Professional Documents
Culture Documents
Vibration Based Electrical Energy Harvesting
Vibration Based Electrical Energy Harvesting
Nonlinear
onlinear Piezoelectric Generators
Relatore Coordinatore
A.A. 2006/2007
ii
…To My Family
iii
Acknowledge
Acknowledgements
considerably at every step of this journey. I’ve appreciated, in particular, his great
thank Dr. Helios Vocca for the assistance, useful discussions he supplied at all levels
of the research project and most of all for his unavoidable irony (specially about the
religious issues). A very special thanks goes out to Dr. Paolo Amico for his skill and
the fruitful conversations and to Dr. Flavio Travasso for his real sympathy, lifestyle
and the many hours spent together in the “cold” laboratory! After the beast, I would
like to thank the beauty of our team who is Dr. Anna Dari for his exigent questions
about all human knowledge! She has been a constant source of inspiration and
incitement for me. Thanks to Chiara Molinelli for his continuous encouragements
and friendship. I would like to thank Dr. Ludovico Carbone for being a great advisor.
Appreciation also goes out to Dr. Leone Bosi and Dr. Igor Neri for the kindness and
all of their computer and technical assistance throughout my doctorate work. I would
like to thank Dr. Michele Punturo who made me know the research team and the
VIRGO project. Again, I have been honored to receive some useful suggestions by
4
My special thanks to my university friends: Alessio, Filippo, Marco, Ruggero for
the great time I had with them and to all my colleagues. I enjoyed their friendship
and their support. I wish to thanks all my nearest friends Carlo, Ciro, Matteo, Daniele
together with all others which have been near me for moral support.
Finally, I doubt that I would not ever have been able to make such adventure
without the fundamental support and affection of my family. I owe them my eternal
gratitude. I also wish to thanks my fiancée, Martina, for her love and encouragement
5
Abstract
Ambient energy harvesting has been in recent years the recurring object of a
powering of small scale electronic mobile devices. Among the different solutions,
vibration energy harvesting has played a major role due to the almost universal
ambient sound down to thermal noise induced fluctuations. Standard approaches are
mainly based on resonant linear oscillators that are acted on by ambient vibrations. In
transducers present severe limitations like narrow bandwidth, need for continuous
frequency tuning, high resonant frequency at MEMS dimensions and low efficiency,
employed for scavenging energy from vibrational noise. Such a concept is shown to
outperform standard linear oscillators and to overcome some of the most of present
approaches. Experimental tests have been carried out on a simple physical model
6
field and are in excellent agreement with all numerical expectations. We demonstrate
behaviour. In effect, the bistable system shows the ability to adsorb vibrational
energy from a wide bandwidth, mostly at lower frequencies. Moreover, this method
is not only restricted to bistable systems but even better to other kind of nonlinear
systems that should be investigated. We prove that the method proposed here is quite
general in principle and could be applied to a wide class of nonlinear oscillators and
different energy conversion principles. Finally, there are also potentials for realizing
work.
7
Table of Contents
8
Chapter 1
Introduction
devices and today micro and nano-scale machines are yet incorporated in all kinds of
electronics has led to large research efforts in size reduction as power consumption
actuators. These micro devices have wide applications that cover from military
9
Chapter 1. Introduction
randomly distributed in the area of interest and they may make use of existing
communication interface and power supply. More recently those devices have been
transports.
source
Streaming Data to/from
the Physical World
destination
Figure 1.1
Multihop Wireless Sensor etwork
10
Chapter 1. Introduction
Wireless autonomous sensors and actuators with sizes under a centimeter and
below already surround us in an almost invisible way and in a forthcoming future for
example intelligent clothing and body-area networks could monitor our health
parameters. The data that's collected by miniaturized pressure sensors built into
buildings, roads, bridges and railways will be used by construction engineers. It's
almost impossible to find areas of our civilization that will not be affected.
increasing density and shrinking the systems, however, arise the problem of scaling
in the same time the on-board power supply [3]. Research continues to develop ultra-
low power circuit [4] and higher energy-density batteries but the amount of energy
available is not infinite and limits the system's life. Extended life is critical in many
b
a
c
Figure 1.2
a) Intel Mote b) Implantable sensor
c) Smart Dust d) Intel chip
11
Chapter 1. Introduction
For a series of motives, these devices cannot be easily powered by ordinary batteries:
It’s need to develop alternative methods of power these wireless microdevices that
must be economical, efficient and ecological. There are different ways to address the
problem:
• develop self-powering nodes that harvest and convert the energy directly
from the ambient [6]
In the last years many efforts have been done to scale down power energy
devices. Fuel cells technologies, micro heat engine or micro-nuclear batteries, for
example, promise energy densities several time higher than chemical batteries and
are capable of far higher maximum power output [7]. Electronic Radio Frequency
Identity tags, smart cards and other many passive electronic devices are yet powered
by a close energy transmitted to them to perform their operations [8]. However, this
12
Chapter 1. Introduction
method is not a good solution when considering dense networks of wireless nodes. In
fact, it is not suitable for distances beyond 5-10 meter where high power transmitter
is required with a consequent efficiency loss. In that case, this technology would
probably present a health risk and may exceeds local or international regulations of
The best solution to avoid battery replacement is that each node must be
continuously from the ambient [9] (solar, vibrations, electromagnetic, thermal), but
this method is that less explored as fully as the powering by storage systems.
For such reasons a large research effort has been devoted in designing on-board
power generators that could supply the necessary amount of energy when and where
kinetic energy available through random vibration is probably the most common
form. Random vibrations come in a vast variety of forms, amplitude, spectral shapes
oscillating mechanical elements that convert kinetic energy into electric energy via
tuned to the ambient dominant mechanical frequency. However, in the vast majority
of cases the ambient vibrations come with their energy distributed over a wide
13
Chapter 1. Introduction
components. This is the case for example of the omnipresent seismic vibrations. In
order to take advantage of such energy spectral distribution it is necessary to tune the
constraints that the dimensions of the device pose, the efficiency of such mechanical
demonstrate that a bistable oscillator, under proper operating conditions, can provide
powering scavenging methods, useful for low-consumption devices at the micro and
sub-micron scale.
with a special focus on wireless sensor node but in perspective open to all MEMS
and NEMS (Nano Electro-Mechanical Systems) world powering issues. In the next
paragraphs of this chapter will be shown a survey of state of the art techniques and
power sources with a focus on vibration noise. In chapter two will be discussed the
14
Chapter 1. Introduction
and will be showed the idea to take advantage from nonlinear dynamics of a bistable
test of some bistable piezoelectric systems will be presented in chapter three. Finally
The Moore’s law states that the transistors doubling every couple of years and the
Bell’s law that a new computing class born every ten years. Although electronics
parallel the need for energy scale-down became a serious challenge. Semiconductor
but this savings is being counteracted by a higher structural density of transistors and
higher power leakage caused by quantum effects (fig. 1.3). In order to decrease the
been push down as much as possible the supply voltage. In the next figures it’s
15
Chapter 1. Introduction
For the mini scale down to nano level the development of ultra-low energy
consumption electronic devices constitutes a great challenge as for the macro scale as
well. A PDA (Personal Digital Assistant) or a cellular with a battery capacity of 1500
mAh is significantly more energy efficient than a PC using about 200-250 W and
16
Chapter 1. Introduction
classified as a function of the overall average linear size but the functionalities and
Since the main subject of this work regards the physics of energy harvesting
systems that is crucial for low-scale powering, we now shift our attention to the
micro and nano-world area. The main important features for sub-centimeter or sub-
is not always critical. The main goals for the wireless sensor nodes are that they must
be smaller than one cubic centimeter, weigh less than 100 grams, and cost
substantially less than one dollar. Even more important, the electronic components
of the node must use ultra-low power to extend the battery life and to avoid frequent
replacement. However, The huge success in reducing the size of MEMS, in effects, is
17
Chapter 1. Introduction
materials are improving the battery technology, its energy density doesn’t follow the
exponential curve as the Moore’s law for miniaturization process and performance.
device many WSN (Wireless Sensor Nodes) researchers have provided new
specifications. The most important specifications for the power supply system are the
total size and average power dissipation of an individual node (i.e. for a PicoNode
size of a node must be overall less than 1cm3 and the target average power
particularly difficult, and it is likely that several technology efforts will be necessary
to achieve this goal but this is the upper limit that will address the energy harvesting
design. This does not mean that power system solutions which don’t meet this
feature are not worthy of further exploration, but simply that this constrain will
constitute a desirable limit or standard for the most projects of wireless sensor nodes.
µm the
Going into more tiny dimensions, in the domain of EMS from 10nm to 3µ
there are even more problems about energy issue. Energetics represent a serious
in the next future metabolize in vivo local glucose and oxygen for energy, or power
18
Chapter 1. Introduction
Freitas [13] makes an estimation using as a first crude approximation the power law
density. Supposing P=100 Watts for an m= 70 kg human body mass and assuming
water density for nanorobots, then P = 23pW for a 1 µm3 nanorobot, therefore a
oxyglucose engine it’s need to consider the fundamental limits on power density that
µm 3.
chemical power density of d~109 watts/m3 or 0.1nW/µ
The current state of research on vastly different potential power sources for micro
and nano systems is not so simple to discuss. Furthermore there are many good
works about this topic[10, 14, 15] and it’s not the aim of this thesis to do an ulterior
deep inspection. Here we want only to make a survey and a comparison among
storage systems and renewable sources. Power sources are distinguished as energy
power source acceptable when it’s capable of providing power density on the order
19
Chapter 1. Introduction
µW/cm3 for at least ten years of duration. For a sake of simplicity the
of 100µ
principal metric which we will use in this work for evaluating power sources is
power per volume, specifically µW/cm3 and J/cm3 for energy density.
Today there are many forms of energy storage that may be used in micro-systems
• Ultracapacitors
• Micro-fuel cells
• Micro-heat engines
From the first zinc-copper cell invented by A. Volta in 1779 to the last years the
never before, it is the most diffused mean of energy storage for medium scale devices
and household hardware. In effects, this device are probably the easiest and most
practical solution for electronic devices because its flexibility and availability.
20
Chapter 1. Introduction
Among its major advantages there is the stability of the output voltage that allow to
the systems to run directly without any power transformation, so reducing the
dissipation of extra energy. As we have already stressed, the energy density and
lifetime are the crucial features that discriminate the various storage systems.
Common batteries are grouped in the two main classes: primary and secondary
potential energy to electrical energy, once the initial supply of chemical agents is
Secondary batteries can be recharged, that is, have their chemical reactions reversed
Alkaline, Mercury, and Lithium. These are most commonly used in portable devices
with light current drain, mostly in circuits where electric power is used intermittently
such as sensors, alarm, radio communication, small calculators but even in a fairly
high and constant consumption electronics like hearing aids and watches. Among
that we have listed, Lithium batteries are the most expensive but they possess high
energy density (2880 J/cm3), high voltage (3-4V per cell) and almost the best
duration. For example, a lithium battery with a capacity of 1000mAh can provide
energy to a wireless sensor node with an average consumption of 100µW for at least
one year. For these reasons the Li-ion battery is one of best ready-to-use solution for
Rechargeable batteries have less energy density (Lithium-ion 1080 J/cm3) than
non-rechargeable ones but for their intensive use are mostly employed in notebook
21
Chapter 1. Introduction
computer, cell phones, PDA’s, digital camera and so on. In the context of micro-
sensor devices another primary power source must be used to charge them. It’s clear
that periodically connect the nodes to a power grid is almost impossible. Indeed it
could be possible to recharge the on-board battery by solar cell as a possible solution.
However, we must taking into account of the extra dissipation due to the control
electronics for charging process. In any case, the more electrolyte and electrode
material there is in the cell, the greater the capacity of the cell. Thus the capacity of a
cell scale with its size a this is the principal reason for that it doesn’t follow the
22
Chapter 1. Introduction
1.3.1.1 Batteries
searching for a new energy source but recently, battery research seems to find a
one electrode and the lithium metal that cover the white side of the film is the other.
The sheets can be twisted, rolled, folded, or modeled in to numerous shapes with no
loss of efficiency or stacked, like printer paper to boost total output (every film can
produce 2.5volts of electrical potential). Their light weight and the inexpensive
material make them attractive for portable electronics, aircraft, automobiles and
cells. This discover can revolutionize the micro-batteries research field. In fact, the
possibility to stack many layers solves the problem of low surface area and low
23
Chapter 1. Introduction
Another promising storage device is the micro fuel cell that new state of the art
fabrication technologies have permitted to realize. The main important feature is that
their energy density is higher than an order of magnitude then conventional batteries
(18 kJ/cm3 Vs 2 kJ/cm3 of ordinary battery) . This technology can drive a cellular
phone on standby for 6 months as opposed to 2 weeks with lithium ion batteries, a
notebook computer for a week. At large scale fuel cells can produce sufficient power
for an electric car engine or an house backup energy system. These devices differ
from conventional electrochemical cells and batteries. Both technologies involve the
cell or battery employs reactions among metals and electrolytes whose chemical
nature changes over time, the fuel cell actually converts the chemicals hydrogen and
oxygen into water or another fuel such as methanol from which extracting hydrogen
(DMFC Direct Methanol Fuel Cells), and in the process it produces electricity
leaving nothing but an empty reservoir or cartridge. A proton membrane separate the
proton from the hydrogen atom and with electrons recombine with oxygen atoms on
the other side. Micro fuel cells also offers an higher power density (100mW/cm2
than micro-battery thanks to high surface to volume ratio but seems to operate good
only at higher temperatures. So, even if at large scale these device have reached 45-
90% of efficiency, at micro scale they show of it only a 20% of maximum efficiency
for methanol type. Although these technologies presents many advantages and are
24
Chapter 1. Introduction
attractive for the micro-scale world, they are not yet mature. There are also
disadvantages like not so small size (the most commercial are of to order of
range.
classical dielectric layer. No chemical reactions are involved in their energy storage
mechanism so that they provide a very high efficiency. Because of their very long
lifespan (even one million recharge cycles), short charging, high performance to
release high power in a short time, they are attractive for many applications even
working in parallel with classical batteries. While they reach significantly higher
power density (10kW/kg) their energy density if only one order of magnitude lower
that need high power peak performance can benefit from ultracapacitors but they
ordinary batteries as a secondary power sources. This imply more energy dissipation
due to the power control electronics. Furthermore, they are already limited by high
costs.
