Development of an Inertial Generator for Embedded

Applications in Rotating Environments

by
Stephen D. Conrad

B.S. Mechanical Engineering

University of Cincinnati, 2005
Submitted to the Department of Mechanical Engineering
in Partial Fulllment of the Requirements for the Degree of
Master of Science in Mechanical Engineering
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2007
c 2007 Massachusetts Institute of Technology.

All rights reserved.

Signature of Author: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Department of Mechanical Engineering
May 21, 2007
Certied by: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Steven R. Hall
Professor of Aeronautics and Astronautics
Thesis Supervisor
Read by:. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
David L. Trumper
Professor of Mechanical Engineering
Department Thesis Reader
Accepted by:. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Lallit Anand
Professor of Mechanical Engineering
Department Committee on Graduate Students
2
 Development of an Inertial Generator for Embedded
Applications in Rotating Environments
by
Stephen D. Conrad

Submitted to the Department of Mechanical Engineering

on May 21, 2007, in Partial Fulllment of the
Requirements for the Degree of
Master of Science in Mechanical Engineering

Abstract
Inertial generators are devices that generate electrical energy from their inertial mo-
tion, and have only one mechanical connection with their surroundings. This makes
them suitable power sources for embedded systems operating in environments that ex-
perience some inertial excitation. Typical inertial generators are designed to generate
electricity from linear vibrations, and are often termed vibrational energy harvesters.
Traditional sprung mass vibrational energy harvesters must be tuned to achieve res-
onance at a specic frequency, and perform poorly when the excitation does not fall
within a narrow band around this natural frequency.
In this thesis, a novel inertial generator is proposed that is specically designed to
take advantage of the unique inertial loads experienced by a system embedded within
a large scale rotating part with a horizontal axis of rotation, such as the propeller of
a large ship. The design process begins with the identication of the inertial path
and then proceeds with the development of a device that takes advantage of the
unique inertial loads experienced along that path. The device is designed to achieve
resonance at any steady state rotation rate, without any active forms of actuation.
This is achieved by utilizing centrifugal forces to produce a natural frequency that
tracks the excitation frequency.
Experimental results from full-scale spin testing verify that the device has a mono-
tonically increasing power output with increasing frequency. This result contrasts
sharply with the frequency response of a traditional sprung mass vibrational energy
harvester, which typically has a single peak at the resonant frequency. Experimental
results are also presented showing that the device can successfully deliver a charging
current to a battery over a wide range of operating speeds.

Thesis Supervisor: Steven R. Hall

Title: Professor of Aeronautics and Astronautics
Margaret MacVicar Faculty Fellow

3
4
Acknowledgments
This research would truly not have been possible without the support, advice and
inspiration that I received from numerous individuals throughout the project. I would
like to extend special thanks to my adviser Prof. Steven Hall for making this project
available to me and supporting this work with his advice and his many resources.
I would like to thank Bob O'Handley, Jiankang Huang, and Kevin O'Handley at
Ferro Solutions, Inc., for their input at all stages of the project, and for sharing their
expertise in vibrational energy harvester design. I would also like to thank Prof. David
Trumper for acting as a reader for this thesis and for teaching me a great many things
in his courses that were useful throughout the completion of this research. I also have
much gratitude to express to Prof. Alan Epstein for allowing me to conduct the spin
testing in the gas turbine laboratory in a safely contained environment.
Along the course of this research, there have been a number of individuals who
have supported my eorts to procure, construct, and build the equipment necessary
for laboratory testing. I would like to extend thanks to James Letendre for supplying
me with every sort of tool imaginable, including several that I have never before
imagined, and entrusting me with their proper use. Many thanks to Todd Billings
for allowing me to fabricate a great deal of hardware in the Aero-Astro machine shop
despite the fact that I kept bringing in large pieces of structural steel into a shop
that is accustomed to aluminum. I also appreciate his instruction on the use of the
various CNC machines that allowed me to improve my designs considerably, and his
help on completing a great many machining tasks that at rst seemed impossible. My
thanks to Dave Robertson for his help locating useful items and occasionally lifting
the especially heavy ones, all while maintaining a good natured hostility toward grad
students and those who study mechanical engineering in particular. Thanks also to
Dick Perdichizzi for help locating hard to nd items and for advice regarding safety
concerns.
The students in the Aerospace Controls Lab in building 41 were wonderful neigh-
bors and supported me in many ways. It was always relaxing to take breaks to see

5
their computer controlled ying contraptions perform increasingly amazing maneu-
vers in such a conned space. I also appreciate the many times they oered their
advice and assistance on microcontroller programming and electrical circuit design.
I would like to recognize my friends Mario and Brett in particular for going out of
their way to get to know me and for sharing their love for ice hockey.
I would like to thank Tony, Tatsuya, and Jinshan for being both great roommates
and great friends, and Tao for making my life a great deal more interesting than it
would otherwise have been. I also appreciate the stress relieving role that my friends
in the SUSHIS hockey team and the MIT Easyriders motorcycle club have played in
my life and will hopefully continue to play in the future.
I owe much to my family who have always supported my endeavors and who has
been the stable platform in my life from which I have always felt secure to reach
toward my dreams. Ultimately I owe everything to my Lord and Savior who has
granted me the opportunity, the ability, and the strength to accomplish every good
thing in my life.

This research was supported by Ferro Solutions, Inc., through funding provided
from the Navy STTR phase II project. Topic: N04-T022 energy scavenging. Contract
number: N00014-05-C-0420.

6
Contents

1 Introduction 11
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Previous Work and Chosen Approach . . . . . . . . . . . . . . . . . . 14
1.3.1 The Inertial Trajectory . . . . . . . . . . . . . . . . . . . . . . 14

2 Device Conguration 19
2.1 Explored Device Congurations . . . . . . . . . . . . . . . . . . . . . 19
2.1.1 Falling Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.2 Simple Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.2.1 Restoring Torques on a Simple Pendulum in a
Centrifugal Potential Field . . . . . . . . . . . . . . . 23
2.1.2.2 Simple Pendulum Performance . . . . . . . . . . . . 24
2.1.3 Distributed Pendulum . . . . . . . . . . . . . . . . . . . . . . 25
2.1.4 Simple Pendulum With Geared Flywheel . . . . . . . . . . . . 27
2.1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Slider Augmented Pendulum . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.1 Conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.2 Slider Augmented Pendulum Dynamics . . . . . . . . . . . . . 33
2.2.3 Natural Frequency Tracking . . . . . . . . . . . . . . . . . . . 33
2.2.3.1 Frequency Tracking in Linearized Model . . . . . . . 33
2.2.3.2 Frequency Tracking in Full Nonlinear Model . . . . . 34
2.2.3.3 Amplitude Dependency of Natural Frequency . . . . 36

7
 2.2.4 Statically Stable Range of Motion . . . . . . . . . . . . . . . . 36
2.2.5 Sensitivity to Misalignment . . . . . . . . . . . . . . . . . . . 39
2.3 Extracting Energy Through Damping . . . . . . . . . . . . . . . . . . 43
2.3.1 Damped Harmonic Oscillator . . . . . . . . . . . . . . . . . . 44
2.3.2 Applying Electromagnetic Damping . . . . . . . . . . . . . . . 46

3 Prototype Design 51
3.1 Choosing Nondimensional System Parameters . . . . . . . . . . . . . 51
3.2 Magnet Assembly and Coil Design . . . . . . . . . . . . . . . . . . . . 55
3.3 Mechanical Linkage Design . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.1 Link Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.2 Symmetrical Conguration . . . . . . . . . . . . . . . . . . . . 60
3.4 Slider Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5 Device Frame and Attachment . . . . . . . . . . . . . . . . . . . . . . 64
3.6 Prototype Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4 Experimental Setup 69
4.1 Spin Stand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2 On-board Variable Loads . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.1 Resistive Load . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.2.2 Switching Load and Mock Battery . . . . . . . . . . . . . . . 81

5 Experimental Results and Discussion 85

5.1 Static Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.1 Coil Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.2 Electromagnetic Coupling Coecient . . . . . . . . . . . . . . 88
5.2 Spin Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.2.1 Spin Testing with Resistive Load . . . . . . . . . . . . . . . . 90
5.2.2 Battery Charging Demonstration with Switching Load . . . . 99

6 Conclusion 105
6.1 Opportunities for Future Research . . . . . . . . . . . . . . . . . . . . 106

8
A Slider Augmented Pendulum Dynamics Derivations 109
A.1 Newton's Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.2 Lagrange's Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
A.3 Linearization about ψ = 0, ψ̇ = 0 . . . . . . . . . . . . . . . . . . . . 117

B Prototype Drafts 121

C Spin Stand Controls 171

9
10
Chapter 1

Introduction

1.1 Background
With the ever-advancing elds of embedded sensors, low power electronics, and wire-
less communications, there is a healthy demand for completely self-contained systems
capable of supplying power to such devices. For some applications, it is feasible to
use batteries as a source of energy; however, when the target device has a lifetime
power requirement that exceeds the capabilities of a battery, and it is not possible or
economical to perform battery changes, it becomes necessary to generate the needed
electricity from the energy available in the devices immediate environment [10]. This
is the case in many applications such as embedded sensor networks, medical implants,
distributed micro computers, and even wearable electronics [17, 15, 5]. Energy sources
that may be available in a remote environment and can be harnessed to generate elec-
trical energy include thermal gradients, solar radiation, ambient ow, and mechanical
vibrations [9].
Generating electricity on location to power an embedded system from vibration
has popularly been approached with the idea that energy can be obtained for free.
The devices used to generate electricity from environmental vibration have often been
called energy scavengers or energy harvesters , giving the impression that energy is
gained without expense [8]. The term energy reclamation gives some acknowledgment
to the fact that vibration in a mechanical structure is often due to undesirable losses in

11
an unrelated system and that converting this vibrational energy into usable electricity
is a way of reclaiming energy that would otherwise be lost [2]. While it is true
that unwanted vibration can yield useful electrical energy, it must be taken into
consideration that the manufacturing cost of the energy harvester itself may greatly
exceed the value of the electricity generated. The true value of such devices is not
their ability to save small quantities of wasted energy, but rather their ability to
supply electrical power to embedded systems indenitely with no external electrical
connections. A more appropriate name for such devices is the term inertial generator,
which both acknowledges that the devices convert energy from one form to another,
and that they depend upon inertial motion to generate electricity, with only one
external physical connection to their environment [17].
Inertial generators use any of several methods to converting inertial motion into
electrical energy, including the use of electromagnetic, piezoelectric, magnetostric-
tive, or electrostatic transducers [8]. Many inertial generators focus on single axis
vibration as a source of energy, but other interesting congurations have emerged
that focus on other excitations such as mechanical impact [13]. The conguration
optimized for impacts is attractive, because it attempts to avoid one of the most
signicant challenges faced by conventional vibrational inertial generators, which is
the uncertainty in the vibrational excitation. Most inertial generators have a sprung
mass conguration that allows them to achieve resonance at a specic frequency, and,
if the excitation frequency deviates from the natural frequency of the device, there is
a sharp decline in power output [12].
One strategy to deal with this narrow range of useful operation is to store energy
when the vibrations fall in the favorable frequency range, and to expend this stored
energy when they do not [16]. An alternate strategy is to modify the device param-
eters to optimize the power conversion at each frequency. Modication of the power
storage circuitry has been shown to alter the damping of the device, allowing the
device to be somewhat more useful at frequencies away from its natural frequency.
However, using electrical actuation to modify the natural frequency itself has been
shown never to result in higher net power conversion [7, 11]. Modifying the physical

12
properties of the device through actuation that only requires power at the time the
change is made, is a feasible solution, but it adds a large amount of complexity to the
system [9].

1.2 Motivation
The motivation for the work contained in this body of research stems from the desire
to develop a device which is capable of powering a system that is completely embedded
within a large rotating part that rotates about a horizontal axis, such as the propeller
of a large ship. The powered subsystem may require any range of power, from the
small amounts required for health monitoring and wireless transmissions, to the large
quantities that would be required to drive actuators for the purpose of modifying the
ow characteristics of the propeller [6]. The inaccessibility of the propeller of a large
ship, the harsh environment during operation, and the long service life required from
the power source make batteries an unacceptable solution. Due to the completely
enclosed environment with no signicant thermal gradients or sources of radiation, it
is necessary to rely on an inertial generator as a source of power.
The goals for this research are to develop and test a device suitable for embedded
installation within a large scale propeller. The device may need to be mounted at
any radius from the hub of the propeller, and a one meter radius was chosen as a
focus for the research as a representative quantity. The target operating rotation
rate of the device was chosen to be 250 RPM. Large ships often cruise for extended
periods of time at a constant throttle setting, so the device should be capable of
achieving resonance and producing appreciable power at steady state operation for a
wide range of rotation rates, both above and below the chosen target. The designed
device should also have a mechanical structure that is capable of operating under the
centrifugal loads experienced at the target radius and rotation rate and have a long
expected service life.

13
1.3 Previous Work and Chosen Approach
Prior to the work presented in this thesis, research was conducted toward this goal
and a device was designed to generate electrical power from the fore-aft vibration of
the propeller blade [1]. This device was designed with linearized natural frequency
that would allow the device to target vibration at ve times the rotation rate. This
approach of using the free vibrations of the propeller was found to be unsatisfactory
due to the low amounts of power that were available in the vibration of the structure.
In order to generate signicantly more power at the low frequency of the target
rotation rate, a more powerful excitation was necessary.
The previous work conducted on this problem was done from the mindset of vi-

brational energy harvesting , and naturally the fore-aft vibration mode of the propeller
would be expected to contain the most unwanted vibrational energy. The work pre-
sented in this thesis takes the more general mindset of inertial generation and arrives
at a much more satisfactory result. From the mindset of inertial generation, it is
natural to rst look at the inertial trajectory of the device, which includes all forced
motions including vibration, and then to design a device that takes the greatest ad-
vantage of that inertial trajectory.

1.3.1 The Inertial Trajectory

For this project, the device that is to generate the electricity is to be embedded at
some radius r in the interior of the propeller of a large ship, and several important
observations about the inertial motion of the device can be made. First, the path
of the device is circular, and that the circular path has a horizontal axis of rotation.
Second, the rotational inertia of a typical large scale propeller is more than sucient
to ensure that any reaction forces caused by the device do not aect the steady rate
of rotation of the propeller. The inertial path of the device is illustrated in Figure 1-1,
which shows the device embedded in the propeller at radius r, denes the rotation
rate of the propeller as ω , and shows the angle of the point of attachment from the
vertical as φ.

14
Figure 1-1: The box in the gure represents the location at which the device is
embedded within the propeller blade as viewed from the inertial frame I, ˆ Jˆ. The
device frame ı̂, ̂ is attached to the device and rotates with the device.

