A Review of Vibration-Based Structural Health Monitoring

The Shock and Vibration Digest
http://svd.sagepub.com
A Review of Vibration-based Structural Health Monitoring with Special Emphasis on Composite

Materials
D. Montalvão, N. M.M. Maia and A. M.R. Ribeiro
The Shock and Vibration Digest 2006; 38; 295
DOI: 10.1177/0583102406065898
The online version of this article can be found at:

http://svd.sagepub.com/cgi/content/abstract/38/4/295
Published by:
http://www.sagepublications.com
Additional services and information for The Shock and Vibration Digest can be found at:
Email Alerts: http://svd.sagepub.com/cgi/alerts
Subscriptions: http://svd.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations (this article cites 99 articles hosted on the

SAGE Journals Online and HighWire Press platforms):
http://svd.sagepub.com/cgi/content/refs/38/4/295
Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

© 2006 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
Articles
A Review of Vibration-based Structural Health

Monitoring with Special Emphasis on Composite
Materials
D. Montalvão, N.M.M. Maia and A.M.R. Ribeiro
ABSTRACT—Structural health monitoring and damage this paper includes many citations common to Doebling et al.
detection techniques are tools of great importance in the off- (1996**) or Sohn et al. (2003); these are herein identified
shore, civil, mechanical and aeronautical engineering com-
munities, both for safety reasons and because of the economic with, respectively, * and ** accompanying the year of publi-
benefits that can result. The need to be able to detect dam- cation.
age in complex structures has led to the development of a vast Other reviews may bring insight to more specific branches
range of techniques, of which many are based upon structural of damage detection and SHM. For example, a state-of-the-art
vibration analysis. In the present article, some of the latest overview on monitoring of rotating machinery is given by
advances in Structural Health Monitoring and Damage Detec-
tion are reviewed, with an emphasis on composite structures Randall (2002). More recently, Carden and Fanning (2004)
on the grounds that this class of materials currently has a presented a review covering structural engineering related
wide range of engineering applications. aspects. Uhl and Mendrok (2004) discuss the applicability of
modal model-based methods in structural diagnosis. Farrar
FOREWORD—It should be noted that this review is not et al. (2004) present a brief yet comprehensive overview of
intended to be a general, all-encompassing review covering
the whole range of structural health monitoring (SHM); it was nonlinear system identification techniques used for damage-
planned as the starting point for a study focusing on damage sensitive feature extraction from measured data.
detection, localization and assessment for certain kinds of In the present paper, special attention will be given to meth-
structure. Thus, the line of thought behind the search and the ods in which composite materials and structures are used as
structure of this review is a result of objectives beyond the case studies. Nevertheless, given the huge range of methods
scope of the paper itself. Nevertheless, it was considered
that, once the above was understood, an updated synopsis described in the literature, and the restrictions mentioned in
such as this could also be useful for other researchers in the the foreword, many damage detection and SHM techniques
same field. will not be explored in detail in this review, even where they
are touched on briefly. These include such examples as rotat-
KEYWORDS: damage detection, structural health monitor-
ing, composite materials ing machinery, online SHM and monitoring systems, updating,
sensitivity-based methods (which are widespread in model
updating), non-linear systems, wireless sensing, civil infra-
structures, joints and couplings, conditioning, chaotic time
series excitation. A positive decision has been made, however,
1. Introduction to include methods based on natural frequencies, frequency
response functions (FRFs), mode shapes, mode shape and FRF
Vibration-based SHM and damage detection is a field of curvatures, auto-regressive moving average (ARMA) family
study of great interest to many people. Important advances models, dynamic flexibility, modal strain energy, transmissi-
in this field have been described by Doebling et al. (1996**, bility, damping, impedance, Lamb waves, time-frequency anal-
1998**) and Sohn et al. (2003), which present very compre- ysis, Hilbert transforms, principal component analysis (PCA)
hensive surveys covering the period until early 2002. Many and singular value decomposition (SVD), neural networks,
of the works cited in these surveys still represent the state of instrumentation and others.
the art, motivating their inclusion in this work. Therefore, A general definition of damage is given by Sohn et al.
(2003) as “…changes introduced into a system that adversely
D. Montalvão Department of Mechanical Engineering, Escola Superior de
affect its current or future performance. Implicit in this defi-
Tecnologia, Polytechnic Institute of Setúbal, Campus do IPS, Estefanilha, nition is the concept that damage is not meaningful without a
2910-761 Setúbal, Portugal. N.M.M. Maia and A.M.R. Ribeiro Department comparison between two different states of the system, one of
of Mechanical Engineering, Instituto Superior Técnico, Technical University which is assumed to represent the initial, and often undam-
of Lisbon, Av. Rovisco Pais, 1049-001 Lisboa, Portugal. (E-mail: nmaia@
dem.ist.utl.pt)
aged, state.” For example, one may have a structural crack
(stiffness change), bridge pillar silting (boundary condition
The Shock and Vibration Digest, Vol. 38, No. 4, July 2006 295–324 change), counterweight balancing loss (mass change) or loose-
©2006 SAGE Publications ness in a bolted joint (connectivity change) (Maia (2001)).
DOI: 10.1177/0583102406065898

296 The Shock and Vibration Digest / July 2006
The need to be able to detect damage in complex structures Cawley (1997**) used an analytical model of a cantilevered
has motivated the development of techniques based upon beam to compare the effect of crack formation to the effect
variations in the structure’s dynamic behavior. For instance, of beam length in the natural frequencies, observing that ther-
the method used to monitor the condition of train wheels is mal expansion (simulated by changes in the beam length) has
much the same today as it was 100 years ago: a ‘modal spe- more significant effects on the natural frequencies than cracks.
cialist’ walks along the train cars hitting the wheels with a Kullaa (2002) tried to eliminate the environmental influences
hammer and listening to their response (Maia (2001)). In con- in modal parameters, using an approach based on factor anal-
ditioning maintenance, vibration is without any doubt more ysis, in which it is assumed that there exist a set of common
straightforward as an indicator of the machine state than most factors with effects on the modal features that can be sub-
other indicators, such as temperature, pressure, flow or tribo- tracted from the observations.
logy features (Wowk (1991)). In another field of application,
Vescovo and Fregolent (2005) use a non-invasive acoustic 1.1. Damage Detection Philosophy
experimental technique to assess the partial detachment of
plaster portions in artistic frescoes. It is even argued by Bon- Rytter (1993*,**) introduced a damage state classification
figlioli et al. (2005) that monitoring and assessing the health system which has been widely accepted by the community
state of infrastructures is one of the largest industries in the dealing with damage detection and SHM. Following these
world. lines, the damage state is described by answering the follow-
The main idea behind damage detection techniques based ing questions (Sohn et al. (2003)):
on structural dynamic changes is the fact that the modal param-
eters (natural frequencies, mode shapes and modal damping) (1) Is there damage in the system? (existence)
are functions of the physical parameters (mass, stiffness and (2) Where is the damage in the structure? (location)
damping) and thus it is reasonable to assume that the exist- (3) What kind of damage is present? (type)
ence of damage leads to changes in the modal properties of (4) How severe is the damage? (extent)
the structure. (5) How much useful life remains? (prognosis)
According to Doebling et al. (1996**), an ideal robust
damage detection scheme should be able to identify damage Generally, identification of the damage type and extent
at a very early stage, locate the damage within the sensor require prior knowledge of the structural behavior in the pres-
resolution being used, provide some estimate of the extent ence of each of the possible expected failure modes for
or severity of the damage and predict the remaining useful future correlation with experimental data, which is normally
life of the structural component in which damage has been achieved by resorting to analytical models. For example, in
identified. The method should also be well suited to auto- operational monitoring, the modal parameters of the dam-
mation, and should be independent of human judgment and aged structure must be compared to the parameters of the
ability. structure in its undamaged state, in what is called global diag-
Betti (2005) points out that no single approach is appro- nostics. Once damage existence is detected, the use of a model
priate for all situations, making the following basic distinc- of the structure in a damaged state may be used to determine
tions between approaches: linear vs non-linear, output only the damage location, in what is called local diagnosis (Uhl
vs input/output, on-line vs off-line, time domain vs frequency and Mendrok (2004)).
domain, parametric vs non-parametric and time varying vs Some difficulties may be encountered in the practical
time-invariant. application of modal models, such as the knowledge of exci-
Operational and environmental conditions, such as tem- tations and loads during machine operation, with several
perature, humidity, loads and boundary conditions, should sources and with unknown distribution along the system.
also be addressed, since in many cases the changes they cause However, several output-only modal identification techniques
can ‘hide’ those resulting from damage. For example, Farrar can be found in the literature; for example Guillaume et al.
et al. (1994*,**), having done several measurements on the (1999), Parloo et al. (2002a), Brincker and Andersen (2003),
I-40 bridge over the Rio Grande in Albuquerque in the state Brincker et al. (2003), Mevel and Goursat (2004), Rodrigues
of New Mexico in the USA, observed that temperature plays et al. (2004) and Galvin and Dominguez (2005).
a major role in the dynamic properties of the bridge. They Prognosis, which is traditionally related to fracture mechan-
introduced four different levels of damage by gradually cutting ics and fatigue, is starting to be brought up by the modal anal-
one of the bridge girders, corresponding to a loss of stiffness. ysis community as a field of interest, as seen in Farrar et al.
However, instead of the expected result, they observed an (2003**).
increase in the fundamental frequency for the first two dam-
age cases, and concluded that such results were mainly due 1.2. The Monitoring Process
to temperature changes and temperature gradients in the
bridge. Many other researchers, such as Woon and Mitchell The monitoring process involves the observation of a sys-
(1996a,b), Andersen et al. (1997), Alampalli (2000), Peeters tem over long periods of time, using samples of experimen-
and De Roeck (2000), Kullaa (2002), De Roeck and Degrauwe tal data acquired periodically with adequate sensors. It also
(2005), Steenackers and Guillaume (2005) and Yan et al. involves sensitivity feature extraction and statistical correla-
(2005a,b) address the effects of the environmental or opera- tions to determine the actual ‘health’ of the system. Farrar et al.
tional conditions in Structural Health Monitoring. Woon and (2001) and Sohn et al. (2003) acknowledge the problem of
Mitchell (1996a) studied the sensitivity of natural frequen- damage detection in the context of a statistical pattern rec-
cies to the relative humidity, but concluded that this had little ognition paradigm. Along these lines, one may describe this
significance when compared to the effects of temperature. paradigm as a four-step process:

Montalvão, Maia and Ribeiro / A REVIEW OF VIBRATION-BASED STRUCTURAL HEALTH MONITORING 297
(1) Operational evaluation. erties, which allow one to distinguish between the damaged
(2) Data acquisition, fusion, and cleansing. and undamaged structural states.
(3) Feature extraction and information condensation. Information condensation becomes increasingly advanta-
(4) Statistical model development for feature discrimination. geous and necessary as the quantity of data increases, par-
ticularly if comparisons are to be made between sets of data
Sections 1.2.1 to 1.2.4 are based mainly on the very com- obtained over the life cycle of a system. Condensation may
prehensive survey conducted by Sohn et al. (2003). be seen as a form of data fusion.
The evolution of the philosophy of damage detection and
1.2.1 Operational Evaluation SHM over the last ten years can be best understood by com-
Operational evaluation answers questions related to the paring the surveys of Doebling et al. (1996**) and Sohn et al.
damage detection system implementation, such as economic (2003). In the first one, the statistical pattern recognition
issues, possible failure modes, operational and environmen- paradigm, around which the second one was written, is not
tal conditions and data acquisition related limitations. For even mentioned; the concepts of operational evaluation, data
example, Aktan et al. (2000**) present and discuss the pre- acquisition, fusion and cleansing and statistical model devel-
requisite issues in creating a successful monitoring program. opment for features discrimination were matters whose rele-
Ruotolo and Surace (1997**) were among the first to vance was not deemed to be noteworthy of exclusive and
explicitly address the possibility of having multiple baseline directed research, and were addressed only as parts of the
configurations, proposing a technique to distinguish between feature extraction process.
changes caused by working conditions and resulting from
damage, based on the singular value decomposition (SVD). 1.2.4 Statistical Model Development for Feature
More recently, Vanlanduit et al. (2005) also used a SVD Discrimination
based system to detect damage in structures subjected to This is the part of the SHM process that had received the
different operational conditions, including different working least attention prior to the end of the last century. When data
excitation levels, geometrical uncertainties and surface treat- are available from both the undamaged and damaged struc-
ments. ture, the statistical pattern-recognition algorithms fall into the
general class referred to as supervised learning. Unsuper-
1.2.2 Data Acquisition, Fusion and Cleansing vised learning refers to the class of algorithms that are applied
Data acquisition is concerned with the quantities to be meas- to data not containing examples from the damaged structure
ured, the type and quantity of sensors to be used, the loca- (obtained, for instance, from models) (Sohn et al. (2003)).
tions where these sensors are to be placed, sensor resolution, An important issue in the development of statistical mod-
bandwidth, and hardware. This part of the process is applica- els is to establish the model features sensitivity to damage
tion specific and heavily dependent on economic factors. and to predict false damage identification. In this context, one
Another consideration is the periodicity for data acquisition. may have either false-positives, in which damage is identi-
For instance, if the growth of a fatigue crack is to be moni- fied though it did not happen, or false-negatives, in which
tored, then the data must be measured in an almost continu- damage is not identified though it is present. Despite the fact
ous way. On the other hand, if measurements are to be made that both these situations are undesirable, the second one can
under varying operational and environmental conditions, data obviously lead to more severe consequences.
normalization helps distinguish signal changes caused by
operational and environmental conditions from those caused 1.3. Damage in Composite Materials
by damage. Sohn et al. (2001a**) observed that the natural
frequency of a bridge over the Alamosa canyon in the state The use of fibre reinforced plastics (FRP) as an alternative
of New Mexico in the USA suffered 5% deviations over a 24 to conventional materials, such as metallic alloys, is under-
hour period due to temperature variations. going increasing growth, especially in the aeronautical, naval
Sohn et al. (2003) acknowledged that, according to Klein and automotive industries, because of their excellent mechan-
(1999**), data fusion first appeared as a result of defense ical properties, low density and easy of shaping. Neverthe-
organizations attempting to formalize procedures for inte- less, composite materials are very different from metals in
grating information from diverse sources, with the purpose their micromechanical interactions and their failure modes,
of determining battlefield situations and preventing threats. which (for composites) include matrix cracking, fibre break-
Data fusion, as a discipline of SHM, is the ability to integrate age, interlaminar delamination or voids (Sanders et al. (1992*),
data acquired from the various sensors in the measurement Jacob et al. (1997**) and Matthews (1999)).
chain. Composite materials possess strength/density ratios much
Data cleansing is the process of selecting significant data higher than those of the most widely used metallic materi-
from the multitude of information, i.e., the determination of als, such as steel, aluminum and titanium. However, the
which data is necessary (or useful) in the feature selection extreme sensitivity of composite materials to impact loads
process. constitutes a hindrance to their utilization. In aeronautical
structures, for example, the components may have to undergo
1.2.3 Feature Extraction and Information (i) low energy impacts caused by dropped tools or mishan-
Condensation dling during assembly and maintenance, (ii) medium energy
Within the paradigm defined by Farrar et al. (2001) and impacts caused in-service by foreign objects such as stones
Sohn et al. (2003), the field of damage detection which has or birds and (iii) high energy impacts caused by military
received the highest attention is feature extraction. Feature projectiles (Matthews (1999), Silva (2001) and Carvalho
extraction is the process of identifying damage sensitive prop- (2003)).