25
Chapter 1. Introduction
The energy density of a liquid hydrocarbon fuel is of the order of 30kJ/cm3 so, if
a tiny engine can convert it into electrical power, such a technology would provide at
almost 10 times the energy density of a Li-ion battery. Nowadays, the micro heat
engine research field is one of the most funded all over the world. Many approaches
Wankel rotary engine[22, 23], free and loaded piston internal combustion engine[24],
a B
The P3 is based on a thin film piezoelectric PZT transducer that convert fuel
energy into electrical. It is capable to run off a variety of sources, from diesel fuel to
solar energy or even waste heat from an hot surface or an exhaust pipe. A voltage 4V
and 1mW of power have been achieved with membrane generators 9mm2 in area and
2-3 µm thick[27]. The power performance predicted for micro heat generators ranges
However, this so large power output does not always represents a benefit, specially
for wireless sensor micro node where the power request is small whereas their
26
Chapter 1. Introduction
autonomy in time, that depends on energy storage density, is more important. So, the
efficiency doesn’t overcome 20% the energy density of about 7 kJ/cm3 gains only a
factor 2-3 relative to a Li-Ion battery. Furthermore, devices that burn fuel potentially
low-energy beta emitter can provide energy to MEMS for decades dependently only
material could be in both solid and liquid form: current best candidates as nuclear
63
sources of electric charges are, for example, i, 3H, 210
Po. Evidently, no gamma
emitters are possible sources such as 238U for commercial application because of the
heavy shielding needed to avoid heath risks and electronics damages. There are two
main viable methods under investigation and at micron-scale some prototypes have
that make use of betavoltaic effect. The second concept is based on self-
electrical potential due to net positive charge flow of the β-particle induced electron
27
Chapter 1. Introduction
hole pairs (EHPs). As electric field of the depletion region sweeps the induced
charges across the junction a resulting current is created from n to p-type layer.
a b
Figure 1.6
a) Betavoltaic microbattery based on a pn-junction with the the inverted pyramid tank of
Liquid 63iCl/HCl solution. b) View of the bulk-icromachined inverted pyramid array.
c) Picture of a packaged sample of betavoltaic micro-battery based on a planar Si pn-diode with
electroplated 63i [28].
cantilever plate faced in front of radioisotope source. The increasing electric force
created deflects the beam until it contacts a ground-electrode. After this point the
beam initiate to oscillate while the plate restarts to collects other charges for another
cycle. In this mode, the cantilever became like an intermittent oscillator and its
28
Chapter 1. Introduction
b
Figure 1.7 – a) Drawing of a 63i radioisotope piezoelectric cantilever.
b) Sensor microchip with on-board PZT nuclear generator [30].
they have no high efficiency: only 0.5% it’s been demonstrated for betavoltaic
Moreover, they are not yet suitable for high power devices. The electric power of
µW/cm3 of PZT
them range from 10nW/cm3 for betavoltaic technology to about 10µ
viable method in addition to energy stock method. In this case, the energy density or
power density per unit volume is not a good metric to measure the performances
because the power received primarily depends by the efficiency of the power
29
Chapter 1. Introduction
during the early 1900s. Nikola Tesla designed his own transmitter with power-
1975 a microwave ray of 30Kw was beamed over a distance of 1 Mile at 84%
efficiency. After that, far field wireless power transfer systems based on traveling
microwaves have had no great success because of the health and safety risks due to
with the use of resonant coupling, wavelengths produced are far lower making it no
more dangerous than being exposed to radio waves. For instance, WiTricity
based on near field inductive coupling through magnetic fields like RF ID tags. They
were able to transfer 60 watts with ~40% efficiency over distances at about 2 meters.
Nevertheless, this technology is not suitable for mid-range and long-range (5-10
meters and beyond) power distribution. The power transmitted to a node is expressed
by P(r) = Po λ2/(4πr2) where ‘Po‘ is the transmitted power, ‘λ’ is the wavelength of
the signal and ‘r’ is the distance between transmitter and receiver. Assuming a
maximum distance of 10 meters in the frequency band of 2.4-2.485 GHz and that a
single node consumes at max 100µW, the power transmitter needs to emit 10-14
30
Chapter 1. Introduction
transmitted fall more realistic as 1/r4 indoor the efficiency will follow the same trend
as well.
The advantages of distribute power to sensor nodes by means of wires are very
devices where the power grid for sensors should be foreseen in the design. But for
dense sensor network this way is not practicable due to high costs, prohibitive
Extracting energy from acoustic wave could be feasible power source only for
power density less then 1µW/cm2 that is far from the 100µW/cm3 target discussed
before. Finally, if at first glance deliver energy to sensors with a laser directly
focused toward them is possible, this method does not present so many benefits and
failure of its own electro-mechanical components. On the other hand, these methods
31
Chapter 1. Introduction
are not yet deeply explored because of the complex variety of environments, each
one with its forms of renewable energy source. So that, there is not a unique solution
suitable for all environments and applications. Unlike energy reservoirs, power
scavenging sources provide the energy for the time during which the source is in
operation. Therefore, they are primarily characterized by the power density, rather
Instead, the lighting power density in indoor environments ranges from about
efficiencies of about 15% and up to 20-40% for the state of the art expensive
research cells. However, these type of solar cells are not suitable for indoor
environments because they are affected by severe degradation of the open circuit
voltage[31]. Thin-film polycrystalline cells are not expensive but show efficiencies
of only 10 – 13%. A cadmium telluride (CdTe) thin-film cell type has a very wide
spectral response. So, it has good performance in both indoor light conditions and
32
Chapter 1. Introduction
a
Figure 1.8
a) Integrated beacon circuit with an on-board 3x2cm2 Panasonic BP-213318 CdTe solar cell [32]
b) 16 mm3 mock-up with an integrated millimeter solar panel (Smart Dust [33])
For these reasons, it is selected which best candidate as a power generator that can
applications (fig. 1.8a). Even though it needs a proper power electronics to transform
the current for the battery and to optimize its lifespan. More cheaper plastic organic
2.5%. This is too low for our scopes but researchers promise to reach values like that
The wind flow power goes as cubic power of its velocity “v” and it is direct
proportional to air density “ρ” and cross sectional area “A” by the relation
velocity between 2m/s and 6m/s the power density ranges from about 20µW/cm2 to
10mW/cm2 for a conversion efficiency of 20%. But we must take into account that at
33
Chapter 1. Introduction
low velocity the efficiency normally does not overcome 5%. Unlike large-scale
windmills have reached efficiency of 40% thanks to even more sophisticated material
technology and shapes, at small scale the research is currently quite poor because the
All of those have the efficiency related to the Carnot law expressed by equation
the heat power that flow through conduction along a 1cm length for a ∆T=5°C is
117mW/cm2. At first sight this could be seems an excellent result but the real devices
have efficiencies well below the simple Carnot rule. Exploiting the Seebeck effect
efficiency factor of 0.83 µW/K2cm2. A more efficient approach that has already been
that approaches the ideal vapor Carnot cycle. Mechanical power is converted into
34
Chapter 1. Introduction
The energy burnt by an average human body every day is about 10.5MJ that
801600 Watts. Many research groups and industries are currently working on the
most efficient technologies to tap the energy worn by human body. For example,
piezoelectric insert embedded into a shoe can capture energy "parasitically" from
footfalls (theoretically available from 5-8 Watts up to max 68W) while walking[38].
However, the efficiency of this technology does not overcome on the average 17%,
excluding some advanced prototype, so that the effective mean power harvested is
about 1.5 Watt. Wristwatches powered by both the kinetic energy of a moving arm
and the heat flow from the surface of the skin are yet available. These make use of
the so called Inertial Power Generators based on 2 gram “proof” mass mounted off-
center on a spindle. As the user moves during the day, the mass rotates on the spindle
and winds the mechanism. Some models such as ETA Autoquartz Self-Winding
Electric Watch or Seiko AGS system Seiko creates 5µW on average when the watch
is worn and 1mW when the watch is forcibly shaken. But generally, also scaled up
these kind of scavenging energy systems do not produce more that 10mW. While it is
35
Chapter 1. Introduction
sensors still remain. So, these ways could be both impractical and not cost efficient
Another renewable power source could comes from atmospheric pressure and/or
thermal variation. The possible energy available E for a fixed volume V and a
varies of 677 Pa once per day the available power density would be 7.8nW/cm3.
While, considering that a pressure changing due to temperature variation for a fixed
volume of ideal gas follow the state equation ∆P=mR∆T/V where m is the mass of
the gas (i.e. Helium) and R a gas constant, for a ∆T=10°C thermal variation per day
the corresponding energy change would be 1.4 Joules, which is about 17 µW/cm3.
So, we are tens time below desirable power density for sensor nodes. Although there
are yet some devices that incorporate power supply system that make use of phase
changing of a fluid like “Atoms clock”, however, there are no recent advances in
1.3.3.6 Vibrations
Vibrations
environments, military devices and so on. Their characteristics are various: spectral
36
Chapter 1. Introduction
shape from low to high frequency, amplitude and time duration are manifolds
shows that the power density that can be converted from vibrations is about
µ/cm3. Another strong point is that as solar the vibration source is a renewable
300µ
source as well, so, it has no lifetime and in addition it is not limited to the sunlight
areas. In order to establish how much power comes from vibrational excitation
In figure 1.9 are shown a list of vibration sources measured with a standard
Frequently, the most of vibrational energy is located at fairly low frequencies (below
1.10, shows sharp peaks in magnitude around 120Hz and 250Hz. Even for a milling
37
Chapter 1. Introduction
machines the peaks of fundamental mode fall around 70Hz with few higher
harmonics close to 150 and 200Hz. Likewise for a wooden deck the first vibration
modes appear at 350Hz and at 240Hz for a refrigerator. The sharp peaks at low
frequency indicate the fairly sinusoidal shape of displacement and acceleration signal
oscillating system.
the power spectrum tends to fall off as ω2. Other source such as water jet assisted
with an accelerometer by a group[39] have two main resonant peaks to consider and
those peaks at about 400 Hz and 1400Hz. Many meso and micro-scaled energy
38
Chapter 1. Introduction
scavenging generators have been developed in the last five years by an increasing
number of research groups[4, 9, 10, 16, 38, 40-42]. Three are the principal concepts
electro-magnetic. These models will be more deeply analyzed in the next chapter.
The predicted power density that a such kind of micro-scaled generators can extract
of the source, where the most of energy of the vibration spectrum is sited. This
constitute a great limit. In facts, although larger structures can achieve relatively
higher power densities (for instance, a simple shake-driven flashlight can delivers 2
elastic constant and m the inertial mass). Far from fundamental frequency of the
environment, that’s for hypothesis within a band below 300-400Hz, the resonator is
a
Figure 1.11 a) SEM photo of the fabricated cantilever prototype[43].
39
Chapter 1. Introduction
piezoelectric resonator with a inertial Nickel mass 0.02 grams has a natural
frequency of about 608Hz and its power output is 2.16 µW. In order to obviate to
batteries and fuel cells, and renewable energy sources, it is difficult to use the same
metric. Because some sources like batteries are benchmarked by energy density
while other are characterized by power density such as solar cells. Some devices do
not need of the third dimension so, they can be characterized by power per square
40
Chapter 1. Introduction
Roundy et al. have analyzed the characteristic of various power sources so far
discussed (tab. 1.2, fig.1.12). This is a starting point for the choice of optimal way for
sensors. How it has been already stated before, it’s improbably that any single
solution will satisfy all applications, because each method has its own constraints.
Solar cells require sunlight, thermal gradients need sufficient temperature variation,
sources are generally omnipresent and can be readily found in inaccessible locations
41
Chapter 1. Introduction
It can be noted that solar and vibration power density can be range within an
office light condition for solar cells, low level vibrating environment) but they are no
function of lifetime. While both energy drain and leakage determine a variation in
time for chemical batteries with inflection point for rechargeable types. So that,
within one year the batteries can support efficiently a wireless sensor nodes
(assuming a 100µW of consumption) but going beyond 2 year arise the refilling
problem for those rechargeable and over 5 year the primary batteries cannot provide
the same power level of solar cells or vibration-based generators. Others interesting
indicators for a comparison among the various methods (especially for batteries) are
the specific power defined as power over weight ratio and specific cost per watt.
42
Chapter 1. Introduction
Besides the combustion engines that have the highest specific power (fig. 1.13),
transducers present performances comparable with fuel cells. In any case, these
43
Chapter 1. Introduction
methods do not have the problem of high noise output, toxic exhaust fumes and
instability of electrical power from the system typical of combustion engines. On the
front of cost-effective the piezoelectric seems to be the worst solution (fig 1.14) but
this research was been conducted for power demands in 100mW–30W, then for
meso-scaled devices. Instead, for micro devices both specific power and cost-
effective are not so critical, while, size scalability and lifespan are the most important
parameters.
bistable oscillators and vibration noise source as a basis from which simulations and
44
Chapter 2
Vibration driven
microgenerators
There are three possible devices that can transform ambient vibrations into
electrical energy:
45
Chapter 2. Vibration driven microgenerators
These three methods are commonly used for inertial sensors (i.e. accelerometers)
as well as for actuators. The best transducers systems should be those that can
maximize the coupling between the kinetic energy of the source and the conversion
vibrations. Vibration kinetic energy is best suited to generators with the mechanical
component attached to an inertial casing which acts as the fixed frame. The case
A brief analysis of strength and weakness points of the existent transducer models
will be outlined in this paragraph. Further details on dynamic and equivalent circuit
models will be explained in the next paragraph 2.2. with particular focus on
in two plates which are electrically isolated from each other by a dielectric (typically
air, vacuum or an insulator). Unlike the simple fixed capacitor the metallic plates of
variable capacitor can be in motion in order to vary its capacitance. As the separation
between the plates (typically nanometer or microns for a MEMS) varies the energy
stored in the charged capacitor changes due to the work done by an external vibrating
force. The capacitance for a parallel plates capacitor in term of the insulator
46
Chapter 2. Vibration driven microgenerators
A
C = kε 0
d
(2.1)
where A is the plate surface, d the relative distance, ε and ε0 are the permittivity of
the dielectric material and vacuum, while, the voltage across the plates is expressed
by definition from
V = Q/C (2.2)
hence,
Qd
V= (2.3)
ε0 A
1 1 1
E = QV = CV 2 = Q 2C (2.4)
2 2 2
At constant voltage, in order to vary the energy it’s needed to counteract the
1 AV 2
Fe = ε 2 (2.5)
2 d
then, the mechanical work against this electric force done by an external force like
the capacitance. A current flow through a load shunted to plates in order to balance
the fixed voltage. A similar method like fixed voltage is that of charge constrained,
with the difference that if a constant charge is held into the plates (i.e. by means of a
1 2d
Fe = Q (2.6)
2 εA
47
Chapter 2. Vibration driven microgenerators
but in general, the voltage constrained offers more energy than the charge
constrained approach. A manner to increase the output electrical energy for the
charge constrained method is add a capacitor in parallel with the variable harvesting
capacitor. This parallel storage capacitor effectively constrains the voltage on the
energy harvesting capacitor. A base circuit was designed by Roundy (fig. 2.1) where
Cv is the variable capacitor, Cpar the parasitic capacitance associated with the variable
capacitance and interconnections, finally, the switches which transfer the electric
current toward the storage capacitor and regulate the charging that can be substitute
by diodes.
