FINAL YEAR PROJECT-ORNITHOPTER

DEPARMENT OF MECHANICAL UET LAHORE

DESIGN AND FABRICATION OF UNMANNED AERIAL VEHICLE (UAV)ORNITHOPTER
GROUP MEMBERS:
SAQIB IMRAN
FAISAL SALEH RAZA
SYED ABBAS RAMEEZ
MAHMOUD HUSSAIN
ADVISOR:

2010-ME-212
2010-ME-79
2010-ME-219
2010-ME-236

PROF DR. SYED FAYYAZ HUSSAIN SHAH

Abstract

In recent years the subject of flying vehicles propelled by applying wings,

also
known as ornithopters, has been an area of interest because of its
application to
micro aerial vehicles (MAVs). These miniature vehicles seek to mimic small
birds
and insects to achieve never before seen agility in flight. This renewed
interest has
raised a host of new problems in vehicle dynamics and control to explore.
In order to better study the control of flapping wing flight ,we shall develope
a
large scale ornithopter. It is capable of carrying a heavy (400 gram)
computer and sensor package and is designed specially for the application
of controls research. The design takes special care to optimize payload
capacity, crash survivability, and field repair abilities. This project covers the
design and fabrication process of an ornithopter that shall be used for the
advancement of our country in the field of technology.

Introduction

Natural fliers like birds and insects have captivated the minds of human
inventors

through history. The ease and grace with which they take to the air vastly
surpasses the state of the art in aircraft and their control systems.
This is not to say that modern aircraft designs are ineffective, they are
excellent
in many respects. Propellers and turbines are very efficient methods of
producing
thrust and airfoils efficiently produce lift. A Boeing 747 achieves a
dimensionless cost of transport (energy used divided by weight times
distance) of 0.1, equivalent to a soaring albatross and does it with amazing
reliability, but it will never match
the maneuverability of the albatross.
The problem mirrors legged versus wheeled locomotion well. Wheels
provide a
stable, easy to analyze, and very efficient way of getting around with the
sacrifice
of a large amount of agility. Legs are notoriously difficult to control and
current
implementations are energy inefficient and applying wing flight parallels this
well.
The unsteady fluid dynamics of applying wings are poorly understood and
it's difficult to get an ornithopter (the term used henceforth to refer to a
applying wing vehicle) to maneuver as desired.
Interest in the design and control of ornithopters has grown in recent years
as
interest has grown in the area of Micro Aerial Vehicles or MAVs. These small
flying
machines have struck the imaginations of many as ideal platforms for a
variety of tasks
including systems monitoring and surveillance where a swarm of tiny agents
would be
unobtrusive and have better access to confined areas than larger flying
vehicles.

Project Purpose

Perhaps the first question that must be asked about the project is what it
has to do with Micro Aerial Vehicles and why the work is a reasonable
research direction, especially because a large ornithopter bears little
resemblance to the kinds of miniature machines many researchers are
focused on. While it is true that at a 1.8 meter wingspan the ornithopter
isn't a step towards the resolution of the problems that currently plague
MAV research like energy storage and actuator miniaturization, it is a big
step toward solving the problems that these projects will encounter soon,
those of control. Even this is many times the size of MAVs under develop-

ment we conjecture that the problems inherent in stabilizing and coaxing

agility out of them are retained in large part with the scaling up.
Working with a much larger robot allows us to sidestep the problems of
miniaturization and instead work on the problems that interest the Robot
Locomotion
Group specifically, those of dynamics and control. Instead of working in
series with
researchers attempting to build the miniature machines we are working in
parallel
with comparatively easy to build hardware and of the shelf electronics to lay
the
groundwork on the dynamics and control work which will be central to the
next steps
in these projects. This is not to say that the dynamics will be the same, they
may be
as far apart as the differences between how a hawk and bumble bee fly, but
the body
of work covering the control of any sort of applying wing flight is extremely
small so
any conclusions reached will be valuable additions as work moves forward.
In order to become familiar with manually controlled ornithopters and get an
intuitive feel for the problems an ornithopter named as The Kestrel has
been made
and represents the best of available designs in terms of performance and
reliability.
With the designs in hand lab staff member Stephen Proulx built and the
machine
was successfully own the Kestrel ornithopter along with lab member Derrick
Tan,
an expert radio control pilot.
Instrumentation is required to acquire the data needed for analysis and the
eventual implementation of a controller. Instrumentation could be done
either onboard with sensors like an inertial measurement unit.
Scaling down would require even smaller servo motors and would
completely rule out
the possibility of any onboard control later because of the reduced payload
capacity.
The manual flight testing would be conducted to know about the weight
capacity which could be used to accommodate a sensor and computer
package. This leads to two options, either make an extremely light and
small sensor package, or increase the payload capacity of the ornithopter to
handle the weight of the sensors commonly used in the lab. A preliminary

analysis of the payload and carrying capacity will show that it would be
feasible to make an ornithopter large enough. In addition to the advantage
of using a computer and sensors we are already familiar with a larger
vehicle would have slower dynamics which be easier to
analyze and control when the project is advanced. Increased payload to
accommodate
the weight would also allow the expansion of the system with the addition of
more
sensors further down the line and improved performance.