15
 At this point the problem can be viewed from two perspectives. The rst per-
ˆ Jˆ reference frame shown in Figure 1-1. This is a xed point of
spective is in the I,
reference that sees the propeller rotating in the counterclockwise direction about its
center, and sees the device both traveling in a circular path and rotating as it pro-
cesses along that path. From this xed point of reference, gravity is a conservative
potential eld, and the net work done on the device can only be found by evaluating
the reaction forces acting on the device over a full revolution.
A perspective that oers much more intuition on the situation is to view the prob-
lem from the ı̂, ̂ reference frame. This reference frame is attached to the device and
rotates with it. From this perspective, the device is stationary, and the centrifugal
forces that the device experiences can be viewed as being the result of a conservative
potential eld that extends radially outward from the center of rotation of the pro-
peller. The gravitational acceleration vector now rotates in the clockwise direction
with a positive rate of ω , as shown in Figure 1-2. Now it is clear that, along a given
axis of the device, gravity produces a sinusoidal excitation at a xed amplitude of 1 g
and at an exact frequency of ω . Viewed from the device frame ı̂, ̂, the excitation force
due to gravity is not conservative, and can do net work on the internal components
of the device. It is now possible to take advantage of the fact that the excitation
acceleration is both large and well known in the development of a device that will
eciently take in energy from this excitation and convert it into usable electrical
energy.
In Chapter 2, various device congurations are evaluated, and a conguration is
chosen that takes advantage of the circular inertial path experienced in the stated
application. Chapter 3 presents the details of the design of a prototype that is used
to verify the expected dynamics and performance of the chosen conguration. The
experimental equipment developed for use in testing the prototype device is presented
in Chapter 4. Chapter 5 discusses the key results of the experimental testing and
discusses their importance. The concluding remarks, summarizing the ndings of
this body of research are included in Chapter 6.

16
 ˆ Jˆ and the gravitational
Figure 1-2: Viewed from the device frame ı̂, ̂ the xed frame I,
acceleration vector are seen to rotate clockwise at an angular rate of ω , and the
centrifugal acceleration appears as a potential eld.

17
18
Chapter 2

Device Conguration

2.1 Explored Device Congurations

As explained in Chapter 1, any device that has internal motion as a result of being
driven through an inertial trajectory can be used as an inertial generator by damping
the motion through electromagnetic or piezoelectric means. In this section, several
devices of various congurations are compared to determine the characteristics that
are important to the function of the inertial generator. In order to make fair compar-
isons between the devices that will be explored, it is necessary to develop a method of
comparing the amount of energy that each conguration can generate per revolution
at a given mass and length scale.
The maximum amount of work that the excitation force can do on the device twice
per revolution is the product of the gravitational constant, the mass of the device,
and the distance over which the center of gravity of the device can move. To draw
fair comparisons between the devices that are explored below, they are normalized
such that they all have the same moving mass M , and a maximum length L. With
these constraints in place, the maximum displacement of the center of gravity of the
moving mass can be used as a measure of the maximum amount of work that the
excitation force can do on the device.

19
 Figure 2-1: Diagram of falling mass conguration.

2.1.1 Falling Mass

One simple conguration is a magnet of mass M constrained to slide in a tube of
length L as shown in Figure 2-1. At slow speeds, the acceleration of gravity causes
the mass to slide back and forth hitting the ends of the tube, and energy can be
removed from the system through a coil that damps the motion of the mass. The
problem with this class of systems is that, at high frequencies, there is not enough
time in each half cycle for the mass to travel the entire length of the tube.
The dynamics of the falling mass system are

mẍ = f (2.1)

in the time domain, and taking the Laplace transform, the system transfer function
can be written as
X 1
= (2.2)
F ms2

20
The target operating speed is 250 RPM, which is an angular velocity of

rad
ω = 26 (2.3)
sec

At this frequency, and for an excitation of 1 g, the magnitude of the displacement of

the mass is found to be
g
 
|x| =  2  = 0.0143 m (2.4)
s
This displacement of 14 mm is only achieved when no damping is applied, and there-
fore no power is produced. The application of damping will further reduce the am-
plitude, and the amount of work done on the system by the excitation force, which is
proportional to the displacement, is limited. For a system on the general length scale
of a approximately 10 cm to eectively take in energy at frequencies in the range of
the the target operation condition of 250 RPM, it must have a resonant response that
boosts the amplitude of motion signicantly. Adding a spring that connects the mass
to the frame of the device as shown in Figure 2-2 results in a system with natural
frequency s
k
ωn = (2.5)
m
When the mass spring system is driven at its natural frequency, large amplitudes can
be achieved and greater amounts of work can be done on the device in each revolution,
however when the driving frequency is not suciently near the natural frequency the
amplitude, and power output drops back to unsatisfactory levels.

2.1.2 Simple Pendulum

The simple pendulum shown in Figure 2-3 has a natural frequency for small motion
is given by
g
r
ωn = (2.6)
L
when it is swinging about a xed point in a gravitational eld. This is interesting,
because the natural frequency changes as the strength of the potential eld changes.

21
 Figure 2-2: Diagram of sprung mass conguration.

Figure 2-3: Diagram of simple pendulum conguration.

22
Figure 2-4: Diagram of a simple pendulum of length L swinging about point O in a
potential eld that expands radially from point A which is at a xed distance of r
from point O.

In considering the pendulum as a possible conguration, it is necessary to determine

how the pendulum will behave when it is swinging in a centrifugal potential eld as
opposed to a gravitational potential eld.

2.1.2.1 Restoring Torques on a Simple Pendulum in a

Centrifugal Potential Field
As shown in Figure 2-4, the force acting on the mass of the simple pendulum is applied
at an angle θ from vertical due to the curvature of the potential eld. The restoring
torque acting on the pendulum is given by

T = −L sin (ψ − θ) F (2.7)

23
and the magnitude of the force due to the centrifugal potential eld can be calculated
as
F = ω 2 m (r cos θ + L cos (ψ − θ)) (2.8)

From the geometry of the system, it can be shown that

r sin θ = L sin (ψ − θ) (2.9)

Solving for L, we have

sin θ
L=r (2.10)
sin (ψ − θ)
Combining Equation (2.7), Equation (2.8), and Equation (2.10), and using stan-
dard trigonometric identities, the expression for the restoring torque can be simplied
to

T = −mω 2 rL sin ψ (2.11)

Comparing this result to the restoring torque of a simple pendulum in a gravitational

eld which is given as
T = −mgL sin ψ (2.12)

it is interesting to nd that the two relations have similar form despite the diering
geometries of the potential elds.

2.1.2.2 Simple Pendulum Performance

From the similarity between Equation 2.11 and Equation 2.12, it can be shown that
a simple pendulum attached to a rotating part, as shown in Figure 2-4, will have a
natural frequency for small motion given by
s
ω2r r
r
ωn = =ω (2.13)
L L

The key result is that the system has a natural frequency that is always proportional
to the driving frequency. However, in the case of interest, r  L, and the natural

24
Figure 2-5: Diagram of a counterbalanced pendulum with unequal point masses and
equal arm lengths.

frequency of the pendulum is always higher than the driving frequency. This mismatch
of natural frequency leads to the same problems seen in the falling mass conguration
and amplitude and power intake suer.

2.1.3 Distributed Pendulum

The natural frequency of a distributed pendulum for small angles of motion when
swinging in a gravitational eld is given by
s
mgd
ωn = (2.14)
IO

where d is the distance between the pivot O and the center of gravity of the pendulum.
The rotational inertia about the pivot is found from

Z
IO = ρl2 dV (2.15)
V

25
where ρ is the density of the part, l is the distance from the dierential unit of volume
to the rotational axis passing through the point O, and the integration is taken over
the volume V of the pendulum. In the case where the pendulum is swinging in a
centrifugal potential eld, the natural frequency can be written as
s s
mω 2 rd mrd
ωn = =ω (2.16)
IO IO

For the case where the relation

mrd = IO (2.17)

is satised, the natural frequency will match the rotational rate under all operating
conditions, and resonance can be achieved.
A simple conguration that nearly optimizes the distributed pendulum congu-
ration with length constraint L and mass constraint m is shown in Figure 2-5. The
mass of the pendulum is represented by two point masses m1 and m2 situated at a
distance of
L
l= (2.18)
2
from the pivot and the rotational inertia about the pivot becomes

2
L mL2

IO = (m1 + m2 ) = (2.19)
2 4

By varying the distribution of the total mass m between the two points m1 and
m2 the distance d between the pivot and the center of gravity can be set to satisfy
Equation (2.17). The resulting value of the distance from the center of gravity to the
pivot is
IO L2
d= = (2.20)
mr 4r
At the target radius of 1 m and taking the length scale L of the device to be 0.1 m,
evaluation of Equation (2.20) yields an oset distance of

b = 0.0025 m (2.21)

26
Figure 2-6: Diagram of a simple pendulum which swings about the point O that drives
a ywheel which rotates about point B. The thickest lines in the gure represent the
teeth of the limited angle gear attached to the pendulum and the full gear attached
to the ywheel.

This result shows that for a distributed pendulum, even though resonance is attainable
at any input frequency and large angular motion can be achieved, the maximum
motion of the center of gravity of the device remains quite small, resulting in poor
performance.

2.1.4 Simple Pendulum With Geared Flywheel

Another method that will increase the rotational inertia of a pendulous system is
to employ a ywheel driven from the pendulum through a gear ratio, as shown in
Figure 2-6. The kinetic energy of the system can be written in terms of the rotational
inertia of the pendulum about its pivot IO , the rotational inertial of the ywheel about
its axis of rotation IB , and the angular velocities of the pendulum and the ywheel
ωP and ωF respectively. Using the gear ratio R to relate the rotational velocities of

27
the pendulum and the ywheel, the ywheel's rotation rate can be written as

ωF = RωP (2.22)

With use of this relation, the kinetic energy of the system can be written in terms of
only the angular velocity of the pendulum as

1 1 1 
EK = IO 2 + IB 2 = IO + IB R2 ωP2 (2.23)
2 2 2

and the equivalent rotational inertia IO0 can be dened as

IO0 ≡ IO + IB R2 (2.24)

The natural frequency of the system can then be expressed as

s s
mω 2 rL mrL
ωn = 0
=ω (2.25)
IO IO0

and the relation that must be satised to achieve resonance at every driving frequency
becomes
mrL = IO0 = IO + IB R2 (2.26)

These relations show that, by taking advantage of a gear ratio, a ywheel can be
employed to set the natural frequency of the system to be always equal to the driving
frequency of the excitation. Furthermore, if a conventional generator of rotational
inertial IB is used in place of the ywheel, the energy can be conveniently removed
from the system at that point. This system does not suer from any of the problems
that have reduced the motion of the center of gravity of the previously considered
congurations. The theoretical maximum displacement amplitude of the center of
the pendulum mass is the full length parameter L.

28
2.1.5 Conclusions
From the analysis of the congurations presented above, it has been seen that it
is possible to devise a system that when spun in the vertical plane at a radius r
will exhibit a resonant response to the excitation of gravity at any steady rotation
rate ω . It has been shown that this resonant response is key to achieving large
amplitudes of the internal motion that is necessary for the inertial excitation to do
net work on the device. Both the distributed mass pendulum and the pendulum
with the geared ywheel can achieve the requirements for resonance, but, among
the presented congurations, only the pendulum with the geared ywheel has the
potential of generating signicant amounts of energy per cycle.
The pendulum with geared ywheel does, however, have several key drawbacks
when the operating conditions of the system are considered. The gear teeth will be
subjected to cyclic loading, making them vulnerable to wear and fatigue. The contact
between the gears is also a point of frictional energy loss. If a belt were used instead,
the deterioration of the belt itself over time would limit the life of the device. The
use of a conventional generator in the place of a ywheel also has the drawbacks of
the frictional losses and wear associated with commutation brushes. Taking these
drawbacks into consideration, a new type of device was developed with long lifespan
as a primary design goal.

2.2 Slider Augmented Pendulum

The slider augmented pendulum was designed to preserve the desirable features of
the pendulum with geared ywheel conguration by both achieving resonance at
any rotation rate, and also allowing for large motion of the center of gravity of the
pendulum. The design does not include any of the wear items mentioned above,
and is expected to have a long life of service. This conguration relies solely on the
centrifugal forces acting on the device to produce restoring torques, and is therefore
capable of operating at a natural frequency that is matched to the excitation for all
driving frequencies. In the previously explored devices, the constant of proportionality

29
between the natural frequency and the driving frequency was set to unity by increasing
the rotational inertia of the system. In this conguration, the opposite approach is
taken. The rotational inertia is left essentially unchanged, and the natural frequency
is altered by reducing the net restoring torques acting on the pendulum by means of
an additional linkage and mass.

2.2.1 Conguration
The diagram of the the slider augmented pendulum system, shown in Figure 2-7,
ˆ Jˆ. One blade
gives a broad view of the system from the inertial reference frame I,
of the propeller is shown with the device attached such that the pivot point O of
the pendulous link is xed at a radius r from the hub of the propeller at point A.
The device reference frame ı̂, ̂ is xed to the point O and translates and rotates
in the inertial frame, but is stationary as observed from the attachment point of
the device. Figure 2-8 shows the device in greater detail as viewed from the device
reference frame. The device consists of a pendulous link of mass m and length L1 ,
and a secondary shorter link of length L2 which attaches the pendulous link to a slider
of mass mS , which is constrained to linear motion along the midline of the device.
Though called a slider, the mass mS can be constrained by exures, due to its small
motion, and will not have the losses and wear associated with a sliding constraint.
The distance between the center of gravity of the pendulous link and the point O is
not depicted in the diagrams, but appears as the parameter L in the equations of the
system dynamics. Also appearing in the dynamic equations are the parameters FC
and LC , which represent the electromagnetic damping force exerted by a coil onto a
magnetic portion of mass m and the moment arm at which it the force is applied,
respectively. The system employed to provide electromagnetic damping to the device
is described in detail in Chapter 3.

30
Figure 2-7: Slider augmented pendulum system overview showing the device attached
ˆ Jˆ. The device frame ı̂, ̂ is
to a propeller blade as viewed from the inertial frame I,
attached to the device and rotates with the propeller. The frame ı̂0 , ̂0 is attached to
the pendulous portion of the device.

31
 (a) (b)
Figure 2-8: Slider augmented pendulum system showing the linkage in the centered
state (a) and the deected state (b). Both views are shown in the device frame ı̂, ̂.
The slider of mass mS is constrained to linear motion along the midline of the device.