The maximum potential energy per cycle that can be harvested by this configuration
1 Cmax + C par
E = Vin2 ∆C (2.7)
2 Cmin + C par
1
E = VmaxVin ∆C (2.8)
2
with ∆C=Cmax-Cmin and Vmax which represents the maximum allowable voltage
across a switch.
48
Chapter 2. Vibration driven microgenerators
Up to now, there are three kinds of electrostatic generators (fig.2.1) that are based
on both constrained charge and voltage[44, 45]: In-plane overlap varying, In-plane
a b c
For the in-plane overlap topology (a) the capacitance changes by changing overlap
area of interdigitated fingers that implements the multi-plates capacitor. While for
the other two types the capacitance changes by changing gap between fingers (b) or
large plates (c). In Table 2.1 it is shown the electrostatic force variation for the three
For the Out-of-plane gap closing type there are several problems. The gap x must
become very small in order to obtain a large capacitance change but, as the fluid
damping force is proportional to 1/x3, the loss becomes very large as the plates move
49
Chapter 2. Vibration driven microgenerators
close together. A possible solution may be to set the MEMS device under very low
contact as the plates get close together. In-plane gap closing converter solves this
problem. For this type of configuration the motion of plates is in the plane of the
substrate, therefore, the minimum dielectric gap, and thus the maximum capacitance
by Roundy[47] (fig. 2.2) the In-plane gap closing type offers the highest power
output with an optimized design producing 100 µW/cm3; out-of-plane gap closing is
the next highest and the last in performances is in-plane overlap varying. It can be
noted that the maximum power occurs at very small dielectric gaps.
a b
integrated with silicon based microelectronics that are readily available. Therefore,
the scalability of its size through MEMS technology is the first reason why
electrostatic converter is attractive. On the other hand, one of the principal negative
50
Chapter 2. Vibration driven microgenerators
recently some groups [48] reports on MEMS electrostatic converter with high
electrical damping capable to operate over a wide low frequency range (<100 Hz): a
power of 70 µW.
Many other group are focusing in realization of low frequency operating converters
generator that harnesses ventricular motion operating at heart beat frequency 1-2Hz
with the aim of driving a cardiac pacemaker permanently. Anyway the power output
for a square centimeter variable capacitors that have been developed so far range
from 10 to 100µW.
51
Chapter 2. Vibration driven microgenerators
connected to a spring with elastic constant k that is attached to a rigid case (fig. 2.5).
The ambient vibration excites the housing which transmits the mechanical
displacement to the inertial mass. The consequent variation of magnetic flux through
the coil generates an inducted current in accordance with Faraday’s law. In this way
the part of kinetic energy stored in the movement of mass-spring system is converted
a
Figure 2.5 – a) Drawing of inductor generator, Amirtharajah [9]
b) Cross-section of the wafer-scale electromagnetic generator proposed by Williams[50]
There are many other preferable configurations: for example with the magnets
dΦB
ε =− (2.9)
dt
hence, for a coil moving through a perpendicular constant magnetic field, the
52
Chapter 2. Vibration driven microgenerators
dx
Voc = &Bl (2.10)
dt
where & is the number of turns in the coil, B is the strength of the magnetic field, l is
the length of a winding and x is the relative vertical distance between the coil and
magnet. Making a few assumptions: baseline vibrations of 2.25 m/s2 at 120 Hz,
maximum device size is 1cm3, and about the magnetic field intensity and coil design,
it can easily be shown that output voltage does not overcome 100mV. Far more
realistic estimates of present technology range within 50mV, otherwise always less
then 1 Volt. For a typical 5mm x 5mm x 1mm device, the predicted power generation
was 1µW for an excitation frequency of 70Hz, and 100µW at 330Hz. Though the low
high output current levels are achievable. Second, unlike electrostatic conversion, no
separate voltage source is needed to get the process started. Moreover, the almost
total absence of mechanical contact between any parts improves reliability and
various types of material[50-53] that are well suited and proven in cyclically stressed
applications. Some of those with size of 5mm x 5mm x 1.5mm achieves 35mW of
minimum line and space for coils fabrication and most of all limited amplitude of
vibrations (∼10µm).
53
Chapter 2. Vibration driven microgenerators
Piezoelectric ceramics have been used in many applications for many years to
convert mechanical energy into electrical energy. The direct piezoelectric effect was
early demonstrated by Jacques and Pierre Curie in 1880. They found that when
certain ceramic crystals were subjected to mechanical strain, they became electrically
polarized and the degree of polarization was proportional to the applied strain.
which show piezoelectricity are widely available in many natural and man-made
forms: single crystal quartz, cane sugar, Rochelle salt, piezoceramic materials (e.g.
Lead Zirconate Titanate, PbTiO3, BaTiO3 composites [54]), thin film (e.g. sputtered
zinc oxide), screen printable thick-films based upon piezoceramic powders [55],
material[57].
Figure 2.6 - Piezoelectric elementary cell; (1) before poling (2) after poling.
54
Chapter 2. Vibration driven microgenerators
The origin of the piezoelectric phenomenon is due to the asymmetry in the cell unit
clouds from their individual atomic center and this lead to a macroscopic net
polarization of the crystal lattice (fig. 2.6). The compressive and tensile stresses
along one single direction will generate a parallel electric field and a consequent
force that opposes to the length variation. It is also reciprocal, the same crystal
exposed to an electric potential will experience an elastic strain causing its length to
decrease or increase according to the field polarity. Each of these effects result
almost linear within small length variation relative to the crystal size.
Groups of dipoles with parallel orientation form the so called Weiss domains
(fig.2.7). the raw piezoelectric material has these domains randomly oriented.
Applying an electric field (> 2KV/mm), the material expands along the axis of the
field and contracts perpendicular to that axis. After poling action the material
55
Chapter 2. Vibration driven microgenerators
mechanical, thermal and electrical limits of the material) and it is grown in the
dimensions aligned with the field and it’s contracted along the axes normal to the
the Weiss domains increase their alignment proportional to the voltage causing the
The coupling between the electrical and mechanical behaviour of the material has
variables:
S = s ET + dE
(2.11)
D = dT + ε T E
56
Chapter 2. Vibration driven microgenerators
(z) 3
direction of polarization
5
2 (y)
4
(x) 1
The axes 4, 5 and 6 identify rotations (shear). The direction of polarization (3 axis) is
established during the poling process by a strong electrical field applied between two
extract the voltage the d33 coefficient determines the electro-mechanical coupling
produce much more high voltage then that transverse, however, is not a practical
In general, the elements of piezoelectric beams or films are coupled in the transverse
direction because such a configuration is more practical and it multiplies the applied
mechanical stress.
57
Chapter 2. Vibration driven microgenerators
Table 2.2 shows the fundamental constants that characterize piezoelectric material
Eie
kij = (2.12)
E mj
between the electrical energy Eie stored along the i-axes and the mechanical input
energy Eim along the j-axes. This describes the efficiency of energy conversion of
the material between electrical and mechanical form in a given direction. The matrix
gij is defined as the electric field generated per unit of mechanical stress, or the strain
the material which is defined as the dielectric displacement per unit electric field and
s which is the compliance matrix namely the strain produced per unit of stress. In
58
Chapter 2. Vibration driven microgenerators
Piezoelectric generator
Cp Rp
Vp RL
The voltage source represents the voltage that develops due to the excess surface
charge on the crystal. The series capacitor Cp represent the capacitance of the
piezoelectric layer which is proportional to the film permittivity and area and
resistance and RL a purely resistive load. Assuming that the mechanics take place
along a single axis then each variable or constant is treated as a single scalar quantity
rather than a tensor. The open circuit voltage that results from an external mechanical
d ⋅t
Vout = − σ (2.13)
ε
while the average power dissipated by a simple resistive load will be PL=VL2/2RL. In
direct generation of suitable voltages and currents [44], small mechanical damping
material and unlike the variable capacitors they do not necessitate of separate voltage
59
Chapter 2. Vibration driven microgenerators
piezoelectric MEMS with thin film under one millimeter[59] but some problems of a
A fundamental starting point to benchmark the three methods so far showed can
be made by considering the theoretical inherent energy density and summarizing the
strength and weakness side of each one. The following table realized by Roundy et
al. [10] shows the practical and theoretical maxima of energy density for each
transduction techniques.
density of piezoelectric mechanism that is 17.7mJ/cm3 and the other techniques that
produce only 4mJ/cm3. Finally, the advantages and disadvantages points of each
60
Chapter 2. Vibration driven microgenerators
Table 2.4 - Summary of the comparison of the three transduction mechanisms. [15]
From this comparison it is clear that the most desirable conversion method results
that piezoelectric one which presents the major number of advantages. So, it is for
these reasons that this is currently the best choice to realize the micro vibration-
On this basic choice a generic theory of linear and non-linear resonator with
piezoelectric coupling will be discussed in the next paragraphs. Anyhow, the nucleus
of this thesis, that is the idea to exploit the non-linear dynamics to enhance the
capacitive techniques.
61
Chapter 2. Vibration driven microgenerators
Making use of the linear system theory it is possible to express a simple generic
model for the conversion of kinetic energy of a mass that undergoes to a vibrating
spring of stiffness, k.
k
x
m
dm+de
y(t)
The energy losses by friction are related to the internal mechanical damping which is
the electro-mechanical conversion. For simplicity, it can be assumed that the inertial
mass m of generator is much smaller then the vibrating mass of ambient (wall, floor,
machine) and that the vibration source is an infinite energy reservoir. In this mode
vibration moves out of phase with the inertial mass when the generator housing is
vibrated at resonance resulting in a net displacement that we call for simplicity x(t),
between the mass and the frame. If we consider the displacement y(t) of the vibrating
62
Chapter 2. Vibration driven microgenerators
housing rather than the external force, the differential equation of motion for the
the force on the mass is equal to the force on the mass-spring-damper, that, for a
The simple steady-state solution for the mass displacement of the equation (2.14) is
ω2
x(t ) = Y0 sin(ωt − φ ) (2.16)
2 2
k 2 (d e + d m )ω
−ω +
m m
setting dT =dm+de the total damping coefficient, the phase angle φ is given by
dT ω
φ = tan −1 (2.17)
k −ω m
2
Maximum energy can be extracted when the excitation frequency is tuned to the
ωn = k / m (2.18)
The instantaneous kinetic power p(t) transferred to the mass is the product of the
p (t ) = − myɺɺ(t )[ yɺ (t ) + xɺ (t )] (2.19)
Taking the Laplace transform of equation 2.14 and 2.19, the transfer function is
X (ω ) ω2
H xf (ω ) = = (2.20)
Y (ω ) −ω 2 + 2iω (ζ e + ζ m )ωn + ωn2
63
Chapter 2. Vibration driven microgenerators
ζT=(ζe+ζm)=dT/2mωn, is expressed by
2 2
Pdiss (ω ) = mζ T ωn Xɺ = mζ T ωnω 2 f ⋅ H xf (2.21)
that is
3
ω
mζ T Y0 ω 3
2
Pdiss = ωn (2.22)
2
ω 2 ω
2
1 − + 2ζ T
ωn ωn
mY0 2ωn3
Pdiss = (2.23)
4ζ T
mA02
Pdiss = (2.24)
4ωnζ T
For steady-state solutions like these, power remains limited and does not tend to
infinite as the damping ratio tends to zero. Separating parasitic damping ζm and
mζ e A2
Pel = (2.25)
4ωn (ζ m + ζ e )2
it can be regulated to improve the conversion mechanism. As we can see from the
64
Chapter 2. Vibration driven microgenerators
graph 2.11, for a sufficient acceleration, increasing the total damping coefficient will
results in a broader bandwidth response or the oscillator which loss the power
transferred. So, the damping factor control the selectivity of the device.
ζ=0.05
5 ζ=0.1
ζ=0.2
ζ=0.3
4
Power
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω/ωn
It is clear from 2.22 that the inertial mass of generator should be maximized within
the geometrical constrains in order to obtain the maximum electrical output. For a
Furthermore, it is necessary to know the spectral shape of vibration noise source well
to properly design the transfer function of a linear resonator. So, it is critical that the
driving vibrations. Unlikely, it is not always simple to find a source that concentrate
the vibrational energy around a single frequency constantly. Often, the most part of
65
Chapter 2. Vibration driven microgenerators
wide bandwidth that changes in time. For instance, inside a car tyre or on aircraft the
spectral power density could vary a lot in time domain. Up to now, there are no so
many solutions to face up this limitation. Systems with active tuning or multimodal
resonator are under investigation [60, 61] but they present some problems of size
scaling and energy losses due to the control electronics and mechanical configuration
This linear model proposed by Williams and Yates is only a first approximation
and neglects the details of transduction mechanism. It is a fairly good model for
coupling term is not always linear (most of all for piezoelectric system) and
and this model can be a starting point to compare the efficiency between linear and
nonlinear harvesting devices. In facts, it’s quite simple to calculate the electro-
mechanical induced damping term ζe for each type of conversion mechanism in first
approximation.