Related Work

Ornithopters have been a relatively obscure area of research in comparison

to
fixed wing aircraft and field of ornithopter design is sparsely populated.
Much of the
research done has been performed by hobbyists such as Sean Kinkade.
Kinkade is
the designer of a wide range of radio controlled ornithopters both smaller
and larger
than the Park Hawk that the Kestrel is based on and holds a patent on the
design. Even though these machines have been designed both the
ornithopters and plans are exceedingly hard to obtain. Both original designs
and designs that build on Kinkade's that build are in progress in the hobbyist
community but very little work has been published. Ornithopter designs in
similar form factors that focus on adding additional degrees of freedom to
the wings have been published in addition to a variable amplitude wing
design produced by the Robot Locomotion Group previously.
An extensive analysis of the wing design has been performed with a motion
capture system by Robyn Harmon of the Morpheus Lab at the University of
Maryland which explains many of the aerodynamic properties of this type of
ornithopter. The Morpheus Lab has also been working with the simple kinds
of ornithopter control. Paralleling this work similar results have
been shown by the University of Arizona working with a 74cm wingspan
ornithopter. Both of these projects make use of a very limited onboard
computer in order to
be lightweight enough to fly.
James Delaurier's work forms much of what has been accomplished in larger
scale ornithopter design and analysis. A project to build a piloted ornithopter
has been ongoing for many years and as of July 8, 2006 achieved several
seconds of sustained flight.

Project Requirements

When examining what is required to build a ornithopter a layered system

becomes apparent.
At the base is the mechanical ornithopter system which has the main
requirement
of flying acceptably. Acceptable in this most preliminary case is defined as
being able
to sustain flight in calm conditions. Key to this is the ability to carry all the
additional
weight of the sensors and computer. Branching out from this base
requirement are
several secondary requirements. Because this is a controls research
platform it can be
expected that the ornithopter will end most of its flights with crashing at
least at the
beginning stages, this makes crash survivability of great importance in
addition to it
being a reliable machine in less severe conditions. An emphasis is placed on
designed
points of failure to isolate damage to parts easily replaced in t
he field. In addition to
this all of the systems need to be easy to tie into the computer controller.
The sensing and computing equipment which is taken along for the ride with
the mechanical platform is usually considered secondary, but is key to the
intended

research use for the ornithopter. In order to measure as much of the

ornithopter's
state as possible to perform the desired system identification and control
sensors are
needed. The absolute position of center of mass in space isn't important at
this stage
because it isn't a factor in the dynamics, however the orientation of the
body and the
position of the wings and tail in relation to the body is very important. In
addition
to the sensing of this data a computer is required to log and eventually
process it into
commands to be sent to the actuators. Ideally this computer would have
sufficient
computational power to implement advanced control algorithms unlike
many small
scale autopilots based on microcontrollers.
Even though the project requirements are pretty straightforward and simple,
the
engineering process to realize them required many iterations to develop
them into final
specifications which further required design iterations to produce a
realization of those
specifications.

Electronics

While the electronics on the ornithopter aren't a critical system as far as the
mechanical function of the machine performs they do make up the one of
the most
important specifications for the project, the minimum payload capacity.
Because the
rest of the sizing and design of the ornithopter depends on this the weight
of the
computer, interface equipment, sensors, and battery must be determined
first.

Computer
The selection of the computer to base the rest of the system on is one of the
defining
decisions of the design process because it makes up the largest amount of
additional
payload weight. Because the scaling of the robot depends on the payload
capacity

needed the computer choice will have a long chain of effects down the
design process.
Two main computer options may be selected , one for its ease of
programming,
the other for its light weight.
This software hooks in with Simulink and Matlab on a host computer for fast
and user friendly control system development. On the downside, the
computer is heavy by itself, 106 grams, but on top of that it needs
additional IO boards to interface with the servos and any wireless
communication which add even more weight. There is an ARM based
computer
which runs the Open Embedded Linux distribution. The system is much
more compact and lightweight, the set of boards weighs 43.5 grams, includes the
necessary IO
equipment to interface with the servos and a wireless networking module.
The major
cost in this option is that development for the system is vastly more
difficult. The
software development environment has a steep learning curve, the
operating system
and drivers are unreliable, and all of the software needs to be written in C.

Battery
The battery is a three cell lithium polymer pack, the standard for high
performance machines like airplanes on this scale because it has the best
power and energy to weight ratios available. Power density is the main
concern at the initial stages of the project because flights aren't expected to
last for long which would make power output the limiting factor for the
battery. A lithium polymer battery can be discharged at up to 10 times its
capacity (commonly referred to as 10C) with specially selected batteries
capable of continuous discharge up to 25C which makes it able to deliver
the short bursts of power that the ornithopter needs to apply during short
test flights.
Lithium phosphate batteries produced by A123 are also available because
they have similar characteristics but the cell sizes available at the time were
too large to make a suitably lightweight pack at the correct voltage. Since
then a smaller series of cells has become available, the 18650, which
matches the application well. Four cells would produce a pack at 13.2 volts
at 156 grams. This works out comparably to the lithium polymer pack with
two important advantages. The cell price of the lithium phosphate batteries
works out to a pack at about half the price of the lithium polymer pack. In
addition to this the lithium phosphate cells are much less sensitive to
overcharging which causes lithium polymer batteries to violently explode.