32
2.2.2 Slider Augmented Pendulum Dynamics
The equations of motion for the slider augmented pendulum were derived with two
assumptions. First, it is assumed that the slider mass mS is constrained to move pre-
cisely along the center line of the device, as shown in Figure 2-7, without friction. The
second assumption is that the secondary link connecting the slider to the pendulous
link has both negligible mass and rotational inertia. With these two assumptions, the
equations of motion for the slider augmented pendulum were derived with both New-
ton's Method and Lagrange's Method. The derivation by each method is presented
in detail in Appendix A. Three parameters

q
Z≡ L22 − (L1 sin ψ)2 (2.27)
!
L1 cos ψ
Y ≡ L1 sin ψ −1 (2.28)
Z
and
L2 sin2 ψ L21 cos2 ψ
" ! !#
L1 cos ψ
A ≡ L1 cos ψ −1 + 1 −1 (2.29)
Z Z Z2
are dened during the derivation to allow the nal equation of motion to be written
in the compact form

−mL (rω 2 sin ψ + g sin (φ + ψ)) + mS Y ω 2 (r + L1 cos ψ − Z)

ψ̈ =
I + mS Y 2
 O
mS Y −ψ̇ 2 A + g cos φ + LC FC
+ (2.30)
IO + mS Y 2

2.2.3 Natural Frequency Tracking

2.2.3.1 Frequency Tracking in Linearized Model
In order to consider the changes in the natural frequency of the system with changes
in the rotation rate for small angles of deection, the excitation force and external
electromagnetic forces in Equation (2.30) are set to zero, and the free undamped
equation of motion is linearized about the point ψ = 0, ψ̇ = 0. The linearization
procedure is documented in detail in Section A.3, and the linearized dynamics are

33
found to be h  i
L1
ω 2 −mLr + mS L1 (r + L1 − L2 ) L2
−1
ψ̈ = ψ (2.31)
IO
By similarity to the classic undamped mass spring system, the linearized system has
a natural frequency of
v h  i
u ω 2 mLr − mS L1 (r + L1 − L2 ) L1 − 1
u
L2
ωn =
t
IO
v  
u mLr − mS L1 (r + L1 − L2 ) L1 − 1
u
L2
= ω (2.32)
t
IO

This result shows that, within the linear regime, the natural frequency ωn is
directly proportional to the rotation rate ω . Now, all that is required to make the
linearized natural frequency equal to the forcing frequency for any rotation rate is to
set the constant of proportionality to unity, which leads to the requirement that

L1
 
mS L1 (r + L1 − L2 ) − 1 = mLr − IO (2.33)
L2

Frequency tracking can therefore be achieved by selecting appropriate parameters mS ,

L1 , and L2 for any pendulous link with parameters m, L, and IO mounted at a radius
of r. For the general case where r  L, it is naturally expected that mLr − IO > 0,
and therefore L1 > L2 , as depicted in Figure 2-7.

2.2.3.2 Frequency Tracking in Full Nonlinear Model

In order to verify the frequency tracking ability of the nonlinear device, the system
was simulated numerically from the full nonlinear equations of motion shown in Equa-
tion (2.30). The simulation was performed in MATLAB using the ode45 solver, with
no gravitational forces or Lorentz forces acting on the system. Simulations were run
from the initial conditions of ψ0 = 10◦ and ψ̇0 = 0 to the point at which the linkage
reached the midline of the device at ψ = 0. The time at the termination of the simu-
lation, representing one fourth of the total period of oscillation, was used to nd the
undamped natural frequency of the system. Representative values were used for the

34
 350

Frequency of Undamped Oscillation ωn (RPM)

300

250

200

150

100

50

0
0 50 100 150 200 250 300 350
Rotation Rate ω (RPM)

Figure 2-9: Nonlinear simulation results showing the natural frequency of the un-
damped system when released from rest at an initial deection of ψ0 = 10◦ and no
forcing excitation g = 0. Representative system parameters used and mS hand tuned
at a single frequency.

physical parameters of the system. The slider mass mS was hand tuned at a single
rotation rate of 250 RPM, such that the natural frequency of the system matched the
driving frequency precisely at that rate of rotation. The system was then simulated
over a range of driving frequencies and the period and natural frequency of the system
was recorded at each rotational rate.
The simulation results are presented in Figure 2-9, and verify the precise frequency
tracking of the nonlinear system operating at a xed amplitude of motion. The fre-
quency tracking ability shown by the system was expected, because all of the potential
energy terms in Equation 2.30, viewed from the device frame ı̂, ̂, are proportional to
ω2.

35
2.2.3.3 Amplitude Dependency of Natural Frequency
For the system parameters used in the nonlinear simulations, it was noted that the
slider mass mS required to tune the device did not satisfy Equation (2.33), which was
found to be the requirement for natural frequency tracking in the linearized model.
This apparent discrepancy is due to the fact that the natural frequency depends on
the amplitude of the motion, and the natural frequency for small displacements found
in the linearized model is not the same as the natural frequency when released from
ψ0 = 10◦ .
It is interesting that the system shows its nonlinearities in the relation between
the natural frequency ωn and the angle at release ψ0 at each rotation rate ω , but at
any given ψ0 there is a linear relationship between ω and ωn . Plotting the ratio ωn /ω
as a function of the angle at release for any rotation rate produces the single curve
shown in Figure 2-10. It should be noted that the shape of this curve is dependent
on the parameters of the system, but can be expected to start at a higher frequency
for small motion and drop to lower frequencies as the motion increases, due to the
increased eectiveness of the force produced by the slider mass as the angles θ and
(ψ − θ) increase with increasing ψ .

2.2.4 Statically Stable Range of Motion

At every position within its allowed range of motion, it is important that link 1 will
have a restoring torque that will accelerate it back toward the ψ = 0 to ensure that
the linkage cannot become trapped in some local energy minima. As the rotation rate
is increased, the centrifugal forces increasingly outweigh the excitation force provided
by gravity and the force provided by the coil, so these forces are not taken into
consideration in the analysis of the stable range of motion. It is, however, necessary
to note that at relatively low rotation rates, where the excitation of gravity exceeds all
other forces, the linkage will have large deections, and will move until the mechanical
stops are reached. It is, therefore, guaranteed that the linkage will, at some time,
reach every point within its mechanically limited travel and, due to the unknown

36
 1.5
Frequency Ratio ωn / ω

0.5

0
0 5 10 15
Displacement at Release ψ°0

Figure 2-10: Nonlinear simulation results showing the amplitude dependence of the
frequency ratio ωn /ω for the system with the same parameters as used in Sec-
tion 2.2.3.2. The system was released from rest at various initial deections ψ0 and
at various rotation rates ω , with each rotation rate producing an identical curve.

37
transient accelerations ω̇ , the linkage may conceivably achieve any deection at any
rotational rate. It is therefore necessary to ensure that the linkage is stable for any
deection within its mechanically limited range of motion for the conditions g = 0,
FC = 0.
Stability can be tested by nding the range of deection for which there is a
restoring torque that tends to bring the linkage back to center. From the equation of
motion given in Equation (2.30) and setting g = 0, FC = 0, and ψ̇ = 0, and assuming
a positive deection in ψ , the stability condition can be written as

−mLrω 2 sin ψmax + mS Y ω 2 (r + L1 cos ψmax − Z)

<0 (2.34)
IO + mS Y 2

Recognizing that IO > 0 and mS > 0 and rearranging terms simplies the condition
to
mLr sin ψmax − mS Y (r + L1 cos ψmax − Z) > 0 (2.35)

Substituting in the value of Y from Equation (2.28) gives the stability condition as
!
L1 cos ψmax
mLr sin ψmax − mS L1 sin ψmax − 1 (r + L1 cos ψmax − Z) > 0 (2.36)
Z

In the expected case that L1 > L2 , sin ψmax > 0 due to the geometric constraints
of the linkage itself, and further simplication leads to the relation
!
mLr L1 cos ψmax q 
> −1 r + L1 cos ψmax − L22 −Z (2.37)
m S L1 Z

Substituting the value of Z from Equation (2.27) the stability condition becomes
 
mLr L1 cos ψmax
 q 
>  q −1 r + L1 cos ψmax − L22 − (L1 sin ψmax )2
m S L1 L2 − (L1 sin ψmax )2
2
(2.38)
This result gives some insight on the system by showing that as mS is increased, the
range of stable motion is decreased. It can be shown that the linkage will become
unstable at some ψmax < sin−1 (L2 /L1 ) < (π/2) by substituting ψ = sin−1 (L2 /L1 )

38
into the right side of Equation (2.38) and recalling that for ψ < (π/2), cos ψ > 0.
With this substitution, the right hand side of Equation (2.38) becomes
!
L1 cos ψ
− 1 (r + L1 cos ψ) = ∞ (2.39)
0

which clearly cannot satisfy the stability condition.

Since the right side of Equation (2.38) has been shown to increase without bound
while ψ < sin−1 (L2 /L1 ), the linkage will always have some deection at which stabil-
ity is lost before inversion can occur. Mechanical stops must be placed to ensure that
the linkage does not travel beyond the stable range of motion, or the device could
become trapped at the end of its range of motion until the rotation rate is lowered
to the point that the excitation force becomes large in comparison to the centrifugal
force.

2.2.5 Sensitivity to Misalignment

The slider mass and link 2 can be thought of as a subsystem that acts as a posi-
tive feedback loop, reducing the natural stability of the pendulous link 1. Using this
perspective from the controls discipline, it is reasonable to expect that this positive
feedback will cause the system to become more sensitive to disturbances. This char-
acteristic is seen in the increased sensitivity of the system to misalignment, which can
cause the linkage to become trapped against its mechanical stops during high speed
operation.
Figure 2-11 shows the device mounted to the base part with a misalignment angle
of γ between the midline of the device and the true radial ray that passes from the
center of rotation of the base part outward through the main pivot of the device
at point O. Starting with Equation (2.30) and evaluating the expression at ψ̈ = 0,
ψ̇ = 0, and FC = 0, the static torque balance about point O is obtained as

mLr sin ψ = mS Y (r + L1 cos ψ − Z) (2.40)

39
Figure 2-11: Diagram of device mounted at a misalignment angle γ between the
device center line and true radial line passing through the center of rotation of the
propeller at A and the main pivot of the device at O.

40
To nd the relation between the misalignment angle γ and the resulting deection ψ
at which static equilibrium is achieved, the torque balance can be written as

mLr sin (ψ − γ) = mS cos γY (r + L1 cos ψ − Z) (2.41)

to accommodate the misalignment angle. Substituting the trigonometric relation

sin (ψ − γ) = sin ψ cos γ − cos ψ sin γ (2.42)

and rearranging terms yields

sin ψ mS
tan γ = − Y (r + L1 cos ψ − Z) (2.43)
cos ψ mrL cos ψ

Substituting for Y from Equation (2.28) and further simplifying gives the condition
for equilibrium as
" ! #!
m S L1 L1 cos ψ
γ = tan −1
tan ψ 1 − − 1 (r + L1 cos ψ − Z) (2.44)
mrL Z

Linearizing about the point ψ = 0 with the approximation that


∂γ 
γ ≈ γ|ψ=0 +  ψ (2.45)
∂ψ ψ=0

was performed. The rst term in the approximation

m S L1 L1
    
γ|ψ=0 = tan −1
0 1− − 1 (r + L1 − L2 ) = tan−1 (0) = 0 (2.46)
mrL L2

and the second term can be found by dening a variable for substitution
" ! #
m S L1 L1 cos ψ
B ≡ 1− − 1 (r + L1 cos ψ − Z) (2.47)
mrL Z

and using the relation

d tan−1 u 1 du
= (2.48)
dx 1 + u2 dx

41
Substitution allows the partial derivative to be written as
!
∂γ 1 B ∂B
= + tan ψ (2.49)
∂ψ 1 + (tan ψB) cos2 ψ ∂ψ

which can then be evaluated as

 !
∂γ  1 ∂B
 = B|ψ=0 + 0
∂ψ ψ=0 1 + (0) ∂ψ
m S L1 L1
 
= 1− − 1 (r + L1 − L2 ) (2.50)
mrL L2

Assembling Equation (2.45), Equation (2.46), and Equation (2.50) the linearized
sensitivity transfer function relating the misalignment angle γ to the equilibrium angle
ψ is obtained as
ψ 1
=   (2.51)
γ 1− mS L1 L1
− 1 (r + L1 − L2 )
mrL L2

When the system is tuned for frequency tracking by satisfying Equation (2.33), we
have that
ψ 1 mrL
= mrl−IO = (2.52)
γ 1 − mrL IO

Noting that the natural frequency of an uncompensated pendulum (without the slider
and linkage) is s s
mω 2 rL mrL
ωn U = =ω (2.53)
IO IO
The sensitivity transfer function can be expressed as

ψ mrL ω2
= = n2 U (2.54)
γ IO ωn C

where ωn C is the compensated natural frequency. When the pendulum is approxi-

mated to have no rotational inertia about its own center of gravity, so that

IO = ml2 (2.55)

42
we nally obtain
ψ r
= (2.56)
γ L
This result, that the sensitivity to misalignment is directly proportional to the
square of the ratio of the uncompensated and compensated natural frequencies, be-
comes an important factor in designing and mounting a device of this conguration.
In the point mass approximation of the pendulum this sensitivity is proportional to
the ratio of the radius at which the device is mounted to the length of the pendulous
link. This clearly shows that alignment will be a primary concern for the system if at-
tempts are made to reduce the length scale of the device to very small fractions of the
radius at which it is mounted. It also implies that the device, at a given length scale,
becomes more robust as the mounting point is brought toward the axis of rotation of
the part on which it is mounted.

2.3 Extracting Energy Through Damping

The slider augmented pendulum system has been shown to be able to achieve a
resonant response to a steady input of any frequency for a xed target amplitude
of motion. This resonance allows energy to build within the system to produce and
maintain large motion, and also indicates that the phase relationship between the
excitation and the motion is such that the excitation is always transmitting energy
into the system. It has also been shown that mechanical stops must be utilized to
limit the travel of the linkage to its stable range of motion. In order to obtain usable
electrical energy from the system, a coil and magnet pair are used to damp the motion
of the linkage. From Lenz Law, the relative motion between the magnet and the coil
will cause a current to ow through the coil, creating an electromagnetic eld that
oppose the relative motion. This section focuses on investigating the parameters that
are important to the optimization of the power transfer to the load.

43
Figure 2-12: Damped harmonic oscillator system which captures the important dy-
namics of the slider augmented pendulum system when a frequency dependent spring
rate is applied.

2.3.1 Damped Harmonic Oscillator

Due to the complexity and nonlinearities of the slider augmented pendulum system,
it is somewhat dicult to gain an intuitive grasp of the optimization of the power
transfer to the load. For this reason, it is benecial to rst consider instead a linear
mass spring system that captures the important aspects of the slider augmented
pendulum conguration. Such an analogous system is shown in Figure 2-12. The
block of mass m is acted upon by a sinusoidal excitation force

f (t) = f sin (ωt) (2.57)

a reaction force from a spring with a frequency-dependent spring constant

k (ω) = ω 2 m (2.58)

and a force due to the damping b. The choice of the frequency-dependent spring
constant gives the simple linear system the same frequency tracking and matching
properties as the slider augmented pendulum, with
s
k
ωn = =ω (2.59)
m

44
 The transfer function of the system

X (s) 1
= (2.60)
F (s) ms2 + bs + k

evaluated at s = jω , gives
X (jω) 1
= (2.61)
F (jω) jωb
and shows that the amplitude of motion is inversely proportional to both the damping
constant and the driving frequency. The transfer function

V (jω) X (jω) jω 1
= = (2.62)
F (jω) F (jω) b

between the velocity of the system and the excitation shows the relation between
velocity and excitation is the damping value itself. The relation can be brought into
the time domain as
ẋ (t) 1
= (2.63)
f (t) b
In the time domain, the instantaneous power being damped from the system is given
by
f (t)2 f 2 sin2 (ωt)
P (t) = f (t) ẋ (t) = = (2.64)
b b
and the average power can be found by integrating over one complete cycle and
dividing by the period, to arrive at

2π
f2 Z ω ω f2
P = sin2 (ωt) dt = (2.65)
b 0 2π 2b

From the expression for the average power, it can be seen that the power increases
without bound as the damping is reduced toward zero, but at the same time the
amplitude given in Equation (2.61) also grows without bound.
In any physical system, there will be a limit to the allowable mechanical motion,
and some minimum damping level will be required to prevent the device from hitting
its mechanical stops. This optimal damping level is found from the system transfer

45
 2

1.8

1.6

1.4

1.2
b*xmax / f

0.8

0.6

0.4

0.2

0
0 2 4 6 8 10
ω (rad/sec)

Figure 2-13: Normalized optimal damping as a function of frequency.

function as a function of frequency to be

f 1
bO = (2.66)
xmax ω

and is plotted in Figure 2-13. Combining this result with Equation (2.65), the optimal
power output of the device is found to be

f2 f xmax ω
PO = = (2.67)
2b 2

and increases linearly with frequency.