66
Chapter 2. Vibration driven microgenerators
For the purpose to explain the main idea on which is based our research, we show
the basic dynamic model of a bistable damped oscillator which is forced by random
dU ( x)
mxɺɺ(t ) = −η xɺ (t ) − + σξ (t ) (2.26)
dx
where m is the inertial mass that we can normalize to 1, η is the damping coefficient,
σξ is the stochastic external excitation and the conservative Duffing potential is the
quartic well
x2 x4
U ( x) = −a +b (2.27)
2 4
from such a potential the 2.26 differential equation becomes the Langevin equation
x (t ) = −δ xɺ (t ) + ax − bx 3 + σξ (t )
ɺɺ (2.28)
Accounting the driven random force, for instance, a Gaussian distributed white noise
differential equation 2.28 cannot be calculated analytically but only numerically. For
now, here we only give the basic equations and qualitatively description. As it can be
seen from figure 2.12, for a fixed b>0 and a<<0 the potential resemble to quadratic
harmonic well and in fact the oscillator behaves likewise a linear resonator. With a
close to zero the potential still has only one equilibrium position at x=0, but the basis
shape of the well becomes even more flat until the critical condition a=0 is reached.
67
Chapter 2. Vibration driven microgenerators
350 U Hx L
300
250
a<< 0
200
150
100
a= 0
50
x
-4 -2 2 4
U Hx L
60
40 a> 0
20
-4 -2 2 4 x
barrier height
-20 a>> 0
Figure 2.12 – Biquadratic potential well for fixed b>0 and various values of a.
When a>0 bistability arises with the formation of two relative energy minima at
standard deviation σ and its autocorrelation time, a certain dynamical regime can
outperforms the kinetic energy transfer from the source to the system.
68
Chapter 2. Vibration driven microgenerators
and it will able to adsorb vibrating energy from a narrow or wide frequency
bandwidth. Bistable systems have been extensively studied in the presence of noise
both in the classical[63] and in the quantum domain[64], but some aspects related to
the energy spectrum and dynamics requires a deeper insight into the stochastic
dynamics of the oscillator because analytical descriptions do not exists. These will be
explained with the help of simulations and experiments discussed in the next chapter.
We are going to expose in the next paragraphs the basic models of bistable vibration
based transducers which were used in the simulations and experimental tests.
realizing at macroscopic scale. So, it has been chosen to test the main concept of the
present work. If part of its bar is made by a flexible piezoelectric beam it realizes a
attached at the upper tip of a rigid bar with a global moment of inertia I. The
clamping base is forced to vibrate by a random shaking force f(t). Looking the
system from to the reference frame of the clamping base, for the sake of simplicity,
the inertial mass m sees as a stochastic force ξ(t) applied to itself. Furthermore, the
mass is subjected to a restoring elastic force relates to the effective stiffness Keff and
69
Chapter 2. Vibration driven microgenerators
m Y
ξ(t) x
θ
ξ(t)
-mg k
x
l m
k η
f(t) X
I f(t)
a) b)
Vp
piezoelectric element
Approximating the periodic term with the Taylor series expansion around θ=0
sin θ ≈ θ − (1/ 6)θ 3 to the second order, we obtain the stochastic Duffing equation
70
Chapter 2. Vibration driven microgenerators
From the equation 2.31 we have two cases. One equilibrium position exists for
mgl<Keff (a<0) that corresponds to linear behavior with m exactly above the rotation
axis for θ=0. Two equilibrium positions: one at right and one left when mgl>Keff
(a>0) that may be approximately calculated equating the opposing torques and are
structural geometry (relations 2.35) of the piezoelectric bender [6, 65, 66], the
Where Ip is the current flowing in the equivalent circuit that could be assumed a
purely resistive parallel load impedance (figure 2.9b) so that Ip(t)=Vp/RL. Otherwise,
it could be used a more complex AC-DC harvesting electronic circuit (i.e. rectifier
sketched in fig. 2.14) and with a more efficient current regulator system such as
Ip
Vp
Ce Rout Vc
Piezoelectric transducer
71
Chapter 2. Vibration driven microgenerators
tungsten mass attached to its tip. This is the configuration that we have chosen for
tsh
s
m
lm
deflection
bimorph
strain
x
piezo-bender
lb
m
tp
configuration of figure 2.15 rather than ‘θ’, the dynamical coupled equations (2.33)
K eff Kv 1
x(t ) = Fa − x(t ) − 2δ K eff / m * xɺ (t ) + Fb x (t ) − Vp (t ) + ⋅ σ ⋅ ξ (t )
3
ɺɺ
m m m
(2.34)
Vɺ (t ) = K xɺ (t ) − V p (t )
p c
RL Cb
where it has been considered that the ratio x/lb<<1, hence, from the Taylor series
derived making use of modal analysis, Euler beam equation piezoelectric linear
72
Chapter 2. Vibration driven microgenerators
setup. Using the equations of piezoelectric linear theory 2.11 applied to piezo-
cantilever modeling[61, 67] the electromechanical coupling constants are given by:
K eff d 31a
a) K v =
2t p k1
t p d31YpE k1
b) K c =
aε p
(2.35)
3(t p / 2 + th / 2)
c) k1 =
lb (lc + lb )
4I
d) k2 =
(t p / 2 + th / 2)(4lb + 3lm )
The parameters Fa and Fb of 2.34 are used to tuning the shape of elastic-gravity
bistable potential and they depend in this case by the gravity acceleration g and
stiffness Keff. Moreover, d31, εp, YEp and I are respectively dielectric displacement
moment of inertia of the beam. While, the other are geometrical parameters showed
in fig. 2.15.
later, it could be possible to imagine oscillating systems which work in other kinds of
73
Chapter 2. Vibration driven microgenerators
anharmonic potentials for our scope. For example, oscillators moving on a periodic
investigated.
2.4Bistable Piezoelectric
iezoelectric beam in a repulsive
magnetic field
Even though we chose a commercial material with an high mass density (i.e.
energy harvesting generator along only the axis of gravity force. An alternative
bistable system feasible on both the macro and micro scale is an oscillator in a
constant magnetic field that creates the bistable potential. A magnetic force acting to
permanent magnets and/or coils. This force can be repulsive or attractive in order to
counteract or reinforce the restoring elastic force of the beam. In this way, we can
adjust the shape of the potential well passing from mono to bistable dynamics. In
74
Chapter 2. Vibration driven microgenerators
hm
permanents
∆ magnets M2
ξ(t) excitation r θ
lm m
M1
‘xd’ deflection
lb Piezo bender
X
a) b)
In the hypothesis of a relative distance ∆>>hm larger then the size of each magnet,
we can assume the interaction between two permanents magnets likewise of two
“point” magnetic dipoles. The force between two dipoles having the same magnetic
Fm = ± ar 3µ 0 M 2 / 2π r 4 (2.36)
is the unit radial coordinate vector. This supposition is not ever valid since it depends
by the specific configuration. For short distance, in facts, the magnetic force goes as
square of the distance. Nevertheless, the experimental test was resulted in better
agreement with 2.36 rather than the inverse square of distance law. Making another
75
Chapter 2. Vibration driven microgenerators
strong assumption that the deflection xd is small relative to the beam length lb, we can
consider the two dipole always aligned. In this case the tangential magnetic force can
be neglected, then, the radial component of the magnetic force sees by tip magnet M1
3µ 0 M 1 M 2 x
Fm = (2.37)
2π ( x + ∆ 2 )5 / 2
2
µ MM
K eff ( x 2 + ∆ 2 ) + 0 2 1 22 3/ 2
1
U ( x, ∆ ) = (2.38)
2 2π ( x + ∆ )
a) ɺɺ
x(t ) = −( K eff / m) x(t ) − 2δ K eff / m * xɺ (t )+ ≈
3µ0 M 1M 2 x(t )
≈ +(1/ m) ⋅ − ( K v / m)V (t ) + (1/ m) ⋅ σ ⋅ ξ (t ) (2.39)
2π ( x(t ) + ∆ 2 )5 / 2
2
V (t )
b) Vɺ (t ) = K c xɺ (t ) −
RL Cb
The dynamics and stability points of the system are now controlled by the relative
magnets distance ∆ that now plays the same role of the parameter a in Duffing
potential 2.27 as it can be seen from the plot 2.17. Adjusting this parameter the
system passes from quasi-linear monostable to bistable behavior. For large ∆ the
system oscillates around the minimum located at zero displacement. When ∆ reaches
a critical value the potential well becomes flat, thus, the system remains monostable
but shows an anharmonic dynamics. After ∆ has overcomes this critical value, the
potential shows two minima separated by a rising barrier with the decreasing of ∆.
76
Chapter 2. Vibration driven microgenerators
0.14
∆=5
0.12 ∆=7
∆=8
∆ = 10
0.1 ∆ = 15
∆ = 20
U(x)
0.08
0.06
0.04
0.02
-10 -8 -6 -4 -2 0 2 4 6 8 10
x (a. u.)
∂U ( x, ∆) −
5
= K eff x − 3 xK m ( x + ∆ ) 2
2 2
(2.40)
∂x
∂ 2U ( x, ∆) −
5
−
7
= K − 3 K ( x 2
+ ∆ 2
) 2
+ 15 K x 2
( x 2
+ ∆ 2
) 2
(2.41)
∂x 2
eff m m
77
Chapter 2. Vibration driven microgenerators
1
3K 5
K eff − 3K m ∆ −5 < 0 ⇒ ∆ < m (2.43)
K
eff
Setting α = (3K m / K eff )1/ 5 , when ∆ > α there is an absolute minimum and the
In the limit case for ∆ → ∞ the angular frequency becomes ω02 = K eff / m namely the
For ∆ < α the potential has three different zeroes derived by the equation U’(x)=0,
( 3K / K eff )
2/5
xm = ± m − ∆2 (2.45)
3
1 5 K eff 5
∆U = K eff ∆ 2 + K m ∆ −3 − K m (2.46)
2 2 3K m
Where for a decreasing ∆ the cubic term dominates the concurrent square term.
Consider now the energy balance. Let 2.39a be multiplied by xɺ and 2.39b be
78
Chapter 2. Vibration driven microgenerators
tf
1 tf 1 tf 1 tf
∫ σξ xdt
ti
ɺ = m xɺ 2 + K eff x 2 + C p V p2 + ≈
2 t t
i 2 t i 2 i
tf tf tf
(2.47)
≈ − K m ∫ x( x 2 + ∆ 2 )−5/ 2 xdt
ɺ + ∫ η xɺ 2 dt + ∫ Vp I p dt
ti ti ti
where, considering a purely resistive load the electrical converted energy term is
tf t t
1 2 tf
f
1 2 tf
f
V p2
∫t v p
K V xdt
ɺ =
2
C V
p p t
i
+ ∫t p p 2 p p ti ∫t RL dt
V I dt = C V + (2.48)
i i i
and the last integral represents the electrical energy dissipated on pure resistive load
RL. The physical meaning of all terms in energy balance equation (2.47) is described
∫ σξ xdt
ti
ɺ Input energy
1 tf
m xɺ 2 Kinetic energy
2 ti
1 tf
K eff x 2 Elastic energy
2 ti
tf
− K m ∫ x( x 2 + ∆ 2 ) −5 / 2 xdt
ɺ Magnetic energy
ti
tf
∫ η xɺ dt
2
Mechanical losses
ti
tf tf
1 2 tf
∫t k vV p xdt
ɺ =
2
C p V p ti
+ ∫ V p I p dt
ti
Converted electrical energy
i
79
Chapter 2. Vibration driven microgenerators
rate of the intra-well jump e the distribution P(x) of the position as a function of
noise strength. On the other hand, being the excitation a random force, the motion
equations become stochastic nonlinear differential equations that can be solved only
through numerical methods whose results will be exposed in the next section. The
energy balance and efficiency will be computed through numerical evaluation of the
velocity xɺ . As it be easily seen from the expression 2.46 and numerically computing
the discrete average value of the power terms in table 2.5, in order to evaluate the
enhancement of the electrical and mechanical energy, the important observables such
related to the root-mean-squared value of xrms and Vrms. These were numerically
simulated and in the next section will be shown and compared with experimental
results.
80
Chapter 3
Numerical Analysis And
Experimental Results
(2.33), since the oscillator was assumed to be driven by random excitation. In order
81
Chapter 3. &umerical Analysis And Experimental Results
of this method, we want to describe now the statistical properties of the vibration
noise source, the equations used and the structural parameters of the prototype
appendix A.1. Unlike the linear analysis of precedent works in which a vibration
source tuned to the natural frequency of the harvesting system was postulated, here
d ξ e (t ) 1
= − ξ e + σ g w (t ) (3.1)
dt τc
where Gw(t) is a Gaussian white noise with zero mean and δ-autocorrelation
g w (t ) = 0
(3.2)
g w (t ) g w (t ') = δ (t − t ')
and correlation time τc. While the autocorrelation function of process ξe is given by
with ξe2 = σ 2τ c / 2 .
inverted pendulum; therefore, we can start from the non-coupled prototype Duffing
x (t ) = ax(t ) − bx 3 (t ) − 2γ xɺ + σξ (t )
ɺɺ (3.4)
82
Chapter 3. &umerical Analysis And Experimental Results
stochastic differential equation SDE. Let the time interval [ 0,T ] be divided by &
points tj for j=1,..,& with & positive integer and t j = j (T / & ) = j ∆t , and, let the
above second order SDE be turned into a system of two first order differential
equations
Xɺ 2 = aX 1 − bX 13 − 2γ X 2 + σξ
ɺ (3.5)
X 1 = X 2
t t
t0 t0
t
(3.6)
X (t ) = X (t ) + X ( s)ds
1 1 0 ∫ 2
t0
Adopting the Euler-Maruyama method (A.11), we are now able to rewrite it in terms
code. A proper time step ∆t has been chosen sufficiently small to prevent the
summation from diverging at the end point t=T since from (A.12) the error results
e∆strong
t := Ε X & − X (T ) ≤ C ∆t1/ 2 . Furthermore, we accounted for a sampling error that
neglected the inherent errors in the random number generator and the floating point
roundoff errors.
83
Chapter 3. &umerical Analysis And Experimental Results
This choice for code language was made for various reasons: the ease of use, the
mathematics tools for signal analysis, random generators function and flexible
plotting procedures.