Because of potential cost and safety advantages of the lithium phosphate

cells over the lithium polymer cells they may be further investigated in the
future. Because the current draw from the motor is not yet known a
standard battery, 1200mAh at 11.4V and 93 grams, can be chosen for the
weight calculation. Eventually a three cell 2200mAh battery pack can be
selected for actual use.

Sensors
Ideally full state information for the ornithopter would be available to the
control system, but because sensors are a significant part of the weight
budget only those of immediate need to the control system envisioned can
be included. Orientation and velocity are typically used to describe the state
of an aircraft because the body's position doesn't generally play a part in
the vehicle dynamics. This of course discounts environmental effects like
wind and obstructions.
The MicroStrain 3DM-GX1 is the standard orientation sensor used by the lab
and is widely used in the field. It's a fully integrated gyroscope,
accelerometer, and magnetometer sensor package that performs the sensor
filtering and fusion onboard and communicates orientation and acceleration
over an RS232 data line. It was chosen as the primary sensor for the
ornithopter because of its reliable and accurate performance in addition to
its small and lightweight construction. The sensor package is fully
integrated onto two small boards with all surface mount construction,
making it surprisingly light at 25.8 grams for the amount of equipment
packed onto it.
While velocity information is theoretically available from the sensor through
the integration of reported accelerations in practice that method is
extremely prone to drift and a pitot tube or hot wire velocity sensor is a
much better sensor for the application. Both of these could be added at a
later point at a small weight cost but because the very first control
experiments would only be using pitch data in a single-input-single-output
fashion the decision to maximize flying performance by minimizing weight
was chosen.
This also applies to the position data that will eventually be needed for
autonomous flights over longer distances and involving turns. There are
many extremely lightweight GPS units that can be integrated at a later time
for a small weight penalty once initial performance concerns are worked out.

Hardware Interface
There are three main elements to the control system interface: the servo
motor controller, drive motor controller, and communications to systems
onboard the robot.
Conventional hobby servos like those used on model aircraft use a pulse
width modulated signal with a period of about 40hz and a pulse width of

between 1 and 2 milliseconds which corresponds to the desired absolute

position of the servo's output drive. Because the computer system's main
interface is over RS232 serial, a converter is needed. Serial to PWM
converters designed specifically for controlling hobby servos are widely
available because of their widespread application in robotics. For this project
the Pololu Micro Serial Servo Controller can be chosen because of its
extremely small size and weight; the smallest available at about one square
inch and about five grams. The device uses a custom programmed
Microchip PIC microcontroller to receive serial data and produce the PWM
signal. While a similar device could be produced custom for the project the
advantage over the already extremely small size and weight of the Pololu
controller would be very small.
The actual test flights conducted are a combination of radio control and
computer control, both for procedural and safety reasons. While testing
different controllers it's desirable to be able to control the robot via a
conventional radio controller, either in full or just specific channels like the
throttle while the computers handles the control surfaces. It's also used to
guide the robot into initial conditions where it's safe to switch over to the
computer controller. There are multiplexers designed for the specific
purpose of robot control in which the signal source selected is controlled by
a separate PWM input from the same radio receiver that traditionally
controls the servos. A digital output line is also connected to the control
computer so that the control loop can be initialized correctly once control is
switched over. The multiplexer used on it can be RxMux board made by
Reactive Technologies which weighs
about 30 grams. Since the time of the original design Pololu has released a
new servo signal multiplexer that would fit this application much better with
significantly size and weight savings over the Reactive Technologies board.
Next to the multiplexer board is the radio receiver.
The drive motor controller sits between the serial to PWM servo controller
and the motor and uses that signal to control a high current driver suitable
for the power requirements of the large motor. This is again a standard radio
control airplane part and a JETI Model Advance 30 brushless motor speed
controller can be chosen for this application because of its past use in the
lab. The speed controller is designed to handle 30 amps of current which is
considered to be an upper bound of power required, about 300 watts at the
11.4 volts the battery pack runs at, and weighs 35 grams.

Wireless Communications
The last major piece of interface equipment is on the other side of system,
communication between a base station computer and the robot. While a
standard 802.11 network link is desired the PC/104 based system running
XPC Target doesn't support a USB wireless transceiver. This problem can be

sidestepped through the use of a wireless bridge which connects to the

wired ethernet on the computer making the connection transparent to the
system. An Asus WL-330g, the smallest device capable of the job at the
time, was used. This allows data to be onboarded and control system
parameters to be changed without hooking an umbilical cable up to the
robot.
The fact that 802.11 wireless networking is the widely used standard in
retrospect has led to many unforeseen problems. The extremely high
proliferation of wireless networks in the Stata Center and around the MIT
campus results in several access interference that causes latency problems
and dropped packets from qualitative experience. While measures can be
taken to increase the strength and directionality of the desired signal it
doesn't seem to make much difference with respect to the problems
experienced. Depending on the fragility of the protocol used results of these
latency issues and dropped packets can range from insignificant to
catastrophic, the best example of this being the communications used by
the Simulink XPC system which completely fails in the presence of any
dropped packets.

