9
I‘£.‘\
..

.¥V-L.Ct.nmn
”lawman"

. i

p

. 2.0.. 1.: .. V I
‘ .. . inn“!!! .
VI\....: .157 5135.11.39?! 1. I:
til.) ; {trivt It.
, o .k).

v». v..: Rx...

.-...im

 

    

Ila“!

_ ”as. Aiumfikfimfiwfi

c .., n

 

L Lurnn|v\

«3... . . . ‘ H.
2. .v. hunt-9.. . ‘ .. 1 _ . L x
s“. t n... . .. ‘ . . ‘ ‘ ‘...Nn.mm\m13mn KWwfmmfii :4.

. ,.
e . 620:: .

émafiisfimafie

  

  
   

 

 

 

III
1"

 

 

'mmm

 

 

2 UBRARY
zoo 2 Michigan State
, University

 

 

This is to certify that the
dissertation entitled

SELF- POWERED SENSING IN STRUCTURAL HEALTH AND
' USAGE MONITORING

presented by

NIZAR LAJNEF

has been accepted towards fulfillment
of the requirements for the

 

 

 

 

PhD. degree in Civil Engineering
7/
fi/f r (NIELL €qu) WWW
”7; Major Professor’ 3 Signat’ure
/ 08/1 1/08
Date

MSU is an affinnative-action, equal-opportunity employer

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

“MAWZWI
U

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K.IProj/Acc&Pres/CIRCIDaleDue.indd

 

 

 

SELF-POWERED SENSING IN STRUCTURAL HEALTH AND USAGE
MONITORING

By

Nizar Lajnef

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment ofthe requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Civil and Environmental Engineering

2008

ABSTRACT

SELF -POWERED SENSING IN STRUCTURAL HEALTH AND USAGE
MONITORING

By

Nizar Lajnef

Fatigue and overload of mechanical, civil and aerospace structures remains a major
problem that can lead to costly repair and catastrophic failure. Long term monitoring of
mechanical loading for these structures could reduce maintenance cost, improve
longevity and enhance safety. However, the powering of these sensors throughout the

lifetime of the monitored structure remains a major problem.

The ability to convert ambient energy into electric power would eliminate the problem
of drained electrical supply, and would allow indefinite monitoring. This work first
answers the key question: can sufficient electrical energy be produced from typical
structural motions? Realistic earthquake, wind and traffic loads are used to calculate the
theoretical maximum energy levels that can be extracted from these dynamic events. The
same dynamic loads are used to calculate and experimentally measure the electrical
energy produced by a realistic piezoelectric generator.

The collected energy levels are compared to the energy requirements of various

. . . 3
electronic subsystems in a WIreless sensor. For a 5 cm sensor node (the volume of a

typical concrete stone), it is found that only extreme events such as earthquakes can

provide sufficient energy to power currently available wireless sensors. For most typical

. . . . 3
dynamic events, it would be impOSSIble to harvest enough energy to power a 5 cm

wireless sensor. The results show that achieving continuous, self-powered, monitoring
will require the development of a sensor node that can operate within a power budget of

luW.

The implementation of a novel self-powered fatigue monitoring sensor is presented.
The sensor is based on the integration of piezoelectric transduction with floating gate
avalanche injection. The miniaturized sensor enables self—powered, continuous
monitoring and time-to-failure predictions of mechanical and civil structures. The sensor
exploits a log-linear response of a current starved hot-electron injection process on a
floating—gate transistor biased in the weak-inversion region. The measured response is
shown to be minimally invariant to device mismatch and temperature fluctuations. By
configuring an array of floating-gate transistors to respond to different amplitude levels
of the input signal, the proposed circuit implements a level counting algorithm which is
widely used in many usage monitoring techniques. Measured results from a fabricated

integrated circuit in a 0.5-um CMOS process demonstrate that the prototype can sense,

7 . . . . .
store and compute over 10 loadmg cycles. The power dlSSlpaIIOH of the prototype IS
measured to be 800nW which makes it ideal for autonomous long-term operation. The

prototype is interfaced with different piezoelectric transducers and is tested in the

laboratory to demonstrate its applicability for real-time usage monitoring.

Copyright by
NIZAR LAJNEF
2008

DEDICATION

To my parents, without whom this work could not have been completed

ACKNOWLEDGEMENTS

I would like to express my gratitude to Drs Niell Elvin and Shantanu Chakrabartty for
their support, ideas and guidance throughout my stay at Michigan State University. Their
vision and expertise guided me in the development of this research. I also would like to
extent my appreciations to Dr. Ronald Hariehandran and Dr Rigoberto burgeneo for
reading this dissertation and offering constructive comments.

Thanks to Dr Karim C hatti for his friendship and valuable advices. He always gave me
back my confidence whenever I lost it. Thanks to my parents, my wife and my family for
their continuous emotional support.

I am indebted to Amit Gore for the innumerable discussions that we had together, and
for his support in the testing of the analog floating gates. Finally, many thanks to my
fellow graduate students at the AIM lab and the Civil Engineering Department.

The research in this dissertation has been supported by the NSF grant CMMI 0700632.

vi

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. ix
LIST OF FIGURES ............................................................................................................ x
CHAPTER 1
INTRODUCTION .............................................................................................................. 1
1.1 Motivation ........................................................................................................... 1
1.2 Background ......................................................................................................... 3
1.2.1 Fatigue And Fatigue Prediction .................................................................. 3
1.2.2 Smart Sensors For Structural Monitoring ................................................... 8
1.3 Objectives And Overview ................................................................................. 1 1
CHAPTER 2
PIEZOELECTRIC ENERGY HARVESTING ................................................................ 14
2.1 Introduction ....................................................................................................... 14
2.2 Piezoelectric Basics .......................................................................................... 14
2.3 Piezoelectric Modeling ..................................................................................... 17
2.3.1 Strain Energy Harvesting .......................................................................... 17
2.3.2 Vibration Energy Harvesting .................................................................... 20
2.3.2.1 Background ........................................................................................... 20
2.3.2.2 Simplified Mechanical And Equivalent Electrical Models .................. 23
2.3.2.3 Experimental Validation Of Models ..................................................... 24
2.4 Conclusion ........................................................................................................ 34
CHAPTER 3
FEASIBILITY OF STRUCTURAL MONITORING WITH VIBRATION POWERED
SENSORS ......................................................................................................................... 35
3.1 Introduction ....................................................................................................... 35
3.2 Electronics And Subsystem Power Consumption ............................................. 38
3.3 Practical Considerations .................................................................................... 39
3.4 Vibration Energy Harvesting In Typical Civil Environments .......................... 40
3.4.1 Harvestable Power .................................................................................... 43
3.4.2 Maximum Theoretical Harvestable Power ............................................... 46
3.4.3 Piezoelectric Energy Conversion Efficiency ............................................ 48
3.5 Energy Harvesting In High Vibration Civil Environments .............................. 52
3.5.1 Optimal Energy Harvesting ...................................................................... 54
3.5.2 Example .................................................................................................... 55
3.5.3 Optimal Energy Harvesting Results .......................................................... 56
3.6 Discussion ......................................................................................................... 59
3.7 Conclusion ........................................................................................................ 64

vii

CHAPTER 4
SUB-MICROWATT PIEZO-POWERED ANALOG CIRCUITS: MODELING AND

ANALYSIS ....................................................................................................................... 66
4.1 Introduction ....................................................................................................... 66
4.2 Floating Gate Transistors .................................................................................. 67
4.3 F loating-Gate Injector And Its Mathematical Model ........................................ 72
4.4 Cmos Current References ................................................................................. 76
4.5 Design OfA Complete Floating-Gate Sensor Array ........................................ 80
4.6 Conclusion ........................................................................................................ 84

CHAPTER 5

RESULTS AND DISCUSSION ....................................................................................... 85
5.1 Introduction ....................................................................................................... 85
5.2 System Robustness ............................................................................................ 87
5.3 Experimental Measurments .............................................................................. 89

5.3.1 Initialization Of The Injectors Array ........................................................ 89
5.3.2 Threshold Adjustment And Levels Detection ........................................... 91
5.3.3 Level-Crossing Counter ............................................................................ 95
5.3.4 Counting Resolution ............................................................................... 102
5.3.5 Power Dissipation ................................................................................... 102
5.4 Design Examples For Systems Monitoring — Sensor And Transducer
Interfacing ................................................................................................................... 104
5.5 Discussion ....................................................................................................... I 10
5.6 Conclusion ...................................................................................................... 1 1 1

CHAPTER 6

CONCLUSIONS AND RECOMMENDATIONS ......................................................... 113

APPENDIX

MICROGRAPHS AND LAYOUT DETAILS ............................................................... 1 18

BIBLIOGRAPHY ........................................................................................................... 126

viii

LIST OF TABLES

Table 1.1 Cycle counting methods [19, 20, 21, 22] ............................................................ 7

Table 2.1 Comparison of material properties of two commonly used piezoelectric
materials .................................................................................................................... 17

Table 2.2 Generator parameters assuming an Euler-Bernouli cantilever bimorph. The
displacement shape function is assumed to be, w = u(t)[l — cos(71x /(2L)] where x is
the horizontal distance from the support. The electrical potential shape function is
given byI/l = zv/ h , where z is the vertical distance from the beam center, i.e. the
beam’s displaced shape. E is the modulus of elasticity of the beam, 63; is the
piezoelectric coupling coefficient, 8 is the electric permittivity, I=bh3/12 is the

second moment of area and 772‘ is the mass per unit length. ...................................... 23
Table 2.3 Piezoelectric sensor properties .......................................................................... 30
Table 3.1 Comparison of energy scavenging sources [52] ............................................... 37

Table 3.2 Approximate energy consumption in a wireless sensor node (references in
brackets). Volatile memory (VM) is assumed to be replaced by low-power non-

volatile memory (NVM) [56] .................................................................................... 39
Table 3.3 Piezoelectric and circuit sensor properties ........................................................ 58
Table 4.1 Component sizes used in circuit in Figure 4.10 ................................................ 81
Table 5.1 Summary of measured specifications ............................................................... 86

ix

LIST OF FIGURES

Figure 1.1 Example of an S-N curve (redrawn with parameters adapted from [18]) ......... 5

Figure 2.1 Schematic depiction of a piezoelectric material pulled by a force F. The
material undergoes an axial strain S ......................................................................... 17

Figure 2.2 (a) Schematic of a piezoelectric scavenger of length L, width b and height h;
(b) Equivalent circuit diagram (the dashed block) for a piezoelectric material under
slow dynamic loading ............................................................................................... 19

Figure 2.3 Theoretical energy harvester with an equivalent electric damper ................... 21

Figure 2.4 Schematic of a piezoelectric vibration scavenger of mass (M), length (L),
width (b) and height (h). The equivalent mechanical single degree of freedom
system is also shown ................................................................................................. 21

Figure 2.5 The equivalent circuit for a single degree of freedom piezo-generator. The
mechanical force input is simulated as a voltage generator element with
magnitude Miig ......................................................................................................... 24

Figure 2.6 Method of experimentally measuring piezoelectric generator parameters. The
vibrator is driven over a range of frequencies to obtain the open-circuit voltages. The
accelerometer is used to measure the base excitation iig directly ............................ 25

Figure 2.7 Open-circuit piezoelectric voltage transfer function (the least square fitted
model is shown as a solid line, and the experimental results are shown as dots). The
non-linear least square fitted parameters are K=985N/m, Q=6x104C/m, and §=l.3%.
The measured parameters are M,,-p=9.4g and C=25.2nF. The coefficient of
correlation is 170.98 .................................................................................................. 26

Figure 2.8 A full diode bridge piezoelectric circuit for storing electrical charge on a
capacitor (Cs), the circuit leakage is represented by the resistor (R4) ....................... 27

Figure 2.9 (a) Measured and calculated voltages for the piezoelectric unimorph with
properties calculated from Figure 2.7. The input accelerations are compared at 5
seconds after the linear stroke motor is switched on. The circuits elements
properties are: storage capacitor CS = 3.3uF; drain resistance Rd = IOMQ, and diodes
1N4148. The line represents equal measured and calculated voltages, the coefficient
of correlation, assuming the simulated voltage is equal to the measured voltage is
r=0.97; (b) a simulated (solid) and experimental (dashed) voltage time history on the
storage capacitor ....................................................................................................... 28

Figure 2.10 Schematic representation of a PVDF vibration energy scavenger ................ 31

Figure 2.1 1 Experimental setup to test the piezoelectric bimorph generator .................. 31

Figure 2.12 Experimental (dots) and theoretical (line) acceleration—voltage transfer
function for piezoelectric energy harvester ............................................................... 32

Figure 2.13 Theoretical and experimental results for the piezoelectric energy generator
subjected to the El Centro earthquake (only 6 s is shown for clarity) ...................... 33

Figure 3.1 Schematic of a piezoelectric vibration scavenger ........................................... 39

Figure 3.2 (a) Time-history and (b) frequency spectrum for the acceleration
measurements from the deck of a concrete bridge under traffic loading [67] .......... 41

Figure 3.3 (a) Ground acceleration time-history and (b) frequency spectrum from the El
Centro earthquake (North—South direction) .............................................................. 42

Figure 3.4 Theoretical average power output for various electric damping ratios. The
mechanical damping is assumed to be 5% ................................................................ 44

Figure 3.5 (a) Maximum power and (b) bandwidth for various electrical and damping
ratios .......................................................................................................................... 45

Figure 3.6 Maximum theoretical vibration energy for various structures and loading
conditions .................................................................................................................. 48

Figure 3.7 Electric energy output from a piezoelectric bimorph for various dynamic
excitations ................................................................................................................. 5 1

Figure 3.8 Approximate serviceability structural vibration limits (based on [74], [75] and
[76]). The shaded region represents the limits based on various exposure times. The
lower boundary represents longer duration vibrations; the upper boundary represents
shorter duration vibrations. The solid line at 1.5m/52 represents extreme vibration
conditions .................................................................................................................. 53

Figure 3.9 Investigated structure consisting of a steel I—beam with an overhead moving
crane. Position of accelerometer node is as shown ................................................... 55

Figure 3.10 A typical input acceleration time history and the associated frequency
spectrum for a lightweight crane moving at 0.8m/s, measured at mid-span on the
supporting beam ........................................................................................................ 57

Figure 3.11 Maximum extractable electrical energy for a resistive circuit (solid lines) and

diode-bridge circuit (circles), for the crane moving from left to right at 0.8m/s. The
three solid lines represent acceleration histories from three different crane runs ..... 58

xi

Figure 3.12 Generated electrical energy from the El Centro earthquake for a bimorph
PZT with various resonant frequencies ..................................................................... 61

Figure 3.13 Generated electrical energy for various traffic loadings on a concrete bridge,
assuming perfect electro-mechanical conversion ..................................................... 62

Figure 4.1 Principle of the piezo-powered floating—gate sensor ....................................... 67

Figure 4.2 (a) Top and cross-sectional view of a floating-gate transistor and (b) the
measured drain current when injection pulses of different amplitude are repeatedly

applied across the drain-source terminal ................................................................... 69
Figure 4.3 Capacitive model of a pMOS floating-gate transistor ..................................... 71
Figure 4.4 Simplified circuit model for piezo-driven floating-gate sensor [83] .............. 73

Figure 4.5 Response of a floating—gate injector based on the mathematical model in

Equation 4.10 ............................................................................................................ 75
Figure 4.6 Response of the floating-gate injector for different values of (a) parameter K2

and (b) parameter K1 ................................................................................................ 77
Figure 4.7 Schematic of a CMOS current reference ......................................................... 78
Figure 4.8 F loating-gate reference array ........................................................................... 82

Figure 4.9 Simulation result using the circuit in Figure 4.8, showing the current through

the floating gate cells for different peak excitation input voltage ............................ 83
Figure 5.1 Micrograph of the prototype floating-gate sensor ........................................... 86
Figure 5.2 Voltage responses measured at various source currents .................................. 88
Figure 5.3 Voltage responses measured for various chips in the same run ...................... 88
Figure 5 .4 Voltage responses measured for various temperatures ................................... 89
Figure 5 .5 Flow-chart describing the sensor initialization algorithm ............................... 90
Figure 5.6 Measured output voltages (O 1. —O4) response vs. input voltage ...................... 93

Figure 5.7 Measured source-to-drain voltage response across floating-gate elements at
Vdd=5.3 V ................................................................................................................... 93

Figure 5.8 Measured source-to-drain voltage response across floating-gate elements at
Vdd=6.1 V ................................................................................................................... 94

xii

Figure 5.9 Measured source—to-drain voltage response across floating-gate elements at
Vddz6. 9V ................................................................................................................... 94

Figure 5.10 Measured output voltage response for the sensor array when subjected to a
loading cycle represented by a piezoelectric output with injection duration 3:2:1
................................................................................................................................... 97

Figure 5.11 Measured output voltage response for the sensor array when subjected to a
loading cycle represented by a piezoelectric output with injection duration 322.521
................................................................................................................................... 98

Figure 5.12 Measured output voltage response for the sensor array when subjected to a

loading cycle represented by a piezoelectric output with injection duration 3:2:0.5
................................................................................................................................... 99

Figure 5.13 Measured output voltage response using the sensor array in which the first
1000 cycles have been used as calibration intervals and the subsequent three level
loading cycles with the duration of injection are in the ratio 3:221 ........................ 100

Figure 5.14 Measured output voltage response using the sensor array in which the first
1000 cycles have been used as calibration intervals and the subsequent three level
loading cycles with the duration of injection are in the ratio 3:2.5:1 ..................... 101

Figure 5.15 Measured results to calculate the resolution for event counting: (a) Change in
source voltage measured for different relative counts of voltage pulses for over
100,000 events; (b) Inset showing zoomed area between 10,000 to 40,000 events

................................................................................................................................. 103
Figure 5.16 Supply current drawn by the sensor at different supply voltages ................ 104
Figure 5.17 (a) Voltage generated by a piezoelectric transducer in response to periodic

loading; (b) Measured response of the floating-gate injector array ........................ 106
Figure 5.18 Setup environment for MTS machine ........................................................ 108

Figure 5.19 Voltage responses measured when the prototype is interfaced with a PVDF
transducer and subjected to controlled cyclic strain levels with magnitude 2107 us
................................................................................................................................. 108

Figure 5.20 Voltage responses measured when the prototype is interfaced with a PVDF
transducer and subjected to controlled cyclic strain levels with magnitude 2546u8
................................................................................................................................. 109

xiii

Figure 5.21 Voltage responses measured when the prototype is interfaced with a PVDF
transducer and subjected to controlled cyclic strain levels with magnitude 2809ue

................................................................................................................................. 109
Figure 5.22 Slope deviation observed for injection currents less than 1 electron per

second ..................................................................................................................... 110
Figure 6.1 System architecture of a complete implanted device that integrates the

proposed floating-gate sensor array ........................................................................ 116
Figure A.1 Micrographs of different prototypes ............................................................. 120
Figure A.2 Layout of the prototype tested in Chapter 5 ................................................. 121
Figure A.3 Layout of the current reference .................................................................... 122
Figure A.4 Layout of the diodes array ............................................................................ 123
Figure A.5 Layout ofthe floating—gate injectors array ................................................... 124
Figure A.6 Layout ofa single floating-gate injector ....................................................... 125

xiv

CHAPTER 1
INTRODUCTION

1.1 MOTIVATION

Innovative sensors and sensor systems represent one of the foundations of smart
structures technology deve10pment. The design, fabrication, and implementation of these
systems are one of the ultimate challenges to engineering researchers today. Structural
monitoring is a primary application for new sensor technologies. Indeed, much attention
has been focused in recent years on the declining state of the aging civil infrastructure in
the United States. These concerns apply not only to civil engineering structures, but also
to other types of mechanical systems, such as aircrafts, rotating machines and
biomechanical implants. A critical parameter is the prediction of the remaining life of
structural components which is an ongoing research and commercial enterprise that can
lead to substantial cost savings. For example, the US Air Force estimates that there will
be a cost saving of nearly $1 billion for the next 15 years for the rotor of the F100
aircraft, due to the procedures that were first implemented in 1986 to detect and predict

fatigue damage [1].

However, in many cases the current techniques that have been developed for the
evaluation of structural components remain expensive, unreliable or impractical. For
example, currently there is no completely reliable means of monitoring sub-surface
damage in civil infrastructure, though many expensive techniques such as thermography

[2] and ground penetrating radar [3] are currently being developed with various degrees

of success [4]. In many structures the cost—savings of fatigue monitoring cannot be
realized since these advanced fatigue life prediction strategies require damage detection
through regular non—destructive inspection and relatively sophisticated fatigue life
prediction algorithms [5]. This makes these techniques too expensive (or unreliable) for
many low-cost (or single unit) structures such as commonly found in civil infrastructure
for example (railways, pavements and bridges).

It is largely due to the cost and maintenance time needed for classical non-destructive
imaging techniques that the field of Structural Health Monitoring is currently being
investigated [6, 7, 8]. The use of embedded sensors, offers several benefits over non-
destructive imaging techniques. These include the ability for continuous monitoring
without the disruption of use of the structure for inspection. Though structural monitoring
is attractive for these reasons, several issues still hamper their wide acceptance. The first
disadvantage of wireless embedded sensors is their relative expense. Even at $100 per
sensor node [9] (probably an unrealistically low number), instrumenting an entire
structure with many sensors might be prohibitively expensive for many applications.
Secondly, the useful life of the sensor nodes is dependent on the battery power [9].
However with current battery operation it is unlikely that the wireless sensor nodes would
last more than several years before being replaced (The best shelf—life of batteries is on
the order of 5 to 10 years). The expense of periodically replacing batteries, especially for
sensors embedded within the structure, remains a major problem for the wide acceptance
of wireless-sensor networks. (Other problems such as network size are also a major

concern [10, 11]).

