HEALTH MONITORING OF STRUCTURES

The last few decades have witnessed the construction of vast infrastructural
facilities in India and other parts of the world. Now, the ageing of these structures is
creating maintenance problems for which engineers are not logistically prepared. In
addition, it is very important to assess the condition of a structure immediately after
extreme loading such as an earthquake. This is mainly to ensure that the building is
still safe. These considerations are prompting the development of automated structural
health monitoring (SHM) and non-destructive evaluation (NDE) systems, which can
provide cost-effective alternative to traditional visual inspection. The concept of
SHM, which was originally applied to aerospace and mechanical systems, is now
being extended to civil-structures. However, the unique feature of most civil
engineering structures is that they are much larger and more complex than the
common mechanical or aerospace systems.

Farrar and Jauregui (1998) defined four distinct levels of damage detection
a) To ascertain that damage has occurred or to identify damage
b) To identify damage and to determine its location
c) To identify and locate damage, and to determine its severity (i.e to
quantify it).
d) To identify, locate and quantify damage and to determine remaining useful
life of the structure

There are various approaches for addressing the problem of health monitoring
and damage detection in structures. Basic principle is that any flaw/ damage change
the characteristics/ response of the structure. Researchers have proposed various
techniques based on different kinds of structural characteristics or responses. The
various methods reported in the literature can be broadly classified into the following
categories:
(i) Global static response based techniques
(ii) Global dynamic response based techniques
(iii)Local damage detection techniques
(iv) Techniques using ‘smart materials’ and ‘smart system’ concept.
 The following sections describe briefly the major developments in recent years
in all the above mentioned areas.

GLOBAL STATIC RESPONSE BASED TECHNIQUES

A technique based on static displacement response was formulated by Banan
et al. (1994). The technique involves applying static forces to the structure and
measuring the corresponding displacements. It is not necessary to select entire set of
forces and displacements. Any convenient subset may be selected. However, several
load cases might be necessary to obtain sufficient information. The resulting
equations have to be solved recursively to arrive at a set of member constitutive
properties or the structural parameters. Any change in parameters from the baseline
healthy state gives the indication of the presence of damage.

The method has a number of shortcomings, particularly in the way of practical

implementation. Measurement of displacements on a large-scale structure is not an
easy task. As a first step, one needs to establish a frame of reference. For contact
measurements, this can entail the construction of a secondary structure on an
independent foundation. Employing a number of load cases can be very time-
consuming. Also, the method requires a lot of computational effort.

Sanayei and Saletnik (1996) proposed a technique based on static strain

measurements. The advantage of this technique over the previous one is that strain
measurements can be made with a higher degree of precision as compared to
displacements. In addition, the establishment of an independent frame of reference is
also circumvented. Strains on the surface are caused by both bending and axial
deformations, and therefore are able to capture the element behavior well. However,
like the displacement approach, this technique is also based on the least squared
minimization of strain error function. The resulting equations are required to be
recursively solved to obtain the vector of change in the structural parameters.
Therefore the method faces similar computational problems. Besides, its application
in real life structures is bound to be very tedious.

The application of large loads to cause measurable deflections (or strains)

warrants huge machinery and power input. As such, these methods are too tedious and
expensive to enable a timely and cost effective assessment of the health of real-life
structures.

GLOBAL DYNAMIC RESPONSE BASED TECHNIQUES

In these techniques, the test-structure is subjected to low-frequency
excitations, either harmonic or impulse, and the resulting vibration responses
(displacements, velocities or accelerations) are picked up at specified locations along
the structure. The vibration pick-up data is processed to extract the first few mode
shapes and the corresponding natural frequencies of the structure, which, when
compared with the corresponding data for the healthy state, yield information
pertaining to the locations and the severity of the damages. A damage in a structure
alters its modal parameters such as modal frequencies, modal damping, and mode
shapes associated with each modal frequency. Changes also occur in structural
parameters, namely stiffness and damping matrices. These techniques have obvious
advantages over the static response techniques and they are easy to implement. Even
the ambient vibration data can be utilized to derive modal data.

Zimmerman and Kaouk (1994) developed “change in stiffness method” based

on this principle. It identifies changes in the structural stiffness matrix resulting from
damage. The stiffness matrix [K] is determined from mode shapes and modal
frequencies derived from the measured dynamic response of the structure, as
 n 
[K] = [M][  ][  ][  ]T[M] = [M]    i2  i  iT  [M] (1)
 i 1 
where [M] is the mass matrix and [  ] the mode shape matrix. n is the total
number of modes considered. Let [  K] represents the change in stiffness matrix due
to damage. The indicator of damage in this method is a damage vector defined by
[Di] = [  K] [  i ] (2)
where [Di] is the ith damage vector obtained from the ith mode shape vector
[  i ]. The location of damage is indicated by the maximum parameter in the damage
vector.

This method has a significant drawback. Higher modes are more important in
the estimation of stiffness matrix, as can be seen from Eq. (1). The modal
contribution to [K] increases as the modal frequency increases. Unfortunately, the
dynamic vibration testing can yield only first few mode shapes. Hence, the estimated
[K] is likely to be inaccurate.

This problem was addressed by Pandey and Biswas (1994), who developed the
change in flexibility method. The basic principle used in this approach is that damage
in a structure alters its flexibility matrix, which can be used to ‘identify’ damage.
Secondly, damage at a particular location alters the respective elements differently.
The relative amount by which the various elements are altered is used to ‘localize’ the
damage. Like the change in stiffness method, mode shape vectors and resonant
frequencies obtained from the dynamic response data (collected before damage and
after damage) are used to obtain the flexibility matrix [F], which may be expressed
as
n
1
[F] = [  ][  ]-1[  ]T = 
i 1
2
 iiT (3)
i

As can be seen from Eq.(3), [F] is proportional to the square of the inverse of
the modal frequencies. Therefore it converges rapidly with increasing frequencies.

50% damage in element 8

Change in flexibility

50% damage in
element 16
1 6 11 16 21 26 31
Node

Fig. 1 Variation of flexibility for two damage locations

Hence, only a few lower modes are sufficient for an accurate estimation of [F]. The
maximum of the values at a particular column  j of [F] is taken as an index of

flexibility for that particular degree of freedom. Any change in this parameter is
regarded as an indicator of damage at that particular degree of freedom. In the case of
a simply supported beam, the change is at maximum near the damaged element. The
amount of change is a measure of the severity of damage. Fig. 1 shows the qualitative
variation of flexibility index for a simply supported beam consisting of 30 elements
and 31 nodes. Curves for two different damage locations are shown. 50% damage
indicated in the figure implies that Young’s modulus for the element has reduced by
50%. It is obvious from the figure that the location of damage can be easily inferred to
be the point of maximum flexibility change. Although the technique is an
improvement over the change in stiffness method, the researchers did not investigate
the case of multiple damage locations.

Stubbs and Kim (1994) developed the damage index method. The damage
index  is calculated based on the strain energy stored in the structure when it
deforms in a particular mode shape. For the j th location and the ith mode, the damage
index  ij is defined as

 b *'' L
 L
  [ i ( x )]2 dx   [ i*'' ( x)] 2 dx  [ i'' ( x)] 2 dx
  
 ij   ab 0  0 (4)
 L
 L
  [ i'' ( x )]2 dx   [ i'' ( x)] 2 dx   [ i*'' ( x)] 2 dx
 
a 0  0

where  i'' ( x ) and  i*'' ( x ) are the second derivatives of the ith mode shape
corresponding to the undamaged and the damaged structures, respectively. Here, “a”
and “b” are limits of a segment of a beam where damage is being evaluated. This
technique has an advantage over the previous techniques in that it has specific criteria
for determining whether damage has occurred at a particular location. Other
techniques look for the largest change in a particular parameter.

Farrar and Jauregui (1998) did a comparative study of the above methods on a
single structure, the I-40 Bridge over Rio Grande in Albuquerque, New Mexico, USA.
The bridge has a maximum span length of about 50m and a girder depth of about 3m.
Damage was induced by means of a torch cut, at the middle of the 50m span and was
started from the mid depth of web as a 0.6m long by 10mm wide crack . It was then
extended in three stages to the entire bottom half of the web and the bottom flange.
The bridge was excited by a hydraulic shaker (both in undamaged and damaged
conditions) and the modal data was extracted. Acquisition

The major conclusions drawn based on the analysis of the response data by
various methods are as follows:
(i) Standard modal properties such as mode shapes and resonant
frequencies are poor indicators of damage. No noticeable change in the
measured resonant frequencies and mode shapes was observed until
the final level of damage.
(ii) All the methods identified the damage location correctly for the most
severe damage case, i.e a cut through the entire bottom half of the
beam.
(iii) In some of the methods human judgment was required to correctly
identify the location of damage. If they were applied blindly, it would
have been difficult to locate the damage.
(iv) When the entire set of tests are considered, the damage index method
is the one with the best performance.