(The) listing 1 in the file wise_duffing.m (A.2) is structured as a function with the
following sections:
1. Declaration of constants
4. Plotting
5. Saving
The input parameters of the function: T, damp, Fa, Fb, sigma, tau represent
respectively the time interval, the damping coefficient, quartic potential parameters
(a, b), noise intensity and autocorrelation time of the exponentially correlated
Gaussian noise. Keeping fixed parameter b of the quartic well and varying a, we
examined the dynamics of the Duffing oscillator for different potential shapes and
noise strength σ. The parameters used for the first test are listed in table 3.1.
84
Chapter 3. &umerical Analysis And Experimental Results
Parameter Value
σ [0.025,1] with step 0.025
γ 0.05
m 1 [Kg]
T 100-500 [s]
∆t 10-4 [s]
τ 0.1 [s]
a [-1,1] with step 0.025
b 1
Table 3.1 –Parameters for Duffing model.
negative coefficient ‘a’ in fig. 3.1 and for its positive values in fig. 3.2. It is obvious
that the quartic well is similar to the parabolic one (fig. 2.12) while a<0, therefore,
the oscillator behaves like a linear spring-mass damped system and most of the noise
85
Chapter 3. &umerical Analysis And Experimental Results
a = -1
1
displacement
-1
0 50 100 150 200 250 300 350 400
time [sec]
a =-0.5
1
displacement
-1
0 50 100 150 200 250 300 350 400
time [sec]
a=-0.25
1
displacement
-1
0 50 100 150 200 250 300 350 400
time [sec]
a=-0.05
1
displacement
-1
0 50 100 150 200 250 300 350 400
time [sec]
Figure 3.1 – Displacement [m] Vs. Time at noise magnitude σ=0.1, γ=0.05 and a<0.
The well base becomes even more flattened as the ‘nonlinear’ parameter ‘a’
approaches to zero. According to the relation (2.44), The first vibration mode
parameter a is derived from the moment of inertia and the effective elastic constant
(2.32); thus, if we want to vary only this parameter taking b fixed, it is sufficient to
86
Chapter 3. &umerical Analysis And Experimental Results
a = 0.05
1
displacement
-1
0 50 100 150 200 250 300 350 400
time [sec]
a = 0.25
2
displacement
-2
0 50 100 150 200 250 300 350 400
time [sec]
a = 0.5
2
displacement
-2
0 50 100 150 200 250 300 350 400
time [sec]
a=1
2
displacement
-2
0 50 100 150 200 250 300 350 400
time [sec]
Figure 3.2 - Displacement Vs. Time at noise magnitude σ=0.1, γ=0.05 and a>0.
Displacement traces for positive values of a shown in fig. 3.2 give reason of the
intra-well jumps between the two minima. These jumps occur with even more low
frequency as the barrier height increases. Indeed, when the potential barrier grows,
the jump probability decreases exponentially according to Kramer’s escape rate [71]
given by
87
Chapter 3. &umerical Analysis And Experimental Results
where σ and τc are respectively the intensity and the autocorrelation time of the
excitation noise. After that, a critical value of barrier height has been crossed, the
system is constrained to oscillate mostly around one of the local minima with very
rare jumps. The variance of the random variable x accounts for the average
magnitude of vibrations around its mean value along a time series and it is given by
2
σ x2 = x 2 T
− x T
= xrms
2
− x2 (3.9)
This is an important quantity for analyzing the dynamics and energy balance, as we
will see later. The experimentally measured value of the variance (or rms if we have
zero mean) is thus affected by the measurement time. If the measurement time is
long enough, compared to the inverse of Kramer’s rate, then the measured rms
coincides with the rms of the equilibrium process. On the contrary, if the
measurement time is significantly shorter than the inverse of Kramer’s rate, the
system is confined around one local minima and thus a lower rms is measured.
Figure 3.3 below contains the measured standard deviation STD(x)=σx versus a at
series of & = T / ∆t = 500 /10 −4 = 5 ⋅106 samples for 100 values of a. We must point
out that the time step ∆t affects in a crucial way the numerical convergence of the
solution of the SDE. Taking a sufficient small integration step size ∆t is one of the
(3.4), but at the same time it is necessary to find a trade off with the computing time.
88
Chapter 3. &umerical Analysis And Experimental Results
0.7
σ = 0.025
0.6
σ = 0.05
σ = 0.075
0.5 σ = 0.1
0.3
0.2
0.1
0
-1 -0.5 0 0.5 1 1.5
a
This plot shows clearly the dynamical behavior discussed above. Three distinct
growing of the vibration amplitude when passing from the quasi-linear to the
2) a > 0. A raising peak in the nonlinear region around zero where the dynamics
3) a >> 0. The potential is bistable with a very pronounced barrier between the
minimum. As we have stated before, this depends on the length of the finite
89
Chapter 3. &umerical Analysis And Experimental Results
time series on which we measure the rms. In fact, for an infinite time interval,
the jumps between the two minima would not terminate and the variance
filtering effect, indeed, the low frequencies are primarily generated by intra-
Figures 3.4 and 3.5 show the x standard deviation Vs parameter a and
Figure 3.4 – STD(x) versus ‘a’ and square of noise strength σ=[0.05,0.2]
b =1 and autocorrelation time τc=0.1[s].
90
Chapter 3. &umerical Analysis And Experimental Results
Figure 3.5 – Top view. The solid black line traces the fitting curve a = 3 σ2 from
theoretical prediction of the maximum shift.
The top view of the 3D plot 3.4 clearly shows the evolution of the maximum σx. The
solid line is a theoretical prediction obtained with the following argument. The root
contributions: i) the raising, mainly due to the growth of the separation between the
two minima; ii) the drop, mainly due to the decrease in the jump probability
measured by the crossing probability defined in (3.13), caused by the increase of the
potential barrier height ∆U. The solid black line remarks the dependence of the
maximum from parameters a and σ whose form is explained in the next theoretical
analysis. Since in a purely open circuit case the output voltage Vp of a Duffing
equations (2.33) and (2.34), the larger the standard deviation σ x = STD ( x ) is, the
larger average extracted voltage Vp is. Indeed, for the open circuit case we get
91
Chapter 3. &umerical Analysis And Experimental Results
Kp K
x 2 − x 2 = p σ x2
2
σ V2 = Vp2 − Vp = (3.10)
p
Cp Cp
The remarkable result shown by fig. 3.2-3.3-3.4 is that the vibration magnitude in
nonlinear region is six times larger than that in a linear region for some value of
noise amplitude. The increasing of the average mechanical and converted electrical
power output will follow the same trend since they are proportional to the root mean
square of the displacement and electric potential. These results will be discussed in
An attempt of theoretical interpretation of the curve in fig. 3.2 and 3.3 could be
made starting from qualitative considerations and statistical properties of the bistable
system under observation. First of all, since it is simulated and measured in a finite
time interval, we cannot consider a stationary state persisting for an infinite time.
Therefore, the rate of intra-well jumps in a bistable regime depends on the rate
between Kramer’s time and the total interval of time ∆T over which the dynamics is
generated. For a<<0 we identify quasi-linear behavior or RMS of x that can be easily
characterized; for a>>0 we are in an entrapment region of x around one of the two
minima and even in this case the dynamics is quite similar to that of a linear
oscillator oscillating close to its stable equilibrium position. Instead, for the intra-
well jumping region at positive a not far from zero, the interpretation appears more
92
Chapter 3. &umerical Analysis And Experimental Results
complex. So we can try to approximate the variance σx using the root mean square of
the state variable x for a stationary bistable system[71, 72] driven by a additive
random noise, multiplied by Kramer’s jump rate (3.8). The approach is the
following.
2 2
σ x2 = x 2 st
− x = xrms
2
; st − x (3.11)
∫ x P ( x)dx
2
st
2
x
rms ; st = x 2
= −∞
∞
(3.12)
st
−∞
∫ P ( x)dx
st
1 a b
Pst ( x) = & 0 exp(−U ( x) / σ ξ2τ c ) = & 0 exp 2 x 2 − x 4 (3.13)
σ ξ τ c 2 4
with &0 as a normalization constant. Assuming positive values for a,b, σ ξ2τ c and
I 3 ( r ) + I 1 ( r ) + I 1 ( r ) + I 3 (r )
a − −
x 2
= 4 4 4 4
(3.14)
st
2b I 1 (r ) + I 1 (r )
−
4 4
with r expressed by
1a 2 1 ∆U
r= = (3.15)
2σ ξ2τ c 2b 2 σ ξ2τ c
which represents the half ratio between barrier height and noise strength.
93
Chapter 3. &umerical Analysis And Experimental Results
the square position of the minimum x±2 = a / b , keeping the other parameters fixed.
2
To make qualitative considerations we equate xrms ≈ x±2 as a first approximation.
Moreover, by virtue of the arguments discussed above, we can consider the squared
Pjump and x±2 , because, as the intra-well jumps decrease, the oscillations will happen
even more around one of the two local minima. Otherwise, this mean will be null for
2 2
x T
≈ (1 − Pjump ) x± (3.16)
2
σ x2 = xrms
2
;st − x ≈ & x±2 − (1 − Pjump ) x±2 = & ⋅ ( Pjump x± ) 2 (3.17)
and, computing the jump probability times x±2 for the Duffing potential, it results
a 1 a2
σ x2 ≈ x±2 ⋅ & ⋅ Pjump = ⋅ & exp (3.18)
σ ξ τ c 4b
2
b
where & is a constant that can be properly chosen in order to fit the numerically
simulated curve σ x2 = f ( a; b, σ ξ ,τ c ) .
For the sake of identifying the dependence of the maximum position, we compute
∂σ x
= 0,
∂a
∆U ∆U (3.19)
1 − 2
σ τ a a 1 − σ ξ2τ c
e ξc+ − e =0
2 ab b 2b σ ξ2τ c
94
Chapter 3. &umerical Analysis And Experimental Results
a = bσ ξ2τ c (3.20)
As it can be seen from the solid line in figure 3.5, the theoretical prediction
bistable system our idea is based on. This model, anyway, does not yet incorporate
the same designed in fig. 2.9; so, we have to simulate the two coupled equations
(2.34) with coefficient relationships (2.35) in the case of a pure resistive load. The
model geometry is the same exposed in figure 2.15. The governing equations are
bistability.
Note that in the case of the inverted pendulum this term arises from the gravity
force applied on the mass, therefore, the coefficients Fa and Fb are function of mass
and moment of inertia, hence, they are not uncoupled. It is not possible to vary these
95
Chapter 3. &umerical Analysis And Experimental Results
unpractical to use. Nevertheless, apart from the gravity, an other conservative force
directed toward the clamping base can reproduce the repulsive effect that
counterbalances the elastic restoring force. This mechanism can be realized by using,
for example, elastic spring, electrical or magnetic field. Thus, for now, in order to
study the dynamical and electrical behavior more generically, we have opportunely
simulation much easier. After a little algebra, we can express the new form with
reduced coefficients:
d 2 xɶ (ɶt ) ɶ
ɶ dxɶ (t ) − ΘVɶ (tɶ ) + σξɶ (tɶ )
ɶ2 = axɶ (tɶ ) − bxɶ (tɶ ) − ζ
3
dt dtɶ
(3.21)
dVɶ (tɶ ) dxɶ (tɶ )
dtɶ = Θ − ΩVɶ (tɶ )
dtɶ
ξ (tɶ ⋅ t )
ξɶ (tɶ ) =
ξ
x(tɶ ⋅ t )
xɶ (tɶ ) = (3.22)
x
V (tɶ ⋅ t )
Vɶ (tɶ ) =
V
ɶt = t / t
96
Chapter 3. &umerical Analysis And Experimental Results
t = m/ K
ξ = K / m⋅ x
x Kv
=
V Kc K
a = Fa − 1
Fb 6
b= con b = 2 (3.23)
b x
ζɶ = 2ζ
Kv Kc
Θ=
K
t m / K τ 0 ωC
Ω= = = =
RLC p RL CP τ C ω0
where Ω represents the ratio between the piezoelectric cutoff frequency ωc and
the natural frequency at short circuit ω0 = K / m in the limit of linear oscillator that
is for a=-1 and b=0. This influences the electrical power dissipated on load. So, Ω is
the piezoelectric element works as an high pass filter and, as it will be shown next,
the bistability pushes the power absorption from noise towards low frequencies.
relatively high, for a high bistability it brings a loss of electrical power transferred to
the situation of infinite load resistance for real model or infinite capacitance and the
strength and derives from the electrical and geometrical piezo-bender characteristics.
97
Chapter 3. &umerical Analysis And Experimental Results
equations (2.34), where the first is multiplied by mass and velocity and the second
one by the electrical potential, then, using the set (3.23) we obtain the following
The resulting mean power terms and efficiency are simulated for a set of plausible
parameters listed in table 3.3. Simulations were made simultaneously running many
time series for equal and/or different parameter space on a 20-nodes computer
cluster. Then, averaging the statistical observables of the same constants and
parameters series.
98
Chapter 3. &umerical Analysis And Experimental Results
parameters
σ 0.01– 0.2 step 0.0025 noise strength
tɶ 0 – 2000 step 0.001 normalized time
ζɶ 0.02 – 0.4 damping
a (-1,1) step 0.01 potential parameter
b 1 potential parameter
Ω 0.011–1 dimensionless cutoff frequency
Θ 0.336 coupling coefficient
macroscopic model similar to our prototype test model (exposed in the next
paragraph), but these can be varied in order to investigate many different geometry
and material properties. Let us give a glance of some displacement and voltage
simulated outlines that show in time domain the dependency between them as a
99
Chapter 3. &umerical Analysis And Experimental Results
1
a = -1
displacement
0.5
-0.5
-1
0 50 100 150 200 250 300 350 400 450 500
normalized time
0.4
a = -1
0.2
voltage
-0.2
-0.4
0 50 100 150 200 250 300 350 400 450 500
normalized time
2
a = 0.8
displacement
-1
-2
0 50 100 150 200 250 300 350 400 450 500
normalized time
1
a = 0.8
0.5
voltage
-0.5
-1
0 50 100 150 200 250 300 350 400 450 500
normalized time
Figure 3.6 – Displacement and voltage traces of PDG model with Ω =0.014
for linear case (upper) a= -1, and strong nonlinear (lower) a=0.8, ζ =0.1, σ =0.2.