Another possibility of structural monitoring is the so called ‘Health and Usage
Monitoring System’ (HUMS). This system senses and stores the mechanical usage of a
structural component [12, 13, 14]. Though often inaccurate in predicting remaining life, it
is still an important tool in the aircraft industry for scheduling further inspection and
maintenance. However, current HUMS systems are expensive, bulky and also require a
continuous source of power.

This work verifies and tests the hypotheses that a low-cost, self-powered usage
monitoring sensor, that can compute and store cumulative history of strain experienced
by the structure, can be achieved by combining piezoelectric transduction and low power
analog signal processing circuits. The developed system would serve as an important

intermediate solution for fatigue life monitoring of many structures.

1.2 BACKGROUND

1.2.1 Fatigue and fatigue prediction

The term fatigue was first introduced by Poncelet [15] in 1839. In the early twentieth
century, due to the increasing use of ferrous structure especially in railway systems,
significant work has been done to understand and model this phenomenon. Although
these models can predict, to some extent the experimental results, they remain inaccurate
for certain real applications. For example these models cannot take into account
environmental conditions and material imperfections.

Mechanical fatigue is formally defined as the accumulation of damage in a structure
under applied fluctuating stresses. Even though the magnitudes of the applied stresses

could be less than the tensile strength of the material, the progressive fatigue damage may

ultimately lead to complete mechanical failure. The fatigue life is defined as the number
of constant amplitude load cycles necessary to induce fatigue failure in an initially
undamaged specimen. Several approaches can be used to characterize fatigue life, such as
the energy-based approach and the stress or strain life approach.

A simple energy based prediction of fatigue life can be expressed as follow:

Nf =—- (1.1)

where Q is the total mechanical work absorbed in the material, q is the cyclic energy

and N f is the number of cycle to failure. This criterion is based on the assumptions that

q is approximately constant during the fatigue life and Q is constant for the imposed
external conditions. Q might be also regarded as the maximum amount of mechanical

work the material could absorb untill failure.

For a stress/strain based approach, the fatigue life is typically expressed using S-N
curves. An example of an S-N curve is shown in Figure 1.1. Higher load levels require
fewer cycles to cause structural failure whereas low load levels require many more cycles
to cause failure.

Mathematically, the SN curve can be modeled by the following equation [16]:
= f 2N
0 a 0f f (1.2)
where 0'; is the fatigue strength coefficient, and b is the fatigue strength exponent (or the

Basquin exponent) that are determined experimentally [16].
In its strain-life form, the S-N curve can be characterized based on the plastic strain

amplitude using the Coffin-Manson relation [17] given by:

—A—:E=£;(2Nf)c (1.3)

where A5 p / 2 is the plastic strain amplitude, as; is the fatigue ductility coefficient and c

is the fatigue ductility exponent. Using (1.2), the total strain amplitude can then be related

to the fatigue life as follow:

0;
—2-=af(2Nf)C+—E-(2Nf)b “-4)

Where E is the young's modulus.

1.00

0.10

    

 

 

A a . \
Annealed\ Hardened
. ~ \4340 steel
0.010L— 4340 Steel \ o x . ~
\ .
\ .
x ' u
" ' ' " ‘ " T=a7i§u3 EifiEfBr’éfi's’t‘ée‘l
0.001 I I I J
10° 102 104 105 103
N.

Figure 1.1 Example of an S-N curve (redrawn with parameters adapted from [18])

An important consequence of the S-N curve is that loading cycles less than the so-

called fatigue limit (Figure 1.1) should not cause the material to fail (ofien taken as a

. 7 . . . . .
stress level that can w1thstand 10 cycles). In the context of fatigue monitoring, it IS

therefore unimportant to measure strains less than the fatigue limit of the material since
they do not contribute to overall damage of the material. The monitoring system (sensor)
could therefore ignore input signals below a level that would be calibrated to match the
fatigue limit of the inspected structure. For many construction steels, the fatigue strain
limit is greater than 1500ue [18].

However, the S-N curve can not be directly used to detcnnine fatigue in real-world
conditions, since the magnitude of the applied cyclic loads vary depending on the
environmental conditions. Several algorithms have therefore been devised for fatigue
life-time predictions under non-periodic loading conditions. The most common and
widely used algorithm is the Palmgrcn-Miner’s Rule [16] often simply called the Miner’s
rule. In its basic form, the Miner’s rule assumes that each strain cycle with magnitude i
uses I/N,- of the total fatigue life, where N,- is the number of fatigue cycles taken to fail the

material at a strain level ofi. Algebraically this can be stated as:

"i _

2 if] - 1 (1.5)
The Miner’s rule algorithm is by no means perfect (because of its omission of load
sequence effects) and was not developed for all possible loading conditions. However it
is still routinely used in practice to estimate the fatigue life for structural components.

A pre-processing step is required before Miner’s rule can be applied to calculate
statistics of arbitrary loading cycles. This typically involves counting methods which
reduces a spectrum of varying stress-strain measurements into a set of simple stress-strain
reversals. The objective is to identify the range and mean values of the loading events,
within the random spectrum, which contribute to the specimen fatigue so that the proper

amount of damage due to each event can be accounted for. Table 1.1 presents an illustrat-

Table 1.1 Cycle counting methods [19, 20, 21, 22]

 

 

Name Illustration Description
Strain A
€33 W A All maximums above the mean and all
3 Mean 1 A minimums below the mean are counted

 

 

 

Mean crossing
peak

Strain A

 

Mean/\/\/\/\IL

Only the largest peak between successive

crossing of the mean is counted

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Strain j
50 i 4 . . .
E 3 All posrtive slope level crossmg above the
g - \ 2 mean, and negative slope level crossing
g Mean \/ 1 47 below the mean are counted
--1 t

Strain A
$3 4 Similar to level crossing except that only
E - .V - -\ ........ 3 . 4 one count is made between successive
o -
g, y """" \T """" f ' 3 crossing of a lower level associated with
5 Mean """" 33;} """" 2 ¢ .
u. .................. I t each counting level

Strain ‘A Each range (the difference between
30 r r3 r4 /\ successive peak values) is counted as 1/2
= l . . .
g 0 r V V1 t> cycle, the amplitude of which 18 half the

range

Strain A
C
8 m4
E m3 A Ranges are counted as above and the mean

I

g 0 v1 4 value of each range is also considered
: V t
62

 

 

 

 

 

ion and a small description of six commonly used empirical cycle counting methods. In
this thesis, the level-crossing method is selected. The algorithm counts the total number
of occurrences when the magnitude of strain exceeds a pre-defined threshold. The
thresholds are typically quantized strain levels as shown in Table 1.1 which also shows
the threshold crossing events (denoted by circles). The total count of events denoted by

n,- is then normalized by N,- (from the S-N curve) according to the Miner’s rule (Equation

(1.5)).

1.2.2 Smart sensors for structural monitoring

A smart sensor differs from a standard sensing unit mainly by its on-board intelligence
capabilities (on-board computations). Three important features typically characterize a
smart sensor: an on-board processing unit, wireless capability, and networking
capability.

Some of the first efforts in developing a smart sensor for civil engineering applications
were presented in [23, 24, 25]. Two types of monitoring systems were identified: extreme
event, and long-term monitoring systems. The designed hardware consisted mainly of a
microprocessor, radio modem, data storage, and batteries. Data analysis programs were
developed to analyze the acquired data and facilitate damage detection diagnosis. While
the developed systems provided ease of installation and portability, several problems
were identified such as synchronization of communications signals between nodes and
data collection system, battery durability/longevity, and significant variations of the host
structure characteristics due to environmental changes such as loading, boundary

conditions, temperature, and humidity.

Several research groups [7, ll, 26, 27, 28, 29] have focused on solving the wireless
communication problems. In [26], a bridge monitoring system, consisting of a battery
operated sensing interface and a small radio transponder, was investigated. The complete
system consisted of a data collection system and a network of transducers distributed
throughout the structure. The micro-sensors monitoring systems developed in [27] use a
multiplexer to allow data from multiple transducers to be communicated over a single
communication channel. Data is transmitted at the rate of 50Kbps over a range of
approximately 0.25 miles. Damage is detected via variations in the natural frequencies of
the structure.

The battery life of these sensors remained a major issue. For example, the prototype
presented in [28] operates only for 15 hours continuously on two 9V batteries. In order to
conserve power the sensors are put in an ultra low-power sleep mode, only
communicating when a significant loading event occurs or at predefined fixed time
intervals.

The devises presented in [7, 25] used mixed-mode analog and digital on—board
electronics, and were characterized by a number of important attributes such as bi-
directional communication, on board local processing, pre-programmed user defined
algorithms, and onboard storage memory. The prototype wireless sensor in [7] used
standard integrated circuit component (8-bit Motorola 68HC11E1 microcontroller,
ADXL210 accelerometer). The whole system was accommodated within a sealed
package with an approximate size of 5” by 4” by 1” and was validated through controlled

experiments in the laboratory

Driven by the desire to solve to powering issue, energy harvesting has been a topic
given great attention in recent years [30, 31, 32, 33, 34, 35]. A number of potential self-
powering energy sources have been identified, (e.g, solar power, thermal gradient,
piezoelectric, vibrational), yet few are capable of providing the ~ IOOuW of continuous
power widely believed to be the minimum required to operate a single sensor ([36],
PicoRadio Project UC Berkeley). The quantity of energy that can be converted is directly
related to the volume of the energy-converter. However in many civil engineering
structures, the sensor unit would be embedded into the structure (such as in concrete or
asphalt). A volume greater than 5 cm3 could lead to significant weakening of the
structure. In this thesis, all presented results are for a limited sensor size of 5 cm3, which
is the typical size of concrete aggregate [37].

Several research groups [36, 38, 39] are currently working on decreasing the power-
requirement for wireless motes, but even these low-power wireless sensor network nodes
would require a minimum energy level of 100 uW. In order to decrease sensor-power,
on-board memory storage is an attractive alternative [40]. It should be also noted that in
the context of civil and mechanical system fatigue monitoring, it is important to be able
to capture each loading cycling. Many civil engineering structures are only subjected to
significant loading cycles infrequently. During the times between significant loading
cycles, it is possible that the sensor would experience periods of brown—outs and black-
outs since any on-board battery or ultra-capacitor might lose the charge between these
intermittent loading cycles. This further explains the need for using non-volatile memory

storage, i.e, volatile memory storage would be lost during power-black outs. Current non-

10

volatile memory storage (for example floating gate transistors found in FLASH) requires
311.1 to write 1 byte [University of Berkeley Mote measurement].

Previous research in [41, 42] has shown that it is possible to self-power some basic
electronic switching and RF circuits using a 28pm x 4cm2 PVDF piezoelectric material
under a single trapezoidal pulse of applied strain (approximately 1500118, in magnitude
and 0.5 seconds in duration). The estimated total energy generated by the loading is
approximately 3 n]. It should be noted that this corresponds to an average operational

power-levels of 3 nW! The piezoelectric generated voltage is approximately 15 V.

1.3 OBJECTIVES AND OVERVIEW

The main objective of this work is to design and develop a novel structural fatigue
monitoring sensor based on the integration of a piezoelectric transducer with ultra-low
power computational analog circuits. The miniaturized sensor will enable continuous
battery-less monitoring of engineering structures (for example civil, mechanical, and
biomechanical). This will facilitate efficient maintenance and repair schedules by

localizing possible failure points in a structure. The novel features of the sensor are:

- Self-powered, continuous and autonomous sensing. The proposed sensor will
use novel analog signal processing circuits that will require less than 1 uW of
power (to our knowledge this is two orders of magnitude less power than
previously demonstrated self-powered sensors [39]). The operational energy will
be harvested directly from the sensing signal that is generated by a piezoelectric

transducer.

11

Autonomous computation and non-volatile storage of sensing variables. The
sensor will use floating gate transistor injection principles [43] for computing
cumulative mechanical strain patterns experienced by a structure. The method
will obviate the use of data converters and digital computational circuitry, thus

reducing the power requirement of the sensor.

Even though all the subcomponents of the sensor will be integrated, the scope of this

thesis work will not include fatigue tests and challenges involved in packaging of the

microelectronic components. The aim is to demonstrate practical applications of the

sensor to autonomous sensing by validating and benchmarking its performance in

realistic laboratory simulated structural monitoring scenarios. The specific objectives are:

Objective 1: Piezoelectric characterization with respect to electrical power
generation.

Objective 2: Development of ultra-low power electronic sensing system which
can sense, compute and store cumulative statistics of signals generated by the
piezoelectric transducer.

Objective 3: System integration, calibration and testing of a proof-of-concept

prototype under laboratory conditions.

Chapter 2 presents simple mechanical and equivalent electrical models of the

piezoelectric generator under different loading configurations. The presented models are

validated using experimental results.

12

In chapter 3, the electrical energy that can be generated from typical civil structures
such as bridges and buildings is evaluated under normal and extreme conditions. Results
are obtained using models developed in chapter 2. It is shown that power converted from
average dynamic events such as wind, traffic and low-scale earthquakes is approximately
1 pW, which is insufficient to run currently available sensors which require a minimum
of 100 11W of continuous power.

In chapter 4, the design and theoretical model of a novel self-powered event
monitoring system based on a combination of piezoelectric transduction and floating gate
injection are presented. Cadence based spice simulations of different components confirm
the desired behavior of the proposed design.

Using a fabricated prototype, the design and models presented in chapter 4 are
validated in chapter 5. A complete system, piezoelectric transducer and the fabricated
prototype, is tested under realistic low—frequency, civil engineering loading in the
laboratory. It is shown that the devc10ped sensor can sense store and compute over 105

loading event cycles at a total power dissipation ofless than 800 nW.

The work reported in this dissertation demonstrates the feasibility of a self—powered
smart sensor for continuous monitoring of mechanical systems. This work provides an
initial step toward the implementation and deployment of a low cost and reliable ‘smart’

networks for the long term monitoring of civil infrastructure.

13

CHAPTER 2
PIEZOELECTRIC ENERGY HARVESTING

2.1 INTRODUCTION

Currently, three major approaches are being investigated to convert mechanical energy
to electrical energy for self-powering sensor electronics. These include: (1)
electromagnetic, (2) electrostatic, and (3) piezoelectric harvesters. Though this chapter
focuses only on piezoelectric power conversion, it should be noted that the other
vibration self—powering approaches can be investigated with similar methods.

Piezoelectric energy harvesting, which is the ability of converting ambient mechanical
loading into electrical energy, can potentially transform electronic sensing by providing a
renewable and continuous energy source. Since a number of different approaches have
been suggested for modeling piezoelectric energy harvesters ([35, 44, 45, 46]), a new
researcher or designer in the field of piezoelectric energy harvesting might be
overwhelmed by the number and complexity of the various models.

In this chapter, simple mechanical and equivalent electrical models that can be easily
used for a general piezoelectric generator analysis are presented. The electrical models
could be easily implemented into existing circuit simulation programs (such as SPICE)
and provide reasonably accurate estimates of the generator performance under a wide

variety of ambient loading environments.

2.2 PIEZOELECTRIC BASICS
The direct piezoelectric effect is the ability of certain crystalline materials to generate

electric charge from an applied mechanical stress. Since most piezoelectric are

14

anisotropic, their physical constants (elasticity, permittivity etc.) are tensor quantities and
relate to both the direction of the applied stress, electric field etc., and to the directions
perpendicular to these. For this reason the constants are generally given two subscript
indices which refer to the direction of the two related quantities (e.g. stress and strain for
elasticity, displacement and electric field for permittivity). A superscript index is used to
indicate a quantity that's kept constant. In the discussed examples in this work,
stresses/stains are applied along the X-direction noted ‘1’. The direction of positive
polarization is chosen to coincide with the Z-axis ofa rectangular system and is noted ‘3’.
The physical constants used in piezoelectric modeling are (1) pemiittivity e (or dielectric

constant), defined as the dielectric displacement per unit electric field, (2) the compliance

sij ofa material is the strain produced per unit stress. It's the reciprocal ofthe modulus of

elasticity. The first subscript refers to the direction of strain, the second to direction of

stress, (3) the piezoelectric charge constant dij, defined as the electric polarization

generated in a material per unit mechanical stress applied to it. Alternatively, it is the
mechanical strain experienced by the material per unit electric field applied to it. The first
subscript refers to the direction of polarization generated in the material (at E = 0) or to

the applied field strength, the second refers respectively to the direction of the applied

stress or to the direction of the induced strain, (4) the piezoelectric voltage constant gij,

defined as the electric field generated in a material per unit mechanical stress applied to
it. Alternatively, it is the mechanical strain experienced by the material per unit electric

displacement applied to it. The first subscript refers to the direction of the electric field

15

generated in the material or to the applied electric displacement, the second refers

respectively to the direction of the applied stress or to the direction of the induced strain.

Consider a piezoelectric material with dimensions L x b x h polled through its thickness
with electrodes on the top and bottom surfaces (Figure 2.1). A mechanical force (F) is
applied along the materials length. The generated voltage can be expressed in a number

of different ways as shown below:

E
SY d31h

szé’git25yE/7g3l = (2.1)

where V is the generated voltage, g3] and d3} are piezoelectric constants, S is the applied

mechanical strain, YE is the short circuit elastic modulus and e is the electrical

permittivity. Another important property of piezoelectric materials is their capacitance.

The capacitance of a piezoelectric generator is given by:

_ Lbs
h

C (2.2)

Two different piezoelectric material classes are commonly used for electric energy
harvesting. The first class of material is a ceramic composed of lead zirconate titanate
(PZT), the second is a semi—crystalline plastic polyvinylidene fluoride (PVDF). These
two materials have very different properties and are typically used for different
applications. The mechanical and electrical properties of these two materials are
compared in Table 2.1.

To illustrate the difference between the two materials, consider two piezoelectric strips
30 mm long and 12 mm wide. PVDF is typically formed in 28 um thickness, while PZT

is manufactured in thicknesses of 0.1 mm or thicker.

16

(1) For L=30 mm, b=12 mm, T=28 um
PVDF film: c = 1.36 nF; v = 12x10“3 V/ue

(2) For L=30 mm, b=l2 mm, T=100 um
PZT: C = 38 nF; V = 65 V/ue
Though PZT can generate significantly more voltage than PVDF per unit strain, PZT is
a brittle ceramic and thus can fracture at relatively low strains. Furthermore, tensile
strains as low as 500 118 can cause fatigue damage in PZT. On the other hand, PVDF is a
flexible plastic that can withstand at least 10,000 us of applied tensile strain.

«L/ b/ F

IP th

 

 

 

L

A
T

 

 

 

Figure 2.1 Schematic depiction of a piezoelectric material pulled by a force F. The
material undergoes an axial strain S

Table 2.1 Comparison of material properties of two commonly used piezoelectric
materials

 

 

 

 

 

 

 

Property PVDF PZT-5H
Young’s Modulus (YE) 2x109 N/mz 60x109 N/m2
Permittivity (a) 106x 10'12 F/m 106x 10'10 F/m
Piezo Strain Constant ((131) 23x10-12 C/N 110x10” C/N
Piezo Stress Constant (g31) 216x103 m/C 10x10”3 m/C

 

 

 

2.3 PIEZOELECTRIC MODELING

2.3.1 Strain Energy Harvesting

Piezoelectric material is typically used in two energy harvesting modalities. The first

modality is by attaching the material to a substrate and applying a slowly changing

17

dynamic load (Figure 2.2(a)). A mathematical generator model can be expressed using a
set of equations of the form:

{Ku—Osz :1 25L (2.3)

®u+Cv=I
t

Energy = [v] dt
0

where K is the stiffness, Q is the electromechanical coupling coefficient, C is the

piezoelectric capacitance, u is the displacement of the generator induced by the imposed
strain S, F is the applied force, v is the generated voltage, and I is the circuit current. The
mechanical and electromechanical parameters can be derived using a simplified

Rayleigh-Ritz [44] approach and are given by:

 

E
K ___ Y th

E (2.4)
o = Y d31b

In this modality, the piezoelectric element can be modeled using a simple high-pass
filter circuit as shown in Figure 2.2(b). The piezoelectric voltage generated is given by
Equation (2.1). The capacitance is given by Equation (2.2). The nature of the high-pass
filter circuit shows that very little voltage is generated for slow loading rates. The
adequate matching of both the generated voltage and the capacitance to the applied
mechanical load magnitude and frequency content is therefore critical for maximal
electrical energy generation. It should be noted that the method of attachment (bond) of
the piezoelectric material to its substrate in this modality is critical, and significant

voltage loss (a factor as high as two to four) is common.

18

If the sensor circuit in Figure 2.2 is modeled by a resistor RL, then for a harmonic

loading of the piezoelectric transducer at a frequency f Hz, the magnitude of voltage

across the sensor is given by:

 

 

 

2 CV
VL (f) = ”RL (2.5)
2 2 2 2 / 2
1+ 47! f RLC
and the power delivered to the sensor can be expressed as:
P (f) leU]2 472‘2f2RLC2V2
L = : 4 ‘
RL 1+ 47r2f21rfcz) (2-6)

 

 

Piezoelectric Element h (a)
.1 a}...