Several other related publications can be found, reporting the use of improved
algorithms, modern wireless technology and high speed data processing (Singhal and
Kiremidjian, 1996; Skjaerbaek et al., 1998; Pines and Lovell, 1998; Aktan et al.,
1998, 2000; Lynch et al., 2003a). However, in spite of rapid progress in the hardware
and the software technologies, the basic principle remains the same, which is to
identify changes in the modal and the structural parameters (or their derivatives)
resulting from damages. The main limitations of the global dynamic techniques can
be summarized as follows

(i) These techniques typically rely on the first few mode shapes and the
corresponding natural frequencies of structures, which, being global in
nature, are not sensitive enough to be altered by localized incipient
damages. For example, Pandey and Biswas (1994) reported that a 50%
reduction in the Young’s modulus of elasticity, over the central 3%
length of a 2.44m long beam (used by the investigators as an example),
only resulted in about 3% reduction in the observed first natural
 frequency. Changes of such small order of magnitude may not be
considered as reliable damage indicators in real-life structures, in light
of experimental errors of about the same order of magnitude.

The global parameters (on which these techniques heavily rely) do not alter
significantly due to local damages. In physical terms, the reason for this is attributed
to the fact that the long wavelength stress waves associated with the low-frequency
modes may cross a local damage (such as a crack), without sensing it. It is for this
reason that Farrar and Jauregui (1998) found that the global dynamic techniques failed
to identify damage locations for less severe damage scenarios in their experiments. It
could be possible that damage, just large enough to be detected by global dynamic
techniques, may already be critical for the structure in question.
(ii) These techniques demand expensive hardware and sensors, such as
inertial shakers, self-conditioning accelerometers and laser velocity
meters. For a large structure, the overall cost of such sensor systems
could easily run into millions of dollars. For example, the Tsing Ma
suspension bridge in Hong Kong was instrumented with only 350
sensors in 1997 with a total cost of over US$ 8 million.
(iii) A major limitation of these techniques is the interference caused by the
ambient mechanical noise, besides the electrical and the
electromagnetic noise associated with the measurement systems. Due
to low frequency, the techniques are highly susceptible to ambient
noise, which also happens to be in the low frequency range, typically
less than 100Hz.
(iv) The pre-requisite of a high fidelity ‘model’ of the test structure restricts
the application of the methods to relatively simple geometries and
configurations only. Because evaluation of stiffness and damping at the
supports (which are often rusted during service), is extremely difficult,
reliable identification of a ‘model’ is quite difficult in practice.
(v) Often, the performance of these techniques deteriorates in multiple
damage scenarios (Wang et al., 1998).
LOCAL SHM TECHNIQUES
Another category of damage detection techniques is formed by the so-called
local methods, which, as opposed to the global techniques (static/ dynamic), rely on
localized structural interrogation for detecting damages. Some of the methods in this
category are the ultrasonic techniques, acoustic emission, eddy currents, impact echo
testing, magnetic field analysis, penetrant dye testing, and X-ray analysis.

The ultrasonic methods are based on elastic wave propagation and reflection
within the material for non-destructive strength characterization and for identifying
field inhomogeneities caused by damages. In these methods, a probe (a piezo-electric
crystal) is employed to transmit high frequency waves into the material. These waves
reflect back on encountering any crack, whose location is estimated from the time
difference between the applied and the reflected waves. These techniques exhibit
higher damage sensitivity as compared to the global techniques, due to the utilization
of high frequency stress waves. Shah et al. (2000) reported a new ultrasonic wave
based method for crack detection in concrete from one surface only. Popovics et al.
(2000) similarly developed a new ultrasonic wave based method for layer thickness
estimation and defect detection in concrete. In spite of high sensitivity, the ultrasonic
methods share few limitations, such as:

(i) They typically employ large transducers and render the structure
unavailable for service throughout the length of the test.
(ii) The measurement data is collected in time domain that requires
complex processing.
(iii) Since ultrasonic waves cannot be induced at right angles to the surface,
they cannot detect transverse surface cracks (Giurgiutiu and Rogers,
1997).
(iv) These techniques do not lend themselves to autonomous use since
experienced technicians are required to interpret the data.

In acoustic emission method, another local method, elastic waves generated by

plastic deformations (such as at the tip of a newly developed crack), moving
dislocations and disbonds are utilized for analysis and detection of structural defects.
It requires stress or chemical activity to generate elastic waves and can be applied on
the loaded structures also (Boller, 2002), thereby facilitating continuous surveillance.
However, the main problem to damage identification by acoustic emission is posed by
the existence of multiple travel paths from the source to the sensors. Also,
contamination by electrical interference and mechanical ambient noise degrades the
quality of the emission signals (Park et al., 2000a; Kawiecki, 2001).

The eddy currents perform a steady state harmonic interrogation of structures

for detecting surface cracks. A coil is employed to induce eddy currents in the
component. The interrogated component, in-turn induces a current in the main coil
and this induction current undergoes variations on the development of damage, which
serves an indication of damage. The key advantage of the method is that it does not
warrant any expensive hardware and is simple to apply. However, a major drawback
of the technique is that its application is restricted to conductive materials only, since
it relies on electric and magnetic fields. A more sophisticated version of the method is
magneto-optic imaging, which combines eddy currents with magnetic field and
optical technology to capture an image of the defects (Ramuhalli et al., 2002).

In impact echo testing, a stress pulse is introduced into the interrogated

component using an impact source. As the wave propagates through the structure, it is
reflected by cracks and disbonds. The reflected waves are measured and analysed to
yield the location of cracks or disbonds. Though the technique is very good for
detecting large voids and delaminations, it is insensitive to small sized cracks (Park et
al., 2000a).

In the magnetic field method, a liquid containing iron powder is applied on the
component to be interrogated, subjected to magnetic field, and then observed under
ultra-violet light. Cracks are detected by appearance of magnetic field lines along
their positions. The main limitation of the method is that it is applicable on magnetic
materials only. Also, the component must be dismounted and inspected inside a
special cabin. Hence, the technique not very suitable for in situ application.

In the penetrant dye test, a coloured liquid is brushed on to the surface of the
component under inspection, allowed to penetrate into the cracks, and then washed off
the surface. A quick drying suspension of chalk is thereafter applied, which acts as a
developer and causes coloured lines to appear along the cracks. The main limitation of
this method is that it can only be applied on accessible locations of structures since it
warrants active human intervention.

In X-ray method, the test structure is exposed to X-rays, which are then re-
caught on film, where the cracks are delineated as black lines. Although the method
can detect moderate sized cracks, very small surface cracks (incipient damages) are
difficult to be captured. A more recent version of the X-ray technique is computer
tomography, whereby a cross-sectional image of solid objects can be obtained.
Although originally used for medical diagnosis, the technique is recently finding its
use for structural NDE also (e.g. Kuzelev et al., 1994). By this method, defects
exhibiting different density and/ or contrast to the surroundings can be identified.

Table.1 summarises the typical damage sensitivities of the local NDE methods
described above. A common limitation of the local methods is that usually, probes,
fixtures and other equipment need to be physically moved around the test-structure for
recording data. Often, this not only prevents autonomous application of the technique,
but may also demand the removal of finishes or covers such as false ceilings. Moving
the probe everywhere being impractical, these techniques are often applied at very
selected probable damage locations (often based on preliminary visual inspection or
past experience), which is almost tantamount to knowing the damage location a
priori. Generally, they cannot be applied while the component is under service, such
as in the case of an aircraft during flight. Computer tomography and X-ray
techniques, due to their high equipment cost, are limited to very high performance
components only (Boller, 2002).
 Table 1 Sensitivities of common local NDE techniques (Boller, 2002).
Method Minimum High probability Remarks
detectable crack detectable crack
length length (>95%)
Ultrasonic 2mm 5-6mm Dependent upon structure
geometry and material
Eddy currents 2mm 4.5-8mm Suitable for thickness
(low-frequency) <12mm only
Eddy currents 2mm (surface) 2.5mm (surface)
(high-frequency) 0.5mm (bore holes) 1.0mm (bore holes)
X – Ray 4mm 10mm Dependent upon structure
configuration. Better for
thickness > 12mm
Magnetic particle 2mm 4mm surface
Dye penetrant 2mm 10mm surface

ADVENT OF SMART MATERIALS, STRUCTURES AND SYSTEMS FOR

SHM AND NDE
The SHM/ NDE methods described so far are the conventional monitoring
techniques. They typically rely on the measurement of stresses, strains, displacements,
accelerations or other related physical responses to identify damages. The
conventional sensors, which these techniques employ, are passive and bulky, and can
only extract secondary information such as load and strain history, which may not
lead to any direct information about damages (Giurgiutiu et al., 2000).

However, the past few years have witnessed the emergence of ‘smart’ materials,
systems and structures, which have shown new possibilities for SHM and NDE. Due
to their inherent ‘smartness’, the smart materials work on fundamentally different
principles and exhibit greater sensitivities to any changes in the environment. The
next section briefly describes the principles and the recent developments in SHM/
NDE based on smart structures and materials.

Definition of Smart Systems/ Structures

The definition of smart structures was a topic of controversy from the late
1970’s to the late 1980’s. In order to arrive at a consensus for major terminology, a
special workshop was organised by the US Army Research Office in 1988, in which
‘sensors’, ‘actuators’, ‘control mechanism’ and ‘timely response’ were recognised as
the four qualifying features of any smart system or structure (Rogers, 1988). In this
workshop, following definition of smart systems/ structures was formally adopted
(Ahmad, 1988).