2.3.2 Applying Electromagnetic Damping

In the calculations above, the optimal damping and optimal power output were found
as a function of frequency for a system of limited range of motion. Next, we consider

46
Figure 2-14: Magnet coil conguration that results in electromagnetic damping of the
motion of the magnet assembly relative to the xed coil. RC represents the internal
resistance of the coil and RL the impedance of the attached electrical load.

how damping is applied to the system electromagnetically. The system shown in

Figure 2-14 has a conguration often used in maglev linear actuators, and consists of
a two magnet assembly which moves within a xed coil of wire [3, 14]. By Faraday's
Law, the motion of the magnet will produce an electromotive force in the coil

VC = V BLT N (2.68)

which is proportional to the velocity V , the density of the magnetic eld B , and the
linear length of wire in the eld given by the product of the length per turn LT and
the number of turns N . The Lorentz force

FEM = IBLT N (2.69)

relates the electromagnetic force FEM exerted on the magnet to the current in the
coil using the same parameters. Rearranging terms, the electromagnetic constant of
the magnet-coil pair can be dened as

VC FEM
KEM ≡ = = BLT N (2.70)
V I

47
With the assumption that the reactance of the coil can be considered purely resistive
at the excitation frequency, the resistances in the coil and in the load can be used
to relate the voltage and current across the coil, and the damping constant can be
written as
2
KEM
b= (2.71)
RC + RL
This result is important because it introduces two key deviations from the ideal case
investigated in Figure 2-12. First, the resistance of the coil introduces losses in the
system, so that even if ideal damping is applied, only a portion of the damped power
will make it to the load. The second deviation is that the damping has a nite upper
limit, set by the internal resistance of the coil

2
KEM
bmax = (2.72)
RC

and at low frequencies the system will not be able to achieve the high levels of damping
required to prevent collision with the mechanical stops.
At high frequencies, when the damping is set to the optimal value in Equa-
tion (2.66) to maximize the total power removed through damping, the portion of
the power that is delivered to the load is given by

f 2 RC
!
2 2
PL (t) = P (t) − PC (t) = f xmax ω sin (ωt) − IC(t) RC = f xmax ω − sin2 (ωt)
KEM
(2.73)
and the average power delivered is found to be

f xmax ω f 2 RC
PL = − 2
(2.74)
2 2KEM

The rst term in this expression is the maximum rate at which energy can be removed
through damping, and the second term represents the losses in the coil due to its
internal resistance. The second term shows that the coil losses are constant when
optimal damping is applied to the system.
At low frequencies, the maximum damping that the coil can provide is insucient

48
to prevent collision with the mechanical stops, and if maximum damping is applied
by directly shorting the leads of the coil, no power is transmitted to the load. This
condition occurs when PL in Equation (2.74) reaches zero, showing that the coil loss
term is equal to the rate at which the excitation does work on the system.

f RC
ωA = 2
(2.75)
xmax KEM

At these low frequencies, the mechanical stops will be reached by the mass, and the
electromagnetic forces will have little impact on the motion of the device. Under
these conditions, the motion of the device, and therefore the voltage across the coil,
can be considered to be of constant amplitude, and maximum power is delivered to
the load when the load is impedance matched to the coil by setting

RL = RC (2.76)

Mid-range frequencies exist at which it is possible to provide enough damping that

the system will not impact the mechanical stops. However, the load resistance must
be lowered below the coil resistance. Because impedance matching is desirable at low
frequencies, it is reasonable to use the resistance of the coil as the minimum load
resistance for all frequencies, and therefore allow the mechanical stops to be reached
at these mid-range frequencies as well. The frequency ωB at which the damping
applied by impedance matching the load to the coil is the optimal damping, is found
from Equation (2.71) and Equation (2.66) to be

2f RC
ωB = 2ωA = 2
(2.77)
xmax KEM

To reduce coil losses at frequencies above ωB and to shift both ωA and ωB to lower
frequencies, it is desirable to maximize the ratio KEM
2
/RC .

49
 2.5
Pin at Optimal Damping
PL at Optimal Damping

2
P K2EM / (f 2 RC)

1.5

0.5

0
0 1 2 3 4 5
ω xmax K2EM / (f RC)

Figure 2-15: Normalized curves showing power taken into the system and power
delivered to the load as a function of frequency. On the normalized plot ωA = 1 and
ωB = 2.

50
Chapter 3

Prototype Design

3.1 Choosing Nondimensional System Parameters

With the dynamics and performance of the device and the dependency on each param-
eter laid out in detail in the previous chapter, the task becomes to choose a desirable
set of parameters for use in the design of a prototype device. It is convenient to begin
with the parameters that can be represented in non-dimensional form and therefore
have the least dependence on the details of the nal mechanical structure. It has been
shown that the necessary frequency tracking can be achieved for a system of any given
set of parameters through the choice of the slider mass mS and the link lengths L1
and L2 . Taking a representative set of system parameters that includes the target
radius of 1 m and appropriate values for m, L, and IO , numerical simulation using
various lengths L1 , and L2 can determine the required value of mS and the natural
frequency of the system ωn as a function of displacement ψ .
By dening the linkage ratio as L2 /L1 and the frequency ratio as ωn /ω , the am-
plitude dependence and stable range of motion can be evaluated in terms of these
nondimensional parameters. Figure 3-1 shows the amplitude dependence on the fre-
quency ratio for various linkage ratios. Each curve begins at zero displacement and is
extended to the point at which static stability is lost. The key result from this family
of curves is that, for a set desired amplitude, the linkage ratio denes the deviation
from unity of the frequency ratio at low amplitudes. This result is important to the

51
 2.5
L2/L1 = .25
L2/L1 = .35

2 L2/L1 = .45
L2/L1 = .55
Frequency Ratio ωn / ω

L2/L1 = .65
L2/L1 = .75
1.5
L2/L1 = .85
L2/L1 = .95

0.5

0
0 2 4 6 8 10 12 14
Displacement at Release ψ°0

Figure 3-1: Plot of the ratio between the natural frequency and the driving frequency
ωn /ω as a function of release amplitude ψ0 for several values of the linkage ratio
L2 /L1 . For each curve a slider mass mS has been chosen to achieve a frequency ratio
of unity at an amplitude of ψ = 10◦ for a representative set of system parameters.
Each curve extends to the limit of static stability.

dynamics of the device because if the mismatch is too large, the device will not be
capable of building up amplitude from rest at the centered position while spinning at
high frequencies. It is therefore desirable to have a large linkage ratio to minimize the
deviation of the frequency ratio from unity. Considering the endpoints of the curves,
at which point stability is lost, it is seen that a large linkage ratio oers the largest
stable range of motion.
When the proportions of link 1 are held constant, the rotational inertia IO scales
linearly with the mass m and it becomes meaningful to dene the mass ratio as mS /m.
The linkage ratio, for an otherwise constant set of parameters, will dictate the mass
ratio required to set ωn equal to ω . This relationship is presented in Figure 3-2,
and shows that the mass ratio increases without bound as the length ratio of unity

52
 10

7
Mass Ratio mS / m

0
0 0.2 0.4 0.6 0.8 1
Linkage Ratio L2 / L1

Figure 3-2: Plot of the required mass ratio mS /m as a function of linkage ratio L2 /L1
for a frequency ratio of unity ωn = ω at an amplitude of ψ = 10◦ for a representative
set of system parameters.

is approached. The required travel of the slider mass as a function of linkage ratio
is shown in Figure 3-3, and shows the opposite trend. Small slider displacements
are desirable to enable the use of exures as a means of constraining the slider with
minimal losses. From these results, the choice of the linkage ratio becomes a trade o
between reduced nonlinearities in natural frequency, enhanced stability, and reduced
slider motion at large ratios and reduced slider mass at low ratios. Weighing all of
the benets for a target amplitude of ψ = 10◦ , a length ratio of L2 /L1 = 0.75 was
chosen for use in the prototype design, resulting in a mass ratio of mS /m = 1.88, and
a normalized slider displacement of 0.005.

53
 0.1

0.09
Normalized Slider Motion (x Max−x 0) / L1

0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0
0 0.2 0.4 0.6 0.8 1
Linkage Ratio L2 / L1

Figure 3-3: Relationship between normalized slider maximum displacement

(xmax − x0 ) /L1 and linkage ratio L2 /L1 .

54
3.2 Magnet Assembly and Coil Design
The magnet assembly consists of two magnets placed on either side of a spacer,
as shown in Figure 2-14, with a ux return path provided around the outer loop to
contain the eld and to reduce losses associated with eddy currents that may otherwise
be induced in nearby components. Because the magnet assembly has signicantly
more weight than the coil, the coil is held stationary and the magnet assembly is
integrated into the the pendulous link. In order to achieve a range of motion of ψ =
±10◦ while still holding the tight clearances that are desired to maximize eciency,
the magnet assembly was designed with the curvature shown in Figure 3-4. The
design consists of many at plates to reduce the manufacturing complexity of the
steel parts. Simplifying the geometry of the magnet assembly did, however, pose
challenges on the shaping of the magnets themselves. The magnets were cut using a
wire EDM processes after the magnet stack was assembled.
The magnet assembly is designed to provide a very concentrated magnetic eld
at a right angle to the motion of the coil to obtain a large eective eld density B .
This high eld density helps to increase the electromagnetic constant KEM dened
in Equation (2.70), and therefore increases the eciency of the device as described in
Section 2.3.2 by increasing the ratio

2
KEM (BLT N )2
= (3.1)
RC RC

The coil itself was also designed to maximize this ratio to achieve better eciency.
Because the length of a single turn of wire in the coil LT was xed by the geometry of
the magnet assembly, coil design focused on maximizing the ratio N 2 /RC . The cross
sectional area of the coil is limited by the tolerances between the coil and magnet
assembly and the maximum number of turns that will t into the available space is
given by
AS
N = rP (3.2)
AW
where As is the area of the cross section of the coil, AW is the cross section of the wire,

55
Figure 3-4: Integrated pendulum and magnet assembly structure showing magnet
orientations and coil position in centered (top) and deected (bottom) state. The
coil is held xed by an external structure not included in the diagram.

56
and rP is the packing ratio. The internal resistance of the coil RC can be calculated
as a function of the resistivity of the copper in the wire ρ, the cross sectional area of
the wire AW , and the total length of wire N LT as

ρN LT
RC = (3.3)
AW

Combining these equations shows that

N2 N AW r P AS
= = (3.4)
RC ρLT ρLT

and the packing ratio is the only term that can be altered to decrease the internal
losses in a copper coil of given geometry.
The coil design reects the desire to maximize the packing ratio by allowing the
wire to round the corner of the coil with a large enough radius that the windings can
be maintained in a at and parallel orientation. A section view of the coil situated
in the magnet assembly is presented in Figure 3-5 and shows the tight tolerances
that are held between the coil and the surrounding structures. To further increase
the packing ratio, magnet wire with a thin non-bonding insulation was used, and
the coil was wound and epoxied by hand on a custom CNC cut winding xture. It
should be noted that while only the packing ratio aects the eciency of the coil, the
choice of wire will aect the number of turns, and will therefore aect the balance
of voltage and current delivered to the load. The choice of wire can therefore be
made to minimize the losses in the load. For the prototype device, it was desirable
to match the coil voltage output to the input range of the data acquisition system,
and 32 AWG magnet wire was selected. A photograph of the wound coil is shown in
Figure 3-6.

3.3 Mechanical Linkage Design

The physical design of the mechanical linkage is a critical challenge of its own. At
the testing condition of r = 1 m and rotation rates up to the desired 350 RPM, the

57
Figure 3-5: Section view of coil and magnet assembly along the midline of the de-
vice showing the t of the coil between the magnet stack and the side plates of the
assembly.

Figure 3-6: Photograph of the hand wound coil showing the organized orientation of
the windings to maximize the packing ratio.

58
device will experience centripetal accelerations of

2
2π m

ω 2 r = 350 = 1343 = 137 g (3.5)
60 s2

Because of these high loads, it is necessary to carefully design the linkage to avoid
mechanical failure and to prevent the high loads from causing friction or deections
that would prevent the device from preforming as desired. The design was completed
with a factor of safety of n ≥ 2 against failure in both shear and bending at all critical
points in the design.

3.3.1 Link Forces

The force in each link varies with time and has components due to both the centrifugal
forces and reaction forces produced by the dynamic motion of the linkage itself. The
forces in each link and the reaction forces of the slider against its constraint were
found from the dynamics of the system using the Newtonian approach. The resulting
tensile forces in each link are

FL1 = mACG ˆ0 + FS cos (θ − ψ)

Z L1
 
= m (ACG ı̂ sin ψ + ACG ̂ cos ψ) − FL2 cos ψ + sin2 ψ
L2 L2
Z L1
  2   
2
= m L ω + ψ̇ + rω cos ψ − FL2 cos ψ + 2
sin ψ (3.6)
L2 L2

and

−mS  
FL2 = −FS = g cos φ − ψ̈Y − ψ̇ 2 A + ω 2 (r + L1 cos ψ − Z)
cos θ
m S L2  
= ψ̈Y + ψ̇ 2 A − ω 2 (r + L1 cos ψ − Z) − g cos φ (3.7)
Z

The force exerted by the slider mass on its constraint in the tangential direction is
!
L1 sin ψ
FmS = FL2 sin θ = FL2 (3.8)
L2

59
 2560
FL1
Cos φ
2550

2540

2530
Force (N)

2520

2510

2500

2490
0 π 2π 3π 4π 5π 6π 7π 8π 9π 10π
Position φ (rad)

Figure 3-7: Tensile forces in link 1 as a function of propeller position calculated

through numerical simulation at the maximum desired test speed of 350 RPM.

Numerical simulation was employed to generate the time varying values of the link
and constraint forces plotted as a function of propeller position φ in Figure 3-7,
Figure 3-8, and Figure 3-9.

3.3.2 Symmetrical Conguration

Due to the high forces in the links, it was necessary to use a symmetrical congura-
tion that prevents bending moments from being transferred through the connections
between links. The symmetrical conguration of the design is shown in Figure 3-10.
The device is anchored by a plate and bearing block that sit between the two side
plates of the integrated link 1 magnet assembly, with minimal clearance to reduce
the bending moments exerted on the main pivot shaft. For logistical purposes, the
paired secondary links are placed on the outside of the slider mass side plates and

60
 −1620
FL2
Cos φ
−1630

−1640

−1650
Force (N)

−1660

−1670

−1680

−1690
0 π 2π 3π 4π 5π 6π 7π 8π 9π 10π
Position φ (rad)

Figure 3-8: Tensile forces in link 2 as a function of propeller position calculated

through numerical simulation at the maximum desired test speed of 350 RPM.