 

 

TI .1 (b)

Figure 2.2 (a) Schematic of a piezoelectric scavenger of length L, width b and height h;
(b) Equivalent circuit diagram (the dashed block) for a piezoelectric material under slow
dynamic loading

The maximum power that can be delivered at a loading frequency can be obtained by

optimizing Equation (2.6) with respect to RL which can be expressed as the condition:

19

dPL =147z2f2C7-RLVX_ 16;:4f4c4R2V :0 (2 7)
dRL I1+4”2f2RiC2l (1+47z2szfcz)2 '

 

 

The optimal value of the load resistance R L is then given by:

l

= BEEC— (2.8)

RL

For a loading frequency of 1 Hz and a transducer capacitance of 10 nF the optimal load
according to Equation (2.8) is 15 MD. For a 5-V open circuit voltage, the optimal resistor

would produce a current of 300 nA.

2.3.2 Vibration energy harvesting
2.3.2.1 Background

The second mode of energy generation is through higher frequency vibration. In this
case the piezoelectric material is driven by ambient mechanical vibration.

The theoretical maximum energy that can be extracted from a vibration source can be
estimated by calculating the energy dissipated through an equivalent damper. Figure 2.3
shows the schematic for a single-degree-of—freedom mass subjected to ground motion. An
electrical energy harvester is included via an equivalent damper in parallel to the
mechanical damper.

The governing equations for the theoretical energy harvester shown in Figure 2.3 are

given by the standard dynamic equations:

Mii + (cm + C6012 + ku = —Miig (2.9-3)
1

Energy = [08122 dt (2.9—b)
0

20

where M is the vibrating mass, cm is the mechanical damping, ce is the equivalent electric
damping, k is the spring stiffness, x is the unknown relative displacement of the mass, and
rig is the prescribed ground acceleration.

A basic layout for a piezoelectric energy harvester is shown in Figure 2.4. The ambient
vibration produces base acceleration, fig, and this induces motion in the piezoelectric
cantilever beam.

Spring (k)

    

Dampers (cm)
Mass (M)

//

Dampers (Ce)

////////

W

1.4.

..

fl

Figure 2.3 Theoretical energy harvester with an equivalent electric damper

 

It,

 

 

 

 

 

 

 

 

' I

 

 

   

Figure 2.4 Schematic of a piezoelectric vibration scavenger of mass (M), length (L),
width (b) and height (h). The equivalent mechanical single degree of freedom system is
also shown

The material strains are converted into electrical charge through the piezoelectric effect.
The generated charge is dissipated through the electronic circuit and thus acts to dampen

out the vibration. The mathematically rigorous generator modeling approach typically

leads to a set of equations of the form:

21

Mii+cmri+kpu—®v=—Mii
Ori+Cv=I

g

t (2.10)
Energy = [vl'dt
0

where kp is the beam stiffness, O is the electromechanical coupling coefficient, C is the

piezoelectric capacitance, x is the tip displacement of the beam with respect to the base, v
is the generated voltage, and I is the piezoelectric current. The remaining symbols have
the same meaning as in Equation (2.9). For a full description of piezoelectric beam
modeling, the reader is referred to work in the literature [47 , 48].

In general the complexity of solving Equation (2.10) arises from including the circuit
current (I). By assuming that only a resistor (R) is connected to the generator, i.e. the
“Circuit” in Figure 2.4 is made up of an equivalent single resistor, the current is
expressed as I=-v/R, and Equation (2.10) can be solved using a numerical approach such
as the Runge—Kutta method. In most instances however, a more complicated non-linear
circuit is connected to the generator. Such circuits are significantly more difficult to
analyze. The response of the electric circuit is typically solved. using standard electrical
engineering software such as SPICE. Accounting for the piezoelectric element within the
circuit simulator remains a problem. Several two port models [45, 49, 50] and a 5 port
piezoelectric model [46] have been proposed.

This work presents a simple one degree of freedom model (based on the previous work
of [46, 48, 49]) that can be readily incorporated into any circuit simulation program (e.g.
SPICE [89]) by using an equivalent circuit for the mechanical structure of the
piezoelectric sensor. The presented model provides an accurate description for a single

vibration mode that includes the backwards and forwards coupling between the electrical

22

and mechanical systems. The performance of the piezoelectric model attached to a

relatively complex non-linear circuit is also compared with experimental results.

2.3.2.2 Simplified Mechanical and Equivalent Electrical Models
Most ambient vibration environments consist of a relatively narrow band vibration
source characterized by a single dominant frequency. In this case a single degree of

freedom damped oscillator, as shown in Figure 2.4, can capture the mechanical response
adequately. The mechanical and electromechanical parameters (K, M, C,,,, and 0) can be

either derived using simplified approach such as the Rayleigh-Ritz ([44, 47]) or by more
complicated finite element methods. In many cases rectangular piezoelectric bimorph
cantilevers generators are used. Table 2.2 summarizes the equivalent mechanical
parameters and the Rayleigh-Ritz approximation. The mechanical parameters can also be

measured experimentally as discussed below.

Table 2.2 Generator parameters assuming an Euler-Bemouli cantilever bimorph. The
displacement shape function is assumed to be, w: u(t)[l—cos(nx /(2L)] where x is the

horizontal distance from the support. The electrical potential shape function is given
byl/l = zv/h, where z is the vertical distance from the beam center, i.e, the beam’s
displaced shape. E is the modulus of elasticity of the beam, e31 is the piezoelectric

coupling coefficient, a is the electric permittivity, I=bh /12 is the second moment of area
and 17 is the mass per unit length.

 

 

 

 

 

 

 

 

 

 

Parameter Approximation for Piezo Bimorph
K EIrr4
32L3
M 372‘ - 8
M - +thL
G 8311);”?
2L
C
26(yi]
h

 

23

Equation (2.10) can be represented by the equivalent electric circuit of the mechanical
and electromechanical nature of the harvester as shown in Figure 2.5 (see [46] and [35]).
The advantage of this model is that it can be analyzed easily using only a standard circuit

simulation program such as SPICE.

N:1=o:1 M c

 

l/K

-t
i I

Figure 2.5 The equivalent circuit for a single degree of freedom piezo—generator. The
mechanical force input is simulated as a voltage generator element with magnitude Miig

 

 

2.3.2.3 Experimental Validation of Models
A- Using Lead Zirconate Titanate (PZT) in a High Frequency Environment

A relatively simple experimental method of measuring the generator parameters is
illustrated in Figure 2.6. The piezoelectric generator is attached to a mechanical vibrator
and its open—circuit voltage response is measured over a range of input frequencies, in the
region of the resonant frequency of the generator. The open-circuit voltage can be
measured using a high impedance oscilloscope. This-experimental method is applicable
for the measurement of any generator configuration. The tip mass and capacitance of the
piezoelectric generator can be measured directly. From Table 2.2, approximately 20% of
the piezoelectric mass can be added to the tip mass to give a sufficiently accurate

estimate of the effective mass.

24

 

I - Accelerometer

 

Funcfion
Generator

 

 

 

Oscilloscope l \ I

l

 

 

 

L

Linear
Stroke
Motor

 

 

Figure 2.6 Method of experimentally measuring piezoelectric generator parameters. The
vibrator is driven over a range of frequencies to obtain the open-circuit voltages. The
accelerometer is used to measure the base excitation iig directly

For the open—circuit case, Equation (2.10) can be reduced to

Mii + emu + (K +02 m). = —Mz'ig

 

v=_911 (2.11)
C
|TF|=®MI 1 I

C I— Ma)2 + me," + (K + o2 /C)I

where ITF] is the magnitude of the voltage-acceleration transfer function and a) is the
driving frequency of the excitation.

Figure 2.7 shows an experimentally measured transfer function for a piezoelectric
35mm diameter unimorph PZT disk (CEB-35D26, CUI lnc., Tualatin, OR) clamped on

one edge. A non-linear least square curve fitting routine can be used to estimate the
stiffness (K), damping ratio (g) (cf =Cm/2xl MK ) and mechanical coupling coefficient

(8). It is suggested that the stiffness and damping ratio are then independently checked.
by measuring the open circuit voltage in a ring-down (free-vibration) test.
The equivalent circuit model shown in Figure 2.5 can be validated by using a standard

full-bridge generator circuit as shown in Figure 2.8. The full-bridge diode circuit is typic-

25

|Transfer Funetionl (Vlg)
0|
9

 

40;
30;-
20
10;
48 50 52 54 56 58 60
Frequency (Hz)

Figure 2.7 Open-circuit piezoelectric voltage transfer function (the least square fitted
model is shown as a solid line, and the experimental results are shown as dots). The non-

linear least square fitted parameters are K=985N/m, 0=6xlO-4C/m, and §=I.3%. The

measured parameters are M,,-p=9.4g and C=25.2nF. The coefficient of correlation is
r=0.98

ally used to collect the generated electric charge on the storage capacitor (C5). A drainage
resistor (Rd) can also be attached to the storage capacitor to represent leakage through the

rest of the circuit and through CS.

Using the same mechanical vibration setup as shown in Figure 2.6, except now
connecting the harvesting and storing circuit of Figure 2.8, the voltage on the storage
capacitor can be measured for a set time over a relatively wide range of acceleration input

frequencies and magnitudes. As explained above, an electric simulation program (e.g,

26

SPICE) can be used to predict the response of the harvester and electronic circuit. The
measured and calculated voltages over the storage capacitor are compared in Figure 2.9-a
for the disk unimorph piezoelectric generator described in Figure 2.7. The simulated and
experimental time history of the voltage stored on the capacitor show good agreement
(Figure 2.9-b). For smaller storage voltages, the simulated and measured results tend to
show worse agreement. A number of factors contribute to this worse agreement,
including the resolution error of the measurement equipment (accelerometer etc.), and the

low-voltage behavior of the diode model.

Rd I—CS

 

 

 

 

 

%

 

 

 

1.1....
J'l'l
a;

Figure 2.8 A full diode bridge piezoelectric circuit for storing electrical charge on a

capacitor (CS), the circuit leakage is represented by the resistor (Rd)

It should be noted that, in general, PZT-based energy harvesters would be significantly

more efficient in converting mechanical energy to electrical energy due to their higher

electromechanical coupling coefficient (e = 20 C m_2 for PZT compared to 0.07 C m—2

for PVDF). However, the brittleness of PZT-based sensors makes them difficult to use
for practical applications. The damage strain for PZT is on the order of 500 pa (compared
to a yield strain of 1.5% for PVDF). Once over—strained, PZT loses a significant fraction

of its piezoelectric properties, even without any visible cracking. Several PZT bimorph

27

Simulated Voltage (V‘)

 

99 (V)

C Volta

 

 

Figure 2.9 (a) Measured and calculated voltages for the piezoelectric unimorph with
properties calculated from Figure 2.7. The input accelerations are compared at 5 seconds
after the linear stroke motor is switched on. The circuits elements properties are: storage

capacitor CS = 3.3 11F; drain resistance Rd = 10 M0, and diodes 1N4148. The line
represents equal measured and calculated voltages, the coefficient of correlation,
assuming the simulated voltage is equal to the measured voltage is r=0.97; (b) a
simulated (solid) and experimental (dashed) voltage time history on the storage capacitor

28

specimens were over-strained during routine handling in the laboratory. Furthermore, the
elastic modulus of PZT is approximately 25 times greater than that of PVDF. Thus, for
the same geometric dimensions, a PZT beam would have a resonant frequency of
approximately five times greater than PVDF. This becomes a significant effect in terms
of energy conversion for low-frequency excitations that are typical for most civil
engineering structures. It is possible to overcome the brittleness issue of PZT by using
fiber geometries and embedding them in epoxy to form advanced composite materials.
However, this procedure significantly complicates their use, and is beyond the scope of

this thesis.

B- Using Polyvinylidene Fluoride (PVDF) Under Low Frequency Environment

To check the validity of the energy harvester model under dynamic loading, simplified
experiments were performed. The experimental results were then compared to the
predictions of the theoretical model. The first experiment consisted of measuring the
voltage transfer function of a piezoelectric scavenger. The piezoelectric beam was subject
to base excitation. In the second experiment, the piezoelectric scavenger was subject to a
simulated earthquake ground acceleration history. The time response of the generated
voltage was then measured.

In all the experiments, a standard commercial unimorph piezoelectric sensor (LDTI-
028-from MSIUSA Inc.) was used. The piezoelectric unimorph consists of a single layer
of active piezoelectric material (PVDF) attached to mylar plastic. The properties of the

material (from MSIUSA Inc.) for the sensor are summarized in Table 2.3. The sensor

29

geometry is shown in Figure 2.10. This material is a polymer and hence much easier to

handle and use than the brittle PZT ceramic.

Table 2.3 Piezoelectric Sensor Properties

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Piezoelectric (PVDF) Properties

Modulus of Elasticity Cp=3GPa
Capacitance C=l .38nF

Piezoelectric coupling coefficient E=0_07C/m2

Density p_p=1780kg/m3

Piezoelectric thickness hp=28um

Length L = 31.7mm
Width b=16mm

Backing (Mylag Properties
Modulus of Elasticity cb=3,79GPa
Backing thickness h b :17me
Density pb=l 390kg/m3

Resistance 10x106§2

Tip Mass Mn'p =1.2g

Beam Stiffness (Calculated) K =3.6 N/m
Lumped Mass (Calculated) M =1 .23g

Electromechanical Coupling (Calculated) 3:3.69x10’6 C/m
Capacitance (Measured and Calculated) CEXle-3811F and CC A [Fl-41rd:
Damping (Measured) B=5.1x10'3Nm/s
Natural frequency (Measured and Calculated) FEXP = 8.7Hz and fC A L=8.6Hz

 

For the theoretical model, a shape function of the form w = d(tXl — cos(7zx/2L)] for the

deflection of the beam and a potential shape function of (11 = vz/hp are assumed. The

stiffness (K), mass (M), electromechanical coupling (9) and capacitance (C) can be
readily calculated, assuming an Euler—Bemoulli beam and using the Rayleigh—Ritz

approach. These results are summarized in Table 2.3. The damping coefficient is

30

calculated by fitting the theoretical voltage—acceleration transfer function to the
experimentally measured transfer function.

Figure 2.11 shows the experimental setup. A servo-hydraulic actuator was used to
provide a controlled displacement to the unimorph piezoelectric sensor. The piezoelectric
sensor was attached to the actuator’s cross-head using epoxy. Both the voltage across the
resistor (R) and the displacement from the servo-hydraulic machine were measured and

acquired by a data acquisition system.

 

 

 

 

 

1|]—

«g/
g 1

“Wu- L

Figure 2.10 Schematic representation of a PVDF vibration energy scavenger

V

 

Controlled
Displacement

 

 

 
 
 

 

 

 

 

 

Actuator
/ Cross-Head

Data Ac uisition . .
q W\Prezoelectnc
Sensor

 

 

 

 

 

 

 

 

 

 

 

 

Sensor Voltage

 

Resistor R

 

 

Figure 2.1 1 Experimental setup to test the piezoelectric bimorph generator

31

The transfer function was obtained by manually inputting a number of different
sinusoidal displacements and measuring the output sensor voltages. The magnitudes of
the experimental and theoretical voltage—acceleration transfer fimetion are shown in

Figure 2.12. The theoretical model shows good agreement with the experimental results.

 

O5
O

Voltage/Acceleration (Vlg)
A
O

 

 

 

20 4
O
o . . a . .
4 6 8 10 12 14 16
Frequency (Hz)

Figure 2.12 Experimental (dots) and theoretical (line) acceleration—voltage transfer
function for piezoelectric energy harvester

In the second experiment, the 1940 El Centro earthquake’s displacements were input
into the actuator. The El Centro earthquake’s displacement time-history is obtained by
doubly integrating the acceleration record and filtering the very low—frequency
components (a cut-off frequency of 0.5 Hz was used). Time-history records of El Centro
are shown in Chapter 3 (Figure 3.3). The filtering eliminates the large deflection error

(i.e. drifi) caused by the numerical double integration of the acceleration record. Since

32

only a small fraction of the El Centro earthquake’s energy is less than 0.5 Hz, this
filtering should not have a significant effect on the energy calculations. Figure 2.13
shows the theoretically simulated and experimentally measured voltages for part of the El
Centro earthquake. The calculated converted electrical energy for the entire 503 El Centro
earthquake is 0.43 ,uJ, while the measured electrical energy is 0.49 ”I. These values are in

good agreement.

 

Theoretical

I

carer»

Voltage(V)

 

 

 

 

Experimental

Voltage(V)

 

 

 

Time(s)

Figure 2.13 Theoretical and experimental results for the piezoelectric energy generator
subjected to the El Centro earthquake (only 6 s is shown for clarity)

33

2.4 CONCLUSION

This chapter has presented simple methods of modeling and estimating the performance
of piezoelectric energy harvesters. A two step approach can be used to estimate first the
electromechanical piezoelectric properties (either using a simple Rayleigh-Ritz model or
by experimental measurements). The estimated piezoelectric properties can then be
incorporated into an equivalent electric network that can be readily solved by any circuit
simulation program such as SPICE.

The proposed mechanical and electrical models show good agreement with measured
circuit performance for a typical non-linear circuit.

Since only one degree of freedom models are used for the vibration piezo-harvester, it
is limited to analyzing relatively narrow band excitations and generators without closely
spaced mechanical vibration modes. For the case of a rectangular cantilever generator
analyzed in this study, it can be shown that the second generator mode has a resonant
frequency of approximately nine times the first. Thus in the unlikely situation where the
vibration source has a significant frequency component at the second generator mode, it

can be ignored without sacrificing the accuracy of the proposed method.

34

CHAPTER 3

FEASIBILITY OF STRUCTURAL MONITORING WITH
VIBRATION POWERED SENSORS

3.1 INTRODUCTION

The ability to monitor infrastructure using a sensor network could provide significant
benefits. For example, the load history could be used to predict the life expectancy of the
structure. The structure’s vibration signature can be used to identify the location and
extent of damage. Currently, a number of techniques exist to monitor large-scale
infrastructure [26, 27, 29]. These include conventional resistive strain gauging, embedded
or attached optical fiber sensors, accelerometers and linear variable displacement
transducers. The major issues with these monitoring methods are implementation cost,
ruggedness/robustness, data interpretation and, most importantly, powering. The
implementation cost of a single sensor is gradually decreasing with the introduction of
wireless sensor networks which can eliminate some of the implementation problems and
costs associated with providing hardwired communication lines [9].

One of the major problems associated with implementing wireless (and wired) sensor
networks is providing continuous power to each of the sensor nodes. To date, batteries
have been the primary means of powering wireless devices. Battery power can last
anywhere from a few days to 10 years. For practical infrastructure monitoring, it is
desirable that the sensors last the service lifetime of the structure; typically, a structure’s
lifespan is several decades, but it can be as long as 100 years. For example, one study of
concrete bridges has shown that their expected lifetimes range from 44 to 83 years [51].

Even with the continued trend of decreasing the power consumption of electronic

35

systems, it is likely that battery self-discharge would limit the useful lifetime of any
sensor to 10 years. For structures and embedded sensors, it is impractical and costly to
change batteries periodically through the lifetime of the structure.

The problems associated with powering wireless sensors have led several research
groups to investigate self-powering as a means of providing continuous power to remote
sensors. In the context of this thesis, self-powering refers to the ability of the device to
‘harvest’ (sometimes referred to as ‘scavenge’) ambient energy from its environment, and
convert it to electrical energy.

A number of energy harvesting techniques are being investigated and implemented.
These techniques are based on one of the following: solar energy, thermal gradients, and
vibration energy. The most successful energy harvesters to date use solar power.
However, for structural monitoring, where sensors could be embedded (as in concrete
structures), it is not always feasible to use solar powering. Using heat or thermal
gradients provide energy levels that are too low to power sensors for infrastructure
monitoring [52] (see Table 3.1). Currently, several research groups are deve10ping energy
harvesters based on ambient vibration energy [52, 48].

The aim of this chapter is to investigate the electrical energy that can be generated from
motion in high vibration civil environments and typical civil structures such as bridges
and buildings. The generated energy levels will be produced by: (1) routine (service)
loading, and (2) extreme loading conditions, in particular earthquakes. Vibration histories
used in this work have been gathered from published data. The mechanical—to-electrical

energy conversion for a piezoelectric generator is modeled theoretically (Chapter 2). The

36

 

 

 

 

 

29233. 38556 on com ooeseoev 23o Loam
WE ooom .862 m9. mow Aofiwao “no: 80E; 05—080
oBflfi>< bees—Eco o v.2 €555 oEmomemaucb mocotmm
29m=§< bquEoU N. mm A8255 $358-85 motofimm

$2 SON 8:88“; use xenonm
0mm 0mm 3.50283me Egm

:2 09.: 5&3

 

Essa» o. 2 © 2

Seem o. 2 ® 2

E2930 Sagan—Eek

 

 

 

 

booze S 2 Sega; aeoe ram
so 2: ® 03 so 2: ® Rd
852 05:84.
page E E @ 83 so we ® 80.0
.925 as. 8583.2. 832 832 access.
“soEtoaxm x3e ooEo -o 8328 850 .0 @8025 50m

 

ofimam>< b85800

Eu .3503 -02

as see -80.:

new €320 -02

:3. see -80.:

980959 Eom

 

 

5.5—.585 «o 3.55

 

oars: a; 2

8.5393 £23: .538.—

 

oEfloE :3? —

$323.: .3259 eaten

 

 

 

32 mooSom wimfigeom 38am mo acmteaaoo mm 2an

 

37

required energy consumption for performing various electrical processes in a low-power
sensor is also reported. These required energy levels are compared to the generated

electrical energy.