“A system or material which has built-in or intrinsic sensor(s), actuator(s)

and control mechanism(s) whereby it is capable of sensing a stimulus, responding to
it in a predetermined manner and extent, in a short/ appropriate time, and reverting
to its original state as soon as the stimulus is removed”

Previously, the words ‘intelligent’, ‘adaptive’ and ‘organic’ were also used to
characterize smart systems and materials. For example, Crawley and de Luis (1987)
defined ‘intelligent structures’ as the structures possessing highly distributed
actuators, sensors and processing networks. Similarly, Professor H. H. Robertshaw
preferred the term ‘organic’ (Rogers, 1988) which suggests similarity to biological
processes. The human arm, for example, is like a variable stiffness actuator with a
control law (intelligence). However, many participants at the US Army Research
Office Workshop (e.g. Rogers et al., 1988) sought to differentiate the terms
‘intelligent’, ‘adaptive’ and ‘organic’ from the term ‘smart’ by highlighting their
subtle differences with the term ‘smart’. The term ‘intelligence’, for example, is
associated with abstract thought and learning, and till date has not been implemented
in any form of adaptive and sensing material or structure. However, still many
researchers use the terms ‘smart’ and ‘intelligent’ almost interchangeably (e.g. In the
U.S.-Japan Workshop: Takagi, 1990; Rogers, 1990), though ‘adaptive’ and ‘organic’
have become less popular.

The idea of ‘smart’ or ‘intelligent’ structures has been adopted from nature,
where all the living organisms possess stimulus-response capabilities (Rogers, 1990).
The aim of the ongoing research in the field of smart systems/ structures is to enable
such a structure or system mimic living organisms, which possess a system of
distributed sensory neurons running all over the body, enabling the brain to monitor
the condition of the various body parts. However, the smart systems are much inferior
to the living beings since their level of intelligence is much primitive.
Smart Materials
Smart materials are new generation materials surpassing the conventional
structural and functional materials. These materials possess adaptive capabilities to
external stimuli, such as loads or environment, with inherent intelligence. In the US
Army Research Office Workshop, Rogers et al. (1988) defined smart materials as
materials, which possess the ability to change their physical properties in a specific
manner in response to specific stimulus input. The stimuli could be pressure,
temperature, electric and magnetic fields, chemicals or nuclear radiation. The
associated changeable physical properties could be shape, stiffness, viscosity or
damping. This kind of ‘smartness’ is generally programmed by material composition,
special processing, introduction of defects or by modifying the micro-structure, so as
to adapt to the various levels of stimuli in a controlled fashion. Like smart structures,
the terms ‘smart’ and ‘intelligent’ are used interchangeably for smart materials.
Takagi (1990) defined intelligent materials as the materials which respond to
environmental changes at the most optimum conditions and manifest their own
functions according to the environment. The feedback functions within the material
are combined with properties and functions of the materials.

Optical fibres, piezo-electric polymers and ceramics, electro-rheological (ER)

fluids, magneto-strictive materials and shape memory alloys (SMAs) are some of the
smart materials. Fig. 2 shows the associated ‘stimulus’ and ‘response’ of common
smart materials. Because of their special ability to respond to stimuli, they are finding
numerous applications in the field of sensors and actuators. A very detailed
description of smart materials is covered by Gandhi and Thompson (1992).

Smart materials can be either active or passive. Fairweather (1998) defined

active smart materials as those materials which possess the capacity to modify their
geometric or material properties under the application of electric, thermal or magnetic
fields, thereby acquiring an inherent capacity to transduce energy. Piezoelectric
materials, SMAs, ER fluids and magneto-strictive materials are active smart
materials. Being active, they can be used as force transducers and actuators. For
example, the SMA has large recovery force, of the order of 700 MPa (105 psi)
(Kumar, 1991), which can be utilized for actuation. Similarly piezoelectric materials,
which convert electric energy into mechanical force, are also ‘active’.
 The smart materials, which are not active, are called passive smart materials.
Although smart, these lack the inherent capability to transduce energy. Fibre optic
material is a good example of a passive smart material. Such materials can act as
sensors but not as actuators or transducers.

Smart Materials: Future Applications

(1) Stress (1) Electric Charge

Piezoelectric
(2) Electric field Material (2) Mechanical strain

Heat Shape Memory Original Memorized Shape

Alloy

Electro-rheological Change in viscosity

Electric field
Fluid (Internal damping)

Temperature, pressure, Change in Opto-

Optical Fibre Electronic signals
mechanical strain

Magnetic field Magneto-strictive Mechanical Strain

material

Fig. 2 Common smart materials and the associated stimulus-response.

Seasoned researchers often share visionary ideas about the future of smart
materials in conferences and seminars. According to Prof. Rogers (Rogers, 1990),
following advancements could be possible in the field of smart materials and
structures.
 Materials which can restrain the propagation of cracks by automatically producing
compressive stresses around them (Damage arrest).
 Materials, which can discriminate whether the loading is static or shock and can,
generate a large force against shock stresses (Shock absorbers).
 Materials possessing self-repairing capabilities, which can heal damages in due
course of time (Self-healing materials).
 Materials which are usable up to ultra-high temperatures (such as those
encountered by space shuttles when they re-enter the earth’s atmosphere from
 outer space), by suitably changing composition through transformation (thermal
mitigation).

Takagi (1990) similarly projected the development of more functional and

higher grade materials with recognition, discrimination, adjustability, self-diagnostics
and self-learning capabilities.

PIEZOELECTRICITY AND PIEZOELECTRIC MATERIALS

The word ‘piezo’ is derived from a Greek word meaning pressure. The
phenomenon of piezoelectricity was discovered in 1880 by Pierre and Paul-Jacques
Curie. It occurs in non-centro symmetric crystals, such as quartz (SiO 2), Lithium
Niobate (LiNbO3), PZT [Pb(Zr1-xTix)O3)] and PLZT [(Pb1-xLax)(Zr1-yTiy)O3)], in
which electric dipoles (and hence surface charges) are generated when the crystals are
loaded with mechanical deformations. The same crystals also exhibit the converse
effect; that is, they undergo mechanical deformations when subjected to electric
fields.

Constitutive Relations
The constitutive relations for piezoelectric materials, under small field
condition are (IEEE standard, 1987)

Di   ijT E j  dimd Tm (5)

Sk  d cjk E j  skm
E
Tm (6)

Eq. (5) represents the so called direct effect (that is stress induced electrical
charge) whereas Eq. (6) represents the converse effect (that is electric field induced
mechanical strain). Sensor applications are based on the direct effect, and actuator
applications are based on the converse effect. When the sensor is exposed to a stress
field, it generates proportional charge in response, which can be measured. On the
other hand, the actuator is bonded to the structure and an external field is applied to it,
which results in an induced strain field. In more general terms, Eqs. (5) and (6) can be
rewritten in the tensor form as (Sirohi and Chopra, 2000b)

D  T d d  E 
S    c   (7)
   d s E  T 
 where [D] (3x1) (C/m2) is the electric displacement vector, [S] (3x3) the
second order strain tensor, [E] (3x1) (V/m) the applied external electric field vector

and [T] (3x3) (N/m2) the stress tensor. Accordingly, [  T ] (F/m) is the second order
dielelectric permittivity tensor under constant stress, [dd] (C/N) and [dc] (m/V) the

third order piezoelectric strain coefficient tensors, and [ s E ] (m2/N) the fourth order
elastic compliance tensor under constant electric field.

Commercial Piezoelectric Materials

Previously, piezoelectric crystals, which used to be brittle and of large weight,
were used in practice. However, now the commercial piezoelectric materials are
available as ceramics or polymers, which can be cut into a variety of convenient
shapes and sizes and can be easily bonded.

Piezoceramics
Lead zirconate titanate oxide or PZT, which has a chemical composition
[Pb(Zr1-xTix)O3)], is the most widely used type piezoceramic. It is a solid solution of
lead zirconate and lead titanate, often doped with other materials to obtain specific
properties. It is manufactured by heating a mixture of lead, zirconium and titanium oxide
powders to around 800-1000oC first to obtain a perovskite PZT powder, which is mixed
with a binder and sintered into the desired shape. The resulting unit cell is elongated in
one direction and exhibits a permanent dipole moment along this axis. However, since
the ceramic consists of many such randomly oriented domains, it has no net polarization.
Application of high electric field aligns the polar axes of the unit cells along the applied
electric field, thereby reorienting most of the domains. This process is called poling and
it imparts a permanent net polarization to the crystal. This also creates a permanent
mechanical distortion, since the polar axis of the unit cell is longer than other two axes.
Due to this process, the material becomes piezoelectrically transversely isotropic in the
plane normal to the poling direction i.e. d31 = d32  d33; d15 = d24, but remains
mechanically isotropic (Sirohi and Chopra, 2000b).