Typical failure modes in laminated composite materials are explored. Kisa (2004) affirms that the approach followed
caused by impacts may be found in Carvalho (2003). In a high can be used for the analysis of non-linear effects, such as the
energy impact, penetration is total and the damaged area is interface contact that occurs when the crack closes.
generally small. In a medium energy impact, it is usually Shu and Della (2004) studied the free vibration of beams
possible to visually detect the damage location, which is char- with multiple enveloping delaminations using an analytical
acterized by a local indentation. In a low energy impact (with model based on Bernoulli-Euler beam theory. The influence
enough energy to cause damage), only a very slight indenta- of the delamination size and its location on the first two nat-
tion will be seen on the impact surface. This level of damage ural frequencies and mode shapes is also discussed.
is often referred to as barely visible impact damage (BVID). Ostachowicz and Zak (2004) presented some results on
Generally, carbon fibre-reinforced plastics (CFRP) are very damped vibration of a laminated cantilever beam with a sin-
sensitive to medium and low energy impacts. Matrix cracks gle closing delamination. The vibration of the beam is studied
will appear and interact, leading to a delamination process. in the time domain using a dynamic contact algorithm devel-
Furthermore, on the opposite side of the component to the oped by those authors, based on the Newmark method and
impact (an area subjected to traction loads), it is possible that incorporating a procedure based on the Newton-Raphson
fibre breakage will occur. This matter has received some method for solving the equation of motion. They analyze the
attention for purposes of damage prediction; for example, vibration in the frequency domain. The vibration responses
Morais et al. (2005) present a study on the influence of lam- of the beam due to various harmonic and impulse excitations
inate thickness on the resistance to repeated low energy are also considered, with different delamination locations and
impacts of glass, carbon and aramid fabric reinforced com- for different delamination lengths, as well as changes in the
posites for two levels of impact energy. damping dissipation energy due to delamination.
Delamination significantly reduces the component stiff- Le Page et al. (2004) developed bi-dimensional (plane
ness and buckling load capacity, which in turn influence the strain) finite element (FE) models to analyze matrix cracking
structural stability. The influence on the resonant frequen- development in woven fabric laminates, as a function of the
cies of the contact forces between delaminated layers, extent number of reinforcing layers. It was shown that stiffness is
of delamination and delamination location were studied by relatively insensitive to these geometric variations, although
Zak et al. (1999**). the strain energy release rate related to the crack formation
Matthews (1999) presents an overview which addresses is significantly influenced by its location. Notably, where the
the nature, consequences and detection of defects in FRP crack formation is associated with local bending, the energy
materials. He distinguishes between two types of damage: release rate increases with deformation. The inclusion of
manufacturing defects and damage occurring during han- geometric non-linearities has been shown to affect the results
dling, assembly and service. The consequences of damage for models which exhibit higher bending deformation. Also,
and defects are also addressed, with respect to the composite FE models of cross-ply laminates were used both to serve as
performance when subjected to compressile, tensile, flexural a comparison and to evaluate whether the behavior of woven
and shear loadings. Finally, some damage detection techniques fabric laminates can or cannot be described in terms of
applied to composite materials are summarized, embracing cross-ply laminates.
both manufacturing and quality control processes. Tippetts and Hemez (2005) developed FE models to pre-
Although composite materials have received special atten- dict impact damage in composite plates. The failure modes
tion in vibration-based damage detection techniques, other addressed are ply splits (a special combination of matrix
non-destructive techniques are being developed, such as ESPI cracking and debonding with fibre breakage) and delamina-
(electronic speckle pattern interferometry) and the SQUID tion. The plate structural model uses a finite element formu-
(superconducting quantum interference device) (Hiroshi et al. lation with a cohesive zone model (CZM) to simulate the
(2005)). possible fracture surfaces characteristic of impact damage.
This model was validated with experimental tests.
1.4. Composite Materials: Damage Effects and Using a basis of linear fracture mechanics, the Castigliano
Modeling theorem and classical lamination theory, Wang et al. (2005)
investigated the coupled bending and torsional vibration of
Maheri and Adams (2003) used basic laminate plate the- a fiber-reinforced composite cantilever beam with an edge
ory in conjunction with a numerical Rayleigh-Ritz method surface crack. Some important conclusions were drawn: the
for estimation of the modal damping of anisotropic laminates natural frequency shifts, along with observations on the mode
in free vibration, concluding that a good correlation between shape changes, may be used to detect both the crack location
the theoretical and experimental results was achieved. Though and its depth for on-line SHM. Also, the model and results
not explicitly related to damage detection techniques, stud- presented may be useful for predicting flutter speed reduction
ies such as these can help researchers in the field of damage in aircraft with composite wings due to fatigue cracking.
detection, since damping may represent a useful feature for
damage evaluation in composite materials. 2. Some Damage Detection Techniques
Kisa (2004) investigates the effects of multiple cracks on
the dynamic properties of a cantilever CFRP beam. The the- 2.1. Natural Frequencies and FRFs
oretical model integrates fracture mechanics and substructure
coupling, where cracks are modeled as rotational springs. The development of modal analysis techniques for damage
The effects of the location and depth of the cracks, and the detection and SHM arose from the observation that changes
volume fraction and orientation of the fibres, on the natural in the structural properties have consequences on the natural
frequencies and mode shapes of a beam with transverse cracks frequencies. Nevertheless, the relatively low sensitivity of

natural frequency to damage requires high levels of damage modes is lower than the number of the damage parame-
and measurements made with high accuracy in order to ters.
achieve reliable results. Moreover, the capacity to locate Sanders et al. (1992*) use the frequency sensitivity method
damage is somewhat limited, as natural frequencies are glo- developed by Stubbs and Osegueda (1990a*), combined with
bal parameters and modes can only be associated with local an internal-state-variable theory, to detect, locate and quan-
responses at high frequencies. tify damage in CFRP beams. This method includes parameters
Methods based on natural frequency shifts often fall into which indicate two types of damage: matrix micro-cracking
one of two categories: the forward and the inverse problem. (identified by changes in the extensional stiffness), and
The forward problem consists in determining what the natu- transverse cracks in the 90-degree plies (by changes in the
ral frequency changes due to a known damage case (which bending stiffness). Despite the sensitivity equations used only
may include its location, extension and type) will be. Typi- being valid for viscous damping, they argued that the damp-
cally, damage is modelled numerically and the natural fre- ing had little influence in the application of the methodology.
quencies are measured experimentally and compared to those Due to the fact that damage was distributed uniformly along
related to each of the damage cases initially predicted. The the beam length, it was not possible to evaluate the ability of
inverse problem consists of determining damage parameters, the method to locate damage.
such as crack length or location, from changes in the natural Silva and Gomes (1990*) performed an extensive experi-
frequencies. mental dynamic analysis of free-free beams, with slots ranging
According to Doebling et al. (1996**), Lifshitz and Rotem in depths from 1/8 to 1/2 of the beam thickness, introduced
(1969*) present what may be the first journal article to pro- in both dimensions (x and y) of the cross section and in dif-
pose the use of vibration measurements for damage detec- ferent locations. Later, Silva and Gomes (1991) compared the
tion. They search for changes in the dynamic moduli, which experimental and theoretical methods, based on the premise
can be related to shifts in the natural frequencies, to detect that the development of a crack, at a certain location, corre-
damage in elastomers. sponds to a sudden reduction of the bending stiffness of the
Cawley and Adams (1979*) give a formula for damage beam at the same location. The crack was modeled as a tor-
detection, localization and quantification, based on the ratio sional spring representing the bending stiffness at the crack
between frequency shifts for modes i and j, ∆ω i ⁄ ∆ω j . location.
Location is determined by minimizing the error between the Following on from this, Silva and Gomes (1994*) sug-
measured frequency shifts of a pair of modes and those pre- gested a technique for damage detection in beams, based on
dicted by a local stiffness reduction model. In this formula- the estimation of frequency shifts as a function of the crack
tion, neither multiple damage sites nor damping changes are length and position. An optimization algorithm, developed
considered, though the authors agreed that damping might by Gomes and Silva (1992) and designated CRACAR, was
suffer an increase with damage. The results are based on FE used to search through combinations of crack lengths and
models of aluminum and CFRP plates. locations. To overcome the differences between the experi-
Tracy and Pardoen (1989*) presented the experimental mental and analytical models, they suggested a simple cor-
results of a study into the effects of the length of a midplane rection quotient, based on the natural frequencies, which
delamination in the frequency shifts of the first four modes they have shown to be a reliable tool.
of orthotropic graphite-epoxy laminate beams. The analyti- Hearn and Testa (1991*) developed a damage detection
cal model divides the beam into a total of four sections: method which evaluates the ratio of frequency shifts for var-
above, below and on the left and right sides of the delamina- ious vibration modes. Assuming that the mass does not change
tion. For midplane centered delamination, it was observed as a result of damage, and neglecting second-order terms in
that its presence degraded the even numbered vibration modes the formulation, they showed that the change ∆ω i in the ith
more rapidly than the odd numbered vibration modes. Unless natural frequency that results from damage can be related to
the delamination was in a region of the mode shape where the matrix of change in element N stiffness [ ∆k N ] and to the
the shear forces were high, the natural frequency would not element deformation vector { ε N ( Φ ) } evaluated from the
be significantly modified. Thus, in the presence of delami- mode shape vectors:
nation in areas where the mode shape exhibits high curva-
tures (second derivative of deflection), only minor frequency { ε N ( Φ i ) } T [ ∆k N ] { ε N ( Φ i ) }
shifts take place; natural frequency variations will be most ∆ω i2 = --------------------------------------------------------------
- (1)
obvious close to an area of higher shear force (related to the { Φi } T [ M ] { Φi }
third derivative of deflection).
th
Stubbs and Osegueda (1990a*,b*) developed damage where [ M ] is the mass matrix and { Φ i } is the i mass-nor-
detection methods based on modal changes of specific struc- malized mode shape vector. They noted that equation (1),
tural components such as beams, plates and shells. The which resembles Rayleigh’s quotient, seems to support the
method relates frequency shifts to changes in member stiff- idea that if an element must develop a high level of potential
ness using a sensitivity relationship. Stiffness reductions energy because of the deformations imposed by a vibration
can be located solving an inverse problem, since damage is mode, then damage in that member will have a strong influ-
defined as a stiffness reduction of one of the elements form- ence on the natural frequency of the mode shape. The converse
ing the structure. They concluded that it is possible to locate is also true, and because the natural mode shapes are distinct,
multiple damage sites, at least in a beam. False positives the contribution to potential energy made by any one mem-
occurred, however, although generally at a far smaller number ber will be different for each mode. In the case where dam-
of sites than those where damage was actually present. age is limited to a stiffness component of the element and if
Moreover, this method has difficulties when the number of one takes the ratio between the frequency shifts for modes i

