines (Passeriformes) tend to glide near the lower bound (Fig. 1.1c). These
5
Table 1.1: Phylogenetic grouping of avian species.
trends are largely a function of the overall size of the birds within these clades. Figure 1.2a
shows Reynolds number as a function of the total mass of the bird where the increasing trend
is expected because heavier birds tend to have larger mean projected chords (c, Fig. 1.2b) and
fly at higher speeds (U∞ ) [117].
Next, I separated the avian Reynolds number regime into four sub-regimes that Carmichael
defined for smooth airfoils (Table 1.2) [118]. Birds glide within subcritical and supercritical
sub-regimes . In a subcritical regime (Reynolds number ≤ 7 × 104 ), once the laminar boundary
(a) (b)
7 × 105 0.500
0.250 Passeriformes
Mean projected chord (m)
2 × 105 Australaves
Reynolds number
Coraciimorphae
Strigiformes
7 × 10 4 0.100 Accipitriformes
Aequorlitornithes
Gruiformes
3 × 104 0.050 Columbaves
Strisores
104 0.025
0.01 0.10 1.00 10.00 0.01 0.10 1.00 10.00
Mass (kg) Mass (kg)
Figure 1.2: Reynolds number and chord correlates with body mass. (a) Reynolds number
and (b) mean projected chord (c) tend to increase as body mass increases.
6
Table 1.2: Reynolds number sub-regimes for gliding birds. The bold line separates subcrit-
ical from supercritical regimes as defined for smooth airfoils by Carmichael [118].
layer separates from the surface, the airfoil chord is too short relative to the flow velocity to
allow the separated boundary layer to transition to turbulent flow and reattach [118, 119]. This
severely degrades performance. Carmichael noted that for Reynolds number ≤ 3 × 104 the
flow is extensively laminar which results in higher aerodynamic efficiency, albeit at lower lift
coefficients, than at the higher subcritical sub-regime [118].
The supercritical regimes used by birds are normally characterized by laminar separation
bubbles (LSBs). LSBs occur when the laminar boundary layer separates from an airfoil, tran-
sitions to turbulent flow, and then reattaches to the airfoil as a turbulent boundary layer [119].
The region of separated flow is the “bubble” and degrades wing performance depending on its
stability and length [118, 119]. LSBs can be partially or completely eliminated by artificially
7
tripping a laminar boundary layer to transition to turbulent flow. Introducing surface roughness
can trip a boundary layer and feather roughness performs a similar function [33, 120]. For
Reynolds number ≥ 2 × 105 , LSBs still exist although the performance markedly improves
over the lower regimes [118].
Note that Table 1.2 is meant to support a qualitative understanding, but bird wings are three-
dimensional, rough, porous, flexible, and have a variable chord length. As such, a single bird
wing may have airfoils simultaneously operating in multiple Reynolds number regimes that do
not correspond to the Reynolds number regimes experienced by a smooth airfoil.
This survey does not encompass all avian biodiversity because some birds do not glide and
not all birds that do glide have been studied. Of note, I identified only one published gliding
measurement for the Galloanserae clade (Table 1.1), the barnacle goose (Branta leucopsis) that
was gliding at an airspeed of approximately 14.2 m/s (Reynolds number = 1.5 × 105 and Mach
number = 0.04) [121].
8
(a)
URCUNINA
10.00
Mass (kg)
1.00
0.10 Colorado
Black Widow
0.01
(b)
30
URCUNINA
15
Mass/S (kg/m 2 )
Colorado
Alula evo
1
(c)
Maximum lift-to-drag ratio
25 SBXC
20
URCUNINA
15
10
5 Black Widow
CICADA
Colorado
0
5
104 3 × 104 7 × 104 2 × 10 7 × 105
Reynolds number
Birds Wind tunnel experiments
UAVs Free flight measurements
CFD
Figure 1.3: Selected UAVs for comparison to birds. (a) UAV and bird masses (b) UAV and
bird wing loading (c) UAV aerodynamic efficiency by measurement type
methods. The uncertainty of each method can be highly variable and is discussed in more de-
tail in [122]. I found that the maximum lift-to-drag ratio of UAVs that fly in the same regime
as birds, ranges from 3–14 with a few exceptions (Fig. 1.3c). The URCUNINA-UAV and an
SBXC glider have a high reported efficiency of 18 and 24, respectively [8, 53]. This is not
a surprise as URCUNINA-UAV was designed for volcano monitoring and utilizes a pusher
propeller configuration while the SBXC is a hand-launched unpowered glider.
9
25
Types of indirect
Maximum lift-to-drag ratio
20 Passeriformes
Australaves measurements on live birds:
Coraciimorphae glider based
15
Strigiformes wind tunnel
Accipitriformes
airspeed tracking
10 Aequorlitornithes based (measured)
Gruiformes airspeed tracking
Columbaves based (estimated)
5
Strisores
Figure 1.4: Aerodynamic efficiency of birds. Maximum lift-to-drag ratio of gliding birds.
Solid lines connect measurements on the same individual.
Next, I surveyed the biological literature to identify measurements of the maximum lift-
to-drag ratio of gliding birds (Fig. 1.4). Similar to UAVs, these results were obtained with
different measurement methodologies that each have different sources of uncertainty. These
are discussed in detail in [122]. Of note, I found that there was evidence of extremely high
experimental uncertainty in the theoretical predictions and glider based studies (upside down
triangles and crosses in Fig. 1.4). Further, measurements on live birds in steady, gliding flight in
a wind tunnel were noted to represent a minimum bound on the birds’ aerodynamic efficiency
due to behavioral implications [122].
With this collected data it is now possible to compare the aerodynamic efficiency of gliding
birds to UAVs. Figure 1.5a incorporates all published estimates of avian and UAV aerodynamic
efficiency. However, due to the high experimental uncertainty in theoretical predictions and
glider-based studies [122], the avian data points were limited to wind tunnel measurements
(Fig. 1.5b). Note that live bird studies are affected by the individual bird’s behaviour and as a
result these avian data points represent a minimum bound on avian aerodynamic efficiency and
it is possible that birds are physically capable of more efficient glides[122]. With this reduced
data set, there was at least one UAV design that had a higher aerodynamic efficiency than any
bird flying within the same supercritical Reynolds numbers sub-regime. This observation is
10
(a) (b)
25
Maximum lift-to-drag ratio
20 UAVs
Passeriformes
Australaves
15
Coraciimorphae
Strigiformes
10 Accipitriformes
Aequorlitornithes
5 Gruiformes
Columbaves
Strisores
0
5
104 3 × 104 7 × 104 2 × 10 7 × 105 104 3 × 104 7 × 104 2 × 10
5
7 × 105
Reynolds number Reynolds number
Figure 1.5: Aerodynamic efficiency of birds compared to UAVs. (a) All published avian
efficiency metrics and (b) avian efficiency measurements from wind tunnel only.
based on the airfoil sub-regimes defined by Table 1.2 with divisions demarcated by the x-axis
ticks in Fig. 1.5b [118, 119, 123].
Reconsidering the full dataset in Fig. 1.5a, it remains possible that birds can outperform
UAVs operating within supercritical regimes. Note that even with that complete dataset, a
UAV (the SBXC) is the most efficient glider [53]. Unfortunately, there is high uncertainty in
the published avian theoretical estimates and wind tunnel results are not able to measure birds
that are expected to be highly efficient (such as an albatross). Therefore, I cannot state with
absolute certainty that birds are more or less efficient than UAVs in the supercritical Reynolds
number regimes. At best it can be noted that the efficiency of birds and comparable UAVs in
supercritical Reynolds number regimes appear to be within a similar range.
I was only able to identify one comparable UAV with a published aerodynamic efficiency
metric operating in the subcritical Reynolds number regimes. This is possibly because the
subcritical regimes are dominated by rotary or flapping wing designs. The CICADA UAV (a
low-cost, disposable glider) had a maximum lift-to-drag ratio of 3, substantially less than the
western jackdaw (Coloeus monedula), laggar falcon (Falco jugger), and common swift (Apus
apus) that fly within this same regime (Fig. 1.5b) [39, 54, 92, 124]. Subcritical UAVs are often
designed to minimize their wing loading (thus maximizing the wing area and chord for a fixed
wingspan) to provide controllability and to operate close to critical Reynolds numbers to allow
11
sufficient chord length for flow reattachment [80, 125]. However, this leads to reduced aero-
dynamic efficiencies than could be achieved in the lower subcritical range [118]. In addition
to controllability requirements, the CICADA and other fixed-wing UAV designs must satisfy
constraints on their geometry such as the need to stack easily within a specified enclosure [54].
Unlike the CICADA, birds can morph their wings to both adjust their stability characteristics
and fold their wings flat to their body when needed. It is likely that birds’ adaptable wing
geometry allows geometric and stability constraints to be satisfied without needing to sacrifice
performance when flying in these subcritical regimes.
In all, although this literature review suggests that there are many exciting discoveries yet to
be made about avian aerodynamic efficiency, my goal is to begin by analyzing the characteris-
tics of bird flight that readily outperform comparable UAVs: maneuverability and adaptability.
12
dorsal
z ventral
proximal distal
anterior
alula
x finger
elbow wrist
y
shoulder
posterior
Figure 1.6: Key bird wing morphing parameters. Simplified view of a bird’s wing highlight-
ing the skeletal structure which can be actively controlled by activating the wing muscles. The
skeletal drawing is adapted from [126].
remiges) including the primary and secondary feathers. They are attached to the bones and
each other via a complex system of ligaments, tendons, and muscles [134]. Primaries are
distributed along the length of the carpometacarpus and onto the digits. They are numbered
from the most proximal to the most distal feather (Fig. 1.7, I - X). Secondaries are attached to,
what are effectively, protrusions on the ulna and are numbered from the most distal to the most
proximal feather (Fig. 1.7, 1-10). Different species of birds have different amounts of primary
and secondary feathers.
The simplification of modeling the wing as a six-bar linkage does not capture that the
elbow and wrist joints do have some capacity for rotation about three degrees of freedom:
extension/flexion, pronation/supination (twist) and elevation/depression [126], although each
joint’s range of motion varies based on the species [126, 130, 135]. The shoulder joint is also
important in bird flight and results in a solid body rotation of the wing shape about the root.
This joint is discussed within multiple chapters of this work, but future directed studies on the
13
digit I
radiale carpometacarpus
humerus ulna
proximal distal
a
alula feathers
propatagium interremigial smooth muscle
b
interremigial elastic
ligament
X
IX
VIII
3 2 1
7 6 54 I
10 9 8
II VII
III
a
IV
VI
V b
secondary feathers primary feathers
(1-10) (I - X)
Figure 1.7: Simplified anatomical drawings of a bird’s wing bones and feather attach-
ments. Drawing is based on a turkey vulture (Cathartes aura). The musculature is not in-
cluded for clarity. This anatomical view is for reference only, as there is a lot of diversity in
bone and feather shapes. The proximal airfoil shape (dashed line, a-a) differs substantially
from the distal airfoil shape (dashed line, b-b).
role of the shoulder joint in avian flight are required. In addition, birds can actively control two
of their digits: digit I (a thumb-like appendage) and digit II (a pointer-finger-like appendage)
(Fig. 1.7) [133–135]. Here, I focus on the role of the elbow and the wrist predominately as they
have been shown to be responsible for the majority of wing shape variation in birds [122, 126].
14
1.6 Maneuverability vs. stability
Avian flight capabilities are, in part, permitted by neurological control [130, 136] combined
with the physical capability to dynamically morph their wing or tail shape [127, 137]. In this
thesis, I focus on deciphering the role that wing morphing plays in adaptable and maneuverable
bird flight. Although I have defined maneuverability as the ability to adjust the velocity vector
direction and/or magnitude [19, 20], it is still challenging to quantify maneuverability with a
single metric. Therefore, throughout this work I leveraged another flight trade-off, that between
stability and maneuverability.
shoulder
(a) (b)
joint
stable
NP behind
(c) (d)
unstable
NP in front of
: center of gravity
NP : neutral point
Figure 1.8: Simplified diagram of static stability responses. (a) A stable flyer tends to
return back towards it’s trim condition after a perturbation, similar to (b) a ball on a concave
surface. (c) In contrast, an unstable flyer tends to move even further from it’s trim condition
after a perturbation, similar to (d) a ball on a convex surface. The outlined shapes represent the
equilibrium condition for both the bird and the ball.
Flight stability traditionally quantifies a flyer’s response to being disturbed (or perturbed)
from an equilibrium condition, also known as trimmed flight. Stability is evaluated by consid-
ering both the static and dynamic response after a flyer has been disturbed from it’s initial trim
condition. A gust is a common example of a flight disturbance. A flyer is statically stable if,
after a disturbance, the flyer initially tends back towards it’s trim condition (Fig. 1.8a). This
response is similar to a ball after being nudged when it is resting on a surface that is concave
(Fig. 1.8b). A flyer is statically unstable if, after a disturbance, the flyer tends to move even
15
further from the initial condition (Fig. 1.8c), like a ball on a convex surface (Fig. 1.8d).
Static stability is often investigated for either the longitudinal or lateral degrees of freedom.
This differentiation can be made for a rigid aircraft with six degrees of freedom that is given
the constraints of symmetric flight and undergoes only small disturbances. In this case, the
longitudinal and lateral degrees of freedom can be decoupled [138, 139]. Throughout this work,
I make these assumptions to permit a focus on the longitudinal degrees of freedom specifically.
Longitudinal static stability requires that the center of gravity of the flyer is behind the neutral
point of the flyer (Fig. 1.8a). The neutral point is the location where the pitching moment
is independent of the angle of attack. Of note, static stability is independent of time and is a
necessary, but insufficient condition for complete stability.
To be completely stable, a flyer must also be stable over time, often called dynamic stability.
This is the case if, after a disturbance, the flyer eventually returns to its trim condition without
the use of any control inputs. The path followed after the disturbance can be quantified by
calculating the eigenvalues of the resultant dynamic system [138, 139]. Once the eigenvalues
(λ) are known the natural frequency and damping ratio can be calculated as:
λ = (re) ± j(im)
p
ω = re2 + im2 (1.2)
−re
ζ=
ω
Where re and im represent the real and imaginary parts of each eigenvalue respectively.
Whether the dynamic response is stable depends on the sign of the eigenvalue’s real com-
ponent. If the real component is positive, the flyer is unstable (Fig. 1.9c and f). If the real
component is negative, the flyer is stable (Fig. 1.9a and d). If the eigenvalue has no real com-
ponent, the response will be considered neutrally stable. Neutral stability indicates that there
would not be any amplification or decay of the natural frequency amplitude that resulted due
to a disturbance (Fig. 1.9b and e). Note that if the eigenvalue has no imaginary component,
the response will be steady and non-oscillatory (Fig. 1.9d, e, f). However, most commonly
in aircraft analyses the roots will be a complex pair, which signifies an oscillating response
after a disturbance (Fig. 1.9a, b, c). The dynamic response associated with each eigenvalue
16
(a) stable (b) neutral (c) unstable
time
Imaginary axis
Real axis
Figure 1.9: Simplified diagram of dynamic stability responses. The dynamic stability of a
system is quantified through the eigenvalues, which can be plotted on the real (x) and imaginary
(y) axes. If the eigenvalues are complex pairs, the response is oscillatory. The stability of
the system depends on the sign of the real component, where (a) negative is stable, (b) zero
is neutrally stable, and (c) positive is unstable. Note only the complex pair with a positive
imaginary value is shown in this diagram. If the eigenvalues are real numbers, the response
is steady and non-oscillatory. Again, the stability depend on the sign of the real number. The
response is (d) stable if the root is negative, (e) neutrally stable if the root is zero, and (f)
unstable if the root is positive.
of a linear system can be superimposed to obtain the complete response of the aircraft. For
example, the longitudinal response of rigid body aircraft undergoing small disturbances will be
characterized by four eigenvalues, often composed of two complex pairs.
A stable flyer inherently develops restorative moments to return to its trim position after
a disturbance, whether that disturbance is accidental or purposeful. However, to perform a
maneuver requires a purposeful departure from the trim condition and thus a stable flyer must
generate sufficient moments that overcome the inherent restorative moments. These restorative
moments lead to the expected trade-off between stability and maneuverability. It is for this
reason that fighter aircraft, which prioritize maneuverability, are designed to be more unstable
than passenger aircraft, which prioritize safety and fuel efficiency.
The advantage to approaching maneuverability through a stability-based analysis is that
there are well-established methods to evaluate the static and dynamic stability of an aircraft
17
[138–140]. These approaches are implemented and discussed in detail throughout this work.
18
tain this capability to shift between stability modes. This work focused specifically on static
stability characteristics. A version of this chapter has been published [142].
Finally, Chapter 4 combines the aerodynamic approach introduced in Chapter 2 with the
inertial approach introduced in Chapter 3 to quantify the complete dynamic stability of an indi-
vidual gull. This analysis accounts for the different wing configurations that can be achieved by
manipulating the gull’s shoulder, elbow, and wrist angle. I included a variable shoulder angle
in this work to enable a trimmed flight condition. Of the identified configurations that could
trim, those with high wrist angles had non-oscillatory, unstable responses. Within the stable
configurations, I found strong interactive effects between the shoulder, elbow, and wrist angles.
Compared to UAV metrics, the gull wing configurations are expected to be controllable by a
human pilot, albeit with a heavily damped phugoid mode. This heavily damped response indi-
cates that to effectively perform a given maneuver a gull may need to morph into an unstable
configuration.
In all, this thesis provides the first complete investigation of the longitudinal stability asso-
ciated with morphing the avian elbow and wrist. These analyses were completed in a manner
that allow for comparison to traditional aeronautical metrics, which in turn allows the identifi-
cation of the aspects of bird flight that may best enhance future UAV designs.
19
Chapter 2
2.1 Summary
Birds dynamically adapt to disparate flight behaviors and unpredictable environments by
actively manipulating their skeletal joints to change their wing shape. This in-flight adaptability
has inspired many uncrewed aerial vehicle (UAV) wings, which predominately morph within
a single geometric plane. In contrast, avian joint-driven wing morphing produces a diverse set
of non-planar wing shapes. Here, we investigated if joint-driven wing morphing is desirable
for UAVs by quantifying the longitudinal aerodynamic characteristics of gull-inspired wing-
body configurations. We used a numerical lifting-line algorithm (MachUpX) to determine the
aerodynamic loads across the range of motion of the elbow and wrist, which was validated
with wind tunnel tests using 3D printed wing-body models. We found that joint-driven wing
morphing effectively controls lift, pitching moment and static margin, but other mechanisms
are required to trim. Within the range of wing extension capability, specific paths of joint
motion (“trajectories”) permit distinct longitudinal flight control strategies. We identified two
unique trajectories that decoupled stability from lift and pitching moment generation. Further,
extension along the trajectory inherent to the musculoskeletal linkage system produced the
largest changes to the investigated aerodynamic properties. Collectively, our results show that
gull-inspired joint-driven wing morphing allows adaptive longitudinal flight control and could
20
promote multifunctional UAV designs [141].
2.2 Background
A bird can begin its day foraging in a slow glide, suddenly needing to evade a predator,
only to later fly home battling an incoming storm. The adaptability demonstrated by birds is in
part due to their ability to morph the shape of their wings, both actively and passively [85, 127,
137, 143]. Previous research has shown that active wing morphing allows birds to dynamically
adapt their aerodynamic performance and stability characteristics in response to changing flight
conditions or requirements [127, 128]. In comparison, fixed-wing UAVs are often designed to
satisfy specific functions, such as high altitude surveillance or long endurance flights, and
efficient operation is limited to their intended mission parameters [6]. The adaptability offered
by avian wing morphing is highly desirable for UAVs as it may broaden the efficient operational
range, reduce operating costs as well as offer enhanced or novel capabilities [6, 143–145].
In addition, UAVs often face aerodynamic control challenges including the need to adapt to
variable environmental conditions [146] or maneuver through complex territories [147], while
birds complete similar tasks with apparent ease. Therefore, it is of no surprise that as engineers
strive towards the objective of an adaptive, multifunctional UAV, bird wings have directly and
indirectly inspired many morphing wing designs [6, 29–33].
The majority of current engineered morphing wings adjust their wing geometry discretely
within one or two planes such as span, sweep, dihedral, etc. [29, 30, 32, 33]. In contrast, bird
wings are composed of an underlying musculoskeletal system that can be approximated as a
non-planar six-bar linkage system [133]. When the applicable muscles are activated, the non-
planar musculoskeletal linkage causes three-dimensional (3D) changes to the overall wing ge-
ometry [126, 128, 133]. Active manipulation of only two skeletal joints, the elbow and wrist is
responsible for the majority of this wing shape change (Fig. 2.1) [126, 128]. These joints have
three degrees of freedom i.e. the ability to extend/flex, pronate/supinate and elevate/depress
[126]. However, within gliding flight extension/flexion dominates the range of motion and
thus we have limited our study to focus only on the range of motion of extension and flexion
for the elbow and the wrist (hereafter referred to as “joint-driven wing morphing”). Therefore,
21
in this work the elbow and the wrist represent an approximated minimum set of coordinates
that is required to define the overall wing shape. In this case, traditional geometric properties
including distributions of wing twist, sweep, dihedral and the final wingspan can be approx-
imated as functions of the joint positions [128, 148]. Further, these non-planar wing shapes
likely have aerodynamic characteristics that differ from comparable planar wing aerodynamic
theory [122], as highlighted by the Hyper Elliptical Cambered Span (HECS) wing inspired by
gulls, which had improved aerodynamic efficiency (higher lift-to-drag ratio) compared to an
equivalent planar wing [149].
Despite the physical differences, no engineered morphing wing designs have implemented
biologically accurate joint-driven wing morphing. This discrepancy is likely due to the many
challenges associated with implementing a non-planar morphing wing including complex man-
ufacturing methods as well as additional mechanism weight and structural rigidity considera-
tions [6]. In addition, because we cannot assume that a bird’s wing has been optimized for
flight [150], it follows that we cannot assume that a biologically accurate morphing wing will
provide any advantage over a planar morphing wing. This leads to the main question of this
first chapter: does avian-inspired joint-driven wing morphing provide sufficient aerodynamic
benefits to warrant implementation in a future UAV wing design?
To address this question, we investigated the benefits of gull-inspired joint-driven wing
morphing by quantifying longitudinal aerodynamic stability and control, which are critical for
any successful flight, be it high-altitude surveillance or evasive maneuvers [137, 138, 140].
We assumed a symmetric glider with no sideslip, to permit longitudinal (pitch, motion in the
x-z plane, Fig. 2.1) and lateral (roll and yaw, motion in the x-y or y-z planes) components to
be decoupled [140]. In this work, longitudinal control refers to a morphing wing’s ability to
actively adjust its generated lift force and pitching moment. Traditional aircraft can control lift
through wing flap deflections and the pitching moment through elevator deflections [138, 140].
In addition to longitudinal control, we considered the effects of joint-driven wing morphing
on longitudinal stability and balance. Longitudinal stability is the tendency of an aircraft to
return to its equilibrium after an external disturbance, which requires an evaluation of both the
static (initial) and dynamic (time-dependent) response [140]. Here, we focused solely on static
stability, a necessary but insufficient condition for full stability. Finally, longitudinal balance is
22
(a) x z (b)x (c)x (d)
NACA 0020
y y y y x
z
Liu S20 y
ε ε ε
Lift Drag
Liu S40 U
ω ω ω Pitch
0.52 m
ε elbow angle
ω wrist angle NACA 3603
centre of gravity
peripharal landmark
joint landmark
Figure 2.1: Gull wings inspired our analyses of how avian joint-driven wing morphing
affects longitudinal aerodynamic control, stability and balance. (a) Wings from gull ca-
davers (n = 3) were manually manipulated throughout the range of motion of the elbow and
wrist while tracking the 3D position of seven peripheral landmarks (black points) and four joint
landmarks (white points). (b) The identified wing shapes were simplified using linear approx-
imations between the peripheral landmarks. Four airfoils were used to create the wing-body
configurations. (c) The simplified wing shapes were reflected about the x-z plane and investi-
gated using MachUpX and cosine clustering to distribute the control points along the span. (d)
To validate the numerical results, nine wing shapes were 3D printed and tested in a 2ft-by-2ft
low-speed wind tunnel.
the ability for a glider to fly at an equilibrium, also known as trimmed flight.
To quantify the longitudinal control, stability, and balance associated with gull-inspired
joint-driven wing morphing, we first identified the 3D simplified wing shapes associated with
the extension and flexion of the elbow and wrist for hybrid glaucous-winged (Larus glaucescens)
× western (Larus occidentalis) gulls (n = 3, Fig. 2.1a). We selected gulls as our study species
because their non-planar wing shape is known to be actively controlled by elbow and wrist
manipulation [128]. In addition, gulls are a good model species for multifunctional UAVs as
they are generalist flyers, using a wide variety of flight styles from steady glides to sudden ma-
neuvers [112, 151]. Next, we aligned and simplified each extracted wing shape (n = 1031) and
connected these wings to a gull-shaped body (Fig. 2.1b). With each final wing-body configura-
tion, we predicted the aerodynamic properties using MachUpX, a low-order numerical general
lifting-line model (Fig. 2.1c) [152–154]. We validated the outputs from MachUpX with exper-
imental wind tunnel measurements on 3D-printed half-span equivalent wing-body models (Fig.
2.1d). Finally, we investigated four quasi-steady joint extension trajectories within the range
of motion that could be implemented to improve the adaptability of a gull-inspired joint-driven
23
morphing UAV wing.
2.3 Methodology
Wings from gull cadavers (n = 3) were manually manipulated throughout their full range
of elbow and wrist extension and flexion while filming eleven key landmarks with three high
speed cameras. Seven landmarks on the wing periphery (black points, Fig. 2.1a) provided an
outline of the overall wing shape through the range of motion and four landmarks on the wing
joints (white points, Fig. 2.1a), allowed us to calculate the elbow and wrist angle associated
with each wing outline. The videos were digitized using DLTdv5 software [155] to return the
3D position of each landmark. For a detailed methodology on how the 3D landmarks were
obtained please refer to Harvey et. al [128].
The final extracted 3D landmarks were reoriented using a custom R script to align the wrist
joint with the humerus head on the y- and z-axis (Fig. 2.1b). We defined 0° angle of attack by
rotating the wing about the y-axis until the tip of the first secondary feather (S1) was aligned
with the wrist joint along the x-axis. Next, we exported the aligned landmarks into a custom
Python script to ease interfacing with MachUpX and limited the configurations to those with
elbow and wrist angles above the minimum angles used in vivo by gliding gulls [128]. Next,
the wings were segmented along the x-axis at each peripheral landmark. For each segment,
the custom script returned simplified leading and trailing edge points, and a value of sweep,
dihedral, and twist within that segment. The last 5% of the wing segment span was linearly
blended into the next segment to avoid sharp changes in wing geometry. These geometric
properties were selected to minimize error between the true peripheral landmarks and the final
simplified wing shapes. The maximum summed error for all landmarks was found to be 2.1mm.