Wings

The wing design chosen for use is a proven design by Sean Kinkade and is
used throughout his line of ornithopter designs which shows that it scales
well.
Working from the years of development on this wing design allows the focus
of our project to be on the parts we have more experience in such as the
gearbox, electronics, and controls. The wings have a triangular support
structure made from carbon rods. A main spar runs along the leading edge
of the wing and a strut connects from the rear of the ornithopter's body to a
point near the tip of the main spar. From this strut there are several smaller
carbon rods that project to the edge of the wing which are somewhat free to
move. This results in a fanning motion from the trailing edge of the wing
that produces thrust while the leading edge is applying up and down which
directly contributes a part of the lift in addition to the conventional lift
coming from airflow over the wing.
Working from the specifications of the ornithopter and the new payload
capacity necessary an approximate size for the scaled up wing can be
found based on the wing loading. With its 0.22 square meter wing area and
overall weight of 395 grams there is a wing loading of 1.78 kg/m2 . Scaling to a
larger machine that even the size of is known is a pretty difficult proposition,
but a few assumptions can be applied to clean up the situation. The
payload fraction, or amount of payload divided by the overall weight of the
ornithoper is assumed to be constant. For this, number is 0.334 with a
payload of 132 grams. The 132 gram payload is estimated by looking at the
difference between the weights of onboard components like the batteries as

equipped and the maximum weighs allowed (using lithium polymer

batteries vs nickel metal hydride for example). The estimated required
payload is 400 grams which, using the same payload fraction, comes to a
total weight of 1197 grams. With this weight in order keep the same wing
loading as it is, it will need a wing area of 0.672 square meters. By scaling
the original wing the desired wingspan comes to about 2 meters. The actual
wingspan used in the design is shortened to 1.8 meters in order to make the
ornithopter easier to handle through the lab with the option left open to
increase the wingspan with longer spars if necessary.
The wings are of similar construction to a modern kite. The fabric can be
used is 1/2oz polycarbonate coated ripstop polyester which provides
excellent strength and resistance against tearing for very little weight. The
sections where the wing interfaces with spars or the ornithopter frame are
constructed from Dacron fabric which is glued to the polyester fabric to form
pockets and reinforced sections, for the most part directly scaled up from
the wing design. Instead of stitching the fabric which creates stress
concentrations in the fabric gluing is the preferred construction method now.
The gluing is done with the application of a very thin double sided tape
called Very Highly Bonding by 3M. The tape is applied with the backing still
on to one piece of fabric and worked into the surface for full contact, then
the backing is pulled away and the mating piece of fabric is applied.
Selecting the correct wing spars for the scaled up version was one of the
larger engineering leaps of faith because not only do they have to be strong
enough to not break, but the wings rely heavily on their compliance to
produce the kinematics that have been experimentally found to produce the
best performance.
This is because in the case of a beam bending due to a moment acting on
the end, the static situation most closely modeling the ornithopter,
deflection at all points grows linearly with the moment applied. The scaling
relationship for this moment is obtained in the gearbox design section
further on, applied here, leading to the desired moment of inertia four times
greater than that of the spars of the ornithopter.
The insertion ends of the wing spars are a somewhat difficult engineering
problem because they are the point at which the driving torque is exerted.
While the spar is easily able to handle the torque, any piece attached to the
end to drive the spar will create a stress concentration that will create a
point of failure. In addition to this the spars need to be able to be removed
from the shoulder assembly both for assembly of the wing and when they
break. In order to distribute the loading along a length of the spar instead of
concentrating it at a point an extremely stiff rubber is used to make a
threaded connector that screws into a delrin shoulder. This connector allows
the spar to flex and because it is soft it is unable to create a significant
stress concentration, instead relying on a large area of lower force to exert
the torque on the spar.

Gearbox

The most critical part of the ornithopter is the drive mechanism that
converts the electric power from the battery to the applying motion of the
wings. This system is the most complex to design and fabricate because it
must withstand very large forces which reverse direction several times a
second while at the same time being extremely light and durable. Because
of the loads it must be made from may be aluminum which makes it
beneficial to perform careful analysis and trim as much weight as possible.
The drive system can be further broken down into four sections, the electric
motor, a gear reduction stage, a linkage to convert the high torque rotation
into a reciprocating motion, and the connection to the wing spars. While a
highly integrated design is needed in order to maximize the power to weight
ratio, the analysis of these parts breaks down well.