and j, it is possible to show that the effects of the damage idea had already been proposed by Adams et al. (1978*) for
will be reduced simply to a function of damage location, axial vibrations, and was extended in their article to include
∆ω i2 ⁄ ( ∆ω j2 ), illustrating the characteristic influence of each flexural vibrations in one-dimensional beams, allowing bet-
member on the natural frequencies of the structure. The char- ter results to be obtained. Since this model is based on the
acteristic influence can be determined from pre-damaged Bernoulli-Euler beam theory, the authors pay special atten-
modal properties. Damage is located by selecting the mem- tion to the problem of accurately knowing the values of the
ber for which the characteristic influence is closest to the ratio material properties, in particular the Young’s modulus, which
∆ω i2 ⁄ ( ∆ω j2 ). can be overcome by calculating its ‘effective’ value as sug-
Penny et al. (1993*) applied a statistical method to identify gested by Adams et al. (1978*).
the most likely damage location, using a generalized least- Sampaio et al. (2003) and Sampaio and Maia (2004)
squares theory. The method uses the ratio of natural fre- propose the detection and relative damage quantification
quencies from both the measured and analytical data. The indicator (DRQ), based on the use of the frequency domain
damage case is indicated by the minimal error in this fit. assurance criterion (FDAC), as an effective damage indica-
Messina et al. (1992) proposed the damage location assur- tor, capable of distinguishing a positive occurrence from a
ance criterion (DLAC) in location j, which is a correlation false alarm. The column vector j of the receptance matrix
similar to the modal assurance criterion (MAC) of Allemang [ α j ( ω ) ] is the operational deflection shape (ODS) which
and Brown (1982), and is given by: describes the shape (in space) exhibited by the structure at
each excitation frequency, ω , given by the responses nor-
{ ( ∆ω ) X } { ( ∆ ω ) A }
T 2 malized by the applied forces. When a structure is damaged,
DLAC ( j ) = ------------------------------------------------------------------------------------------------
- (2) its stiffness and damping change and, in consequence, so
{(∆ω ) } {( ∆ω ) } {( ∆ω ) } T {( ∆ω ) } T
T
does the receptance matrix [ α ( ω ) ] . Thus, it is reasonable to
 X X
 A A
 assume that the smaller the degree of correlation between
d
where {( ∆ω ) X } is the experimental frequency shift vector and the column vectors (ODSs) of [ α ( ω ) ] and [ α ( ω ) ], where
{( ∆ω ) A } is the analytical frequency shift vector obtained the superscript d stands for damaged, the larger is the damage.
from an analytical model for a given damage location. A zero To measure the degree of correlation between the ODSs, Pas-
value indicates no correlation and a unity value indicates per- cual et al. (1997) proposed the FDAC:
fect correlation between the vectors involved in the DLAC
FDAC j ( ω 1, ω 2 )
relationship. Damage location and dimension is identified
2
by maximizing this objective function. n d
Messina et al. (1996) showed that it is possible to locate ∑ αij ( ω 2 )αij* ( ω 1 )

(3)
i=1
damage more accurately if the frequency shifts are normalized = ---------------------------------------------------------------------------------------------------------
n n
-
∑
d d *
relative to the undamaged structural frequencies, so that one
i=1
α ij ( ω 2 ) α ij ( ω 2 ) ∑ ij 1 ij 1α ( ω )α * (ω )
i=1
may weigh the contribution of each vibration mode, since
higher frequencies tend to suffer more pronounced shifts. where n is the total number of co-ordinates or measurement
The multiple damage location assurance criterion (MDLAC), points. A simplified form of the FDAC is referred to as the
proposed by Messina et al. (1998**), is an extension of the response vector assurance criterion (RVAC) (Heylen et al.
DLAC to detection of multiple damage sites. Using the same (1998)), with only one applied force (so that the receptance
principles, the damage state is indicated by search of a dam- matrix becomes a vector) and pairs of ODSs at the same fre-
age vector { ∆D } which maximizes the MDLAC. This for- quency ω :
mulation allows for relative quantification of the damage in
n d 2
each location, but not for absolute quantification. However,
since the experimental frequency shift vector {( ∆ω ) X } can
∑
αi ( ω )α *i ( ω )
i=1
RVAC ( ω ) = --------------------------------------------------------------------------------------------
n
- (4)
be calculated, it is possible to estimate a scalar severity coef- n
∑ i
d d *
ficient C, in such a way that C ⋅ { ∆D } gives the actual dam-
α (
i=1
ω ) α i ( ω ) ∑ i α ( ω )α *(ω)
i
i=1
age present. This scaling constant C can be estimated by either
a first or second order approach, but these authors note that, From this definition, Sampaio et al. (2003) and Sampaio
although the second order approximation provides better and Maia (2004) formulate the DRQ, which is nothing more
results, for routine monitoring purposes where precise knowl- than an arithmetic average of the RVACs along the frequency.
edge of the defect size is less important than its location, the They also proposed a normalization algorithm for the DRQ,
first order approximation is likely to be an adequate choice based on various damage cases, which is referred to as nor-
due to its relative simplicity. malization of the maximum occurrences, DRQi. Finally, and
Boltezar et al. (1998) devised a method for locating trans- because some of the best known damage localization meth-
verse cracks in flexural vibrations of free-free beams by fol- ods, such as the damage index and mode shape curvature
lowing an inverse problem. The method is based on the (MSC) (Sampaio et al. (1999) and Maia et al. (2003)), use the
assumption that the crack stiffness does not depend on the second spatial derivative of the ODSs to locate damage, one
frequency of vibration (i.e., the values of the crack stiffness, can instead calculate the DRQ″ based on these deriva-
which is modelled as a linear torsional spring, must be the tives and using, for the ODSs, a central difference approxi-
same at the crack position for all of the measured natural fre- mation. It was found that the DRQ indicator is able to detect
quencies). As a result, by plotting the relative stiffness along damage, distinguish a true damage state from a false one
the length of the beam for at least two distinct natural fre- and distinguish adequately between damage cases of differ-
quencies, the crack location can be identified by the inter- ent severity. It was also found that the versions of the DRQ
section of these curves. These authors mentioned that this indicator using the second spatial derivatives of the ODSs

did not perform much better than the versions using only the where the X and A subscripts refer to experimental and ana-
ODSs. lytical vibration modes, respectively. This correlation results
Zang et al. (2003a) presented two criteria to correlate in a single scalar number with a value ranging from zero to
measured frequency responses from multiple sensors and pro- unity. A unity value means that the mode shapes are identical,
posed using them as indicators for structural damage detec- whereas a zero value indicates orthogonality between the vec-
tion. The first criterion is the global shape correlation (GSC) tors, i.e., the mode shapes are ‘totally different’. Although the
function, which is sensitive to mode shape differences but MAC can provide a good indication of the disparity between
not to relative scales, being defined as: two sets of data, it does not show explicitly where in the
structure is the source of discrepancy. The co-ordinate MAC
H x1 { ( ω ) } { H x2 ( ω ) }
H 2 (COMAC) has been developed from the original MAC. It is
GSC ( ω ) = ----------------------------------------------------------------------------------------------------------------
- (5) the reverse of the MAC in that it measures the correlation at
{ H ( ω ) } { H ( ω ) } { H ( ω ) } H { H ( ω ) }
H
each degree-of-freedom (DOF) averaged over the set of cor-
 x1 x1
 x2 x2

related mode pairs. The COMAC identifies the co-ordinates
at which two sets of mode shapes do not agree, and is
where { H x1 ( ω ) } is a column of FRF baseline data measured defined as (Lieven and Ewins (1988)):
initially at frequency ω and { H x2 ( ω ) } is a column of the 2
current measured FRF data. This function exists for all fre-  nCMP 
quencies and the sum is over all locations. The second crite-  ∑ ( φ A ) ij ( φ X* ) ij 
rion, based on actual response amplitudes, is the global j=1 
COMAC ( i ) = n-------------------------------------------------------
- (8)
amplitude correlation (GAC): CMP n
2 CMP
2
∑ ( φ A ) ij ∑ ( φ X ) ij
H j=1 j=1
2 H x1 { ( ω ) } { H x2 ( ω ) }
GAC ( ω ) = ----------------------------------------------------------------------------------------------------------------------
- (6)
{ H ( ω ) } H { H ( ω ) } + { H ( ω ) } H { H ( ω ) } where nCMP is the number of correlated mode shapes.
 x1 x1
  x2 x2
 Although the COMAC can be useful as a tool to locate a
discrepancy, it does not have a physical basis. It is advisable
Each of the correlation criteria given by (5) and (6) is a to use this method in cases where the structure is tested and
function of frequency and uniquely maps a set of complex modelled in a free-free configuration, because grounding of
responses to a real scalar between zero and unity. These the structure will amplify the effect of anti-nodes away from
authors argue that a simultaneous use of both correlation the region of constraint and dominate the error prediction
criteria should be able to quantify the closeness of two sets (Maia and Silva (1997a)). On the other hand, although the
of vibration data. The averaged integrations of the GSC and COMAC is a useful algorithm to detect errors in spatial mod-
GAC functions along the frequency points over the measure- els, it does not provide a unique solution to localize the source
ment range are also proposed, referred to as averaged integra- of error in the structure. Another form of the MAC is the par-
tion GSC (AIGSC) and averaged integration GAC (AIGAC), tial MAC (PMAC), which correlates parts of modal vectors
as damage indicators for SHM. An experimental test on a and allows easy comparison of the selected parts. Kim et al.
bookshelf structure was conducted, but it was concluded (1992*) propose the combined use of the COMAC and the
that further studies would be needed to develop approaches PMAC concepts, referred to as the Total MAC, so that one
that could accurately assess structural states and damage. can isolate the structural area where damage has occurred.
An intelligent decision making method based on the radial Parloo et al. (2002b, 2004) compared input-output and
basis function (RBF) neural network is suggested by Zang output-only damage identification setups for an aluminum
et al. (2003b). beam structure suffering from fatigue induced crack formation.
As shown in the literature, there are several methods availa-
2.2. Mode Shape Changes ble for the calculation of sensitivity (derivatives) of eigenval-
ues and eigenvectors. Expressions for the sensitivity of mode
According to Doebling et al. (1996**), West (1984*) shapes to local changes on mass and stiffness can be found,
presents what is possibly the first systematic use of mode for instance, in Vanhonacker (1980), Maia and Silva (1997a)
shape information for the location of structural damage and Parloo et al. (2003). For example, the sensitivity of
without the use of a prior FE model. West (1984*) uses the DOF j of mode shape i to a local change in stiffness between
MAC to determine the level of correlation between modes DOFs p and q can be given by:
from the test of an undamaged Space Shuttle Orbiter body
flap and the modes from the test of the flap after it has been ∂φ n
1 φ pr – φ qr-
exposed to acoustic loading. ---------ji- = ( φ pi – φ qi ) ∑ --------------- ------------------- φ jr (9)
Kim et al. (1992*) investigated the use of the MAC and ∂k pq λ
r = 1, r ≠ i r
– λi ar
some of its forms for locating structural damage. The MAC,
which is probably the most common way of establishing a with λ r being the system poles and a r the modal scaling
correlation between experimental and analytical models, is factors. An equivalent formulation can be given for a local
defined by Allemang and Brown (1982) as: change in mass. These expressions are exact if all mode
2
shapes of the test structure are taken into account, though a
{ Φ X } iT { Φ A* } j good approximation can be obtained in practice, where only
 
MAC  { Φ X } i, { Φ A } j = --------------------------------------------------------------------------- (7) a limited number of modes are available. In practice, the
 { Φ } T { Φ * }   { Φ } T { Φ * } stochastic errors due to the presence of measurement noise
 X i X i  A j A j