There are a few key assumptions on the final wing shapes. First, we assumed that through-
out elbow and wrist extension and flexion, the digits (including the alula) remain fixed. Next,
we elected to not consider the capability to elevate/depress or supinate/pronate along joints
within the wing or to adjust overall wing position via rotation at the shoulder joint. Addition-
24
ally, we neglected wing porosity, roughness, and flexibility to isolate the effects solely due
to the shape change created by joint-driven wing morphing. Due to these assumptions and
the alignment routine, the final wing shapes are not necessarily representative of an in-flight
configuration used by live gulls.
Each final wing shape was attached to a gull-inspired body. We selected a NACA0020
airfoil based on an estimated body length (41cm) and body height that was estimated using
an allometric relationship with a body mass of 0.91kg [156, 157]. The shoulder joint was
positioned at the quarter-chord of the body. The body width and position of the shoulder joint
along the y-axis (Fig. 2.1b) was determined from allometric relationships with the same body
mass [157]. To create an avian-like body shape, we reduced the body chord from a maximum at
the body’s center to the wing root chord length at the body’s edges using a cosine distribution.
For each wing-body configuration, we quantified a few traditional geometric properties
including the total and projected wing area, as well as wing tip twist, sweep, and dihedral (Fig.
2.5c). The total wing area is the total wing-body area and the projected wing area is the wing-
only area projected onto the y-x plane at 0°, the body projected area remains approximately
constant. The wing tip twist, dihedral and sweep values were calculated for the most distal
wing segment within our custom Python script. Note that all tested wing shapes had wing tips
with some degree of backwards sweep and anhedral.
Next, we selected three airfoils to distribute along the wingspan. We extracted two airfoils
from a previously published scan of a gull wing at 20% (Liu S20) and 40% (LiuS20) span (Fig.
2.1b) [158]. As the species was not identified in that study, we assumed that the airfoil shape
is similar between different gull species. We defined the airfoils for each segment based on the
locations of the skeletal joints as illustrated in Fig. 2.1b. For segments more distal than the
wrist joint, we assigned a NACA 3603 airfoil because the scan did not capture the wing after
77.2% span [158]. We selected the NACA 3603 after compiling a list of measured distal avian
airfoils [159–161]. All aerofoils were linearly blended to create a smooth wing surface within
the MachUpX framework [152].
MachUpX requires the airfoil performance data (lift, drag and pitching moment) at an ap-
25
propriate Reynolds number as inputs. We used XFOIL to predict the 2D airfoil properties
for each of the four selected airfoils used on the wing and body [162]. Following XFoil’s
implementation of the en method for transition prediction, we assumed a critical N-factor at
transition of 9. This value is commonly used to model the turbulence intensity in an average
wind tunnel. We estimated the range of Reynolds numbers that would be experienced by each
airfoil and stored each data set separately. For Reynolds numbers above 100,000 we incre-
mented by 50,000 (to a maximum of 300,000 on the body) and below 100,000 we incremented
by 10,000 (to an enforced minimum of 10,000 on the wing tip). Our custom Python script
predicted the Reynolds numbers experienced by each wing segment and selected the closest
Reynolds numbers number file for the segment airfoil. This methodology accounts for changes
in aerofoil performance at variable Reynolds numbers across the wingspan. MachUpX linearly
interpolates the 2D airfoil properties across wing segments from the provided files.
26
(a) 180 Experimental (b) (c)
MachUpX 1 2 3
Wrist angle
4 5 6
140
(90°,140°) (116°,140°) (145°,140°)
120
7 8 9
0.4
0.0 −0.2
−0.4 4 5 6 −0.4 4 5 6
1.2 0.2
0.8
0.0
0.4
−0.2
0.0
7 8 9 7 8 9
−0.4 −0.4
−10 −5 0 5 10 −10 −5 0 5 10 −10 −5 0 5 10 −0.4 0.0 0.4 0.8 1.2 −0.4 0.0 0.4 0.8 1.2 −0.4 0.0 0.4 0.8 1.2
Angle of Attack (°) CLmorph
Figure 2.2: MachUpX effectively predicted the aerodynamic forces and moments for the
investigated wing shapes. (a) The in vivo range of motion used by gliding gulls was in-
vestigated using MachUpX (n = 1031, translucent gray points; higher-sampled configurations
accordingly appear darker) and validated with wind tunnel tests of 3D printed wing shapes
distributed across the range (white squares). (b) Experimental wings (photo, left) were tested
at an 80% scale from the numerical wings (front view of Fig. 2.1b, right). (c) Planform views
of the 3D printed wing shapes corresponding in location/arrangement to the white squares in
(a) and the plots in (d) and (e). Wing shapes in (b) and (c) are described by (elbow angle, wrist
angle). (d) MachUpX predicted the lift force within the expanded uncertainty range at low
angles of attack. (e) The numerical results for lift and pitching moment lie within the expanded
uncertainty range on both variables. The error bars represent the expanded uncertainty range,
approximately a 95% confidence interval.
27
of attack for 1031 wing-body configurations with a total of 9720 converged test cases. The
majority of the convergence was at lower angles of attack.
We used the MachUpX outputs to investigate the lift and pitching moment for the wing-
body configurations. We were not able to investigate the maximum lift produced by each con-
figuration due to the limitations of the numerical method for the complex shapes. Additionally,
we did not investigate drag as it was outside the scope of this study, but it likely plays a variable
role in avian wing morphing and warrants future investigation. Note that MachUpX predicted
the experimental drag within the expanded experimental uncertainty range for all wings except
for configurations 6, 7, 8, and 9 (Fig. 2.2c).
To allow comparison between the experimental and numerical lift and pitching moments,
we non-dimensionalized the outputs by the dynamic pressure ( 12 ρU∞
2
), maximum total wing
area (Smax ) and maximum wing-body mean chord (cmax ) for each specimen across all mor-
phed configurations to obtain CLmorph and Cmmorph , respectively. These adjusted aerodynamic
coefficients allow comparisons across different wing specimens without filtering out the wing
area change due joint-driven wing morphing. As there is relatively little information about
the location of a bird’s center of gravity, we evaluated the pitching moment about the body’s
quarter-chord (aligned with the shoulder joint); this effectively assumes that the center of grav-
ity is located at the body’s quarter chord (Fig. 2.1a).
To validate the numerical results from MachUpX, we 3D printed nine wing-body config-
urations across the range of motion (Fig. 2.2a, b and c and Fig. 2.3). The wing shapes are
identical to those tested in MachUpX due to a feature that outputs the DXF files for a specified
wing shape [152]. These files were prepared into a 3D model and 3D printed on a CON-
NEX500 printer in rigid plastic (VeroWhitePlus). We printed half-span wing-body models at
80% scale to ensure that the largest wingspan only extended to 71% of the tunnel width. The
printed wings were lightly sanded to minimize surface roughness.
The final wings were tested in the University of Michigan 2ft-by-2ft low-speed wind tunnel
at 12.5m/s (Fig. 2.3). This enforced a constant Reynolds number (Re) between the experimen-
tal and numerical tests of approximately 1.4-1.5×105 based on the maximum mean chord for
28
Figure 2.3: All final 3D printed wings. (a) Organized from left to right based on Fig. 2.2c as
configuration 7, 8, 9, 4, 5, 6, 1, 2, 3.
the wing-body specimen. The wings were mounted to the top of the wind tunnel on a 6-axis
load cell (ATI Delta) that sampled at 4kHz for 45 seconds. There was a 0.5-inch gap between
the body and the tunnel edge. The load cell was installed on a rotary table (Parker 30012-S)
connected to a motor (VEXTA PK266-03B) and a VELMAX motor controller. This installa-
tion method used the reflection plane methodology and blockage constraints were neglected
due to the minimal size of the wings relative to the tunnel test section [163]. Using a custom
MATLAB script, we performed an angle of attack sweep from 0° to 24° (∆2°), 25° to 1° (∆2°),
0° to -3° (∆1°) and -4° to -20° (∆2°).
To determine the experimental uncertainty, we implemented the definition and procedures
outlined in the Guide to Uncertainty Measurement [164] through the R package errors [165].
The calculated uncertainties were multiplied by a coverage factor k = 2 to determine an ap-
proximate 95% confidence level on the experimental data and are included as error bars in Fig.
2.2d and e [164].
Finally, to investigate the differences between the wind tunnel and numerical results, we fit
a first order linear model to the numerical and experimental results from the nine tested wing
shapes. The lift and pitching moment data were the independent variables for the models and
29
Figure 2.4: Wings installed in the wind tunnel. (a) Front view of configuration 6 (Fig. 2.2c)
(b) Side view of configuration 1 (Fig. 2.2c).
the elbow, wrist, angle of attack (or lift for the pitching moment model) and method were the
dependent variables.
Turbulence intensity
30
attack, we expect that the role of turbulence was relatively minor. But as the wing approaches
stall, turbulent flow in the wind tunnel may have contributed to improved flow attachment and
thus higher lift than was predicted by the numerical results.
2.4 Results
The MachUpX lift and pitching moment results fell within the expanded experimental un-
certainty range for low angles of attack (Fig. 2.2d and e)[164]. Disregarding the expanded
uncertainty range, there was an average absolute error in the lift and pitching moment of 0.08
and 0.04 at 0°, respectively. Note that this is largely manifested by MachUpX under predicting
the lift for lower wrist angles and higher elbow angles (Fig. 2.2d, configurations 6, 8, and 9).
By 5°, the error in both metrics more than doubled. We found that there was no significant
effect of methodology (experimental or numerical) on the predicted coefficient of regressions
for either the elbow or the wrist. However, there was a significant effect of methodology (p-
value < 0.001) on the predicted coefficient of regression for the angle of attack for both the
lift and pitching moment models. Specifically, at high angles of attack MachUpX either did
not converge, or under predicted the magnitude of lift and pitching moment, likely due to more
stalled regions. Informed by these outcomes, the rest of our analyses were limited to angles of
attack less than 5°.
With the outputs from MachUpX, we evaluated the lift and pitching moment at a constant 0°
angle of attack for all wing-body configurations (Fig. 2.5a and b). We found that the majority of
the configurations produced a negative (nose-down) pitching moment and a positive lift force.
Figures 2.5a and b reveal that wings with high elbow and low wrist angles had the highest
lift and lowest pitching moment, representing the highest absolute loading condition. We used
the R package GGally [167] to determine the normalized values of the wing tip geometry,
wing area parameters (that were first normalized for each specimen’s maximum wing area) and
31
aerodynamic forces and moments as shown in Fig. 2.5c. Unexpectedly, we found that these
highly loaded wing-body configurations did not have the highest wing area (total or projected)
but instead had the most positive twist angle at the wing tips (Fig. 2.5c).
Longitudinal control
(a) 180 (b) 180
160 160
Wrist angle (°)
0.2 −0.10
increasing increasing
120 120
0.1 force moment
−0.15
0.0
−0.20
100 100
80 100 120 140 160 80 100 120 140 160
Elbow angle (°) Elbow angle (°)
(c) 0.6 max max max max wash-in backward dihedral*
Normalized value
0.3
0.0
−0.3
Figure 2.5: Joint-driven wing morphing provides a reliable method to control the lift and
pitching moment. (a) and (b) Constant lift and pitching moment contour lines overlaid on
the numerical results at 0° angle of attack (n = 414) highlight a region of high magnitude lift
and pitching moment. (c) Normalized geometric properties, centered on 0° twist angle (black
line). The black dashed line represents the wing configuration with the maximum projected
wing area. The color scheme is based on the normalized value of CLmorph .
Positive twist known as wash-in, occurs when the wing tips are at a higher angle of at-
tack than the wing root. Traditional aircraft are designed with wash-out so that the wing root
will stall before the wing tip, which improves the handling characteristics of the aircraft [168].
The importance of twist in lift and pitching moment production is supported by a previous
experimental study that demonstrated that active wing twist morphing alone provides effec-
tive longitudinal control of a tailless aircraft [169]. For the tested configurations, we found
that wash-in was associated with a highly swept wing while the less swept (extended) wings
exhibited wing tip wash-out. Although wing tip twist was a good predictor of the developed
32
forces and moments, models informed by the elbow and wrist also successfully predicted the
developed lift and pitching moment.
To quantify and visualize the longitudinal control characteristics of the wings from the
MachUpX outputs, we fit linear mixed effects models with an interaction term between all the
first order fixed explanatory variables to CLmorph and Cmmorph . The fixed explanatory variables
included the elbow and wrist angle for both models, and the angle of attack for the lift model
which was replaced with CLmorph for the pitching moment model. In addition, we included the
wing specimen identification number as a random effect in each model. Higher order terms on
each explanatory variable were included and the final polynomial forms were selected using
Akaike Information Criterion (AIC) to compare the weights of each model. The goodness-of-
fit of a linear mixed effects model can be assessed through the marginal R2 , which quantifies
the variance explained by the fixed effects and the conditional R2 , which quantifies the variance
explained by the full model [170]. We calculated the marginal and conditional R2 using the
R package MuMIN [171]. Both the lift and pitching moment models fit our numerical results
well. The lift model had a marginal R2 of 0.81 and a conditional R2 of 0.90 while the pitching
moment model had a marginal R2 of 0.97 and a conditional R2 of 0.98. The good model fits
suggest that an engineered joint-driven morphing wing could provide a reliable method for
controlling the lift and pitching moment in flight, similar to twist control.
Our results have important consequences for estimations of aerodynamic forces on live
gliding birds. Specifically, the wing area may not be an effective metric to differentiate between
the lift and pitching moment produced by different wing shapes used by the same bird (Fig.
2.2c). Instead, if they can be obtained the wing tip twist and sweep angle can provide improved
metrics.
Next, we assessed longitudinal balance and stability using the pitch-stability derivative
(slope) and zero-lift pitching moment (y-intercept) from the linear model:
dCmmorph
Cmmorph = CLmorph + Cmmorph,L=0 (2.1)
dCLmorph
A flyer is statically stable if the pitch-stability derivative is negative, and a flyer can be
balanced while statically stable if the zero-lift pitching moment is positive. To quantify the
pitch-stability derivative, we fit a linear model (Eqn. 2.1) to the outputs of each wing-body
33
configuration if it had converged for a minimum of four angles of attack. We verified that the
dCm
data fit was linear (R2 > 0.99). We then extracted the slope ( dCL morph ) and the y-intercept
morph
Cm morph
Cm morph, L = 0
dC m morph
dC L morph CL morph
trim
Figure 2.6: Representative example of the relationship between lift and pitching moment
for a statically stable wing. Longitudinal static stability can be quantified with the zero-lift
pitching moment and pitch-stability derivative.
First, we evaluated the static stability, which requires that a stable glider perturbed from
its equilibrium by an external disturbance (Fig. 2.6) will develop a change in pitching mo-
ment with an opposite sign to the change in the lift force. This returns the glider towards its
equilibrium. We found that all the investigated wing-body configurations had entirely negative
pitch-stability derivatives and thus were statically stable (Fig. 2.7a). The magnitude of the
static stability was higher for more folded wing configurations. Next, we evaluated the tradi-
tional aircraft static margin metric which is equal and opposite to the pitch-stability derivative:
dCmmorph xCG − xN P
static margin = − = (2.2)
dCLmorph cmax
Where xCG is the location of the center of gravity and xN P is the neutral point of the wing-
body configuration. The neutral point is the location where the pitching moment is independent
of the angle of attack. The negative pitch-stability derivatives (positive static margins) for
all our tested configurations indicated that xN P is substantially aft of the body quarter-chord.
34
When we considered the magnitude of the change, our results revealed that joint-driven wing
morphing permits a maximum static margin shift of 24% of cmax , or approximately 5.1 cm.
Moreover, if the gull-inspired wing-body xCG was relocated between 4.1 to 9.2 cm behind
the shoulder joint, this sizable static margin shift would allow the glider to shift between stable
and unstable configurations. This large static margin shift would result in significantly different
handling qualities between configurations, which is usually an undesirable condition for human
pilots [138].
In addition, we found that joint-driven wing morphing provides an effective method to con-
trol the static margin as the elbow and wrist angles were good predictors of the pitch-stability
derivative. We fit a mixed effect model to all of our wing-body configurations following a simi-
lar procedure as used for the lift and pitching moment models with the elbow and wrist angle as
fixed explanatory variables. We found that this model provided a good fit for the pitch-stability
derivative with a marginal R2 of 0.69 and a conditional R2 of 0.83.
This analysis assumes a fixed xCG , but in actuality morphing a wing can cause the center
of gravity to shift. To estimate an approximate xCG shift associated with joint-driven wing
morphing, we investigated the solid 3D printed wings and found a maximum backwards shift
of 1.3 cm. Note that this will not necessarily be comparable to an engineered morphing wing
because the weight distribution will depend on the manufactured design. However, the xCG
shift for a real bird wing-body is likely smaller than the 3D printed wings due to the lightweight
nature of feathers compared to the musculoskeletal system and body. To obtain an approximate
estimate of the static margin shift, we assumed that the xCG moves opposite to the xN P . This
situation would reduce the maximum static margin shift to 3.8 cm (18%). Note that we non-
dimensionalized the static margin using cmax , but traditional static margin analyses use the
mean aerodynamic chord which will be smaller than cmax . In all, we expect that our calculated
static margin shift of 18% represents a conservative estimate.
35
Longitudinal stability and balance control
(a) 180 (b) 180
160 160
Wrist angle (°)
−0.25 −0.02
increasing
−0.30
120 stability 120 −0.04
−0.35
−0.40
−0.06
100 100
80 100 120 140 160 80 100 120 140 160
Elbow angle (°) Elbow angle (°)
Figure 2.7: Joint-driven wing morphing provides a reliable method to control static sta-
bility characteristics, but not the trim state. (a) constant zero-lift pitching moment and (b)
constant pitch-stability derivative contour lines overlaid on the results obtained from all wings
that converged at more than four independent angles of attack (n = 1012).
Next, we investigated the longitudinal balance condition which ensures that while in trimmed
flight a glider can create sufficient lift to support its weight when there is no net moment (math-
ematically represented by a positive x-intercept of Eqn. 2.2, Fig. 2.6). Because we require a
negative slope for static stability, this in turn requires the y-intercept or zero-lift pitching mo-
ment to be positive. We found that none of our wing-body configurations had a positive zero-lift
pitching moment (Fig. 2.7b). As a result, a gull-inspired wing-body with the wings aligned
following our convention could not be balanced at 0° while flying in a stable condition. This
result is identical to a positively cambered aircraft wing which has a negative zero-lift pitching
moment [140]. Furthermore, our results showed that the elbow and wrist would not provide a
reliable method to control the trim position because they were poor predictors of the zero-lift
pitching moment (marginal R2 of 0.39 and a conditional R2 of 0.42).
However, like fixed-wing aircraft a horizontal tail with control surfaces could be added
to gain control. Interestingly, unlike fixed wing aircraft, birds not only have a controllable
horizontal tail, but they also can rotate their wing about their shoulder joint. This is especially
relevant because birds are capable of sustained flight without their tails, possibly suggesting
36
an alternative method can be used to maintain trimmed flight [137]. It follows that to have
controllable, balanced flight for the wing-body configurations investigated, it will be necessary
to investigate the possible roles of both a controllable horizontal tail and a controllable shoulder
angle.
2.5 Discussion
Joint-driven wing morphing creates a two-dimensional morphing space (Fig. 2.8a, shaded
region) that encompasses a wide variety of aerodynamic properties available to a single wing.
In flight, a joint-driven morphing wing can follow any continuous joint trajectory through this
space. We defined a joint trajectory as a specific set of elbow and wrist angles obtained by
following any continuous line within the shaded region of Fig. 2.8a, and a joint extension
trajectory as a specific subset of joint trajectories that progress from left to right on Fig. 2.8a.
Note that it is also possible to implement these trajectories in reverse (flexion), however in this
work we focus on the effects of extension alone. Each specific joint extension trajectory will be
associated with differing gradients in aerodynamic properties where gradients can be visualized
by considering the contour lines in Fig. 2.5a, b, and 2.7. Contour lines were extracted from
the lift, pitching moment and pitch stability derivative models using the R package ContoureR
[172].
To identify joint extension trajectories that could be useful for a gull-inspired morphing
UAV, we examined four unique trajectories at a fixed 0° angle of attack while assuming a
quasi-steady extension that neglects any unsteady aerodynamic effects (Fig. 2.8a). The first
three trajectories were selected by individually extracting constant contour lines from the lift
(number 1), pitching moment (number 2) and pitch-stability derivative models (number 3)
(Fig. 2.8a). The final trajectory considered is the linkage trajectory (number 4) which is the
set of coupled elbow and wrist angles that are provided by the mechanical advantage of the
gull wing’s six-bar linkage system (Fig. 2.8a) [126, 133]. These angles were determined in a
previous study by fixing the humeral head of a specimen’s wing and manually applying a point
37
force within an approximate x-y plane (Fig. 2.1) to a point on the wrist; this causes both the
elbow and wrist joints to extend due to the linkage coupling [126]. This approach allows us
to extract the kinematics of the coupled linkage system that is caused by the displacement of
a single point, the point on the wrist in this case. Finally, we visualized the wing shape at the
start and end of each trajectory from the ventral view (Fig. 2.8c) and head-on view (Fig. 2.8d).
We found that both the constant lift and constant pitch-stability derivative trajectories de-
coupled stability characteristics from load production. First, the constant lift trajectory (Fig.
2.8a, number 1) created the highest absolute change in the pitch-stability derivative, exhibiting
a linear response (R2 > 0.99) throughout the extension (Fig. 2.8b and e). This linear trend
indicates that the control effectiveness remains constant, removing the need for a controller to
know the exact position of the wing. In addition, this trajectory had a minimal effect on the
pitching moment, where the instantaneous control effectiveness (instantaneous slope from Fig.
2.8e) of the pitching moment was below 5 × 10−4 /◦ until over halfway extended but increased
as the wing neared maximum extension. In all, extension along the first half of the constant
lift trajectory would allow a simple trajectory for a morphing UAV to adjust its static margin
without affecting the lift or pitching moment. This extension trajectory could allow a gull-
inspired morphing UAV to shift from a stable to an unstable configuration without affecting its
longitudinal position or orientation. Decreasing static stability may be useful when a flight en-
vironment becomes gustier because lower static stability reduces the strength of the inherently
developed pitching moment and may reduce path oscillations [128]. As a result, this morphing
trajectory may allow an active form of gust rejection.
The second trajectory that decouples stability from load production is the constant pitch-
stability derivative trajectory (Fig. 2.8a, number 3). This extension trajectory created a large
magnitude increase in lift and decrease in pitching moment both with linear responses (R2 >
0.99) throughout extension (Fig. 2.8b and e). This trajectory acts similar to a traditional sym-
metric flap deflection because a downward flap deflection increases lift and creates a nose-down
pitching moment, which can be counteracted with a controllable horizontal tail. Thus, as with
flaps this extension trajectory could be used to steepen the approach angle during landing. Also
similar to flaps that do not change the overall wing area, this trajectory avoids the undesirable
change to an aircraft’s handling qualities that is caused by a static margin shift [138]. However,
38
(a) 1 constant C L morph (c) (d) (e)
2 constant C m morph x ventral view 100
y head-on view Cmmorph,L = 0
3 constant dC m morph /dC L morph y
z 75
180 4 linkage trajectory CLmorph
1 end end 50
dCmmorph
160
Wrist angle (°)
0
0 20 40 60 80 increased:
100 - stability
- balance
0.0 75 - loads
end end 50
4
0.5 m
start start 25
reduced:
0 - stability
-2.5 - balance
0.6 m 0.6 m 0 20 40 60 80
- loads
1 2 3 4 Trajectory (°)
Figure 2.8: Constant joint extension trajectories allow variable longitudinal control
strategies. (a) The entire range of motion investigated (shaded region) overlaid with the four
investigated trajectories: 1) constant lift force, 2) constant pitching moment, 3) constant pitch-
stability derivative, and 4) linkage trajectory. (b) The total control effectiveness of the trajectory
(y-axis) is defined as the change in a parameter between the start and end per degree of exten-
sion along the entire trajectory arc length. Planform view (c) and head-on view (d) of the wing
shapes at the start of each trajectory overlaid with the shapes at the end. (e) The instantaneous
control effectiveness of each longitudinal quantity, represented by the instantaneous slope at
each point, is highly variable. Each quantity is scaled between its absolute minimum and max-
imum based on the results at 0° angle of attack. 0% represents the lowest loads, least stable
and least balanced configuration.
because the relationship between lift and pitching moment is directly affected by the location
of xCG , a successful implementation of this morphing trajectory in a gull-inspired morphing
UAV will need to include a detailed trade-off study investigating the most beneficial placement
of xCG .
Our results showed that minor variations off the identified trajectories recouples the stabil-
ity and loading characteristics. Consider the constant pitching moment trajectory (Fig. 2.8a,
39
number 2) which appears to be slightly askew from the constant lift trajectory (Fig. 2.8a, num-
ber 1). Yet, the constant pitching moment trajectory has a strong non-linear coupling between
the developed lift and static stability (Fig. 2.8b and e). Thus, precise joint angle control will
be needed for a UAV to gain the discussed benefits from the two decoupled trajectories. This
required precision increases the challenge of manufacturing an effective joint-driven morphing
wing.
Finally, we investigated extension along the linkage trajectory (Fig. 2.8a, number 4) and
found that this extension does not cause the largest change to any individual parameter but
does cause the largest magnitude change across all parameters compared to the other investi-
gated trajectories. Interestingly, there was a non-linear response, such that the instantaneous
control effectiveness differs substantially from the total control effectiveness (Fig. 2.8e and
b). Specifically, at the start of the extension trajectory CLmorph increases by 6.3×10−3 /° and
Cmmorph decreases by 2.2×10−3 /°, increasing the absolute load on the wing. Near the end of the
extension trajectory CLmorph decreases by 4.7×10−3 /° and Cmmorph increases by 2.2×10−3 /°,
decreasing the absolute load on the wing. Thus, extension alone allows a method to both
increase and decrease aerodynamic loads, solely dependent on the wing’s position along the
extension. Note that these instantaneous control effectiveness values are a larger magnitude
compared to those obtained by all other investigated trajectories (Fig. 2.8b and e). The strong
variability in the response possibly allows the linkage trajectory to serve many different func-
tions in flight such as initiating and maintaining complex manoeuvres. Moreover, due to the
mechanical advantage of the linkage system, following this trajectory requires input from only
a single actuator which would simplify the manufacturing process of such a wing. However,
the lift and pitching moment are strongly coupled to the balance and stability characteristics
for the linkage trajectory. As such, flight control using joint-driven wing morphing along the
linkage trajectory would be undesirable for human pilots without an additional controller to
account for the shifting static margin.