ɶ
Note that at equal noise intensity the bistable regime shows an amplified
oscillation due to the mix of intra-well jump and vibration close to each local
minimum. Unless a little discharge effect, the voltage trace follows with almost
direct proportionality the displacement of the bender. This result is quite obvious
considering the second linear equation (3.21), because, a relative cutoff frequency
100
Chapter 3. &umerical Analysis And Experimental Results
2
norm. displacement
a = 0.4
-2
0 50 100 150 200 250 300 350 400 450 500
normalized time
0.2
norm. voltage
-0.2
0 50 100 150 200 250 300 350 400 450 500
normalized time
This high pass filtering effect is straightway visible in figure 3.7, where a
frequency Ω = ωc / ω0 > 1 was chosen and, jumps apart, the voltage bistability
completely disappears. Even more clearly, the following plots 3.8 and 3.9 remark a
can be seen from these figures, while the position variance does not experience an
strongly depends by it. In fact, for Ω <<1 the power peak in the bistability region
(a>0) is maximized like that of variance, whereas, for even more Ω ~1 close to one,
it drops and there is no difference in harvesting efficiency between the linear and
101
Chapter 3. &umerical Analysis And Experimental Results
0.7
Ω = 0.01
0.6 Ω = 0.1
Ω = 0.5
0.5 Ω=1
2
0.4
<x >-<x>
2
0.3
0.2
0.1
0
-1 -0.5 0 0.5 1 1.5 2
a
0.035
Ω = 0.01
0.03 Ω = 0.1
Ω = 0.5
0.025 Ω=1
Normalized Power
0.02
0.015
0.01
0.005
0
-1 -0.5 0 0.5 1 1.5 2
a
Figure 3.9 –Mean electrical power computed as RMS normalized voltage (see tab. 3.2)
at different ratios Ω and for ζɶ =0.1 and σ =0.1.
102
Chapter 3. &umerical Analysis And Experimental Results
0
10
-1
10
Efficiency
-2
10
Ω = 0.01
Ω = 0.1
Ω = 0.5
Ω =1
-3
10
-1 -0.5 0 0.5 1 1.5 2
a
Figure 3.10 – ormalized conversion efficiency
at different ratios Ω and for ζɶ =0.1 and σ =0.1.
highly desirable to project the piezoelectric device and harvesting circuit with an
Ω << 1 as low as possible. Also the noise spectrum and, therefore, its autocorrelation
time is an important parameter to know. Indeed, the lowest τ ξ is, the more noise
energy lies at high frequencies, probably far away from the cutoff frequency for a
certain design. This second optimization condition can be defined for dimensional
bistable system to this parameter together with noise strength σ ξ is quite complex
and needs to be further investigated. Actually, the probability of diffusion over the
intra-well barrier depends on the product σ ξ2τ ξ and we must bear in mind that the
system is not linear, so that its behavior changes with the noise strength. As it was
103
Chapter 3. &umerical Analysis And Experimental Results
2
seen in the previous paragraph 3.1, the position variance x2 − x of a bistable
strength and autocorrelation time make possible the diffusion over the potential
displacement x for a bistable system is a well known phenomenon, but looking at the
next plots 3.10-3.11 it is remarkable how much the harvested electrical power gets
104
Chapter 3. &umerical Analysis And Experimental Results
105
Chapter 3. &umerical Analysis And Experimental Results
It is noticeable that, for a fixed noise intensity σ 2 , bistable sector (a>0) shows up
an enhancement around a factor five-six of the harvested power with respect to the
monostable zone (a<<0). In terms of conversion efficiency (fig. 3.12) defined as the
ratio between electrical power dissipated on load and input power supplied by the
mechanical noise (as described in table 3.2), we have the same amplification in
nonlinear zone. In this plot it is better visible that the peak shifts as the square root of
A spectral analysis helps us to confirm what we have so far stated and permits to
frequency for different potential shape. At a fixed RMS of the Gaussian noise σ =0.1
and b=1, varying the potential parameter -1<a<1, we performed the spectra of
in figure 3.15 and 3.16. When a=-1, the conservative potential is parabolic-like and
106
Chapter 3. &umerical Analysis And Experimental Results
for small vibration close to equilibrium the oscillator behaves like a linear harmonic
oscillator. As the potential well becomes even more flattened as a∼0 approaches to
zero, the peak of resonance moves toward low frequencies and widens.
In the range of bistability (a>0), until the barrier height is relatively small, the
intra-well jumps are frequent and we get a broadening of the power peak down to
very low frequency in both position amplitude and voltage. As a consequence within
107
Chapter 3. &umerical Analysis And Experimental Results
a frequency band ∆ωɶ the mean converted electrical power increases being equal to
2
the integral of square voltage amplitude Pel (∆ωɶ ) = ∫ω Vɶ (ωɶ )
∆ɶ
d ωɶ . This is directly
reconfirmed by the previous graphs 3.12-14 where the average power was computed
in time domain. The entrapment of the system within one of the two potential
minima roughly happens for a≥0.5 and depends on the choice of factor σ 2τ ξ as we
have previously mentioned. After that, increasing a the two wells become even more
narrow, the system restarts to become harmonic-like and the amplitude peaks of
displacement and voltage shift toward higher frequency. Thus, the harvested power
quickly goes down as the subtended area becomes even more narrow.
conjunction with a piezoelectric transducer to show the effectiveness of the key idea
on improving the energy harvesting from vibrations. But, by virtue of theoretical and
system. Let us discuss the experimental observations of such a model compared with
108
Chapter 3. &umerical Analysis And Experimental Results
magnetic tip coupled with magnetic coil or permanent magnet placed at a certain
distance. Setting properly the magnetic dipole alignment and the distance or field
force. The physical model has been exposed in paragraph 2.4 and the geometry is the
same drawn in picture 2.16. The choice of a vertical pendulum configuration was
made to reduce at the minimum the influence of the gravity force on inertial mass
interaction rather than gravity which can be neglected in this case (the inertial mass
as regards to Duffing oscillator to be able to control only one parameter (for example
the relative distance between magnets) to tune the potential and regulate the
small scale yet. Indeed, at sub-millimetric size and small masses, the gravity force is
Now, we’re going to describe the real macroscopic model used for the
experimental tests, the experimental setup and the comparison of numerical and
experimental results.
109
Chapter 3. &umerical Analysis And Experimental Results
The following picture shows a schematic draw of the apparatus employed in the
experiment. Here are described the piezoelectric bender, the excitation system and
xy micrometric stage
tip magnets
∆
xm tip displ.
reading displ. xr
CCD Laser
displacement sensor
Ch1 displ.
excitation magnets “x”
clamping
Vout - Piezoelectric
T434-A4-302 4-Layer Bender by Piezo system inc.) made of Lead Zirconate Titanate
110
Chapter 3. &umerical Analysis And Experimental Results
(PSI-5A4E) 40 mm long, clamped at one end. The pendulum base position can be
adjusted via a micrometric xyz displacement system. The pendulum mass is a 60mm-
long steel beam with three magnets (dipole magnetic moment M=0.051±0.002 Am2)
attached. The tip magnet is faced by a similar magnet with inverse polarities placed
b)
a)
frequency 1Mhz (National Instruments interface DAQ). The output voltage signal
placed in parallel with output contact of beam and sampled by the DSPB. The
111
Chapter 3. &umerical Analysis And Experimental Results
elaboration. The DSPB is also employed to drive a current generator that produces,
through a couple of coils, the magnetic excitation that mimics the ground vibration.
The signal generated by the DSPB is filtered and conditioned in order to reproduce
the desired statistical properties. The design parameters of the test model are
112
Chapter 3. &umerical Analysis And Experimental Results
Some of them are derived from the relations (2.35) and modal analysis, others
from the technical data of the piezoelectric material and by measurements. The
damping factor ζ has been extracted from averaging a series of ring down
measurements carried out at open circuit. Figure 3.16 shows a sample of fitted ring
1,0
Ring down
0,8 ζ= 1/ω0τ = 0,016 Fit
f0= 6,67 Hz
Data: QFACTOR_B
0,6 Model: ExpDec1
Chi^2/DoF = 0.00055
rms (volt)
R^2 = 0.99155
0,4 y0 0 ±0
A1 0.8543 ±0.02275
t1 1.46707 ±0.07524
0,2
0,0
0 2 4 6 8 10 12
Time (s)
1000
2
900 Fit ω =k(1/m) k=27,09 (err 0,8)
800
700
600
ω (s )
-1
500
2
400
300
200
25 30 35 40 45 50 55
-1
1/m (Kg )
113
Chapter 3. &umerical Analysis And Experimental Results
The effective elastic constant and mass was retrieved by modal analysis, that is
varying the inertial mass attached to the piezo-bender and measuring the variation of
The dynamical behavior of the real model and powering speculations are
supported by finite element analysis carried out both on magnetic force interaction
the magnetic forces between two real permanent magnets in relative motion at
different configuration and angles, we first computed the axial repulsive force with
2D finite element analysis of the magnetostatic field and then searched a suitable
fitting function.
100
90 Data: Br33dic2_B
Model: magnetic_force
80 Equation: 6e-7*(P1^2)/abs((P2+x))^P3
Weighting:
70 y No weighting
60 Chi^2/DoF = 13.94691
R^2 = 0.91776
force (N)
50
P1 0.05 ±0
40 P2 0.00562 ±0.00038
P3 4.83888 ±0.0531
30
20
10
0
-10
0,000 0,005 0,010 0,015 0,020 0,025
∆ (m)
114
Chapter 3. &umerical Analysis And Experimental Results
Figure 3.20 on the left shows the magnetic field flux streamline and intensity in
air domain computed via finite element modelling for one of the many
configurations, whereas on the right it shows the fitting of repulsive force data along
the vertical axis for magnets with aligned dipoles using the relation (2.37). We must
specify that this analytical relation is valid for magnetic force between two aligned
dipoles at relative long distance (r>>d where d is the linear dimension of a magnet),
whereas for short distance the proper law approximately goes as the inverse square of
the distance [68]. Anyway, for small oscillations and at medium relative distance ∆,
the fourth order polynomial function seems to fit better the FEA and experimental
data than the inverse square law. Actually, we should also consider the term which
accounts for the angle between the two magnetic dipoles, but in first instance it is
possible to neglect it. The resulting dynamical behaviour and modal analysis of our
model has been tested even with mechanical FEM of the piezo-pendulum.
f0=6.58 Hz
first vibrational mode
20000 degree of
freedom
Figure 3.21 – FEM and tip displacement Vs time from monostable to bistable dynamics.
115
Chapter 3. &umerical Analysis And Experimental Results
In figure 3.21 there are plots of dynamical response of the pendulum subjected to
distance ∆∼10mm) before the total trapping around one stable position. The
smaller distance is clearly visible. The mechanical power density spectrum (fig. 3.22)
of the two extreme situations: quasi-linear behavior for negligible repulsive force
Figure 3.22 – Mechanical power spectrum for ∆>20mm (norm. barrier height w=0.001)
and ∆∼10mm (w=0.1).
Evidently, the subtended area of the bistable power spectrum is much more than
that of linear case at low as well as at high frequencies, hence, the finite element
116
Chapter 3. &umerical Analysis And Experimental Results
model further confirms the efficiency enhancement on power transfer to the system
analysis
nalysis has been carried out on two signals: displacement signal x (channel 1) and
piezoelectric output voltage Vout across the resistive load RL (channel 2) made by ad
hoc LabView program. Whereas the noisy input force is generated via software
controlled NI-DAQ
DAQ (channel 3) and then the signal conditioned by pass-band
pass filter in
figure 3.23 below.. We have conjectured as a first step this kind of uniformly
distributed force in frequency domain for two principal reasons: we want to speculate
on how much energy could be extracted form a wide spectrum and the most part of it
117
Chapter 3. &umerical Analysis And Experimental Results
The displacement variance and RMS voltage were computed on time interval ∆T of
piezo_magnetic.m listed in (A.2) seems to be very good. Indeed, as we can see from
fig. 3.25, the experimental distribution of moving position x of the mass is very close
to the one expected from the numerical simulation: say within the relative
experimental error: about 7-8% in both axis directions (error bars are not plotted to
avoid confusion due to the large number of experimental points). This plot reports
the distribution of moving mass for effective driving noise rms σ=1.2 mN at three
and high intra-well barrier height potential when the pendulum is bounded around
one minimum. The relative maxima of displacement distribution (fig. 3.24) perfectly
follow the stable equilibrium positions corresponding to the potential energy (fig.
3.24b). This is derived from the expression (2.38) slightly adjusted considering some
geometrical scaling factors which connect the various position points (x, xr, xm) of the
s2 = (lm + lb) / lb .
118
Chapter 3. &umerical Analysis And Experimental Results
Y b)
∆ xm
lm
xr
CCD
lxr
x
lb
X
a)
0,7
∆ = 7 mm
0,6
∆ = 10 mm
∆ = 24 mm
0,5
0,4
distribution
0,3
0,2
0,1
0,0
-15 -10 -5 0 5 10 15
displacement (mm)
119
Chapter 3. &umerical Analysis And Experimental Results
Finally, the most important confirmation of what we have claimed so far, that is the
fig. 3.26 we plot the standard deviation σ x computed on values of the stochastic
motion (solid line) and by the experimental data (scatter) for three different rms of
effective noisy force ( σ =0.3,0.6,1.2 mN) applied to effective mass of the oscillator.
The agreement between the experimental data and the model is very significant.
8 σ = 1,2 mN
σ = 0,6 mN
7 σ = 0,3 mN
σx (mm)
0
5 10 15 20 25
∆ (mm)
120
Chapter 3. &umerical Analysis And Experimental Results
squared voltage Vrms (fig. 3.27) across the resistive load RL (100 MOhm for this
test) show the same trend with the same maximum peak position as well as σ x .
6
σ = 1,2 mN
Vrms (Volt)
σ = 0,6 mN
5 σ = 0,3 mN
0
5 10 15 20 25
∆ (mm)
Figure 3.27 - – (Scatter) experimental data of voltage RMS across the load
Vs magnets vertical distance ∆. (solid) numerical prediction.
position of the pendulum. This condition accounts for the usual performances
121
Chapter 3. &umerical Analysis And Experimental Results
2) when ∆ is very small, the potential energy is bistable with a very pronounced
barrier between the two wells. In this condition and for a fixed amount of
noise, the pendulum swing is almost exclusively confined within one well and
condition, the pendulum dynamics is highly nonlinear and the swing reaches
its largest amplitude with noise assisted jumps between the two wells.
of the rms exceed by a factor that ranges between 2 and 6 the value obtainable
when the magnet is far away and thus the pendulum operates in linear condition.