Load Case Development

In order to figure out what loads the gearbox needs to be designed for the
torques to be exerted at the wing shoulders need to be known. While a
relationship for scaling the torque from the ornithopter can be created, the
actual torques on it need to be known too. Because the gear ratios and
motor characteristics are known these torques can be found by measuring
the current through the electric motor and calculating the torques at each
point in the transmission to the wings.
With the ornithopter frame fixed a shunt resistor was used to measure the
maximum current at full throttle through the brushed DC motor that flaps
the wings.
This represents the maximum loading seen by the transmission. An
oscilloscope can be used to measure the voltage across the resistor over
time along with the voltage across the motor. With this data it is found that
the maximum torque at the motor is 0.00832 Newton-meters and that the
wings flap at 6hz in the conditions during measurement. This torque at the
motor is translated through the 97.8:1 ratio gearbox and the applying
linkage to produce 4.43 Newton-meters at the shoulder.
With the torque at the shoulder of the ornithopter and the new wingspan
needed by
the ornithopter the one piece that remains is the scaling relationship
between wingspan
and torque at the shoulder. This is accomplished with a relatively simple
analysis. The torque at the shoulder is the product of the distance to the
center of pressure(moment arm) times the force produced by the pressure
on the wing concentrated at that point. Because the wing loading is being
fixed for this scaling process and the wings are roughly square the force
increases with the square of the wingspan.
In addition to this the distance to the center of pressure increases linearly
with the

wingspan as long as the wings remain loaded in a similar fashion along their
span. This suggests that the torque on the shoulder will grow with the cube
of the wingspan. The new design of a wingspan 1.56 times longer and so
the shoulder torque is expected to be 3.79 times higher or 7.78 Newtonmeters.
The rate at which the wings need to be flapped is also important to
selecting gearbox components. While exact data to base this on isn't
available a video of a similarly scaled ornithopter based on Kinkade's design
can be found in which the flapping frequency could be roughly estimated
from. Compared to the Kestrel known to fly at 6Hz the ornithopter in the
video is judged to flap at about 4Hz, roughly in line with an inverse scaling
of flapping frequency with wingspan.

Drive Linkage, Wing Joining Gears and Spar Mounts

In order for the forces on the wings to balance each other out to produce
only
thrust and lift rather a rolling and yawing moment the wings need to move
together
in the same motion. This can be accomplished in a couple different ways,
either by
having a second, opposite four bar linkage which is the strategy used on the
Kestrel,
or with a set of gears that slave the motion of one wing to the other similar
to the
design in Kinkade's patent.
In the case of the Kestrel it made sense to have two separate linkages
because the
lightly loaded gearbox was able to be built directly into the fiberglass frame
of the
ornithopter and was configured longitudinally so setting up two linkages
that don't
interfere with each other is relatively easy. Combined with the fact that they
don't
need to be very strong because of the small scale making two linkages from
lightweight
and cheap materials is the natural choice. One major downside to the
configuration
is that the rotation of the crank and wing are on different axes which means
the links
need to have additional degrees of freedom to accommodate the 3D
motion. This is

accomplished with plastic ball joints that are common to model airplanes
but creates
a major source of broken parts in practice because the joints are not
designed for such
loads.
With the vastly increased load in the gearbox a different strategy is
required. The linkage movement must at very least be kept in plane so that
proper bearings can be used at the joints instead of ball joints, this means
that the linkage must be longitudinal which in turn means that having two
opposing linkages is much harder to make work. Luckily slaving one of the
wings to the other is also an option.
By putting gear teeth on the shoulders they can be made to turn with each
other with
very little extra weight over just moving one wing. With one linkage driving
both
wings at the same time the loads on it are slightly more than doubled, but
since the
motion is in plane the problem is reduced to a relatively simple problem of
selecting
suitable bearings and designing the links to be lightweight and strong in the
right directions.
The linkage is designed to match the wing trajectory with the wing lifting 30
degrees above the horizontal and 20 degrees below. Flat links constructed
from titanium allow them to be designed to be strongest in the directions
that matter and extremely lightweight overall. The geometry of the linkage
can be determined using the Solidworks COSMOSMotion package and then
the links can be optimized using COSMOSMotion FEA. The links can also be
analyzed for buckling because their structure is very susceptible to it, the
load factor of 8.3 predicts that buckling will not occur.
The wing shoulders have three functions, forming the second fixed point of
the main drive linkage, driving one wing to match the motion of the other,
and providing an attachment point for the wing spars. Both wing shoulders
rotate on titanium shafts supported by miniature ball bearings. These
bearings are commonly used in small machinery, most notably model cars,
and are available cheaply in almost any size, these coming from Boca
Bearing. The miniature bearings are lightweight and small allowing the
overall gearbox size to be kept to a minimum.
The shoulder piece itself is machined delrin and has a threaded end to
accept the screw in spar discussed in the wing section. The threaded wing
spar is essential because it allows the spars to be quickly changed out when
they break and allows for the compliant rubber of the spar end to mate with
the shoulder which must be much more rigid. A set of pins join the delrin
piece to the aluminum gear that drives the other wing in addition to the
rocker link that drives both wings. The gears that connect the wings were

selected using the Lewis formula for gear tooth strength to find the most
lightweight gears that fit the form factor of the gearbox. Stockgears were
ordered and waterjetted to the desired shape so that the precise gear teeth
wouldn't have to be machined.