reveal that calculating mass sensitivities can be numerically these conclusions may not seem remarkable, they confirm
more stable than the calculation of the sensitivity to stiff- some hypotheses that have been discussed during this work.
ness. If the reference condition of a structure is altered These authors also propose a mode shape based damage index
by either a local change in mass or in stiffness, one can for damage localization.
write first-order approximations with respect to the damage Kim, B. S., et al. (2005) used several damage detection
parameter. However, it is noted that these approximations criteria, including COMAC, ECOMAC (Enhanced COMAC,
are only valid for relatively small damage; this poses no which is based on the average difference between the ana-
problem, since one wants to detect early damage. In the lytical and experimental modal vector components, and first
case where frequency sensitivities to a change in a struc- introduced by Hunt (1992)), ADSM (Absolute Difference of
tural parameter are considered, Parloo et al. (2004) use Strain Mode shapes, which is based on the absolute difference
an iterative weighted pseudo-inverse algorithm technique of strain mode shapes, proposed by Qin and Zhang (1998))
which allows to better locate damage (Guillaume et al. and a new method based on the absolute difference of the
(2002)). absolute value strain mode shapes, with a focus on finding
In order to reduce the calculated sensitivity errors, the differences of these damage detection criteria caused by dif-
estimated modal parameters from output-only data should be ferent geometry and boundary conditions of gusset plates with
as accurate as possible. For this purpose, Guillaume et al. cracks. Considering crack localization, these authors concluded
(1998, 1999) propose the use of a maximum likelihood esti- that the new method generally performed better than the
mator (MLE), which is an unbiased estimator taking into other methods for the studied example, though it is hard to
account the noise information in the data. Instead of using find its physical meaning.
FRFs, the MLE uses spectral density functions.
Vecchio et al. (2002) present an experimental validation 2.3. Mode Shape and FRFs Curvatures
of a modal model based approach to structural damage detec-
tion, where output-only vibration data is used. However, oper- Mode shape and FRFs curvatures are widely used as
ational working conditions are not necessarily stationary, an alternative to damage identification from mode shape
and if small deviations are observed in the modal parame- changes.
ters, it is very difficult to decide whether damage is present or Salawu and Williams (1994*) evaluated the performance
not. The approach requires continuous monitoring of the sys- of some procedures for locating damage using mode shape
tem. The damage detection algorithms reduce time data to curvature (MSC) and mode shape changes. The first method
covariance data and perform a statistical 1 2-test that assesses estimates the mode shape curvatures using a central differ-
whether the recent data fit the reference model. In a previ- ence approximation as proposed by Pandey et al. (1991*):
ous work, Vecchio and Van der Auweraer (2001) identified
the modal parameters based on the stochastic subspace φ ( j + 1 )r – 2φ jr + φ ( j – 1 )r
method, but, in order to improve pole extraction (making it v jr ″ = ---------------------------------------------------
2
- (10)
faster and easier) and with the aim of achieving more auto- h
mation in the damage detection process, Vecchio et al.
(2002) introduced a new approach based on the MLE. This Here, h is the distance between the measurement co-ordi-
approach is applied to a reticular structure that was built for nates and φ jr are the modal displacements for mode shape r
several dynamic tests in the context of the EC Brite-Euram at the measurement co-ordinate j. Since a local reduction in
SAM project. stiffness results in a local increase in the curvature v″ , it is
Park and Park (2003) introduced a damage detection and reasonable to suppose that these can be used to detect, locate
localization technique based on incomplete experimental and quantify damage. The second method studied by Salawu
models, since it is not possible in practice either to measure and Williams (1994*) is based on the mode shape relative
the responses in all DOFs or to have a reference FE model differences proposed by Fox (1992), in which a graphical
that can represent the real structure in a precise manner, unless comparison of displacement mode shapes is used to indi-
one is studying simple structures. However, this method has cate damage position. Those authors also use the MAC and
the disadvantage of requiring heavy experimentation, because COMAC to correlate mode shapes, but although these cri-
one needs to measure all the FRFs of the reference structure teria proved to be sensitive to damage, they did not provide
for each considered DOF. Furthermore, this technique is not a clear localization of the damage position in a cantilev-
effective over the whole frequency range, and solutions are ered beam. The performance of both the MSC and mode
presented for the frequency range choices by error analysis. shape relative differences methods was not as good as desired,
Lazarov and Trendafilova (2004) investigated the influence since it is necessary to decide which are the most adequate
of damage on the dynamic characteristics of linear elastic mode shapes to use. Only the MSC method was able to give
plates. This study was based on the plate theory of Kirchoff, an indication of simulated multiple damage locations.
and tests were carried out on a square plate. Damage is intro- In a paper where the applicability of the MSC changes to
duced as thickness changes and as saw cuts perpendicular to damage detection in a prestressed concrete bridge is investi-
the boundary contour of the plate, and its influence on the gated, Wahab and De Roeck (1999) introduce the curvature
mode shapes and the modal frequencies of the structure are damage factor (CDF), where the difference in the MSC for all
then derived and analyzed. They first concluded, from both modes can be summarized in one number for each measure-
the numerical and experimental models, that the natural fre- ment point. This technique was then applied to a real structure:
quencies of thin plates are not influenced noticeably by dam- the Z24 bridge which lies between the villages of Koppigen
age, but that the mode shapes undergo considerable changes and Utzenstorf and crosses the highway A1 between Bern
as a result of the presence of damage. Despite the fact that and Zurich in Switzerland. The CDF is based on the MSC

central difference approximation (equation (10)) introduced Battipede et al. (2001) extend the gapped-smoothing tech-
by Pandey et al. (1991*), and is given by: nique to bi-dimensional models or plate-like structures, show-
ing that the method is able to locate single and multiple
damage sites of medium and great extent. This technique,
1 N
CDF = ---- ∑ ν i ″ – ν i ″ d (11) which had already been applied to one-dimensional models
Ni = 1 by Ratcliffe and Bagaria (1998) and Ratcliffe (2000), takes
advantage of the presence of an irregularity in the curvature
where N stands for the total number of modes to be consid- shape in order to detect damage. The displacement shape is
ered, ν i ″ is the curvature of the ith mode shape of the intact converted into curvatures by applying a second order finite
th
structure and ν i ″ d is the curvature of the i mode shape of the differentiation procedure:
damaged structure. These authors note that a relatively dense
measurement grid is required in order to get a good estima- Y ( x i + 1, y j ) – 2Y ( x i, y j ) + Y ( x i – 1, y j )
tion of the curvatures of higher modes. When more than one C xx ( x i, y j ) = -------------------------------------------------------------------------------------
- (17)
fault exists, it is not possible to detect the various damage h x2
locations from a single vibration mode, i.e., all modes
should be carefully examined in order to locate all the exist- in which Y(xi, yj) is the measured displacement (perpendicu-
ing flaws. lar to the plate plane) and hx is the uniform spatial separation
Ho and Ewins (2000**) attempted to evaluate whether of the measurement sensor grid along the x direction. The
the presumption that damage is located at the point where curvature along the y direction, Cyy, is evaluated similarly.
the mode shape change is the greatest is valid using both In the absence of damage and other irregularities, the curva-
simulated and experimental data, since the differentiation ture has a smooth shape and can thus be represented as a pol-
process enhances the experimental variations inherent to mode ynomial function. If the smooth function and the measured
shapes. They addressed five methods based on mode shapes curvature are different when evaluated at the same location,
and their derivatives: flexibility index (FI), mode shape cur- (xi, yj), this can be an indicator of damage at that location.
vature (MSC), mode shape curvature square (MSCS), mode Furthermore, the fact that one is dealing with a bi-dimen-
shape slope (MSS) and mode shape amplitude compari- sional model allows the cross curvature, Cxy, to be taken into
son (MSAC). If more than one mode is defined, it follows account so that a damage index can be formulated as:
that: 2
= a xx  C xx ( x i, y j ) – C xx ( x i, y j )
(m)
δ i, j
 
∑
d 2 2
FI i = φ ij – φ ij (12) 2
+ a yy  C yy ( x i, y j ) – C yy ( x i, y j )
j (m) (18)
 
∑ φ ij″ – φ ij ″
d
MSC i = (13) 2
+ a xy  C xy ( x i, y j ) – C xy ( x i, y j )
(m)
j
 
∑ φ ij″ 2 – φ ij ″ 2
d
MSCS i = (14) where the aij coefficients can be set to 0 or 1, to either con-
j sider or neglect the corresponding curvature. Battipede et al.
(2001) considered axy = 0, since it is difficult to estimate Cxy
∑
d
MSS i = φ ij ′ 2 – φ ij ′ 2 (15) from experimental data. For an undamaged structure, the dif-
j ference between measured and fitted data is mainly due
to measurement noise, so that in an experimental case the
∑
d
MS i = φ ij – φ ij (16) damage index δij is irregular without any sharp peak. If the
j structure is faulty, however, some clear peaks will appear,
corresponding to the sensors located around the damaged
The last two methods, corresponding to equations (15) region of the model under consideration. The major advan-
and (16), are presented by Ho and Ewins (2000**). The first tages of this technique, when applied to a plane rectangular
(15) corresponds to an attempt to introduce the mode shape homogeneous plate, are (i) the possibility of locating multi-
deflection as a feature sensitive to damage but relatively ple damage sites without knowledge of a reference model of
insensitive to experimental variation. The second equation the structure and (ii) only a single test is needed to deter-
(16) has the advantage of not requiring the mode shape dif- mine the current state of the structure, without knowledge of
ferentiation. In calculating the derivatives of mode shapes, its history. In practice, however, measured data tend to be
instead of using finite difference approximation, polyno- polluted by noise, detrimentally affecting the effectiveness
mial functions are used: a local polynomial is fitted through of the technique.
every set of four consecutive measurement points and the Yoon et al. (2005) proposed the generalization of the one-
resulting polynomial is differentiated. These authors con- dimensional gapped smoothing technique to bi-dimensional
clude that the experimental results show that higher deriva- models. As before, it is shown that it is possible to locate
tives are more promising for damage identification, but that damage using only data from the damaged structure, i.e.,
false damage indications may be observed at mode shape without resorting to a reference model. Provided that it is
nodal points or where the quality of the measurements is known that the structure, in its undamaged state, is homoge-
relatively poor. It was also observed that this may also occur neous with respect to stiffness, the procedure will detect the
at the boundaries. non-homogeneous areas that are caused by damage. Varia-

bility indices are generated for each test point on the structure, change of a mode shape function is greatest. The second group
after which statistical treatment and outlier detection ena- distinguished by Maia et al. (2003), the FRF-based damage
bles discrimination of areas with significant variation in stiff- detection methods, includes the FRF-based mode shape, the
ness. The structural irregularity index is calculated as: FRF-based mode shape slope, the FRF-based mode shape
curvature (also presented by Sampaio et al. (1999)), the FRF-
2
δ i, j = ∇ Ψ i, j – C i, j (19) based mode shape curvature square and the FRF-based dam-
age index; these methods are equivalent to the mode shape
where ∇ 2 Ψ i, j is the curvature of either a mode shape or an detection methods, but using FRFs instead of mode shapes.
ODS and Ci, j is the smoothed curvature shape. It has been In Maia et al. (1997b) it is noted that when calculating the
shown that this method allows for the determination of the damage indices for the FRF-based methods, as we progress
size and location of relatively small localized stiffness reduc- along the frequency range, compliing information, the results
tions in homogeneous plate-like structures. Analytical mod- eventually begin to degenerate instead of improving. To
els or reference experimental models of the structure in the overcome this, the location where the difference of damage
undamaged state are not needed. The results generated using and undamaged case is a maximum is looked up for each
the broadband ODS data showed superior performances to frequency and an occurrence is counted at that location. The
those produced using the mode shapes. However, in the case occurrences are summed up along the frequency range for
of large areas or uniform damage (when compared to the each location, and the location where the greatest number of
spacing and number of measurement points), the algorithm occurrences has been counted is proposed as the damage site.
identifies only the edges of the damage, which can make the Maia et al. (2003) found that the methods based on the cur-
interpretation of results more difficult. It was concluded that vatures performed better, although false damage indications
further research is required in order to establish a quantita- remain a problem requiring close attention. Maia et al. (2003)
tive relationship between the local stiffness reductions and suggested that some improvements could be made in the
the irregularity indices. interpolation process, in defining a noise level under which
Maia et al. (2003) present a series of numerical simulations the results are not considered, in the method to calculate the
as well as an experimental example using a simple beam in maximum occurrences, in applying statistics to the results,
order to compare various damage detection methods based on and, finally, in the set-up and skill of the experimentalist.
mode shape changes. They also proposed a generalization of Sampaio et al. (1999) compared the FRF curvature method,
these methods to the whole frequency ranges of measurement, which encompasses the first three levels of the damage iden-
i.e., methods based on mode shapes become based on ODSs. tification process (existence, location and extent), with two
This way, one may avoid modal identification, which can be of the most cited methods in literature: the mode shape cur-
a time consuming process and can filter relevant informa- vature (MSC) method and the damage index method. The
tion. The studied methods fall into two major groups: mode FRF curvature method is based only on the measured data
shape damage detection methods and FRF-based damage without need for any modal identification, and is an extension
detection methods. In the first group, they included all the of the method of Pandey et al. (1991*) to include all fre-
methods overviewed by Ho and Ewins (2000**), except the quencies in the measurement range and not just the modal
flexibility index (FI) method. They also included the damage frequencies, i.e., it uses FRF data rather than just mode shape
index method, developed by Stubbs et al. (1995*), which is data. In fact, the method uses something like an ‘operational
based on the modal strain energy decrease between two struc- mode shape’ defined, for each frequency, by the frequency
tural DOFs for linear-elastic Bernoulli-Euler beams, and response at different locations along the structure. The abso-
given by: lute difference between the FRF curvatures of the damaged
and undamaged structure at location i, along the chosen fre-
 b dφ ″ ( x ) 2 dx + L dφ ″ ( x ) 2 dx L [ φ ″ ( x ) ] 2 dx
 ∫a j ∫0 j  ∫0 j quency range ω , is calculated, for an applied force at point j,
β ij = ----------------------------------------------------------------------------------------------------------------------- (20) by:
 b [ φ ″ ( x ) ] 2 dx + L [ φ ″ ( x ) ] 2 dx L dφ ″ ( x ) 2 dx
 ∫a j ∫0 j  ∫0 j
∑ω
d
∆α ij ″ = α ij ″ ( ω ) – α ij ″ ( ω ) (22)
where a and b are the limits of a segment i of the beam
where damage is being evaluated and L is the total length of Finally, one can sum up results from several force location
the beam. For the purpose of the article presented by Maia cases. This technique was tested using experimental data
et al. (2003), the discrete formulation of equation (20) is gathered from the I-40 bridge over the Rio Grande in Albu-
used. For more than one mode, the damage index can be querque, New Mexico, USA, following previous work done
defined as: by Jauregui and Farrar (1986a*, b*) of the Los Alamos
National Laboratory (LANL) in the USA. From work on the
DI i = ∑j βij (21) FRF curvature method, Sampaio et al. (1999) point out the
following: (i) the method works better for a range before the
first anti-resonance or resonance frequency (whichever comes
Cornwell et al. (1997) generalized the damage index first), since for wider frequency ranges (including several
method to plate-like structures characterized by a two- modes) the difference between the curvatures of the dam-
dimensional curvature. One of the advantages of this strain- aged and undamaged models becomes less significant when
energy based method is that it does not require a well corre- compared with the amplitude difference arising from the
lated FE model and/or mass normalized mode shapes. The resonances’ frequency shift, because of the loss of stiffness;
mode shape detection methods group is based on the assump- (ii) the position of the exciting force seemed not to be an
tion that damage is located between the DOFs where the important influence, according to the numerical model; (iii)