61
 400
Fm2 i
300 Cos φ

200

100
Force (N)

−100

−200

−300

−400
0 π 2π 3π 4π 5π 6π 7π 8π 9π 10π
Position φ (rad)

Figure 3-9: Reaction forces between slider of mass mS and its constraint in the
tangential direction calculated through numerical simulation at the maximum desired
test speed of 350 RPM.

62
Figure 3-10: Top view of device showing the linkage conguration. Not shown in the
diagram are the upper and lower plates that connect the two slider mass side plates
to form a rigid box structure. All linear dimensions shown in inches.

are connected to studs that are press t into blocks attached to the magnet assembly.
This press t conguration was chosen because the use of a through shaft was not
possible, and the press t oers much higher strength than a threaded connection.
The slider mass itself is a box structure formed by the two side plates shown, and
upper an lower plates that bridge across them to form a single rigid structure that
prevents the linkage from warping. Various masses can be attached to these upper
and lower plates to adjust the overall mass of the slider during testing.

63
3.4 Slider Constraints
The motion of the slider is small due to the choice of the linkage ratio in Section 3.1,
and is plotted as a function of propeller position from numerical simulation as shown
in Figure 3-11. The motion of the slider was deliberately designed to be small enough
that exures could be used as a low loss approximation of the sliding constraint.
Flexures would be the method of choice for a production device, but a mechanical
linkage based constraint was chosen for the prototype device to reduce design time
and prototyping cost. The linkage design approximates linear motion over the small
deections of the slider mass by constraining it with three long links as shown in
Figure 3-12. The upper ends of the links, as seen in the gure, are attached to
the frame of the device and the lower ends are attached to the sides of the slider
mass. Three links were chosen, two on the back side and one on the front, to prevent
over-constraining the slider assembly.

3.5 Device Frame and Attachment

An outer framework for the device was designed to provide several important func-
tions. The primary function of the framework is to rigidly support the internal com-
ponents of the device, and to provide a secure attachment to the test stand. The
centripetal forces acting on the linkage itself are transmitted directly through the
back wall of the frame from the test stand. The remaining portions of the frame only
need to rigidly support: the coil assembly with the integrated mechanical stops, the
slider mass support links, and an encoder that measures deection of the main pivot
shaft. Having only light loads acting through the majority of the frame allows for a
relatively lightweight aluminum construction that keeps overall device weight down,
reducing the centrifugal loading on the test stand.
The frame must fully enclose all of the moving portions of the device, to prevent
airow from disturbing their motion, but still allow for quick modications and safety
checks to be performed during testing. The open box construction shown in the

64
 0.35
X
0.3 Cos φ

0.25
Slider Displacement ∆X (mm)

0.2

0.15

0.1

0.05

−0.05

−0.1
0 π 2π 3π 4π 5π 6π 7π 8π 9π 10π
Position φ (rad)

Figure 3-11: Slider deection as a function of propeller position calculated through

numerical simulation.

65
Figure 3-12: Side view of device showing coil assembly, linkage conguration, and
slider constraints. All linear dimensions shown in inches.

66
Figure 3-13: View of assembled prototype device showing the internal linkage installed
in the frame, and the entire assembly mounted to the test stand arm.

Figure 3-13 incorporates four clear polycarbonate panels that allow for easy visual
inspection, and can be easily removed to modify the mass of the slider. The two steel
bars on the front of the device are the only components of the frame that must be
removed for full linkage removal and disassembly, ensuring that the frame geometry
does not change during disassembly and reassembly. These two steel bars also give
the frame added strength to prevent any debris from escaping the enclosure should
any component of the linkage fail.

3.6 Prototype Parameters

A summary of the parameters that were chosen for use in the design of the device
that appear in Equation (2.30) are the link lengths L1 = 2.5 in and L2 = 0.75 in.

67
Figure 3-14: Photo of the assembled prototype device mounted to the test stand arm.

The radius at which the coil interacts with the magnet assembly is LC = 2 in. The
mass of the pendulous link m, the distance L between the pivot point O and the
center of gravity of the link, and its rotational inertia IO were found from the 3D
solid model shown in Figure 3-13. From the solid model of the link 1 assembly,
the mass, location of the center of mass, and the rotational inertia about the pivot
were measured as m = 1.39 lbm, L = 1.755 in, and IO = 5.584 lbm in2 respectively.
Through simulation, the required slider mass to achieve a frequency ratio of unity was
found to be ms = 2.621 lbm. Detailed drafts of the individual parts of the prototype
system are shown in Appendix B.

68
Chapter 4

Experimental Setup

The goals for laboratory testing of the prototyped system were primarily to verify
the dynamics of the device, and to characterize and quantify the performance of the
system. It was also desirable to demonstrate the ability of the device to perform
the task of charging batteries. In order to support these goals, it was necessary to
provide the device with proper excitation conditions and an adjustable load. Testing
required instrumentation capable of measuring the deection of the device, its phase
relative to the excitation, and the power output of the device given by the voltage
and current delivered to the load. Conducting the desired testing required the design
and development of custom equipment including a computer controlled spin stand to
provide the designed trajectory, and two dierent microcontroller based adjustable
loads. This chapter gives an overview of the capability of these systems, and their
integration with the data acquisition and control hardware.

4.1 Spin Stand

The primary requirement of the spin stand is the ability to hold the device securely
at a 1 m radius and smoothly spin it at speeds up through 350 RPM, with enough
torque to overcome wind drag and provide the necessary accelerations. Full computer
controlled closed loop control was desired to ensure that the device was spinning at the
precise target speed, and to monitor the cyclic errors in the rotation rate. A means

69
of transmitting the necessary data between the rotating device and the stationary
instrumentation was required. Sensors are required to both close the control loop
around the rotation rate of the stand ω , and provide the data acquisition system the
actual angular position φ of the arm and the device deection ψ in real time for phase
measurements.
The spin stand conguration is shown in Figure 4-1 with key dimensions and
features labeled. The mechanical structure was designed using a factor of safety of
n = 2 against immediate failure in the event of a 10 lbm out of balance condition at
350 RPM. For the desired range of payload, an innite expected fatigue life of the
drive line was achieved by minimizing the bending moments in each of the shafts. The
drive line of the spin stand consists of a Leeson model N182D17FK1B 2 hp permanent
magnet DC servo rated for continuous operation at 1750 RPM, 9.5 A of current, and
a supply of up to 180 V DC. The motor is coupled to a short shaft that drives the
timing belt noted in the diagram. The 1 in wide H series timing belt is driven by a
16 tooth pulley, and a 60 tooth pulley is attached to the main shaft, giving a drive
ratio of 3.75. The arm of the device consists of a 6 ft × 4 in standard section 6061
aluminum I beam connected to the main shaft with a custom designed hub. The hub
allows the arm to freewheel on the shaft in the event of a drive line seizure without
allowing it to break free. The hub design also aids in the routing of the wiring from
the twelve channel Lebow 6105-12 slip ring, through a collar between the shaft and
the bearing, and out onto the rotating arm.
The power amplier used to supply current to the motor is the Advanced Motion
Controls model 30A20AC Brush Type PWM Servo Amplier. This amplier was
selected for its ability to draw power directly from the standard single phase 120V
supply. The amplier accepts an analog input command, and applies a proportional
current through the motor windings, which is in turn proportional to the torque
applied on the arm of the spin stand. Using the current sensing capabilities of the
amplier, the arm torque and steady state rotation rate were measured at various
command levels, and the coecient of drag of the bare beam was found to be CD =
1.981. This high aerodynamic drag limited the rotation rate to less than 300 RPM

70
Figure 4-1: Simplied assembly diagram of the spin stand showing the overall dimen-
sions, drive line conguration, and basic instrumentation.

71
 35
Beam Without fairing
Fit: Cd=1.981 Dm=2.266
30 Beam With fairing
Fit: Cd=0.136 Dm=2.337
Faired Beam With Prototype
25 Fit: Prototype Cd=0.992
Arm Torque (N m)

20

15

10

0
0 50 100 150 200 250 300 350 400 450
Rotation Rate ω (RPM)

Figure 4-2: Curves characterizing the aerodynamic drag on the arm of the spin stand
without fairing, with fairing, and with prototype attached. The coecient of drag
CD and constant mechanical drag DM values are given for each t curve.

at the maximum continuous arm torque of 30.5 N m, as shown in Figure 4-2. An

aerodynamic fairing was fabricated using a CNC wire foam cutter to produce a NACA
0012 prole large enough to encase the beam. The fairing reduced the coecient of
drag by an order of magnitude to CD = 0.136, and then the rotation rate was limited
by the rail voltage of the servo amplier at 440 RPM. With the unfaired prototype
attached to the faired beam, the spin stand is capable of operating at speeds up to
the desired 350 RPM with ample available torque for acceleration and control.
Two incremental digital optical encoders are used in the experimental setup to
measure the angles φ and ψ . The angle ψ is measured by a US Digital model E5S-
900-250-I encoder, that, with 900 full quadrature counts per revolution, can measure
the displacement of the pendulous link to a tenth of a degree. The encoder on the
main shaft of the spin stand is a US Digital model E6S-1800-1000-IH and has twice

72
Figure 4-3: Photo of spin stand arm with CNC cut NACA 0012 fairing for reduced
aerodynamic drag.

73
 Magnitude (dB) (deg/(N m))
50

−50

−100 Experimental
Modeled
−150
−1 0 1 2
10 10 10 10

45
0
Phase (deg)

−45
−90
−135
−180
−225
−1 0 1 2
10 10 10 10
Frequency (Hz)

Figure 4-4: Frequency response plot measured from torque command to main shaft
encoder measured displacement.

the resolution. Both encoders are interfaced directly to a dSPACE DS1103 real-
time controller development board that is used for both data acquisition and closed
loop control of the spin stand. Using code developed in the Precision Motion Control
Laboratory at MIT, the frequency response of the system was taken with the dSPACE
card and t to the model, as shown in Figure 4-4 [4]. Taking a loop shaping approach,
a controller was designed from the measured data that stabilizes the system with a
1 Hz crossover. The controller consists of a gain, a lead term centered at 1 Hz, and
a complex pair of poles at 3 Hz, with a damping ratio of ζ = 0.5. This gives the
controlled system a phase margin of over 40◦ , a gain margin of over 2 dB, and phase
stabilizes the resonant peak. The resonant peak is attenuated to -30 dB as measured.
The actual controller as implemented in the Simulink model that is compiled to the
dSPACE board can be seen in Figure C-4. The frequency response of the controlled
system is shown in Figure 4-5.

74
 Magnitude (dB) (deg/(N m))

50

−50

−100
−1 0 1 2
10 10 10 10

90
45
0
Phase (deg)

−45
−90
−135
−180
−225
−270
−315
−1 0 1 2
10 10 10 10
Frequency (Hz)

Figure 4-5: Frequency response plot of the open loop transfer function with the
controller applied, showing system stability.

75
 Implementing velocity control with digital incremental encoder feedback can be
dicult, due to the discontinuous position signal from the encoder. Using the change
in position between samples as a measure of the velocity becomes increasingly prob-
lematic as the velocity becomes low compared to the sampling rate. At high sample
rates the velocity appears to be zero while the encoder is between counts and spikes to
large values when each count is received. To eliminate this problem, position control
was implemented, and the system was made to follow a position trajectory in time
that results in the desired velocity. This introduces the problem of variable overow
as the position becomes large during continuous rotation. For this reason, the po-
sition error signal that goes into the controller is not generated from the dierence
between the absolute desired position and the actual position. The position error is
stored in an integrator that is updated each time sample with a delta position desired
and delta position achieved. Saturation is applied to this error storage integrator to
prevent excessive windup.
In order to protect the device from sudden changes in acceleration and to prevent
saturation of the amplier during transitions, a trajectory generator is implemented.
The smooth trajectories in velocity that are generated allow the system to grace-
fully transition from one operating speed to another. The trajectory generator was
implemented in a Stateow chart within the Simulink model and is compiled to the
dSPACE board with the controller and other model components. The Stateow chart
also automates the open loop indexing process that initializes the encoders. This is
accomplished by initially rotating the arm using only the open loop feed forward con-
troller that is based on the aerodynamic measurements. Once the encoders have both
been indexed, the closed loop controller is initialized to the current rotation rate, and
a smooth trajectory is followed to the desired operating speed. Sample output for
the the Stateow chart can be seen in Figure 4-6, and the chart itself is shown in
Figure C-3. The complete Simulink model that is compiled to the dSPACE board
includes other features such as subsystems that keep track of the true arm angle and
provide safeguards against exceeding the safe operating speeds. The full Simulink
model is presented in Appendix C.

76
 50
Set Point
Command (RPM)

40 Open Loop
30 Closed Loop

20

10

0
0 5 10 15 20 25 30 35

Drive

Index

Control

0 5 10 15 20 25 30 35
Time (sec)

Figure 4-6: Sample output of Stateow trajectory generator. Set Point is the desired
operating speed set by the user, Index is the state of the encoders. Open and Closed
loop command are outputs from the Stateow chart to the control system, Drive

enables the servo amplier, and Control enables the closed loop controller.

77
 (a) (b)
Figure 4-7: Photographs of the spin stand with the prototype device mounted for
testing.

78
4.2 On-board Variable Loads
Two dierent variable loads were constructed and used in the testing of the prototype
device. One load was designed to provide a purely resistive load for the purpose
of characterizing the device and measuring the maximum power output at various
rotation rates and damping levels. The second load was designed to demonstrate the
capability of the device to charge a battery through a high speed switching circuit.
Both loads were designed to be mounted on the beam within the fairing to avoid the
multiple connections and long stretches of wire that would be required to bring the
coil leads out through the slip ring to the stationary frame. An Atmel ATmega168
microcontroller was mounted on the beam and used to control each of the loads. The
microcontroller established two way RS232 communications through the slip ring with
the PC that operates the spin stand to display its status, and to receive commands
from the operator.

4.2.1 Resistive Load

The resistive load circuit shown in Figure 4-8 has a simple structure of resistors and
relays. The six resistors placed in series range from 4 Ω to 128 Ω increasing by
factors of two. The six independently controlled Hamlin HE721A0510 reed relays
used to short out the resistors are switched by a parallel communications port on
the microcontroller. This conguration allows the microcontroller to digitally select
a load resistance ranging from short circuit to 252 Ω with a 4 Ω resolution. An
additional relay is used to open or close the entire circuit, allowing the voltage of
the coil to be measured under open circuit conditions. To prevent centrifugal loading
from aecting the operation of the relays, the circuit is mounted near the hub of the
arm with the relays oriented such that their reeds point radially outward. The circuit
was found to operate without error under the maximum centrifugal loads encountered
during testing.