PZT is a very versatile smart material. It is chemically inert and exhibits high
sensitivity of about 3V/Pa, that warrants nothing more sophisticated than a charge
amplifier to buffer the extremely high source impedance of this largely capacitive
transducer. It demonstrates competitive characteristics such as light weight, low-cost,
small size and good dynamic performance. Besides, it exhibits large range of linearity
(up to electric field of 2kV/cm, Sirohi and Chopra, 2000a), fast response, long term
stability and high energy conversion efficiency. The PZT patches can be manufactured
in any shape, size and thickness (finite rectangular shapes to complicated MEMS
shapes) at relatively low-cost as compared to other smart materials and can be easily
used over a wide range of pressures without serious non-linearity. The PZT material is
characterized by a high elastic modulus (comparable to that of aluminum). However,
PZT is somewhat fragile due to brittleness and low tensile strength. Tensile strength
measured under dynamic loading is much lower (about one-third) than that measured
under static conditions. This is because under dynamic loads, cracks propagate much
faster, resulting in much lower yield stress. Typically, G1195 (Piezo Systems Inc., 2003)
has a compressive strength of 520 MPa and a tensile strength of 76 MPa (static) and 21
MPa (dynamic) (Zhou et al., 1995). The PZT materials have negative d31, which implies
that a positive electric field (in the direction of polarization) results in compressive strain
on the PZT sheet. If heated above a critical temperature, called the Curie temperature,
the crystals lose their piezoelectric effect. The Curie temperature typically varies from
150oC to 350oC for most commercial PZT crystals. When exposed to high electric fields
(>12 kV/cm), opposite to the poling direction, the PZT loses most of its piezoelectric
capability. This is called deploing and is accompanied by a permanent change in the
dimensions of the sample.

Due to high stiffness, the PZT sheets are good actuators. They also exhibit
high strain coefficients, due to which they can act as good sensors also. These features
make the PZT materials very suitable for use as collocated actuators and sensors.
They are used in deformable mirrors, mechanical micropositioners, impact devices
and ultrasonic motors (Kumar, 1991), sonic and ultrasonic sensors, filters and
resonators, signal processing devices, igniters and voltage transformers (Zhu, 2003),
to name only a few. For achieving large displacements, multi layered PZT systems
can be manufactured, such as stack, moonie and bimorph actuators.

However, due to their brittleness, the PZT sheets cannot withstand bending
and also exhibit poor conformability to curved surfaces. This is the main limitation
with PZT materials. In addition, the PZT materials show considerable fluctuation of
their electric properties with temperature. Also, soldering wires to the electroded
piezoceramics requires special skill and often results in broken elements, unreliable
connections or localized thermal depoling of the elements. As a solution to these
problems, active piezoceramic composite actuators (Smart Materials Corporation),
active fibre composites (Massachusetts Institute of Technology) and macro fibre
composites, MFCs (NASA, Langley Centre) have been developed recently (Park et
al., 2003a). The MFCs have been commercially available since 2003. These new
types of PZTs are low-cost, damage tolerant, can conform to curved surface and are
embeddable. In addition, Active Control eXperts, Inc. (ACX), now owned by Mide
Technology Corporation, has developed a packaging technology in which one or more
PZT elements are laminated between sheets of polymer flexible printed circuitry. This
provides the much robustness, reliability and ease of use. The packaged sensors are
commercially called QuickPack® actuators (Mide Technology Corporation, 2004).
These are now widely used as vibration dampers in sporting goods, buzzer alerts,
drivers for flat speakers and more recently in automotive and aerospace components
(Pretorius et al., 2004). However, these are presently many times expensive than raw
PZT patches.

Piezopolymers
The most common commercial piezopolymer is the Polyvinvylidene Fluoride
(PVDF). It is made up of long chains of the repeating monomer (-CH2-CF2-) each of
which has an inherent dipole moment. PVDF film is manufactured by solidification
from the molten phase, which is then stretched in a particular direction and poled. The
stretching process aligns the chains in one direction. Combined with poling, this
imparts a permanent dipole moment to the film. Because of stretching, the material is
rendered piezoelectrically orthotropic, that is d31  d32, where ‘1’ is the stretching
direction. However, it still remains mechanically isotropic.

The PVDF material is characterized by low stiffness (Young’s modulus is

1/12th that of aluminum). Hence, the PVDF sensors are not likely to modify the
stiffness of the host structure due to their own stiffness. Also, PVDF films can be
shaped as desired according to the intended application. Being polymer, it can be
formed into very thin sheets and adhered to curved surfaces also due to its flexibility.
These characteristics make PVDF films more attractive for sensor applications, in
spite of their low piezoelectric coefficients (approximately 1/10th of PZT). It has been
shown by Sirohi and Chopra (2000b) that shear lag effect is negligible in PVDF
sensors.

Piezo-rubber, which consists of fine particles of PZT material embedded in

synthetic rubber (Rogers, 1990), has appeared as an alternative for PVDF. The piezo-
rubber shows much higher electrical output due to larger thickness, which is not
possible in PVDF. The piezo-rubber is used in piezoelectric coaxial cable as a vehicle
sensor. It has much longer life and is immune to rain water.

Applications of Piezoelectric Materials

Since this thesis is primarily concerned with piezoelectric materials, some
typical applications of these materials are briefly described here. Traditionally,
piezoelectric materials have been well-known for their use in accelerometers, strain
sensors (Sirohi and Chopra, 2000b), emitters and receptors of stress waves (Giurgiutiu
et al., 2000; Boller, 2002), distributed vibration sensors (Choi and Chang, 1996;
Kawiecki, 1998), actuators (Sirohi and Chopra, 2000a) and pressure transducers (Zhu,
2003). However, since the last decade, the piezoelectric materials, their derivative
devices and structures have been increasingly employed in turbo-machinery actuators,
vibration dampers and active vibration control of stationary/ moving structures (e.g.
helicopter blades, Chopra, 2000). They have been shown to be very promising in
active structural control of lab-sized structures and machines (e.g. Manning et al.,
2000; Song et al., 2002). Structural control of large structures has also been attempted
(e.g. Kamada et al., 1997). Other new applications include underwater acoustic
absorption, robotics, precision positioning and smart skins for submarines (Kumar,
1991). Skin-like tactile sensors utilizing piezoelectric effect for sensing temperatures
and pressures have been reported (Rogers, 1990). Very recently, the piezoelectric
materials have been employed to produce micro and nano scale systems and wireless
inter digital transducers (IDT) using advanced embedded system technologies, which
are set to find numerous applications in micro-electronics, bio-medical and SHM
(Varadan, 2002; Lynch et al., 2003b). Recent research is also exploring the
development of versatile piezo-fibres, which can be integrated with composite
structures for actuation and SHM (Boller, 2002
 The most striking application of the piezoelectric materials in SHM has been
in the form of EMI technique, which evolved during the last decade. The next section
covers the various aspects of this technique in detail.

ELECTRO-MECHANICAL IMPEDANCE (EMI) TECHNIQUE

Physical Principles
The EMI technique is very similar to the conventional global dynamic
response techniques described previously. The major difference is with respect to the
frequency range employed, which is typically 30-400 kHz in EMI technique, against
less than 100Hz in the case of the global dynamic methods.

In the EMI technique, a PZT patch is bonded to the surface of the monitored
structure using a high strength epoxy adhesive, and electrically excited via an
impedance analyzer. In this configuration, the PZT patch essentially behaves as a thin
bar undergoing axial vibrations and interacting with the host structure, as shown in
Fig. 3 (a). The PZT patch-host structure system can be modelled as a mechanical
impedance (due the host structure) connected to an axially vibrating thin bar (the
patch), as shown in Fig. 3 (b). The patch in this figure expands and contracts
dynamically in direction ‘1’ when an alternating electric field E3 (which is spatially
3 (z) 2 (y)
1 (x) Alternating electric
PZT Patch field source 3
2
l l Host E3
1
structure PZT patch

Point of Z w Z
mechanical h
fixity
Structural
l l
Impedance

(a) (b)
Fig. 3 Modelling PZT-structure interaction.
(a) A PZT patch bonded to structure under electric excitation.
(b) Interaction model of PZT patch and host structure.

uniform i.e. ∂E3/∂x = ∂E3/∂y = 0) is applied in the direction ‘3’. The patch has half-
length ‘l’, width ‘w’ and thickness ‘h’. The host structure is assumed to be a skeletal
structure, that is, composed of one-dimensional members with their sectional
properties (area and moment of inertia) lumped along their neutral axes. Therefore,
the vibrations of the PZT patch in direction ‘2’ can be ignored. At the same time, the
PZT loading in direction ‘3’ is neglected by assuming the frequencies involved to be
much less than the first resonant frequency for thickness vibrations. The vibrating
patch is assumed infinitesimally small and to possess negligible mass and stiffness as
compared to the host structure. The structure can therefore be assumed to possess
uniform dynamic stiffness over the entire bonded area. The two end points of the
patch can thus be assumed to encounter equal mechanical impedance, Z, from the
structure, as shown in Fig. 3 (b). Under this condition, the PZT patch has zero
displacement at the mid-point (x= 0), irrespective of the location of the patch on the
host structure. Under these assumptions, the constitutive relations (Eqs.5 and 6) can
be simplified as (Ikeda, 1990)

D3   33
T
E 3  d 31T1 (8)

T1
S1   d31E3 (9)
YE

1
Fig. 4 A piezoelectric material sheet with conventional 1, 2 and 3 axes.

where S1 is the strain in direction ‘1’ (see Fig. 4), D3 the electric displacement
over the PZT patch, d31 the piezoelectric strain coefficient and T1 the axial stress in

direction ‘1’. Y E  Y E (1  j ) is the complex Young’s modulus of elasticity of the

PZT patch at constant electric field and  33

T
  33
T
(1  j ) is the complex electric

permittivity (in direction ‘3’) of the PZT material at constant stress, where j   1 .
Here,  and  denote respectively the mechanical loss factor and the dielectric loss
factor of the PZT material.