the method is considerably insensitive to noise; (iv) it can be periodic structures, in which slopes and curvatures of mode
used to detect, locate and quantify damage, though this last shapes are used to localize damage, and natural frequencies
aspect deserves further developments; (v) the method has the are then used to quantify its extent. The equations of sensi-
great advantage of being simple and not requiring prior modal tivity coefficients of mode shapes, slopes and curvatures of
identification. With respect to the comparison established a mono-coupled periodic system are first derived in terms of
between the three methods, Sampaio et al. (1999) argue that receptances of a periodic element. The numerical and exper-
the FRF curvature method had the best overall performance, imental examples show that curvatures of modal shapes are
although the MSC method produces better results for higher the most sensitive to damage, but slopes of mode shapes seem
order modes. to provide a better indication of the damage location. Also,
Pirner and Urushadze (2004) compared the CAMOSUC the larger changes in the natural frequencies imply higher sen-
(change of mode surface curvature) with earlier methods for sitivity of these modes to damage, which is useful for choos-
detection of the magnitude and location of damage; specifi- ing a few of the lower frequency modes to localize damage.
cally MAC, COMAC, and the error-matrix. The CAMOSUC The method allows for detection and localization of multi-
is defined as the difference of the mode shape curvatures ple damage sites, with different simultaneous extents, using
between the undamaged and damaged states. Thus, each term a few of the slopes and curvatures of the lower order mode
in the curvature damage factor (CDF) equation (11) is the shapes, though the accuracy with which the damage size can
CAMOSUC itself. These authors present detailed applica- be quantified depends on the number of frequencies used.
tions of some of these methods to real cases, such as a square
Plexiglas plate, a Plexiglas model of a railroad bridge sec- 2.4. ARMA Family Models
tion, a three-span prestressed concrete bridge, floor slabs of
a department store and a hyperbolic cooling tower. Brincker et al. (1995a*) used a statistical analysis based
Hamey et al. (2004) approached the experimental aspects on shifts in the natural frequencies to detect damage in two
of methods which make use of mode shape curvatures, com- concrete beams with different reinforcement ratios. The modal
paring four methods: the absolute difference of curvature parameters (natural frequencies and modal dampings) are esti-
mode shape method, the CDF method, the damage index mated from the response time series (accelerations) using the
method and the FRF curvature method. However, instead of so called auto-regressive moving average (ARMA), identi-
obtaining approximations to the second derivatives of the cal to the one used by Brincker et al. (1995b*) in an offshore
displacements, as usual, they proposed the use of direct meas- platform. Referring to Davies and Hammond (1984), Pandit
urements of the curvatures of the mode shapes obtained from and Metha (1985) and Safak (1991), Brincker et al. (1995a*)
piezoelectric materials. They used piezoceramic patches for mention that the ARMA models have been developed mainly
sweep-sine continuous excitation and a hammer for impulse for application in economics and electrical engineering, but
excitation. Sixteen polymeric PVDF sensors were distrib- since they are considered to be a more effective way of esti-
uted along the length of one face of the beam. They tested mating modal parameter than Fast Fourier Transform (FFT)
several cases of damage: a transverse crack, three scenarios based techniques, their use on structural systems has been
of controlled non-symmetric delamination and impact damage, increasing during recent years. A more complete version is
and observed that, in general, the impulse excitation produced the ARMA with exogenous input (ARMAX). The [n, m] order
slightly better results than the continuous excitation. With ARMAX model that describes the time response y(t) as a
respect to the identification methods, they observed that the linear combination of the last n responses as well as the last
damage index method performed better than the others both m unknown inputs and p known inputs is given by:
on damage detection and localization. According to these
n p
authors, the FRF curvature method turned out to be the one
which produced the worst results. Also, they observed that
y(t) = ∑ c ( i )y ( t – i ) +
i=1
∑ b ( i )x ( t – i )
i=1
when the sensors are placed on the opposite side of the beam (23)
m
to the delamination, it was possible to obtain better results
for the cases where the delamination had greater dimensions,
– ∑ d ( i )e ( t – i ) + e ( t )
i=1
although sensor position seemed not to affect the results for
the other damage cases. where c(i) is the auto-regressive parameter describing the
Kim, B. H. et al. (2005) attempted to extend the MSC and response y(t) as a linear regression of the last n values and
damage index methods by resolving some deficiencies of d(i) and b(i) are the moving average parameters which
these methods. They solved a set of linear algebraic equations describe the response y(t) as a linear regression of the last
referred to as flexural damage index equations (FDIE), which values of, respectively, an unknown input time series, e(t),
avoid the singularity problem near the inflexion points and which may be assumed as white noise, and a known time
the mode selection problem of the MSC and damage index series, x(t). The c(i) (AR), d(i) (MA) and b(i) (X) parameters
methods. The FDIE shows that there is a strong relationship are determined by minimizing the variance of the noise time
between the damage and the curvature of flexibility. This series e(t), i.e., based on the measured response and on the
method is also able to detect, locate and determine the size of predicted response given by (23). When the AR parameters
structural damage in a plate-like structure using the measured are estimated, the roots of the characteristic polynomial equa-
modal parameters. However, a very small interval between tion that contains these parameters can be related to the nat-
sensors is required if one wants to get an accurate estimate of ural frequencies and modal dampings. Hence, the ARMAX
the severity of damage. can be utilized to examine these parameters’ variations with
Zhu and Xu (2005) present a sensitivity-based method for time. Brincker et al. (1995a*) also defined a statistical sig-
localization and assessment of damage in mono-coupled nificance indicator which takes account of the changes in

the natural frequencies, so that they can give a quantifiable obtained using a FE model) and damaged states. Due to the
estimate of the structural changes. It was shown that the signif- inverse relation to the square of the modal frequencies, the
icance indicators are sensitive to structural damage, but they dynamic flexibility matrix is very sensitive to changes in the
are not able to provide an accurate estimate of damage location. lower order modes, whereas the stiffness matrix is more sen-
Finally, it was observed that, for the studied beams, the dif- sitive to higher order modes. Based on this, Doebling et al.
ferences between the ARMA and ARMAX were quite small. (1996**) distinguished five different approaches: compari-
Thus, it was concluded that knowing the input signal is not son of flexibility changes, unity check method, stiffness error
essential for the detection of damage in the studied beams. matrix method, effects of residual flexibility and changes in
Wei et al. (2005) used nonlinear ARMA with exogenous measured stiffness matrix.
input (NARMAX) models to assess internal delamination in Park et al. (1988*) suggested the use of the weighted error-
multi-layer composite plates. These authors state that the matrix method, comparing it to the error-matrix method (Sidhu
relationship between input and output is nonlinear for multi- and Ewins (1984) and He and Ewins (1985*)). The stiffness
layer composite panels. According to them, the NARMAX error matrix can provide valuable information on the posi-
model was proposed by Leontaritis and Billings (1985a, b) tion of the damaged area, being defined as:
as a general parametric form for modeling nonlinear systems.
NARMAX models describe the nonlinear systems in terms [ E ] = [ K2 ] – [ K1 ] (24)
of difference equations relating the current output to combi-
nations of inputs and past outputs. The structure of the pro- where [K2] and [K1] are the stiffness matrices of the evalu-
posed model is used to estimate model parameters for damaged ated members in the damaged and undamaged states,
vibration systems, and internal delamination is detected by respectively. Although [K1] can be estimated from a FE
comparing the model parameters of the intact and damaged model, it is possible to determine [E] using only experimen-
systems. The results show that, for the case studied, NAR- tal data. When the stiffness error matrix [E] is plotted, the
MAX models allow investigation of the behavior of com- highest peak will indicate the most probable damage loca-
posite models as well as assessment of damage (especially tion. Due to measurement noise, this method is only reliable
delamination) in composites. when the measured natural frequency shifts and modal vec-
Lu and Gao (2005) proposed an approach formulated as a tor changes exceed the magnitude of the measurement
predictive auto-regressive model with exogenous input (ARX), errors. This lead Park et al. (1988*) to propose the weighted
based on the linear dynamic system theory. After some sim- error-matrix method, which magnifies the amount of stiff-
plifications, the model is expressed in such a way that only ness error only at certain nodal points related to the dam-
response signals are involved, with a response at one location aged element. Although it has been shown that the weighted
chosen as the input of the model. The residual error of the error-matrix method is more sensitive to damage, since it
established model, when applied to actual measured signals, was possible to accurately detect the damaged structural
reflects the structural change, and the standard deviation of member at lower levels of damage, its complexity does not
the residual error is found to be a damage sensitive feature. It make it an attractive method. In order to reduce the time
was observed that the proposed ARX model can behave required for computation and experiments, an iterative
undesirably when the location of the chosen input response method, in which the number of elements used is refined at
is near the location of the damage. For this reason, they rec- each step, has been proposed.
ommend that in a diagnosis situation two separate runs of the
procedure are carried out using two different input locations. 2.6. Modal Strain Energy
It was also noted that the standard deviation of the residual
error of the ARX model, although sensitive to the presence Choi and Stubbs (2004) developed a method to locate and
of damage, does not give a precise indication of the degree determine the size of damage in a structure measuring time-
of damage, and further research is therefore required. domain responses in a set of measurement points. The mean
In an earlier work, Sohn and Farrar (2001b) had pre- strain energy for a specified time interval is obtained for
sented a two iteration AR-ARX model in order to predict each element of the structure, and used in turn to build a
the time series and subsequently use the standard deviation damage index that represents the ratio of the stiffness param-
ratio of the residual errors as the damage sensitivity feature. eters of the pre-damaged and post-damaged structures:
The residual error is the difference between the actual meas- NT
urements and the predicted responses from the combination ∑ { Vi }
T
[ K ] { Vi }
of the AR and ARX models. i=1
β j = ----------------------------------------------
NT
d T
2.5. Dynamic Flexibility ∑ { Vi } [ K ] { V id}
i=1
Another class of damage detection methods is based on  NT NT

d  (25)
 ∑ { V i } [ C j ] { V id} + ∑ { V i } [ K ] { V i }
d T d T
the use of the dynamic flexibility matrix to estimate changes
i=1 i=1 
in the static behavior of the structure. The dynamic flexibil-
×  ----------------------------------------------------------------------------------------------------
ity matrix is a transfer function, defined as the inverse of the  NT NT 
 { Vi } [ Cj ] { Vi } + ∑ { Vi } [ K ] { Vi } 
T
∑
T
static stiffness matrix. Thus, each column of the flexibility  
matrix represents the displacement pattern of the structure  i=1 i=1

associated with a unit force applied at the associated DOF.
Damage is then identified by comparison of the flexibility In this equation, β j is the damage index for the element j,
matrices of the structure in the undamaged (which may be NT is the number of sampling points, {Vi} is the displace-