There are few comparably-sized engineered morphing wing aircraft with published longi-
tudinal characteristics. One great example is a goshawk-inspired drone with a maximum wing
40
span (1.05m) that is 87% of the scale of our largest wing (1.21m) [29]. This aircraft saw a min-
imal change in the lift and pitching moment produced between a swept and extended position
(with a furled tail) around 0° but, at higher angles of attack the extended wing had substantially
higher loads with an absolute variation up to roughly 0.5 in CL and 0.4 in Cm . Note that Cm
was non-dimensionalized by the mean aerodynamic chord which is smaller than cmax and thus
our range of pitching moment cannot be compared directly. Our morphing wing-body numeri-
cal results showed an absolute variation of 0.48 in CLmorph and 0.18 in Cmmorph at 0°. Further
investigation of our experimental results indicated that the absolute variation remains relatively
constant across the investigated angles of attack with the range only beginning to reduce be-
low -5°. This suggests that a joint-driven morphing wing UAV may effectively provide lift and
pitching moment control across a broader range of angles of attack when compared with planar
sweep-only morphing.
The preceding discussions of longitudinal stability and balance have assumed that the gull-
inspired morphing UAV would be controlled with open-loop stability (as is done when a glider
is statically stable). Instead, it is possible to use closed-loop stability to successfully fly while
in an unstable configuration [138]. For the gull-inspired morphing UAV, the xCG would have
to be shifted backwards by over 9.2 cm from the body quarter-chord to render the entire range
of motion unstable. In this configuration, the pitch-stability derivative will be positive and
thus, a negative zero-lift pitching moment would permit a trimmed position. In traditional
aircraft design this is called relaxed static stability. Such an aircraft benefits from improved
drag performance and maneuverability characteristics but requires a high degree of control
to avoid the potentially serious consequences of an unstable response to external inputs such
as gusts [138]. Our current results cannot comment on whether live birds utilize open-loop
or closed-loop control, but for a gull-inspired morphing UAV, closed-loop control offers an
alternative to installing a horizontal tail. In this case, the trim position would still need an
effective control method. This control could possibly be provided by the shoulder angle similar
to hang gliders [137, 140].
41
2.6 Limitations
It is important to note that our previous study of real prepared gull wings found the oppo-
site relationship between elbow angle and the pitch-stability derivative compared to the current
rigid wing results [128]. We expect that differences between the two studies may be caused
by feather flexibility, feather porosity, different wing alignments, and/or the inclusion of wing-
body interaction effects within the current study. Further investigation to understand the differ-
ences between a rigid 3D printed wing shape and a real gull wing will be necessary.
To this end, many assumptions were required to allow a targeted analysis of the effects of
elbow and wrist morphing and will require investigation in future studies to approach a general
understanding of how birds fly. For example, in this study we did not include variations in
velocity or Reynolds number. Birds use an intermediate Reynolds number and it is therefore
possible that shifting into a lower or higher regime could have a measurable effect on the lon-
gitudinal characteristics of an avian-inspired wing [122]. Additionally, it will be necessary to
evaluate the coupled role of shoulder joint control with elbow and wrist morphing to develop a
holistic understanding of flight control due to avian wing morphing. Further, our work assumed
quasi-steady extension, however in reality, birds can manipulate their joints very quickly. This
quick motion could result in induced flow along all major directions. A detailed mechanistic
study is warranted in the future to determine the presence and the role of specific unsteady aero-
dynamic effects. Finally, we only investigated a single species, but birds have a broad range of
species diversity, each of which may offer unique insights on how to efficiently design UAVs.
In particular, we expect different control effectiveness values between different bird species
due to variable wing range of motion, linkage structures and overall geometry [122, 126, 148].
2.7 Conclusion
We investigated the potential benefits of gull-inspired joint-driven wing morphing for future
UAV applications. First, we determined a set of simplified wing shapes across the range of
motion of the elbow and wrist used by gliding gulls. Next, we used a numerical general lifting-
line model (MachUpX) validated with wind tunnel experiments to determine the longitudinal
42
characteristics of the wing-body configurations. Our results showed that wings with the highest
load production had low wrist and high elbow angles and were associated with wing tip wash-
in. Additionally, although the inherent response to an external disturbance for all wing-body
configurations was stable, we found that a controllable horizontal tail or shoulder angle would
be necessary to successfully provide open-loop control. Importantly, we found that the elbow
and wrist angle could provide a reliable method to control the lift, pitching moment, and overall
static margin but would not be sufficient to control the zero-lift pitching moment alone.
Our study revealed that the two-dimensional morphing space allowed by the elbow and
wrist joints permits a wide variety of flight control strategies. In particular, we identified two
trajectories that decoupled longitudinal static stability and longitudinal load production. One
trajectory (Fig. 2.8a, number 1) linearly adjusts static stability without affecting the load pro-
duction and the other (Fig. 2.8a, number 3) linearly adjusts the lift and pitching moment
without affecting stability, in a manner similar to an aircraft flap that does not change the wing
area. Moreover, the identified linear response is highly advantageous for a simplified controller
design. However, we found that a unique but similar trajectory (Fig. 2.8a, number 2) recouples
the loads and stability, suggesting that precise control of the elbow and wrist would be neces-
sary to realize these aerodynamic benefits in a UAV. Finally, the linkage trajectory (Fig. 2.8a,
number 4) afforded by the gull’s musculoskeletal linkage system yielded the highest instanta-
neous control effectiveness of all our investigated trajectories and represents a simple actuation
trajectory that can quickly adjust the longitudinal characteristics. However, the load produc-
tion and stability are highly coupled for this trajectory and other control mechanisms would
be required to negate this effect. In all, investigation of these unique trajectories highlights the
multifunctional capabilities of gull-inspired joint-driven wing morphing.
Despite the identified aerodynamic benefits of a joint-driven morphing wing, a major chal-
lenge for any bio-inspired UAV is to design an efficient actuation mechanism that can realize
the proposed benefits in practice. In the past, a non-planar wing design that was indirectly
inspired by gulls, the HECS wing (discussed in the Section 2.2) was shown to yield mini-
mal aerodynamic benefits when it was actively morphed into its furled configuration [173]
despite promising rigid model results [149]. This emphasizes the multidisciplinary challenges
associated with effective morphing wing design. Successful implementation of gull-inspired
43
joint-driven morphing wings will require detailed structural analyses, flight tests, and multidis-
ciplinary investigations to determine if the benefits identified within this study could effectively
and efficiently be realized in a morphing wing UAV.
Do the benefits provided by joint-driven wing morphing outweigh engineered morphing
wing designs? Our results show that gull-inspired joint-driven wing morphing creates a sim-
ilar magnitude of control effectiveness as an equivalent aircraft with a sweeping mechanism,
but with the added multifunctional capabilities permitted by the variable joint extension tra-
jectories. This is especially promising because we found that a joint-driven morphing wing
can produce a similar aerodynamic response to traditional flaps but would not be limited to
this singular functionality. Combined with future multidisciplinary investigations, we expect
that gull-inspired joint-driven wing morphing could provide a future generation of UAVs the
unique ability to adapt on the fly by morphing along the specific joint trajectory that realizes
the desired aerodynamic characteristics.
44
Chapter 3
3.1 Summary
Birds morph their wing shape to accomplish extraordinary maneuvers [174–177], which
are governed by avian-specific equations of motion. Solving these equations requires infor-
mation about a bird’s aerodynamic and inertial characteristics [138]. Avian flight research
to date has focused on resolving aerodynamic features, whereas inertial properties including
center of gravity and moment of inertia are seldom addressed. Here, we used an analytical
method to determine the inertial characteristics of 22 species across the full range of elbow and
wrist flexion and extension. We find that wing morphing allows birds to substantially change
their roll and yaw inertia but has a minimal impact on the center of gravity position. With
the addition of inertial characteristics, we derived a novel metric of pitch agility and estimated
the static pitch stability, which revealed that the agility and static margin ranges are reduced
as body mass increases. Surprisingly, our results provide quantitative evidence that evolution
selects for both stable and unstable flight, a result that contrasts with the prevailing narrative
that birds are evolving away from stability [178]. This comprehensive analysis of avian iner-
tial characteristics provides the key features required to establish a theoretical model of avian
maneuverability.
45
3.2 Background
There is currently no theory that provides hypotheses to guide studies of avian maneuver-
ability. This is not due to a lack of physical understanding; maneuverability can be broadly de-
fined as a bird’s ability to change the magnitude and/or direction of its velocity vector [19, 20].
Like comparable UAVs, a bird’s flight dynamics, and thus maneuverability, are dictated by its
governing equations of motion. For example, aircraft dynamics depend on a minimum of six
equations; three translational and three rotational that can be derived from Newton’s second
law and it’s rotational counterpart [138, 179]:
d(mv)
F= (3.1)
dt
d(Iω)
M= (3.2)
dt
Where v is the velocity vector and ω is the angular velocity vector. These equations can be
dv dω
combined to solve for a flyer’s acceleration (translationally: dt
and rotationally: dt
), but this
requires knowledge of both the aerodynamically informed external forces (F) and moments
(M) as well as the inertial characteristics including the mass (m) and moment of inertia tensor
(I). However, avian inertial characteristics are not currently available with sufficient breadth or
resolution.
Therefore, avian flight maneuverability is often evaluated experimentally by tracking in-
dividuals to measure accelerations during observed maneuvers [176, 177]. However, tracking
data do not provide a bird’s maximal maneuvering capabilities or allow extrapolation to un-
observed behaviors. Determining these attributes requires a robust and general framework for
maneuverability, equivalent to the maneuverability equations for aircraft [19, 180]. Obtaining
generalizable data is further complicated because aerodynamic and inertial characteristics vary
substantially within and among species, and even dynamically for an individual bird [126, 174].
For example, birds can initiate maneuvers by changing the orientation and shape of their wings,
body and tail, known as morphing [20, 137, 181]. To progress towards a theoretical formula-
tion of avian maneuverability, there has been a marked and justifiable focus on resolving the
aerodynamic characteristics of a bird in flight [127, 128, 141]. However, studies often overlook
46
the equally essential inertial properties (Fig. 3.1a) or use static morphology approximations for
individual species [51, 137, 182–184]. Here, we fill this gap by investigating the variable in-
ertial characteristics of flying birds to provide the necessary next step towards establishing a
general framework of avian maneuverability.
Another challenge to solving a flying bird’s equations of motion is how to properly for-
mulate the equations. For example, the equations can be simplified by defining the origin at
the center of gravity (Fig. 3.1a), which is equivalent to the center of mass within a constant
gravitational field [179]. If the center of gravity moves substantially relative to the body, addi-
tional terms in the equations are required to properly capture flight dynamics [180]. Physically
shifting a bird’s morphology will shift the center of gravity but, it is not currently known how
much the center of gravity moves as a bird morphs. In addition, the rotational inertia, quanti-
fied by the mass moment of inertia tensor (I) about the origin will also be affected by morphing
(Fig. 3.1a and b). This symmetric matrix describes the body mass distribution where diagonal
elements quantify the distribution relative to the major axes (Ixx : roll, Iyy : pitch, Izz : yaw) and
off-diagonal elements quantify distribution within the three major geometric planes (only Ixz
is non-zero for symmetric configurations, Fig. 3.1a) [179].
humeral
(a) head
NP CG
pitch
roll
yaw
(b) lift
drag
aerodynamic
inertial
Figure 3.1: Inertial properties must be determined to quantify avian maneuverability. (a)
A bird’s center of gravity (CG) is the position about which weight is equally distributed, and
the neutral point (NP) is where aerodynamics forces can be modeled as point forces and the
pitching moment is independent of angle of attack. The moment of inertia components (I) are
obtained by integrating differential mass elements (dm) over the entire bird. (b) Flight dynam-
ics are affected by adjusting either inertial or aerodynamic characteristics. (c) We modeled
birds as a composite of simple geometric components.
47
We calculated a bird’s center of gravity and I to evaluate avian maneuverability through
the lens of agility and static stability. Agility encompasses a bird’s ability to perform linear
accelerations (axial agility) and angular accelerations (torsional agility) [20], and depends on
both the center of gravity and I [138]. In contrast, static stability refers to the initial tendency
to return towards an equilibrium after a disturbance [128]. We quantified static pitch stability
with the static margin, which is the distance between the center of gravity and neutral point
(NP, Fig. 3.1a) [138, 141]. If the neutral point is aft of the center of gravity, the static margin
will be positive and thus stable. Often, stability is inversely related to agility because larger
maneuvering forces and moments are sometimes necessary to overcome stabilizing forces and
moments [128].
48
(Fig. 3.3, Section 3.3.4).
(a) (b)
dorsal 0.0 y (m) 0.2 0.0 y (m) 0.1
origin head-on view
0.0
z (m)
joint CG primary feather CG muscle/bone CG
0.1 peripheral CGwing secondary feather CG skin/coverts CG
elbow
(c) 0.0 dorsal view (d)
wrist
Pigeon
x (m)
150
100
50
50 100 150
Elbow angle (°)
Figure 3.2: Elbow and wrist joint range of motion informed our analysis. (a) We modeled
birds as a composite of simple geometric components. Each component’s center of gravity
varies as a wing morphs from an extended (a and c) to a folded (b and d) configuration. (e)
Convex hulls showcase the range of motion (ROM) of the elbow and wrist for 22 species.
We obtained morphological data for 36 adult specimens representing 22 species (Fig. 3.4a)
from frozen cadavers acquired from the Cowan Tetrapod Collection at the Beaty Biodiversity
Museum (University of British Columbia, Vancouver, Canada). Sample size was a function
of the availability and quality of specimens from the museum as we could only rely on fully-
intact, well-preserved specimens. The cadavers were inspected to ensure adequate condition
and completeness, after which we measured the full body mass, wingspan, and body length.
49
Next, we disarticulated the wing at the shoulder joint, taking care to ensure that each wing’s
skin, propatagial elements, and feathers remained intact. One wing from each cadaver was
used to determine wing ROM and corresponding wing shape change (see following section).
The cadaver was further dissected to obtain length and mass measurements for the head, neck,
torso, wing components, legs, and tail. We obtained the center of gravity coordinates for the
torso (body without head, neck, tail, wings) by manually balancing the torso and measuring
the distance from the clavicle reference point to the balanced position. Note that because of
the preservation of the storm petrel specimens, we estimated the mass based on humerus bone
length and the torso center of gravity as being proportional to that of the gull. Finally, we
individually weighed and photographed each flight feather allowing geometric parameters to
be extracted using ImageJ software [186]. Refer to the publicly available data repository for
details on all assumptions used for extracting the morphological measurements. This study
consisted of a single experimental group and thus randomization and blinding was not neces-
sary.
To determine the wing ROM and corresponding shape change, we actuated the cadaver
wings throughout the full range of extension and flexion of the elbow and wrist joints by hand
(following methods established by Baliga et. al [126], Fig. 3.2e). We tracked the location of
ten 4-mm diameter, reflective markers (gray and white points in Fig. 3.2a-d) with automated
3D data capture at 30 frames per second via a four- or five-camera tracking system (OptiTrack:
NaturalPoint, Inc.). Using NaturalPoint, Inc. tools, each recording was calibrated to have less
than 0.5 mm overall mean reprojection error. Joint angles were calculated as the interior angle
defined by three key points: points 1, 2 (vertex), and 3 for the elbow, and points 2, 3 (vertex),
and 4 for the wrist (Appendix A).
50
running R (version 4.0.3) [188]. A high-level overview of the code methodology follows in
this section. Further details are provided in Appendix A, as each individual component of the
avian models required specific procedures and approximations.
To allow a generalized approach, we used a common methodology from mechanics to es-
timate the center of gravity and inertia components using simple geometric shapes [179]. We
elected to use as many elements as possible to allow the best resolution. For each species, we
first modeled the bird’s body without the wings as a composite of five components: head, neck,
torso, legs, and tail. To determine the inertial properties of the wings, we aligned each wing
configuration extracted from the ROM measurements so that the wrist joint was in line with
the shoulder joint along the y- and z-axes and so that the wrist joint was aligned with the first
secondary feather (S1) along the x-axis (extended wing: Fig. 3.2a and c; folded wing: Fig.
3.2b and d). Note that this positioning results in a different shoulder angle between each wing
configuration and wings with extremely low elbow angles and high wrist angles may be at
substantially different incidence angles than the body. Each wing was then modeled as a com-
posite of twelve components: bones (humerus, radius, ulna, carpometacarpus/digit, radiale and
ulnare), muscles (brachial, antebrachial and manus groups), skin, coverts, and tertiary feathers.
In addition, each primary and secondary feather was modeled and positioned individually as
a composite structure of five components: calamus, rachis (cortex exterior and medullar inte-
rior), and distal and proximal vanes. AvInertia permits a variable number of flight feathers.
With our methodology, a bird with 10 primaries and 10 secondaries that flies with an extended
neck will be represented by a composite model with 232 individual simple geometric shapes.
In our study, we investigated only symmetric wing configurations for a full bird and considered
the effects of a single wing independently. We assumed that anisotropic effects such as the air
space within the body would have a minimal impact on the overall center of gravity [189].
To calculate the final inertial characteristics of this composite bird, each component’s shape,
mass, and positioning was informed by its corresponding morphological measurements. We
began by determining the center of gravity and I for one of the basic geometric shapes with
respect to an origin and frame of reference that simplified the formulation of the center of
gravity and I for that shape. Next, AvInertia computed the mass-weighted summation of the
center of gravity of each object and shifted the origin to the bird reference point, located at the
51
center of the spinal cord when cut at the clavicle. The center of gravity was then transformed
into the full bird frame of reference, which is defined by Fig. 3.2a-d. We used the parallel
axis theorem and the appropriate transformation matrices to transform I to be defined about the
final center of gravity within the full bird frame of reference.
We validated our methodology by comparing the maximum rotational inertia about the roll
axis for a single wing (Ixxwing , origin at the humeral head) to data from previous experimental
studies that measured Ixxwing by cutting an extended wing into strips (Fig. 3.3) [190, 191]. Our
95% confidence intervals on the exponent of body mass marginally overlapped with Berg and
Rayner’s predictions [190] but were significantly lower than Kirkpatrick’s predictions [191].
However, Kirkpatrick used 10 wing strips while Berg and Rayner later found that at least 15
strips were necessary to minimize systematic error [190, 191]. Next, we directly compared
results for the pigeon (Columba livia), the only species in common between the studies, and
found Ixxwing (×104 ) was between 1.42 to 1.92 kg-m2 , which encompasses results from previ-
ous studies (1.72 and 1.83 kg-m2 ) [190–192]. The pigeon wing’s maximum center of gravity
position along the y-axis (yCG wing ) was only 3% of the half span more proximal than Berg
and Rayner’s measurement [190]. We expect minor differences because strip methods enforce
that all wing mass is contained within the x-y plane while AvInertia accounts for out-of-plane
10
−1
(kg−m2)
10
−2
10
−3
10
−4
Maximum
Kirkpatick
10
−5
Berg & Rayner
10
−6 AvInertia
Figure 3.3: Our results agreed with previous estimates of the wing’s moment of inertia.
The computed maximum Ixxwing was comparable to published estimates. 95% confidence in-
tervals visualized by transparent ribbons (n = 36 individual specimens).
52
morphology (Fig. 3.3).
All phylogenetically informed analyses were carried out using the time-calibrated Maxi-
mum Clade Credibility tree from Baliga et al.[126], which was pruned to the 22 focal taxa in
this study (Fig. 3.4a). To determine the linear trends with body mass, we fit first-order phyloge-
netic generalized linear mixed models (PGLMM) to the data using the R package MCMCglmm
[193] where the random effects are informed by the phylogeny (Table 3.1). These results will
be discussed in further detail in the following sections. All PGLMM models had priors speci-
fied with the inverse Wishart scaling parameters V = 1 and ν = 0.02 and used 1.3×107 Markov
chain Monte Carlo (MCMC) iterations. To determine the significance and effect of the elbow
and wrist on the center of gravity and I components, we independently fit first order inter-
active models to each specimens’ data with a constant scaling on the elbow and wrist angle.
We calculated the effect size of the elbow and wrist using the R package effectsize [194] and
independently fit first order interactive models to each specimens’ data with scaled and mean
centered elbow and wrist angles.
With our validated results, we first asked: what effect does the elbow and wrist ROM have
on a bird’s center of gravity when its wings are held symmetrically? Our results revealed that
the ROM had a minimal effect on the center of gravity position (Fig. 3.4b, opaque polygons).
The maximum shift along the x-axis and z-axis (xCG and zCG , normalized by the full bird’s
length) was 3% (great blue heron, Ardea herodias, 2.0 cm) and 2% (barn owl, Tyto alba, 0.7
cm), respectively (Fig. 3.4b). Despite the small magnitude, wrist extension consistently shifted
xCG forwards (p-values < 0.002) and the wrist angle explained a high amount of variance in
the data leading to a high effect size, quantified by partial eta-squared (η 2 ) [194, 195]. We
found partial η 2 was greater than 0.34 for all species (Fig. 3.4e). Similarly, elbow extension
53
Table 3.1: MCMCglmm outputs for inertial phylogenetic generalized linear mixed models
(PGLMM).
tended to shift xCG forwards, but its effect size varied across species. Both elbow and wrist
extension predominately shifted zCG dorsally, but the magnitude and effect size varied. We
could not differentiate the log-transformed mean xCG or zCG position from those expected if
birds were simply scaled by preserving all length scales, known as isometry (Fig. 3.4f, Table
3.1). As visualized by Fig. 3.4f, we cannot definitively exclude the possibility that the lower
95% confidence interval on xCG may be positive, which would indicate that xCG scales greater
than isometric predictions. However, multiple MCMCglmm runs returned an insignificant
result.
The small effect of the elbow and wrist on the center of gravity location led us to question
if this would carry over to shoulder joint motion as well. To obtain a conservative estimate,
we assumed that wings could rotate about the humeral head by 90° forwards, aft, up, and
down (Fig. 3.4b, transparent squares). This revealed that the maximum ∆xCG and ∆zCG shift
was 18% (10.9 cm) for the great blue heron, approximately sixfold greater than that achieved
with elbow and wrist morphing alone. Such a large center of gravity shift likely could not be
neglected when formulating the equations of motion. At the other extreme, the Lady Amherst’s
pheasant (Chrysolophus amherstiae) had a negligible shift of 1% (1.4 cm) with shoulder joint
motion. Across the full range of taxa, we found a significant positive relationship between
∆xCG due to shoulder motion and the ratio of maximum wingspan to body length (Fig. 3.4d,
54
(a) Mallard (b) (c) 0
(%)
(Anas platyrhynchos)
10
Canada goose −20 0
(%)
(Branta canadensis) (%) 0
−10
Himalayan monal
(Lophophorus impejanus) 0 30
(%)
Lady Amherst's pheasant
CG range due to
(Chrysolophus amherstiae) max CG range due elbow and wrist ROM
to shoulder rotation (per individual)
Silver pheasant (d)
(Lophura nycthemera)
25
(% of body length)
20
Pigeon 15
(Columba livia)
Black swift 10
(Cypseloides niger)
5
Common nighthawk
(Chordeiles minor) 0
Glaucous−winged gull
(Larus glaucescens)
Cooper's hawk 10
(Accipiter cooperii) Desnity 5
0
Sharp−shinned hawk
(Accipiter striatus) 0.0 0.5 1.0
Partial η2
Barn owl (f) null slope
20 (shoulder origin)
(Tyto alba) PGLMM
*(dorsal origin)
Northern flicker
(% of full length)
(Colaptes auratus) 10
Mean CG
Merlin
(Falco columbarius)
Peregrine falcon
(Falco peregrinus)
Belted kingfisher 2
(Megaceryle alcyon) (g)
(% of halfspan)
(Corvus corax)
(%)
(%)
Steller's jay
−10 0
10 0
(Cyanocitta stelleri) 10
80 60 40 20 0 −20 0 0 30 10−2 10−1 100 101
Time (mya) (%) (%) Body mass (kg)
Figure 3.4: A bird’s center of gravity is minimally affected by elbow and wrist flexion and
extension. (a) Time-calibrated phylogeny for 22 species (mya, million years ago). The elbow
and wrist ROM (opaque polygons, convex hulls) affect (b) xCG and zCG (over bar indicates
normalization by body length), and (c) yCG wing (over bar indicates normalization by maximum
half span). The center of gravity range is overlaid with the maximum bounds due to 90◦
shoulder rotation (transparent polygons), which (d) increase with increasing ratio of wingspan
to body length. e, Effect size (partial η 2 ) of elbow, wrist, and interaction on each center of
gravity component per specimen. The log-transformed mean values of (f) xCG , zCG ∗ (∗ denotes
the z position relative to the dorsal origin defined by Fig. 3.2a) and (g) yCG wing did not scale
with body mass as the phylogenetic generalized linear mixed model (PGLMM; solid line) did
not differ significantly from the null slope (dashed line). 95% confidence intervals visualized
on d, f, and g by transparent ribbons (n = 36 individual specimens).
55
Table 3.1). This trend suggests that proper modeling of flight dynamics for birds with wings
substantially longer than their body length will require an estimation of the expected center of
gravity shift to verify if a fixed center of gravity is an appropriate assumption.
Although the full bird’s center of gravity defines its symmetric flight dynamics, the wing-
only parameters can give insight into asymmetric configurations. We found that the elbow and
wrist ROM caused the center of gravity to shift along the y-axis (yCG wing , normalized by the
maximum half span) from 10% (black swift, Cypseloides niger) to 27% (American white peli-
can, Pelecanus erythrorhynchos) (Fig. 3.4c), where the most distal yCG wing was 28% (western
grebe, Aechmophorus occidentalis). Additionally, ∆yCG wing was positively associated with
the arm-to-hand wing ratio (Table 3.1), such that birds with longer hand wings than arm wings
(like the swift) would have a reduced capacity to shift the wing’s CG. The center of gravity shift
was largely driven by elbow extension (p-values < 0.001, partial η 2 > 0.51, Fig. 3.4e) whereas
the effect of the wrist varied across species. These results highlight a well-conserved proximal
location of the wing center of gravity across species. Contrary to a previous study [190], we
did not find that the log-transformed mean yCG wing differed from isometric expectations (Fig.