The one-dimensional vibrations of the PZT patch are governed by the

following differential equation (Liang et al., 1994), derived based on dynamic
equilibrium of the PZT patch.

 2u  2u
YE   (10)
x 2 t 2

where ‘u’ is the displacement at any point on the patch in direction ‘1’.
Solution of the governing differential equation by the method of separation of
variables yields

u  ( A sin x  B cosx)e jt (11)

where  is the wave number, related to the angular frequency of excitation ,

the density  and the complex Young’s modulus of elasticity of the patch by


  (12)
YE

Application of the mechanical boundary condition that at x = 0 (mid point of

the PZT patch), u = 0 yields B = 0.
Hence, strain in PZT patch
u
S1 ( x)   Ae jt cos x (13)
x
u
and velocity u ( x)   Aje jt sin x (14)
t

Further, by definition, the mechanical impedance Z of the structure is related

to the axial force F in the PZT patch by
F( x  l )  whT1( x  l )   Zu( x  l ) (15)

where the negative sign signifies the fact that a positive displacement (or
velocity) causes compressive force in the PZT patch (Liang et al., 1993, 1994).
Making use of Eq. (15) and substituting the expressions for strain and velocity from
Eqs. (14) and (15) respectively, we can derive
Z aVo d31
A (16)
h cos(l )( Z  Z a )

where Za is the short-circuited mechanical impedance of the PZT patch, given

by

whY E
Za  (17)
( j ) tan(l )
Za is defined as the force required to produce unit velocity in the PZT patch in short
circuited condition (i.e. ignoring the piezoelectric effect) and ignoring the host
structure.

The electric current, which is the time rate of change of charge, can be
obtained as

I   D 3 dxdy  j  D3 dxdy (18)

A A

Making use of the PZT constitutive relation (Eq. 8), and integrating over the
entire surface of the PZT patch (-l to +l), we can obtain an expression for the
electromechanical admittance (the inverse of electro-mechanical impedance) as

wl  T  Z a  2 E  tan l 
Y  2j ( 33  d 31Y )   d 31Y 
2 E
 (19)
h   Z  Za   l 

This equation is same as that derived by Liang et al. (1994), except that an
additional factor of 2 comes into picture. This is due to the fact that Liang et al. (1993,
1994) considered only one-half of the patch in their derivation.

In the EMI technique, this electro-mechanical coupling between the

mechanical impedance Z of the host structure and the electro-mechanical admittance
Y is utilized in damage detection. Z is a function of the structural parameters- the
stiffness, the damping and the mass distribution. Any damage to the structure will
cause these structural parameters to change, and hence alter the drive point
mechanical impedance Z. Assuming that the PZT parameters remain unchanged, the
electromechanical admittance Y will undergo a change and this serves as an indicator
of the state of health of the structure. Measuring Z directly may not be feasible, but Y
can be easily measured using any commercial electrical impedance analyzer.
Common damage types altering local structural impedance Z are cracks, debondings,
corrosion and loose connections (Esteban, 1996), to which the PZT admittance
signatures show high sensitivity. Contrary to low-frequency vibration techniques,
damping plays much more significant role in the EMI technique due to the
involvement of ultrasonic frequencies. Most conventional damage detection
algorithms (in low-frequency dynamic techniques), on the other hand are based on
damage related changes in structural stiffness and inertia, but rarely in damping
(Kawiecki, 2001).

The electromechanical admittance Y (unit Siemens or ohm-1) consists of real

and imaginary parts, the conductance (G) and susceptance (B), respectively. A plot of
G over a sufficiently wide band of frequency serves as a diagnosis signature of the
structure and is called the conductance signature or simply signature. Fig. (5)
shows the typical conductance and susceptance plots for a PZT patch bonded on to the
bottom flange of a steel beam (Bhalla et al., 2001). The sharp peaks in the
conductance signature correspond to structural modes of vibration. This is how the
conductance signature identifies the local structural system (in the vicinity of the
patch) and hence constitutes a unique health-signature of the structure at the point of
attachment.

Since the real part actively interacts with the structure, it is traditionally
preferred over the imaginary part in the SHM applications. It is believed that the
imaginary part (susceptance) has very weak interaction with the structure (Sun et al.,
1995). Therefore, all investigators have so far considered it redundant and have solely
utilized the real part (conductance) alone in the SHM applications.
 0.0008 0.008

Conductance (S)

Susceptence (S)
0.0007

0.0006 0.006

0.0005

0.0004 0.004
140 142 144 146 148 150 140 142 144 146 148 150

Frequency (kHz) Frequency (kHz)

Fig. 5 Conductance and susceptance plots of a PZT patch bonded to

bottom flange of a steel beam.

Method of Application
In the EMI technique, a PZT actuator/ sensor patch is bonded to the surface of
the structure (whose health is to be monitored) using high strength epoxy adhesive.
The conductance signature of the patch is acquired over a high frequency range (30-
400 kHz). This signature forms the benchmark for assessing the structural health. At
any future point of time, when it is desired to assess the health of the structure, the
signature is extracted again and compared with the benchmark signature.

The signature of the bonded PZT patch is usually acquired by means of

commercially available impedance analyzers, such as HP 4192A impedance analyzer
(Hewlett Packard, 1996). The impedance analyzer imposes an alternating voltage
signal of 1 volts rms (root mean square) to the bonded PZT transducer over the user
specified preset frequency range (for example 140-150 kHz in Fig. 5). The magnitude
and the phase of the steady state current are directly recorded in the form of
conductance and susceptance signatures in the frequency domain, thereby eliminating
the requirements of domain transforms. Besides, no amplifying device is necessary. In
fact, Sun et al. (1995) reported that higher excitation voltage has no influence on the
conductance signature, but might only be helpful in amplifying weak structural
modes.

Major Technological Developments During Last ten Years

Major developments and contributions made by various researchers in the
field of EMI technique during the last nine years are summarised as follows. (A very
detailed review of the various case studies and applications is covered by Park et al.,
2003b).
(1) Application of the EMI technique for SHM on a lab sized truss
structure was first reported by Sun et al. (1995). This study was then
extended to a large-scale prototype truss joint by Ayres et al. (1998).
(2) Lopes et al. (1999) trained neural networks using statistical damage
quantifiers (Area under the conductance curve, root mean square
(RMS) of the curve, root mean square deviation (RMSD) between
damaged and undamaged curves and correlation coefficients) using
experimental data from a bolted joint structure. The trained neural
networks were found to successfully locate and quantify the damages
inflicted on the test structure in a different experiment.
(3) Park et al. (2000a) reported significant proof-of-concept applications
of the EMI technique on civil-structural components such as composite
reinforced masonry walls, steel bridge joints and pipe joints. The
technique was found to be very tolerant to mechanical noise and also to
small temperature fluctuations.
(4) Park (2000) extended the EMI technique to high temperature
applications (typically > 500oC), such as steam pipes and boilers in
power plants. Besides, he also developed practical statistical cross-
correlation based methodology for temperature compensation. This
paved way for application of the technique to real situations, where the
effects of damage and temperature are mixed.
(5) Soh et al. (2000) established the damage detection and localization
ability of piezo-impedance transducers on real-life RC structures by
successfully monitoring a 5m span RC bridge during its destructive
load testing. Besides, criteria were outlined for transducer positioning,
damage localization and transducer validation.
(6) Park et al. (2000b) were the first to integrate the EMI technique with
wave propagation modelling for thin beams (1D structures) under
‘free-free’ boundary conditions, by utilizing axial modes. The
conventional statistical indices of the EMI technique were used for
locating the damages in the frequency range 70-90 kHz. The damage
severity was determined by spectral finite element based wave
propagation approach, in the frequency range 10-40 kHz. However,
this combination necessitated the use of some additional hardware and
 sensors, such as accelerometers, which are not accurate at ultrasonic
frequencies. Also, the application of the wave propagation approach
demands additional computational effort, which could restrict the
application to simple structures only. Besides, the integration of the
EMI technique with wave propagation approach was not seamless in
true sense.
(7) After the year 2000, numerous papers appeared in the literature
demonstrating successful extension of the technique on sophisticated
structural components such as restrengthened concrete members (Saffi
and Sayyah, 2001) and jet engine components under high temperature
condtions (Winston et al., 2001).
(8) Inman et al. (2001) proposed a novel technique to utilize a single PZT
patch for health monitoring as well as for vibration control
(9) Abe et al. (2002) developed a new stress monitoring technique for thin
structural elements (such as strings, bars and plates) by applying wave
propagation theory to the EMI measurement data in the moderate
frequency range (1-10kHz). This has paved way for the application of
the technique for load monitoring, besides damage detection. The
major advantage is that owing to localized wave propagation, the
technique is insensitive to boundary conditions and can make accurate
stress identification. However, the suitable frequency band for this
application is very narrow, and generally difficult to identify. Also, the
method is prone to high errors, especially in 2D components, due to
imprecise modelling of the interfacial bonding layer.
(10) Giurgiutiu et al. (2002) combined the EMI technique with wave
propagation approach for crack detection in aircraft components.
While the EMI technique was employed for near field damage
detection, the guided ultrasonic wave propagation technique (pulse
echo) was used for far field damage detection.
(11) Peairs et al. (2003) developed a novel low-cost and portable version of
impedance analyzer, the major hardware used in the EMI technique,
paving way for significant cost-reduction. Integration of the EMI
technique with wireless technology and development of stand-alone
sensor cum processor cum transmission units based on MEMS and
 inter digital transducers (IDT) is also underway (Park et al., 2003b)
which would enable large-scale instrumentation and monitoring of
civil-structures.