Motor Selection
The main drive motor is one of the most important parts to be selected
because it makes up one of the major weight components of the drive
system. The highest power density possible is desirable for obvious reasons,
but second to that is a high torque density. This is because gear reductions
require both additional weight and space in the transmission. The gears
need to be able to handle large torques so they also must be strong and
hence large and heavy. Several types of miniature high performance motors
are available for model airplanes but one in particular sticks out for this
application, the brushless outrunner, which is primarily used in direct
propeller drive configurations because of the high torques supported.
A brushless outrunner motor is essentially an inside out version of the
conventional brushless motor, instead of the coils being placed around an
internal rotating magnet assembly the coils are wound around a stationary
internal post and the magnets are mounted on the outside motor housing
which is free to rotate.
This puts the magnets that the electromagnetic force created by the coils
acts on as Pins connect the delrin spar receptacle to both the driving gear
and the drive linkage rocker, far away from the axis of rotation as possible.
This creates the largest amount of torque possible in the space constraints
of the motor.
The two outputs from the motor and gear reduction system are already
known, the torque transmitted down through the drive linkage becomes
17.7 Newton-meters and the desired speed of 4hz becomes 240RPM. Sadly
this does not fix the variables enough to select a motor or design a gearbox
right of the bat because a wide range of gear ratios and motor windings are
available. Rather than simply fixing a design parameter with no basis to
proceed, some reasoning about the system was done.
Because of the large torque required at the transmission output a steep
gear reduction will be required along with large strong gears in the stage
close to the output. A gearbox with two reduction stages of the maximum
possible ratio was designed as a starting point. In this case the maximum
reduction at each stage was limited by the strength of the gear teeth, what
was kept in stock by the supplier, and what could be packaged feasibly into
a transmission. At the end a maximum of a 24:1 reduction was achieved
which made motor selection very easy. The 240RPM output shaft speed
through the 24:1 reduction becomes a target of 5760RPM at the

motor shaft. At the maximum voltage across the motor of the 11.4 Volts of
the three cell LiPo pack the desired motor speed constant comes out to
505RPM/V .
The electric motor speed constant is inversely proportional to the motor's
torque constant, a measure of the amount of torque produced per unit
current. Because of this relationship it makes sense that the small
lightweight motors desired for this application will have high speed
constants. Windings available on even the the highest torque motors
suitable start well above this speed constant because of their inability to
produce a large amount of torque in such a small space. The law of
conservation of energy when applied to this case shows that for the same
energy output the speed of the motor must increase if the torque produced
decreases.
With these specifications the field of possible motors became very small.
The selection of windings available in the brushless outrunner motors used
in model aircraft is quite limited because of the specialized application but a
motor that fit the desired specifications closely enough was found. The PJS
3D 1200 motor, conveniently used in previous experiments in the lab has a
speed constant of 848RPM/V and a torque constant of 0:01126Nm/A . The
expected torque required at the motor shaft of 0.7375Nm requires 65A of
current through the motor. While this current seems too high the options to
lower it, either by adding a gearbox stage or using a heavier motor are as
unpalatable as handling the high current. This decision was one of the major
engineering compromises of the project but the stacked safety factors in the
gearbox loading and conservative scaling estimates used so far make the
predicted loadings
especially brutal at this stage. The decision to go with the PJS 3D 1200
motor and deal with the consequences of high current was made with the
option of using a larger motor if testing indicates the necessity.

Gearbox Frame
The gearbox frame is a component whose performance, not just existence,
is critical to every function of the gearbox. The distance between the gears
and hence the quality of their meshing depends on the accurate
construction and stiffness of the frame. The quality of the gear mesh is of
utmost importance because the flapping motion puts periodically reversing
loads on the gears and backlash in that case leads to very fast tooth
destruction in addition to sloppy wing motion at the highest and lowest
parts of the stroke. The pursuit of accuracy and stiffness can quickly lead to
a very heavy and expensive construction.
In order to make the frame as lightweight as possible a titanium sheet
design was chosen. The process of designing a complex frame in CAD,
cutting it out with a CNC waterjet cutter, then bending and welding the
structure together is very inexpensive and fast because little time is spent

machining large amounts of material away. By paying careful attention to

the design of box and truss structures and how they are bent from the
parent sheet surprising stiffness can be achieved and with the addition of
tabs that guide assembly and welding the entire process is very fast.
The key to making the process come together is welding the individual
pieces together. In order to make the strongest and cleanest welds the
Tungsten Inert Gas (TIG) process is used. This welding process relies heavily
on the skill of the operator to guide the torch and apply the right amount of
heat to the part in the form of an arc between a tungsten electrode and the
base metal inside an inert argon atmosphere. Titanium is extremely
susceptible to atmospheric contamination when heated so special steps
outside the normal welding process must be taken.
Titanium welding is often performed inside a completely purged glovebox
for this reason. Without one available additional shielding gas can be
applied and attention not to let the areas of the part exposed to the
atmosphere heat up too hot must be observed.
The accuracy of the bearing pocket placement comes from cutting out the
holes undersize on the waterjet cutter and once the bending and welding of
the frame is complete they are reamed to their finished sizes. The reaming
process is performed with the frame mounted in the vise of a milling
machine, by using the machine's table to move the part to the correct hole
locations they can be machined accurate to less than a thousandth of an
inch.
When finished the entire frame weighs a total of 58.1 grams and stands up
to both the high internal forces of the gearbox and large amounts of
external abuse, usually from impacts to the ground during crashes. Because
large shock loads are expected from frequent crashes a protective face
plate is attached to the main frame by several screws. Made from much
cheaper thin fiberglass laminate it's intended to flex and absorb the shocks
and break instead of the main frame breaking. Multiple faceplates can be
kept on hand during testing sessions and replaced quickly whereas the main
frame is cut from much more expensive material and would take hours in
the lab to transplant.