3.4g, Table 3.1).
The center of gravity is crucial to formulating the governing equations, but their solution
depends on a bird’s rotational inertia. Like the CG, we found that a bird’s rotational inertia (log-
transformed mean diagonal components of I) scaled isometrically with body mass (Fig. 3.5a,
Table 3.1). However, we found that elbow and wrist extension provided over an 11-fold Ixx
increase (heron) and 3-fold Izz increase (heron and owl, Fig. 3.5c). This capability was largely
driven by elbow extension (Fig. 3.5b), which had a significant effect on both Ixx (p-values
< 0.001, partial η 2 > 0.23; except Leach’s storm petrel, Hydrobates leucorhous) and Izz (p-
values < 0.009, partial η 2 > 0.45). The absolute values of Iyy and Ixz were minimally affected
by joint extension and the effect size varied substantially across species (Fig. 3.5b). We next
computed the contribution of each major body part to the overall rotational inertia for birds
with wings at maximum elbow and wrist extension (Fig. 3.5d, e, and f). Because the wings
were extended along the y-axis, this captures approximately the lowest wing contribution to Iyy
56
(a)10 −1 (b) Dorsal view
Effect size: elbow wrist interaction
yaw
10
−2 roll
pitch
tail
Mean (kg−m )
2
10
−3
torso
−4
neck
10
head
skin
10
−5
10 muscle
Desnity
5 feathers
10
−6
0
0 1
bones
10
−2
10−1
10 10 0.0 0.5 1.0
2
Body mass (kg) Partial η
(c) (d) (e) (f)
Mallard
Canada goose
Himalayan monal
Lady Amherst's pheasant
Silver pheasant
Western grebe
Pigeon
Black swift
Common nighthawk
American white pelican
Great blue heron
Leach's storm petrel
Glaucous−winged gull
Cooper's hawk
Sharp−shinned hawk
Barn owl
Northern flicker
Merlin
Peregrine falcon
Belted kingfisher
Common raven
Steller's jay
Figure 3.5: Wing morphing, specifically driven by the elbow, has a strong effect on roll
and yaw inertia components. (a) All log-transformed mean diagonal components scaled iso-
metrically with body mass (PGLMM model for each component; solid line). 95% confidence
intervals visualized by transparent ribbons (n = 36 individual specimens). (b) Elbow extension
has the largest effect on Ixx and Izz but joint angles were not strong predictors of Iyy or Ixz .
(c) The ability to adjust I varies substantially across species. At maximum wing extension,
the wing components (bones, feathers, muscle, skin) had the largest contribution to (d) Ixx
while body components (head, neck, torso, tail) played a larger role for (e) Iyy and (f) Izz .
Components are colored following the bird schematic.
but the highest wing contribution to Ixx . The percentage contribution of each body part varied
substantially across the species, but as expected the wings were responsible for the majority of
Ixx . These results indicate that elbow and wrist ROM provides substantial inertial control over
the roll and yaw axes (Ixx , Izz ), but less so for the pitch axis (Iyy ), although species-specific
differences were also apparent in our results. Incorporating the shoulder joint ROM would
increase the wing’s contribution to inertial pitch control.
57
3.5 Methodology - Aerodynamic characteristics
Next, to obtain information about the static stability of the birds, we investigated the aero-
dynamic characteristics associated with wing morphing. To estimate the neutral point of a
bird’s wing-body configuration, we leveraged our previous study on rigid gull wing-body con-
figurations across the in vivo range of motion of elbow and wrist flexion and extension (Chapter
2)[141]. In this previous study, we extracted the neutral point of the wing-body configurations
by fitting a linear model to the change of the pitching moment with the lift force [138, 140]
and provided the morphological information about the associated wing shapes. We used the
same wing orientation as the current chapter. The aerodynamic results were calculated using a
numerical lifting line model which was validated with wind tunnel tests on 3D printed wings.
As it was not feasible to replicate this analysis for each species in this study, we investigated if
there was a metric that could be used to appropriately estimate the neutral point using morphol-
ogy alone. We assumed that the bird’s neutral point could be approximated by the wing-body
configuration neutral point. This approximation is appropriate if the wing produces the major-
ity of the lift as is expected with a furled tail [196], but Section 3.7.5 discusses the implications
of incorporating the tail.
Our approach was informed by traditional aerodynamic theory which predicts that the aero-
dynamic center of a 2D thin airfoil will be at the quarter-chord location [197]. This result can
be extended to lifting line theory for steady flight conditions (commonly used in gliding flight)
or blade element theory for revolving airfoils (commonly used in flapping flight). However, 2D
thin airfoil approximations do not hold for the thick airfoils known to be used on the proximal
wing sections of bird wings[158] nor for 3D wing shapes [197]. Although advances in analyt-
ical methods have resulted in mathematical relationships that account for constant taper ratios
or sweep [198, 199], little information exists for wing shapes as complex as bird wings that
have substantial and often nonlinear distributions of geometric twist, taper, sweep, and dihe-
dral. Therefore, we investigated six different chord-based metrics[140, 199, 200] to establish
58
which would best approximate the neutral point. The origin for all metrics within this study
was set at the shoulder joint. This investigation included:
1. Root chord. A simplistic approach that estimates the neutral point to be at the quar-
ter chord of the root chord. This approach cannot account for a neutral point shifting
forwards of the root leading edge.
Sw
cm = , (3.3)
b
Where Sw is the total wing area and b is the wingspan. Next, we located the most interior
span section that had a chord equal to cm . At that span section we extracted the leading
edge position along the x-axis and added it to the quarter chord of cm while accounting
for any wing twist at this span section. This final value was taken as the quarter chord
position of the mean chord.
3. Mean projected chord. A similar approach as the mean chord but the wing area used is
that calculated by projecting the wing periphery onto the x-y plane.
4. Mean aerodynamic chord. A similar approach as the mean chord but the mean aerody-
namic chord is calculated as [199]:
R b/2
c(y)2 dy
M AC = R0 b/2 , (3.4)
0
c(y)dy
Where c(y) is the chord length along the span as a function of the location on the y-
axis. The mean aerodynamic chord is calculated numerically by disctretizing each wing
along the y-axis into 1000 segments. All integral equations that follow used the same
discretization.
5. Centroid area chord. In this approach we first numerically calculated the position of
59
the wing’s centroid along the y-axis as [199]:
R b/2
c(y)ydy
ycen = R0 b/2 (3.5)
0
c(y)dy
Then, at this span location we calculated the chord of the wing. Similar to the approach
in the mean chord we next determined the quarter chord and leading edge position.
6. Standard mean chord. This method numerically calculates the quarter chord position
directly as [199]:
R b/2
0
c(y)xc/4 (y)dy
x̃c/4 = R b/2 , (3.6)
0
c(y)dy
Where xc/4 (y) is the quarter-chord location as a function of the location on the y axis.
For each of the output quarter chord positions from the six different metrics, we normalized
by the maximum root chord (crmax ) of the specimen. This normalization ensures that the result
could be scaled for different sized individuals. To assess the fit of each normalized metric to
the gull’s measured neutral point we fit log-transformed linear models in R:
|xN Pwb | |x? |
ln ∼ A ln + B. (3.7)
crmax crmax
Note that for all configurations investigated in our previous study, the neutral point had a
negative position on the x-axis (aft of the shoulder joint), which allowed us to take the absolute
value of the data. Surprisingly, the best fit to our data was the standard mean chord (metric 6)
as shown by a low model offset (B) and higher adjusted R2 (Table 3.2). The other mean chord
parameters (metrics 2 and 3) were a particularly bad fit to our data because often the mean
chord was located distally. In that case, folding the wrist caused a substantial aftward shift of
the estimated quarter-chord location. With this information, we calculated the relationship for
every other species in our study with the exponent of 0.8 and assuming that e−0.052 = 0.949 ≈ 1
as: 0.8
|x̃c/4 |
|xN Pwb |
≈ . (3.8)
crmax crmax
We imported this relationship into our comparative analysis and computed the quarter chord
position of the standard mean chord for each specimen and configuration. We checked if the
60
Table 3.2: Linear model fit results for each investigated chord metric.
quarter-chord position was in front of the shoulder and if so we switched the signs of the output.
This essentially reflects the exponential trend that was established for the gull wings into the
positive neutral point region.
In all, this result provides insight into the aerodynamic implications of morphing however
this was informed by an aerodynamic analysis of only one species for rigid wing-body config-
urations. It will be important for future studies to account for inter-specific differences as well
as a neutral point shift due to flexibility and porosity.
We next asked if inertial characteristics could be used to estimate a bird’s pitch agility.
However, because both inertia and aerodynamics are fundamental to flight dynamics, we first
used aerodynamic theory and data from a rigid gull wing [141] to obtain an estimate for the
neutral point, and thus the static margin for each configuration. We developed a pitch agility
metric that estimates the change of the angular acceleration about the y-axis (∆q̇, known as the
time rate of change of the pitch rate) due to a degree change in the angle of attack (∆α) as:
0.8
x 2
crmax − xCG (m0.12 ) Smax
ec/4
∆q̇ c rmax
∝ . (3.9)
∆α Iyy
Where m is the body mass, crmax is the maximum root chord for the specimen, Smax is the
maximum single wing area for the specimen and xCG is the center of gravity position on the
x-axis measured from the humeral head. This equation was derived beginning from the rigid
aircraft y-axis rotational equation of motion assuming a symmetric configuration undergoing
61
small disturbances [138]:
∆M = Iyy ∆q̇. (3.10)
From this equation, we estimated the change in pitching moment (∆M ) with a Taylor
series expansion method assuming that the largest effect is due to angle of attack and then
non-dimensionalized as follows [138]:
∂M
∆M = ∆α (3.11)
∂α
1 2 ∂CM
= ρU∞ (2Smax )crmax ∆α (3.12)
2 ∂α
1 2 ∂CM ∂CL
= ρU∞ (2Smax )crmax ∆α. (3.13)
2 ∂CL ∂α
Where U∞ is the freestream scalar velocity, and CL and CM are the coefficients of lift and
pitching moment, respectively. Because the pitching moment slope ( ∂C
∂CL
M
) is proportional to
static margin [141, 197], we estimated each configuration’s neutral point using equation 3.8
(see Section 3.5). With the estimated neutral point, we calculated the static margin as:
0.8
x
ec/4
xCG − crmax
crmax
∂CM
static margin = − = . (3.14)
∂CL crmax
For the pitch agility metric, we incorporated a previously established allometric scaling
[117] of cruise velocity (U∞ ∝ m0.12 ). We assumed a constant air density (ρ) and constant lift
slope ( ∂C
∂α
L
) across species to obtain the final proportional relationship as:
" 0.8 ! #
x
ec/4 2
∆M ∝ crmax − xCG m0.12 Smax ∆α. (3.15)
crmax
This result was then returned to equation 3.10 and rearranged to obtain the pitch agility metric
as seen in equation 3.9. Note that agility in a stable configuration indicates that the developed
acceleration would tend to return the bird towards an equilibrium position.
62
(a) (b)
15 high agility 1.5
petrel & unstable
Figure 3.6: Modern birds exhibit highly variable pitch agility characteristics. We derived a
pitch agility metric that highlights (a) that heavier birds are less agile even when (b) normalized
as if the birds are flying at the same speed and have the same body length. Maximum and
minimum values for each individual due to elbow and wrist ROM are plotted.
63
Table 3.3: MCMCglmm outputs for agility and stability phylogenetic generalized linear
mixed models (PGLMM).
To investigate the phenotypic optimum of the pitch agility and stability traits, we indepen-
dently fit both Brownian motion (BM) and Ornstein Uhlenbeck (OU) models to the absolute
data using the R package geiger [203]. We assumed that all species belong to the same regime
Figure 3.7: The elbow and wrist angle configurations that yielded the maximum and min-
imum static margin for each species. The ROM investigated for each species with the maxi-
mum static margin and minimum static margin identified with a black diamond on each species.
Note that the diamonds are colored by the static margin.
64
and thus, fit single-peak evolutionary models. We found that four species were entirely stable,
one species was entirely unstable, and 17 species had the capacity to shift between stable and
unstable flight (Fig. 3.8b and e). Using these data, we found that an Ornstein Uhlenbeck (OU)
model was significantly favored over a Brownian motion (BM) model, for both the maximum
(∆AICc = -8.24, Fig. 3.9b) and minimum static margin (∆AICc = -5.01, Fig. 3.9c), where
(a) 60 swift
(b) min max
stable Mallard
Static margin (% of c rmax )
−50 0 50 15 13 11 9 7 5 3 1
higher lower
unstable neutral stable
agility agility
Figure 3.8: Evolution selects for both pitch stability and instability, but modern birds ex-
hibit highly variable pitch agility and stability characteristics. (a) We found that heavier
birds have a reduced static margin range. Maximum and minimum values for each individual
due to elbow and wrist ROM are plotted. (b) The investigated species exhibited a wide variety
of static margins and absolute pitch agility. Dot color and size represents the mean maximum
and minimum value for each species. An Ornstein Uhlenbeck model provided evidence of
selection pressures acting on an unstable minimum (dashed line: min. θsm ) and a stable maxi-
mum (dashed line: max. θsm ) static margin, and xCG (θCG ). This xCG position is (c) stable if
the neutral point is behind this position and (d) unstable if the neutral point is in front of this
position.
65
AICc is the Akaike information criterion with correction for small sample sizes. Further, we
found that the optimal static margin phenotype (θsm ) was stable for the maximum static margin
(26% of crmax , strength of selection (αOU ) = 0.53, variance (σ 2 ) = 14.2 × 10−3 ) whereas the
optimal phenotype for the minimum static margin was unstable (−15% of crmax , αOU = 0.06,
σ 2 = 2.7 × 10−3 ) (Fig. 3.8b). This suggests that evolutionary pressures act to maintain birds’
ability to transition between stable and unstable flight. The strength of selection (αOU ) was
relatively low, but our results were robust to measurement errors (details in Sections 3.7.2 and
3.7.3) and to a preliminary estimation of a neutral point shift due to the tail (details in Section
3.7.5). Further, an OU model was a good fit for the mean xCG such that the phenotypic opti-
mum (θcg ) was 10% of the body length aft of the humeral head (∆AICc= -8.23, αOU = 0.11,
σ 2 = 0.1 × 10−3 , Fig. 3.9a). The stability of this center of gravity position depends on the
location of the neutral point (Fig. 3.8c and d).
Because of the smaller sample size of our study [204], we ran a Monte Carlo simulation
(n = 5000) with the R package pmc [205] to validate that selecting the OU model over the
BM model was appropriate (Fig. 3.9). This method returned a distribution of likelihood ratios
(twice the difference of the maximum log likelihood for each model) when the traits have been
simulated n times under each model. These distributions are then compared to the observed
likelihood ratio (black dashed vertical lines in Fig. 3.9). For details refer to Boettiger et. al
[205]. We found that the likelihood ratio predicted by a BM model was more extreme than
the observed ratio for the minority of simulations (xCG : 0.2%, maximum static margin: 0.1%,
minimum static margin: 1%). Further we had sufficient power to differentiate the two models
as the majority of the simulations under the OU model fell outside of 95th percentile of the BM
distribution (xCG : 73.8%, maximum static margin: 77.2%, minimum static margin: 67.2%).
95% confidence intervals were constructed for each reported metric of each trait (Table 3.4).
Together these results provide confidence that the observed likelihood ratio of each trait is more
likely to occur under an OU model than a BM model.
66
(a) 1.5
1.0
Density
0.5
0.0
(b) 1.5 maximum static margin
1.0
Density
0.5
0.0
(c) 1.5 minimum static margin
1.0
Density
0.5
0.0
0 10 20 30
Likelihood ratios
observed likelihood ratio
Likelihood ratios with data simulated by:
Brownian motion (BM) model
Ornstein Uhlenbeck (OU) model
Figure 3.9: A power analysis confirmed the validity of Ornstein Uhlenbeck (OU) models
for three key traits.We used a Monte Carlo-based method to investigate if our phylogeny pro-
vides support for the use of the OU model. This returns the distribution of likelihood ratios
under each model for (a) xCG , (b) maximum static margin, and (c) minimum static margin.
Comparing the likelihood ratios distributions produced under both a Brownian motion (BM)
model (gray) and OU model (yellow) to the observed likelihood ratio (dashed black line) re-
vealed that the OU model was a better fit for our three key traits (see methods).
Because both the pitch agility and stability metrics directly depend on xCG , we investigated
the sensitivity caused by shifting the combined torso and tail center of gravity forwards and aft
67
Table 3.4: 95% confidence intervals on the OU metrics reported for each investigated
trait.
by up to 15% of the torso. Note that for some species there was a physical limit to the ability to
relocate the center of gravity while maintaining the known morphological properties and if the
shifted distance was larger than 4cm we removed it from the analysis as that was assumed to be
an overestimate. The final estimated shift of the relative maximum and minimum static margin
is shown in Fig. 3.10. This sensitivity analysis revealed a minor effect on the parameters.
In addition, we wanted to investigate the potential effect of error in our measured center of
gravity metric on our key evolutionary results. To this end, we used a custom bootstrapping
68
code (n = 5000) and randomly sampled (with replacements) from each specimen’s center of
gravity error range used for the sensitivity analysis to recalculate the mean value of the mini-
mum and maximum static margin for each species. With each of these new trait distributions,
we re-fit an OU model and extracted the optimal phenotype (Fig. 3.11). We found that even
allowing for this substantial center of gravity error, all minimum static margin cases had an un-
stable optimum and all maximum static margin cases had a stable optimum (Fig. 3.11). Note
that this analysis is equivalent to accounting for the same magnitude shift in the neutral point
with a fixed center of gravity as well as accounting for possible inter-specific variation within
the error bounds shown in Extended Fig. 3.10.
unstable stable
50
minimum static margin
stability boundary
Density
25
0
−0.30 −0.15 0.00 0.15 0.30
Phenotypic optimum, θ
Figure 3.11: Bootstrapping our results within the conservative center of gravity measure-
ment error supported our results. We used a Monte Carlo method to investigate the impact
of center of gravity measurement error on the phenotypic optima for the maximum static mar-
gin (green) and minimum static margin (purple). Our results for a stable maximum and an
unstable minimum are confirmed with over a 95% confidence interval.
We verified that our key static margin findings were not substantially affected by the expo-
nent selected for the neutral point (Eqn. 3.8) by performing a sensitivity analysis. We found
that despite varying the exponent from 0.7 to 1.1, all the optimal phenotypes for maximum
static stability were stable whereas the optimal phenotypes for minimum static stability were
unstable. Specifically, as the scaling parameter increased, we found that the optimal phenotype
for the maximum static margin shifted from 31% to 15% of the maximum root chord and the
69
minimum static margin shifted from -13% to -19%. Thus, increasing the exponent effectively
serves to shift the results towards instability but our evolutionary findings remain supported. As
the gull represents the only species with which we are able to estimate the relationship between
geometric parameters and the neutral point across the full range of flexion and extension, we
selected to proceed with the gull-informed exponent of 0.8.
CLαt
d
xN P = xN Pwb − 1− ηt VH cm (3.16)
CLα dα
lt St
VH = (3.17)
cm Sw
Note that mean chord (cm ) introduced in equation 3.17 is canceled out in equation 3.16.
For plotting purposes, we selected to use each individual specimen’s mean value of cm across
all morphed configurations. This allows a comparable metric to the traditional aircraft metrics
published by Raymer [206]. We estimated that the tail aerodynamic center was at 25% of the
tail length, although future work will need to account for the expected shift due to low tail
aspect ratio. Figure 3.12 shows the estimated tail volume per individual as a function of the
70
101
10−2
10−3
Figure 3.12: Tail volume coefficients as a function of the body mass for the investigated
specimens.
overall body mass. The range for each specimen is largely caused by the changes in the total
wing area (Sw ). Of note, traditional aircraft tail volume coefficients usually vary between 0.4
(black horizontal line, Fig. 3.12) to 1 and many of these bird tail volumes are substantially
lower [206, 207].
Next, we must estimate the factors that multiply the tail volume coefficient in equation
3.16. First, it is known that the dynamic pressure at the tail will be lower than that at the
wing due to the resultant wing wake, which indicates that ηt is less than one [138]. Next,
because the tail will have a substantially lower aspect ratio than the wing while it is furled,
CLαt
we estimated that the lift-slope of the tail will be lower than that of the wing (i.e. CLα
is less
d
than one). Finally, dα
is also always below one due to the downwash from the main wing
[138, 206]. For traditional aircraft in subsonic flight, this value decreases as the wing aspect
ratio and taper ratio increase and as the distance between the tail aerodynamic center and the
center of gravity increases [206]. Each of these characteristics varies substantially across bird
d d
species. Because dα
is subtracted from one in equation 3.16, decreasing dα
will effectively
increase the tail contribution to the neutral point shift. Collectively, because all three discussed
multiplying factors are expected to be less than one, we selected each value to be 0.9 for birds:
CLαt d
CLα
(1− dα )ηt ≈ 0.73. We selected 0.9 as it is expected to overestimate this multiplying factor,
71
which will return the most aft neutral point shift and provide a highly conservative view on the
validity of our evolutionary results.
With these inputs we solved equation 3.16 and 3.14 to obtain the maximum and minimum
static margin for each specimen and then calculated the mean value for each species. Next,
we fit a Ornstein Uhlenbeck (OU) model to the data following the similar procedure detailed
in Section 3.7. This analysis revealed that even when accounting for the tail’s effect on the
neutral point there is evidence that evolution selects for a stable maximum static margin (38%
of crmax , αOU = 2.718, σ 2 = 0.255) and an unstable minimum static margin (−6% of crmax ,
αOU = 0.395, σ 2 = 0.022). As expected, the phenotypic optimum values of both the maximum
and minimum static margin models shifted towards increased stability, but there was evidence
of stronger selection pressure (αOU ) than in the wing-body configurations alone. As with our
key results, this again suggests that birds have the ability to shift their neutral point in front
of their center of gravity to balance the positive tail lift that is required for weight support in
slow gliding flight [208]. It is important to highlight that we expect inter-specific variation
within the multiplying parameters and that the selected value substantially overestimates the
tail’s contribution and likely results in a more stable output. In all, these results are expected to
provide a preliminary estimation of the tail’s contribution.
3.8 Limitations
It is important to highlight that further work is required to incorporate the inter- and intra-
specific aerodynamic capabilities, shoulder and tail ROM, and in vivo configurations to defini-
tively confirm the optimal phenotype(s) for static pitch stability. We expect that the shoulder
joint will enhance the available pitch control and the ability to shift between modes due to
an increased static margin range; the extent of this enhancement will depend on each species’
shoulder ROM [30, 138, 202]. Future work is also required to extend this analysis into the
roll and yaw axes to discuss lateral agility and stability, which will need to account for aerody-
namic and inertial coupling [138]. Finally, 23% of the species in our study were unable to shift
between stable and unstable modes with the elbow and wrist alone, and thus there are many
combinations of stability characteristics within modern birds.
72
3.9 Conclusion
Although studies have suggested that modern birds may be capable of stabilized flight
[128, 137, 141], it is widely believed that birds have evolved to be unstable in pitch to enhance
maneuverability [178]. Our results offer a new perspective on the evolution of avian flight:
evolutionary pressures may be maintaining the ability to shift between stable and unstable
configurations. Elbow and wrist flexion and extension alone offers birds the capacity to shift
between these pitch stability modes. But, if and when a flying bird does shift between these
modes remains to be seen. As highlighted by Thomas and Taylor [137], dynamically switching
between stable and unstable modes likely requires substantially different control algorithms
and thus, switching between these modes would necessitate a complex flight control system.
Further, our findings offer insight on how birds perform slow glides with positive tail lift [208].
By maintaining the capacity to relocate the wing-body neutral point in front of the CG, birds
may achieve an equilibrium, albeit unstable, flight condition.
In summary, our results revealed that elbow and wrist ROM had a small relative effect
on the center of gravity location and pitch inertia, but had a substantial effect on the roll and
yaw inertia. Although inter- and intra-specific variation was apparent, we found that the mea-
sured range of wrist and elbow motion alone is sufficient to enable switching between stable
and unstable flight in 17 out of 22 bird species. Further, an evolutionary analysis showed
that the phenotypic optimum maximum and minimum static margin supports the ability to
transition between stable and unstable flight, suggesting the need for a complex flight control
system. Collectively, investigating the inertial characteristics of flying birds throughout elbow
and wrist ROM brings us one step closer to establishing a fundamental theory to quantify and
then evaluate avian maneuverability.
73
Chapter 4
4.1 Summary
Birds perform astounding aerial maneuvers by actuating their shoulder, elbow, and wrist
joints to morph their wing shape. This maneuverability is desirable for similar-sized UAVs
and can be analyzed through the lens of dynamic flight stability. Quantifying avian dynamic
stability is challenging as it is dictated by aerodynamics and inertia, which must both account
for birds’ complex and variable morphology. To date, avian dynamic stability across flight
conditions remains largely unknown. Here we fill this gap by quantifying how a gull can use
wing morphing to adjust its longitudinal dynamic response. We found that it was necessary to
adjust the shoulder angle to achieve trimmed flight, and that most trimmed configurations were
longitudinally stable, except for configurations with high wrist angles. Our results showed that
as flight speed increases, the gull could fold its wings or sweep its wings backwards to trim.
Further, a trimmed gull can use its wing joints to control the frequencies and damping ratios
of the longitudinal vibrational modes. We found a more damped phugoid mode than similar-
sized UAVs, possibly reducing speed sensitivity to perturbations such as gusts. Although most
configurations had controllable short period flying qualities, the heavily damped phugoid mode
indicates a sluggish response to control inputs, which may be overcome while maneuvering by
morphing into an unstable flight configuration. Our study shows that gulls use their shoulder,
wrist, and elbow joints to negotiate trade-offs in stability and control and highlights the path
forward for designing UAVs with avian-like maneuverability.
74
4.2 Background
Imagine UAVs performing social aerial aerobatics like ravens [209], rapidly diving like
gannets [210], and skillfully maintaining their position in high wind and gusty conditions like
kestrels and gulls [37, 211]. These nature-documentary-worthy feats often exceed the maneu-
verability of modern comparable UAVs, especially non-rotary designs [129]. The ability for
UAVs to effectively maneuver is becoming increasingly important as UAVs are more often
operating close to or within crowded environments, such as urban centers [31, 212].