3.2 ELECTRONICS AND SUBSYSTEM POWER CONSUMPTION

The power consumption and size of integrated circuits are continuing to decrease, while
their speeds are increasing. It is now becoming feasible to embed electronics in everyday
objects to enhance their usability and performance. In structural monitoring, a typical
wireless sensor node would consist of the sensor, an embedded microprocessor, digital
logic circuits, radio receiver, radio transmitter, a timer and a multi-channel analog-to-
digital converter. Current commercial electronics have sleep-power consumptions as low
as 200 nW (for example, the MSP43O microprocessor by Texas Instrument). The same
processor is capable of operating at 0.5 million operations per second, consuming 350
,uW. Commercially available whole sensor nodes (such as Telos by Moteiv Corp.) require

approximately 50 mW to run their sampling, processing and radio transmission functions.

. . . . 3
These commerc1al sensor nodes typically process and transmlt Wireless data at 250x10

bits s—l. This would allow the sensor to process and transmit approximately 500 bytes of

data per millijoule. The central question is how much energy can a power harvester
generate?

Table 3.2 shows the approximate energy consumption of commercial and research
electronics. It should be noted that the power consumption shown in Table 3.2

corresponds to a continuous energy supply.

38

Table 3.2 Approximate energy consumption in a wireless sensor node (references in
brackets). Volatile memory (VM) is assumed to be replaced by low-power non-volatile

 

 

 

 

 

 

 

 

 

 

memory (NV M) [56]

Operation Commercial Research
Microprocessor Operation+ 350 pW [62] 20 [1W [57]
Receive 1 bit 200 n] [65] 12 p] [59]
Transmit 1 bit 200 n] [60] 16 p] [59]
Retain Volatile Memory (VM) — 1 bit 100 pW [62] Replaced by NVM
Write 1 bit to NVM 200 n] [61] 25 p] [63]
Analog—Digital (A/D) Conversion — lbit 2 n] [62] 50 pJ [59]
Sleep 300 nW [62] 5 nW [59]
Digital Signal Processing+ 200 pW [64] 20 uW [57, 58]
Pin Leakage 100 nW [62] 2.2 nW [66]

 

 

 

 

 

+ Operation is scaled to processor speeds of 500 kHz

3.3 PRACTICAL CONSIDERATIONS

The vibration harvester should allow volume for the beam element, suspended mass,
electronic circuits and space for deflection under vibration. For example, consider a
piezoelectric energy harvester shown in Figure 3.1. Significant volume is taken up by

actual displacements of the beam and mass under base excitation.

   

Electrical
Circuit

  

Figure 3.1 Schematic of a piezoelectric vibration scavenger

Practical manufacturing considerations would also have to be taken into account. For

example:

39

(a) It might be impossible to match the optimum beam stiffness with current ceramic
piezoelectric materials.

(b) Handling thin ceramic piezoelectric wafers is difficult due to their brittleness.

3.4 VIBRATION ENERGY HARVESTING IN TYPICAL CIVIL
ENVIRONMENTS

Vibrations are induced in structures due to commonly occurring dynamic loads such as
wind, traffic movement, and more severe loadings such as ground motion (earthquakes).
F igures 3.2 and 3.3 are included to demonstrate the magnitude and duration of typical
ground accelerations to which the power harvester can be subjected. It is well known that
most civil engineering structures have fundamental vibration modes that occur at
frequencies less than 5 Hz. Furthermore, typical dynamic load frequency spectra range
from 0 to 5 Hz. Structural accelerations have been measured for different structures,
subject to typical loading conditions by many research groups. One example acceleration
time trace and frequency spectrum for a concrete bridge with traffic loads is shown in
Figure 3.2 [67]. Monitoring behavior and performance under severe loading conditions
such as earthquake or blast loading is paramount to assess the safety of the structure. A
collection of ground accelerations for a large data set of earthquakes is available from the
US Geological Survey; this data set can readily be used to test power harvesting schemes
in monitoring civil structures. One well studied and commonly used earthquake time-
history benchmark—the 1940 El Centro earthquake—is shown in Figure 3.3(a). Figure
3.3(b) shows the acceleration frequency spectrum for this earthquake.

In general, acceleration magnitudes at a particular location of a structure are affected by

40

 

0.04 - —

Deck Acceleration (3)
° P
O O
O M

.b
O
N

 

 

 

 

 

  

 

 

 

 

0.04 ~ ~
0 5 10 15 20
Time (5)

0.0030 . . .

0.0025 — -
R 0.0020 — —
O
‘0
.3 0.0015 ~ -
E
G
E

0.0010 L

0.0005 —

o I 1
0 5 10 15 20
Frequency (Hz)

Figure 3.2 (a) Time-history and (b) frequency spectrum for the acceleration
measurements from the deck of a concrete bridge under traffic loading [67]

41

0.4

 

P
N

  

 

Ground Acceleration (9)
g: o
N O

 

 

-0.4 r r I .

 

Time (s)

 

I

0.016 _

0.014 ~ ‘

 

Magnitude (g)
9 .° P
g 3 3
o N

E

 

 

P

0 5 10 15 20 25
Frequency (Hz)

Figure 3.3 (a) Ground acceleration time-history and (b) frequency spectrum from the El
Centro earthquake (N orth—South direction)

42

loading magnitude, structure type, construction materials and methods.

3.4.] Harvestable Power

For this study, published data has been used as input to the convertible electrical energy
model during service and extreme structural dynamic loading. As both energy scavenging
techniques and power conversion efficiencies improve with future research, the
maximum energy conversion will not exceed the maximum theoretical power level

calculated from Equation (2.9). In this study, the maximum volume of one sensor node is

constrained to be 5 cm3. This relatively small volume is approximately the size of a large

concrete aggregate (roughly 20 mm in diameter [37]). Thus, existing concrete
manufacturing methods can be used to incorporate the sensor within the structure. This is
also the approximate size of other currently developed embedded sensors such as the
‘Smart Pebble’ [68]. It has been clearly shown that the mechanical properties of
inclusions in concrete have a significant effect on its strength [69]. Thus, limiting the size
of the sensor would reduce the impact of the sensor on the overall strength of the
structure. Neither sensor cost nor implementation feasibility is considered.

For a simplified pure sinusoidal structural vibration, with acceleration magnitude A and
frequency a), the maximum possible mean power that can be extracted is readily derived

from Equation (2.9) and is given by:

 

P- CeszzMz

— 3.1
2[a)2(ce +cm)2 +(Ma)2 -k)2] ( )

The power (P) given by Equation (3.1) is plotted for various vibration frequencies, (1), in

Figure 3.4. In this Figure it is assumed that the mechanical vibration resonant frequency

43

is 1 Hz, the amplitude of the input acceleration is 0.1g, the mass is 10 g, and the
mechanical damping ratio (Cm) is 5%. Figure 3.5(a) shows the maximum vibration
magnitude (when the resonant frequency of the energy harvester is equal to the frequency

of the applied mechanical vibration) for various mechanical (Cm) and electrical damping

ratios (Ce). Figure 3.5(b) shows the normalized frequency bandwidth for the various
mechanical and electrical damping ratios. The bandwidth (BW) is taken as the frequency
difference between the points where the power is Pmax /\/2 . In general, the bandwidth

increases with increasing total damping.

2000. _ WM -__

l n

 

 

Power (uW)
8
G
__‘?_d a

 

 

(540%

 

 

 

 

-Ww-fi# ,_.__2_r2 --.2. W - L_._.___"_‘ __ We“

0.0 0.5 1.0 1.5 2.0 2.5 3.0
Frequency (Hz)

Figure 3.4 Theoretical average power output for various electric damping ratios. The
mechanical damping is assumed to be 5%.

44

 

 

 

 

 

 

 

 

 

 

 

 

 

I ("1:20/0
E 3000] a]
a | z
E I 0.54% I
o 2000] l
n.
I \ zm=6°/o ]
I \zm=8% I
E
OIL # - 2 - . -2 ..-- 2 _-a_ .2 .. -_ 0 an? L. ,2 * -i‘__ _W, ,2 ,_L_,.'
0 10 20 3 4 50
Electric Damping RatioCe(°/0)
(a)
oarw~-~- -_,i-a,-p, .21----2]
a l
i
0.6I~

Bandwidth (Hz)
0
‘3

 

 

0.2 F
I l
0l___-. _ _. 1.- 2 .2... .2 1-.“ __- _2 l_-_ __ -__ __L F- __.__ _ I
0 10 2 30 40 50
Electric Damping Ratio §e(°/o)

(b)
Figure 3.5 (a) Maximum power and (b) bandwidth for various electrical and damping
ratios

45

From Figure 3.5, it can be seen that there is an optimal electrical damping ratio for

which the magnitude of the generated power (Pmax) is a maximum. For the values

assumed, the optimal electrical damping ratio (C6) is 5%. It should also be noted that the

bandwidth of the harvested power is also an important metric since, in general, the input
spectrum for dynamic loading is not given by a single frequency but by a frequency
spectrum. Thus it is possible that the maximum power that can be generated from a
particular dynamic excitation would be greater for an energy scavenger with less
maximum power output at resonance but with a broader bandwidth. Equation (3.1) shows
that increasing the sensor vibrating mass increases the maximum power that can be

generated.

3.4.2 Maximum Theoretical Harvestable Power
In this work, half the sensor volume is assumed to be occupied by the mass. The rest of

the volume is taken by electronics and free space to allow for vibration. Assuming a

density of 10 000 kg m—3 (the approximate density of heavy metals such as silver, lead

etc), the 5 cm3 volume constraint gives a maximum tip mass is 25 g. Structural damping
is typically between 2 and 5% of critical damping, thus the mechanical damping is taken

as 4%. The maximum deliverable power for any vibration excitation can be found by

searching for the optimum stiffness (k) and electrical damping (Ce).

In the calculation for the magnitude of energy that is harvestable under various loading
conditions, the Wilson—Theta time-step integration method was applied to Equation (2.9-

a), to obtain the response of the single—degree-of-freedom system. The total kinetic

46

energy for the loading time-history was then computed by numerically integrating
Equation (2.9-b).

Three dynamic loading types were considered: earthquake, wind and traffic. The
earthquake data was obtained from 32 different earthquakes (COSMOS - Consortium of
Organizations for Strong Motion Observation Systems). Acceleration data for wind and
traffic was obtained directly from researchers [70, 67] and from published data [71, 72,
73].

The resulting maximum harvestable energy is plotted in Figure 3.6. The results are
segregated by structure type, subject to various loading conditions. It should be noted that
the extreme earthquake event (labeled ‘1’ in Figure 3.6) is for the Llolleo earthquake
which occurred in Chile in 1985. The Llolleo earthquake is one of the longest large-
magnitude earthquakes in recorded history. Also note that the maximum energy that
would be harvested for tall buildings (labeled ‘2’) is for a 44—story building subjected to a

small earthquake [71]. Point 3 in Figure 3.6 is the maximum energy that would be

harvested from the flexible Calgary tower subject to 100 km h“1 wind gusts; this wind

speed corresponds to 62% of the design wind load [73].

As noted previously, the sensor stiffness and electrical damping levels were Optimized
to provide for maximum electrical energy output. In Figure 3.6, the optimal stiffness for
maximum electrical energy output ranges from 0.1 to 20 N/m. The optimal electrical
damping coefficient ranges from 5% to 43%. Electrical energy is presented instead of
electrical power since, in all investigated cases, the loading events tend to be discrete (i.e,
lasting a finite time). Thus it might be misleading to quote average power, which could

imply that continuous electrical power is available for the sensor node. For the wind-

47

loading cases that tend to be more continuous in nature, the loading record was calculated

for a period of 305. In the earthquake and traffic loading, all loading durations ranged

 

 

 

 

 

 

 

 

 

 

from 30 to 60s.
100 'I ——————————————— 1: T ————————— —l
. O Earthquake
. [:1 Wind
2 .
A 10 - —————————— =o== ~~~~~ —:—_._é.l[§ffl§____
'1
E. a 3
a l
a; 1 7 —————————— —.-fir — — —- — ——————————
F I
u.l —A—
a B i
0.1- —————— §‘--———D—-———4—— __________
A a
Bridges Tall Earth-

Buildings quakes

Figure 3.6 Maximum theoretical vibration energy for various structures and loading
conditions

3.4.3 Piezoelectric Energy Conversion Efficiency

. . . . 3

The energy levels shown 1n. Figure 3.6 are the maxrmum achievable values for a 5 cm
vibration scavenger (i.e, it assumes that all the available kinetic energy is converted to
electrical energy). In general, the amount of energy available to drive the circuit sensor is

dependent on how efficiently the vibration is converted to electrical energy. This

conversion is a function of the electromechanical coupling of the generator. Assuming no

48

losses in the circuit caused by unwanted leakages and nonlinearities, Equation (2.10) can
be integrated to calculate the generated electrical energy.

The PZT-5H piezoelectric material properties are used in the current simulations. This
material has one of the highest electromechanical coupling coefficients, thus providing a
good approximation of the upper bound to the electrical power that can be harvested by
commercially available piezoelectric materials.

The geometry of the sensor is assumed to be as follows. The sensor has overall

dimensions of 50 X 10 X 10 mm3 (the volume is 5 cm3). The actual beam dimensions are:

length L = 30 mm, thickness h = 0.2 mm, and width b = 10 mm. The maximum allowable
tip-deflection of the beam is constrained to 2 mm since, under larger deflections, the
mechanical strains can cause the PZT to lose its piezoelectric properties. The maximum
tip mass is assumed to be M = 25 g, as explained in section 3.4.1 above.

For these dimensions and mass, and a PZT piezoelectric cantilever beam, the resonant

 

frequency is approximately 7 Hz (Given byf z JEbh3/l6zz‘2L3M , where E is the elastic

modulus for the piezoelectric material). However, most civil structural dynamic loadings
have extensive vibration components at around 1 Hz. Since this frequency is far from
resonance, it is evident that the sensor will not be effective in converting most of the
mechanical vibration to electrical energy. Although matching the dominant frequencies
might be possible, by using different sensor geometry and materials, the maximum
allowable tip displacements and piezoelectric failure strains would significantly constrain
the magnitude of generated energy.

The generated electrical energy is calculated by solving the piezoelectric equations

(Equation (2.10)) using a Runge—Kutta time-step integration routine. It has been shown

49

previously [48] that the extractable electric power is strongly dependent on the load
resistance. The optimal load resistance to maximize the generated charge is found using
the optimization algorithm given as follows:

(1) Estimate or measure the base acceleration time history.

(2) Set the tip mass, length, width, material and damping properties of the generator.

(3) Choose an initial estimate for circuit resistance (R).

(4) Calculate generator parameters K, C m, @and C.

(5) Solve for the total energy in Equation (2.10) using a numerical integration
technique such as the Runge-Kutta algorithm.
(6) Modify the resistance (R) and repeat step (4) and (5) until the energy is
maximized
The results for the piezoelectric extractable energy are presented in Figure 3.7; the same
loading histories and structures as in Figure 3.6 are used.

Comparing Figure 3.7 to Figure 3.6 shows that, in general, the converted energy is
much lower due to weak electromechanical coupling of the piezoelectric material. The
coupling is strongly dependent on the resonant frequency of the host structures and on the
resonant frequency of the harvester. For example, the tall buildings analyzed here
generally have resonant frequencies under 0.5 Hz. Since the energy harvester has a
natural frequency of approximately 7 Hz, it is evident from the transfer function shown in
Figure 3.4, that very little energy will be coupled from the structure to the harvester. This
results in the extractable energy from tall buildings being approximately 104 times less
than the maximum available energy. For earthquakes, the dominant excitation frequency

ranges from as low as 0.2 Hz to as high as 10 Hz. At the higher excitation frequencies,

50

more energy is coupled into the harvester and thus more energy can be extracted. On
average, the energy extracted from ground motion is only approximately 30 times smaller
than the maximum theoretical value. For the case of bridges, the natural frequency of
many medium-span bridges is around 5 Hz. Since this frequency is close to the 7 Hz
resonant frequency of the harvester, which can be considered as moderate coupling; there

is approximately 10 times less energy than the theoretical maximum for bridges.

 

 

 

 

 

 

 

 

 

 

102
.Fr' 0 Earth uake
101 - —————————————— —44———————.——q————
l Wind
100 _ _____________ _____ __A__T@ff_l<.3____
:;
E, 10-1 ~ ———————————————— — ——————————
>4
9 t
2 10-2- _____ . ~ __________ L3...— _________
J.I.I 3
10-3 4 _____ _______..D=./_.2__.-_ __________
._l.._.
0
10-4 J_____4 _______ __ _____ ___ __________
10-5 E
Bridges Tall Earth-

Buildings quakes

Figure 3.7 Electric energy output from a piezoelectric bimorph for various dynamic
excitations

For the theoretical case (Figure 3.6), the structure loaded by earthquakes (point 2)
shows larger output energy than for wind loading (point 3). The case is the reverse for the
more realistic piezoelectric bimorph (Figure 3.7). This is due to the fact that the

fundamental frequency of the Calgary tower (point 3) is 0.36 Hz, while for the 44 story

51

building (point 2) it is 0.23 Hz. This again illustrates that even small relative frequency
shifts of the host structure, towards the resonant frequency of the sensor (7 Hz) can

produce large changes in generated electrical energy.

3.5 ENERGY HARVESTING IN HIGH VIBRATION CIVIL ENVIRONMENTS

Typical civil engineering structures have their first fundamental natural frequency
below 5H2. Furthermore, dynamic loading such as wind, traffic and earthquakes have
significant frequency components below 10 Hz. In the previous section, it was shown that
under such low frequency excitation and using commercially available piezoelectric
materials, it is difficult to generate more than l m] of electrical power even under
earthquake loading. Only very limited sensor operation can be performed with such low
power levels. This section focuses on evaluating the harvestable electrical energy in some
unusual high vibration civil environments.

Previous work [35, 44] has shown the feasibility of continuous circuit powering under
mechanical vibration at high amplitude (greater than 0.1 g) and high frequency excitation
(greater than 50 Hz). Such high frequency loading, though not typical, can be found in
industrial civil engineering structures subjected to impact loading and supporting rotating
and traveling machinery. The actual base excitation spectrum applied to the piezoelectric
generator is then dependent on factors such as the frequency of the machinery, motor-
imbalance, mounting of the machine, and the natural frequencies and damping of the
support structure. In most cases the vibration magnitudes in the supporting civil structure
are constrained by building design codes that include: (1) the ultimate strength of the

structure, and (2) the structures’ serviceability that limits distortion of the structure and

52

the physical discomfort of the buildings’ occupants. In most cases the serviceability
requirements are more limiting than the strength requirements. These requirements are
dependent on the frequency and amplitude of the excitation and whether the vibration is
continuous or transient. Most serviceability design standards (for example ISO-2631 [74],

and AISC ll [75]) limit the low frequency accelerations (defined to be in the 2 Hz to 10
Hz range) to between 0.05 m/s2 and 0.1 Iii/32 depending on the duration of the vibration
(Figure 3.8). Tolerable acceleration amplitudes increase with increasing frequency. In
general peak acceleration amplitudes greater than 1.5 m/s2 are considered to be extreme

[76]. For many civil engineering structures, high vibration loads are ofien transient, and

thus might not be able to provide continuous powering to electronic sensors.

 

 

 

 

 

 

 

 

10 ----fi--, - Hm
"E 10‘ r
E .
~§ 1.5m/s2
E
2 10‘
O
O
O
<
.1:
8 -1
n. 10
l
l
10" r . . W . .
10" 10° 10‘ 102
‘ Frequency (Hz)

Figure 3.8 Approximate serviceability structural vibration limits (based on [74], [75] and
[76]). The shaded region represents the limits based on various exposure times. The
lower boundary represents longer duration vibrations; the upper boundary represents

shorter duration vibrations. The solid line at 1.5 m/s represents extreme vibration
conditions

53

3.5.1 Optimal Energy Harvesting

For a given base excitation, the piezoelectric generator parameters (M, K, Cm, 0, and

C) in Equation (2.10) can be optimized to yield the maximum converted electrical

energy. In general, increasing the tip mass (M), increasing the electromechanical, coupling

(6), and decreasing the mechanical damping coefficient (Cm) tends to increase the

generated energy. However this optimization problem is constrained by the actual
material properties, the strength and the size of the generator. Although not considered in
this study, the cost of the piezoelectric material will ultimately also play an important
role. In order to compare the harvested power to results in section 3.4, the size and shape

of the piezoelectric material are assumed to be the same as in section 3.4. The generator

srze fits into a 5 cm volume, whrle allowrng space for the electronic Circuitry and

generator displacement under dynamic loading. The generator material is assumed to be
PZT—5H which gives the highest electromechanical coupling coefficient of commercially
available piezoelectric materials. The mechanical damping is assumed to be 4% of
critical.

Under the given constraints, this optimization problem has two free parameters. The
first free parameter is the height of the beam (h) which affects: the beam stiffness (K),
generator capacitance (C) and electromechanical coupling coefficient (0). The second
free parameter is the impedance of the sensor, which influences the current (1) drawn by
the circuit. For example, for a purely resistive load (R), the current drawn is given by

I = v/ R . The optimal energy is determined using the same optimization procedure as in

section 3.4.3.

54

As previously mentioned, a myriad of factors influence the harvester’s base
acceleration and thus the input excitation might be difficult to estimate a priori. The
vibration history might change under different operating conditions. Furthermore, the
structural and machine properties could change with time because of damage and wear. It
is thus advisable to optimize the performance of the generator with various estimated or

measured inputs.

3.5.2 Example

To demonstrate both the optimization algorithm and to estimate the amount of
convertible energy in a high vibration civil industrial environment, the support beam
under an overhead crane is investigated in this study.