This behavior amounts at a potential gain for energy harvesting between 200%
and 600% compared to the standard linear oscillators, depending on the noise
intensity and on the other physical features of the pendulum. Moreover we would
like to briefly comment about two other important features: a) the maximum
position shifts toward larger distances ∆ when the noise intensity increases. b) In
122
Chapter 3. &umerical Analysis And Experimental Results
the low ∆ regime, the rms values seem to reach a plateau that is smaller than the
40
35
σ = 1,2 mN
σ = 0,6 mN
Electrical Power (10 watt)
30 σ = 0,3 mN
-7
25
20
15
10
5 10 15 20 25
∆ (mm)
These features further support all the considerations presented in chapter 2 about
the stochastic dynamics and potential energy. In particular, when ∆ is very large the
∆=30mm). Both the measured value 6.6±0.1 Hz and that simulated by FEM
123
Chapter 3. &umerical Analysis And Experimental Results
increases. The potential becomes bistable when ∆ is lower than the critical value
( 3K / K eff )
2/5
x± = ± m − ∆ 2 separated by a maximum at x=0. The agreement between
the analytical and experimental values is almost perfect: for example, at ∆=0.007mm,
analytical minima results xr± =±0.0067 (where the modified formula with geometrical
( 3K s 2 / K eff )
2/5
ratios xr± = ±( s1 / s2 ) m − ∆2 has been used in fig. 3.29 and
experimental ones are xr± =±(0.0067±0.0001). From this point, the oscillator jumps
now between the two minima and the σ x increases proportional to xr± . As
124
Chapter 3. &umerical Analysis And Experimental Results
The continuous growth of the σ x is interrupted when, due to the increase of the
barrier height, the noise induced jump dynamics between the two minima becomes
less and less effective. By further decreasing ∆ the barrier height is so large that the
confined within one well. Such a trapping condition happens at smaller values of ∆
(larger barrier) when the noise is larger. This explains the observed shift of the
maximum position toward smaller ∆ as observed in a). Here the dynamics is almost
linear with small oscillations around the potential local minima (x+ or x-) with σ x
All these considerations about the nonlinear stochastic dynamics are inline with
those previously exposed for the Duffing piezoelectric oscillator. The sole difference
is the expression of the conservative potential. The role of the distance ∆ is played in
the Duffing case by the parameter a. Basically, the three dynamical regimes are the
same as well as the noticed efficiency enhancement. Also in this case, we notice a
2
maximum values of the Vrms a factor 3-6 grater than the value obtainable in quasi-
harmonic domain.
In the next chapter we will synthesize again the key results of the thesis subject
and take a glance to the possible solutions based on this concept. The implementation
designs and miniaturization issues are also outlined with some design example
125
Chapter 4
Conclusions
4.1Effectiveness of nonlinear
nonlinear approach versus
linear
In spite of the fact that a growing interest and many studies have been recently
focused on vibration energy harvesting, most of these are left restricted within the
these inspections have been based on the maximization of the kinetic energy transfer
around its first resonant mode and of electro-mechanical coupling constants. This
126
Chapter 4. Conclusions
common approach is clearly optimal for a well known shaking source whose energy
is primarily centered around single frequencies, however, most of the power of many
random vibration sources is spread over a large spectral bandwidth and most of all at
frequencies under hundred hertz [10, 31, 38, 61, 74, 75]. This is also the reason
dimension with a consequent mismatch between the driven frequency and natural
frequency of the device. Some attempts to design energy scavengers with a wider
bandwidth under investigation concern the use of a chain of & different spring-mass-
dampers [61]. The various oscillators in this sort of chain have resonant frequencies
almost overlapped such that at least one element is in resonance over the desired
frequency range. Others groups are improving the signal processing and storing
damping[76]). By the way, both these approaches present severe limitations such as
the scale reduction, dispersive control electronics and poor efficiency in the case of
wide band multimodal excitation. Moreover, the linear model elaborated by Williams
et al. [50] can describe quite well the behaviour of electromagnetic transducer whose
efficiency depends basically by the velocity of inertial mass, whereas, it’s not
completely valid for piezoelectric beam conversion technique that needs to maximize
its displacement and to reduce its cutoff frequency as much as possible in order to
The pivot idea we’ve presented here concerns the use of nonlinear oscillators
instead of linear ones. The exploitation of the their dynamical features permit to
127
Chapter 4. Conclusions
strongly improve the absorption of the ambient vibrational energy from a wideband.
Our work has been focused in particular on the use of a Duffing-like bistable
oscillator coupled with piezoelectric element, but the key principle can be explored
stochastic oscillators.
grater than a factor six with respect to a linear oscillator concept. The
maximum peak of the generated power is obtained for a certain trade off
among noise variance, intra-wells distance and barrier height, while the
energy filtering.
frequency, but it’s able to extract kinetic energy from a wide bandwidth
128
Chapter 4. Conclusions
Early theoretical considerations have been done about the power peak
systems cover a wide range of physical and biological phenomena so that they are
under crescent interest far beyond their application as energy harvesting methods.
investigations, even though these studies are mainly focused on materials research
and various kinds of microlithography techniques [57, 77]. Our concept can be easily
applied to every system even at small scale. Indeed, while the physics remains
regulated under the classical mechanics laws, our results could be easily transferred
inspected even in quantum domain, but this lies outside the scope of this work.
129
Chapter 4. Conclusions
4.2Miniaturization perspectives
perspectives of nonlinear
power harvesting systems
Finally we would like to emphasize that the results obtained here can led to a
etching, wet chemical etching and UV-LIGA etc.) a crescent number of research
systems for energy scavenging. For example, Jeon et al.,[40] developed a 170µm x
260µm PZT beam that is able to generate 1µW of continuous electrical power to a
5.2MΩ resistive load at 2.4V dc. Although, the corresponding energy density
obtained under a driving force precisely tuned on its first resonant frequency of
a) b)
Figure 4.1 – a) SEM of the fabricated PMPG device with bond pads (Jeon [40]).
b) Micrograph of a 500 µm PMPG of TIMA labs, Ammar et al. [78]
130
Chapter 4. Conclusions
design has been developed by Fang et al.[79], which results in about 2.16µW (0.89V
silicon generator with a 300µm wide paddle beam that gives a natural frequency of
6.4 kHz. Although its output voltage of 0.36V could be amplified, even in this case,
the actual power density is not still promising. Going down to more tiny scales,
at the interface between atomic force microscope tip and the ZnO NW.
vessels, acoustic waves, flowing blood or a beating heart. With some adjustments,
the nonlinear dynamical concept shown here can be quite easily applied to all these
present linear designs, even down to nanoscale, in order to make them working on
been yet realized and studied for other purposes by some research groups [80, 81].
fig. 2.16, it’s not difficult to tailor a suitable bistable potential to the bandwitdth of
131
Chapter 4. Conclusions
permanet magnets
stochastic force
piezo beams
+ − + − + − + −
Vout
Iout
stochastic force
permanent
magnets
piezo beams
In the picture above shown there are two possible configuration of permanent
while in b) the piezo-beams counteract each other their elastic restoring force via
132
Chapter 4. Conclusions
piezoelectric membranes. Each one with an appropriate inertial mass attached to the
centre. The bistable dynamics could be created, without accessory magnets, simply
contracting one of its edge on its parallel plane as outlined for example in fig.4.3.
stochastic force
behaviour through finite element modelling of such idea. A 20mm length x 0.1 mm
thickness PZT membrane model with a steel sheet inside of thickness 0.05mm is
plotted in fig. 4.4. It was compressed setting its vertical edge displacement by 0.01-
(sinusoidal and/or stochastic) pointed to its center, for different compression levels of
the edges, we’ve computed the time series distribution of central point displacement.
settle the symmetry of bistability. This design is similar to that of a pressure sensor
and in effects it can works both as generator/sensor and actuator, with the essential
difference that, in this case, its dynamic is bistable and not monostable (see fig. 4.5).
133
Chapter 4. Conclusions
between the two sides without encounter the typical problems of calibration of
134
Chapter 4. Conclusions
∆P = 100mbar
∆P = 1000mbar
∆P = 2000mbar
1,0
0,5
displacement (mm)
0,0
-0,5
-1,0
0 2 4 6 8 10
Time (s)
Actually, one of the major difficulty of such a kind of model thought as energy
harvester is to know exactly the elastic potential which is not perfectly conservative.
could be a promising and simpler solution than magnetic piezoelectric cantilever and
135
136
Appendices
A.1 Euler-
Euler-Maruyama method
dX t
= F ( X t , t ) + G ( X t , t )ξt (A.1)
dt
the deterministic evolution and the term G ( X t , t )ξt is a stochastic Gaussian term with
t t
X t = X 0 + ∫ F ( X s )ds + ∫ G( X s )ξ s ds (A.2)
0 0
∫ ξ ds = W
0
s t (A.3)
137
1) W (0) = 0 with probability 1;
From this relations the stochastic term of the (A.2) can be expressed as
Wt
∫ G( X )dW
0
s s (A.5)
n
S n = ∑ Gτ ( n ) (Wτ ( n ) −Wτ ( n ) ) (A.6)
i i i −1
i =1
Where τ i( n ) ∈ ti(−n1) , ti( n ) . But the summation limit depends by the choice of these
criterion. Among the various consistent methods of integration the most successful
where the Itô integration corresponds to equal α=0, while the Stratonovich
the equation (A.2) on Brownian process W (t ) , thus, in the Wiener increment dW(t)
138
t t
X (t ) = X 0 + ∫ F ( X s ( s), s)ds + ∫ G ( X s ( s ), s)dW ( s) (A.8)
0 0
these two integrals can be discretized as follows. Let a time interval [ 0,T ] be divided
by & points tj for j=1,..,& with & positive integer and t j = j (T / & ) = j ∆t .Then, as
& −1 & −1
∑ F ( X (t j ), t j )(t j +1 − t j ) = ∑ F ( X (t j ), t j )∆t
j =0 j =0
(A.9)
& −1 & −1
∑ G ( X (t j ), t j )(W j +1 − W j ) = ∑ G ( X (t j ), t j )dW j
j =0 j =0
(A.10)
& −1 & −1
X (T ) = X 0 + ∑ F ( X (t j ), t j )∆t + ∑ G ( X (t j ), t j ) dW j . (A.11)
j =0 j =0
Ε X n − X (τ ) ≤ C ∆t γ (A.12)
for any fixed τ = n∆t ∈ [ 0, T ] and ∆t sufficiently small. Keeping in mind that X (τ )
is the possible analytic value and Xn is the simulated variable. It can be shown that
139
A.2 Simulation programs
Listing n.1
File: wise_duffing.m
function wise_duffing(serial,T,damp,Fa,Fb,sigma,tau,plt1,plt2,sav)
try
tic
% dependent parameters
epsilon=zeros(1,N);
for g = 1:N
epsilon(g+1) = epsilon(g)*(1-h/tau) + randn*h;
end
epsilon = epsilon/std(epsilon);
epsilon = epsilon - mean(epsilon);
for k=1:length(sigma)
140
pos = 0.0;
Y(1)= 0.0; % initial conditions
for j=1:length(Fa)
X(1)= pos;
Y(1)= 0.0;
for i=1:N
Y(i+1)=Y(i)+Fa(j)*X(i)*h-Fb*X(i)^3*h-2*damp*Y(i)*h...
+sigma(k)*epsilon(i)*dW(i);
X(i+1)=X(i)+Y(i)*h;
end
pos=X(length(X));
if plt1
figure(1);
clf
plot(t,X,'r')
hold on
xlabel('time [sec]');
ylabel('displacement');
end
end
end
%% plotting
if plt2
figure(1);
clf
plot(Fa,Std_x);
xlabel('nonlinear parameter');
ylabel('Xrms');
figure(2)
surf(sigma.^2,Fa,Std_x,'EdgeColor','none');
xlabel('sigma^2');
ylabel('nonlinear parameter');
zlabel('Xrms)');
end
%% saving in Pavg.mat
if sav
end
141
toc
catch
lasterror
end
Listing n.2
File: piezo_magnetic.m
function []=piezo_magnetic(T,damp,mag,dy,scale,sigma,plt1,plt2,sav)
try
tic
%% CONSTANTS DECLARATION
mu0=4*pi*1e-7;
142
Tp=0.33e-3; % thickness of piezoelectric layer
Tsh=0.2e-3; % Thickness of the center shim
thk=Tp*2+Tsh; % thickness of the beam
% dependent parameters
epsilon=zeros(1,N);
tau=0.1; % colored noise correlation time [sec]
143
for g = 1:N
epsilon(g+1) = epsilon(g)*(1-h/tau) + randn*h;
end
epsilon = epsilon/std(epsilon);
epsilon = epsilon - mean(epsilon);
sf2 = (lb+lm)/lb*scale;
k_mag = 3*mu0/(2*pi)*mag^2*sf2;
% initialization
for k=1:length(sigma)
pos = 0.0;
Y(1)= 0.0; % initial velocity
V(1)= 0.0; % initial voltage
for j=1:length(dy)
Z(1)= pos;
Y(1)= 0.0; % initial velocity
V(1)= 0.0; % initial voltage
for i=1:N
Y(i+1)=Y(i) -
(K/m)*Z(i)*h+3*mu0/(2*pi*m)*mag^2*Z(i)*sf2/((sf2^2*Z(i)^2+(dy(j))^2)^2.5)*h-
2*damp*sqrt(K/m)*Y(i)*h...