Details of PZT Patches

In the EMI technique, the same PZT patch serves the actuating as well as the
sensing functions. Fig. 6 shows a typical commercially available PZT patch suitable
for this particular application (PI Ceramic, 2003). The characteristic feature of the
patch is that the electrode from the bottom edge is wrapped around the thickness, so
that both the electrodes are available on one side of the PZT patch, while the other
side is bonded to the host structure. PZT patches of sizes ranging from 5mm to 15mm
and thickness from 0.1mm to 0.3mm are best suited for most structural materials such
as steel and RC. Such thin patches usually have thickness resonance frequency of the
order of few MHz. Hence, the frequency response signature is relatively flat in 30-400
kHz frequency range.

10mm
10mm

Top electrode film

Bottom electrode film
wrapped to top surface

Fig. 6 A typical commercially available PZT patch.

Selection of Frequency Range

The operating frequency range must be maintained in hundreds of kHz so that
the wavelength of the resulting stress waves is smaller than the typical size of the
defects to be detected (Giurgiutiu and Rogers, 1997). Typically, for such high
frequencies, wavelengths as small as few mm are generated. Contrary to the large
wavelength stress waves in the case of low frequency techniques, these are
substantially attenuated by the occurrence of any incipient damages (such as cracks)
in the local vicinity of the PZT patch.
 Sun et al. (1995) recommended that a frequency band containing major
vibrational modes of the structure (i.e. large number of peaks in the signature), such
as the one shown in Fig. 5, serves as a suitable frequency range. Large number of
peaks signifies greater dynamic interaction between the structure and the PZT patch.
Park et al. (2003b) recommended a frequency range from 30 kHz to 400 kHz for PZT
patches 5 to 15mm in size. According to Park and coworkers, a higher frequency
range (>200 kHz) is favourable in localizing the sensing range, while a lower
frequency range (< 70 kHz) covers a large sensing area. Further, frequency ranges
higher than 500Hz are found unfavourable, because the sensing region of the PZT
patch becomes too small and the PZT signature shows adverse sensitivity to its own
bonding condition rather than any damage to the monitored structure. It should also be
noted that the piezo-impedance transducers do not behave well at frequencies less
than 5kHz. Below 1kHz, the EMI technique is not at all recommended (Giurgiutiu
and Zagrai, 2002).

Sensing Zone of Piezo-Impedance Transducers

The PZT patches have a localized sensing zone of influence. This is because a
PZT patch vibrating at high frequencies excites ultrasonic modes of vibration the
structure, which are essentially local in nature. Besides, damping is much more
significant at high ultrasonic frequencies, leading to localization of the waves
generated by the vibrating PZT patch. Esteban (1996) carried out extensive numerical
modelling based on wave propagation theory, as well as conducted comprehensive
parameteric studies to identify the sensing zone of the piezo-impedance transducers.
However, at such high frequencies, exact quantification of energy dissipation proved
very difficult and hence the sensing zone could not be exactly identified. However, it
was found that this zone depends on the material of the host structure, its geometry,
the frequency of excitation and the presence of structural discontinuities. It was
concluded that structural discontinuities acting as the sources of multiple reflections
cause maximum attenuation to the propagating waves.

However, based on experimental data from a large number of case studies,

Park et al. (2000a) claimed that the sensing radius of a typical PZT patch might vary
anywhere from 0.4m on composite reinforced structures to about 2m on simple metal
beams. Tseng and Naidu (2001) reported the sensing range to be greater than 1m in
their experiments on thin aluminum beams. Therefore, for effective damage
localization, in general, the structures must be instrumented with an array of PZT
patches.

Due to a localized sensing region, the technique shares a rare ability to detect
damages without being affected by far field boundary conditions, external loading or
normal operating conditions (Esteban, 1996). However, this advantage comes at the
cost of a limited sensing area.

Effects of Temperature
The conductance signatures of piezo-impedance transducers have been found
to be temperature sensitive (Sun et al, 1995; Park et al., 1999). In real situations, the
effects of damage and temperature are bound to be mixed. This necessitates a method
to decouple the two. Fortunately, over a small frequency band, the overall effect of
temperature has been observed to be a superposition of uniform horizontal and
vertical translations of the signature (Sun et al., 1995). This is absolutely different
from the signature deviation resulting from any damages, which causes an abrupt and
local variation. It was observed by Pardo De Vera and Guemes (1997) that the
horizontal shift is not uniform and depends on frequency. However, if the frequency
band is rather narrow, it can be assumed to be uniform.

Park et al. (1999) proposed statistical cross-correlation based methodologies

for temperature compensation. Bhalla (2001) studied temperature effects using finite
element simulation. It was found that the major effects of temperature on the
signatures are the horizontal shift, due to change in the host material’s Young’s
modulus, and the vertical shift, due to variations in 33 and d31 of the PZT patch. All
the shifts were found to vary linearly with temperature over narrow frequency bands.
Out of these, the most critical was the vertical shift due to change in 33. A simple
temperature compensation methodology was proposed which required the acquisition
of the baseline signatures at two different temperatures.
Effects of Noise and Other Miscellaneous Factors
Most low frequency vibration based SHM/ NDE methods on real-life
structures are likely to encounter the presence of noise. The noise could be (a)
mechanical noise, caused by sources such as vehicle movement or wind; (b)
electrical noise, generated by variations in the power supply; or (c) electromagnetic
noise, caused by communication waves, which affect the signal acquisition and
transmission through cables and other susceptible circuitry (Samman and Biswas,
1994a).

The greatest advantage of the high frequency EMI technique is that the signal
(in few hundred kHz frequency range) is not likely to be affected by mechanical
noise, since this type of noise is dominant in the low frequency ranges only (typically
less than 100Hz). Electrical noise too is not crucial in the EMI technique since the
power required by each PZT patch is in the low milliwatt range, which does not call
for the deployment of high power generating sets. Rather, it makes possible the
development of battery operated sensors (Park, 2000). The only possible noise could
be the electromagnetic noise, which can be minimized by using coaxial cables.

Another source of error could be the parasitic electrical admittance of the

connection wires. It can be accounted for by performing zero-correction in the
impedance analyzer, prior to taking measurements. However, it could be problematic
for large arrays where each PZT patch may have a different wire length. It is
recommended that the same set of connection wire be used for recording both the
baseline signature as well as the signature at any future point of time, so that the
residual conductance (if not properly accounted for in the zero correction) is the same
in both cases. The change in signature, if any, will be due to structural damage alone.
It should also be noted that extensive experimental study by Raju (1998) found that
the method can still work well in-spite of variable test wire lengths.

Park et al. (2000a) demonstrated that the technique is insensitive to distant

boundary condition changes and mass loading. The technique is also insensitive to
arbitrary ambient inputs to the structure. This is very important, especially for the in-
flight monitoring of aircraft or bridges, while under service.
 However, it should be noted that care must be exercised in applying the EMI
technique on structures which are instrumented with ultrasonic transducers for
purposes of NDE. The high frequency excitations from these transducers could
generate a high frequency noise for the EMI technique. Hence, it should be made sure
that these are turned off before applying the EMI technique.

Multiple Sensor Requirements

Since the EMI technique is essentially acousto-ultrasonic in nature, the
number of sensors necessary depends upon the geometry and material of the
monitored component. The number of sensors is small in thin beams and plates where
the acoustic waves can easily travel long distances through the material medium.
However, in complex structures with holes, notches, discontinuities and thickness
variations, a large number of sensors may be required due to greater losses on account
of energy dissipation. Also, the same would be true for materials such as composites
or concrete, which are characterized by high material damping. In such scenarios, it is
important that such a multi-sensor architecture to have a built-in redundancy such that
one or more sensors may be allowed to fail without making the entire system
ineffective (Boller, 2002). Also, it is important to consider issues like sensor
validation, data pre-processing, feature extraction and pattern recognition.