Tail

The tail section of the ornithopter is responsible for both of the controllable
degrees of freedom aside from the ability to throttle the drive motor. The tail
of the Kestrel is set up with one servo directly connected to the tail at an
angle and another further up the body which rocks it via a linkage. Mounting
the rudder servo at an angle is important because with a single control
surface the elevator and rudder are naturally coupled, moving the rudder
servo makes the tail also move in the vertical direction unless it's at zero

angle where it doesn't have any control authority anyway. With the tail at an
angle to the rudder servo it allows the servo to held into the zero angle
position with respect to the ornithopter body and while the tail stays near
the trimmed position for horizontal flight. This causes the tail to move in a
bowl shaped trajectory decoupled from the elevator action and is much
easier to control by a human pilot. It does not solve the problem when the
elevator servo moves the tail out of the trimmed horizontal position, but
being able to exploit that is a big advantage where it applies.
A second point to make about the Kestrel tail is that the linkage for the
elevator servo appears to be unnecessary but serves two important
purposes. First is that the servo range of motion is much larger than what is
necessary for the tail. By using a linkage not only does the range of motion
get reduced but the mechanical advantage allows a smaller servo to be
used. In addition to this, the servo, a major point of weight on the
ornithopter, is moved forward in the frame. While not obviously an issue it
becomes important once the center of gravity must be placed because the
frame ends up rear-heavy.
The tail of the Phoenix is designed with computer control in mind and so
doesn't use the configuration the Kestrel exploits to make human control
easier, but it does use the same tail fan design. The servos are directly
joined to each other so that as little weight is spent on linkage and
mounting as possible, again using lightweight welded titanium construction.
At a glance it can be seen that the tail cannot follow the same path as the
Kestrel tail, but it should be able to act similarly aerodynamically.
Because of the large surface area of the tail and the fact that tail is attached
directly at the servos output horns very high torque HiTec 5995TG servos
with titanium gearsets were chosen. The high strength gears are much less
likely to be damaged in the event of a crash that impacts on the tail and
come at a slight weight savings over lower torque servos with a steel
gearset.

Main Frame

The main frame of the ornithopter is a surprisingly simple component from a

design standpoint. Because the flapping mechanism is contained fully within
the gearbox frame the main frame of the machine serves mainly to provide
mounting locations for the rear wing mounts, electronic components,
battery, and tail assembly. There are two directions to go with the frame
design, either a single at plate which relies on its own thickness for stiffness,
or a three dimensional design made from much thinner material that gets
its stiffness from the truss-like structure.
If the frame had to be very stiff the second option would make for a much
lighter and stronger option, but in this case the frame really doesn't have to
be very stiff in all directions. The rear wing mount actually holds the frame
from flexing in the direction a sheet would be most weak because the wing

spars form a very large and stiff triangle. What this amounts to is that while
the more complicated structure is often the higher stiffness and lower
weight option it doesn't come with much advantage in this case. The at
frame is vastly easier to design and fabricate and may even be lighter.
Because the frame is really just a collection of mounting locations for the
rest of the components it's relatively easy to design. The one major
specification that must be paid attention to is the location of the center of
gravity. Because the data needed to work out the correct location
theoretically isn't available this was also scaled up from the Kestrel where
the location was experimentally determined from flight testing to be about 4
inches behind the main wing spar. If this is scaled linearly with wingspan the
desired center of gravity should be about 6 inches behind the main spar.
In order to facilitate the locating of all the parts the weights and sizes of all
the components were modeled in Solidworks and component configurations
were iterated until the center of gravity was in the right location. This turned
out to be more difficult than expected because the machine tends to be rear
heavy with all the computer equipment distributed on the frame. In order to
counteract this the main battery back was moved in front of the gear box, a
less than ideal option but necessary to get the balance right. The completed
machine weighs 1200 grams.

Control System

For the very first flights a PD control was selected for several reasons. First
because of its widespread application in conventional aircraft autopilots,
second because it's very simple to establish reasonable starting gains and
tune by hand. The proportional and derivative action is easy to visualize and
intuitively understand which is very important in this case where making as
few tests as possible is desired and catastrophic failure isn't. Applying PD
control to pitch was the first step taken because the other degrees of
freedom are relatively passively stable and pilots had little luck controlling
the ornithopter in pitch themselves. It isn't obvious that a simple PD control
should work with a an ornitopter because of the vehicle's periodic dynamics
from flapping. One way of looking at the situation is that the controller
needs respond to larger changes in pitch rather than the fast motions due to
flapping. In order to accomplish this the time constant of the controller can
be adjusted so that the response to fast changes in pitch is minimized.
The control system used on the successful flights so far is written in C and
runs on the Gumstix computer at 40Hz. 40Hz is the target update rate
because it's the maximum that the servos can take commands since the
frequency of their variable duty cycle pulses is 40Hz. The pitch and
derivative of the pitch come directly from the inertial measurement unit and
do not undergo further filtering. In tests so for the desired pitch has been