79
 80
Figure 4-8: Resistive load circuit diagram showing arrangement of resistors and reed relays.
4.2.2 Switching Load and Mock Battery
The switching load and mock battery circuit, shown in Figure 4-9, was designed to
demonstrate the capability of the device to charge batteries, and to investigate any
changes in the dynamics of the device caused by the nonlinear load. The circuit has
the ability to accurately measure the power output of the device that is delivered to
the load, but the circuit itself is externally powered, and circuit eciency was not a
consideration in the design. The basic structure of the circuit is to rectify the AC
voltage from the coil and to use it to charge a large storage capacitor. This large
capacitor acts as a low pass lter, maintaining a nearly constant voltage despite the
pulsating input current from the magnet assembly. The stored energy in the capacitor
is then supplied to a high speed switching DC-DC converter, that eciently brings the
voltage down to the levels appropriate for charging a battery. In the true application,
the battery would be used for long term storage and would supply power, at a stable
voltage, to the attached systems.
The power conditioning portion of the circuit uses Schottky diodes in a full bridge
conguration to charge the large storage capacitor. The DC-DC converter was de-
signed with an IRFBG20 power MOSFET as the switching element, and a smaller
2N7000 MOSFET to drive the gate of the power MOSFET. The smaller MOSFET is
driven by an adjustable duty cycle 25kHz PWM signal generated by the ATmega168.
An inverting buer is used to oer added protection to the microcontroller. This
simple switching circuit was convenient for the desired test and functioned well for
the known voltage range of the coil. In general, for high or unknown coil voltages, it
would not be sucient to drive the power MOSFET gate from a high voltage supply,
and commercially available bootstrapping MOSFET driver chips would oer a better
solution. After the switching stage, the standard DC-DC converter structure, using
an inductor, capacitor, and diode to ground, is used to step down the voltage and
step up the average current.
The use of an actual battery in the circuit is not convenient for laboratory testing,
because the state of charge of the battery would become an additional variable in the

81
system. A real battery would have varying resistance and voltage characteristics at
varying levels of charge and temperature, and would not be suitable for continuous
testing due to the dangers of overcharging. For these reasons, a mock battery circuit
was designed to approximate the electrical characteristics of a battery without the
extra variability. The circuit consists of a small sense resistor that approximates the
internal resistance of the battery and a 2.4 V zener diode that is biased with a current
from an external source. This conguration gives the circuit the characteristics of a
standard battery model of a voltage in series with a resistance, and also provides
direct measurement of the power delivered by the prototype device.
Two identical buer amplier circuits with a gain of 2 were used to measure the
voltage before and after the mock battery resistance, to provide amplication, and to
drive the long lines running through the slip ring assembly and back to the control PC
without inuencing the circuit. The rst buer amplier provides the voltage across
the mock battery circuit, and the dierence between the two measurements is used
to calculate the current passing through the circuit from the DC-DC converter. This
conguration allows the true power delivered to the mock battery to be calculated.

82
83
Figure 4-9: Switching load circuit diagram consisting of power conditioning circuitry, mock battery, and buer ampliers.
84
Chapter 5

Experimental Results and Discussion

Laboratory testing was conducted to measure the parameters of the prototype inertial
generator experimentally and to characterize the performance of the device during
operation. Static tests, conducted with the device xed to a stationary level surface,
included the measurement of the coil impedance as a function of frequency and mea-
surement of the electromagnetic coupling coecient as a function of displacement.
These measurements were used to complete the model of the device that was devel-
oped in Chapter 2. Spin testing employed the use of the spin stand to swing the
device through the circular trajectory. Spin testing was conducted with the resistive
load and the high speed switching load circuits that were presented in Chapter 4.
These spin testing experiments were conducted to verify the function of the device,
quantify its performance, and demonstrate its ability to perform the representative
task of battery charging.

5.1 Static Testing

5.1.1 Coil Impedance
In Chapter 2 the damping of the system was given in Equation (2.71) with the as-
sumption that the impedance of the coil could be considered purely resistive over
the frequency range of operation. To verify this assumption, the impedance of the

85
coil was measured as a function of frequency to establish the resistance of the coil
RC and to measure the inuence of the coil's inductance and parasitic capacitance
throughout the frequency range of operation. The same code that was used to mea-
sure the frequency response of the spin stand was used with the dSPACE system to
measure the impedance of the coil [4]. The sinusoidal excitation output by the DAC
was amplied with a simple op-amp buer amplier circuit and was used to drive the
coil and a sense resistor in series. The voltage across the coil was measured, and the
voltage across the sense resistor was used to calculate the current passing through
the coil, allowing the impedance of the coil to be found.
The impedance of the coil was measured with the coil installed within the fully
assembled magnet assembly, so that the impedance would be measured with the
proper core permeability. Measuring the impedance of the coil in this conguration
adds the complication that the current passing through the coil tends to excite the
magnet assembly, which introduces additional dynamics from the system. To isolate
the dynamics of the coil from the dynamics of the mechanical system, the measure-
ments were taken under two conditions. First, the magnet assembly was physically
constrained to minimize the relative motion between the coil and magnet assembly.
This procedure pushes the mechanical resonances of the system to relatively high
frequencies. A second set of measurements was taken with the magnet assembly un-
constrained, allowing the mechanical resonances to drop to lower frequencies. From
the overlay of the two sets of data, shown in Figure 5-1, the discrepancies between
the two curves can be attributed to the mechanical resonances of the system, and the
common features are the true coil characteristic. The results show that the coil has a
DC resistance of RC = 47.9 Ω, and the impedance can be considered purely resistive
for frequencies up to nearly 100 Hz. The desired speeds for physical testing fall at
much lower frequencies, so that the purely resistive model used to represent the coil
impedance is adequate.

86
 3
10
With Constraint
Without Constraint
Magnitude

2
10

1
10
0 1 2 3
10 10 10 10

100

50
Phase (deg)

−50

−100
0 1 2 3
10 10 10 10
Frequency (Hz)

Figure 5-1: Coil impedance as a function of frequency when installed within the mag-
net assembly. Measurements were taken with and without physical restraints applied
to magnet assembly to distinguish between actual coil impedance and mechanical
resonances.

87
5.1.2 Electromagnetic Coupling Coecient
The electromagnetic coupling coecient KEM was measured as a function of the
displacement ψ by using the relationship between the velocity ψ̇ and coil voltage V
expressed in Equation (2.68). The prototype device was placed on a at surface and
the magnet assembly was manually displaced and released. As the magnet assembly
swung back and forth through the stationary coil, the open circuit coil voltage and the
encoder position ψ were recorded by the dSPACE system. The data was processed
in MATLAB to determine the coupling coecient and position at each time sample,
and the data corresponding to each position was averaged over many cycles. The
results from this analysis are presented in Figure 5-2 and show the expected trend,
with the coupling reaching the highest values when the device is centered and the coil
is in the strongest portion of the magnetic eld. In this central region, the average
value of the coupling coecient was found to be

V V
KEM = 0.276 = 15.8 (5.1)
deg/sec rad/sec

This distribution of the coupling coecient is quite acceptable, considering that the
velocity ψ̇ is highest in this central region, making it the most critical for the extrac-
tion of energy through damping. The maximum damping that the coil can provide is
found from Equation (2.72) to be

 2
V
2
KEM 15.8 rad/sec Nm
bmax = = = 5.22 (5.2)
RC 47.9 Ω rad/sec

As discussed in previous chapters, improvements in this maximum available damping

lead to increased power transmitted to the load across the entire range of frequencies.
It has also been shown that for frequencies above ωB , given by Equation (2.77),
the power output is maximized when the optimal damping is applied, and for lower
frequencies it is best to match the impedance of the load to the impedance of the coil.

88
 0.04

0.035

0.03
K (V /(Deg/sec))

0.025

0.02

0.015

0.01

0.005

0
−10 −5 0 5 10
Postion Ψ (Deg)

Figure 5-2: Measured electromagnetic coupling coecient KEM as a function of swing

angle ψ .

89
5.2 Spin Testing
5.2.1 Spin Testing with Resistive Load
Spin testing with the variable resistive load was conducted over a range of rotation
rates and load resistances. The power output results of this testing are presented
in Figure 5-3, plotted versus rotation rate. The results show the expected trend of
increasing power output with increasing frequency. It should be noted that all of the
data points shown in the gure were taken in a single spinning test, and that the only
changes in the system were the rotation rate and the load resistance. The output
of the device contrasts sharply with that of the traditional sprung mass vibrational
energy harvesters, which must be physically modied for excitation at each frequency
to produce a smiler result. This monotonically increasing trend in the data and the
power levels observed show that the natural frequency tracking concept can be used
to achieve resonance over a broad range of frequencies.
Figure 5-4 presents the same power output data plotted versus load resistance,
where it can be more clearly seen that the power transmitted to the load increases
with increasing load resistance up to the point at which the load resistance is approx-
imately impedance matched with the coil resistance at 48 Ω. A further increase in
the load resistance does not tend to increase the transmitted power. This indicates
that the system output is being limited by the electromagnetic coupling coecient
KEM , and the mass of the pendulous link is not a limiting factor. At the highest fre-
quency, the sharp drop in power output, as the load resistance is lowered from 28 Ω
to 24 Ω, indicates that the damping of the system has increased beyond the optimal
damping given by Equation (2.66). This observation is supported by the swing mag-
nitude measurements shown in Figure 5-5. The amplitude loss that appears at load
resistances below 24 Ω at 300 RPM indicates that this rotational rate falls between
ωA and ωB on the normalized power curve presented in Figure 2-15. The fact that
the other rotation rates do not show this loss of amplitude as the load resistance is
lowered toward zero indicate that they fall below ωA .
Assuming sinusoidal motion of the linkage at a xed amplitude, it would be ex-

90
 60Ω
160
56Ω
52Ω
140 48Ω
44Ω
120 40Ω
36Ω
Power (mW)

100 32Ω
28Ω
80 24Ω
20Ω
60 16Ω
12Ω
40 8Ω
4Ω
20

0
100 150 200 250 300
Rotation Rate ω (RPM)

Figure 5-3: Steady state average power dissipated in the variable resistive load plotted
versus rotation rate for various load resistances.

91
 180
300RPM
275RPM
160
250RPM
225RPM
140 200RPM
175RPM
150RPM
120
Power (mW)

100

80

60

40

20
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
External Load Ω

Figure 5-4: Steady state average power dissipated in the variable resistive load plotted
versus load resistance for various rotation rates.

92
 25

20
Peak to Peak Motion ψ (Deg)

300RPM
15 275RPM
250RPM
225RPM
200RPM
10 175RPM
150RPM

0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
External Load Ω

Figure 5-5: Steady state peak to peak swing magnitude plotted versus load resistance
for various rotation rates.

93
 14
300RPM
275RPM
12 250RPM
225RPM
200RPM
10 175RPM
Peak to Peak Voltage (V)

150RPM

0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
External Load Ω

Figure 5-6: Steady state peak to peak load voltage plotted versus load resistance for
various rotation rates.

pected that the voltage output from the coil with a xed load would increase linearly
with frequency. The experimental data shown in Figure 5-6 directly contradict this
expectation. The data show a dependence on the load resistance, but almost no de-
pendence on the rotation rate. In order for this behavior to occur, it must be the case
that the pendulous magnet assembly is reaching nearly the same maximum velocity
regardless of the rotation rate. This phenomenon is caused by the nonlinearities in
the linkage and the collisions with the mechanical stops at the end of the allowed
travel, and is best illustrated by sample plots of the time domain data.
Figure 5-7 shows the displacement of the pendulum ψ plotted versus time for two
dierent operating speeds with the same 48 Ω resistive load. The upper plot shows
the device in operation at 150 RPM and the lower at 300 RPM. In the time domain,
it can be seen that the motion at high frequency is approximately sinusoidal, but
that the motion at low frequencies is clearly nonsinusoidal. At low frequency, the

94
system begins to behave much more like a falling mass, with collisions at the limits of
travel and delays between periods of motion. The interesting thing to notice is that
the deviation from sinusoidal motion at low frequencies actually improves the power
output of the device. At a constant damping, the total work done over a xed distance
is proportional to the velocity at which the distance is traversed. The average power
can be expressed as the product of the work done over the distance and the number
of time the distance is traversed per second. From this perspective, it becomes clear
that the high velocity motions between delays that appear in the low frequency data,
actually cause more energy to be transduced by the electromagnetic damping than
would purely sinusoidal motion.
The cosine of the arm angle φ is overlaid on the time domain swing angle plots
in Figure 5-7 for use as a phase reference between the motion of the device and the
excitation force provided by gravity. At high frequency, the gure shows that the
proper resonant phase relation exists, and the excitation of gravity is therefore doing
nearly the optimal amount of work of the device in each revolution. The low frequency
plot shows a surprisingly similar phase relationship despite the large deviation from
sinusoidal motion. This similar resonant phase relationship across a wide range of
frequencies further demonstrates the eectiveness of the frequency tracking properties
of the device dynamics.
The peak to peak voltage shown in Figure 5-6 showed little variation with changes
in frequency, and the time domain swing angle data gave some explanation of how
this occurred. Figure 5-8 shows the time domain voltage measured across the load,
which is proportional to the angular velocity ψ̇ , for sample data sets at 150 RPM
and 300 RPM. In addition to the similar peak to peak voltages, it can be seen that
the width of the peaks are also comparable. The time domain sample data of the
power transmitted to the load is presented in Figure 5-9, and shows that changes
in operating frequency tend to change the spacing between the individual spikes in
power more than the magnitude or width of the spikes themselves. This leads to the
rather linear trend in output power as a function of frequency that was observed in
Figure 5-3.

95
 15
Angle ψ
5Cos(φ)
Swing Angle ψ (Deg) 10

−5

−10

−15
0 0.5 1 1.5 2
Time (sec)

15
Angle ψ
5Cos(φ)
10
Swing Angle ψ (Deg)

−5

−10

−15
0 0.5 1 1.5 2
Time (sec)

Figure 5-7: Sample plots of the time domain swing angle measurements taken at
150 RPM (top) and 300 RPM (bottom) during operation with a 48 Ω resistive load.
The cosine of the arm angle φ is overlaid as a phase reference.

96
 6
Voltage
Cos(φ)
4

2
Coil Voltage (V)

−2

−4

−6
0 0.5 1 1.5 2
Time (sec)

6
Voltage
Cos(φ)
4

2
Coil Voltage (V)

−2

−4

−6
0 0.5 1 1.5 2
Time (sec)

Figure 5-8: Sample plots of the time domain coil voltage measurements taken at
150 RPM (top) and 300 RPM (bottom) during operation with a 48 Ω resistive load.
The cosine of the arm angle φ is overlaid as a phase reference.

97
 700 Power (mW)
100Cos(φ)
600

500
Power Output (mW)

400

300

200

100

−100
0 0.5 1 1.5 2
Time (sec)

700 Power (mW)

100Cos(φ)
600

500
Power Output (mW)

400

300

200

100

−100
0 0.5 1 1.5 2
Time (sec)

Figure 5-9: Sample plots of the time domain power output measurements taken at
150 RPM (top) and 300 RPM (bottom) during operation with a 48 Ω resistive load.
The cosine of the arm angle φ is overlaid as a phase reference.