This structure consists of a continuous steel I-beam, (height: 252mm, width: 115mm,

flange thickness: 13mm, and web thickness: 8mm) supporting a 3 ton crane (Figure 3.9).

 

Figure 3.9 Investigated structure consisting of a steel I-beam with an overhead moving
crane. Position of accelerometer node is as shown

The crane can operate at two speeds (0.8 m/s and 0.3 m/s). The accelerations at mid—span

of the supporting beam (span 4.8m) was measured using a wireless, custom-built,

55

lightweight data recorder (ZeroPoint Technology, Johannesburg, South Africa) at a
sampling rate of 2000 Hz. The accelerometer in the ZeroPoint device was a 3 axis
accelerometer (ADXL330, Analog Devices Inc., Norwood, MA). Throughout this
experiment, all accelerations are measured by this device.

A typical time history and associated frequency spectrum for the acceleration at the
mid-span of the beam is shown in Figure 3.10. The crane is started and stopped at the
midpoints of the adjacent beam spans. The time history and frequency spectrum shows
the transient nature of the excitation (lasting approximately 15 seconds), with a very wide

excitation frequency bandwidth (0 to 200 Hz).

3.5.3 Optimal Energy Harvesting Results
In many cases it is not possible to match the impedance of the circuit to the ambient

vibration. Further, the generated energy is often stored for future use. In one such circuit,

the converted voltage (V5) is stored onto a capacitor (C5) through a full diode bridge as

discussed in section 2.3.2.3. The sensor circuit can then be represented by a drainage

. . . l
resrstor (Rd). In this case, the generated energy is E = ‘Z’CSng .

Assuming the material and geometry properties of the generator shown in Table 3.3, the
maximum energy can be calculated for each beam’s fundamental resonant frequency
using the optimal resistance. The result of this optimization for the over-head crane
moving from left to right at its higher speed is shown in Figure 3.11. Three separate crane
base accelerations were used for the resistive circuit simulation showing good

consistency from case to case. The mechanical natural frequency of the generator can be

56

 

0.5
35
r: 0
.2
8
2
3 -0.5
o
<

 

 

 

 

5 10 15
Time (s)

0M~-fl_h44me_fi____-m

0.015

|Acceleration| (g)
G
3

0.005 -

 

 

 

0‘ 50 ’"400 A 150—‘ 200
Frequency Hz

Figure 3.10 A typical input acceleration time history and the associated frequency

spectrum for a lightweight crane moving at 0.8 m/s, measured at mid-span on the
supporting beam

57

Table 3.3 Piezoelectric and Circuit Sensor Prgperties

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Piezoelectric (PZT-5H) Properties

Modulus of Elasticity CD=63 GPa
Permittivity 6:1 .21x10'8 F/m
Piezoelectric coupling coefficient 9:1 13 C/m2
Piezoelectric Density p=7 500 kg/m3
Tip Mass Mtip=25 g
Length L=30 mm
Width b=10 mm
Damping coefficient {=4%

Circuit Elements

Diodes lN4148
Storage Capacitor Cg=3.3 [1F
Drainage Resistor R filo MQ

 

 

 

 

f fir l 1 l

— Resistive circuit
11 . O Diode bridge circuit _

Energy (mJ)
.0 9
O Q d

.°
4:-

0.2

 

 

 

 

o I I l I I l
20 40 60 80 1 00 1 20 140 1 60

Frequency (Hz)

Figure 3.11 Maximum extractable electrical energy for a resistive circuit (solid lines) and
diode-bridge circuit (circles), for the crane moving from lefi to right at 0.8m/s. The three
solid lines represent acceleration histories from three different crane runs

58

related approximately to the generator’s height, h, using a Rayleigh-Ritz approach by:

~ £1: Ehb (3.2)

f L 1536M

The maximum resistive energy that can be extracted from the crane is approximately
1.17mJ when the vibrator’s mechanical natural frequency is 76Hz, corresponding to h =
0.99mm. The maximum diode-bridge energy that can be extracted is approximately
0.5mJ when the vibrator’s natural frequency is 68Hz, corresponding to h = 0.92mm.
These values are approximately the same when the crane moves in the opposite direction
(right to left), with the maximum extractable energy being 10% less than the left to right

case.

3.6 DISCUSSION

Good electromechanical coupling between the host structure and sensor is critical for
maximum electrical power conversion. For example, the commonly used El Centro
earthquake generates approximately 0.5 )1] from a PVDF unimorph generator. Using a
PZT generator would produce approximately 1.2 m]. In comparison, an electric

generator, as described in section 3.4.1, with perfect coupling would produce 13.9 m].

Sensor nodes need energy to wake up and start collecting and transmitting—this was
not considered in this study. The energy levels for sensor start-up are of the order of 10 p]
[77]. Therefore, using today’s integrated circuits, 10 ,uJ is the absolute minimum energy
required to be converted from structural vibrations for the sensor node to even switch on.

Thus, for PVDF generators scavenging vibration energy on civil structures subject to the

59

El Centro earthquake, no useful electronic sensor firnctions can be performed. For a PZT-
based generator, approximately 5000 data bits could be transmitted, compared to 70 000
bits if perfect electromechanical coupling is realized. Choosing a greater tip mass in a
piezoelectric generator could increase the power harvested. This increase is achieved by:
(a) Directly increasing the kinetic energy (as shown in Equation (2.10)), and
(b) Reducing the resonant frequency of the beam to more closely match the
spectral energy content of the loading.

However, increasing the tip mass would also increase the maximum tip deflection, and
the maximum strain in the beam. As discussed above, this might overload the beam,
resulting in a significant reduction or loss of piezoelectric properties.

The effect of changing the resonant frequency of the beam to match the spectral energy
of the input loading is shown in Figure 3.12 for the El Centro earthquake. By increasing
the length and decreasing the width of the bimorph shown in Figure 2.4 and described in
section 3.4.3, it is possible to decrease the resonant frequency of the beam without
changing the volume of the sensor. Figure 3.12 shows the significant increase in
harvested energy for the El Centro earthquake with better frequency matching. As
mentioned previously, the maximum theoretical harvestable energy for the El Centro
earthquake is 13.9 mJ. Figure 3.12 shows that the optimal resonant frequency of the PZT
generator is between 1.5 and 2 Hz. This closely matches the dominant frequency range
for the El Centro earthquake, as shown in Figure 3.3(b). However, the energy numbers in
Figure 3.12 are theoretical since for a real PZT bimorph the very large deflections (and

strains) at the matched frequency (i.e. 1.5—2 Hz) could lead to sensor damage.

60

 

 

{r —"’— — — ” fl —— "‘ _ — _ ”‘_ T—TT ____ "—T— ~_~ Tl—T—T ”—1

”l 1
l

1 l

3r 'l

A l l

a g j

E 61 ‘l
B r

3 l j
5 I

41? 1

' l

l

2— 1

l

o I

0 2 4 6

Resonant Frequency (Hz)
Figure 3.12 Generated electrical energy from the El Centro earthquake for a bimorph

PZT with various resonant frequencies

Figure 3.13 shows the maximum theoretical energy that can be generated by a range of
traffic loading on a concrete bridge with increasing tip mass (data from [67]). The
shading in Figure 3.13 corresponds to the range of dynamic loading from heavy and light
traffic loads. The average curve corresponds to the average generated energy from all
traffic loading. The generated energy scales approximately linearly with mass. As shown
from the piezoelectric analysis, the maximum convertible energy is of the order of 10 to
100 times less than the theoretical maximum. Assuming that 50 p] is required for a
significant data transmission (10 p] for circuit start-up, 20 ,uJ for data transmission, 10 ,ul
for sensor operation, and 10 ,uJ for microcontroller energy), a mass of 100—1000 g is

needed for adequate circuit powering under average traffic loading. Under the same mass

61

density for the sensor as used in section 3.4.2, this would correspond to a volume
between 20 and 200 cm3. For applications where the generator is embedded in the
structure, this volume might be too large, thus creating problems: (a) in manufacturing

the structure and effectively embedding the node, and (b) by weakening the structure at

the location of the node.

 

3.0 . . .

    
 
 
 

Large dynamic loading

1"
or

.N
0

Average dynamic loading

Energy (mJ)
;

Small dynamic loading

.s
O

0.5 r

 

 

 

0 50 100 150 200
Mass (9)

Figure 3.13 Generated electrical energy for various traffic loadings on a concrete bridge,
assuming perfect electro—mechanical conversion

For the high vibration environment considered (i.e, a moving crane), the optimization
shows that a 5 cm3 piezoelectric generator can convert approximately 1.2 m] of electrical
energy from a single 0.8 m/s crane run. The magnitude of the generated energy is similar
to the optimum energy that can be extracted fiom a mid to large size earthquake. Under
continuous crane movement (backwards and forwards at maximum speed) the average

power that can be harvested is 125 uW. This amount is eight times less than the energy

62

required to operate a basic sensor node continuously. This sensor node would consist of
an efficient commercial microprocessor, data acquisition system, and wireless
transmission link (section 3.2). As previously discussed, the generated electrical energy
scales approximately linearly with generator volume (Figure 3.13). Thus an optimally
designed 40 cm3 generator could continuously power a wireless sensor node today. With
the rapid advances in wireless sensor nodes, it is likely that the 1 mW power level will be
achieved in commercial electronics components within a few years. Thus the use of
piezoelectric power harvesters as energy sources for sensor nodes in high vibration

environments can be expected in the near future.

One important note is that the overall vibration environment of the piezoelectric
generator could change both in magnitude and frequency throughout the lifetime of the
building. For example, the vibration under a crane load can be influenced by the quality
and alignment of the supporting rail, the balance of the wheels, the speed of the crane,
and the stiffness of the supporting structure. To prevent de-tuning, a good piezoelectric
generator design should take into account the possible changes in the input accelerations
throughout the lifetime of the structure. For example, changing support conditions
throughout the life of the structure could lead to de-tuning of the harvester. For a single
dominant base excitation frequency and for a simple sensor circuit (such as a purely
resistive load), the de-tuning effect can be readily calculated [52, 78] as shown in Figure
3.12. However for a relatively wide-spectrum excitation (such as is common in civil
engineering structures), the effect of changing structural support conditions on energy

harvesting has not been investigated adequately. The performance of different harvesting

63

circuits under these changing conditions also needs to be considered. Future research

should focus on quantifying the de—tuning effect.

3.7 CONCLUSION

This chapter set out to answer the following question. Can power harvesters based on
vibration energy be used to power wireless sensors and sensor networks? It was found
that, assuming perfect electromechanical coupling (the results shown in Figure 3.6),
vibration harvesters can produce in the range of 1 ml. (This range ignores extreme
earthquake events, which can produce significantly more energy). This amount of energy
can power a basic wireless sensor; approximately 500 bytes of information can be
transmitted.

Realistically however, when imperfect electromechanical coupling is taken into
account, power harvester’s efficiency can be reduced by as much as 100 times (See
Figure 3.7 for results). Assuming a 5 cm3 sensor node, the 10 ,uJ of energy converted
from average dynamic events such as wind, traffic and low-scale earthquakes, is
insufficient to run modern wireless sensors.

Piezoelectric-based vibration harvesters do produce enough energy from large-scale
earthquakes and high vibration environments; these events can allow for approximately
500 data samples to be collected and wirelessly transmitted by a sensor node.

The wake-up energy of a wireless sensor is on the order of 10 )1] [77]. Thus, for most
typical dynamic events, it would be impossible to harvest enough energy to power a 5
cm3 wireless sensor. Self-powered wireless sensor communication could become more

feasible if this start—up energy barrier is reduced.

64

Increasing the mass (and thus volume) of the energy converter will increase the
maximum power generation. This work shows that, with a circuit energy consumption of

approximately 50,11], the volume of a realistic energy generator would be between 20 and

200 cm3. The actual volume will depend on the available kinetic energy and the

electromechanical coupling between the generator and the vibrating structure.

Current research is striving to reduce sensor node power consumption to the picojoule
range [36]. If this is achieved, then small wireless sensor nodes could process and
transmit data even for very low-energy wind loading on tall structures. However, even if
the sensor electronic power consumption is reduced, significant effort would still be
needed to reduce losses during the energy conversion process. Examples where such
losses occur are voltage rectification, and capacitor and resistor leakages.

Lastly, it should be noted that there are several practical considerations in energy-
converter design. The fracture resistance of the piezoelectric ceramic, the loss of the
piezoelectric properties, and the maximum deflection under dynamic loading could all
play major roles in reducing the overall power generation capacity of electromechanical

energy harvesters.

65

CHAPTER 4

SUB-MICROWATT PIEZO-POWERED ANALOG CIRCUITS:
MODELING AND ANALYSIS

4.1 INTRODUCTION

The previous chapters discussed some of the limitations of self-powering using
piezoelectric transduction. Even though the transducer can generate a large open load
voltage (> 15 V), its capacitive nature (see Figure 4.6) limits the total current that can be
delivered to its electrical load. Typically this current is in the order of nanoamperes for a
loading cycle of 1 Hz or less (section 2.3.2).

The properties of the piezoelectric transducer, however, are attractive for operating
analog floating-gate circuits that typically require high voltage and very low current for
programming. A floating gate is a poly-silicon gate surrounded by an insulator, which in
the standard semiconductor fabrication process is silicondioxide [43, 79]. Because a
floating gate is surrounded by high—quality insulation, any electrical charge injected onto
this gate is retained for long intervals of time (> 8 years). The basic principle behind the
integrated piezoelectric floating-gate sensor is illustrated in Figure 4.1, in which the
floating gate is coupled to a p-channel metaloxide-semiconductor (pMOS) transistor. The
electrical energy produced by the piezoelectric transducer is used to inject electrons from
the transistor channel onto the floating gate. The repeated accumulation of charge on the
floating gate is indicative of the history of mechanical loading and therefore can be used
as an integrated platform for sensing, computation, and storage in structural monitoring
applications. The following sections present an overview of the developed structural

monitoring micro-system based on floating-gate VLSI circuits.

66

SOU I'CG

    
 

i Control
Piezoelectric ' node
transducer :
Floating
gate

drain

Figure 4.1 Principle of the piezo-powered floating—gate sensor

4.2 FLOATING GATE TRANSISTORS

In a standard CMOS process, a floating gate is a poly—silicon gate surrounded by

silicon-dioxide that acts as insulator. When the floating gate is coupled to a gate of a

transistor, as shown in Figure 4.2(a), the cumulative charge stored on the floating gate

can be sensed by measuring the transistor’s drain current.

The charge on the gate can be modified using either hot electron injection or using

tunneling [79]. Hot electron injection in a pMOS transistor occurs when a high—electric

field is formed at the drain-to-channel depletion region. Due to this high-electric field, the

holes gain significant energy to dislodge electrons by hot-hole impact ionization. The

67

released electrons travel back into the channel region, gaining energy and when their
kinetic energy exceeds the silicon and silicon-dioxide (> 3.2 eV) barrier, the electrons are
injected into the oxide. As more electrons are added to the floating gate, its potential
decreases, which results in an equivalent increase in the drain current through the
transistor. Hot-electron injection in a pMOS transistor is a feedback process because an
increase in injection onto the floating gate (decrease in gate potential) reinforces the
injection process. Thus injection is required to be carefully controlled to be able to
perform any useful and sustained computation. Empirically, the injection current has
been found to be proportional to the source current through the pMOS cell and drain-to-

channel (source) potential and can be expressed as:

Iinj = [Sef(Vdc) (4.1)

where f(VdC) is a smooth function of drain-to-channel potential Vdc and I S is the

source current through the pMOS transistor. Figure 4.2(b) shows measured injection
characteristics of a pMOS floating-gate transistor when the source-to-drain terminal is
subjected to 1 second pulses of varying voltage amplitude. For each excitation pulse, the
pMOS transistor injects a packet of charge on the floating gate that results in an increase
in the drain current. Subsequently, a larger packet of charge is injected per pulse as
evident from Figure 4.2(b) reiterating the positive feedback nature of pMOS injection.
This mechanism has been used by several researchers [43, 79, 80] to accelerate
programming of the floating gate to a pre-determined voltage. It has also been
demonstrated that the injection current is practically independent of the floating-gate-to-
channel potential, as long as the gate potential is greater than the drain potential, which is

naturally satisfied for pMOS transistors operating in weak-inversion [43]. For a pMOS tr-

68

Bulk connection control Gate

/r’l/ .
/ ’1

//

\ \ .~ .
.~\ \ y/
\ / / , . ~ "
/ //”/.v‘ / //
ang/flflot/M xxx; //./'/..//./%

 

 

 

s 5‘0» 5,." “\§\\“
E ~08) "'--..,,_Vds =’ 4:55'v ‘~~.
o
t -1 ~
3 .
o
-12 _
-1 4 — Vds =.4.5_\I:""'..
-1.6 —
-1.8 ‘ ' I ‘ I
o 2 . 4 6 8 10
Time (seconds)

0))
Figure 4.2 (a) Top and cross-sectional view of a floating-gate transistor and (b) the
measured drain current when injection pulses of different amplitude are repeatedly
applied across the drain-source terminal

69

ansistor fabricated in a 0.5pm CMOS process and biased in weak-inversion, drain-to-
source voltages greater than 4.2 V have been found to be sufficient for onset of hot-
electron injection [43]. It can also be seen from Equation (4.1) that the injection
efficiency (ratio of injection current to source current) is practically constant over
different values of source current [43]. Therefore, compared to digital FLASH
programming, analog floating-gate programming can be performed with power
dissipation as small as 100 pW during the write cycle. Even though Equation (4.1) has
been expressed in terms of drain—to-channel potential [81, 82], different variants of
Equation (4.1) have been proposed that employ measurable potentials that are the drain,
source, and gate potentials. In this study, it was found that a simplified empirical model is

sufficient for injection-based circuits and it is expressed as:

Vsd
V. .
Iinj “16156 m]

 

(4.2)

with ,6 and V inj are bias-dependent parameters. Hot-electron injection is typically a one-

way process in which electrons can be added to the floating gate.

Removal of electrons from the floating gate can be achieved either through exposure to
ultraviolet (UV) radiation or by use of tunneling. Tunneling is a quantum mechanical
phenomenon whereby the electrons, instead of surmounting an energy barrier, propagate
through it. A tunneling capacitor as shown in Figure 4.2(a) is coupled to the floating gate
and a large potential is applied to initiate tunneling. The magnitude of the tunneling

current has been empirically determined as:

‘17: (4.3)

70

where 1tun0 is a pre-exponential current, Vox is the voltage across the oxide and Vf is a

constant that depends on the oxide thickness. In addition to the tunneling capacitor, the
floating gate is capacitively coupled to other terminals. A model of a pMOS floating-gate

transistor with capacitive elements is shown in Figure 4.3.

 

Figure 4.3 Capacitive model of a pMOS floating-gate transistor

The control gate capacitance CCg serves as the dominant capacitance and is used for

modulating the drain-to-source current. The tunneling capacitor in Figure 4.3 is denoted

by C “m , the parasitic capacitance between the supply voltage and substrate is denoted by
C dd and C sub respectively and the overlap capacitance between the floating gate to the
drain and source terminal is denoted by C dg and C58 . The auxiliary parasitic
capacitance to the neighboring buses is denoted by a lumped capacitance C pa, . All these

capacitors play key roles in achieving uniform and matched injection response in an array
of floating-gate transistors. Based on the circuit model in Figure 4.3, the voltage on the

floating gate V fg can be expressed as:

71

_ Qfg + chch + CddVdd + Csuszub + Cpaerar + Cdng + CsVs

fg _ CT

 

(4.4)

where CT = ch + Cdd + C5"), + Cpa, + Cdg + CS and Qfg is the residual charge on the

floating gate modulated by injection and tunneling. The source-to-drain current through

the pMOS transistor when biased in weak-inversion can then be expressed in terms of the

floating-gate voltage Vfg , drain voltage Vd and source voltage VS as:

W V/g V Vd
I =1 —ex —K—‘— ex —‘3— —ex — 4.5
d5 0 L p( Ur)l NUT) 14%)] ( )

where W and L are the width and length of the transistors, K is the floating gate

efficiency, 10 is the specific current and UT is the thermal voltage (26mV at room

temperature). For Vsd > 3UT , Equation (4.5) simplifies to

W Vfg V
I d, = I 0 —exp(—K———) exp(—S—) (4.6)
L UT UT
which is the saturated mode of operation for the pMOS transistor biased in weak-

inversion.

4.3 F LOATIN G~GATE INJECTOR AND ITS MATHEMATICAL MODEL

The schematic of a single—channel piezo-powered floating—gate sensor is shown in

Figure 4.4. The injection rate of the floating gate transistor Q1 can be controlled by
limiting the source current in Equation (4.2) using a constant current reference I b- Also,

as shown in section 4.5, the drain-to-source voltage can be controlled by inserting diodes

between the current reference and the floating-gate transistor. Let Vg denote the voltage

of the floating gate at a fixed control gate bias voltage V8- The total floating gate

72

capacitance is given by CT according to Equation (4.4). The current reference biases the

floating transistor in weak-inversion satisfying the following relationship:

V
I), = 10 Kexp{— K——g—Jexp[-I:‘fl] (4.7)
L UT UT

 

o-l

F
<

JI

 

O
—'

 

 

 

 

 

Figure 4.4 Simplified circuit model for piezo-driven floating-gate sensor [83]

Provided that conditions are suitable for hot-electron injection (hi gh-electric field at the
drain-to-channel region), the injection current charges the floating gate capacitor
according to:

(IV V
.Iin- = ——CT __g = ,BIb ex fl (4.8)
1
dt Vinj
where Equation (4.2) was used.