To determine how to best improve UAV maneuverability based on insights from birds, we
must first quantify maneuverability. Maneuverability is broadly defined as the ability to change
the magnitude and/or direction of a flyer’s velocity vector [19, 20]. Although, there are mul-
tiple ways to evaluate flight maneuverability, many traditional methods quantify an aircraft’s
stability and control characteristics across relevant flight conditions [138, 213]. This is often
done by linearizing the governing equations of motion about an equilibrium condition (trim
state) and solving the resultant eigenvalue problem to extract information about the aircraft’s
response to small perturbations [138, 139].
For complete stability, the flyer must be both statically and dynamically stable. A flyer is
statically stable if after a disturbance, the flyer’s initial tendency is to return towards its trim
state (time-invariant), while a flyer is dynamically stable if it eventually returns to its trim
state after a disturbance (time-variant). Static stability is a necessary but insufficient condition
for dynamic stability. In the longitudinal plane (x-z plane, Fig. 4.1), the dynamic response
is commonly characterized by two superimposed modes: the short period and phugoid. For
traditional aircraft, the short period mode is a heavily damped, high frequency vibration in
the angle of attack and pitch rate (Fig. 4.1a), while the phugoid mode is a lightly damped,
low frequency response in the flight speed and pitch angle (Fig. 4.1b) [138]. The associated
damping ratio and natural frequency of these modes dictate how “sluggish” or “sensitive” an
aircraft is to control inputs, and are used to define an aircraft’s flying qualities [139, 214–218].
This stability-based approach to quantifying maneuverability requires knowledge of the
aerodynamic and inertial characteristics across all flight conditions and configurations. Ob-
taining this data for birds is challenging because of their complex and variable geometries. As
75
(a) Short period mode
(α and q)
(c)
e
lift pitching erlin
drag cent
moment
horizon
glide pitch
angle,−γ angle, −θ0 pitch rate, q
0
angle of
c attack, α0
c weight center of gravity
neutral point
speed, U0
Figure 4.1: In the longitudinal plane, gliding flight dynamics are usually dominated by
the short period and phugoid modes. (a) The short period mode largely affects the angle of
attack (α) and pitch rate (q) and is visualized by the oscillation of the center line (dotted grey
line) about the fixed velocity vector (solid green line). (b) The phugoid mode largely affects
the flight speed (u) and pitch angle (θ) and is visualized by the oscillation of the centre line
(dotted grey line) about the fixed horizon (solid black line). (c) Side view of gull with key flight
parameters illustrated.
a result, there are few studies that have quantified the dynamic flight response of gliding birds
throughout wing morphing. Instead, studies of gliding maneuverability often leverage observa-
tions of live birds by tracking and analyzing their morphology and flight path [177, 219, 220].
For example, in their thesis Durston [219] used live birds, 3D printing techniques, and X-ray
computed tomography (CT scanning) to show that three species of raptors are dynamically un-
stable in the longitudinal axis while gliding towards their handler. Although this work provides
the first detailed investigation of avian dynamic stability, the results are limited to the wing
shapes and behaviors that the birds used during the recorded flights.
Here, we investigated a gull’s longitudinal dynamic stability across the full range of elbow
and wrist flexion and extension used in gliding flight. Our dynamic analyses were informed
our two previous studies on hybrid glaucous-winged (Larus glaucescens) × western (Larus
76
occidentalis) gulls. The aerodynamic results were obtained with an open-source numerical
lifting line method (MachUpX) following similar procedures to Chapter 2 [141] and the inertial
characteristics were obtained with an open-source method that models birds as a composite of
simple objects (AvInertia) [142] following similar procedures to Chapter 3. As Chapter 2
showed that gull wing-body configurations were unable to trim at a fixed shoulder angle with
no sweep or dihedral, we incorporated a furled (unspread, Fig. 4.2b) tail and two new degrees
of freedom: shoulder dihedral (Fig. 4.2a) and sweep angle (Fig. 4.2b). By coupling these
extended aerodynamic and inertial results with the traditional stability-based dynamic analysis
framework, we derived the small perturbation equations of motion across the in vivo range of
motion of the elbow and wrist for a gliding gull [128]. Next, we investigated the free vibrational
response of the gull with its wings in each morphed configuration, which allowed us to extract
the natural frequencies and damping ratios of the system. Finally, to visualize the effect of
wing morphing on the gull’s time response, we investigated two types of simplified gusts: 1) a
simplified transverse-gust modelled by an initial offset in the angle of attack and 2) a simplified
streamwise-gust modelled by an increase in the forward speed with a 1-cosine profile [214].
(a)
elbow joint
wrist joint dihedral
angle
c shoulder joint
c wrist
angle
−sweep
elbow angle
angle
Figure 4.2: Gulls can morph their wings to adjust key flight parameters. (a) Front view
of gull, visualizing the shoulder dihedral angle where a positive angle is an upwards deflected
wing. (b) Dorsal view of gull, visualizing the shoulder sweep angle where positive is a back-
wards swept wing. Elbow and wrist angles are always positive with higher angles as the wing
extends.
77
4.3 Methodology
We developed the governing equations of motion for a gliding gull in the longitudinal plane.
We assumed a rigid, non-porous, symmetric gull undergoing small perturbations in a quasi-
steady state. These assumptions allowed us to evaluate longitudinal (i.e., pitch) characteristics
separate from the lateral (i.e., roll and yaw) characteristics and obtain a state-space representa-
tion of the longitudinal governing equations. We formulated the equations of motion following
procedures similar to those outlined in aeronautical texts [138, 139]. Any deviations from these
texts are due to assumptions on the aerodynamic derivatives and are detailed in the following
sections. Note that although we modeled the bird as a rigid body, we accounted for a change in
the aerodynamic and inertial characteristics between each different wing configuration. This
approach of solving the dynamic response for each fixed configuration independently is simi-
lar to the approach used to establish the operating parameters for aircraft across different flight
conditions independently [139].
There are four frames of reference to consider when discussing traditional aircraft dynam-
ics. First, we require an inertial frame to resolve the applicable equations of motion. We
selected an earth-fixed frame as our inertial frame due to the slow gliding speeds of birds
[122, 138]. Next, the body-fixed frame is often defined so that the origin is fixed on the center
of gravity of the glider with the x-axis pointing out of the nose of the aircraft, the z-axis pointing
downwards from the aircraft’s ventral surface and the y-axis pointing towards the right wing
tip (Fig. 4.1c, xc -y-zc axis). This body-fixed frame is defined relative to the inertial frame by
the traditional 3-2-1 Euler angles φ, θ, and ψ. Within this body axis, the traditional angular
velocities are defined: roll rate (p), pitch rate (q) and yaw rate (r).
From the body axis, we can define a stability frame of reference which is defined by a
rotation about the body’s y-axis (known as the angle of attack, α) so that the new stability
x-axis is in line and opposite to the incoming wind velocity (Fig. 4.1c, x-y-z axis). Finally,
we can define a wind frame of reference by rotating about the stability z-axis (known as the
78
sideslip angle, β) until the new wind x-axis is directly in line and opposite to the incoming
wind velocity. It is within the wind frame of reference that the resultant aerodynamic loads can
be decomposed into lift (L), drag (D), side force, pitch (M ), roll, and yaw.
To simplify the formulation of the equations, we made a few key assumptions. First, we
assumed that when trimmed and at t = 0, the body-fixed frame of reference is aligned with the
stability axis (i.e. xc is parallel to x). This ensures that the entire incoming velocity is equal and
opposite to the x-axis. Next, to properly implement a rigid body dynamic analysis, we assumed
that the bird maintains a constant configuration as it undergoes perturbations, which ensures a
constant moment of inertia and center of gravity. In addition, we modeled only a single gull
specimen. We selected to use the specimen that was used to extract inertial measurements for
Chapter 3 [142].
Fx = m(U̇ + qW − rV )
Fy = m(V̇ + rU − pW )
Fz = m(Ẇ + pV − qU )
(4.1)
Mx = ṗIxx − Ixz (pq + ṙ) + qr (Izz − Iyy )
To investigate the dynamic stability about the equilibrium condition, we assumed a sym-
metric flight condition (i.e., V = p = q = r = φ = ψ = β = 0) and that the gliding bird
would only experience small perturbations about a given equilibrium condition, known as the
79
trim point (notified with a subscript of 0), defined as:
U = U0 + ∆U q = ∆q Fx = ∆Fx
V = ∆V r = ∆r Fz = ∆Fz
W = ∆W p = ∆p My = ∆My
For this small disturbance model any higher order terms (such as ∆q∆W ) are assumed to
be negligible. These assumptions reduced equation 4.1 to:
∆Fx = m(∆U̇ )
∆My = ∆q̇Iyy
Note that decoupling the longitudinal equations from the lateral equations is only an ap-
propriate assumption if we limit the glider to symmetric flight and low magnitude maneuvers
within the longitudinal axis [138]. Next, the translational velocity components can be rear-
ranged to simplify their formulation as:
∆U
u=
U0
∆W
w= = sin(∆α) ≈ ∆α
U0
This leads to the final form of the small disturbance longitudinal equations as:
∆My = ∆q̇Iyy
Because there is a dependence of the forces on the pitch angle (θ) there are effectively four
unknowns and only three equations. Therefore, these equations are traditionally supplemented
by the definitional relationship:
∆θ̇ = ∆q (4.4)
80
The final steps now include determining the inertial and aerodynamic characteristics that
are required prior to solving the system represented by equations 4.3 and 4.4.
The solution of this linear dynamic system requires an equilibrium condition or trim point.
Trim is defined so that Fx0 = Fz0 = My0 = 0. Evaluating the free body diagram shown in Fig.
4.1c assuming the stability axis is aligned with the body-fixed axis at t = 0, allows us to write
the full trimmed equations as:
My0 = 0
These equations show that to trim while supporting the glider’s weight during flight, there
must be a positive lift force to balance out the opposing body weight when the pitching moment
is equal to zero. However, in Chapter 2 we found that wing-body configurations were unable
to trim when the wings were held at a shoulder angle with 0◦ dihedral and 0◦ sweep [141]
(Fig. 4.2). These configurations could not trim because they generated a negative zero-lift
pitching moment and a negative pitch stability derivative, thus there was no angle of attack that
generated positive lift while the pitching moment was zero, which is necessary to trim.
However, the existence of a trim state is necessary for most stability analyses. To address
this issue, we first adjusted the model to include a furled, static tail (Fig. 4.2b). The tail
was modeled as a flat, thin, rectangular wing behind the body with a NACA 0006 airfoil and
dimensions based on previously obtained furled tail measurements from the same gull species
[142]. We found that the tail had a minor stabilizing effect but alone was not sufficient to
trim. We expect that the tail will have a larger impact when spread and/or if it is rotated at an
incidence angle relative to the body, like an aircraft’s elevator.
As gulls are capable of gliding with their tail furled, we next included two new degrees of
freedom: the sweep and dihedral angle at the shoulder joint (Fig. 4.2). We investigated setting
the shoulder dihedral angle at 0◦ , 10◦ and 20◦ and the shoulder sweep angle at -20◦ , -10◦ , 0◦ ,
81
10◦ , and 20◦ . Because these new parameters required 15-fold more configurations to be tested,
we sub-sampled the elbow and wrist configurations that we ran in MachUpX to 200, down
from 1031 in the previous aerodynamic study [141]. We ensured that the configurations were
equally distributed by binning in increments of 5◦ of elbow angle by 5◦ of wrist angle and
randomly selecting one configuration from each bin. In addition, to increase the convergence
speed for these complex wing shapes, we implemented a custom line search that leverages an
inverse parabolic interpolation [221] to calculate the optimal relaxation factor for each iteration
of MachUpX’s Newton method [152, 154]. We verified that this update returned the same
converged result as MachUpX’s fixed relaxation factor. In addition to the updated aerodynamic
results, we used outputs from our previous inertia study [142] to recompute the center of gravity
and moment of inertia of each wing configuration allowing for the new degrees of shoulder
motion (see Section 4.3.4 for details).
Because the modeled gull was not the same individual used to estimate the aerodynamic
parameters, we extracted the coefficient of lift (CL ) and pitching moment (CM ) from the nu-
merical lifting-line method. Note that these are similar to the coefficients calculated in Chapter
2 but we decided to not use the “morph” subscript for brevity. Further in this chapter, the coef-
ficient of pitching moment was non-dimensionalized with the specimen’s maximum root chord
rather than the wing-body mean chord. In Chapter 2, we found that the numerical results best
agreed with experimental data within the expanded uncertainty range for angles of attack below
5◦ [141]. Therefore, in this work we limited our aerodynamic analysis to configurations that
could achieve a trimmed configuration below 5◦ . In addition, numerical results were limited to
angles of attack greater than or equal to -10◦ to reduce the amount of models.
To extract the aerodynamic coefficients for the lift and pitching moment, we used linear
models fit to the outputs from MachUpX as:
CL = A0 + A1 e + A2 w + A3 α + A4 ew + A5 eα + A6 wα + A7 ewα + A8 α2 +
82
CM = B0 + B1 e + B2 w + B3 CL + B4 ew + B5 eCL + B6 wCL + B7 ewCL +
B14 ewΛCL + B15 ΛCL + B16 e2 + B17 e3 + B18 w2 + B19 Γ + B20 ΓCL (4.7)
Where α is the angle of attack, e is the elbow angle, w is the wrist angle, Γ is the shoulder
dihedral angle, and Λ is the shoulder sweep angle. These models allow for interactive effects
between the key joint angles. The linear model coefficients (Ai and Bi ) were determined using
a linear model fit in R and exported into the Python dynamic analysis code. The models had
an adjusted R2 value of 0.87 and 0.91, respectively. The linear models agree with the forms
determined in Chapter 2 [141] but were updated to include the effects of the shoulder sweep
and dihedral angle.
Note that the coefficient of pitching moment was adjusted to be calculated about the center
of gravity for each specific configuration. This required the center of gravity location for the
specimen used for the aerodynamic results. Thus, we assumed that the center of gravity of the
aerodynamic specimen would be located at the same distance from the shoulder joint as the
gull specimen used in the data analysis. This value ranged between 2.7 to 5.8 cm backwards,
and 0.00 to 0.02 cm downwards from the shoulder joint, depending on the wing joint angles.
This location was then used as the origin for recalculating the pitching moment coefficient.
We validated the fit of these models by estimating the lift and pitching moment of the ex-
perimental data. We found that the average absolute error between the linear model prediction
and the experimental measurement was 0.015 for the coefficient of pitching moment and 0.082
for the lift coefficient. In addition, we found that the average absolute error between the linear
model prediction and the numerical result was 0.012 for the coefficient of pitching moment
and 0.058 for the lift coefficient. These results highlight that there is error between our model
fit and both the experimental and numerical results. We investigated using machine learning
regression techniques to improve the model predictions however, there was no significant re-
duction in the error. As the overall error is relatively low, we decided to proceed with our
analysis, however improved model fits will improve the accuracy of our methods for future
work.
Next, we used a different approach for the drag because MachUpX drag predictions did
83
not agree well with our experimental results for wing configurations that were heavily swept
backwards [141]. Therefore, we estimated the coefficient of drag (CD ) directly from the ex-
perimental data as:
This model had an adjusted R2 value of 0.89 and was used to estimate the drag for each wing
configuration. We found that the average absolute error between the linear model prediction
and the experimental measurement was 0.014 for the drag coefficient. Note that we did not
include the effect of shoulder sweep and dihedral in this model. This is because the experi-
mental results were only obtained for nine wings with a single fixed shoulder configuration.
Therefore, throughout this work we effectively assumed that the effect of the shoulder sweep
and dihedral angles on the drag force is negligible. For this reason, we limited the maximum
angles for the shoulder to 20◦ to minimize the effects on the drag production. Future research
is necessary to investigate the role of how the shoulder angle variation on complex gull wing
shapes will affect the drag force.
Trimmed configurations
With the linear models for each aerodynamic coefficient (equations 4.6, 4.7, and 4.8), we
reformulated equation 4.5 to equation 4.9. This system of equations allowed us to calculate the
trim angle of attack (α0 ), trim speed (U0 ), and trim glide angle (γ0 ) (Fig. 4.1c). We iterated
through all possible combinations of elbow, wrist, and shoulder angle in the in vivo gull gliding
range to calculate the trim position of each configuration (23). The elbow angle was varied
from 86◦ to 164◦ (∆2◦ ), wrist angle from 106◦ to 178◦ (∆2°), sweep angle from -20◦ to 20◦
(∆5◦ ), and dihedral angle from 10◦ to 20◦ (∆5◦ ).
1
0 = − ρU0 2 SCD − mgsin (γ0 )
2
1
0 = − ρU0 2 SCL + mgcos(γ0 ) (4.9)
2
1
0 = ρU0 2 ScCM
2
Where ρ is the air density, S is the maximum wing-body area across all morphed configura-
84
tions, c is the maximum wing root chord across all morphed configurations, and m is the mass.
To ensure compatibility across all metrics, the reference area, chord, and mass are all from one
gull specimen with 0° shoulder sweep and dihedral angle that was investigated in our previous
inertial study [142]. Similar to the derivation of the linear models, we limited our outputs to
configurations that could trim at α0 < 5◦ because MachUpX best matched the experimental
data at low angles of attack [141]. If a configuration with a given combination of joint angles
could not trim within our set parameter space, it was not included in the analysis.
We found that increasing the shoulder dihedral angle and sweeping the wings forward al-
lowed more wing configurations to trim due to an increased zero-lift pitching moment. Because
we found that the majority of elbow and wrist configurations could trim at 20◦ dihedral with
forward swept wings, we limited our results to these shoulder angle parameters for the remain-
der of the study. Note that higher dihedral angles would allow all combinations of elbow and
wrist angles to trim, but we limited our analysis to 20◦ to minimize the effects on drag esti-
mation. Even with these limitations some of the configurations required extremely steep trim
glide angles (γ0 ), likely closer to terminal velocity than true gliding flight. Therefore, we lim-
ited our results to configurations that had γ0 < 45◦ . In total, 1457 configurations both satisfied
our imposed limitations and were able to trim (Fig. 4.3F).
The key inertial characteristics for the longitudinal dynamic response are the mass (m),
center of gravity, and moment of inertia about the y-axis (Iyy ). We determined these inertial
characteristics using the gull specific outputs from Chapter 3 [142] and expanded the results
to account for the shoulder sweep and dihedral angle of each tested configuration. To do this,
we first recalculated the location of the center of gravity relative to the shoulder joint for each
morphed wing configuration. This location is important as it must be used as the new origin
for the pitching moment and moment of inertia.
Next, to adjust the moment of inertia, we extracted the wing-only inertial value, rotated the
wing about the shoulder joint to the appropriate dihedral and sweep angles and added it back
to the rest of the body moment of inertia. We then shifted the resultant moment of inertia so
that the origin was at the newly determined center of gravity.
85
Because we wanted to investigate any general wing configuration, we fit a linear model to
the final moment of inertia calculated about the center of gravity. The elbow, wrist and shoulder
sweep and dihedral angles were the explanatory variables within this model and interaction
terms were included. This model was used to extract the final moment of inertia for each
wing configuration and had an adjusted R2 of 0.999. We found that the average absolute error
between the linear model prediction and the numerical estimates was 6.12×10−5 kg-m2 for the
moment of inertia.
The key aerodynamic characteristics are encapsulated by the left-hand side of equation
4.3. Following traditional aeronautical methods we recast these variables using a Taylor series
expansion as follows:
Note that we have only incorporated the stability derivatives within this expansion, but con-
trol derivatives are normally also included [138, 139]. In this work, we selected not to treat the
elbow, wrist, and shoulder angles or the tail as control inputs, but rather to investigate each pos-
sible configuration independently as a rigid-body flyer. This simplified approach was intended
solely to provide an initial understanding of the dynamic stability of each morphological con-
figuration. Future work is warranted to investigate the role of the control derivatives associated
with each degree of freedom and to capture the dynamics of morphing between configurations.
To investigate the flight dynamics, each of the aerodynamic stability derivatives in equation
4.10 must be estimated. This can be simplified by investigating the free body diagram of the
configuration in question under a perturbed condition. From this perturbed aerodynamic state
we can next estimate the key derivatives.
86
Forward speed, u
If we assume a perturbation in the forward speed from the equilibrium (Eqn. 4.5) [138], we
obtain:
Fx = ∆Fx = −∆D
(4.11)
Fz = ∆Fz = −∆L
This equation allows us to calculate the effect of a change in the flight speed on the initial
equations [138]:
During our experimental analysis in Chapter 2 we tested the gull-inspired wings at two
different biologically relevant Reynolds numbers (approximately 1.5×105 and 2.2×105 ) and
found a statistically insignificant effect of velocity on the coefficient of lift, drag and pitching
moment for angles of attack used within the analysis (−10◦ < α < 5◦ ) [141]. This outcome
was expected since the testing was limited to speeds that birds are known to fly at, which are
well within the subsonic regime [122]. These simplifications led to the final form of the speed
derivatives:
∂Fx 1
≈ ρU02 Smax (−2CD0 )
∂u 2
∂Fz 1
≈ ρU02 Smax (−2CL0 ) (4.12)
∂u 2
∂My
≈0
∂u
87
Angle of attack, α
Next, we assumed a perturbation in the angle of attack and a small angle assumption to
obtain the perturbed equations as:
∆Fx ≈ L0 ∆α − ∆D
(4.13)
∆Fz ≈ −∆L − D0 ∆α
∂Fx 1 ∂CD
≈ ρU02 Smax (CL0 − )
∂α 2 ∂α
∂Fz 1 ∂CL
≈ ρU02 Smax (− − CD0 ) (4.14)
∂α 2 ∂α
∂My 1 ∂CM
= ρU02 Smax cmax
∂α 2 ∂α
To estimate the derivatives with the angle of attack we took the derivative of our linear
model fits for the coefficient of lift (Eqn. 4.6), coefficient of pitching moment (Eqn. 4.7) and
coefficient of drag (Eqn. 4.8). It is possible that these angle of attack derivatives of the co-
efficient of lift have a lower magnitude than expected due to error between the numerical and
experimental results [141]. By limiting our analysis to lower angles of attack, we expect this er-
ror to be reduced but additional studies are required to improve the angle of attack relationship
with lift. Note that for the pitching moment derivative, we assumed that:
This approximation is valid at low angles of attack within the linear region of the pitching
moment and lift relationship.
Because our gull is modeled with a furled tail, we neglected all of the aerodynamic deriva-
tives related to the rate of change of the angle of attack (α̇). This is equivalent to assuming
that the flow remains attached over the wings and is an acceptable assumption for a gliding
88
configuration at small angles of attack. Note that for the force equations, this is commonly
done in large scale aircraft due to the small contribution of this term even with a tail [139].
Pitch rate, q
∂Fx 1 ∂CD
≈ − ρU02 Smax
∂q 2 ∂q
∂Fz 1 ∂CL
≈ − ρU02 Smax (4.16)
∂q 2 ∂q
∂My 1 ∂CM
≈ ρU02 Smax cmax
∂q 2 ∂q
∂CL ∂CM
We directly calculated ∂q
and ∂q
from numerical results for the same nine wings con-
figuration that were tested in the wind tunnel. This was done by running MachUpX for each
wing held at each shoulder sweep and dihedral angle configuration while varying q from -0.5
to 0.5 rad/s. Given our problem formulation in MachUpX, while we estimated the stability
derivatives q was defined to be about the shoulder joint. Therefore in this work, we implicitly
assume that the change in the pitching moment coefficient about the center of gravity due to a
pitch rate at the shoulder joint is approximately equal to that due to a pitch rate about the center
of gravity. Similarly, we implicitly assume that the change in the lift coefficient due to a pitch
rate at the shoulder joint is approximately equal to that due to a pitch rate about the center of
gravity. With these outputs we then fit a linear model that predicted the pitch rate derivatives
for each joint angles as:
∂CL
= E0 + E1 e + E2 w + E3 Λ + E4 ew + E5 eΛ + E6 wΛ + E7 Γ (4.17)
∂q
∂CM ∂CL
= G0 + G1 e + G2 w + G3 Λ + G4 Γ + G5 (4.18)
∂q ∂q
The linear models had adjusted R2 values of 0.93 and 0.80, respectively. We assumed that
the effect of q on the drag ( ∂C
∂q
D
) was negligible following traditional aircraft studies [138].
89
Flight path angle, θ
Finally, we assumed that the pitch angle had been perturbed. Using trigonometric identities
and a small angle approximation on the pitch angle the following equations are obtained:
Further, because γ = θ − α and the change in the pitch angle (θ) is independent from the angle
of attack (α), ∆γ = ∆θ. This leads to:
∂Fx
= −mgcos(γ0 )
∂θ
∂Fz
= −mgsin(γ0 ) (4.19)
∂θ
∂My
=0
∂θ
Finally, all of the above stability derivatives can be integrated into equation 4.10 and equa-
tion 4.3. We can write the final small perturbation state-space representation of each bird
configuration as:
Ẋ = AX
u
∆α
X=
∆q
∆θ
g
−2m̃CD m̃(CL − ∂C D
∂α
) 0 − U0
cos(γ0 )
∂CL ∂CL g
−2m̃CL m̃(−CD − ∂α ) 1 − m̃ ∂q − U0 sin(γ0 )
A=
(4.20)
0 ˜ ∂CM
Iyy ∂α ˜ ∂CM
Iyy ∂q 0
0 0 1 0
90
Where:
ρU0 S ρU02 Sc
m̃ = I˜yy =
2m 2Iyy
These equations model the longitudinal dynamics of the fixed-wing gull as a fourth-order sys-
tem under the simplifications outlined above.
Next, we used the Python Control Systems Library [222] to solve for the time response of
the gull for two simplified gusts.
Transverse gust
First, to model a simplified transverse gust, we solved the free response of the system given
an initial angle of attack of 2◦ . This is mathematically equivalent to an impulse in the angle of
attack. Due to the rigid body assumption, we did not model the joints as independent controls
and thus the input vector (U) is zeros for this case. To extract the time response, we solved the
following system for each trimmed wing configuration:
Ẋ = AX
91
With the following initial condition:
0
π
2 180
Xt=0 =
0
0
Streamwise gust
Second, to model a streamwise gust, we solved for the forced response of a discrete gust
model as implemented within MIL-F-8785C, which uses a 1-cosine velocity profile [214]. We
assumed that the gust velocity was in the positive x direction such that:
ua = u − ug (4.21)
Note that the gust velocity will be subtracted from the body velocity as the governing
equations were developed with motion relative to the atmosphere [139]. This results in the
following state space system:
Ẋ = AX + BU
Where:
2m̃CD
2m̃CL
B=
0
0
h i
U = uf (t)
The initial condition was set to zero for all inputs. Finally, we solved the system by discretizing
in time (following methods outlined in the Python Control Systems Library [222]), so that the
92
gust was modeled as:
0 t=0
uf (t) = ug
2
(1 − cos( tπt
m
)) 0 < t ≤ tm
u g
t > tm
Where ug is the gust amplitude, which was set as 2% of the trim speed (U0 ) and tm was 5
seconds. This model is similar to that introduced by MIL-F-8785C [214], although we defined
the gust relative to time rather than distance traveled.