98
5.2.2 Battery Charging Demonstration with Switching Load
Testing with the high speed switching load, that approximates the loading experienced
during battery charging, successfully demonstrated the ability of the device to charge
batteries. The power transmitted to the mock battery as a function of rotation rate
for each duty cycle is shown in Figure 5-10. One key success is that the device was
able to charge the 2.4 V mock battery at some rate at all frequencies tested despite
a relatively low maximum coil voltage. The ability to continue to charge the mock
battery at low rotation rates is a result of the relatively constant voltage pulsed
power characteristics that were observed at low frequencies with the resistive load.
The power transmitted to the mock battery is also nearly linearly increasing with
increasing frequency. The peak power transmitted to the mock battery is somewhat
lower than the power dissipated in the resistive load, due to the multiple diode drops
associated with the rectier and the switching circuitry. Choosing a smaller diameter
wire to increase the number of turns in the coil would boost the coil voltage and
reduce the losses that occur in the diodes.
In the experimental results, it was seen that the highest performance was achieved
at a duty cycle of 100%, and the switching portion of the circuitry was therefore
unnecessary. For the tested battery voltage of 2.4 V and the tested coil a simple
direct connection between the rectier and battery would be sucient; however, if a
high voltage coil were used to reduce the diode losses, a switching circuit would be
expected to be benecial.
Figure 5-11 and Figure 5-12 show the voltage across the mock battery and the
charging current, respectively. The trend of these two plots veries that the mock
battery has similar characteristics to a real battery. The voltage across the mock
battery changes relatively little with changes in current, and approaches a nominal
2.4 V as the charging current is reduced to zero. The charging current shown is within
the acceptable range to charge a AA two cell NiCad battery pack in a reasonable
amount of time.
The device did not show any signicant changes in its dynamics due to the addi-

99
 110

100

90

80 100%
90%
70 80%
70%
Power (mW)

60 60%
50%
50 40%
30%
40 20%
10%
30 5%
20

10

0
0 25 50 75 100 125 150 175 200 225 250 275 300 325 350
Rotation Rate ω (RPM)

Figure 5-10: Power transmitted to the mock battery portion of the switching load
circuit plotted versus frequency for various duty cycles.

100
 3

2.5

100%
90%
2
80%
70%
Voltage (V)

60%
1.5 50%
40%
30%
20%
1
10%
5%

0.5

0
0 25 50 75 100 125 150 175 200 225 250 275 300 325 350
Rotation Rate ω (RPM)

Figure 5-11: Voltage across the mock battery portion of the switching load circuit
plotted versus frequency for various duty cycles.

101
 40

35

30 100%
90%
25 80%
Current (mA)

70%
60%
20 50%
40%
30%
15 20%
10%
10 5%

0
0 25 50 75 100 125 150 175 200 225 250 275 300 325 350
Rotation Rate ω (RPM)

Figure 5-12: Charging current passed throughout the mock battery portion of the
switching load circuit plotted versus frequency for various duty cycles.

102
tional nonlinearities introduced by the use of the rectier and storage capacitor. The
time domain swing angle measurements taken while the switching load was employed,
shown in Figure 5-13, show no signicant dierence from those shown in Figure 5-7,
which were taken during testing with the purely resistive load.

103
 15
Angle ψ
5Cos(φ)
10
Swing Angle ψ (Deg)

−5

−10

−15
0 0.5 1 1.5
Time (sec)

15
Angle ψ
5Cos(φ)
10
Swing Angle ψ (Deg)

−5

−10

−15
0 0.5 1 1.5
Time (sec)

Figure 5-13: Sample plots of the time domain swing angle measurements taken at
150 RPM (top) and 300 RPM (bottom) while the system was loaded by the mock
battery charging circuit at 100% duty cycle. The cosine of the arm angle φ is overlaid
as a phase reference.

104
Chapter 6

Conclusion

An inertial generator for embedded operation within the propeller of a large ship
has been designed and successfully tested under the expected operating conditions.
The inertial generator was designed specically for the inertial excitation experienced
during rotation in the vertical plane at a radius that is large relative to the size of
the device. The design relies solely on the known inertial trajectory and the presence
of gravity to generate electrical energy. Within the reference frame attached to the
device, gravitational forces can be viewed as the nonconservative excitation. However,
in the inertial frame it is seen that the source of the energy produced by the device
is the work done on the propeller by the drive shaft of the ship, which provides the
reaction forces necessary to drive the device through its circular motion. The designed
system has the ability to reliably provide electrical energy to an embedded system
with only a single physical external connection.
The results of the laboratory testing have veried the expected dynamics of the
device by demonstrating that it can achieve large resonant deections at a wide range
of operating speeds. The experimental power output of the device shows the expected
monotonically increasing power output with increasing rotation rate, which has clear
advantages over the narrow useful operating range associated with traditional sprung
mass vibrational inertial generators. The nonlinear characteristics of the device have
been investigated, revealing that the device produces peak voltages at low rotation
rates that are comparable to the voltages produced at much higher rates. The rela-

105
tively constant peak voltage output eliminates the dropout that other designs exhibit
at low frequencies, when the voltage output falls below the level necessary to charge
the storage element. Laboratory testing has also demonstrated the ability of the de-
vice to charge a battery, with acceptable charging currents delivered across the entire
range of tested frequencies.
The tested prototype has proven its ability to perform at the target radius of one
meter and across the desired range of frequencies. The power output of the device
is limited by the coil assembly, and will increase linearly as the device is scaled in
thickness. As the ratio of the length scale of the device to its radius is raised, the
device becomes more stable, and less mass is required in the slider to tune the device.
As the ratio is lowered by reducing the length scale of the device or mounting it at
a larger radius, the device becomes more sensitive to alignment errors and requires
larger slider masses to tune the system.

6.1 Opportunities for Future Research

Many opportunities exist for further improvement of the currently prototyped de-
vice. In Section 2.3.2, it was shown that increases in the ratio of the square of the
electromagnetic coupling coecient to the resistance of the coil results in improved
eciency. Improvement opportunities in this area may include modications to the
geometry of the magnet assembly, or the development of new construction techniques
that do not require wire EDM machining of the magnets, which may have reduced
their strength. The development of magnetic stops, that limit the travel of the device
without the mechanical impacts that occur when rubber stops are employed, would
reduce both the noise produced by the device and the impulsive loads on the internal
mechanical components. Another interesting area for future research would be the
design of a self aligning device, allowing the device to operate at larger radii and
smaller length scales.
Opportunities for further development exist in optimizing the design for use with
a specic load. The research presented in this thesis has focused on characterizing

106
the devices dynamics and presenting its capabilities. Once a specic load is chosen
the material presented here may be used as a guide for producing a design that is
well matched to the selected load.
The slider augmented pendulum conguration was chosen in this work based on
its many attractive characteristics, especially its simplicity, and the lack of elements
that would be subject to wear or add mechanical drag to the system. In light of the
experimental results that show that the current prototype is limited by its electro-
magnetic coupling, the conguration consisting of a simple pendulum with a geared
ywheel may be a viable alternative. The system includes a wear item in the drive
line and is expected to have increased mechanical drag, but this may be oset by the
increase in electromagnetic coupling gained by integrating the ywheel with a brush-
less AC generator. The device would also have the advantage of a larger stable range
of motion, and would not experience the problems associated with misalignment an-
gles, while still maintaining perfect frequency tracking. Only careful prototyping and
laboratory testing can verify the long term benets of each design.

107
108
Appendix A

Slider Augmented Pendulum

Dynamics Derivations

As stated in Section 2.2.2, the equations of motion for the slider augmented pendulum
were derived with two assumptions. First, it is assumed that the slider mass mS
is constrained to move along the center line of the device as shown in Figure 2-
8 without friction. The second assumption is that the link connecting the slider
to the pendulous link has negligible mass and rotational inertia. With these two
assumptions, the nonlinear equations of motion for the slider augmented pendulum are
derived below using both Newton's Method and Lagrange's Method. The linearization
of the equations of motion for the free undamped response of the system is also
included below, and the results were utilized in the small angle analysis of the systems
natural frequency and stability explored in Equation Chapter 2. Diagrams of the
system and a description of variables are presented in Section 2.2.1.

A.1 Newton's Method

The motion of the center of gravity of the main pendulous link 1 has a velocity given
by

−−−−−→
 
VCG = VO/A + ω + ψ̇ × →
−
r CG/O

109
    
= rωı̂ + ω + ψ̇ L cos ψı̂ + ω + ψ̇ L sin ψ̂
       
= rω + L ω + ψ̇ cos ψ ı̂ + L ω + ψ̇ sin ψ ̂ (A.1)

where −
→
r CG/O is the location of the center of gravity of the link with respect to the
point O in the device reference frame ı̂, ̂, and VO/A is the velocity of the point O with
ˆ Jˆ. The acceleration is given by
respect to the point A in the inertial frame I,

−−−−−→ −−−−−→ 
ACG = aO/A + α × rCG/O + ω + ψ̇ × ω + ψ̇ × −
→
  
r CG/O
 2
= rω 2 ̂ + ψ̈L (cos ψı̂ + sin ψ̂) + ω + ψ̇ L (cos ψ̂ + sin ψı̂)
  2 
= ψ̈L cos ψ − ω + ψ̇ L sin ψ ı̂
h   i
+ rω 2 + ψ̈L sin ψ + ω + ψ̇ L cos ψ ̂ (A.2)

The component of the acceleration of the center of gravity in the ı̂0 direction is

ACG ı̂0 = ψ̈L cos2 ψ − L (ω + ψ)2 sin ψ cos ψ + rω 2 sin ψ

 2
+ ψ̈L sin2 ψ + L ω + ψ̇ cos ψ sin ψ

= ψ̈L + rω 2 sin ψ (A.3)

Taking a torque balance about point O, and utilizing the result derived in Sec-
tion 2.1.2.1, we have

X
ψ̈I = TO
 
= −mL ψ̈L + rω 2 sin ψ − Lmg sin (φ + ψ)

+ FS L1 sin (φ − ψ) + FC LC (A.4)

The radius of the slider mass is given by

rS = − (r + x) ̂

110
and its velocity and acceleration are

vS = ω (r + x) ı̂ + ẋ̂

and
 
aS = (2ω ẋ) ı̂ + −ẍ + ω 2 (r + x) ̂ (A.5)

The force balance on the slider in the ̂ direction can be written as

(A.6)
X
F ̂ = ms as ̂

or
 
FS cos θ − mS g cos θ = mS −ẍ + ω 2 (r + x) (A.7)

The slider position x is dictated by the swing angle ψ , allowing the slider position to
be written as
q
x = L1 cos ψ − L22 − (L1 sin ψ)2 (A.8)

and the velocity as

 
L1 sin ψL1 cos ψ ψ̇ L1 cos ψ
ẋ = −L1 ψ̇ sin ψ + q = L1 ψ̇ sin ψ  q − 1 (A.9)
2 2 2 2
L2 − (L1 sin ψ) L2 − (L1 sin ψ)

Dening the quantities

q
Z≡ L22 − (L1 sin ψ)2 (A.10)

and !
L1 cos ψ
Y ≡ L1 sin ψ −1 (A.11)
Z
allows Equation (A.8) and Equation (A.9) to be written in the more compact form

x = L1 cos ψ − Z (A.12)

and
ẋ = ψ̇Y (A.13)

111
Dierentiation of the expression for the velocity in Equation (A.13) yields the accel-
eration of the slider as

L2 sin2 ψ L21 cos2 ψ

" ! !#
L1 cos ψ
ẍ = ψ̈Y + ψ̇ 2 L1 cos ψ −1 + 1 −1 (A.14)
Z Z Z2

Again, dening a quantity

L21 sin2 ψ L21 cos2 ψ

" ! !#
L1 cos ψ
A ≡ L1 cos ψ −1 + −1 (A.15)
Z Z Z2

allows the more compact representation of Equation (A.14) as

ẍ = ψ̈Y + ψ̇ 2 A (A.16)

Now substituting Equation (A.12), Equation (A.13), and Equation (A.16) into
Equation (A.7) gives the relation

 
FS cos θ = mS g cos φ − mS ψ̈Y + ψ̇ 2 A + mS ω 2 (r + L1 cos ψ − Z) (A.17)

From the geometry of the linkage, it can be shown that

q
L22 − (L1 sin ψ)2 Z
cos θ = = (A.18)
L2 L2

and
L1 sin ψ
sin θ = (A.19)
L2
Using these relations, equation Equation (A.17) and the last term in Equation (A.4)
can be written in terms of ψ alone as

Z  
FS = mS g cos φ − ψ̈Y − ψ̇ 2 A + ω 2 (r + L1 cos ψ − Z) (A.20)
L2

112
and

FS L1 sin (θ − ψ) = FS L1 (sin θ cos ψ − cos θ sin ψ)

L1 Z
 
= FS L1 sin ψ cos ψ − sin ψ
L2 L2
Z L1
 
= FS L1 sin ψ cos ψ − sin ψ
L2 Z
!!
Z L1 cos ψ
= FS L1 sin ψ −1
L2 Z
Z
= FS Y (A.21)
L2

It is now convenient to combine Equation (A.4), Equation (A.21), and Equa-

tion (A.20) to generate the full equation of motion for the system as

 
ψ̈I = −mL ψ̈L + rω 2 sin ψ − Lmg sin (φ + ψ) + FC LC
 
+ Y mS g cos φ − ψ̈Y − ψ̇ 2 A + ω 2 (r + L1 cos ψ − Z)

which simplies to

−mL (rω 2 sin ψ + g sin (φ + ψ)) + mS Y ω 2 (r + L1 cos ψ − Z)

ψ̈ =
I + mS Y 2
 O
mS Y −ψ̇ 2 A + g cos φ + LC FC
+ (A.22)
IO + mS Y 2

where IO is the rotational inertia of link L1 about point O.