Eliminating the variable Vow using Equations (4.7) and (4.8), the first-order differential

equation is obtained in terms of the gate voltage Vg:

73

I+'—‘—“:
dV (1) V"?! V
C __&Z_ J—CX K_g_
T d, ,3 51—7. inj (4.9)
0 L
which is written in its closed form as:
V (t)-——l—lo KK [dr+ex (—K V ) 410
g—K2g12 Pzgo (-)
2'6!
with the values ofK, and K2 given by
_U.£_
CT 10 L
K2 = _£_
Vinj

V go in Equation (4.10) is the initial floating-gate voltage and t denotes the total duration

during which the injector circuit in Figure 4.4 is operational. The output voltage

V0“, (t) can also be expressed in terms of Vg (t) as:

Vout (t)= Vg(t) + K3 (4.11)
with K3 :UT10g(Ib/10).
Figure 4.5 shows a typical response of the output voltage V0", (t) as a function of
injection duration t. The response as shown in Figure 4.5 consists of two distinct regions.

The linear region which is characterized by the condition I << (1/ K 1K 2 )exp(—K ZVgO) and

by using Equation (4.1 1) can be simplified to:

{<le .,
CXPG' Kng0)

 

Vout(t)= Vgo + K3 — (4.12)

where the approximation log(l +x) z x has been used. Thus in the linear region, the change

in the output voltage is linear with respect to the injection duration and therefore is

74

suitable for short-term fatigue monitoring (typically less than 100 loading cycles). The
other region, which is particularly important for long-term monitoring, is the log—linear

region and is characterized by the condition t>> (1/K1K2)exp(—K2Vg0), Which when

applied to Equation (4.1 1) leads to:

 

 

1 1
Vom(’l = K3 ‘k—IOgWIKzl—K—loga) (4.13)
2 2
5,__ - e d - ’ . : -:-;—*__Tj
Linear L ___,W'°de.__'_
4.9, -
E 48: Long-Term Monitoring
3 .
= 3 Log-linear -
§ 4'7l 4 p;
2=2 a
'5 4.6
° l
4.5,.
l
4.4». -_ ”W my“. ___-
4000 8000 12000
Cunulative Injection Time (s)

Figure 4.5 Response of a floating-gate injector based on the mathematical model in
Equation 4.10

Thus, the change in output voltage is a logarithmic function of time and could be used
for long—term event monitoring. The first part in Equation (4.13) is an offset term that
captures the dependence of the output voltage on the biasing conditions, initial

conditions, and ambient temperature.

75

Figures 4.6(a) and (b) show the output voltage V0“, (1) response plotted on a logarithmic
scale for typical values of parameters K1,K2. It can be seen from Figure 4.6(a) that the
slope of the log-linear response is a function of K2 and hence determines the maximum

monitoring duration. Using typical parameters obtained from a 0.5 pm CMOS process,
we have verified that “log-linear” response can easily last beyond a million injection

seconds. The parameterK 1, which is a function of the ambient conditions, only

introduces an offset in the log-linear response. This principle will be important in
compensating the response of the injector for initialization errors that can be seen when
Equation (4.13) is expressed in its incremental form as:

1 t
AVout (t): —E10g(}B—] (4.14)

where AVG,” (t) is the change in voltage measured with respect to voltage at reference

time-instant to . Details related to calibration will be discussed in section 5.3.3 of chapter

4.4 CMOS CURRENT REFERENCES

The current reference in the schematic shown in Figure 4.4 is important for ensuring
stable floating-gate injection. The reference therefore has to establish a bias current that is
insensitive to the voltage fluctuation at the output of the piezoelectric transducer. Also the
reference should be robust to variations in process and ambient conditions (for example
temperature). In the literature most current references fall into either one of the following
categories: (a) references whose output is proportional to absolute temperature (PTAT);

(b) references whose output is temperature independent. In this section, the principles be-

76

Output Voltage (V)
co
01 4s

0)

 

 

 

2.5
2 , A 1. A l 7 I F
10° 102 ‘ 0" 10“
Injection Duration (5)
(a)
4.5 - . l

    
     
 

A

 

K1=5.6x10'18

Output Voltage (V)
0)
01

(A)
f

o K1=5.6x10'19

 

 

 

 

 

>

2.5 o K1=5.6x10'2° ,

2 6 i T 7 I", ,. l 4 Lts 7 a

10 102 1o 10 10
Injection Duration (5)

(b)

Figure 4.6 Response of the floating—gate injector for different values of (a) parameter K2
and (b) parameter K1

77

hind the design of a CMOS PTAT current reference are discussed, and then some of its
performance limitations are illustrated [84, 85, 86].
Figure 4.7 shows a schematic of a basic current reference, that was first proposed using

bipolar transistors [87], and later in its CMOS form [88].

Q2 Q5

 

 

 

 

Startup

 

 

 

 

Figure 4.7 Schematic of a CMOS current reference

If the width-to-length ratio W/L of transistors Q1 and Q2 is denoted by W1 /L1 and

W2/L2 , then 1 ref is given by:

[ref = K12 (4.15)
With K = WiLz/Lle.

78

The current [ref determines the potential drop across the resistor R, which in turn

determines the ratio of the currents flowing through the nMOS transistors Q5 and Q6. If

the transistors are biased in weak-inversion, the current 1 ref can be calculated as:

U T
[ref 2 YlogK (4.16)
which is theoretically independent of the supply voltage and linearly proportional to

temperature through U T-

It can be verified that I ref =0 is also a stable state for the current in the CMOS
reference circuit. The schematic in Figure 4.7, therefore, requires a startup circuit that is
formed by transistors Q5 and Q6. Initially the capacitor C is fully discharged and
therefore Q5 and Q6 are ON. This ensures that the node V M is at a sufficiently higher

potential than the source of Q4 during startup. When the equilibrium condition is

achieved as determined by the Equation (4.16), the capacitor C charges to the supply

voltage, which then turns off the transistors Q5 and Q6. Under ideal conditions (perfectly
matched transistors), [ref does not depend on supply voltage or threshold voltage, but is

monotonic in temperature (approximately PTAT in weak inversion). In practice, the
reference is affected by the supply voltage variation due to finite drain-to-source

impedance of transistor Q2 , and also by threshold voltage mismatch between the pMOS
pair Q1,Q2 and nMOS pair Q3,Q4. The sensitivity of the current reference to the supply

voltage variations can be reduced by either using long transistors and through cascading.

Stability of the current reference is another design consideration that is important during

79

its start up. By limiting the total capacitance at node VM , limit-cycle behavior can be

avoided.

4.5 DESIGN OF A COMPLETE FLOATING-GATE SENSOR ARRAY

In this section, a complete design of an array of floating-gate injectors is described. The
basic sensor circuit given in Figure 4.4 has been extended to an array by inserting MOS
diodes between the current reference and the source of the floating gate transistor. A
schematic implementing a complete floating-gate injector array with an integrated current
reference is shown in Figure 4.8. Table 4.1 provides the respective sizes of the transistors,
resistors, and capacitors used for this prototype. Seven injectors have been integrated on

the chip.

A reference current generator circuit is implemented using transistors M1 — M8 and a

resistor R. If the ratio of the pMOS current mirror transistors M 1,M 5 is denoted by K,

then the magnitude of the reference current is I b = (U T / R )log K , where UT denotes a

thermal voltage (26 mV at room temperature). Transistors S] and 52 form a startup
circuit for the current reference. The reference current is copied by mirrors Pl -P1 3,
which drive the floating-gate cells F l—F 7. Diode-connected transistors arrays Dl—D6, are

used to control the potential drop between the supply terminal and source of the floating-

gate transistors, this ensures that each of the floating-gate cells (F 1-—F 7) start injecting at a

differential supply potential(W—V). For an injector circuit consisting of M diodes

between the current source and the floating—gate transistor, the minimum supply voltage

required for onset of injection is:

80

V+ — ZVdsa, — MUT 10%???) — Voml. = V— (4.17)

where Vdsa, is the drain-to-source voltage drop for the cascaded current mirrors, I do is

specific current for the pMOS diode transistors, and Vow]. is the initial floating-gate

source voltage for nodes 0,-,i =1,..,8.

For transistors biased in weak—inversion Vdsa, Z 3U T , which leads to:

_ I
V+ —V 2 MUT 10g[76—::-]+V0ut0 +6UT (4.18)
For Vomo = 5V , the inequality (4.18) leads to:

V+—V' 20.6M+5.15 _ (4-19)

Therefore the minimum amplitude of voltage pulses required to be generated by the

piezoelectric transducer is 5.15 V.

Table 4.1 Component sizes used in circuit in Figure 4.10

 

 

 

 

 

 

 

 

Component Parameter
Ml 60um/10um
M2, M5-M6, P1-P16 30um/10um
M3—M4, M7-M8 60um/10um
D1-D7 lOum/lOum
F 1-F8 6um/6um
C1-C8 lOOfF
R 1.5M!)

 

 

 

 

81

zeta 8:20on Bum-mEEoE wd ouswi

 

 

[x

 

 

 

 

O
0
II
II

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

astmuw

illlll|l|l|l|l_ _lll||ll|l_

82

Cadence based SPICE simulation of the circuit in Figure 4.8 demonstrates the
activation profile of different floating gate cells F1-F7 at different peak input voltage
amplitude. For this experiment a storage capacitor of 10 nF was chosen, and the duration
of the simulated piezoelectric pulse excitation was set to 2 seconds. The circuit exhibits a
start-up time of about 100 ms, which is sufficient for most civil engineering applications.
The start-up however, can be optimized by appropriately sizing the storage capacitor at
the rectifier but at the expense of lower coupling voltage (rectifier). Simulation results
shown in Figure 4.9 clearly illustrate that different floating-gate cells are activated at
different input levels. The simulation also shows poor current regulation of the reference
circuit due to sub-threshold operation of the circuit, but does not adversely affect the

response of the sensor.

 

 

 

15'-

1o-
A If.
4'
t {f-

.41"
5_ .
Cell1 Cel|2 Cell Cell4 Cel|5 Cell6 Cell?
0 , - - —/fi
8 1O 12 14 16 18 20

Peak piezo vottage (V)

Figure 4.9 Simulation result using the circuit in Figure 4.8, showing the current through
the floating gate cells for different peak excitation input voltage

83

4.6 CONCLUSION

In this chapter, the feasibility of a self-powered event monitoring system based on a
combination of piezoelectric transduction and floating gate injection was shown. The
proposed system exploits computational primitives inherent in a current starved IHEI on
a floating-gate transistor that is biased in weak-inversion region. Comprehensive
theoretical models for different components of the VLSI circuits were discussed. Cadence
based SPICE simulation results confirm the desired behavior of the proposed design.
Experimental measured results, that will verify the theoretical models proposed in the
previous sections, as well as the global performance of the complete sensor array, will be
presented in the next chapter. The integration of the proposed circuits with piezoelectric

generators, under real civil engineering environments, will be also discussed.

84

CHAPTER 5
RESULTS AND DISCUSSION

5.1 INTRODUCTION

In the previous chapters, the feasibility, design and implementation of a novel piezo-
powered analog floating gate sensor was discussed. The proposed sensor would be able to
record cumulative statistics of the stress/strain history undergone by a mechanical system.
The concept of operation is based on the coupling of piezo-electric transducers with
floating gate analog memory transistors. Based on the electrical signal generated by the
transducer due to the induced stress, electrons are injected onto a floating node. The total
charge accumulated on the floating gate would indicate the number of stress-strain
events. This chapter focuses on the testing and validation of the presented concept. The
important validation factors include: the control of the rate of injection of the floating
gate transistor, the monotonic response of the sensor to the cumulative input signal, the
robustness of computation to different artifacts (mismatch and temperature), and the
power dissipation of the complete system and its accordance with the 1 11W power
requirement determined in chapter 2.

Several prototypes of the floating-gate injector array have been fabricated and tested.
Figure 5.1 shows the micrograph of the latest tested version manufactured in a standard
0.5 pm CMOS process. Its specifications are summarized in Table 5.1. (Detailed graphs
of different components and the micrographs of different prototypes are shown in the
Appendix).

A complete sensor will have the piezoelectric transducer outputs directly connected to a

full-wave rectifier, implemented using a standard diode bridge. For the tested prototype,

85

"iv.” “a a” t .‘rww imam".-
- , .-

u, up..-

-' -a.oo'.-o .- .-

 

Figure 5.1 Micrograph of the prototype floating-gate sensor

Table 5.1 Summary of measured specifications

 

 

 

 

 

 

 

Parameter Value
Technology 0.5 pm CMOS
Size 1.5 mm x 1.5 mm
Number of Injectors 7

Injection Range 4.2 V - 8 V
Maximum Current 160 nA

Startup Time <30 ms

 

 

 

 

n+ / p— substrate and p+ / n- well diodes were used, which are naturally integrated on
electrostatic discharge (ESD) protection pads of the VLSI chip. A storage capacitor is

used at the output of the rectifier to filter out unwanted high-frequency components. The

86

size of this capacitor also provides a trade-off between the total hold-time versus the
voltage swing at the sensor. For the prototype, an external capacitor (10 nF) was chosen,
which is equivalent to a voltage swing of up to 8 V across the rectifier when a 20-V pulse
is generated by the piezoelectric transducer. A voltage over-protection and clamping
circuitry (consisting of zener diodes) was integrated at the output of the diode bridge to

prevent damage due to unwanted piezoelectric surges.

5.2 SYSTEM ROBUSTNESS

From Equation (4.14), it can be seen that the differential response of a floating gate

injector, and hence the output of the sensor array, is independent of the parameter K1.

However, the robustness of the parameter K2 to process artifacts still needs to be

addressed. Several experiments were conducted to quantify the robustness of the

parameter K2 to different biasing and mismatch conditions. Figure 5.2 shows the

responses obtained from multiple injectors on the same chip that were biased with

different current sources (lb). The mismatch in the parameter (K2) was calculated to be

less than 10% for a bias current variation greater than 100%. This result is encouraging
since it implies that the precision of the current source is not critical for the operation of
the floating-gate injector. Figure 5.3 shows the responses obtained from multiple
injectors biased using the same source current but implemented on different chips for the
same manufacturing run. The fabrication related mismatch was computed to be less than

15%. Figure 5.4 shows the response of the injector obtained for two different temperature

87

Output Voltage (V)

Output Voltage (V)

 

4.5

 

    
 

 

 

 

4.5,
4.45r
l
4.41. _ 44 _.--
l o l=17n,K2=10.0604
1 1:1 |=13n,K2=10.1420
4'351‘ o |=10n,K2=8.787 l
i A I: 8n,K2=9.2166 J
4 3 gm;- T:Ln_-_,-L.L-L L.-. . Lr _E run; .__-.__L_.-_-1
10° 101 102 1o3 10‘ 105

Time (t)

Figure 5.2 Voltage responses measured at various source currents

 

 

 

 

 

 

 

4.56, 4 -* 4 44 . 4 4 444-4 - -4- -44.,
4' l ‘2 h ' T An.
. H: . i
4.52 - i 1‘,‘ 1
4.5 ‘ _. i
448 :I
073a,“? , 1?, ; 5315.? . ~
4.46: -.
1 :1 Chip 2,K2=10.9104 ..
4.44sl A Chlp 3,K2=10.5166
‘ 0 Chip 4-,K2=10.0577
442 L4 . Lima "“1‘-;"_:1;:-J.1 Egg~_,__,_g,_
10° 101 1o2 103 104

Time (t)

Figure 5.3 Voltage responses measured for various chips in the same run

88

 

E 1

Q l

m l

:3 . 1

3 I I

> |

‘5 .

2- 4.4, l
3 .

4.35l o T= 21.2",c K 159.7857! 1

l

ll :1 T= 321°c,2K=8.szao] ‘

43 V _..__- up”; ,L. i 2; 111.1- ;_:_ril11..1 LEE. “LII -n_i_ 1 Lina-1.

10° 101 1o2 103 104 105
Time (t)

Figure 5.4 Voltage responses measured for various temperatures

levels. We have conducted the experiments over a range of 10°K and have observed less

than 10% variation in the slope parameter K2. The measured results summarized in

Figures 5.2 —— 5.4 thus demonstrate that the parameter K2 is robust to variations in biasing

and ambient conditions.

5.3 EXPERIMENTAL MEASURMENTS
5.3.1 Initialization of the injectors array

Before the prototype is used for monitoring, an equalization procedure is required to
initialize the state of the floating-gate injector array. The equalization algorithm is

summarized as a flow chart in Figure 5.5.

89

 

1
Set Q20 / iiio
Set V. If),

mput : i0

11 1

Measure Q. Tunnel E.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No No

Yes Yes

 

 

 

 

 

 

 

 

iO =i0 +1 Inject Fl.
J

 

 

 

Figure 5.5 Flow-chart describing the sensor initialization algorithm

Initially, each of the floating—gate cells F l—F7 (Figure 4.10) is programmed (using
tunneling and injection) to store a fixed amount of charge and hence a fixed floating-gate

voltage across F1—F7. The channel to be initialized is first selected by connecting the

source of all other channels to the terminal V_. This ensures that these channels are

unaffected when the selected channel is being initialized. For the sake of clarity the

programming transistors have not been shown in Figure 4.8. The IHEI is then initiated on

the selected channel by applying a large potential difference across the terminals V+ and

V—. As a result the potential of the selected output node decreases. FN tunneling is then

used by applying a large potential (>16 V) on the tunneling node (also not shown in
Figure 4.8 — shown as node T in Figure 4.4) which results in an increase in the output

voltage. These gate voltages are indirectly monitored by measuring drain-to-source

90

voltages across each of the floating-gate cells (01‘ ,i = 1,2,.., 7 ). For different supply

voltages Vddia i = 1,2 ,.., 7 , the procedures are repeated until the output node potential

is programmed to V50 = 5 V. The choice of this voltage was to ensure that sufficient

drain-to-channel potential exist to initiate the IHEI process, when input voltage exceeds a

prescribed threshold.

5.3.2 Threshold adjustment and levels detection
The minimum voltage required to initiate injection on a floating-gate cell can be
calculated according to Equation 4.18 where M is the number of series pMOS diode

connected transistor for the channel. It can be seen that the resolution of thresholds can be

controlled either by changing M or by changing the reference currenth (using a different

value of resistance R for the current reference). Figure 5.6 shows the measured results

obtained after the output voltage of the first four injectors in Figure 4.8 were initialized,

using the algorithm shown in figure 5.5, and the input voltage (V+ - V“) is varied from

4V to 8V. It can be seen from Figure 5.6 that the responses of the injectors are offset by
potential difference proportional to the number of series diodes. This implies that
different channels will start the IHEI at different levels of the input voltages. Figure 5.6
also illustrates the heuristic threshold level (drain-to-channel voltage) required to trigger

the IHEI. For the 0.5-um CMOS process, the injection threshold is approximately 4.2 V

which also justifies the choice of the initialization voltage V SO = 5 V.

The level-selection functionality of the prototype was validated by applying voltage

pulses of different magnitude using a programmable signal generator. These voltage

91

pulses were used to emulate the output of a piezoelectric transducer. The startup time for
the current reference and injection circuitry was measured to be less than 30 ms, which is
sufficiently fast for a typical loading cycle (assumed to be 1000 ms). Figures 5.7 to 5.9

show the measured output voltages Ol—O3 corresponding to the three injector cells F1—

F3, for continuous amplitude voltage pulses of different amplitude (Vdd1=5.3 V,

Vdd2=6.1 V and Vdd3=6.9 V). Before each run, the floating-gate cells were initialized

using the algorithm described in section 5.3.1.

It can be seen from Figure 5.7 that only cell F1 injects when pulses of amplitude 5.3 V
are applied. Also the response of the injector F1 is log-linear and produces a change in
output voltage as a function of the duration of injection. The cells F2 and F3 show
negligible injection compared to cell F 1. Figure 5.8 shows the response of the injector
array when voltage pulses with amplitude 6.1 V were applied. Both F1 and F2 inject at an
identical rate, whereas cell F3 shows negligible injection. Thus cells F1 and F2 gather
statistics of occurrence of events when voltage amplitude exceeds 6.1 V. For pulses with
amplitude greater than 6.9 V, all the cells Fl—F3 inject at identical rates (Figure 5.9) and
the responses show excellent agreement with the mathematical model in Equation (4.13)
shown in Figure 4.5. Using the model in Equation (4.13), the total duration that a cell

injects before it approaches the threshold (4.2 V) is calculated to be approximately

70,000 seconds. This period can be easily extended to beyond 107 cycles by varying the

design parameters such as gate capacitance and injection current.