4.4 Results
With the aerodynamic coefficients for each wing configuration, we solved for the trim angle
of attack (α0 ), trim speed (U0 ), and trim glide angle (γ0 ) (Fig. 4.3a, see Section 4.3.3). For the
configurations capable of trimmed flight (n = 1457), we found that the trim speed ranged from
11.8 to 29.8 m/s and the shallowest trim glide angle was -12.2◦ . In-flight measurements of gulls
gliding past an urban environment by Shepard et. al [25] showed an airspeed range from 8.1
to 19.9 m/s, which includes approximately half of our configurations (n = 768). This previous
study investigated gulls in transient flight and did not include behaviors with high glide angles
such as those used in landing flight. As such, we expect that gulls have the capability to trim at
the higher speeds as predicted by our model. Note that due to our imposed limits on the angle
of attack and dihedral angle, it is likely and probable that gulls can also trim at lower speeds
and shallower glide angles. Our identified trim states permit an initial evaluation of dynamic
stability in gull gliding flight.
To determine how the three different wing joints affect the trim state, we fit linear models
where the trim speed and glide angle were the dependent variables and the elbow, wrist, and
sweep angles were independent variables. This analysis revealed strong interactive effects
between the joint angles for both the trim speed and glide angle. Folding the wrist, increased
the trim glide speed for over 80% of tested configurations and folding the elbow, increased
the trim glide speed for 60% of tested configurations. Because there is evidence that many
93
(a) (b) (d) (f)
0 150 −15° sweep only 180 stable
unstable
140
160
Trim glide angle, γ0 (°)
130
−15
90
−45 80 100
0 30 60 0 2 4 6 8 10 80 100 120 140 160
(c) # of trimmed (e) − body -axis (cm) Elbow angle (°)
configurations
configurations
# of trimmed
Figure 4.3: Wing morphing allows gulls to switch between statically stable and unstable
configurations. (a) The gull can fold its wrist to trim as speeds increase. The configurations
capable of trimmed flight (n = 1457) include multiple forward sweep angles (b and c). (d)
Gull wing morphing allows a substantial shift in the static margin largely due to neutral point
control as the motion of the center of gravity remains relatively small. (e) The gliding gull is
unstable if the neutral point is in front of the center of gravity and stable if it is behind. (f)
Adjusting the forward sweep angle ensures that the majority of elbow and wrist configurations
at 20◦ dihedral angle can trim, but most configurations with high wrist angles become statically
unstable (hollow squares).
bird species, including gulls, fold their wing joints as wind speeds increase [39, 52, 82, 85, 92,
100, 124, 128, 223], our results suggest that this wing morphing behavior allows gliding birds
to adjust their trim condition to adapt to different flight conditions. In addition, we found that
reducing the forward sweep at the shoulder joint caused the trim glide speed to increase (and the
glide angle to decrease) for each tested configuration as expected from traditional aeronautical
results [129]. Yet, there is little documented evidence of birds sweeping their wings backwards
at the shoulder joint in response to increased wind speeds. Therefore, it is possible that to trim,
birds preferentially fold their wings rather than just changing the shoulder joint sweep angle.
One benefit to folding the elbow and wrist over sweeping the entire wing would be that folding
the wing both reduces the total wing lifting area and moves the wings closer to the body. These
94
two effects would reduce the wing bending moment, whereas adjusting the shoulder sweep
angle would not change the wing area and only marginally move the wings closer to the body.
A directed study is required to determine if and how birds balance trade-offs between structural
constraints and aerodynamic loading in trimmed flight.
With the known trim state for each configuration, we next investigated the static stability
of each configuration. We quantified static stability with the static margin, a measure of the
distance between the center of gravity and the neutral point. The neutral point is the location
where the distributed forces and moments can be modelled as point loads. It differs from the
center of pressure because the pitching moment about the neutral point is independent of the
angle of attack. If the neutral point is behind the center of gravity, the configuration has a
positive static margin and is statically stable (Fig. 4.3e) [141].
We found that the majority of trimmed elbow, wrist, and shoulder combinations for the gull
were statically stable (solid squares, n = 1331, Fig. 4.3f), but there was a set of configurations
with extended wrist angles that were unstable (hollow squares, n = 126, Fig. 4.3f). We found
that the progression towards instability for the gull wings was largely driven by a shift in the
neutral point rather than the center of gravity, as is expected since wing morphing has only a
marginal effect on shifting the center of gravity (Fig. 4.3d for a constant 15◦ forward sweep
angle) [142]. This result agrees with Durston’s finding that raptors gliding towards their trainers
with fully extended wing configurations were statically unstable [219]. However, our results
expand on this understanding to reveal that a gull can fold its wrist to achieve a stabilized
configuration, allowing a shift between stable and unstable flight conditions.
This capacity to shift stability with wing morphing agrees with a previous finding that most
species can shift between stable and unstable flight [142]. However, that previous study was
limited to 0◦ shoulder sweep and dihedral angles, which yields only statically stable configura-
tions for the gull. Here, we expanded on these results to show that including the shoulder joint
further enhances birds’ ability to transition between statically stable and unstable flight. It is
important to highlight that there is limited data on the true shoulder angles used in bird flight.
However, gulls are often observed flying with swept forwards wings held at a positive dihedral
95
angle [224], therefore this stability shift is likely used by live gliding gulls.
Static stability provides a necessary, but insufficient condition of full stability. To deter-
mine if a gull is completely stable, we next calculated the dynamic characteristics by solving
for the eigenvalues of the rigid gull modelled as a fourth-order system. We found that all the
statically stable configurations had eigenvalues with negative real values for both vibrational
modes (solid points, n = 1331, Fig. 4.4a and b) leading the gull to be dynamically, and thus
completely, stable in the longitudinal axis. Only statically unstable configurations had dynam-
ically unstable responses and exhibited a non-oscillatory divergent response, which was char-
acterized by eigenvalues with only real parts, similar to Durston’s results [219]. This indicates
that the gull was only ever completely stable or unstable and there were no configurations that
exhibited static stability and dynamic instability. Of note, we found that the phugoid mode re-
mained stable even for the statically unstable configurations (Fig. 4.4b) and thus the instability
was entirely due to the unstable short period response.
With the complex and real components of the eigenvalues, we calculated the damping ra-
tio (ζ) and natural frequency (ω) associated with the two vibrational modes for all the stable
Imaginary
−40 −1.25
Real Real
Short period Phugoid
Figure 4.4: Root locus plot of the open-loop system. The poles of the fourth order system are
displayed for a forward sweep angle of 5◦ (triangles) and 15◦ (circles) for the short period (a)
and phugoid mode (b).
96
configurations. First, one mode had a high frequency, highly damped response (Fig. 4.5a and
c) that was independent of the speed (demonstrated by the low magnitude of the teal dots in
Fig. 4.6a). This response is characteristic of the short period mode (Fig. 4.1a). The short
period frequency ranged from 8.9 to 41.0 rad/s, which is approximately half to more than
double the frequency of a similar sized UAV (Fig. 4.5a) [217]. Previous studies have shown
that small UAVs will have higher short period frequencies than large aircraft due to scaling
alone [217, 218]. The gull’s variable frequency response is because wing morphing allows a
substantial shift in the static margin, and thus the static stability, compared to values used in
traditional UAV designs [138, 139]. To this end, wing morphing has a strong effect on the short
period characteristics for gulls. Furthermore, we found significant interactive effects between
the elbow, wrist, and sweep angles (visualized in Fig. 4.5a and c). Despite these interactive
effects, general trends in the short period characteristics are apparent within our investigated
joint ranges. For example, wrist extension decreased the short period natural frequency (Fig.
4.5a) and increased the damping ratio (Fig. 4.5c), when the elbow angle was above 90◦ .
Next, the second identified mode had a substantially lower frequency response (Fig. 4.5b)
that was independent of the angle of attack (demonstrated by the low magnitude of the light
blue dots in Fig. 4.6b). This response is characteristic of the phugoid mode (Fig. 4.1b). The
phugoid mode had a similar or slightly lower frequency than a similar sized UAV ranging from
0.45 to 1.10 rad/s (Fig. 4.5b). However, we found that the phugoid mode was heavily damped,
with a damping ratio on the same order of magnitude as the short period mode (Fig. 4.5d). This
is unlike most comparable UAVs or large scale aircraft [138, 217, 218], although a flight test
on a smaller morphing gull-wing UAV found a similar heavily damped phugoid mode [31]. We
expect that the high damping is because we investigated gliding flight rather than steady, level
cruise and because our configurations were limited to angles of attack below 5◦ , which ex-
cludes the most aerodynamically efficiency results for these wing configurations. Per Pamadi’s
phugoid approximation, these combined approximations will decrease the phugoid frequency
and increase the damping ratio [138]. Therefore, it remains possible that at more efficient trim
conditions a gull configuration could be statically stable but dynamically unstable due to re-
duced phugoid damping. Like the short period mode, we found significant interactive effects
between the elbow, wrist, and sweep angles (visualized in Fig. 4.5b and d). Unlike the rela-
97
Sweep angle (°): −5 Wrist angle (°): 120 140 160
−15
20 1.0
10 0.5
0 0.0
(c) 1.00 (d)1.00
Damping ratio, ζ
0.75 0.75
0.50 0.50
0.25 0.25
0.00 0.00
80 100 120 140 160 80 100 120 140 160
Elbow angle (°) Elbow angle (°)
Figure 4.5: The short period and phugoid mode characteristics are significantly affected
by the wing positioning. Within the in vivo range (above 90◦ elbow angle) wrist extension
tends to decrease the short period (a) natural frequency and increase the (b) damping ratio. The
effect of wrist extension on the phugoid (c) natural frequency and (d) damping ratio depends
on the elbow angle.
tively consistent wing morphing trends for the short period, the effect of wrist extension on the
phugoid frequency (Fig. 4.5b) and damping (Fig. 4.5d) reverses signs within our investigated
ranges. For example, the damping ratio tends to increase with wrist extension at low elbow
angles but tends to decrease with wrist extension at high elbow angles.
To better understand the dynamic response characteristics, we compared the estimated fly-
ing qualities of the gull to known aircraft specifications. We evaluated the flying qualities as
established by the U.S. Department of Defense’s MIL-F-8785C specification [214]. This spec-
ification defines three levels of flying qualities: Level 1 flying qualities are clearly adequate for
98
(a) (b)
90 90
1.00 2.5
0.75 2.0
1.5
0.50
1.0
Magnitude
Magnitude
0.25 0.5
0.00 180 0 0.0 180 0
speed
angle of attack 270 270
pitch rate
pitch angle
Phase (°) Phase (°)
Figure 4.6: Short period and phugoid modes were identified from the magnitude of the
eigenvectors. All magnitudes and phases were normalized to the pitch rate (maroon dot) to
facilitate comparison. (a) The short period was characterized by the high magnitude response
in the pitch rate (maroon dots) and a small magnitude response in the speed (dark teal dots).
(b) The phugoid mode was characterized by a high magnitude response in the speed (dark teal
dots) and pitch angle (orange dots) with a small magnitude response in the angle of attack (pale
blue dots).
the given flight phase; Level 2 necessitates a higher pilot workload and/or degradation of mis-
sion effectiveness; and Level 3 results in an excessive workload or inadequate mission effec-
tiveness. We considered only qualities associated with flight phases that include non-terminal
flight maneuvers such as a gliding descent (Category B per MIL-F-8785C).
MIL-F-8785C defines desirable short period characteristics by the damping ratio (ζ) and a
2
short period frequency metric, which is the ratio of the natural frequency squared (ωsp ) to the
load factor per angle of attack (nα ) (y-axis, Fig. 4.7). Considering the damping ratio limits, we
found that most configurations (n = 1232) satisfied the Level 1 requirements, however some
configurations (n = 99) with a 20◦ forward sweep angle only satisfied the Level 2 requirements
(Fig. 4.7, M, solid vertical lines). Considering the frequency metric requirements, we found
that only seven configurations satisfied Level 2 requirements (Fig. 4.7, M, solid horizontal
lines). Although interactive effects were again significant, we found that sweeping the wing
forwards, extending the wrist, or folding the elbow angle tended to reduce the short period
frequency metric, thus improving the flight quality (p-values < 0.001, R2 = 0.9083, Fig. 4.7).
99
100 FB - Level 2
C - Level 2
/nα
FB - Level 1
M - Level 2
10
M - Level 1
M - Level 1
M - Level 2
Wrist angle (°)
Figure 4.7: Most gull configurations satisfied Level 2 short period requirements for human
pilots per adjusted UAV guidelines. For the damping ratio, all configurations satisfy at least
Level 2 MIL-F-8785C (shortened to M, solid vertical lines). For the frequency metric, no
configurations satisfied the Level 1 MIL-F-8785C requirements (M, solid horizontal lines).
Adjusting for two previously published UAV metrics revealed that 185 configurations satisfied
Capello et. al’s Level 1 upper limits (C, grey dotted lines) and 457 configurations satisfied
Foster and Bowman’s Level 1 upper limits (FB, black dashed lines). All configurations satisfied
Foster and Bowman’s Level 2 upper limits.
All stable configurations exhibited at least Level 3 qualities in the damping ratio and frequency
limits, but this is indicative of a flyer that would be difficult to control [214].
However, studies on small UAVs have shown that the MIL-F-8785C short period frequency
metric guidelines do not accurately capture the flying qualities of small UAVs [215–218]. As a
result, new scaling parameters have been proposed. Incorporating Foster and Bowman’s [218]
scaling we found that all of our stable, trimmed gull configurations (n = 1331) would have
at least Level 2 flying qualities and 457 configurations would have Level 1 flying qualities
(Fig. 4.7, FB, dashed lines). Incorporating Capello et. al’s scaling [217], we found that 1167
configurations would have Level 2 flying qualities and 185 would have Level 1 flying qualities
(Fig. 4.7, C, dotted lines). Note that we used a Cessna 172 as the comparable large scale
aircraft to calculate the scaling constant [217, 225]. Thus, by accounting for known differences
between large scale aircraft and small UAVs, our results suggest that a gull-like UAV design
100
with wings swept forward less than 20◦ would be flyable albeit with a higher pilot workload
for many configurations.
Unlike the short period mode, MIL-F-8785C only provides a minimum criterion on the
phugoid damping ratio and UAV-focused studies tend to agree with the effectiveness of this
parameter [217, 218]. Our results show that the gull was substantially above the Level 1 mini-
mum damping ratio of 0.04 (Fig. 4.5c) and had nearly an order of magnitude higher damping
ratio than comparable UAVs. As discussed previously, this is due both to our gliding analysis
and the lower aerodynamic efficiency in the tested configurations. Future work is required to
determine if these values exhibited by gliding gulls are too heavily damped for effective imple-
mentation in a gliding UAV. A higher damped phugoid mode may be beneficial as this mode
is notorious for pilot-induced oscillations, but the high damping also suggests that there is a
slow response to control inputs for the flight speed and pitch angle. This sluggish response to
elevator inputs was observed for a small gull-wing morphing UAV [31].
These differences in the phugoid modes between gulls and UAVs are intriguing because
they may play a role in avian gust response, which tends to outperform comparable fixed-wing
UAVs. Small perturbations in the forward velocity of a trimmed gliding gull would be quickly
damped out according to our model. However, the gull would need to use larger control inputs
to maneuver away from the equilibrium condition. These results reveal a reason that gulls may
elect to switch from a stable to an unstable configuration. Gulls could use a stable configuration
to reject undesired perturbations from their local environment while in transit or foraging for
food. Then, gulls could extend their wrists to morph into an unstable configuration to gain a
more sensitive reaction to control inputs, which would support rapid maneuvering.
Because the heavily damped phugoid mode pointed to possible gust-related benefits, we
explored the time response of the gull to transverse and streamwise gusts. Note that a bird’s
gust response affects their foraging and landing capabilities [226–228]. Intriguingly, a study of
live gulls in a wind tunnel found that increased turbulence intensity (a measure of variation in
the freestream velocity) had no effect to the overall metabolism of the bird and thus no effect
on their energetic requirements [229]. Since we found that both the phugoid and short period
101
(a) Transverse gust - free response (b) Streamwise gust - forced response
0.02 Wrist angle (°): 0.030
106
116
Change in speed, u (m/s)
126
0.01 136 0.015
146
156
166
0.00 0.000
−0.01 −0.015
−0.02 −0.030
1 0.05
0 0.00
−1 −0.05
−2 −0.10
50 1.0
Change in pitch rate, Δq (°/s)
25 0.5
0 0.0
−25 −0.5
−50 −1.0
Change in pitch angle, Δθ (°)
2.50 2.50
1.25 1.25
0.00 0.00
−1.25 −1.25
−2.50 −2.50
0 5 10 15 20 0 5 10 15 20
Time (s) Time (s)
Figure 4.8: Gull wrist gains allows substantial control over gust response. (a) A simpli-
fied transverse gust modeled as an initial offset of 2◦ angle of attack. (b) A discrete 1-cosine
streamwise gust modeled as an increase of 2% of the initial trim speed (U0 ) over five seconds
(gray box). Only configurations with an elbow angle of 130◦ and a forwards sweep angle of
15◦ are shown.
102
mode are heavily damped, it is possible that a gliding gull in a stabilized configuration does
not require active control to return quickly to an equilibrium condition, which eliminates any
additional energetic costs.
To visualize the effect of small environmental fluctuations, we calculated the dynamic re-
sponses to disturbances modeled with either a 2◦ step change in the angle of attack or a 2%
increase in the forward speed (see Section 4.3.8). We investigated configurations with a fixed
elbow angle (130°) and fixed shoulder sweep angle (15◦ ) but a variable wrist angle (Fig. 4.8).
This range allowed us to explore both stable and unstable configurations across a broad range
of wrist angles. Each wing configuration was at a different trim state. The wrist angle of 156°
had a trim glide angle (γ0 ) of 61◦ and was excluded from the previous analyses as it is steeper
than the imposed limit of 45◦ (see Section 4.3.3). We included it here for completeness.
The time responses of these configurations captured the quickly diverging dynamics asso-
ciated with higher wrist angles and showed that small perturbations in the angle of attack (Fig.
4.8a) or speed (Fig. 4.8b) would be quickly damped out for lower wrist angles. For the stable
configurations with lower wrist angles, the time to half the amplitude varied between 2.22 to
2.52 seconds for the phugoid mode and 0.05 to 0.12 seconds for the short period mode. For
the unstable configurations with higher wrist angles, time to double the amplitude varied from
0.80 to 1.69 seconds for the phugoid mode and 0.03 to 0.04 seconds for the short period mode.
In all, these results show that gulls gain significant control over their dynamic characteristics
through solely adjusting their wrist joint. It is important to again highlight that the strong in-
teractive effects acting between the elbow, wrist, and shoulder joints means that the effect of
each morphing joint depends on the other joint positions and these evaluated configurations are
only a representative sample. These interactive effects indicate that a complex control system
would be required to effectively pilot a gull-like UAV.
4.5 Limitations
There were many simplifications used throughout this study. Predominately, this approach
uses a quasi-steady aerodynamic analysis of a rigid gull undergoing small perturbations in
symmetric flight. Additional work is required to extend this analysis to include a lateral anal-
103
ysis, non-linear and unsteady flight conditions, and larger scale atmospheric gusts. Further,
we used a single gull specimen to focus our study, but there will be individual differences as
well as species-specific differences in the dynamic characteristics of bird flight. To extend this
study, further work will be necessary to improve aerodynamic prediction capabilities at higher
angles of attack to allow an investigation of the most aerodynamically efficient configurations.
Although the discussed flight qualities are a useful comparative tool, the MIL-F-8785C and
the adjusted UAV guidelines are dictated by conversations with pilots [214, 217, 218]. These
flying qualities do not necessarily translate to avian flying qualities, which are unlikely to be
directly comparable to human metrics. Work is underway to investigate the neurological con-
trol mechanisms related to avian flight [136]. Connecting our results with information about
the avian neurological control system will be a necessary next step to understand avian flight
control methodologies.
4.6 Conclusion
Gulls regularly morph their wings in flight, which has been hypothesized to permit en-
hanced maneuverability and control. Our work incorporated existing studies on the aerody-
namic and inertial properties of a gliding gull to provide the first detailed investigation of
dynamic stability characteristics throughout wing morphing. Our results suggest that the gull
could fold its wing joints or sweep its wings backwards to remain in a trimmed state as wind
speeds increase. Further, we showed that most gull wing configurations have a short period
mode that satisfies the minimum controllability requirements for Level 2 human-piloted air-
craft as well as a heavily damped phugoid mode. We suggested that the high phugoid damping
acts to reduce the gull’s sensitivity to small perturbations in the localized environment. How-
ever, this reduced sensitivity suggests that the gull would have a sluggish response to control
inputs needed to effectively maneuver. Thus, we hypothesized that gulls initiate sudden ma-
neuvers by morphing into unstable configurations and shift into a stabilized configuration to
reject non-desirable perturbations to their flight path.
In all, our study confirms that gulls can negotiate trade-offs in stability and maneuverability
by morphing their wings between dynamically stable and unstable configurations and provides
104
a mechanism for how birds exhibit both stable flight and sudden, rapid maneuvers. Our results
should encourage additional engineering investigations into morphing wings that may permit a
substantial shift in the static margin. With this capability, we will be able to identify whether the
ability to shift stability modes is a necessary condition to achieve avian-like maneuverability
and if this approach can be harnessed to advance the maneuverability of future UAVs.
105
Chapter 5
Conclusion
I never am really satisfied that I understand anything; because, understand it well
as I may, my comprehension can only be an infinitesimal fraction of all I want to
understand about the many connections and relations which occur to me, how the
matter in question was first thought of or arrived at. . .
– Ada Lovelace
In Chapter 2, I performed the first experimentally validated lifting line analysis of a mor-
phing bird wing and found that the extension trajectory followed while morphing the wing
substantially affected the resultant aerodynamic characteristics [141].
106
In this chapter, I implemented an open source software (MachUpX) that calculated the lift
and pitching moment of a gull’s wing across the full range of flexion and extension of the
elbow and wrist. To incorporate biologically-relevant wing shapes, I developed a code that
discretized real gull wings into shape parameters that could be implemented in MachUpX.
This methodology was informed by additional XFOIL studies on the airfoils selected to model
the gull wing. These final numerically implemented, gull-inspired wings were then exported
as half-wings that could be 3D printed and mounted in the wind tunnel at the University of
Michigan. I 3D printed nine wings and performed wind tunnel experiments to measure the
loads acting on the wing across a range of angles of attack. With the output data, I quantified the
experimental uncertainty and implemented statistical approaches to analyze how the elbow and
wrist angle affected the aerodynamic loads and stability across the complete range of motion.
My results identified that extending the wing with different combinations of elbow and
wrist angles yielded substantially different effects on the loads and stability. One extension
trajectory linearly decreased the static margin but kept the lift and pitching moment constant,
while another linearly increased the lift and pitching moment but kept the static margin con-
stant. These results are promising for two reasons. First, the linear response indicates that
there are certain extension trajectories that could be used by two joints that would not require
a control system to account for non-linear characteristics. Second, the substantially different
outputs indicate that the same joints can adapt to different control requirements depending on
the flight condition or mission.
However, I found that minor departures from the identified extension trajectories led to a
coupled change in the stability and loads. This indicates a thorough sensitivity analysis of
outputs would be necessary to successfully implement a gull-inspired joint-driven wing in a
UAV. A version of this chapter has been published [141].
In Chapter 3, I developed the first generalized method to calculate the inertial characteristics
of flying birds and used a comparative analysis to investigate the static stability of 22 bird
species [142].
For this work, I leveraged a classic analytical approach to model a flying bird with over 200
107
components. This approach can account for any wing posture and performs a quick calculation
of the moment of inertia tensor and center of gravity. This software is published as an open
source method on CRAN and GitHub. With measurements from 22 species of birds, obtained
by my collaborators at the University of British Columbia, I calculated the center of gravity
and moment of inertia associated with the complete range of motion of the elbow and wrist.
With the outputs from Chapter 2, I estimated the location of the neutral point as a function of
each bird’s full range of elbow and wrist extension and flexion. With the center of gravity and
neutral point, I then calculated the static margin of each species and developed a novel metric
for the torsional pitch agility of a flying bird. Finally, I performed an evolutionary analysis
to identify the time series model that best captured the modern birds’ static margin traits. To
validate and verify these results, I implemented multiple sensitivity analyses to account for
measurement error and statistical variation.
I found that the majority of the investigated species (17 out of 22) had the capacity to shift
between longitudinally stable and unstable flight within the range of motion of the elbow and
wrist. Further, I found evidence of evolutionary pressures acting to maintain this capacity to
shift between stable and unstable flight. This result shifts the field’s general understanding
of avian flight evolution as it was previously believed that birds were evolving towards being
entirely unstable. This outcome instead suggests that there may be adaptive benefits for birds
to maintain the capability to shift stability modes. Such a possibility should inspire future
UAV designs to implement the capacity to shift between stable and unstable flight to approach
avian-like maneuverability.
Although these results specifically indicate the capacity to shift, additional work is required
to identify if, and when, live flying birds shift between stability modes. As discussed previ-
ously, different control algorithms are usually implemented for stable (open-loop) or unstable
(closed-loop) configurations. Effectively implementing a control system in a UAV that can
shift between these distinct stability characteristics will require a directed study. Further work
is also required to identify how birds seamlessly adjust to these disparate flight conditions. A
version of this chapter has been published [142].
108
5.1.3 Wing joints control the dynamic response
In Chapter 4, I performed the first analysis of the dynamic stability of a bird across its full
range of elbow and wrist extension and flexion.
For this work, I merged the aerodynamic analysis methodology established in Chapter 2
with the inertial analysis methodology established in Chapter 3. With these aerodynamic and
inertial inputs, I formulated the governing equations of motion for a rigid gull that undergoes
small disturbances in the longitudinal axis. To determine a trim condition, I incorporated the
shoulder joint in this analysis. I found that a positive dihedral and forward sweep angle at the
shoulder joint increased the amount of trimmed configurations. With the known trim condi-
tions for each configuration, I next calculated the eigenvalues associated with each wing joint
configuration. I found that trimmed wings with high wrist angles were dynamically unstable
with a non-oscillating divergent response. This unstable response was due to a negative static
margin. For wings with lower wrist angles, I found that wings were dynamically stable and
exhibited a traditional short period and phugoid mode. The associated natural frequency and
damping ratio of these modes was found to significantly depend on the wing joints and the
interactive effects between the joints.