Suitable locations for bonding the patches can be easily determined from the
geometry and loading conditions to which the structure is likely to be subjected
during the course of its service by preliminary structural analysis (Soh et al, 2003). It
is recommended to locate the patches at the points of maximum bending moments and
shear, which can be ascertained by the theory of structures.

It may be mentioned here that given an array of PZT patches, it can either be
excited in self-impedance fashion (The EMI technique) or transfer impedance fashion
(Esteban, 1996). In the transfer function method, one PZT patch acts as actuator and
emits acoustic signal into the structure. The signals are picked by another patch acting
as sensor. The main advantage of the transfer impedance method (or the gain-
phase) method is that it provides greater sensing range and hence reduces the number
of sensors required. Besides, this can also enable the determination of mechanical
properties of the monitored component. Impedance analyzer can be easily utilized for
the transfer impedance approach also. However, the ‘gain’ levels encountered in the
transfer impedance approach are much smaller since the waves have to travel longer
distance, besides encountering higher noise (Park et al., 2003b). Increasing the
excitation level could help overcome this problem and this could help the two
methods to supplement each other, since the same sensor array can be utilized for
both the techniques.

Signal Processing Techniques and Damage Quantification

The prominent effects of structural damages on the conductance signature are
the appearance of new peaks in the signature and lateral and vertical shifting of the
peaks (Sun et al., 1995), which are the main damage indicators. Samman and Biswas
(1994a, 1994b) reported many pattern recognition techniques to quantify the
variations occurring in the structural signatures (similar to conductance signatures)
due to damages; such as the waveform chain code (WCC) technique, the signature
assurance criteria (SAC), the equivalent level of degradation system (ELODS) and the
adaptive template matching (ATM). Similar statistical techniques have been
employed by the investigators researching on the EMI technique; such as the root
mean square deviation or RMSD (Giurgiutiu and Rogers, 1998), relative deviation or
RD (Ayres et al., 1998; Sun et al., 1995), the difference of transfer function between
damaged and undamaged conditions (Pardo de Vera and Guemes, 1997) and the mean
absolute percent deviation or MAPD (Tseng and Naidu, 2001).

The RMSD index is defined as (Giurgiutiu and Rogers, 1998; Giurgiutiu et al.,
1999)
N
 (Gi  Gi )
1 0 2

RMSD (%)  i 1
N
x 100 (20)
0 2
 (Gi )
i 1

where Gi1 is the post-damage conductance at the ith measurement point and

G i0 is the corresponding pre-damage value. Similarly, RD is based on the sum of

mean square algorithm, normalized with respect to an arbitrarily chosen maximum
amount of damage, and is defined for the ith patch (in an array) as (Sun et al., 1995)
 N
 (Gik Gik )
1 0 2

RDi  k 1
N
(21)
 (G1k G1k )
1 0 2

k 1

where the numerator represents the mean square deviation at the ith location
and the denominator represents the deviation for the chosen reference maximum
damage location ‘1’. The MAPD index is defined as (Tseng and Naidu, 2001)

100 N Gi1  Gi0

MAPD   (22)
N i 1 Gi0
The covariance (Cov) and correlation coefficient (CC) are respectively defined
as (Tseng and Naidu, 2001)
1 N
Cov(G o , G1 )  
N i 1
(Gi0  G 0 )(Gi1  G1 ) (23)

Cov(G 0 , G1 )
CC  (24)
 0 1
where 0 and 1 are the standard deviations of the baseline signature and the

signature after damage respectively. G 0 and G1 respectively are the mean values of
the baseline signature and the signature after damage.

Following observations by different investigators regarding statistical indices

are worth being taken note of.
(1) The author performed a comparative study of the RMSD, the SAC, the
WCC and the ATM techniques (Bhalla, 2001) and found the RMSD
algorithm as the most robust and most representative of damage
progression among these indices.
(2) Tseng and Naidu (2001) demonstrated the use of MAPD, covariance
(Cov) and correlation coefficient (CC) to quantify damages in thin
aluminium beams. They found Cov and CC to be very good indicators
when quantifying increase in damage size at one particular location.
When the peaks of one signature match with the peaks of the other
signature, the covariance value obtained is positive. When valleys of
one signature match with peaks of the other, and vice versa, covariance
is negative. When values in both signatures are unrelated, covariance is
nearly zero. Thus, the damages can be characterized by the fact that
 when the deviation between the signatures is large, the covariance is
closer to zero or is negative.
(3) Giurgiutiu et al. (2002) reported comprehensive investigations of CC
as damage index in their experiments on thin circular aluminium
plates. It was experimentally found by these researchers that (1-CC)3
decreased linearly as the distance between the sensor and the damage
(a simulated crack) increased.

Although the statistical methods are easy to implement and share the
advantage of being non-parametric (Soh et al., 2000), their main drawback is that they
do not provide any clear picture of the associated damage mechanism or any change
in mechanical parameters of the structure under question. For example, in many
situations, incipient damage and the high order damage can be found to lead to an
RMSD index of the same order of magnitude. As such, the particular “threshold
value” demanding an alarm could vary from structure to structure (Soh et al., 2000).
In such situations, one needs to rely on the slope of the RMSD curve rather than its
absolute magnitude. However, this may also prove unreliable. It is probably for this
reason that Giurgiutiu et al. (2002) have remarked “…Further work is needed to
systematically investigate the most appropriate damage metric that can be used for
processing the frequency spectra successfully…”.

ADVANTAGES OF EMI TECHNIQUE

The major advantages of the EMI techniques over the prevalent global and
local SHM techniques are summarized below
(i) The EMI technique shows far greater damage sensitivity than the
conventional global methods. Typically, the sensitivity is of the order
of the local ultrasonic techniques (Park et el., 2003b). Yet the
technique is very straightforward to implement on large structures as
compared to the local methods, whose application is quite
cumbersome. It does not warrant very expensive hardware like the
ultrasonic techniques and also does not warrant any probe to be
physically moved from one location to other. The data acquisition is
much more simplified as compared to the traditional accelerometer-
shaker combination in the global vibration techniques since the data is
 directly obtained in the frequency domain. Thus, the EMI technique
provides a very nice interface between global vibration based
techniques and local ultrasonic techniques.
(ii) The PZT patches are bonded non-intrusively on the structure, possess
negligible weight and demand low power consumption. Small and non-
intrusive sensors can monitor inaccessible locations of structures and
components. Hence, this could save the expensive time and effort
involved in dismantling machines and structural components for
inspection purposes. Easy installation (no sub-surface installation)
makes the piezo-impedance transducers equally suitable for existing as
well as to-be-built structures.
(iii) The use of the same transducer for actuating as well as sensing saves
the number of transducers and the associated wiring.
(iv) The limited sensing area of the PZT patches helps in isolating changes
due to far field variations such as boundary conditions and normal
operational vibrations. Also, multiple damages in different areas can be
picked easily.
(v) The technique is practically immune to mechanical, electrical and
electro-magnetic noise. This makes the technique very suitable for
implementation during operating conditions, such as in aircraft during
flight.
(vi) The PZT patches can be produced at very low costs, typically US$1
(Peairs et al., 2003) to US$10 (Giurgiutiu and Zgrai, 2002), in contrast
to conventional force balance accelerometers, which may be as
expensive as US$1000 (Lynch et al., 2003b) and at the same time
bulky and narrow-banded.
(vii) The technique is very favourable for autonomous and online
implementation since the requirements for data processing are
minimal. The data is directly recorded in the frequency domain thereby
saving expensive domain transform efforts.
(viii) The method can be implemented at any time in the life of a structure.
For example, the PZT patches can be installed on structures after an
earthquake to monitor the growing cracks or loosening connections.
Many other methods warrant installation of the sensors at the time of
 construction and hence not suitable for existing structures. However, it
should be noted that the PZT patches would be able to detect any
structural damages appearing in the post-installation period only.
Hence, they cannot detect “existing” damages in the structures.
(ix) Being non-model based, the technique can be easily applied to
complex structures.
(x) The PZT patches are orders of magnitude below the stiffness and mass
of the monitored structures. Hence the dynamics of the host structure
are not modified and accurate structural identification is possible.
(xi) PZT sensors are non-resonant devices with wide bad capabilities and
exhibit large range of linearity, fast response, light weight, high
conversion efficiency and long-term stability.
(xii) Commercial availability of portable and low-cost impedance analyzers
will further enhance the applicability of the technique.

It is therefore needless to say that the EMI technique has evolved as a

universal NDE method, applicable to almost all engineering materials and structures.
If the damage location could be predicted in advance (i.e. ‘where to expect damage’),
the EMI technique would be most powerful technique in such applications (Park et
al., 2003b).