92

 

     

Output Voltage (V)

1 —9— Output 01
l + Output 02 ‘ I
‘ + Output 03
‘ + Outout 04

Injection Threshold

 

'4 7 45 s - as e 65 7 is 7 E
v+ (V)

Figure 5.6 Measured output voltages (OI—04) response vs. input voltage

 

4.9

Output Voltage (V)
P .45 9
c» \1 on

P
01

 

 

 

 

44 - A,

. I 1 A 1
101 1o 103 10‘ 10

Loading Cycles

Figure 5.7 Measured source-to-drain voltage response across floating-gate elements at
Vdd=5.3V

93

sew)
.4
oo

Vdd= 6.1 V

(V)

Output Voltage

Output Volta

:5
to

1“
\l

P
0)

 

 

 

 

 

 

 

 

 

 

 

 

 

4.5
4.4 ‘1’ - A4 1 .
1o1 102 1o3 104 105
Loading Cycles
Figure 5.8 Measured source-to-drain voltage response across floating-gate elements at
5 ' "Wi I * 4_4 * 444,
4.9 —- 0 F1 1
0 F2
4.8 ~ ° F3 -:—
4.7 — ‘ -~
4.6 -4 _
4.5 ~ ~
4.4 111 1 1 1111i1l 1 11111 1 1 1 111
101 102 1o3 10“ 105
Loading Cycles

Figure 5.9 Measured source-to-drain voltage response

Vdd= 6. 9 V

94

across floating—gate elements at

5.3.3 Level-crossing Counter

The next set of experiments was designed to evaluate the performance of the injector
array under variable loading conditions and verify the level-crossing algorithm. The
floating-gate elements were first initialized using the algorithm described in Figure 5.5.
The programmable voltage source was used to generate a periodic arbitrary waveform
consisting of three different levels (5.3 V, 6.1 V, and 6.9 V).

A sample voltage waveform used for one of the experiments is shown in Figure 5.10,
where the duration of the three levels were programmed to be in the ratio of 3:221. The
measured response from the injector array is also shown in Figure 5.10. The difference in
duration of injection translates into an offset in the log-linear response and is evident
from the measured results. Figures 5 .11 and 5 .12 show the response of the injector array
for loading cycles with different durations at the various voltage levels. It can be seen
from the measured results that the array indeed can capture the monotonic relationship of
the statistics in the offsets between the measured voltage curves. The offset, however,
also contains information regarding the initial conditions, thus, any error in initializing
the injector array will appear as a measurement error.

A calibration scheme based on Equation (4.14) has been applied to compensate for
offsets introduced by ambient conditions. The calibration scheme exploits the log-linear
response of the floating-gate injector as described by Equation (4.13). When the injector

operates for a long duration t>> exp(—K2VSO)/K1K2 the response of the injector can be

approximated by Vou,(t)z —ln(t)/K2 , which is independent of the initialization
conditions. For initialization errors less than 10%, the shortest duration (referred to as

calibratlon period 1n thls thes1s) requ1red to achleve compensat1on lS approx1mately 10

95

seconds. This is verified using measured responses as shown in Figures 5.13 and 5.14,
where three injectors (corresponding to channels 1-3) are initialized using the procedure
in Figure 5.5. The initialization errors can be seen as the offset in the source voltage
measured for each of the channel at t = 0 s. The calibration procedure entails operating all
the injectors for a duration of 1000 seconds. It can be seen from the measured responses
that all three injectors self-calibrate themselves (converge to the same output voltage) at
the end of the calibration period (1000 cycles).

Subsequently, the cells are subjected to variable loading cycles, whereby each cell
injects only for a fraction of the duration. If a denotes the fraction of total injection
period, then

AVG“, (t) = -El_log[w] (5.1)

2 t0

and the parameter a can be calculated as:

a : exp[— K2 (Voutl (t)- V0ut1(t0))]_ 1
exp[— K2 (VoutZ (0'— Vout2(t0 ))]_l

where V0,,” and Vowz are the output voltage for cells F1 and F2. In Figure 5.13, the

 

(5.2)

parameter a is computed to be equal to 0.30 for cell F3 and 0.71 for cell F2, which is
close to the ratios 1/3 and 2/3. The sources of error include a mismatch in parameter K2
for the different cells, as well as imperfect startup and shutdown of the injector cells.
Figure 5.14 shows a similar result but for a different ratio (322.5:1) of loading cycles. It
can be seen from Figure 5.13 and Figure 5.14 that the offset between injection response
for cells F 1 and F3 is identical, whereas the offset between F1 and F2 has been reduced,

which is consistent with the statistics of the loading cycles.

96

 

 

 

 

 

 

 

 

 

 

 

4.9—
2 4.8—
m .
U)
.9
g 4'77 j : H
'5 :
.9- 1 f 33
8 4.6~ :1
4.5» » 5—
4.4 . ii
101 105
Loading Cycles
15
<——'>
25
41—1»
35 5.3vI 6.1V 6.9V

 

 

 

Figure 5.10 Measured output voltage response for the sensor array when subjected to a
loading cycle represented by a piezoelectric output with injection duration 3:211

97

 

 

 

 

 

 

 

 

 

 

 

5 , .
F1
F3 ;
cu
O) z
a :
>0 4.71- —
*5 ,
9
8 4.6—
4.5 .1
4.4 #1 111111 . 111.1111 i 11L11'15 1 1111111
101 102 103 10‘ 105
Loading Cycles
5.3VI 6.1 v 6.9V

 

 

 

 

 

Figure 5.11 Measured output voltage response for the sensor array when subjected to a
loading cycle represented by a piezoelectric output with injection duration 32.521

98

 

 

5*
CO
I

 

 

 

P
00
l

 

Output Voltage (V)
A
g,

P
on
I

 

 

 

 

 

4.5— .-
4.4 . - i ;.
101 105
Loading Cycles
0.5s
4—>
Zs
<——->
35 5.3VI 6.1 v 6.9V

 

Figure 5.12 Measured output voltage response for the sensor array when subjected to a
loading cycle represented by a piezoelectric output with injection duration 3:2:0.5

99

 

 

 

 

 

 

 

 

 

5 . ,.-H, . [4.4, . ': ..
. . F1
4.9» F2 —
F3
244
cu j '
a) ., .
g '15:
g 4.77 *1
.5 _
e
8 4.6~ 5
4.5— 4
4.4 1 1 1 11111l 1 1 1 111111 1 11111111 1 1111111
101 1o2 103 1o4 105
Loading Cycles
15
4—>
Zs
4—-—>
- 3s _ 5.3vI 6.1 v 6.9V

 

 

 

Figure 5.13 Measured output voltage response using the sensor array in which the first
1000 cycles have been used as calibration intervals and the subsequent three level loading
cycles where the durations of injection are in the ratio 322:1

100

 

 

   
  
 
  

 

 

 

 

 

 

 

 

 

F1
4.9- F2 T
F3
$48— ‘ ‘
a)
a, .
.9 i
g 4.7—
'5' :3;
E." g:
8 4.6—
4.5 ~ 1
105
5.3VI 6.1 v 6.9V

 

 

 

Figure 5.14 Measured output voltage response using the sensor array in which the first
1000 cycles have been used as calibration intervals and the subsequent three level loading
cycles where the durations of injection are in the ratio 322.521

101

5.3.4 Counting Resolution

The next sets of experiments were used to determine the resolution at which a single
floating—gate injector can count the number of events (when the amplitude exceeds a
given threshold). Only the first injector channel was used for this experiment and was
subjected to voltage pulses (generated using a programmable signal generator), each with
a duration of 1 second and amplitude of 5.5V. First, the injector was calibrated by
applying 1000 calibration pulses of a duration of 1 second to eliminate any initialization
errors. The voltage pulses were then programmed according to the following conditions:
for the first set of experiments 256 voltage pulses were directly applied to the injector
(referred to as 256/256 in the measured result); for the second set of experiments only
248 out of 256 pulses were applied to the injector; the process was repeated using 240 out
of 256 pulses until 0 pulses out of 256 were applied to the injector. The measured results
for all of the above set of experiments are summarized in Figure 515(3) and its inset in
Figure 5.15(b). The measured results show a monotonic relationship between the injector
voltage and the count of the voltage pulses applied to the injector. It can also be seen that

the proposed injector can count events at a resolution of more than 5 bits.

5.3.5 Power dissipation

The power dissipation of the sensor (current reference and floating gate array) was
calculated at different voltage levels and is summarized in Figure 5.16. It can be seen that
the total current drawn by the sensor saturates around 6.5 V where all the injectors
become active. Thus, the nominal power dissipation of the sensor at 5.5 V is 800 nW.

More importantly, the impedance of the sensor is 40 MO, implying that the proposed sen-

102

7000/2236 cyfcles‘.‘ ‘
220/256 cycles ‘ ,

O
A
El 240/256 cycles‘‘ 1
+ 1
0

   
  

248/256 cycles l1
. 2567/2535 cycles 1 1

i I
l

   

Output Voltage (V)
.5
a:

4.51
4.41
4.3 1
4.2‘ 1 - 3 - 1 1 1 1; 14 l
10 10
Loading Cycles (n)
(3)

Output Voltage (V)

 

Loading Cycles (n)
(b)
Figure 5.15 Measured results to calculate the resolution for event counting: (a) Change in
source voltage measured for different relative counts of voltage pulses for over 100,000
events; (b) Inset showing zoomed area between 10,000 to 40,000 events

103

sor could also be interfaced with other power harvesting sources that typically have low

current driving capability.

160

 

14o-

Load Current (nA)
8 B

m
0

 

 

 

 

4.5 5 5.5 6 6.5 7
Supply Voltage (V)

er

3 3i5
Figure 5.16 Supply current drawn by the sensor at different supply voltages

5.4 DESIGN EXAMPLES FOR SYSTEMS MONITORING — SENSOR AND
TRANSDUCER INTERFACIN G

For the first set of experiments, the prototype injector array was interfaced with a
piezoelectric transducer (PZT). The transducer was mounted on an extruded acrylic
cantilevered beam. The beam was then subjected to periodic loading cycles. An external
capacitor (100 nF) was chosen which is equivalent to a voltage swing of up to 8 V across
the rectifier when a 20 V pulse is generated by the piezoelectric transducer. An over-
voltage protection and clamping circuitry (consisting of zener diodes) was integrated at
the output of the diode bridge to prevent damage due to unwanted piezoelectric surges.

Figure 5.17(a) shows a sample waveform recorded at the output of the half-wave rectifier

104

integrated on-chip with the injector array, showing that for the loading cycles the
transducer can easily generate voltage levels close to 7 V. The measured response for
each of the injectors in the array is shown in Figure 5.17(b). The results clearly show
monotonic response across the three injectors and that each response is proportional to
the total duration for which the strain levels exceeded the programmed thresholds. The
injector responses were then calibrated against Figure 4.5 and Equation (4.13) to
determine the loading statistics which can then be used in Miner’s rule. For instance, the
cumulative loading statistics calculated at a given time are indicated in Figure 5.17(b),
which shows that 30% of the time the structure experienced strain levels exceeding the
threshold for injector 1, 16% of the time experienced strain levels exceeding the threshold
for injector 2, and less than 1% of the time experienced strain levels exceeding the
threshold for injector 3.

The second set of experiments was designed to test the ability of the sensor to record
low frequency loads. To understand the limitations imposed by low-frequency operation,
consider a simplified equivalent model of the prototype sensor interfacing with the
piezoelectric transducer. The transducer is modeled using an AC voltage source

connected to a decoupling capacitor C (Figure 2.2). The processor has been modeled as a
simple resistive load RZ.

As discussed in section 2.3.1, for a harmonic mechanical loading of the piezoelectric
transducer at a frequency f Hz, the magnitude of the voltage across the sensor is given by

Equation (2.5), and the power delivered to the load (sensor) can be optimized with

respect to the load R2. A typical loading frequency of 1 Hz and a typical PVDF

capacitance of 10nF would yield an optimal load of the processor circuit of about 15 MO.

105

 

h
I

Voltage (V)

 

 

 

 

 

 

 

N-
no
‘9

"o 1 2 3 4 5 6

 

 

 

 

 

 

 

 

 

Tine(s)
(a)
5- i :1 F1
0 F2
a
d)
8’
:5 4.9~ -
>
‘5
a.
S 4.85— -
o
4.8— e
4.75 ‘W’ ...
1o2 10" 10‘
Loading Cycles
(b)

Figure 5.17 (a) Voltage generated by a piezoelectric transducer in response to periodic
loading; (b) Measured response of the floating-gate injector array

106

For a 5 V input voltage, this is equivalent to a load current of 300 nA. The total current
drawn by the fabricated prototype has been measured to be 160 nA at 6.7 V which is
clearly less than the optimal load current and hence is ideal for this application.

The integrated sensor (piezoelectric transducer and floating-gate injector IC) was
attached to a plexiglass beam and the setup was mounted on a testing system (MTS
machine) as shown in Figure 5 .18. The MTS machine was programmed to generate three
distinct levels of displacement at the center of the beam. The averaged strains over the
length of the PVDF transducer were calculated to be 2107 us, 2546 p8 and 2809 us. The
size of the transducer was selected so that the first tested load level would yield enough
voltage to activate only the first injector F 1 (a PVDF sheet of dimensions 62 mm x 12
mm x 28 mm was used). The mechanical loading was cyclically applied to the plexiglass
beam at a loading frequency of 1 Hz. Figure 5.19 shows that when 2107 pg strain was
applied only the first channel records a change in output voltage. When loading cycles
corresponding to 2546 us are applied, both channels 1 and 2 record change in voltage
whereas the channel 3 voltage remains unchanged (shown in Figure 5.20). When the
input load level exceeds 2809 us, the first three cells Fl-F3 are simultaneously active
(Figure 5.21). Although no calibration is performed in advance, the offset in measured
output voltage can be used to determine the count of loading cycles with different strain

levels.

107

 

Output Voltage (V)

1 _J ; ,. ,1; 1
415‘» g —e—Channe|1 ‘77,
‘—E—Channel2
1+Channel3j ‘

451ii

103 104
Loading Cycles (n)

Figure 5.19 Voltage responses measured when the prototype is interfaced with a PVDF
transducer and subjected to controlled cyclic strain levels with magnitude 2107 us

108

 

  

5" ... —- ... _A_A AA A A
4.9‘ ,
§ 1
034.8 1
g
0
>
‘5 4.7
2-
3
O
4.61

   
 
   

+Channel1
45‘ t +Channel2
' +Channel3l

103 104

Loading Cycles (n)
Figure 5.20 Voltage responses measured when the prototype is interfaced with a PVDF
transducer and subjected to controlled cyclic strain levels with magnitude 2546 us

7 7% 77"" , 777—“ __T_' T

 

4.9 1

P
‘1’

Output Voltage (V)
:“
fl

:5
or

451“—e—Channel1
' 1‘—B--Channel2 .
1 ‘ +Channel3j ‘
4.4 ,zzi :,,.. .717..113 . . . . .1 ,4
10 10 10
Loading Cycles (n)

Figure 5.21 Voltage responses measured when the prototype is interfaced with a PVDF
transducer and subjected to controlled cyclic strain levels with magnitude 2809 us

109

5.5 DISCUSSION

Even though the measured results shown in the previous sections have demonstrated

the operation of the floating-gate injectors for up to 105 loading cycles, we have observed
deviations from the log-linear response beyond 105 loading cycles. Figure 5.22 shows a

sample response obtained using three injectors (channels 1-3) operating up to 106 loading

cycles. It can be seen from Figure 5.22, that the response follows a superposition of two
log-linear responses. We believe that a possible explanation to the appearance of the
superposition term could be the contribution from interstitial traps which releases
electrons back in to the transistor channel. At high injection rates (greater than 100
electrons per second) the de-trapping behavior is masked but becomes dominant when the

injection rate reaches as low as l electron per second.

  
   
   
 

5’ l .L,;;.' 7’: FV'ILI
0 Channel1 E
0 Channel 2 ‘
4'8 El Channel 3 1"
4:6 — :1
Slope

Deviation

Output Voltage (V)
.5 A
n c.

h
V .
0

\~.; ‘

1 \..Il

3'8 ‘ i 1 1 1 .1 . . I i 1‘
101 102 1o3 10‘ 105 10°

Loading Cycles (n)

Figure 5.22 Slope deviation observed for injection currents less than 1 electron per
second

110

Even though the power dissipation of the fabricated prototype is less than 800 nW
which makes it suitable for self-powered usage monitoring at a harmonic loading of
around 1 Hz (which is typical of most mechanical loading in civil engineering), there are
applications which require monitoring at much lower frequencies such as the monitoring
of strain cycles due to daily temperature fluctuations (corresponding to a period of about
24 hours). In such cases, the required power dissipation is less than 30 nW. We believe
that the current design can achieve this requirement by starving the injection currents

using a higher value of current reference resistance R.

The minimum amplitude of transducer signal required to initiate injection is measured
to be approximately 4.2 V. Even though piezoelectric transducers can easily generate the
minimum operational voltage, other transducers typically generate sub-volt signals. We
anticipate that this limitation could be overcome by using a sub-100 nm process where
the minimum IHEI voltages is significantly lower than that of required by a 0.5-um
CMOS process. In the future, memristors could be used instead of floating-gate injectors.

However it is not known if the log-linear response could be replicated.

5.6 CONCLUSION

In this chapter, the design and models presented in chapter 4 were validated using a
fabricated prototype. We have demonstrated that the response of the system is robust to
device mismatch and temperature variations. Furthermore, it has been shown that the
level counting circuit can sense store and compute over 105 event cycles with injection
currents less than 1 electron per second. At a total power dissipation of less than 800 nW,

the fabricated prototype is ideal for interfacing with piezoelectric transducers and was

111

shown to count the occurrence of mechanical loading of structures. The proposed sensor
has numerous applications in long-term structural health monitoring, where a large
number of prototypes could be embedded directly inside a material and can

autonomously record statistics of its mechanical loading (usage).

112

CHAPTER 6
CONCLUSIONS AND RECOMMENDATIONS

In this thesis, a novel piezo-powered floating gate array that can sense, compute and
store cumulative statistics of loading cycles experienced by a mechanical structure was
presented. The nominal power dissipation of the integrated circuit is less than 1 uW,
this is 2 orders of magnitude smaller than the best currently available self-powered
sensors. The lower power usage, along with its small form factor makes it ideal for
sensing and long-term monitoring of civil infrastructure. The work reported in this
dissertation provides an initial step toward the implementation of complete low cost and
reliable monitoring networks.

Traditional self-powered wireless sensor systems consist of a microprocessor, radio,
data storage, and batteries. Commercial electronic components are commonly used in
these systems. The harvested energy is accumulated in a capacitor, and a switch is used to
allow the capacitor to charge to a predetermined voltage value at which point the switch
would open and the capacitor is allowed to drain through the circuitry. Once the capacitor
had discharged to a second predetermined value, the switch closes and the capacitor is
allowed to recharge. The entire process is then repeated. Leakage through the switch,
combined with the power requirements of the various components limit the possible
functionalities that could be achieved using the extracted power. For example, the
MSP430 microprocessor by Texas Instrument operates at a 350 ,uW power consumption,
and a 1-bit Analog to Digital (A/D) conversion consumes 2 n] of energy.

Directly using analog signal processing circuits eliminates the need for digital

conversion, thus reducing the power demand. Furthermore, analog computations can use

113

solid-state transistor physics to implement complicated mathematical relations. For
example, a pMOS operated in the sub-threshold regime inherently calculates the
exponential/logarithm relation by simply transforming voltage to current. This alleviates
the need for digital logic components, energy costly microprocessor and leaky switches.
In addition, analog floating-gate programming can be performed with power dissipation
as low as 100 pW during the write cycle, compared to digital FLASH programming that
consumes 100 to 200 nW. The principle of operation of the novel sensor developed in
this work is based on the integration of piezoelectric transducers with ultra low power

analog computation circuits and analog floating-gate memory.

Different approaches to convert mechanical energy to electrical energy for self-
powering sensor electronics were discussed. Piezoelectric energy harvesting was selected
as the focus of this research as it was proven to be the most appropriate technique for the
studied applications. Simple mechanical and equivalent electrical models that can be
easily used for a general piezoelectric generator analysis were presented. The electrical
models could be easily implemented into existing circuit simulation programs (such as
SPICE) and provide reasonably accurate estimates of the generator performance under a

wide variety of ambient loading environments.

. . 3 .
Usmg the developed models and assuming a 5 cm sensor node, it was shown that the

electrical energy generated under typical civil structures service conditions
(approximately 1 uW) is insufficient to run modern wireless sensors. Piezoelectric-based

vibration harvesters do produce enough energy fi'om large-scale earthquakes and high

114

vibration environments; these events can allow for approximately 500 data samples to be

collected and wirelessly transmitted by a sensor node.

The design and theoretical model of a novel piezoelectric based self-powered event
monitoring system was presented. The novel features of the developed system include (1)
self-powered, continuous and autonomous sensing, the sensor uses novel analog signal
processing circuits that require less than luW of power (to our knowledge this is two
orders of magnitude less power than previously demonstrated self-powered sensors), (2)
autonomous computation and non-volatile storage of the sensed variables, the sensor uses
floating gate transistor injection principles for computing cumulative mechanical strain

patterns experienced by a structure.

A complete system, piezoelectric transducer and a fabricated prototype, was tested
under realistic low-frequency, civil engineering loading in the laboratory. It was shown
that the developed sensor can sense store and compute over 105 loading event cycles at a

total power dissipation of less than 800 nW.

One of the future key challenges will be to remotely access the loading statistics
accumulated on the floating gate sensor. Therefore, the sensor array needs to be
interfaced with a data transmission circuit which can be remotely powered and
interrogated using an RF signal. The envisioned system would consist of a network of
low costs sensors distributed through the structure (for example a bridge). Each sensor
node would be self-powered and capable of continuously monitoring and storing the

dynamic stress/strain levels in the host structure. The data from all the sensors would be

115

periodically uploaded wirelessly to a central database using low cost RF transponders.
The monitored host structures can be frequently monitored to detect changes in structural
integrity that may not only foreshadow a severe failure, but also allow for more accurate
maintenance and repair scheduling.