A.2 Lagrange's Method

The position and velocity of the center of mass of the pendulous link L1 is given by

RCG = [r sin φ + L sin (φ + ψ)] Iˆ + [−r cos φ − L cos (φ + ψ)] Jˆ (A.23)

113
and

h   i h   i
VCG = rω cos φ + L ω + ψ̇ cos (φ + ψ) Iˆ + rω sin φ + L ω + ψ̇ sin (φ + ψ) Jˆ
(A.24)
The square of the velocity of the center of gravity of the link can be written as

   2
2
VCG = r2 ω 2 cos2 φ + 2rLω ω + ψ̇ cos φ cos (φ + ψ) + L2 ω + ψ̇ cos2 (φ + ψ)
   2
+r2 ω 2 sin2 φ + 2rLω ω + ψ̇ sin φ sin (φ + ψ) + L2 ω + ψ̇ sin2 (φ + ψ)
 2  
= r2 ω 2 + L2 ω + ψ̇ + 2rLω ω + ψ̇ (cos φ cos (φ + ψ) + sin φ sin (φ + ψ))
 2  
= r2 ω 2 + L2 ω + ψ̇ + 2rLω ω + ψ̇ cos ψ (A.25)

It is convenient to dene the quantities Z , Y , and A as was done in the Newtonian

approach, where
q
Z≡ L22 − (L1 sin ψ)2 (A.26)
!
L1 cos ψ
Y ≡ L1 sin ψ −1 (A.27)
Z
and
L2 sin2 ψ L21 cos2 ψ
" ! !#
L1 cos ψ
A ≡ L1 cos ψ −1 + 1 −1 (A.28)
Z Z Z2
The motion of the slider can be dened in terms of these quantities as was done in
Equation (A.12), Equation (A.13), and Equation (A.16) where

x = L1 cos ψ − Z (A.29)

ẋ = ψ̇Y (A.30)

and
ẍ = ψ̈Y + ψ̇ 2 A (A.31)

114
Several partial derivatives of the slider motion will be necessary in the formulation of
Lagrange's equation, and are found as follows:

L2 sin ψ cos ψ
!
∂x L1 cos ψ
= −L1 sin ψ + q 1 = L1 sin ψ −1 =Y (A.32)
∂ψ L22 − (L1 sin ψ)2 Z

∂ ẋ
=Y (A.33)
∂ ψ̇
and ! !
dY d ∂ ẋ ∂ dẋ ∂ ẍ ∂  
= = = = ψ̈Y + ψ̇ 2 A = ψ̇A (A.34)
dt dt ∂ ψ̇ ∂ ψ̇ dt ∂ ψ̇ ∂ ψ̇
The dynamics of the system can be found from Lagrange's equation, which is
given by !
d ∂T ∂T ∂V
− + = LC FC (A.35)
dt ∂ ψ̇ ∂ψ ∂ψ
where FC is the non-conservative force exerted by the coil on the magnet assembly
and LC is the distance at which the force is applied. The kinetic energy in the system
is

1  2 1 2 1
T = I ω + ψ̇ + mVCG + mS Vm2 S
2 2  2
1  1
2  2   
= I ω + ψ̇ + m r2 ω 2 + L2 ω + ψ̇ + 2rLω ω + ψ̇ cos ψ
2 2
1  
+ mS ω 2 (r + x)2 + ẋ2 (A.36)
2

The potential energy is given by

V = mg (−r cos φ − L cos (φ + ψ)) − mS g cos φ (r + x) (A.37)

Taking the appropriate partial derivatives of the kinetic and potential energies:

∂V ∂x
= mgL sin (φ + ψ) − mS g cos φ = mgL sin (φ + ψ) − mS g cos φY (A.38)
∂ψ ∂ψ

115
 ! !
d ∂T d     ∂ ẋ
= I ω + ψ̇ + mL2 ω + ψ̇ + mrLω cos ψ + mS ẋ
dt ∂ ψ̇ dt ∂ ψ̇
dY
= I ψ̈ + mL2 ψ̈ − mrLω sin ψ ψ̇ + mS ẍY + mS ẋ
   dt 
= I + mL ψ̈ − mrLω ψ̇ sin ψ + mS Y ψ̈Y + ψ̇ 2 A + mS ψ̇Y ψ̇A
2

 
= ψ̈ IO + mS Y 2 − mrLω ψ̇ sin ψ + 2mS ψ̇ 2 AY (A.39)

and

∂T   ∂ ẋ ∂x
= −mrLω ω + ψ̇ sin ψ + mS ẋ + mS ω 2 (r + x)
∂ψ ∂ψ ∂ψ
2 2
= −mrLω sin ψ − mrLω ψ̇ sin ψ + mS ψ̇ AY

+ mS ω 2 Y (r + L1 cos ψ − Z) (A.40)

Now assembling the partial derivatives into Equation (A.35), yields

   
LC FC = ψ̈ IO + mS Y 2 − mrLω ψ̇ sin ψ + 2mS ψ̇ 2 AY

+ (mgL sin (φ + ψ) − mS g cos φY )

+ mrLω 2 sin ψ + mrLω ψ̇ sin ψ − mS ψ̇ 2 AY

− mS ω 2 Y (r + L1 cos ψ − Z) (A.41)

Rearranging terms and simplifying yields

 
ψ̈ IO + mS Y 2 = −mS ψ̇ 2 AY − mgL sin (φ + ψ) + mS g cos φY
 
+ −mrLω 2 sin ψ + mS ω 2 Y (r + L1 cos ψ − Z)

+ LC FC (A.42)

and nally

−mL (rω 2 sin ψ + g sin (φ + ψ)) + mS Y ω 2 (r + L1 cos ψ − Z)

ψ̈ =
I + mS Y 2
 O
mS Y −ψ̇ 2 A + g cos φ + LC FC
+ (A.43)
IO + mS Y 2

116
A.3 Linearization about ψ = 0, ψ̇ = 0
The linearized equations of motion of the systems free undamped response were used
in Chapter 2 to investigate the natural frequency and stability of the system for
small angles of deection. The formal linearization of the equations of motion for
that purpose are shown in detail in this section. The free undamped response of the
system is obtained by setting g = 0 and FC = 0 in Equation (A.22), and the nonlinear
free undamped equation of motion is obtained as

−mLrω 2 sin ψ + mS Y ω 2 (r + L1 cos ψ − Z) − mS Y ψ̇ 2 A

ψ̈ = (A.44)
IO + mS Y 2

Linearizing the equation about the centered position C where ψ = 0 and ψ̇ = 0 is

performed with the approximation
 
 ∂ ψ̈  ∂ ψ̈ 
ψ̈ ≈ ψ̈  +  ψ̇ +  ψ (A.45)

C ∂ ψ̇ C ∂ψ C

Noting that
Y |C = 0 (A.46)

it follows that the rst term is


ψ̈  = 0 (A.47)

C

The partial derivative in the second term can be written as

∂ ψ̈ −2mS Y ψ̇A
= (A.48)
∂ ψ̇ IO + mS Y 2

and evaluation at the centered position yields


∂ ψ̈ 
 =0 (A.49)
∂ ψ̇ C

117
The remaining partial derivative is given by
 
∂ 1
∂ ψ̈ 
IO +mS Y 2
= −mLrω 2 sin ψ + mS Y ω 2 (r + L1 cos ψ − Z) − mS Y ψ̇ 2 A
∂ψ ∂ψ
2
" #
1 ∂ (−mLrω sin ψ) ∂Y  2

+ + mS ω (r + L1 cos ψ − Z)
IO + mS Y 2 ∂ψ ∂ψ
" #
1 2 ∂ (r + L1 cos ψ − Z) 2 ∂Y A
 
+ mS Y ω − ψ̇ (A.50)
IO + mS Y 2 ∂ψ ∂ψ

and evaluation at C results in

 
1
∂
  #
∂ (−mLrω 2 sin ψ) 
"
∂ ψ̈  IO +mS Y 2 1
 = (0) + 
∂ψ C ∂ψ IO + 0 ∂ψ 
C

1 ∂Y   
+  mS ω 2 (r + L1 − Z|C )
IO + 0 ∂ψ C
1 ∂ (r + L1 cos ψ − Z)
+ (0)
IO + 0 ∂ψ
 
2
" #
1 ∂ (−mLrω sin ψ)  ∂Y   
=  +  mS ω (r + L1 − Z|C ) (A.51)
2
IO ∂ψ 
C
∂ψ C

The rst partial derivative in Equation (A.51) can be taken and evaluated as

∂ (−mLrω 2 sin ψ)   
 = −mLrω 2 cos ψ  = −mLrω 2 (A.52)

∂ψ 
C
C

The remaining partial derivative in Equation (A.51) is

 " ! !#
∂Y  L1 cos ψ ∂ L1 cos ψ 
 = L1 cos ψ − 1 + L1 sin ψ −1 

∂ψ C Z ∂ψ Z 
C
!
L1 ∂ L1
 
= L1 −1 +0 −1
Z|C ∂ψ Z
L1
 
= L1 −1 (A.53)
L2

Substituting Equation (A.52) and Equation (A.53) into Equation (A.51) gives the
expression

∂ ψ̈  1 L1
     
 = −mLrω 2 + L1 −1 2
mS ω (r + L1 − L2 )
∂ψ C  IO L2

118
 h  i
L1
ω 2 −mLr + mS L1 (r + L1 − L2 ) L2
−1
= (A.54)
IO

Substituting Equation (A.47), Equation (A.49), and Equation (A.54) into Equa-
tion (A.45) gives the linearized equation of the free undamped motion about the
point C , where ψ = 0, ψ̇ = 0, as
h  i
L1
ω 2 −mLr + mS L1 (r + L1 − L2 ) L2
−1
ψ̈ ≈ ψ (A.55)
IO

119
120
Appendix B

Prototype Drafts

Contained within this appendix are the assembly and part drafts showing the details
of the components of the prototype device. All dimensions are in inches. Purchased
parts are are listed in Table B.1 and are not presented in individual drafts.

121
 122
Figure B-1: Prototype Assembly, Sheet 1 of 5
123
Figure B-2: Prototype Assembly, Sheet 2 of 5
 124
Figure B-3: Prototype Assembly, Sheet 3 of 5
125
Figure B-4: Prototype Assembly, Sheet 4 of 5
 126
Figure B-5: Prototype Assembly, Sheet 5 of 5
127
Figure B-6: Core Assembly
 128
Figure B-7: Magnet Stack Assembly
129
Figure B-8: Slider Mass Assembly
 130
Figure B-9: Spine
131
Figure B-10: Bearing Block
 132
Figure B-11: Front Upper Rail
133
Figure B-12: Front Lower Rail
 134
Figure B-13: Core - Side Plate
135
Figure B-14: Core - End Plate
 136
Figure B-15: Core - Shoulder Screw Support
137
Figure B-16: Core - Magnet Support
 138
Figure B-17: Slider Mass - Side
139
Figure B-18: Slider Mass - Upper Bridge
 140
Figure B-19: Slider Mass - Lower Bridge
141
Figure B-20: Slider Mass - Adjustment Plate - 0.5 in
 142
Figure B-21: Slider Mass - Adjustment Plate - 1.0 in
143
Figure B-22: Slider Mass - Adjustment Plate - 2.0 in
 144
Figure B-23: Slider Mass - Adjustment Plate - 2.5 in
145
Figure B-24: Mounting Plate
 146
Figure B-25: Slider Mass - Shaft
147
Figure B-26: Base Plate
 148
Figure B-27: Back Wall
149
Figure B-28: Top Plate
 150
Figure B-29: End Plate
151
Figure B-30: End Brace
 152
Figure B-31: Link 2
153
Figure B-32: Slider Mass - Constraint Link
 154
Figure B-33: Slider Mass - Lower Rail
155
Figure B-34: Center Spacer
 156
Figure B-35: Lexan Front
157
Figure B-36: Lexan Top
 158
Figure B-37: Lexan End
159
Figure B-38: Core - Shaft
 160
Figure B-39: Coil Support Block
161
Figure B-40: Coil Support Flange
 162
Figure B-41: Magnet Assembly Prior to Cutting
163
Figure B-42: Magnet Assembly - Magnet
 164
Figure B-43: Magnet Assembly Center Iron
165
Figure B-44: Coil Dimensions
 166
Figure B-45: Coil Winding Fixture Assembly
167
Figure B-46: Coil Winding Fixture - Center
 168
Figure B-47: Coil Winding Fixture - Side
Table B.1: Purchased parts used in prototype assembly.

169
170
Appendix C

Spin Stand Controls

The following gures show the details of the Control Desk Layout and the Simulink
model that were used to automate the spin stand.

171
Figure C-1: Spin stand control desk layout

172
Figure C-2: Simulink model compiled to the DS1103 dSPACE board for spin stand
control.

173
 174
Figure C-3: Stateow chart used for open loop indexing of the encoders and smooth trajectory generation.
Figure C-4: Controller subsystem. Discrete time control blocks used for closed loop
feedback control of arm position.

Figure C-5: Feed forward subsystem. Generates expected torque necessary to over-
come wind drag and mechanical friction.

Figure C-6: Scaling subsystem. Converts from torque desired to voltage command
from gear ratio, motor torque constant, and amplier gain. Includes saturation to
protect motor and amp in the event of any failure elsewhere in the diagram.

175
Figure C-7: Rev-limiter subsystem. Generates a signal to cut power to the amplier
if arm speed exceeds the forward or backward limits.

Figure C-8: Wrap subsystem. Keeps track of arm position as 0◦ ≤ φ ≤ 360◦ without
overowing during extended continuous operation.

176
Bibliography

[1] J. Chambers. Personal communication. Documentation and electronic media,

Nov 2004.

[2] A. Kasyap, J. Lim, D. Johnson, S. Horowitz, T. Nishida, K. Ngo, M. Sheplak,

and L. Cattafesta. Energy reclamation from a vibrating piezoceramic composite
beam. In Proceedings of 9th International Congress on Sound and Vibration ,
2002. Orlando, FL, Paper No. 271.

[3] W. Kim and H. Maheshwari. High-precision control of a maglev linear actua-

tor with nanopositioning capability. Proceedings of the 2002 American Control

Conference , 5:42794284, May 2002.

[4] K. Lilienkamp and D. Trumper. Dynamic signal analyzer for dSPACE. In Pro-

ceedings of the dSPACE User's Conference , May 2000.

[5] P. Mitcheson, T. Green, E. Yeatman, and A. Holmes. Analysis of optimized

micro-generator architectures for self-powered ubiquitous computers. In Adjunct
Proceedings UBICOMP 2002, Ubiquitous Computing 4th International Confer-

ence , pages 56, Oct 2002.

[6] D. Opila, A. Annaswamy, W. Krol, and S. Raghu. Biomimetic reduction of wake

decit using tail articulation at low Reynolds number. IEEE Transactions on

Ocean Engineering , 29(3):766776, July 2004.

177
 [7] G. Ottman, H. Hofmann, and G. Lesieutre. Optimized piezoelectric energy
harvesting circuit using step-down converter in discontinuous conduction mode.
IEEE Transactions on Power Electronics , 18(2):696703, Mar 2003.

[8] S. Roundy. On the eectiveness of vibration-based energy harvesting. Journal

of Intelligent Material Systems and Structures , 16(10):809823, 2005.

[9] S. Roundy, E. Leland, J. Baker, E. Carleton, E. Reilly, E. Lai, B. Otis, J. Rabaey,

V. Sundararajan, and P. Wright. Improving power output for vibration-based
energy scavengers. IEEE Pervasive Computing , 04(1):2836, 2005.

[10] S. Roundy, P. Wright, and J. Rabaey. A study of low level vibrations as a power
source for wireless sensornodes. Computer Communications , 26(11):11311144,
July 2003.

[11] S. Roundy and Y. Zhang. Toward self-tuning adaptive vibration-based micro-

generators. In Proceedings of the SPIE, Smart Structures, Devices, and Systems

II, volume 5649, pages 373384, Feb 2005.

[12] R. Torah, S. Beeby, M. Tudor, T. O'Donnell, and S. Roy. Development of a

cantilever beam generator employing vibration energy harvesting. In Proceedings
of The 6th International Workshop on Micro and Nanotechnology for Power

Generation and Energy Conversion Applications , pages 181184, 2006.

[13] M. Umeda, K. Nakamura, and S. Ueha. Analysis of the transformation of me-

chanical impact energy to electric energy using piezoelectric vibrator. Japanese

Journal of Applied Physics , 35:32673273, May 1996.

[14] S. Verma, Won jong Kim, and H. Shakir. Multi-axis maglev nanopositioner for
precision manufacturing and manipulation applications. IEEE Transactions on

Industry Applications , 41(5):11591167, Sept-Oct 2005.

[15] T. von Büren, P. Lukowicz, and G. Tröster. Kinetic energy powered comput-
ing - An experimental feasibility study. In Proceedings of the Seventh IEEE

International Symposium on Wearable Computers , pages 2224, 2003.

178
[16] C. Williams, C. Shearwood, M. Harradine, P. Mellor, T. Birch, and R. Yates.
Development of an electromagnetic microgenerator. Electronics ,
Letters

33(22):18831884, Oct 1997.

[17] C. Williams and R. Yates. Analysis of a micro-electric generator for microsys-

tems. The 8th International Conference on Solid-State Sensors and Actuators ,
1:369372, Jun 1995.

179