LIMITATIONS OF EMI TECHNIQUE

In spite of many advantages over other techniques, the EMI technique shares
several limitations as outlined below

(i) A PZT patch is sensitive to structural damages over a relatively small

sensing zone, ranging from 0.4m to 2m only, depending upon the
material and geometrical configuration. Though sufficient for
monitoring miniature components and mechanical/ aerospace systems,
the small sensing zone warrants the deployment of several thousands
of PZT patches for real-time monitoring of large civil-structures, such
as bridges or high rise buildings. The large number of PZT patches
would warrant significant cost and effort for laying out the wiring
 system, data collection and data processing. Hence, critical locations
must be judiciously decided based on the theory of structures.
(ii) Since all civil and mechanical structures are statically indeterminate,
cracking of a few joints might not necessarily affect the overall safety
and stability of the monitored structure. Thus, a drawback of the EMI
technique as compared to the global SHM techniques is its inability to
assess the overall structural stability. Rather, in this respect, global
SHM techniques and the EMI techniques could easily complement
each other.
(iii) PZT materials and the related technologies are only supplementary
steps in addition to good designs of structures and machines. Many
academicians argue that more research should be focused on
improving material strength and design rather than on sensors. But
even the best-designed structures could have problems, therefore it is
justified to explore the application smart materials to sense or detect
damages in advance (Reddy, 2001).

OTHER SMART MATERIALS

Optical Fibres
Optical fibres are very thin fibres made of glass and silica, which utilize fibre
properties to generate optoelectronic signals indicative of the external parameters to
be measured. An optical transmission system, employing either a laser diode (LD) or
a light emitting diode (LED) creates an appropriate light signal, which is then
conducted into the fibre. This signal is received by an optical receiver system
containing a photodiode (at the other end of the fibre), which converts it back into
electrical signal. The optical fibre may be placed in a region subjected to an external
stimulus (such as strain, pressure, temperature etc.), which can modulate the
frequency, intensity, wavelength, amplitude, phase, colour, modal distribution or
polarization characteristics of the light beam. The receiver system analyzes the
modulated light to yield information regarding the external stimulus.

Of the various types of fibre based technologies, the fibre Bragg grating
(FBG) has proved to be the most versatile (Tjin et al., 2002). A Bragg grating is a
periodic structure, fabricated by exposing a photosensitised fibre to an ultraviolet
light. Fig. 7 illustrates the working principle of the FBG based strain gauge. When
light from a broad band source interacts with the grating, a single wavelength, known
as the Bragg wavelength, is reflected back whereas the rest of the signal is
transmitted. The Bragg wavelength, λB, is related to the grating pitch Λ and the
effective refractive index neff of the grating by
 B  2 n eff (25)

An external mechanical strain in the fibre (a stimulus) shifts the Bragg

wavelength (response of the fibre) through expansion/ contraction of the grating
periodicity and the photo elastic effect. Similarly, a temperature variation (stimulus)
causes thermal expansion/ contraction of the grating periodicity and also changes the
refractive index (responses). These effects provide the means of employing the FBG
written fibres as the sensor elements for measuring strains and temperatures. The
change in Bragg wavelength due to an external disturbance, comprising a strain
change, a temperature change T and a pressure change P, can be quantitatively
expressed as
 B
 K    K T  T  K P P (26)
B
where Kε, KP and KT and are the wavelength sensitivity coefficients for strain
(ε), temperature (T) and pressure (P) respectively.

The optical fibres were originally developed for telecommunication purposes.

However, they have found tremendous recognition as smart sensors in the recent
years, especially since late 1980s onwards. Many novel sensors have been developed
using optical fibres, such as velocity sensors (Ng et al., 1998), pressure sensors (Tjin
et al., 2000), strain gauges and displacement sensors, which find numerous
applications in global SHM techniques (e.g. Storoy et al., 1997; Yamakawa et al.,
1999; Brownjohn et al., 2003).
 UV light

Phase mask

Reflected signal
Optical fibre
Input signal

Λ Grating pitch

Transmitted signal
λ

Fig. 7 Principle of FBG sensors.

Key advantages of the optical fibres are that they are light, non-conductive,
rugged and can be easily embedded into composite structures due to their shape
adaptability. They do not require electrical isolation from the structural material and
are immune from electromagnetic interference. The requirements of wiring are
minimal since multiple sensors can be multiplexed using a single fibre.

Fig. 8 depicts a smart structure featuring an embedded fibre optic sensing

system (Gandhi and Thompson, 1992). In this system, the light source generates an
optical signal, which is introduced into the optical network and is subsequently
modified as it passes through the structure. Any crack in the structure will sever the
transmission path of light beams, from which the location and the extent of the
structural damage can be determined. Similar systems can be used in the development
of smart skins for aircraft and for monitoring the structural integrity and failure
mechanisms in composites.

Although very versatile, the fibre optic sensors share few limitations as well.
Being delicate, they are very prone to damage at the time of installation. For example
in the reinforced concrete (RC) structures, where they are bonded to the rebars, they
are highly susceptible to damage during the pouring of concrete, due to the use of
vibrators or during actual operation (Storoy et al., 1997). Besides, the fibre coating is
not completely immune from alkaline environment inside concrete (Kawiecki, 2001).
Usually, the fibre optic sensors are multiplexed, i.e. a single fibre is used to receive
signals from all sensors in series. A single snap can render all the sensors
dysfunctional and if the fibre has been embedded inside the material, there is no
possibility of repair. Also, the optical fibres are difficult to be used on steel structures
since they may not survive the high temperatures during welding (Kawiecki, 2001).
The fibre optic sensors are not yet mature in terms of standardized field proven
devices (Storoy et al., 1997).

Input fibre Structure being

monitored
Light Source
Embedded optical
fibre grid

Output fibre

Output data processing

and analysis Signal Converter

Fig. 8 A smart structure featuring embedded fibre-optic sensing system.

Shape Memory Alloys (SMAs)

The SMAs were first developed in the 1960s (applied for patent in 1965),
although their practical utilization has begun only in the last decade. These alloys,
when plastically deformed (under low temperatures), show the property of regaining
back their original ‘memorized’ shape when heated above a certain temperature,
called the characteristic transition temperature. The most commercially used SMA is
NITINOL (an alloy of Nickel and Titanium developed by Naval Ordnance
Laboratory). By simply heating the embedded SMA fibres, the modal response of the
structure can be easily altered since heating process increases the material’s Young’s
modulus by 4 times and yield strength by 10 times (Rogers, 1990). Restraining the
material from regaining the memorized shape can create stresses of the order of
100,000 psi (against yield strength of 12, 000 psi under room temperature).
 Some common applications of the SMAs are active vibration control (Reddy
and Barbosa, 2000), active buckling control, shape/ motion control and artificial
muscle. Recently, SMA composites have been fabricated using SMA fibres such that
the material can be easily stiffened or controlled by passing electric current, which
heats the SMA above the transition temperature. Till date, only SMAs have been
widely used for actuation besides the piezo-electrics. A very novel integration of the
two smart materials has recently been achieved by Peairs et al. (2002) in the form of a
self-healing bolted joint.

Electro-Rheological (ER) and Other Smart Fluids

The ER fluids were first developed in 1947. These fluids possess the ability to
alter their flow characteristics on the application of an electric field with a reaction
time of a few milli seconds. The ER fluids are composed of 1 to 100 m particles
dispersed in a carrier fluid, which must be a good insulator. For ER effect, the electric
field required is of the order of 4 kV/mm, which is non-lethal. The application of the
electric field forces the particles move towards the opposite electrode thereby forming
chains and changing the viscosity of the fluid. On the removal of the electric field, the
ER fluids flow like water. In the earlier days, the ER fluids were based on mineral oil
and starch/ silica particles adsorbed in water, which had the tendency of settling
down. Now a days, the particles are based on polymers, minerals and ceramics, which
are much more stable.

Vibration control using ER fluids has been demonstrated using hollow

graphite epoxy cantilever beams filled with ER fluids (Gandhi and Thompson, 1992).
By applying voltage across the beam, internal damping can be increased and
vibrations can be suppressed. Viscosity characteristics can also be exploited to create
a structure with variable stiffness, which can be effectively used to shift system’s
natural frequency away from the excitation frequency.

Some other smart fluids which have recently been developed are magneto-
rheological fluids (which change their flow characteristics on the application of
magnetic fields) and magneto-resistive fluids (which change their electrical resistivity
in the presence of magnetic fields) A magneto-resistive fluid named MARSONPOL
(John et al., 2002). has recently been developed and applied for patent by Defence
Research and Development Organisation (DRDO). The fluid is insulator (resistivity
108 ohm-m) under normal conditions. However, on the application of a magnetic field
(about 0.5T), the resistivity drops down to less than 1 ohm-m. Future applications of
this smart fluid could in magnetic field sensing probes, explosion proof electrical
switches and microwave devices.

CONCLUDING REMARKS
This technical article has presented a detailed review of the state of the art in
SHM, with a special emphasis on the EMI technique. The chapter also introduced the
concept of smart materials and structures. The physical principles underlying the EMI
technique and the details of the previous work undertaken by prominent research
groups of the world (Liang, Rogers and coworkers; Inman, Park and co-workers;
Giurgiutiu and coworkers) have been presented. The needs for further research to
improve this technique were also highlighted. This article is followed by two journal
publications highlighting the application of the EMI technique on RC bridges and RC
frames.

Smart materials and the related technologies are only supplementary steps in
addition to good designs of structures and machines. Many academicians argue that
more research should be focused on improving material strength and design rather
than on sensors. But even the best designed structures could have problems, therefore
it is justified to explore the application smart materials to sense or detect damages in
advance (Reddy, 2001).