ment configuration vector at time ti, and [Cj] is the geometric responses at points around damage. Considering a general
portion of the contribution of the jth element to the system structure with i measuring co-ordinates and j excitation co-
stiffness matrix [K]. Possible locations of damage in a struc- ordinates, it is possible to compute the transmissibility matrix
ture can be identified by utilizing a classification algorithm for both the reference (undamaged) and damaged states. The
with the damage index given by (25) taken as the feature transmissibility matrices are square, with m × m elements
vector. A statistical decision technique of hypothesis testing and a zero diagonal. With this arrangement, it is possible to
is used in the classification algorithm, in which the damage write:
indices given by (25) are normalized, assuming that the
damage index is a random variable and the collection of the {x} = [ T]{x} (27)
damage indices are distributed normally. Once the possible
locations of damage are isolated, corresponding damage sever- This is a special case of the transmissibility matrix, which
ities can be obtained (i.e., the fractional loss in stiffness). relates the set of m responses with itself. The sum of all the
These authors present, as major conclusions, the fact that transmissibility matrices in the frequency range of interest
the time-domain response may be used directly to localize produces a unique matrix for each case, [dST] for the damaged
and determine the size of damage in a structure, and that it is and [ST] for the undamaged case. The difference [∆ST] =
d
possible to reduce false-negatives by lowering the significance [ ST] – [ST] between them will show the location of dam-
level for damage localization, at the expense of increasing age, since its values should display a significant increase near
the number of false-positives. Thus, the balance between them the damaged region. To further clarify the location, the dif-
represents a measure of the sensitivity of the method as well ferent values in each line i of the [∆ST] matrix are summed
as of the quality of the results and the effectiveness of the and the second derivative ∆ST i ″ is calculated by a central
algorithm for damage localization. difference approximation. Matrices [dST] and [ST] can be
Patil and Maiti (2005) provided an experimental verifica- represented in a contour plot, where the rows are the x-axis
tion of an energetic method for prediction of the location and the columns are the y-axis. Any noticeable change of these
and size of multiple cracks based on the measurement of plots may indicate the presence of damage, which will be
natural frequencies for slender cantilevered beams with two located between the co-ordinates of ∆ST i ″ that show major
or three normal edge cracks. In the theoretical model the differences. Sampaio et al. (2000) demonstrated another way
beam was divided into a number of segments and each seg- of applying the transmissibility concept to damage detec-
ment was considered to be associated with a damage index. tion, referred as the pseudo-transmissibility matrix method,
The cracks are represented as torsional springs. The damage which uses FRFs instead of responses. Despite the fact that
index behaves as an indicator of the amount of strain energy this version performed better, these authors note that the
stored in the crack (or torsional spring). The method is based former (using the transmissibility matrix) is more consistent
on the concept that the strain energy U of a beam containing because the pattern change is correlated among the various
a crack is reduced because the beam can deform more easily plots. These authors also developed the frequency of maxi-
to the same extent than an uncracked beam. This reduction mum differences algorithm, to improve the accuracy of the
is equal to the amount of energy stored in the fictious tor- transmissibility method and others, such as the FRF curva-
sional spring (or crack), which, for the case where m cracks ture method, that use raw data directly from the FRFs. The
are present is: transmissibility matrix, pseudo-transmissibility matrix and
FRF curvature methods were applied to a 10-DOF numerical
m
M2 model. It was concluded that using the frequency of maxi-
U = Un – ∑ -------i-
i = 1 2K i
(26)
mum differences algorithm greatly improves the methods
which use FRFs in damage detection and location. However,
where Un is the stored energy in the corresponding uncracked the version which uses the transmissibility matrix for sev-
beam for vibration mode n and Mi and Ki are the bending eral simultaneous forces did not work in the examples tried.
moment at the crack location x = li and stiffness associated Mares et al. (1999**) applied genetic algorithms to dam-
to the crack, respectively. Based on these principles, and age assessment where the objective function is based on the
after performing several simplifications and some mathe- measured transmissibilities. The method was tested on a
matical manipulation, an expression for the location of a numerical model of a four story steel frame structure. The
crack in a segment of the beam is suggested. It is concluded model was developed with the intention of having a high in-
that, on the whole, the accuracy of the method is good for plane rigidity (of the floors) with respect to the out-of-plane
the prediction of the crack location, but lower for the predic- rigidity (of the columns). Four damage scenarios were con-
tion of the crack size. Accuracy decreases as the number of sidered, using combinations of Young modulus reductions
cracks increases. One of the major drawbacks of this method and removal of columns and braces. These authors observe
is revealed when it is observed that it is necessary to meas- that not all the transmissibilities are sensitive to damage in the
ure a number of frequencies equal to twice the number of same way. Therefore, they performed a two-step procedure
cracks in order to predict the location and size of all cracks, to select the ‘best’ transmissibility functions before running
which number it may not, in practice, be possible to deter- the genetic search, in order to enhance the optimization proc-
mine in advance. ess and hence the damage assessment procedure. The proce-
dure used enabled transmissibility functions to be ranked in
2.7. Transmissibility terms of their sensitivity to damage. The transmissibilities
found to be the most sensitive to damage were used in the
Sampaio et al. (2000) showed that a minor stiffness change formulation of the cost function. The optimization process
can result in a significant change of transmissibility relating follows two iterations to obtain the global minimum of the

cost function. First, a genetic search is applied to the cost ∆U

function, and then a classical gradient-based algorithm is run SDC = -------- (28)
U
to refine the solution, so that global minima are correctly
distinguished from local minima. Mares et al. (1999**) where ∆U is the energy dissipated in one cycle and U is the
observed that, according to the numerical simulations they total energy stored in that cycle. The energy stored by the
performed, the use of transmissibility functions provide esti- beam in bending, U, is calculated from the mode shape maxi-
mates of the state of damage that are sufficiently accurate. mum deflection at the center of the beam. The energy dissi-
However, for the steel frame structure model used, damage in pated in one cycle is equated to the energy input. Tests were
braces was easier to identify than damage in columns. carried out both in air and in vacuum, and it was found that
Johnson, T. J. et al. (2004) presented work on smart sen- air damping significantly affected the measurements, partic-
sor arrays with the goal of implementing local vibration-based ularly when the beam possessed off-axis reinforcement.
diagnostic algorithms inside a smart ‘black box’ to demon- Yam et al. (2004) proposed to develop a practical method
strate the feasibility of distributed health monitoring for dam- for the location and prediction of damage in CFRP plates by
age detection, location and quantification. It was shown that means of a combination of the measured modal damping
the transmissibility is a good indicator feature of structural and the computed strain energy distribution. First, a FE model
damage in spite of environmental fluctuations and boundary is established to estimate the modal parameters, such as nat-
condition non-linearities. For this reason, the transmissibil- ural frequencies, mode shapes and modal strain energy. Based
ity-based damage identification technique is referred to as on the numerical model and on the fact that delaminations
the structural diagnostics using non-linear analysis (sDNA) in different regions produce characteristic increases in the
approach. They demonstrated that this approach reduces the modal damping, they were able to experimentally locate
number of false diagnoses due to non-linearities. Experi- damage. These authors note that the numerical simulations
ments were carried out on two large distributed structures; a provide a good explanation for damping increase due to dela-
building and a rotorcraft fuselage. mination, i.e., the energy dissipation is mostly induced by
interfacial slip across the delamination and the tendency
2.8. Damping for penetration between the upper and lower surfaces in
the delamination region.
Based on the observation that modal damping is a parame- Various authors have tried to study the damping mecha-
ter with higher sensitivity to internal delamination of CFRPs nisms in composite materials. For example, Nayfeh (2004)
than the natural frequencies, Keye et al. (2001) developed a developed a model for vibration parallel to the plane of lam-
method which is capable of relating modal damping devia- ination of a symmetric five-layer elasto-viscoelastic sandwich
tions caused by structural damage to the damage location on beam, since it is known that a sandwich beam consisting of
the structure. They call attention to two important aspects of alternating elastic and viscoelastic layers can be designed to
the experiments. First, when only a single sensor (or a small exhibit large damping on flexural vibration in the direction
number) is available, the experimental data provides no spa- normal to the plane of lamination. Berthelot and Sefrani (2004)
tial information on the dynamic response, i.e., on the mode experimentally analyzed damping of unidirectional glass-
shapes, and the process of localizing structural damage has fibre and Kevlar composites, comparing them to the models
to cope with eigenfrequencies and modal damping values, of Adams-Bacon, Ni-Adams and complex stiffness. The pur-
which can be extracted from the measured FRFs. In such a pose was to develop an analysis of the effect of the fibre-rein-
case, a FE model should be used to locate the damage. Sec- forcement orientation in the case of unidirectional layers,
ond, since in practice the damage location is unknown, a since it is known that, at the constituent level, the energy dis-
range of damage locations must be simulated in the numeri- sipation in fibre-reinforced composites is induced by differ-
cal model and the modal data must be computed for each ent mechanisms such as the viscoelastic nature of the matrix
damage case. After testing the method on a plate-like CFRP and fibre materials, damping at the fibre-matrix interface,
structure, they proposed some improvements: (i) to enhance damping due to damage, etc.
the correlation between numerical and experimental data,
updating techniques should be used; (ii) implementation of 2.9. Impedance-Based Methods
a systematic mode selection strategy and the weighting of
individual modes based on their sensitivity to damage local- Park et al. (2003) present an overview of piezoelectric
ization; (iii) optimization of actuator/sensor placement and impedance-based health monitoring where the hardware and
evaluation of their influence on the damage identification software issues are summarized, including a discussion of
capability; (iv) evaluation of other aspects, such as tempera- future research areas and the path forward.
ture effects and influence of the excitation method. Park et al. (1999**) introduced an impedance-based method
Kyriazoglou et al. (2004) explore the use of the specific which uses a piezoelectric transducer (PZT) as both actuator
damping capacity (SDC) for damage detection and localiza- and sensor. It can be shown that the electrical impedance of
tion in composite laminates. One important observation is the PZT is directly related to the structure’s mechanical
that the resonant frequency allows detection of cracks in impedance (Liang et al. (1994)). Thus, a relationship can be
glass fibre-reinforced laminates, while for carbon fibre-rein- found to allow determination of the structural properties;
forced laminates no detectable changes in the resonant fre- specifically, the mass, stiffness and damping. Any change in
quencies could be found. However, large changes were the mechanical impedance, which could be caused by the
found in the SDC. The method consists of measurements and presence of damage, will show up in the electrical imped-
analysis of the SDC of composite beams vibrated in free- ance of the PZT. The damage state is identified when a
free flexure in their first mode. The SDC is defined as: defined metric, given by the sum of the squared differences

of the impedance between the current and reference states, example using a 5-DOF system and an experimental investi-
increases above a predefined threshold. These authors point gation on a multi-story building model, and it was found that
out that this technique possesses the twin advantages of not the algorithm could assess the condition of a structure in a
requiring numerical models and the fact that its high-fre- more quantifiable manner than the traditional impedance
quency excitation nature makes it very sensitive to local approaches. Some issues that need close attention were iden-
structural modifications. tified: (i) the robustness of the proposed algorithm against
Grisso et al. (2004) used an impedance-based SHM system environmental condition changes should be improved; (ii)
to detect the onset of transverse matrix cracking in cross-ply different damage sensitive features could be used, such as
graphite/epoxy composite. Ceramic PZT patches are attached relative and/or absolute changes in exogenous or auto-regres-
to the composite samples (in this case, beams) to simultane- sive ARX coefficients; (iii) a prior curve fitting on the imped-
ously excite the structure with high-frequency excitations (two ance data should be performed before constructing the ARX
frequency ranges, from 10 to 20 kHz and from 40 to 60 kHz) model in order to minimize the effect of leakage in the fre-
and monitor any changes in structural mechanical impedance. quency response measurements; (iv) the differentiating proc-
In order to detect damage severity, these authors use the ess involved in feature extraction often led to amplifications
root mean square deviation (RMSD), using the squared dif- of existing noise in data; (v) the selection of proper EVS
ferences of the impedance between the undamaged and actual distribution for each data set and the evaluation of associated
states. These authors conclude that acoustic emissions test- distribution parameters are still strongly based on an initial
ing showed a good correlation only for thicker specimens, guess and on a trial-and-error method.
while the impedance-based method gave better correlation
for thinner specimens. 2.10. Lamb waves
Moura and Steffen (2004) tried to identify the best fre-
quency bands for impedance-based structural damage iden- One approach that has received much attention in the last
tification in flexible structures. Damage was simulated in a few years is the use of Lamb waves. Lamb waves are a type
cantilevered aluminum beam by adding a mass to the free end of elastic perturbation that can propagate across large areas
of the beam. By observing the results, these authors agreed of a free-free solid plate with low dispersion of energy, even
that, at least for the conditions used, the best low frequency in materials with a high attenuation ratio. This type of wave
value is 15 kHz, while the best bandwidth value is the inter- was first described in theory by Lamb (1917), although he
mediate one (4 kHz). never tried to produce them. Alleyne and Cawley (1992)
Peairs et al. (2005) developed a new modeling technique were among the first to discuss interaction of Lamb waves
for impedance-based SHM, combining the spectral element with defects for non-destructive testing. Saravanos et al.
method (SEM) with electric circuit analysis. These authors (1994*) presented a procedure for delamination detection in
argue that SEM has several advantages over other conven- composite materials using Lamb Waves and embedded pie-
tional FE techniques, for example because SEM more accu- zoelectric sensors.
rately models higher frequency vibrations because the mass is Kessler et al. (2002) maintain that techniques using Lamb
modeled ‘exactly’ and incorporates higher order models more waves have proven to provide more information about dam-
easily. Sensor multiplexing for transfer impedance and high age type, severity and location than previously tested methods
frequency modeling were also investigated. However, mod- using FRFs, since Lamb waves are more sensitive to local
eling at higher frequencies is more complex because of the structural defects. Piezoceramic patches were used to excite
presence of the PZT resonances. the first anti-symmetric Lamb wave (A0 mode). The PZT actu-
Simmers et al. (2005) applied an impedance-based method ators were chosen because of their high force output at rela-
to detect and quantify the onset and growth of pre-crack sur- tively low voltages and also due to their good response
face corrosion. Three metrics were used: one similar to the qualities at low frequencies. Kessler et al. (2002) explored the
one used by Park et al. (1999**), the RMSD and a cross- optimization of Lamb wave methods for damage detection
correlation. The experimental results on an aluminum beam in composite materials, covering the problems of choosing
indicate that the impedance-based method is effective for the appropriate actuating frequency, pulse shape and sensor
corrosion detection and tracking, and that there is a relation- geometry for Lamb wave application. The results were com-
ship between the metrics and corrosion location, surface pared by performing a wavelet decomposition using the
coverage and pit depth. Morlet wavelet, and plotting the magnitude of the coeffi-
Park, G. et al. (2005) proposed the use of statistical pattern cients at the driving frequency. Although Lamb waves dis-
recognition methods to address damage classification and data play great capabilities in damage detection and localization,
mining issues associated with large numbers of impedance these authors point out, as the major disadvantage of this
signals for health monitoring applications. In order to diag- method, that it has to be active; i.e., it requires a voltage supply
nose damage with statistical confidence, the impedance- and function generating signal. Another awkward require-
based monitoring is cast in the context of an outlier detec- ment is the high data acquisition rate needed to gain useful sig-
tion framework. The statistical process control developed nal resolution. Finally, the Lamb wave method should most
consists of a two-stage prediction model that is a combination likely be placed into a SHM system in conjunction with another
of AR and ARX models, the AR-ARX. The damage-sensitive passive detection method, such as an FRF method, in order
feature is then computed by differentiating the measured to conserve power and data storage space and because the
impedance and the output of the ARX model. Because of the Lamb wave data can be more difficult to interpret.
non-Gaussian nature of the extracted features, extreme value Su et al. (2002) state that noteworthy efforts have been
statistics (EVS) is employed to develop a robust damage made since the 1980s in the field of wave propagation based
classifier. The proposed technique was tested in a numerical identification methods. Nevertheless, it is known that the