My results showed that the stable wing configurations had short period characteristics that
would be controllable by a human pilot. To arrive at this conclusion, I incorporated a UAV-
based scaling to the traditional military flying quality specifications. This scaling accounts for
size differences between UAVs and large scale aircraft. Unlike the controllable short period,
the phugoid mode was heavily damped and would likely result in a sluggish response to control
inputs. Therefore, I hypothesized that gull’s ability to shift between stable and unstable flight
would be necessary to allow for effective pitch control in gliding flight. For example, the
heavily stabilized configurations with lower wrist angles could be used when it is desirable to
inherently reject flight path perturbations, whereas a gull could extend its wrist to switch into an
unstable configuration, which would result in more active control over its flight characteristics.
This initial evaluation of avian dynamic flight stability was limited to small disturbances,
longitudinal flight, and a rigid body analysis. However, there are many more characteristics
that must be incorporated to develop a holistic understanding of avian flight maneuverability.
109
5.2 Next steps
In this work, I quantified the effect of the avian elbow and wrist on longitudinal stability
in a manner similar to traditional aircraft controls with the intention of identifying what as-
pects of bird flight control could inspire more maneuverable and adaptable UAVs. This thesis
specifically focuses on gliding flight and assumes non-elastic, quasi-steady characteristics in a
decoupled longitudinal plane. This approach lays the foundation for stability analysis in birds,
but future studies will need to incorporate lateral characteristics and the coupling between lon-
gitudinal and lateral characteristics. Needless to say, there are a multitude of paths that must
yet be explored to completely quantify and then evaluate avian flight stability, maneuverability,
and adaptability. Further, there are many steps that must be taken to effectively incorporate
a biological understanding of bird flight into future UAV designs. Here, I discuss three areas
of study that are necessary components to advance towards a complete understanding of bird
flight maneuverability.
110
addition, actuation of any wing joint requires sufficient control force to properly operate across
a broad range of flight conditions. Therefore, next steps should look towards designs that use
simplified, existing wing morphing mechanisms to shift between stable and unstable flight. By
beginning with an example case, perhaps with a simple wing sweep morphing design, we can
explore the feasibility and effectiveness of a more complicated bio-inspired design. In addition,
we can evaluate if being able to shift stability modes in flight is required to enable avian-like
maneuverability.
Throughout this work, I have approached both the aerodynamic and inertial characteristics
as if birds were rigid. This is certainly not the case. Bird feathers are flexible and result
in different loading conditions across different flight speeds [230]. However, my approach
allows for an initial understanding of the control provided by the wing joints and captures the
implications of morphing the overall wing shape changes.
There is a focus in the aeronautical discipline to understand the role of flexibility in aircraft
performance, stability, and control, however there is much left to understand about the role of
flexibility in bird flight. In particular, it will be an important next step to quantify the role of
flexibility in avian flight control. Specifically, it will be informative to identify how the control
effectiveness of the wing joints changes due to passive morphing at different flight conditions.
Aside from affecting control effectiveness, the passive response of the feathers may also serve
as a form of gust alleviation and when coupled with active morphing may be an integral part to
bird’s ability to adapt to various flight conditions.
To date, my work has entirely focused on gliding flight with the goal of laying a solid foun-
dational understanding. Gliding flight is a valuable first step as it simplifies a direct comparison
to fixed-wing UAVs. However, to truly advance our understanding of avian flight maneuver-
ability, flapping flight cannot be overlooked.
Flapping flight and its associated unsteady characteristics play a significant role in avian
111
flight. In fact, some species of birds are not known to glide but instead use bounding flight,
which involves repeatedly transitioning between flapping flight and tucking their wings against
their body [100]. Such birds may provide inspiration for novel maneuverable UAV designs and
we require more directed studies to better understand this method of flight. Substantial future
work is necessary to advance towards a holistic model of bird flight that accurately captures
the attributes of both gliding and flapping flight.
5.3 Summary
Here, I investigated how varying the avian elbow and wrist affected the longitudinal sta-
bility of a gliding bird. This analysis revealed that the elbow and the wrist range of motion
alone permits a wide range of longitudinal control and stability characteristics. In particular,
the ability to shift between stable and unstable flight was identified as a key characteristic that
should be considered for future UAV designs. In all, investigating how birds control their glid-
ing flight provides new inspiration to advance the design of more maneuverable and adaptable
UAVs.
112
Appendix A
AvInertia Implementation
This chapter provides a detailed breakdown of the assumptions and procedures used within
the development and implementation of AvInertia. The final outputs from AvInertia are re-
turned in the pre-selected origin and axis system: the bird (vehicle) reference point (VRP) and
“full bird” frame of reference (always right-handed axes). In our work, we selected the VRP
to be the location where the neck attaches to the torso (Fig. A.1) . This is approximately the
center of the spinal cord if cut at the clavicle. The x-axis points forwards along the center of the
bird, z-axis points ventrally and y-axis points along the right wing. The selection of the origin
and axis system is user-specific but must be consistently followed for all inputs. All measure-
ments input into the program should be defined relative to this same origin unless otherwise
noted in this document.
Multiple frames of reference are utilized throughout this program. Each individual section
will detail the appropriate frame of reference utilized. The wing is modeled as a composite
structure of bones, muscles, feathers and skin. When computing the final bird inertial properties
it is possible to model both symmetric and asymmetric wing shapes. In this work, we focused
solely on symmetric configurations.
2. Determine moment of inertia tensor (I) and center of gravity (CG) of each component
within a frame of reference and about an origin that simplifies their formulation.
3. Transform I and CG to be within the full bird frame frame of reference with the VRP as
113
Muscles joint
Bones peripheral
Primary feathers CG
x Secondary feathers
Skin/coverts/tertiary feathers
VRP y
7 6 Alula Radiale & Ulnare
12 1
4
3
8
2
11
10
Legs
Figure A.1: Dorsal view of entire bird modeled as a composite of simplified geometric
shapes.
the origin. This procedure is highly variable and the following sub-sections detail how
each component is transformed to be in this final frame of reference.
Note: Parallel axis theorem is only valid between an arbitrary point and the center of grav-
ity, not between two arbitrary points [179]. This was accounted for within the code.
3. Position of the wing defined by ten key landmarks (Fig. A.1). Note that the identity
of Pt11 varies among species; see “birdmeasurements readytorun.csv” in the publicly
available data repository for total feather counts. In addition, we do not include Pt5 and
Pt7 as these positions were not needed within the analysis. We did not renumber to avoid
114
confusion.
• Pt1: Humeral head • Pt9: Distal tip of the fourth-to-last pri-
mary feather (usually P7)
• Pt2: Center of the elbow joint
• Pt10: Distal tip of the first secondary
• Pt3: Center of the wrist joint
feather (S1)
• Pt4: Distal tip of the carpometacarpus
• Pt11: Distal tip of the final secondary
• Pt6: Wing leading edge ahead of the feather
wrist joint
• Pt12: The most proximal location
• Pt8: The distal tip of the final primary along the leading edge of the wing
feather (usually P10)
A.0.3 Assumptions
1. Base geometric shapes are the greatest assumption within this code although it is com-
monly used for estimating the inertial characteristics of complex objects [179]. Bio-
logical specimens are variable and the accuracy of this assumption will vary between
different species.
2. Wings were aligned so that the point on the wrist joint is in line with the shoulder along
the y-axis and along the x-axis. The wing was then rotated so that the point on the feather
tip at the wing root was at the same height (on the z-axis) as the shoulder. However, this
is not inherent to AvInertia and any wing alignment can be input.
115
x
bone
ri ro
frame
z
t end
l bone
Figure A.2: Simplified bone diagram including the referenced frames of reference.
A.1.1 Methodology
mbone
= 2(πro2 tend ) + π(ro 2 − ri 2 )(lbone − 2tend ) (A.1)
ρbone
2. Calculate the mass of each end cap and the hollow cylinder.
3. Determine I and CG of the hollow cylinder (eqn. A.5 and A.13) and two end caps (eqn.
A.6 and A.13) with respect to the bone specific frame of reference and origin (Fig. A.2).
7. Radiale and ulnare estimated as point masses on Pt3 within the full bird frame.
116
A.1.3 Assumptions
1. Density (ρbone ) treated as constant 2060 kg/m3 for all major bones [233].
5. CG is at the center of the measured bone length. This may differ slightly from that
extracted using the Optitrack markers. The start of the bone is assumed to be at the most
proximal Optitrack marker.
A.2.1 Methodology
1. Calculate the cylinder radius based on the mass and muscle density for the current group.
2. Determine I and CG of the cylinder (eqn. A.6 and A.13) within the bone frame of
reference and origin (Fig. A.2).
117
A.2.3 Assumptions
1. Density (ρmuscles ) was assumed to be 1100 kg/m−3 for all groups [235]. This is slightly
higher than 1060 kg/m−3 that was calculated for muscles alone [236] because we did
want to include the tendons and connective tissues in the overall calculation.
2. Muscles are stretched along the length. In reality, muscles will be more heavily grouped
to the start and end section but for simplicity we assumed a constant muscle width along
the bone length.
3. Radius of the muscles is determined based on the muscle group mass, estimated muscle
density and the length of the bones.
A.3.1 Methodology
1. Calculate the approximate vane mass assuming vanes are composed of solid cortex cylin-
drical barbs with a previously determined [239, 240] barb radius (rbarb ) and spacing
(dbarb ) per:
nbarbs = lvane /(dbarb + 2rbarb ), (A.2)
2
mvane = ρcortex nbarbs wvane πrbarb . (A.3)
2. Substract the proximal and distal vane masses from the total feather mass mf to deter-
mine the total mass of the rachis and calamus (mrc ).
118
Vane Rachis Calamus
z Θvane l cal
z
distal proximal
vane frame vane frame
l vane
y l vane ri
y ro
rachis
frame
y
w vane
z z ri
ro
calamus feather
frame frame
y
y
Figure A.3: Simplified feather diagram including the referenced frames of reference.
3. Calculate the inner radius (rical ) of the calamus and width of the interior rachis pyramid
by ensuring that mrc is equal to the mass predicted by the volume and density of the
calamus and rachis components:
4 4
mrc = ρcor (π(ro2cal − ri2cal )lcal + (ro2cal − ri2cal )lvane ) + ρmed ( ri2cal lvane ). (A.4)
3 3
4. Determine I and CG of the calamus (eqn. A.6 and A.13) within the calamus frame
assuming a hollow cylinder.
5. Determine I and CG of the rachis (eqn. A.10 and A.14) within the rachis frame assuming
exterior cortex and interior medullary retangular pyramids.
119
6. Determine I and CG of each vane (eqn. A.11) within their respective vane frames as-
suming flat rectangular plates with a mass calculated in Step 1.
9. Transform rachis and vane I and CG to be expressed in the calamus frame, (rotate by
θvane ) with the origin as the start of the feather.
10. Combine I and CG for the rachis and vane with the calamus components.
11. Transform I and CG to be expressed in the feather frame. Up to this point all of this
inertial data can be computed with no knowledge of the current wing positioning. For this
reason, the code has two seperate functions relating to the feather inertial calculations.
12. Transform I and CG to be expressed in the full bird frame using information about each
individual feather positioning and orientation.
13. Alula feathers estimated as a point mass on Pt6 within the full bird frame.
4. Outer radius of the calamus, rocal 8. Interior angle between calamus and rachis,
θvane
A.3.3 Assumptions
1. Density of cortex (ρcor )[238, 241–244] and of medullary (ρmed )[242, 245] material treated
as constant 1150 kg/m3 and 80 kg/m3 , respectively.
120
2. Shape of feathers was assumed to be constant within a species. We measured the shape
properties for only one specimen but, individually measured the mass of each specimen’s
feathers. Then we assumed isometric scaling to adjust the length and area measurements
for each feather as necessary.
3. Length of medullary part of the rachis extends all the way to the feather tip [238].
4. Mass of vanes is based on previously measured barb radii and distance between barbs
[239, 240].
5. Proximal and distal vane barb properties are treated as constant. Note that previous
work did find slight but measurable differences between the vanes that will be neglected
in this work[246].
6. Alula feathers are treated as a point mass on Pt6 although their structure differs between
species.
7. Feather positioning:
• The base of the secondaries are equally spaced along the ulna and their tips are
equally spaced along the line between Pt10 and Pt 11 (last secondary).
• The base of P1 through P6 are equally spaced along the carpometacarpus and their
tips are equally spaced along the line between Pt10 and Pt9.
• The base P7 and up are located at the end of the carpometacarpus (Pt4) and their
tips are equally spaced along the line between Pt9 and Pt8.
8. Feather orientation:
• Primaries lay flat on the plane defined by Pt3, Pt4 and their tip position as defined
above.
• Secondaries lay flat on the plane defined by Pt2, Pt3 and their tip position as defined
above.
121
A.4 Tertiaries and skin/coverts
The tertiary feathers and skin/coverts are modeled as flat triangular plates. The tertiary
feather sections are defined as two sections, with vertices as follows: 1) Pt12, Pt2 and the
trailing edge of the wing at the body and 2) Pt11, Pt2 and the trailing edge of the wing at the
body. The skin/coverts section vertices are defined by Pt12, Pt2 and Pt6.
A.4.1 Methodology
1. Given the input positions calculate I and CG based on the general polygon formulations
[247] within a frame of reference and about an origin that simplifies their formulation.
2. Transform I and CG to be within the full bird frame and shift so that the VRP is the
origin.
A.4.3 Assumptions
1. Skin density (ρskin ) treated as constant (1060 kg/m3 ) based on a previously measured
muscle-only measurement [236, 248]. This was used to calculate the final skin thickness
based on ensuring that the volume would return the known mass of the section.
2. Tertiary density is treated as constant and equal to the cortex density (ρcor )[238, 241–
244] of 1150 kg/m3 . As with the skin this was used to calculate the final thickness of the
tertiary sections based on ensuring that the volume would return the known mass of the
section.
122
A.5 Head, neck, legs and tail
The head (including the beak) was modeled as a solid cone, the neck was modeled as a
solid cylinder, the legs were point masses and the tail was modeled as a flat rectangular plate.
A.5.1 Methodology
1. Calculate I and CG of the head (eqn. A.8 and eqn. A.14), neck (if used, eqn. A.6 and
eqn. A.13) and tail (eqn. A.11).
2. Calculate I and CG of the legs as point masses placed on the ventral sides of the bird.
3. Transform I and CG to be within the full bird frame and shift so that the VRP is the
origin.
2. Length of the head (tip of beak to neck), lhead 8. Length of the furled tail, ltail
3. Radius of the head (maximum), rhead 9. Width of the furled tail, wtail
4. Mass of the neck, mneck 10. Length of the torso + tail, ltot
5. Length of the neck outstretched, lneck 11. Mass of both legs, mleg
A.5.3 Assumptions
1. Head/beak CG was measured on the specimens and we found that for all of the mea-
sured species the head CG was within 15% of the quarter of the head length (See “Veri-
ficationData.xlsx” tab “HeadCGVerification” in the publicly available data repository).
Thus, we assumed that a solid cone would be a fair approximation of the shape.
123
2. Neck was only included for some species that are known to stretch out their neck while
in a cruise flight configuration. If not outstretched, the neck mass is added to the head
mass. This can be adjusted for general use.
3. Legs modeled as point masses however some birds do stick their legs behind their body
while in flight. Because grebes have very minimal tails and fly with their legs directly
behind their bodies the grebes’ legs were treated as a tail. However, for other species this
effect was neglected in our work and should be investigated in future studies.
A.6 Torso
The torso is modeled as a structure composed of a hemiellipsoid, partial elliptical cone and
either a full elliptical cone (3 individuals) or elliptical cylinder (33 individuals) (Fig. A.1). Due
to the complex structure, the center of gravity position of the torso + tail must be measured and
thus the code largely functions to estimate the associated moment of inertia for the torso. In
the study, we performed a sensitivity analysis for up to 15% error of the total torso length on
the CG measurement.
A.6.1 Methodology
1. Calculate the volume of each component assuming an elliptical cylinder for the back
piece.
2. Option 1: If the calculated average density places the CG within 5% of the measured
value, use the elliptical cylinder and continue to step 5.
3. Option 2: If not, calculate if the average density using an elliptical cone for a back piece
places the CG within 5% of the measured value. If so, use the elliptical cone and continue
to step 5. Figure A.1 illustrates option 2.
4. Option 3: If not, use an elliptical cylinder and an optimization routine to vary the density
between each of the three sections of the torso. The routine minimizes the difference
between the output densities and the calculated average density for the full torso. In
124
addition, we assume that the CG is moved forwards by 5% of the total length. This is
necessary to ensure reasonable densities. We assumed that density measurements could
be very low due to the possibility that the majority of the volume is made from loosely
packed feathers. The lowest density was found to be 42 kg-m−3 for the end section of
a storm petrel. All section densities can be seen within the “VerificationData.xlsx” tab
“TorsoDensities” that is included within the publicly available data repository.
5. One of the three above options will provide the final calculated volume, mass and output
CG.
6. Given these parameters for each section, calculate I (eqn. A.12, A.9 and A.7) and CG
(eqn. A.15 and A.14) for each component in the torso frame, where ztorso = −xf ullbird ,
xtorso = zf ullbird and ytorso = yf ullbird .
7. Transform I and CG to be within the full bird frame and shift so that the VRP is the
origin.
1. Mass of the torso and legs, mtorso 6. Maximum body width, wmax
4. Body width at leg insertion, wleg 9. x-location of the CG of torso + legs, CGx
5. x-location of leg insertion, lleg 10. z-location of the CG of torso + legs, CGz
A.6.3 Assumptions
1. Minimum density allowed by the optimizer for the front section is 200 kg/m3 .
125
A.7 Base moment of inertia tensors
All tensors are listed in a published handbook [249] except for the elliptical cone calculation
which was computed for this project.
1. Hollow cylinder. ro is the outer radius, ri is the inner radius, l is the length and m is
the mass. Origin is at the center of mass. Used for the bone interior, neck and feather
calamus.
1
(3(ro2 + ri2 ) +l ) 2
0 0
12
1
I = m (3(ro2 ri2 ) 2 (A.5)
0 12
+ +l ) 0
1 2 2
0 0 (r + ri )
2 o
2. Solid cylinder. r is the radius, l is the length and m is the mass. Origin is at the center
of mass. Used for the bone end caps and muscles.
1
(3r2 + l2 ) 0 0
12
1
I = m (3r2 2 (A.6)
0 12
+l ) 0
1 2
0 0 2
r
3. Elliptical cylinder. a is half the maximum height along the x direction, b is half the
maximum width along the y direction, l is the length, and m is the mass. Origin is at the
center of mass. Used for the last portion of the body if required.
1
(3b2 +l ) 2
0 0
12
1
I = m (3a2 2 (A.7)
0 12
+l ) 0
1 2 2
0 0 4
(a + b )
4. Solid cone. r is the radius of the cone base, h is the height and m is the mass. Origin is
126
at the center of the cone’s base, not the center of mass. Used for the head/beak.
1 3 2
( r + l2 ) 0 0
10 2
I = m
1 3 2 2
(A.8)
0 ( r
10 2
+l ) 0
3 2
0 0 10
r
5. Elliptical cone. l length until the tip of the cone, A is half the maximum height along
the x direction, B is half the maximum width along the y direction, and m is the mass.
Origin is at the base of the cone, not the center of mass. Used for the back two-thirds of
the body.
1 3 2 2
( B +l ) 0 0
10 2
1 3 2
I = m + l2 ) (A.9)
0 ( A
10 2
0
3 2 2
0 0 20
(A + B )
6. Solid square pyramid. w is the entire width of one side of the pyramid base, h entire
height of the pyramid and m is the mass. Origin is at the center of the pyramid’s base,
not the center of mass. Used for the rachis.
1
(w2 + 2h ) 2
0 0
20
1
I = m (w2 2 (A.10)
0 20
+ 2h ) 0
1 2
0 0 10
w
7. Flat rectangular plate. w is the entire width of one side, h entire height and m is the
mass. Origin is at the center of mass. Used for the feather vanes and tail.
2 2
(w + h ) 0 0
1 2
I = m (A.11)
0 h 0
12
2
0 0 w
8. Solid hemi-ellipsoid. a is half the height along the x direction, b is half the width along
the y direction and c is half the length along the z direction. Origin is at the base of the
127
hemi-ellipsoid, not the center of mass. Used for the front third of the body.
2 2
(b + c ) 0 0
1 2 2
I = m (A.12)
0 (a + c ) 0
5
0 0 (a2 + b2 )
2. Pyramid. (circular, square or elliptical) h is the height. Origin at the center of the base.
0
CG = 0 (A.14)
1
4
h
128
A.9 Illustrations of the required measurements
Figure A.4: Entire bird measurements. Drawn by J.C.M. Wong for [142].
humerus mass
mbone
mass of
primaries
mf
mass of
tertiaries radius mass ulna mass mass of carpometacarpus
mtertiaries mbone secondaries mass
mbone
mf mbone
Figure A.5: Wing specific measurements. Drawn by J.C.M. Wong for [142].
129
vehicle reference point max body
(VRP) height
hmax
x-location of
max body width x-location of
lbmax max body width leg insertion
wmax lleg
body width at
x leg insertion x
wleg
full torso
length
y ltot z
mass of
the head diameter
mhead length of length of of the head
the head the neck 2rhead
lhead lneck mass of
the neck
mneck
diameter
mass of of the neck
the legs 2rneck length of mass of the
mlegs the torso torso and legs
ltorso mtrue
mass of
the tail length of
mtail the tail
ltail
Figure A.6: Body specific measurements. Drawn by J.C.M. Wong for [142].
130
Bibliography
[1] N. Tesla. (1898). Method of and apparatus for controlling mechanism of moving vessels
or vehicles. USA Patent, US613809A. URL: https://patents.google.com/patent/US6138
09A/en.
[2] L. R. Newcome. (2004). Unmanned aviation: a brief history of unmanned aerial vehi-
cles. American Institute of Aeronautics and Astronautics, Reston, VA, USA.
[4] H. Everett. (2015). Unmanned systems of World Wars I and II. MIT Press.
[5] J. F. Keane, S. S. Carr. (2013). A brief history of early unmanned aircraft. Johns Hopkins
APL Technical Digest, 32(3), 558–571.
[7] K. L. B. Cook. (2007). The Silent Force Multiplier: The History and Role of
UAVs in Warfare, in: 2007 IEEE Aerospace Conference, IEEE, 1–7. doi:
10.1109/AERO.2007.352737.
131
[9] T. F. Villa, F. Gonzalez, B. Miljievic, Z. D. Ristovski, L. Morawska. (2016). An
Overview of Small Unmanned Aerial Vehicles for Air Quality Measurements: Present
Applications and Future Prospectives. Sensors, 16(7). doi: 10.3390/s16071072.
[12] E. Salamı́, C. Barrado, E. Pastor. (2014). UAV Flight Experiments Applied to the Re-
mote Sensing of Vegetated Areas. Remote Sensing, 6(11). doi: 10.3390/rs61111051.
[13] J. Kim, S. Kim, C. Ju, H. I. Son. (2019). Unmanned Aerial Vehicles in Agriculture: A
Review of Perspective of Platform, Control, and Applications. IEEE Access, 7, 105100–
105115. doi: 10.1109/ACCESS.2019.2932119.
[14] S. M. Adams, C. J. Friedland. (2011). A survey of unmanned aerial vehicle (UAV) us-
age for imagery collection in disaster research and management, in: 9th International
Workshop on Remote Sensing for Disaster Response, 8, 1–8.
132
optimization of Nigeria vaccine supply chain. International Journal of Scientific & En-
gineering Research, 10(10), 1273–1279.
[18] I. Sung, P. Nielsen. (2020). Zoning a service area of unmanned aerial vehicles for pack-
age delivery services. Journal of Intelligent & Robotic Systems, 97(3), 719–731. doi:
10.1007/s10846-019-01045-7.
[20] R. Dudley. (2002). Mechanisms and implications of animal flight maneuverability. Inte-
grative and Comparative Biology, 42(1), 135–140. doi: 10.1093/icb/42.1.135.
[21] F. Gill, D. Donsker, P. Rasmussen. (2021). IOC World Bird List (v11.1). URL: 10.143
44/IOC.ML.11.1.
133
[26] A. Hedenström, M. Rosén. (2001). Predator versus prey: on aerial hunting and escape
strategies in birds. Behavioral Ecology, 12(2), 150–156. doi: 10.1093/beheco/12.2.150.
[30] A. A. Paranjape, S.-J. Chung, M. S. Selig. (2011). Flight mechanics of a tailless artic-
ulated wing aircraft. Bioinspiration & Biomimetics, 6(2), 026005. doi: 10.1088/1748-
3182/6/2/026005.
[31] M. Abdulrahim, R. Lind. (2004). Flight testing and response characteristics of a variable
gull-wing morphing aircraft, in: AIAA Guidance, Navigation, and Control Conference
and Exhibit, American Institute of Aeronautics and Astronautics, Providence, RI, USA.
134
[35] R. V. Baudinette, K. Schmidt-Nielsen. (1974). Energy cost of gliding flight in herring
gulls. Nature, 248, 83–84. doi: 10.1038/248083b0.
[36] P. L. Richardson. (2011). How do albatrosses fly around the world without flapping their
wings? Progress in Oceanography, 88, 46–58. doi: 10.1016/j.pocean.2010.08.001.
[38] R. Spaar, B. Bruderer. (1997). Optimal flight behavior of soaring migrants: a case study
of migrating steppe buzzards, Buteo buteo vulpinus. Behavioral Ecology, 8(3), 288–297.
doi: 10.1093/beheco/8.3.288.
[41] T. Alerstam. (1975). Crane (Grus grus) migration over sea and land. Ibis, 117(4), 489–
495. doi: 10.1111/j.1474-919X.1975.tb04241.x.
[43] R. Austin. (2011). Unmanned aircraft systems: UAVS design, development and deploy-
ment. John Wiley & Sons Ltd.
[44] J. Bäckman, T. Alerstam. (2001). Confronting the winds: orientation and flight be-
haviour of roosting swifts, Apus apus. Proceedings of the Royal Society of London.
Series B: Biological Sciences, 268(1471), 1081–1087. doi: 10.1098/rspb.2001.1622.