-(Kv/m)*V(i)*h+sigma(k)*epsilon(i)*dW(i)/m;
Z(i+1)=Z(i)+Y(i)*h;
V(i+1)=V(i)+Kc*(Z(i+1)-Z(i))-1/(Rl*Cb)*V(i)*h;
end
pos=Z(length(Z));
Z = Z*sf;
[dist,xout]=hist(Z,-15e-3:0.1e-3:15e-3);
dist=dist/sum(dist)*20;
if plt1
figure(1);
clf
subplot(2,1,1);
plot(t,Z,'r')
hold on
xlabel('time [sec]');
144
ylabel('displacement');
subplot(2,1,2);
plot(t,V,'k')
hold on
xlabel('time [sec]');
ylabel('voltage');
end
Pavg(j,k) = std(V);
Pavg_z(j,k)= std(Z);
end
end
%% plotting
if plt2
figure(1);
clf
subplot(2,1,1);
plot(t,Z,'r')
hold on
xlabel('time (s)');
ylabel('displacement (m)');
subplot(2,1,2);
plot(t,V,'k')
hold on
xlabel('time (s)');
ylabel('voltage (V)');
hold off
figure(2);
clf
subplot(2,1,1);
plot(dy,Pavg);
xlabel('dy');
ylabel('STD(V)');
subplot(2,1,2);
plot(dy,Pavg_z);
xlabel('dy ');
ylabel('STD(Z)');
figure(3);
clf
bar(xout,dist);
xlabel('x (mm)');
ylabel('distribution');
hold on
end
xout=xout';
dist=dist';
%% saving in Pavg.mat
if sav
145
suffix = ['date' date '_num'];
save(['num_' suffix '.mat'],'T','dy','Pavg','Pavg_z','K','damp'...
,'m','mag','Kv','Kc','Cb','sf','sf2','f_n','Rl','k_31','d_31','sigma');
save('distribution.txt','dist','-ascii');
end
toc
catch
lasterror
end
146
Bibliography
147
12. Rabaey, J., et al., PicoRadio: Ad-hoc wireless networking of ubiquitous low-
energysensor/monitor nodes. VLSI, 2000. Proceedings. IEEE Computer Society
Workshop on, 2000: p. 9-12.
13. Freitas Jr, R.A., &anomedicine, Basic Capabilities Vol. I. 1999, Georgetown, TX:
Landes Bioscience.
14. E. Lefeuvre, A.B., C. Richard, L. Petit, D. Guyomar, A comparison between several
vibration-powered generators for standalone systems. Sensors and Actuators A:
Physical, 2006. 126: p. 405-416.
15. Roundy, S., et al., Power Sources for Wireless Sensor &etworks. Energy (J/cm 3).
3780(2880): p. 1200.
16. Paradiso, J.A. and T. Starner, Energy Scavenging for Mobile and Wireless
Electronics. Pervasive Computing, IEEE, 2005. 4(1): p. 18-27.
17. Robert J. Linhardt, e.a. Flexible energy storage devices based on nanocomposite
paper. in Proceedings of the &ational Academy of Sciences of the United State of
America. 2007.
18. Neudecker, B.J., N.J. Dudney, and J.B. Bates, “Lithium-Free” Thin-Film Battery
with In Situ Plated Li Anode. Journal of The Electrochemical Society, 2000. 147: p.
517.
19. Hart, R.W., et al., 3-D Microbatteries. Electrochemistry Communications, 2003.
5(2): p. 120-123.
20. The UltraCell RMFC. [cited; Available from:
http://www.ultracellpower.com/pwp/tech_detail.htm.
21. Epstein, A.H., Millimeter-Scale, Micro-Electro-Mechanical Systems Gas Turbine
Engines. Journal of Engineering for Gas Turbines and Power, 2004. 126: p. 205.
22. Lee, C.H., et al., Design and fabrication of a micro Wankel engine using MEMS
technology. Microelectronic Engineering, 2004. 73: p. 529-534.
23. Fu, K., et al., Design and Fabrication of a Silicon-Based MEMS Rotary Engine.
Proc. 2001 International Mechanical Engineering Congress and Exposition
(IMECE), 2001.
24. Toriyama, T., S. Sugiyama, and K. Hashimoto, Design of a resonant micro
reciprocating engine for power generation. TRANSDUCERS, Solid-State Sensors,
Actuators and Microsystems, 12th International Conference on, 2003, 2003. 2.
25. Nielsen, O.M., et al., A thermophotovoltaic micro-generator for portable power
applications. TRANSDUCERS, Solid-State Sensors, Actuators and Microsystems,
12th International Conference on, 2003, 2003. 1.
26. Richards, C.D., et al., MEMS power- The p 3 system. IECEC- 36 th Intersociety
Energy Conversion Engineering Conference, 2001: p. 803-814.
27. Whalen, S., et al., Design, Fabrication and Testing of the P3 Micro Heat Engine.
Sensors and Actuators, 2003. 104(3): p. 200-208.
28. Guo, H. and A. Lal, &anopower betavoltaic microbatteries. TRANSDUCERS,
Solid-State Sensors, Actuators and Microsystems, 12th International Conference on,
2003, 2003. 1.
148
29. Li, H., Self-reciprocating radioisotope-powered cantilever. Journal of Applied
Physics, 2002. 92(2): p. 1122.
30. Lal, A.D. and R.H. Li, Pervasive Power: A Radioisotope-Powered Piezoelectric
Generator. Pervasive Computing, IEEE, 2005. 4(1): p. 53-61.
31. Randall, J.F., On ambient energy sources for powering indoor electronic devices.
2003, Ecole Polytechnique Federale de Lausanne.
32. Roundy, S., et al., A 1.9 GHz RF Transmit Beacon using Environmentally Scavenged
Energy. optimization. 4(2): p. 4.
33. Warneke, B.A., et al., An Autonomous 16mm 3 Solar-Powered &ode for Distributed
Wireless Sensor &etworks. Proceedings of Sensors’ 02, 2002.
34. Shaheen, S.E., 2. 5% efficient organic plastic solar cells. Applied Physics Letters,
2001. 78(6): p. 841.
35. Stordeur, M. and I. Stark, Low power thermoelectric generator-self-sufficient energy
supplyfor micro systems. Thermoelectrics, 1997. Proceedings ICT'97. XVI
International Conference on, 1997: p. 575-577.
36. Strasser, M., et al., Miniaturized thermoelectric generators based on poly-Si and
poly-SiGe surface micromachining. Sensors & Actuators A, 2002. 3179: p. 1-8.
37. Glatz, W., S. Muntwyler, and C. Hierold, Optimization and fabrication of thick
flexible polymer based micro thermoelectric generator. Sensors and actuators. A,
Physical, 2006. 132(1): p. 337-345.
38. Kymissis, J., et al., Parasitic power harvesting in shoes. Proc. of the Second IEEE
International Conference on Wearable Computing (ISWC), IEEE Computer Society
Press, pages, 1998: p. 132–139.
39. Singh, U.K., Piezoelectric Power Scavenging of Mechanical Vibration Energy, in
Australian Mining Technology Conference. 2007.
40. Y.B. Jeon, R.S., J.-h. Jeong and S.-G. Kim, MEMS power generator with transverse
mode thin film PZT. Sensors and Actuators A: Physical, 2005. 122: p. 16-22.
41. Wang, Z.L. and J. Song, Piezoelectric &anogenerators Based on Zinc Oxide
&anowire Arrays. 2006, American Association for the Advancement of Science.
42. Wright, S.R.a.P.K., A piezoelectric vibration based generator for wireless
electronics. Smart Material and Structures, 2004. 13: p. 1131-1142.
43. Fang, H.B., et al., Fabrication and performance of MEMS-based piezoelectric power
generator for vibration energy harvesting. Microelectronics Journal, 2006. 37(11):
p. 1280-1284.
44. Amirtharajah, R., et al., A micropower programmable DSP powered using a MEMS-
basedvibration-to-electric energy converter. Solid-State Circuits Conference, 2000.
Digest of Technical Papers. ISSCC. 2000 IEEE International, 2000: p. 362-363.
45. Meninger, S., et al., Vibration-to-electric energy conversion. Very Large Scale
Integration (VLSI) Systems, IEEE Transactions on, 2001. 9(1): p. 64-76.
149
46. Beeby, S.P., M.J. Tudor, and N.M. White, Energy harvesting vibration sources for
microsystems applications. Measurement Science and Technology, 2006. 17(12): p.
R175-R195.
47. Roundy, S., P.K. Wright, and K.S.J. Pister, MICRO-ELECTROSTATIC
VIBRATIO&-TO-ELECTRICITY CO&VERTERS. Fuel Cells (methanol). 220: p. 22.
48. Basrour, S., et al., Fabrication and characterization of high damping electrostatic
micro devices for vibration energy scavenging.
49. Tashiro, R., et al., Development of an electrostatic generator for a cardiac
pacemaker that harnesses the ventricular wall motion. Journal of Artificial Organs,
2002. 5(4): p. 239-245.
50. Williams, C.B. and R.B. Yates, Analysis Of A Micro-electric Generator For
Microsystems. Solid-State Sensors and Actuators, 1995 and Eurosensors IX.
Transducers' 95. The 8th International Conference on, 1995. 1.
51. Beeby, S.P., et al., Micromachined silicon Generator for Harvesting Power from
Vibrations. 2004.
52. Kulah, H. and K. Najafi, An electromagnetic micro power generator for low-
frequency environmental vibrations. Micro Electro Mechanical Systems, 2004. 17th
IEEE International Conference on.(MEMS), 2004: p. 237-240.
53. Mizuno, M. and D.G. Chetwynd, Investigation of a resonance microgenerator.
Journal of Micromechanics and Microengineering, 2003. 13(2): p. 209-216.
54. Sághi-Szabó, G., R.E. Cohen, and H. Krakauer, First-Principles Study of
Piezoelectricity in PbTiO_ {3}. Physical Review Letters, 1998. 80(19): p. 4321-
4324.
55. White, N.M. and J.D. Turner, Thick-film sensors: past, present and future. Meas.
Sci. Technol, 1997. 8(1): p. 1-20.
56. Lovinger, A.J., Ferroelectric Polymers. Science, 1983. 220(4602): p. 1115.
57. Wang, X., et al., Direct-Current &anogenerator Driven by Ultrasonic Waves.
Science, 2007. 316(5821): p. 102.
58. Ikeda, T., Fundamentals of Piezoelectricity. . 1996, Walton St, Oxford, UK: Oxford
University Press 263.
59. Jeon, Y.B., et al., MEMS power generator with transverse mode thin film PZT.
Sensors and Actuators A: Physical, 2005. 122(1): p. 16-22.
60. Map, S., P. Room, and S. Cart, Improving Power Output for Vibration-Based Energy
Scavengers. Pervasive Computing, IEEE, 2005. 4.
61. Roundy, S., et al., Improving Power Output for Vibration-Based Energy Scavengers.
Pervasive Computing, IEEE, 2005. 4(1): p. 28-36.
62. Duffing, G., Erzwungene Schwingungen bei veränderlicher Eigenfrequenz. Vieweg,
Braunschweig, 1918: p. 134.
63. Rahman, M., Stationary solution for the color-driven Duffing oscillator. Physical
Review E, 1996. 53(6): p. 6547-6550.
150
64. Grossmann, F., et al., Coherent destruction of tunneling. Physical Review Letters,
1991. 67(4): p. 516-519.
65. Shu, Y.C. and I.C. Lien, Analysis of power output for piezoelectric energy
harvesting systems. Smart Mater. Struct, 2006. 15: p. 1499–512.
66. Sodano, H.A., G. Park, and D.J. Inman, Estimation of Electric Charge Output for
Piezoelectric Energy Harvesting. Strain, 2004. 40(2): p. 49-58.
67. Hagood, N.W., W.H. Chung, and A. Von Flotow, Modelling of Piezoelectric
Actuator Dynamics for Active Structural Control. Journal of Intelligent Material
Systems and Structures, 1990. 1(3): p. 327.
68. Kraftmakher, Y., Magnetic field of a dipole and the dipole–dipole interaction.
European Journal of Physics, 2007. 28(3): p. 409-414.
69. Higham, D.J., An Algorithmic Introduction to &umerical Simulation of Stochastic
Differential Equations. SIAM Review - Society for Industrial and Applied
Mathematics 2001. 43(3): p. 525-546.
70. Fox, R.F., et al., Fast, accurate algorithm for numerical simulation of exponentially
correlated colored noise. Physical Review A, 1988. 38(11): p. 5938-5940.
71. Risken, H., The Fokker-Planck Equation: Methods of Solution and Applications.
1996: Springer.
72. Mei, D.C., Y.L. Xiang, and C.W. Xie, &umerical simulation study of the state
variable correlation function of a bistable system subject to cross-correlated noises.
Physica Scripta, 2006. 74(1): p. 123-127.
73. Gitterman, M., Classical harmonic oscillator with multiplicative noise. Physica A:
Statistical Mechanics and its Applications, 2005. 352(2-4): p. 309-334.
74. Ottman, G.K., et al., Adaptive piezoelectric energy harvesting circuit for wireless
remote power supply. Power Electronics, IEEE Transactions on, 2002. 17(5): p. 669-
676.
75. Shenck, N.S. and J.A. Paradiso, Energy scavenging with shoe-mounted
piezoelectrics. Micro, IEEE, 2001. 21(3): p. 30-42.
76. Guyomar, D. and A. Badel, &onlinear semi-passive multimodal vibration damping:
An efficient probabilistic approach. Journal of Sound and Vibration, 2006. 294(1-2):
p. 249-268.
77. Wang, Z.L. and J. Song, Piezoelectric &anogenerators Based on Zinc Oxide
&anowire Arrays. 2006, American Association for the Advancement of Science. p.
242-246.
78. Ammar, Y., et al., Wireless sensor network node with asynchronous architecture
and vibration harvesting micro power generator. Proceedings of the 2005 joint
conference on Smart objects and ambient intelligence: innovative context-aware
services: usages and technologies, 2005: p. 287-292.
79. Fang, H.B., et al., Fabrication and performance of MEMS-based piezoelectric power
generator for vibration energy harvesting. Microelectronics Journal, 2006. 37(11):
p. 1280-1284.
151
80. Aldridge, J.S. and A.N. Cleland, &oise-Enabled Precision Measurements of a
Duffing &anomechanical Resonator. Physical Review Letters, 2005. 94(15): p.
156403.
81. Almog, R., et al., &oise Squeezing in a &anomechanical Duffing Resonator. Physical
Review Letters, 2007. 98(7): p. 078103.
82. Gammaitoni, L., Private communication. 2006.
83. Werner, M.J. and P.D. Drummond, Robust algorithms for solving stochastic partial
differential equations. Journal of Computational Physics, 1997. 132(2): p. 312-326.
152