Figure 6.1 shows a possible system level architecture of a complete sensor node that
integrates the proposed piezo—floating gate array. The sensor would consist of two sub-
systems: (a) a self-powered sub-system consisting of a floating-gate sensor array, and (b)
a radio-frequency identification (RFID) interface that facilitates remote access to the
stored features. Before the sensor is deployed the floating—gate array should be
calibrated/equalized offline. The operation of a deployed sensor will then consist of two
modes: 1) a continuous monitoring mode where the energy harvested from the applied
mechanical stresses is used for computing and storing loading statistics, and 2) an access
mode where an RF reader is brought in proximity to the sensor which powers up the
analog-to-digital (A/D) converters and transfers the digitized representation of the stored
statistics to a reader over the RF link. The measured statistics can then be used to
determine the history of strain levels experienced by the structure using a calibration
curve that equates the voltage generated by the piezoelectric transducer when subjected to

different strain levels.

 
  

   
  
  

AID

    
 

Piezo
transducer Converter
"1' ‘
*- air-"‘15 '9 Powert I Data

  
 

RF
Interface

    

Self-powered

RF-powered

Loading Features

Figure 6.1 System architecture of a complete implanted device that integrates the
proposed floating- gate sensor array

116

Durability of the sensors both during construction and throughout the lifetime of the
structure is also a critical aspect for successful implementation. A robust package needs
to be developed and tested in order to increase the sensor’s robustness to withstand harsh
environmental conditions. Several commercially available robust RFID tags use an ABS
(acrylonitrile butadiene styrene) plastic package that is durable over a wide temperature
range (~50 degrees to 180 degrees F). This packaging also can withstand mechanical
abrasion and severe chemical attack, thus making it an ideal candidate for sensor
packaging. Another potential candidate for packaging is polyurethane. This epoxy is

commercially used for the manufacturing of RFID tags.

Successful development of the proposed sensor could dramatically transform the
economics of civil infrastructure inspection by facilitating early damage detection and
future condition evaluation. It is envisioned that the piezoelectric strain sensor could be
reduced in cost to the order of $2/sensor in the near term. Given the annual costs of

infrastructure inspection, this sensor cost could rapidly pay for itself.

117

APPENDIX

118

APPENDIX

MICROGRAPHS AND LAYOUT DETAILS

Figure A.1 shows the micrographs of the first three prototypes manufactured and tested
in different optimization iterations. Each prototype represented an improvement with
respect to previous designs based on experimental observations during laboratory testing.

Floating—gates manufactured in the first tested prototype, shown in Figure A.1 (a),
performed as expected, injection curves were measured and matched with the deve10ped
theoretical models. But this prototype had a fimdamental design problem. Since each
floating-gate memory was connected to its own reference circuitry, mismatch between
different transistors induced by the manufacturing process resulted in different startup
behavior between references.

In the second generation shown in Figure A.1 (b), all memory gates were connected to
the same reference, eliminating the startup mismatch. An innovative current reference
was implemented in this prototype. The design was based on the integration of a floating-
gate as a current mirror in the reference circuit in order to accurately control its output
and improve its insensitivity to temperature fluctuations. The lack of control pins at the
reference gate resulted in failed tests. The charge stored on the reference gate could not
be controlled accurately, which resulted in a non-stable behavior of the reference.
Similarly to the previous prototype, all floating-gates performed as expected.

The optimal reference design was included on the third prototype (Figure A.1 (0)). In
order to improve the matching of injection rates between different gates, the bulk of each
cell was connected to a common pin driven by the input voltage. This configuration

disrupted the injection characteristics of the gates due to parasitic capacitances which

119

were not accounted for in the simulations. All optimal designs of the reference and the

floating-gate cells were included in the fourth generation tested in Chapter 5.

DD Floating-gate

Reference

Floating-gate array

- 1.5 mm

Injector
Channels Reference

E]

w.

i
Q.»

 

Current Mirrors
(C)
Figure A.1 Micrographs of different prototypes

120

m 5320 E 358 09389:“ 2: go 3903 N.< 233m

 

121

8:282 “5.55 2: me 3934 m.< ouswE

 

122

 

///7/7/7’/7/77 //7/'///7/7//7/7///////

///7/7/ / ///// 7 /////

//77// //7 // /

 

 

 

 

 

 

 

 

\\\\\A\\\\\\\\\\\\m

 

 

 

II [III r

 

 

 

123

Figure A.4 Layout of the diodes array

.35 E88 .2: Summagou 2: mo 3934 m.< oSME

 

 

124

.5803 Bummzuaom 2wfim a mo 30.34 o.< oeswfl

 

 

.\t

\\\\\\\\\\\\\\\\\\\\‘\\\\\\\\N\\\\\\\\\ \\\\.\N\\\\ \\\\\\\\\\\\\ \N\V\\\\\\\\\\\\\\\_
.

 

 

 

 

 

 

 

 

125

BIBLIOGRAPHY

126

10.

11.

BIBLIOGRAPHY

. Air Force Research Laboratory, “Engine Rotor Life Extension”, Materials and

Manufacturing Directorate, <http://www.ml.afrl.af.mil/mll/p_erle.html>, Accessed
July 2008

Kamoi A, Okamoto Y, Vavilov V, ‘Study On Detection Limit of Buried Defects in
Concrete Structures by Using Infrared Thermography’, Key Engineering Materials.
Vol. 270-273, Part 2, pp. 1549-1555. 2004

Huston D. Hu JQ.; Maser K, Weedon W, Adam C., ‘GIMA Ground Penetrating
Radar System for Monitoring Concrete Bridge Decks’, Journal of Applied
Geophysics, Volume 43, Number 2, March pp. 139-146(8), 2000

Wang X., Chang C-C., Fan L., ‘Nondestructive Damage Detection of Bridges: a
Status Review’, Advances in Structural Engineering, Volume 4, Number 2, 75-91,
2001

Zhao Z, Haldar A, Breen, FL, ‘Fatigue—Reliability Updating Through Inspections of
Steel Bridges’, Journal of Structural Engineering, Vol. 120, no. 5, pp. 1624-1643.
1994

Akyildiz IF, Su W, Sankarasubramaniam Y, Cayirci E, ‘Wireless Sensor Networks: a
Survey’, Computer Networks, 38 (2002) 393—422, 2002

Lynch JP, Sundararajan A, Law KH, Kiremidjian AS, 'Embedding Damage Detection
Algorithms in a Wireless Sensing Unit for Operational Power Efficiency ', Smart
Mater. Struct., 13 800—810, 2004

Tanner NA, Farrar CR, Sohn H, 'Structural Health Monitoring Using Wireless
Sensing Systems With Embedded Processing', Proceedings of SPIE, Vol. 4704, p.
215—224, 2002

Sazonov E, K Janoyan, R Jha, 'Wireless Intelligent Sensor Network for Autonomous
Structural Health Monitoring', Proceedings of SPIE: Smart Structures and Materials,
San Diego 5384 304-14, 2004

Chee-Yee Chong; Kumar SP, ‘Sensor Networks: Evolution, Opportunities, and
Challenges’, Proceedings of the IEEE, Volume 91, Issue 8, Aug. Page(s): 1247 —
1256, 2003

Rajendran V, Obraczka K, Garcia-Luna-Aceves J, ‘Energy efficient, Collision-Free
Medium Access Control for Wireless Sensor Networks’, Proc. lst international conf.
an embedded networked sensor systems, pp. 181-192, 2003

127

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

Hunt SR, Hebden IG, ‘Validation of the Eurofighter Typhoon Structural Health and
Usage Monitoring System’, Smart Material and Structures, v.10 497—503, 2001

Haas DJ, Baker T, Spraeklen D, Schaefer CG, 'Joint Advanced Health and Usage
Monitoring System (Jahums) Advanced Concept Technology Demonstration (actd)’,
AHS International, Annual Forum, 56th, Virginia Beach, Proceedings. Vol. 2; 935-
945, 2000

Chang FK, ‘Structural Health Monitoring’, CRC Press, 1999

Poncelet JV, ‘Introduction 21 la Mecanique Industrielle, Physique et Experimentale’,
pp. 317-8, Deuxieme edition, Imprimerie de Gauthier-Villars, Paris, 1839

Basquin OH, ‘The Exponential Law Of Endurance Tests’, Proceedings of the
American Society for Testing and Materials 10, 625-30, 1910

Coffin LF, ‘A Study of the Effects of Cyclic Thermal Stresses on a Ductile Metal’,
Transactions of the American Society of Mechanical Engineers 76, 931-50, 1954

Mason SS, ‘Fatigue—A Complex Subject’, Experimental Mechanics, Vol. 5, No. 7,
pp. 193—226, 1965

Suresh S, ‘Fatigue of Materials’, 2nd edition, Cambridge University Press, 1998.

Matsuiski M, Endo T, ‘Fatigue of Metals Subjected to Varying Stress’, Japan Society
of Mechanical Engineering, 1969

Dowling NE, ‘Fatigue Failure Predictions for Complicated Stress-Strain Histories’,
Journal of Materials, Vol. 7, No. 1, pp. 71-87, March 1972

ASTME 1049-85, Standard Practices for Cycle Counting in Fatigue Analysis, ASTM
International, 2005

Straser EG, and Kiremidjian AS, ‘A Modular Visual Approach to Damage
Monitoring for Civil Structures’, Proceedings of SPIE v2719, Smart Structures and
Materials. 112—122, 1996

Straser EG., Kiremidjian AS, Meng TH., Redlefsen L, ‘A Modular, Wireless Network
Platform for Monitoring Structures’, Proceedings SPIE 3243:12450—456, 1998

Kiremidjian AS, Straser EG, Meng TH., Law K, Soon H, ‘Structural Damage
Monitoring for Civil Structures’, International Workshop Structural Health
Monitoring. 371—382, 1997

Maser K, Egri R, Lichtenstein A., Chase S, ‘Development of a Wireless Global
Bridge Evaluation and Monitoring System (WGBEMS)’, Proceedings of the

128

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

Specialty Conference on Infrastructure Condition Assessment: Art, Science, Practice.
91—100, 1997

Mitchell K, Sana S, Balakrishnan VS, Rao V, Pottinger HJ, ‘Micro Sensors for Health
Monitoring of Smart Structures’, SPIE Conference on Smart Electronics and MEMS.
3673:351—358, 1999

Agre JR, Clare LP, Pottie GJ, Romanov NP, ‘Development Platform for Self-
Organizing Wireless Sensor Networks’, Proceedings of SPIE 3713:257—267, 1999

Liu RC, Zhou L, Chen X, Mau ST, ‘Wireless Sensors for Structural Monitoring’,
Strong Motion Instrumentation for Civil Engineering Structures, pp. 253—266, 2001

Sirohi J, Chopra 1, ’Fundamental Understanding of Piezoelectric Strain Sensors’,
Journal of Intelligent Material Systems and Structures, Vol. 11, April, pp. 246 — 257,
2001

Rahimi M, Shah H, Sukhatrne GS, Heideman J, Estn'n D, ‘Studying the Feasibility of
Energy Harvesting in a Mobile Sensor Network’ Proceedings of the 2003 IEEE

International Conference on Robotics & Automation, Taipei. Taiwan, September 14-
19, pp. 19-24, 2003

Sodano HA, Inman DJ, Park G, ‘A Review of Power Harvesting from Vibration
Using Piezoelectric Materials’, The Shock and Vibration Digest, Vol. 36, No. 3, May,
pp. 197-205, 2004

Sodano HA, Park G, Inman DJ, ‘Estimation of Electric Charge Output for
Piezoelectric Energy Harvesting’, Strain, Vol. 40, pp. 49-58, 2004

Roundy S, Wright PK, Rabaey J, ‘A Study of Low Level Vibrations as a Power

Source for Wireless Sensor Nodes’, Computer Communications, Vol. 26, pp. 1131-
1144, 2002

Roundy S, ‘On the Effectiveness of Vibration-based Energy Harvesting’, Journal of
Intelligent Material Systems and Structures, Vol. 16, pp. 809 — 823, 2006

Wameke BA, Pister KSJ, ‘Exploring the Limits of system Integration With Smart
Dust’, Proceedings of IMECE’02, New Orleans, Louisiana, 2002

Issa MA, Islam MS, Chudnovsky A, ‘Size Effects in Concrete Fracture: Part 1,
Experimental Setup and Observations’, Int. J. Fracture. 102 1—24, 2000

Otis B, Chee YH, Lu R, Pletcher NM, Rabaey JM, ‘An Ultra-Low Power MEMS-

Based Two-Channel Transceiver for Wireless Sensor Networks,’ IEEE Symp. VLSI
Circuits, June 2004

129

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

51.

EmbedSenseTM Wireless Sensor, Wireless Sensor and Data Acquisition System,
MicroStrain®, 2007

Arms SW, Townsend CP, Churchill DL, Phan N, ‘Energy Harvesting, Wireless,

Strain Sensor for Damage Tracking of Helicopter Components,’ 3rd European
Workshop on Structural Health Monitoring, 2006

Elvin NG, Elvin AA, Spector M, ‘A Self-Powered Mechanical Strain Energy Sensor’,
Smart Materials and Structures, No.10, pp 293-299, 2001

Elvin NG, Elvin AA and Choi DH, ‘A Self-Powered Damage Detection Sensor’, The
Journal of Strain A nalysis for Engineering Design, v.3 8,n2, ppl 15-125, 2003

Dorio C, Hasler P, Minch B, Mead CA, ‘A Single-Transistor Silicon Synapse’, IEEE
Trans. Electron Devices, vol. 43 (11), Nov. 1996

Sodano HA Inman DJ, Park G, ‘Comparison of Piezoelectric Energy Harvesting
Devices for Recharging Batteries’, J Intel Mater Syst Str 16 799-807, 2005

Nakano K, Elliott SJ, and Rustighi E, ‘A Unified Approach to Optimal Conditions of
Power Harvesting Using Electromagnetic and Piezoelectric Transducers’, Smart
Mater. Struct. 16 948-958, 2007

Cho YS, Pak YE, Han CS, Ha SK, ‘Five-Port Equivalent Electric Circuit of
Piezoelectric Bimorph Beam’, Sensors and Actuators 84 140-148, 2000

Lesieutre GA, Davis CL, ‘Can a Coupling Coefficient of a Piezoelectric Actuator be
Higher Than Those of its Active Material?’, .1. Intel]. Mater. Syst. Struct. 8 859—67,
1997

Ottman GK, Hofinann HF, Bhatt AC, Lesieutre GA, ‘Adaptive Piezoelectric Energy
Harvesting Circuit for Wireless Remote Power Supply’, IEEE Trans. Power Electron.
17 669—76, 2002

Roundy SJ, Wright PK, ‘A Piezoelectric Vibration Based Generator for Wireless
Electronics’, Smart Mater Struct 13 1131-1142, 2004

Ajitsaria J, Choe SY, Shen D, Kim DJ, ‘Modeling and Analysis of a Bimorph
Piezoelectric Cantilever Beam for Voltage Generation’, Smart Mater. Struct. 16 447-
454, 2007

Van Noortwijk JM, Klatter HE, ‘The Use of Lifetime Distributions in Bridge

Replacement Modeling’, lst Int. Conf. on Bridge Maintenance, Safety and
Management, 2002

130

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

Roundy SJ, ‘Energy Scavenging for Wireless Sensor Nodes with a Focus on
Vibration to Electricity Conversion’, PhD Thesis University of California, Berkeley,
2003

Stamer, T, ‘Human-Powered Wearable Computing’, IBM Systems Journal, 35: 618—
629, 1996

Shenck N, Paradiso J, 2001, ‘Energy Scavenging with Shoe-Mounted Piezoelectrics’,
IEEE Micro, 21(3): 30—42, 2001

Mehra A, ‘Development of a High Power Density Combustion System for a Silicon
Micro Gas Turbine Engine’, Ph.D. Thesis, Massachusetts Institute of Technology,
2000

Stan MR, Franzon PD, Goldstein SC, Lach JC, Ziegler M, ‘Molecular Electronics:
From Devices and Interconnect to Circuits and Architecture’, Proc. IEEE 91 1940—
57, 2003

Amirtharajah R, Chandrakasan AP, ‘Self-powered Signal Processing Using
Vibration-Based Power Generation’, IEEE JSolid-State Circuits 3 687—95, 1998

Amirtharajah R, Chandrakasan AP, ‘A Micropower Programmable DSP Using
Approximate Signal Processing Based on Distributed Arithmetic,’ IEEE Journal of
Solid-State Circuits, vol. 39, pp. 337—347, Feb 2004

PicoRadio: University of California, Berkeley Project.

X-bow: X-Bow Technologies Mica2Dot Wireless sensor Mote.

Micronz-Micron Technology NOR Flash Memory.

Texas Instrument: MSP430 microprocessor.

Cavendishz-Cavendish Kinetics Corp, MEMS memory.

LSI: - LSI Logic Corp—LSI403LC Digital Signal Processor.

Telos:- Moteiv Corporation.

Kao J, ‘Sub-threshold Leakage Control Techniques for Low Power Digital Circuits’,
PhD Thesis Massachusetts Institute of Technology, 2001

El-Borgi S, Choura S, Ventura C, Baccouch M, Cherif F, ‘Modal Identification and

Model Updating of a Reinforced Concrete Bridge’, Int. J. Smart Struct. Syst. 1 83—
101, 2005

131

68.

69.

70.

71.

72.

73.

74.

75.

76.

77.

78.

79.

80.

Watters DG, Jayaweera P, Bahr AJ, Huestis DL, Priyantha N, Meline R, Reis R,
Parks D, ‘Smart Pebble: Wireless Sensors for Structural Health Monitoring of Bridge
Decks’, Proceedings of .SPIE 5057 20—8, 2003

Giaccio G, Zerbino R, ‘Failure Mechanism of Concrete-Combined Effects of Coarse
Aggregates and Strength Level’, Adv. Cement Based Mater. 7 41—8, 1998

Brownjohn JWM, personal communication, Data measured for the rooftop of the 280
meter tall, 66 story, Republic Plaza building, Singapore, 2005

Celebi M, ‘GPS in Dynamic Monitoring of Long-Period’, structures Soil Dyn.
Earthq. Eng. 20 477—83 COSMOS: Consortium of Organizations, 2000

Li QS, Fang JQ, Jeary AP, Wong CK, Liu DK, ‘Evaluation of Wind Effects on a
Supertall Building Based on Full-Scale Measurements’, Earthquake Eng. Struct.
Dyn.29 1845—62, 2002

Lovse JW, Teskey WF, Lachapelle G, Cannon ME, ‘Dynamic Deformation
Monitoring of tall Structures Using GPS Technology’, J. Surveying Eng 121 35—40,
1995

ISO-2631 Mechanical vibration and shock -- Evaluation of Human Exposure to
Whole-Body Vibration -- Part 2: Vibration in Buildings (1 Hz to 80 Hz), International
Organization for Standards

AISC Design guide 1 1, Floor Vibrations Due To Human Activity, American Institute
of Steel Construction, 1997

Bachmann H, ‘Vibration Problems in Structures’, Birkhauser, Germany 1994

Polastre J, Szewczyk R, Sharp C, Culler D, ‘The Mote Revolution: Low Power
Wireless Sensor Network Devices’, Proc.Hot Chips 16, 2004

Elvin NG, Lajnef N, and Elvin AA, ‘Feasibility Of Structural Monitoring With
Vibration Powered Sensors’, Smart Mater Struct 15 977-986, 2006

Shibata T, Ohmi T, ‘A functional MOS Transistor Featuring Gate-Level Weighted
Sum and Threshold Operations’, IEEE Transactions on Electron Devices, vol. 39 (6),
pp.l444—1455, June 1992

Srinivasan V, Serrano GJ, Gray J, Hasler P, ‘A Precision CMOS Amplifier Using

Floating-Gate Transistors for Offset Cancellation’, IEEE Journal of Solid State
Circuits, vol. 42, no. 2, February 2007

132

 

81.

82.

83.

84.

85.

86.

87.

88.

89.

Machado Gerson AS, Enz CC, Bucher M, ‘Estimating Key Parameters in the EKV
MOST Model for Analogue Design and Simulation’, Proc. IEEE Int. Symp. Circuits
Syst., pp. 1588-1591, Seattle, Washington 1995

Hasler P, ‘Foundations of Learning in Analog VLSI’, Ph.D. Thesis, California
Institute of Technology, Pasadena, California, 1997

Lajnef N, Chakrabartty S, Elvin N, Elvin A, ‘Piezo-Powered Floating Gate Injector
for Self-Powered Fatigue Monitoring in Biomechanical Implants’, IEEE Symposium
on Circuits and Systems (ISCAS’2007), New Orleans, 2007

Rincon-Mora GA, ‘Voltage References: From Diodes to Precision High-Order
Bandgap Circuits’, New York: John Wiley, 2001

Sansen WM, Op’t Eynde F, Steyaert M, ‘A CMOS Temperature-Compensated
Current Reference’, IEEE Journal of Solid-State Circuits, vol. 23, no. 3, pp. 821—824,
June 1988

Gray PR, Hurst PJ, Lewis SH, Meyer RG, ‘Analysis and Design of Analog Integrated
Circuits’, 4th Edition: John Wiley, 2001

Widlar RJ, ‘Design Techniques for Monolithic Operational Amplifiers’, IEEE
Journal of Solid-State Circuits, vol. SC-4, pp. 184—191, 1969

Vittoz E, Fellrath J, ‘CMOS Analog Integrated Circuits Based on Weak Inversion
Operation’, IEEE Journal of Solid-State Circuits, vol. 12, pp. 224—231, 1977

the EECS Department of the University of California at Berkeley, The SPICE page,
http://bwrc.eecs.berke1ey.edu/Classes/icbook/SPICE/, Accessed July 2008

133