135
[45] M. A. Barcala-Montejano, A. A. Rodrı́guez-Sevillano, J. Crespo-Moreno, R. Bardera-
Mora, A. J. Silva-González. (2015). Optimized performance of a morphing micro air
vehicle, in: 2015 International Conference on Unmanned Aircraft Systems (ICUAS),
794–800. doi: 10.1109/ICUAS.2015.7152363.
[50] A. A. Demircali, H. Uvet. (2018). Mini Glider Design and Implementation with Wing-
Folding Mechanism. Applied Sciences, 8(9), 1541. doi: 10.3390/app8091541.
[53] D. Edwards. (2008). Performance Testing of RNR’s SBXC Using a Piccolo Autopilot.
URL: http://www.xcsoaring.com/techPicts/Edwards%20performance%20test.pdf.
136
[54] D. J. Edwards, A. D. Kahn, S. B. Heinzen, T. Z. Young, N. J. Arnold, D. Newton,
B. Eber, S. V. Carter. (2018). CICADA Flying Circuit Board Unmanned Aerial Vehi-
cle, in: 2018 AIAA Aerospace Sciences Meeting, Kissimmee, FL, USA, 1008. doi:
10.2514/6.2018-1008.
[55] J. Grasmeyer, M. Keennon. (2001). Development of the black widow micro air vehicle,
in: 39th Aerospace Sciences Meeting and Exhibit, American Institute of Aeronautics
and Astronautics, Reno, NV, USA, 127. doi: 10.2514/6.2001-127.
[58] M. Hacklinger. (1964). Theoretical and experimental investigation of indoor flying mod-
els. The Aeronautical Journal, 68(647), 728–734. doi: 10.1017/S0368393100080846.
[59] T. Hansen. (2014). Modeling the Performance of the Standard Cirrus Glider using
Navier-Stokes CFD. Technical Soaring, 38(1), 5–14.
[61] A. Hedenström. (1995). Song flight performance in the skylark Alauda arvensis. Journal
of Avian Biology, 26(4), 337–342.
[62] A. Hedenström, S. Åkesson. (2017). Adaptive airspeed adjustment and compensation for
wind drift in the common swift: differences between day and night. Animal Behaviour,
127, 117–123. doi: 10.1016/j.anbehav.2017.03.010.
[63] T. L. Hedrick, C. Pichot, E. De Margerie. (2018). Gliding for a free lunch: biomechan-
ics of foraging flight in common swifts (Apus apus). Journal of Experimental Biology,
221(22), jeb186270. doi: 10.1242/jeb.186270.
137
[64] R. Hoey. (2010). Exploring bird aerodynamics using radio-controlled models. Bioinspi-
ration & Biomimetics, 5(4), 045008. doi: 10.1088/1748-3182/5/4/045008.
[65] N. Horvitz, N. Sapir, F. Liechti, R. Avissar, I. Mahrer, R. Nathan. (2014). The gliding
speed of migrating birds: slow and safe or fast and risky? Ecology Letters, 17(6), 670–
679. doi: 10.1111/ele.12268.
[66] P. Idrac. (1920). Soaring flight in Guinea. Technical Report NACA-TN-13, National
Advisory Committee for Aeronautics. Langley Aeronautical Lab., Langley Field, VA,
United States.
[67] A. R. Jenkins. (1995). Morphometrics and Flight Performance of Southern African Pere-
grine and Lanner Falcons. Journal of Avian Biology, 26, 49–58. doi: 10.2307/3677212.
[68] W. Jin, Y.-G. Lee. (2014). Computational analysis of the aerodynamic performance of a
long-endurance UAV. International Journal of Aeronautical and Space Sciences, 15(4),
374–382. doi: 10.5139/IJASS.2014.15.4.374.
[69] S. Junlei, Z. Zhou, W. Heping, L. Shan. (2017). The Conceptual Design and
Aerodynamic Characteristics Analysis of the Diamond Joined-Wing Configura-
tion UAV, in: 2017 5th International Conference on Mechanical, Automotive
and Materials Engineering (CMAME), IEEE, Guangzhou, China, 275–279. doi:
10.1109/CMAME.2017.8540111.
[71] J. Kosmatka. (2007). Development of a long-range small UAV for atmospheric mon-
itoring, in: 48th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and
Materials Conference, American Institute of Aeronautics and Astronautics, Honolulu,
HI, USA, 2234. doi: 10.2514/6.2007-2234.
138
[72] M. Kovač, J.-C. Zufferey, D. Floreano. (2009). Towards a self-deploying and gliding
robot, in: Flying insects and robots, Springer, 271–284.
[73] R. A. Kroeger, H. D. Grushka, T. C. Helvey. (1972). Low speed aerodynamics for ultra-
quiet flight. Technical Report AD893426, The University of Tennessee Space Institute,
Tullhoma, Tennessee. URL: https://apps.dtic.mil/sti/pdfs/AD0893426.pdf.
[74] L. K. Loftin. (1985). Quest for performance: The evolution of modern aircraft. Scientific
and Technical Information Branch, National Aeronautics and Space Administration.
[76] J. McGahan. (1973). Gliding flight of the Andean condor in nature. Journal of Experi-
mental Biology, 58(1), 225–237. doi: 10.1242/jeb.58.1.225.
[77] S. K. Meyer, R. Spaar, B. Bruderer. (2000). To cross the sea or to follow the coast?
Flight directions and behaviour of migrating raptors approaching the Mediterranean Sea
in autumn. Behaviour, 137(3), 379–399.
[78] R. A. Meyers. (1993). Gliding flight in the American kestrel (Falco sparverius):
an electromyographic study. Journal of Morphology, 215(3), 213–224. doi:
10.1002/jmor.1052150304.
139
[81] P. Panagiotou, P. Kaparos, K. Yakinthos. (2014). Winglet design and optimization for
a MALE UAV using CFD. Aerospace Science and Technology, 39, 190–205. doi:
10.1016/j.ast.2014.09.006.
[83] C. J. Pennycuick. (1971). Gliding flight of the white-backed vulture Gyps africanus.
Journal of Experimental Biology, 55(1), 13–38. doi: 10.1242/jeb.55.1.13.
[85] C. J. Pennycuick. (1968). A wind-tunnel study of gliding flight in the pigeon Columba
livia. Journal of Experimental Biology, 49(3), 509–526. doi: 10.1242/jeb.49.3.509.
[87] A. Raspet. (1960). Biophysics of bird flight. Science, 132, 191–200. doi: 10.1126/sci-
ence.132.3421.191.
[90] M. Rosén, A. Hedenström. (2002). Soaring Flight in the Eleonora’s Falcon (Falco
eleonorae). The Auk, 119(3), 835–840. doi: 10.1093/auk/119.3.835.
140
[91] M. Rosén, G. Spedding, A. Hedenström. (2007). Wake structure and wingbeat kinemat-
ics of a house-martin Delichon urbica. Journal of The Royal Society Interface, 4(15),
659–668. doi: 10.1098/rsif.2007.0215.
[92] M. Rosén, A. Hedenstrom. (2001). Gliding flight in a jackdaw: a wind tunnel study.
Journal of Experimental Biology, 204(6), 1153–1166. doi: 10.1242/jeb.204.6.1153.
[94] E. Sarradj, C. Fritzsche, T. Geyer. (2011). Silent owl flight: Bird flyover noise measure-
ments. AIAA Journal, 49(4), 769–779. doi: 10.2514/1.J050703.
[95] J. Shamoun-Baranes, E. van Loon. (2006). Energetic influence on gull flight strategy se-
lection. Journal of Experimental Biology, 209(18), 3489–3498. doi: 10.1242/jeb.02385.
[98] G. R. Spedding. (1987). The wake of a kestrel (Falco tinnunculus) in gliding flight.
Journal of Experimental Biology, 127(1), 45–57. doi: 10.1242/jeb.127.1.45.
141
[100] B. Tobalske, K. Dial. (1996). Flight kinematics of black-billed magpies and pigeons
over a wide range of speeds. Journal of Experimental Biology, 199(2), 263–280. doi:
10.1242/jeb.199.2.263.
[102] B. W. Tobalske. (1996). Scaling of muscle composition, wing morphology, and intermit-
tent flight behavior in woodpeckers. The Auk, 113(1), 151–177. doi: 10.2307/4088943.
[103] B. Tobalske. (1995). Neuromuscular control and kinematics of intermittent flight in the
European starling (Sturnus vulgaris). Journal of Experimental Biology, 198(6), 1259–
1273. doi: 10.1242/jeb.198.6.1259.
[104] A. Tomaszewski, Z. J. Goraj. (2018). Assessment of a small UAV speed polar graph
by conducting flight tests. Aircraft Engineering and Aerospace Technology, 91(5). doi:
10.1108/AEAT-03-2018-0099.
[105] V. A. Tucker. (1987). Gliding Birds: The Effect of Variable Wing Span. Journal of
Experimental Biology, 133(1), 33–58. doi: 10.1242/jeb.133.1.33.
[107] J. Wakeling, C. P. Ellington. (1997). Dragonfly flight. I. Gliding flight and steady-
state aerodynamic forces. Journal of Experimental Biology, 200(3), 543–556. doi:
10.1242/jeb.200.3.543.
[108] A. Welch. (1973). Development of the competition glider. The Aeronautical Journal,
77(756), 612–619. doi: 10.1017/S0001924000042056.
[109] A. Welch. (1995). Evolution of the high performance glider. The Aeronautical Journal,
99(986), 243–259. doi: 10.1017/S0001924000049630.
142
[110] C. White, S. Watkins, E. W. Lim, K. Massey. (2012). The soaring potential of a micro
air vehicle in an urban environment. International Journal of Micro Air Vehicles, 4(1),
1–13. doi: 10.1260/1756-8293.4.1.1.
[113] T. Wolf, R. Konrath. (2015). Avian wing geometry and kinematics of a free-flying barn
owl in flapping flight. Experiments in Fluids, 56(2), 28. doi: 10.1007/s00348-015-1898-
6.
[114] W. Biesel, H. Butz, W. Nachtigall. (1985). Erste Messungen der Flügelgeometrie bei frei
gleitfliegenden Haustauben (columbia livia var. domestica) unter Benutzung neu aus-
gearbeiteter Verfahren der Windkanaltechnik und der Stereophotogrammetrie. Biona-
report, 3, 139–160.
[115] J. D. Anderson Jr. (1990). Modern compressible flow: with historical perspective.
McGraw-Hill New York.
[118] B. Carmichael. (1981). Low Reynolds number airfoil survey, volume 1. Technical
Report NASA-CR-165803-VOL-1, National Aeronautics and Space Administration,
Capistrano Beach, CA, USA. URL: https://ntrs.nasa.gov/citations/19820006186.
143
[119] P. Lissaman. (1983). Low Reynolds number airfoils. Annual Review of Fluid Mechanics,
15(1), 223–239. doi: 10.1146/annurev.fl.15.010183.001255.
[120] D. Lentink, R. de Kat. (2014). Gliding Swifts Attain Laminar Flow over Rough Wings.
PLoS ONE, 9(6), e99901. doi: 10.1371/journal.pone.0099901.
[121] P. Butler, A. Woakes. (1980). Heart rate, respiratory frequency and wing beat frequency
of free flying barnacle geese Branta leucopsis. Journal of Experimental Biology, 85(1),
213–226. doi: 10.1242/jeb.85.1.213.
[122] C. Harvey, D. J. Inman. (2021). Aerodynamic efficiency of gliding birds vs. compara-
ble UAVs: a review. Bioinspiration & Biomimetics, 16(3), 031001. doi: 10.1088/1748-
3190/abc86a.
[124] V. A. Tucker, G. C. Parrott. (1970). Aerodynamics of gliding flight in a falcon and other
birds. Journal of Experimental Biology, 52, 345–367. doi: 10.1242/jeb.52.2.345.
[126] V. B. Baliga, I. Szabo, D. L. Altshuler. (2019). Range of motion in the avian wing
is strongly associated with flight behavior and body mass. Science Advances, 5(10),
eaaw6670. doi: 10.1126/sciadv.aaw6670.
144
[128] C. Harvey, V. B. Baliga, P. Lavoie, D. L. Altshuler. (2019). Wing morphing allows gulls
to modulate static pitch stability during gliding. Journal of The Royal Society Interface,
16(150), 20180641. doi: 10.1098/rsif.2018.0641.
[131] N. Shubin, C. Tabin, S. Carroll. (1997). Fossils, genes and the evolution of animal limbs.
Nature, 388(6643), 639–648. doi: 10.1038/41710.
[132] H. I. Fisher. (1957). Bony mechanism of automatic flexion and extension in the pigeon’s
wing. Science, 126(3271), 446–446. doi: 10.1126/science.126.3271.446.a.
[134] T. L. Hieronymus. (2016). Flight feather attachment in rock pigeons (Columba livia):
covert feathers and smooth muscle coordinate a morphing wing. Journal of Anatomy,
229(5), 631–656. doi: 10.1111/joa.12511.
145
[137] A. L. R. Thomas, G. K. Taylor. (2001). Animal flight dynamics I. Sta-
bility in gliding flight. Journal of Theoretical Biology, 212, 399–424. doi:
http://dx.doi.org/10.1006/jtbi.2001.2387.
[139] R. C. Nelson. (1998). Flight stability and automatic control. volume 2, WCB/McGraw
Hill New York.
[140] J. D. Anderson Jr. (1989). Introduction to Flight. 3 ed., McGraw-Hill Higher Education.
[143] J. Valasek. (2012). Morphing aerospace vehicles and structures. John Wiley & Sons.
[144] S. Mintchev, D. Floreano. (2016). Adaptive morphology: A design principle for multi-
modal and multifunctional robots. IEEE Robotics & Automation Magazine, 23(3), 42–
54. doi: 10.1109/MRA.2016.2580593.
[146] P. Lissaman. (2009). Effects of turbulence on bank upsets of small flight vehicles,
in: 47th AIAA Aerospace Sciences Meeting including The New Horizons Forum and
Aerospace Exposition, Aerospace Sciences Meetings, American Institute of Aeronau-
tics and Astronautics. doi: 10.2514/6.2009-65.
146
[147] M. W. Orr, S. J. Rasmussen, E. D. Karni, W. B. Blake. (2005). Framework for developing
and evaluating MAV control algorithms in a realistic urban setting, in: Proceedings of
the 2005, American Control Conference, 2005, IEEE, Portland, OR, USA, 6, 4096–
4101. doi: 10.1109/ACC.2005.1470619.
[152] AeroLab. (2020). MachUpX: Fast and accurate aerodynamic modelling using lifting-
line theory. URL: https://www.github.com/usuaero/MachUpX.
[155] T. L. Hedrick. (2008). Software techniques for two- and three-dimensional kinematic
measurements of biological and biomimetic systems. Bioinspiration & Biomimetics,
3(3), 034001. doi: 10.1088/1748-3182/3/3/034001.
147
[156] S. Lee, H. Choi. (2017). Characteristics of the alula in relation to wing and body
size in the Laridae and Sternidae. Animal Cells and Systems, 21(1), 63–69. doi:
10.1080/19768354.2016.1266287.
[157] R. Nudds, J. Rayner. (2006). Scaling of body frontal area and body width in birds. Jour-
nal of Morphology, 267(3), 341–346. doi: 10.1002/jmor.10409.
[158] T. Liu, K. Kuykendoll, R. Rhew, S. Jones. (2006). Avian Wing Geometry and Kinemat-
ics. AIAA Journal, 44(5), 954–963. doi: 10.2514/1.16224.
[162] M. Drela. (1989). XFOIL: An Analysis and Design System for Low Reynolds Number
Airfoils, in: T. J. Mueller (Ed.), Low Reynolds Number Aerodynamics, volume 54 of
Lecture Notes in Engineering, Springer, Berlin, Germany, 1–12.
[163] J. B. Barlow, W. H. Rae, A. Pope. (1999). Low-speed wind tunnel testing. John Wiley &
Sons.
[164] JCGM. (2008). Evaluation of measurement data - Guide to the expression of uncer-
tainty in measurement. Technical Report JCGM 100:2008, Joint Committee for Guides
in Metrology (JCGM/WG 1).
148
[166] J. van Ingen. (2008). The eN Method for Transition Prediction. Historical Review of
Work at TU Delft, in: 38th Fluid Dynamics Conference and Exhibit, Fluid Dynamics
and Co-located Conferences, American Institute of Aeronautics and Astronautics. doi:
10.2514/6.2008-3830.
[168] D. McLean. (2012). Understanding aerodynamics: arguing from the real physics. John
Wiley & Sons.
[169] R. Guiler, W. Huebsch. (2005). Wind tunnel analysis of a morphing swept wing tail-
less aircraft, in: 23rd AIAA Applied Aerodynamics Conference, American Institute of
Aeronautics and Astronautics, Toronto, ON, Canada, 4981. doi: 10.2514/6.2005-4981.
[170] S. Nakagawa, H. Schielzeth. (2013). A general and simple method for obtaining R2
from generalized linear mixed-effects models. Methods in Ecology and Evolution, 4(2),
133–142. doi: 10.1111/j.2041-210x.2012.00261.x.
149
[175] D. R. Warrick. (1998). The turning-and linear-maneuvering performance of birds: the
cost of efficiency for coursing insectivores. Canadian Journal of Zoology, 76(6), 1063–
1079. doi: 10.1139/z98-044.
[178] J. M. Smith. (1952). The importance of the nervous system in the evolution of animal
flight. Evolution, 6(1), 127–129. doi: 10.2307/2405510.
[183] G. Ducci, V. Colognesi, G. Vitucci, P. Chatelain, R. Ronsse. (2021). Stability and Sen-
sitivity Analysis of Bird Flapping Flight. Journal of Nonlinear Science, 31(2), 47. doi:
10.1007/s00332-021-09698-1.
150
[184] R. Mills, H. Hildenbrandt, G. K. Taylor, C. K. Hemelrijk. (2018). Physics-based simula-
tions of aerial attacks by peregrine falcons reveal that stooping at high speed maximizes
catch success against agile prey. PLOS Computational Biology, 14(4), e1006044. doi:
10.1371/journal.pcbi.1006044.
[190] C. Berg, J. Rayner. (1995). The moment of inertia of bird wings and the inertial power
requirement for flapping flight. Journal of Experimental Biology, 198(8), 1655–1664.
doi: 10.1242/jeb.198.8.1655.
[191] S. J. Kirkpatrick. (1990). Short communication the moment of inertia of bird wings.
Journal of Experimental Biology, 151(1), 489–494. doi: 10.1242/jeb.151.1.489.
[192] C. J. Pennycuick, A. Lock. (1976). Elastic energy storage in primary feather shafts.
Journal of Experimental Biology, 64(3), 677. doi: 10.1242/jeb.64.3.677.
[193] J. D. Hadfield. (2010). MCMC methods for multi-response generalized linear mixed
models: the MCMCglmm R package. Journal of Statistical Software, 33(2), 1–22. doi:
10.18637/jss.v033.i02.
151
[194] M. S. Ben-Shachar, D. Lüdecke, D. Makowski. (2020). effectsize: Estimation of Ef-
fect Size Indices and Standardized Parameters. Journal of Open Source Software, 5(56),
2815. doi: 10.21105/joss.02815.
[195] J. Cohen. (1973). Eta-Squared and Partial Eta-Squared in Fixed Factor Anova
Designs. Educational and Psychological Measurement, 33(1), 107–112. doi:
10.1177/001316447303300111.
[196] W. Maybury, J. Rayner, L. Couldrick. (2001). Lift generation by the avian tail. Proceed-
ings of the Royal Society of London. Series B: Biological Sciences, 268(1475), 1443–
1448. doi: 10.1098/rspb.2001.1666.
[199] A. H. Yates. (1952). Notes on the Mean Aerodynamic Chord and the Mean Aerodynamic
Centre of a Wing. The Journal of the Royal Aeronautical Society, 56(498), 461–474. doi:
10.1017/S0368393100129311.
[200] W. S. Diehl. (1942). The mean aerodynamic chord and the aerodynamic center of a ta-
pered wing. Technical Report NACA-TR-751, National Advisory Committee for Aero-
nautics. URL: https://ntrs.nasa.gov/citations/19930091829.
[201] W. Phillips, R. Niewoehner. (2009). Characteristic length and dynamic time scale
associated with aircraft pitching motion. Journal of Aircraft, 46(2), 572–582. doi:
10.2514/1.38724.
152
American Institute of Aeronautics and Astronautics, Palm Springs, CA, USA, 1727. doi:
10.2514/6.2004-1727.
[207] J. R. Rivera Parga. (2004). Wind Tunnel Investigation of the Static Stability and Control
Effectiveness of a Rotary Tail in a Portable UAV. Master’s thesis, Air Force Institute of
Technology, Wright-Patterson Air Force Base, Ohio.
[209] R. Hewson. (1957). Social flying in ravens. British Birds, 50, 432–434.
[211] F. Headley. (1912). Sailing Flight of Birds. Nature, 90, 220. doi: 10.1038/090220b0.
153
[212] N. M. Noor, A. Abdullah, M. Hashim. (2018). Remote sensing UAV/drones and its ap-
plications for urban areas: A review, in: 9th IGRSM International Conference and Exhi-
bition on Geospatial & Remote Sensing (IGRSM 2018), IOP Publishing, Kuala Lumpur,
Malaysia, 169, 012003. doi: 10.1088/1755-1315/169/1/012003.
[214] Department of Defence. (1980). Flying qualities of piloted airplanes. Technical Report
MIL-F-8785C, Department of Defence, Wright-Patterson Air Force Base, OH, USA.
[215] M. C. Cotting. (2010). Evolution of flying qualities analysis: Problems for a new genera-
tion of aircraft, Ph.D. thesis, Virginia Polytechnic Institute and State University, Blacks-
burg, Virginia.
[216] J. P. Kim. (2016). Evaluation of Unmanned Aircraft Flying Qualities Using JSBSim.
Master’s thesis, Air Force Institute of Technology, Wright-Patterson AFB, OH, USA.
[218] T. Foster, J. Bowman. (2005). Dynamic stability and handling qualities of small
unmanned-aerial vehicles, in: 43rd AIAA Aerospace Sciences Meeting and Exhibit,
American Institute of Aeronautics and Astronautics, Reno, NV, USA, 1023. doi:
10.2514/6.2005-1023.
[219] N. Durston. (2019). Quantifying the flight stability of free-gliding birds of prey, Ph.D.
thesis, University of Bristol, Bristol, UK.
154
[221] W. H. Press, W. T. Vetterling, S. A. Teukolsky, B. P. Flannery. (1986). Numerical
Recipes. 1 ed., Cambridge University Press.
[222] R. Murray, et. al. (2021). Control Systems Library for Python. URL: http://github.com
/python-control/python-control.
[223] V. A. Tucker. (1992). Pitching Equilibrium, Wing Span and Tail Span in a Gliding Har-
ris’ Hawk, Parabuteo Unicinctus. Journal of Experimental Biology, 165(1), 21. doi:
10.1242/jeb.165.1.21.
[225] J. Berndt. (2004). JSBSim: An open source flight dynamics model in C++, in: AIAA
Modeling and Simulation Technologies Conference and Exhibit, American Institute of
Aeronautics and Astronautics, Providence, RI, USA, 4923. doi: 10.2514/6.2004-4923.
[226] J. J. Videler, D. Weihs, S. DAAN. (1983). Intermittent Gliding in the Hunting Flight of
the Kestrel, Falco Tinnunculus L. Journal of Experimental Biology, 102(1), 1–12. doi:
10.1242/jeb.102.1.1.
[227] E. Shepard, E.-L. Cole, A. Neate, E. Lempidakis, A. Ross. (2019). Wind prevents cliff-
breeding birds from accessing nests through loss of flight control. eLife, 8, e43842. doi:
10.7554/eLife.43842.
[229] V. A. Tucker. (1972). Metabolism during flight in the laughing gull, Larus atri-
cilla. American Journal of Physiology, 222(2), 237–245. doi: 10.1152/ajple-
gacy.1972.222.2.237.
155
[230] L. L. Gamble, C. Harvey, D. J. Inman. (2020). Load alleviation of feather-inspired
compliant airfoils for instantaneous flow control. Bioinspiration & Biomimetics, 15(5),
056010. doi: 10.1088/1748-3190/ab9b6f.
[231] C. Pennycuick. (1967). The strength of the pigeon’s wing bones in relation to their func-
tion. Journal of Experimental Biology, 46(2), 219–233. doi: 10.1242/jeb.46.2.219.
[233] E. R. Dumont. (2010). Bone density and the lightweight skeletons of birds. Pro-
ceedings of the Royal Society B: Biological Sciences, 277(1691), 2193–2198. doi:
10.1098/rspb.2010.0117.
156
[238] I. M. Weiss, H. O. Kirchner. (2010). The peacock’s train (Pavo cristatus and Pavo crista-
tus mut. alba) I. structure, mechanics, and chemistry of the tail feather coverts. Journal of
Experimental Zoology Part A: Ecological Genetics and Physiology, 313(10), 690–703.
doi: 10.1002/jez.641.
[240] A. M. Rijke, W. A. Jesser. (2011). The water penetration and repellency of feathers
revisited. The Condor, 113(2), 245–254. doi: cond.2011.100113.
[241] B. Filshie, G. Rogers. (1962). An electron microscope study of the fine structure of
feather keratin. The Journal of Cell Biology, 13(1), 1–12. doi: 10.1083/jcb.13.1.1.
[244] R. Fraser, T. MacRae. (1957). Part I: Introductory Survey. Textile Research Journal,
27(5), 379–384. doi: 10.1177/004051755702700506.
[246] A. Elowson. (1984). Spread-wing postures and the water repellency of feathers: a test
of Rijke’s hypothesis. The Auk, 101(2), 371–383. doi: 10.1093/auk/101.2.371.
[247] D. Hally. (1987). Calculation of the Moments of Polygons. Technical Report AD-A183
444, Defence Research Establishment Suffield Ralston, Alberta, CA. URL: https://apps
.dtic.mil/dtic/tr/fulltext/u2/a183444.pdf.
157
[248] A. M. Heers, J. W. Rankin, J. R. Hutchinson. (2018). Building a Bird: Musculoskeletal
Modeling and Simulation of Wing-Assisted Incline Running During Avian Ontogeny.
Frontiers in Bioengineering and Biotechnology, 6, 140. doi: 10.3389/fbioe.2018.00140.
[249] J. A. Myers. (1962). Handbook of equations for mass and area properties of various
geometrical shapes. Technical Report, Naval Ordnance Test Station, China Lake, CA,
USA. URL: https://apps.dtic.mil/sti/pdfs/AD0274936.pdf.
158