propagation characteristics of Lamb waves are relatively while it was observed that the Lamb wave propagation
complicated because of the dispersion phenomenon. These method is effective for thin plates, the impedance method is
authors studied the interaction between the fundamental Lamb more suitable for detecting damage near structural joints or
wave modes and delamination in carbon fibre-reinforced connections.
laminates. Models (FE and experimental) of a plate with a Toyama and Okabe (2004) studied the effects of tensile
delamination were tested, using PZT wafers as sensors/actu- strain and transverse cracks in the propagation velocity of
ators. The calculated and measured dynamic responses were Lamb waves in cross-ply FRP laminates, since the waves’
processed using wavelet transform-based analysis in the propagation velocity is sensitive to changes in the in-plane
time-frequency domain for the purpose of damage evaluation, stiffness of the laminates. It is noted that some compos-
which has proven to be very effective in diminishing the ite materials, such as CFRP, exhibit non-linear stress-strain
influence of broadband noises and structural vibration patterns responses, which makes Lamb wave velocity based damage
and also showed that the response in the time-frequency detection methods more complex under external loading.
space is noticeably sensitive to the delamination location. The Lamb wave propagation velocity was measured in GFRP
Lee and Staszewski (2002) discussed acousto-ultrasonic (glass fibre-reinforced plastics) and CFRP laminates during
wave propagation modeling techniques, focusing on two- tensile tests and the elastic behavior was studied. These
dimensional wave interactions with defects in metallic struc- authors make two interesting observations: First, that it seems
tures. that measuring the first symmetric Lamb wave (S0 mode)
Ricci et al. (2004) presented a methodology for automatic velocity is a more accurate means of evaluating the elastic-
damage identification and localization in composite struc- ity of laminates than the tensile test. Secondly, they noted
tural components. The damage is inflicted in the form of an that the wave velocity depends on both damage and the lam-
impact, and identification is based on the calculation of a set inate elasticity. Finally, they state that one needs to know
of damage correlation indices obtained from the frequency the laminate stiffness as a function of strain when under exter-
analysis of the signals, obtained from PZT sensors, from the nal loadings, for damage detection purposes.
reference and damaged state of the structure. The main idea Sundararaman et al. (2005) use sparse phased sensor/
behind this approach is that elastic waves, propagating from actuator arrays for online damage detection and localization
the site where they are generated to the location where the sig- on heterogeneous and homogeneous plates and discuss the
nals are sensed, carry information about the portion of the theory of narrowband and broadband beamforming of prop-
structure through which they have travelled. Elastic waves agating plate waves in damaged elastic media. According to
can be very sensitive to small local defects. An indicator of these authors, a beamformer is a spatial-temporal filter that
any change in the propagation characteristics of the wave can be used to ‘look’ in the direction of transmitted signal
can be given by: while eliminating interference that cannot be removed through
d T d
temporal filtering or carrier demodulation alone. Beamform-
{ Hi } { H i } ers consist of sensor arrays that are used to extract directional
D i = 1 – -----------------------------
- (29)
{ Hi } T { Hi } damage signatures. The principle on which this method is
based is equivalent to an acoustic impedance discontinuity
where Di is the damage index evaluated at the sensor loca- in the solid medium, so that damage, in the form of debond-
tion i and {Hi} is the frequency response vector at location i. ing, delamination, fiber breakage, matrix cracking, or voids,
These authors argued that this damage index approach can be can be detected, localized and quantified. Experimental eval-
used for detection and, under certain conditions, characteri- uation of this procedure was carried out on glass/epoxy woven
zation of degradation in aircraft, aerospace and civil structures composite, steel and aluminium plates.
and that the analysis of the waveform signals would provide Other recently published works which make use of Lamb
information on the location and nature of small defects. waves for damage detection and SHM include Beadle et al.
Sohn et al. (2004a) proceeded with the work developed by (2005), Fritzen and Mengelkamp (2005), Giurgiutiu et al.
Sohn et al. (2004b) and proposed a multi-scale structural (2005), Hera et al. (2005), Konstantinidis et al. (2005), Lucero
health monitoring approach for detecting defects in com- and Taha (2005), Nieuwenhuis et al. (2005) and Raghavan
posite structures by combining Lamb wave propagation, and Cesnik (2005).
impedance-based methods and time reversal acoustics using
a common active sensing system for local nondestructive 2.11. Neural Networks
evaluation. These authors mention that the ultrasonic research
community has studied Lamb waves for the nondestructive The use of neural networks in damage detection procedures
evaluation of plates since the 1960s. They also define Lamb and SHM has been motivated by the possibility of the exist-
waves as mechanical waves corresponding to vibration modes ence of different types of damage at several different loca-
of plates with a thickness of the same order of magnitude as tions within the same structure, making damage detection a
the wavelength. In the study presented, the wave attenuation complicated process.
feature, identified using a wavelet based damage index, is Hanagud and Luo (1997**) used three layer feed-forward
used to locate the region of damage (Sohn et al. (2004b)). neural networks to identify two different types of damage in
Damage was introduced by firing a small projectile at dif- GFRPs: delamination and stiffness reduction (due to trans-
ferent locations on a graphite fibre/epoxy matrix square plate verse cracks or impact damage). It was assumed that only one
with varying velocities. It was found that the three methods of these defects existed, and analytical models were built to
studied can be complementary because they can use the same predict the dynamic behavior of the structure considering var-
sensors/actuators, which in this case were commercially ious scenarios of damage. The main neural-network identi-
available thin films with embedded PZT sensors. For instance, fies the type of damage present using a non-linear dynamic

response criterion, which directs the problem to one of two tor is based on the analysis of signal energy components in
sub-networks. Both these sub-networks use FRFs as inputs. discrete time-scale domain. Normally, the wavelet coeffi-
Using the same beam as ‘case study’, Luo and Hanagud cients of the impulse response function (IRF) make different
(1997**) propose the dynamic learning rate steepest descent contributions to the considered system response. Thus, if
(DSD) for training neural networks, with the aim of increas- the excitation force is known, it is easy to normalize the
ing the learning convergence speed relative to the simple response relative to the excitation force. Another normaliza-
steepest descent method. tion possibility is to relate each response to the response in a
Krawczuk et al. (2000**) presented the results of the reference location, thus leading to the transmissibility func-
application of a genetic algorithm and a neural network to tion. According to this author, it is possible (and adequate)
detection and location of delamination in a numerical model to consider the transmissibility functions and their wavelet
of a multi-layered GFRP beam. Two different procedures decompositions in a context of SHM when the ambient
were followed to identify the damage location and size: the excitation is unknown. However the analysis so far is based
first is based on FE model updating and error localization, on the assumption of similar excitations for all the observa-
while the second considers a set of possible damage scenarios tions, which means that further research is needed to evalu-
(including damage type, location and size). These authors ate how this restriction can be surpassed without the need
used an objective function, based on changes in the first for a known excitation.
four natural frequencies and on the damage location assur- Savov and Wenzel (2005) used a wavelet approach in order
ance criterion (DLAC) proposed by Messina et al. (1992). to locate damage in civil structures, based on the system’s
The genetic algorithm converges after a reasonable number acceleration time history responses. A FE model of a three
of generations, although these authors considered that this story shear-resisting frame excited by white noise ground
procedure deserved future development by including more acceleration and a prestressed reinforced concrete test beam
processes that are observed in nature, for example elitism. under impact loading conditions were used as test structures.
The neural network’s performance in detecting the delami- The first level fast wavelet decomposition of the measure-
nated layer location across the thickness was poor, possibly ments for both the undamaged and damaged structures was
because the authors considered a relatively small population applied using the Haar mother wavelet, and the approxima-
of delaminated cases for training. tion and detail coefficients were obtained. Reconstruction of
Hatem et al. (2004) also applied genetic algorithms and the actual system response was accomplished by a cross com-
neural networks to damage detection in CFRP composites. bination of the approximation coefficients of the undamaged
Four types of damage were considered in the model of a structure with the detail coefficients of the actual measure-
cantilevered beam: circular holes (with different diameters ment. The sensitive feature is defined as the standard devia-
and locations), delaminations (with different areas and location of the error between this reconstructed signal and the
tions), linear surface cracks (with different lengths, orientations actual measured signal. This method is based on the premise
and locations), and linear through cracks (with different that the detail coefficients carry information about the local
lengths, orientations and locations). Damage type is identified structural integrity in the time history response at the dam-
by the generalized regression network. A special sub-network aged sites. The advantage of this method is that one does not
is used for each damage type, in an ensemble of five: a gen- need to know the excitation mechanism, since the method is
eralized regression network, a linear network, two back- entirely based on the measured responses. However, damage
propagation networks (with and without regularization) and location is compromised by the sensor mesh refinement.
a radial basis network. These authors state that the general- Extension of this approach to quantification of damage, to
ized regression networks successfully classified damage type, prediction of remaining structural life and to evaluation of
with a success rate ranging from 85% to 98%. After damage more complex structures, varying operational and environ-
is classified, an appropriate neural network or genetic algo- mental conditions and non-linear damage phenomena are
rithm is run to detect the remaining damage parameters, issues to which these authors will give further attention.
namely location, size and, in the case of a crack, orienta- Li et al. (2005) present a methodology for determining
tion. Damage size was predicted with good accuracy, but crack location and size in a beam, taking advantage of wave-
the results on damage position and orientation were not par- let finite element methods (WFEM). The idea of the WFEM
ticularly accurate. is to discretize a body into wavelet finite elements which are
Zheng et al. (2004) combined computational mechanics and interconnected at nodal points on the element boundaries.
neural networks, in this case the backpropagation method, Some of the advantages of WFEM over traditional FE methods
for prediction of delamination in CFRP beams. The neural for modal analysis of crack problems are pointed out. The
network was trained with FE models, which were designed natural frequencies of the beam with various crack locations
assuming various delamination sizes and locations. As inputs, and sizes are obtained, through the use of a WFEM, and used
it used the first five natural frequencies. According to these as features in the damage detection process. The accuracy of
researchers, the neural network correctly predicted delami- the method was tested on the experimental data of a beam
nation size and location within a small error margin. studied by Silva and Gomes (1990*). Finally, these authors
suggest that this procedure could be easily extended to com-
2.12. Time-Frequency Analysis plex structures with multiple cracks.
Zabel (2004, 2005a, b) applied a wavelet-based damage 2.13. Hilbert Transform

indicator to damage detection on reinforced concrete struc-
tures, using several case-studies, including the Z24-bridge Testa (2005) argues that the Hilbert-Huang transform
in Switzerland already mentioned in this work. The indica- (HHT) is very useful for non-stationary, non-zero mean and