set to a constant value approximating what has been seen in manual flights
but this can easily be changed as progress is made.
Controlling yaw has not been implemented yet and will require a somewhat
more clever strategy of servo control because the area of the tail presented
to the airow changes with the position of the elevator servo.
A proportional control on the throttle with altitude as an input has been
attempted. The reasoning behind this isn't full altitude control but to
enhance the level flight by bumping up the throttle a little bit when the
ornithopter experiences a drop in altitude. A sonar range sensor was added
to the Phoenix to sense altitude but the data has been too unreliable to
hand control of the throttle over to the controller because of the sensor's
inherent unreliability and the sine errors introduced by the robot pitching
and rolling that must be accounted for. An alternative being investigated is
an extremely sensitive atmospheric pressure sensor in combination with
acceleration data from the IMU.

Testing

Many test flights shall be conducted with the finished ornithopter, first with
an equivalent weight and distribution payload under manual control to
determine whether the machine would actually be able to fly. Initial tests
show that sustained flight is possible but the robot is exceedingly difficult to
control and quickly crashes. Later tests with a PD control on the elevator to
stabilize pitch qualitatively shows promise but difficulties with gearbox and
wing spar reliability plagues the testing process.
This process of breaking things during testing is an essential part of the
design process and leaps of progress are made during this time in tracking
down problems and implementing design solutions to them. Parts of the
gearbox like the connection between the final rocker link and the shoulder
are a common point of failure and receive stopgap design revisions until
enough changes accumulated for a full design iteration of the machine.
Changes are incorporated into the gearbox design, the electronics package
switch over to the much lighter Gumstix based system, and the frame is
reconfigured to balance the new weight distribution properly.
A PD control stabilizing pitch is applied to the elevator with the throttle set
at a fixed point by a remote control operator. The throttle setting is taken
from a successful manual flight. The controller gains and pitch set point
were initially set to a response that seemed to mimic a human pilot and
tuned over the course of several tests. At the end of the flight flapping
power is shut off and the wings passively enter a high dihedral which causes
it to quickly glide to the ground while maintaining roll and pitch stability.
Steady state flight data from one of the successful test is shown below, note
that the length of time is only about two seconds because the section
containing initial conditions is excluded and the area used for testing is
small. Longer tests are planned but have not been conducted yet. The data

shown covers nine wing beats. The graph of orientation shows pitch
maintained with small variations, most likely caused by environmental
conditions and the action of the wings flapping. Both roll and yaw are
uncontrolled but are expected to be somewhat passively stable. The trend in
roll shown in the data seems likely to be caused by a difference between the
wings such as a weakened spar or out of trim tail. It shows peak torques in
the 0.35 Newton-meter range which is about half the maximum predicted.
The peak torque seen here isn't the peak torque that can be produced
however, because the throttle is set at about 70% of maximum for the
steady level flight. Two factors complicate measurement above this point.
First is the speed controller is only rated for 30 amperes so going above this
level results in the speed controller shutting down. Second is that throttle
settings near maximum excite a second mode in the motion of the wing
spars and sets up a standing wave instead the desired flapping motion. Both
of these problems will be subject of further investigation.
Because both current and voltage are measured it is simple to look at the
power it takes for the ornithopter to fly. While efficiency isn't an immediate
goal it's obvious that the current realization of the ornithopter and its
controller is an extremely inefficient way to fly. An estimated forward
velocity of about 4m/s based on total time of flight and distance covered
combined with a vehicle weight of 1.2Kg shows a very rough unitless cost of
transport of about 30.

Conclusion

In this project the case for the construction of a large scale ornithopter
suitable for control systems research is motivated. Performance and weight
constraints imposed by the computers and sensors desired onboard make it
difficult to work with the smaller platforms currently available, let alone
micro UAVs currently in development.
In order to work with the dynamics and controls of a flapping wing flying
vehicle while these future targets are currently in development a version
shall be constructed by the will of GOD. With its larger payload capacity it's
capable of carrying a fully equipped computer and high-end inertial
measurement unit with the option of future additions of GPS or other more
exotic sensors.
The ornithopter shall be designed from the ground up with the needs of
research in mind and for its useful benefits for the society. All components
have been designed to be as lightweight and high performance
as possible so as to maximize payload capacity and are intended to fail in
predicable
and field repairable ways. Examples of this are the screw in wing spars and
replaceable

face plates. In addition to this all parts of the ornithopter are simple and
inexpensive
to fabricate and assemble.
Manual and initial autonomous flight tests have been conducted and show
that the ornithopter is capable of sustained flight with a full load of
electronics and can
be stabilized by simple controllers in common use in aircraft. Flight tests
have also
shown that the planned points of failure work as expected and allow repairs
to be
quickly accomplished in the field.