non-linear real signals. The HHT makes an Empirical Mode sitivities of the orthogonality conditions of the mode shapes,
Decomposition (EMD) of the time signal into narrow band which allow for the calculation of a damage parameter in each
components with zero mean, in which each component is finite element. This damage parameter is directly related to
called an Intrinsic Mode Function (IMF). Unlike their equiva- the stiffness reduction in the damaged finite element. They
lents in the FFT, these components do not have a specific argue that the proposed method allows for detection and
analytical representation, but each component can be asso- location of multiple damage sites. These authors formulate a
ciated with a physical meaning, i.e., the components are set of M = m(m + 1)/2 linear independent equations with N
related to the mode shapes and existence of damage. unknowns, where m is the number of vibration modes and N
Lin et al. (2005) use the HHT technique on the phase I is the number of elements, based on the orthogonality condi-
1 2
IASC -ASCE benchmark problem for SHM. This benchmark tion sensitivities of the damaged structure. In the cases where
has been developed to simplify the side-by-side comparison M ≥ N, the solution can be efficiently found by a Gauss-
of various analysis techniques for damage identification in Jordan elimination method, whereas if the system is under-
structures on a common basis. The scale-model structure is a determined, i.e., M < N, it is possible to find the N – M fam-
3.6 m high four-story steel frame with 2.5 × 2.5 m cross- ily of solutions using the Singular Value Decomposition tech-
sectional area. Johnson, E. A. et al. (2004) present two analyt- nique (SVD) (Golub and Van Loan (1983)). These authors
ical models of the structure, both finite element based: one is argue that the proposed technique allows for an efficient
a 12-DOF shear-building model, the other a 120-DOF model. quantification of damage over either small or larger areas and
Based on the noisy acceleration data (due to ambient excita- that this methodology is a promising tool for the research
tion), Lin et al. (2005) confirm that HHT is capable of identi- community working on experimental damage identification.
fying the natural frequencies, damping ratios, mode-shapes, More recently, Santos et al. (2004, 2005) presented a
stiffness matrix and damping matrix of a structure with a damage identification technique based on FRFs sensitivities.
reasonable accuracy. Damage has been identified by com- Considering that a single-input unit force is applied to the lth
paring the stiffness of each floor before and after introducing DOF of both the undamaged and damaged structure, they
damage. Simulation results show that the HHT technique show that:
can detect, locate and quantify damage with good accuracy.
d d
Finally, these authors acknowledge that, despite the fact that [ H ( ω ) ] [ ∆Z ( ω ) ] { Hl ( ω ) } = { H l ( ω ) } – { Hl ( ω ) } (30)
normal modes are assumed to exist in the current benchmark
d
model, the technique can also be used on linear structures where [∆Z ( ω )] = [ Z ( ω )] – [Z ( ω )] is the difference between
with complex eigenproperties, as demonstrated by Yang the dynamic stiffness matrices of the damaged and undam-
et al. (2003). aged structures and {Hl(ω)} is the lth column of the mobility
Zhang et al. (2005) presented the implementation of a matrix [H(ω)]. After some mathematical manipulation, includ-
method for nonlinear, nonstationary data processing, namely ing a first-order Taylor expansion, the above equation can be
HHT in traditional vibration-based approaches to characterize rewritten in a compact form, where the solution is found
structural damage. Also, the HHT is compared to the Fou- using a least-squares method. In that formulation, it is implicit
rier transfer function technique in detecting local damage that one knows all DOFs of the damaged structure, which is
with a computer model and experiments in two pile founda- not feasible in practice because rotational DOFs are very
tions of the Trinity River Relief bridge in Texas, USA. The difficult to obtain experimentally, as can be seen, for example,
HHT analysis showed a more significant frequency down- in Montalvão et al. (2004a, b). Consequently, the unknown
shift than Fourier-based approaches for measurements made DOFs of the damaged structure should be determined using
near the damage location. These authors observed one point an expansion of the measured DOFs. For example, Santos et al.
which, although only applicable to very specific structures (2003a, b) propose the use of either static or dynamic expan-
such as bridges, remains of interest: by selecting two or more sions for this purpose. Santos et al. (2004, 2005) observed
similar structural members (e.g., two of four columns with that the mobility matrices should be computed using at least
the same size, cross section, and construction in a bridge), the three times the number of natural frequencies and mode shapes
need for data from a reference state may be avoided if only contained in the frequency range of interest. On the other
one of these members has experienced damage. hand, they observed that the best identification results are
Shi and Law (2005a, b) present a HHT-based technique obtained at lower frequency ranges. Also, for damage of small
for parameter (stiffness and damping) identification in dam- extent, the errors are the main influence in the identification
age detection for, respectively, linear time-varying discrete quality, whereas for large damage incompleteness becomes
dynamic systems and nonlinear systems. Both procedures the most important factor. These authors have also used a
are tested using analytical models in free vibration. Shi and procedure for the weighting and elimination of equations,
Law (2005a) identify the non-linear system using a so-called based on Ren and Beards (1995), once the solution depends
skeleton linear model (SLM), which has similar dynamic not only on the quality of the FRF measurements but also on
behavior to the corresponding nonlinear system. the frequencies and co-ordinates in which the vibration ampli-
tude is largest. Finally, it is argued that this technique
2.14. Principal Component Analysis and Singular presents better results than those obtained when using a tech-
Value Decomposition nique based on modal data sensitivities.
Bernal (2002b) used the Damage Locating Vector (DLV)
Santos et al. (1999, 2000) presented a technique for the approach to locate damage, using as case study the 4-story
identification of damage in laminated composite structures steel frame building of the phase II IASC-ASCE benchmark
using, as case study, a numerical simulation of a rectangular problem. The proposed damage characterization strategy is
plate. The algorithm was formulated by considering the sen- grounded in a cascade methodology that contains the fol-

lowing modules: (i) compression of the measured data and those arising from structural damage. Novelty analysis on
identification (stochastic state-space realization); (ii) extrac- the residual errors provides a statistical indication of damage.
tion of flexibility proportional matrices; (iii) localization of The environmental conditions are assumed to have a linear
damage using the DLV approach and (iv) quantification of effect on the vibration features. The application of this method
damage. The DLV is a theoretically based technique for map- to numerical and experimental data of a wooden bridge
ping changes in flexibility to the spatial distribution of dam- shows that different levels of damage can be correctly assessed
age. According to Bernal (2000**, 2002a), the DLVs are a independently from the environmental effects. This method
set of vectors that have the property of inducing stress fields was extended to the non-linear case by Yan et al. (2005b), in
whose magnitude is zero in the damaged elements. Consider which a two step procedure is conceived, consisting of clus-
a system that can be treated as linear in both the undamaged tering the data space into several regions and then applying
and damaged states, with flexibility matrices FU and FD the PCA in each local region. The local PCA-based damage
(respectively) at m sensor locations, and assume there are a detection method is applied, using vibration data measured
number of load vectors, defined in the sensor coordinates, in-situ over a one-year period, in the SHM of the Z24-bridge
which produce identical deformations at the sensor loca- mentioned previously. During the monitoring period meas-
tions in both the undamaged and damaged states. If the line- urements of the environmental parameters (temperature, wind,
arly independent vectors that satisfy this requirement are humidity, etc.) were recorded, and it was observed that the
collected in a matrix L, and according to the compatibility asphalt layer on the bridge surface plays a different role during
of displacements, one may write: warm and cold periods, thus causing non-linearity. For exam-
ple, in an earlier work, Peeters and De Roeck (2000) estab-
[ F U – F D ]L = [ DF ]L = 0 (31) lished ARX models for the first four modes, in order to
obtain a good correlation between temperatures at selected
where DF is computed from an SVD of the change in the locations and the natural frequencies of each mode, with the
flexibility matrix of the system. Equation (31) can be satis- aim of analyzing different damage scenarios. A comparison
fied either if DF = 0 or if DF is rank deficient and L is a basis has been made by Yan et al. (2004) between the work of
for the null space of DF. Bernal (2002a) shows that the first Peeters and De Roeck (2000) and that of Yan et al. (2005b),
possibility implies that either there is no damage (though in leading to the argument that by using the PCA-based damage
practice no one ever computes DF equal to zero) or that dam- detection method the problem is simplified, as only vibration
age is confined to a region of the structure where the stresses features are needed to determine whether the structure is
are zero for any loading sensor coordinates. On other words, damaged.
the null space of the change in flexibility contains vectors that Vanlanduit et al. (2005) introduced the robust singular value
lead to identical displacements at the sensors locations in both decomposition (RSVD) for damage detection from measure-
the undamaged and damaged states. More developments of ments taken under different operational conditions, i.e., dif-
this issue for practice purposes are shown in Bernal (2000**, ferent excitation levels, geometrical uncertainties and
2002a), such as selecting thresholds for the definition of the surface treatments. This method is based on the SVD of a
null space and the ‘zero stress region’, since in fact the stress matrix [H] = [H1, …, HN], where H1, …, HN are observation
level will never be exactly zero in any element. As a result, features taken from N experimental conditions. These fea-
one may point out two drawbacks of this method: it is not tures can be FRFs, response spectra or estimated modal
sensitive to damage which does not introduce changes in the parameters. The RSVD computes the SVD of the intact
flexibility at the measurement points, and identification may structure from a set of observations of both intact and dam-
be poor if the damage region is too large when compared to aged structural states, possibly obtained under different con-
the number of sensors. However, it is worth mentioning that ditions. The RSVD performance is then compared to the
the authors claim that (i) it has the ability to detect both sin- classical least-squares SVD and to the iterative SVD (ISVD).
gle and multiple damage scenarios, (ii) it can operate with The experimental results show that the classical least-squares
an arbitrary number of sensors without reference to DOF SVD is not able to correctly classify damage because it gives
expansion or reduction strategies and (iii) the DLV vectors an incorrect decomposition when both damaged and undam-
are computed strictly from the measured data without refer- aged measurements are used to compute the subspace. This
ence to a mathematical model of the system. Also, this occurs because it is assumed that it is not known which
method is equally applicable to static measurements. Finally, observations are the healthy ones. If a set of observations
since in Bernal (2002b) the inputs could not be measured, he belonging to the undamaged case is previously known, the
used flexibility proportional matrices obtained following authors agree that the SVD performance will be comparable
the procedure introduced by Bernal and Gunes (2002). This to that of the RSVD. It was also shown that the ISVD slightly
latter is an extension of the DLV damage localization tech- improved the classification results, though it was not capable
nique to the output-only case. of identifying the smallest damage case considered. Finally,
Yan et al. (2005a) proposed a damage detection method these researchers claim that the RSVD has been proven to
for SHM under varying operational and environmental condi- provide the most reliable results.
tions based on Principal Component Analysis (PCA), also
known as Karhunen-Loève transform or proper orthogonal 2.15. Instrumentation
decomposition. One advantage of the use of PCA is that it
is not necessary to measure the environmental parameters Lin (1999**) developed a printed circuit film layer of
because they are taken into account as embedded variables. piezoceramic materials which can be embedded into com-
This method relies on the principle that the measured fea- posite structures during the production process. This film can
ture changes due to environmental causes are different from support temperatures above the curing temperature for most

composites, and Lin (1999**) argues that it has little or no Sundaresan et al. (2004) studied a structural health moni-
effect on the mechanical properties of composite structures. toring system that employs an embedded AE (continuous)
Sundaresan et al. (2001a) discuss the use of a scanning sensor, developed by Sundaresan et al. (2001b). One of the
Laser Doppler vibrometer (SLDV) and piezoceramic mate- advantages of this continuous sensor is that it avoids separate
rials for distributed sensing, with the purpose of evaluating cabling and individual support electronics. They monitored
the application of widely distributed sensors for wide area fatigue crack growth in a fiberglass laminate specimen with
health monitoring of composite materials. The SLDV is two circular notches. They stated that the performance of the
used for field tests as a rapid non-contact detector for larger continuous sensor was superior to the traditional single node
hidden damage, and the highly distributed embedded smart sensors, and also used wavelet maps to classify the AE sig-
sensors are investigated for in situ real-time monitoring of nals.
composite structures. Two methods of damage detection using Nichols et al. (2005) reported recent advances in the fields
the laser vibrometer were presented. The first, called bound- of sensing and signal processing, with the aim of damage
ary effect detection (BED), performs post processing of the detection. A nine fibre Bragg grating strain sensor network
operational deflection shapes (ODS), and, according to the was used to measure the vibrational responses of a thin steel
authors, is somewhat akin to the wavelet method, thus allow- plate subjected to successive growing saw-cut damage lev-
ing for small damage detection in the low frequency range. els. The relationship between a reference (or undamaged)
The second method, called operational deflection pattern rec- state and the actual (or damaged) state can be mathemati-
ognition (ODPR), is able to identify faults by evaluating cally established when the structure is interrogated with a
irregularities in the vibration response patterns, avoiding the deterministic chaotic signal. This method relies on the prin-
need for reference pre-damage data. ciple that this function loses differentiability when damage
Fritzen et al. (2002) designed an intelligent self-diagnosis exists. Thus, the use of the local Holder exponent is proposed
system to detect the existence of delaminations in plate-like as a means of quantifying the differentiability of this func-
CFRP structures. The plate is instrumented with an array of tion relating the undamaged to the damaged state responses.
four piezoceramic sensors and is excited with a random signal An algorithm to determine these quantities from time series
by means of a piezoceramic actuator. Different delamination data is also described.
severities were introduced by impacts at different velocities. Park, J. M. et al. (2005) evaluated the performance of dif-
As a damage indicator, they used the subspace-based fault ferent sensors for damage detection in glass fibre/epoxy com-
detection algorithm, based on the work presented by Basse- posites through acoustic emission measurements. The sensors
ville et al. (2000). To determine the eigenfrequencies, damping analyzed were: PZT (piezoelectric lead-zirconate-titanate),
and mode shape changes, the stochastic subspace identifica- PVDF (polyvinylidene fluoride) and P(VDF-TrFE) (poly
tion method was used, following the approach proposed by (vinylidene fluoride-trifluoroethylene)). The PVDF sensors
Peeters and De Roeck (1999). They also performed a statis- wer tested both surface mounted and embedded on the com-
tical 1-test to assess whether the recent data fit the baseline posite structure, whereas the PZT unit was only surface
data. mounted. It was observed that the embedded sensors are
Deraemaeker and Preumont (2004a, b) presented an more sensitive to damage.
approach to vibration based damage detection based on Motivated by recent research advances, technological
the use of a large number of sensors to which a programmable changes and commercial activities in sensors and acquisition
linear combiner is attached, using a simply supported beam systems for monitoring, Bonfiglioli et al. (2005) studied the
as a test case. The linear combiner is programmed to work as measurement errors of electric strain gauges (ESG) on FRP.
a modal filter, i.e., the converted time-domain to frequency They extended a theoretical deterministic approach devel-
output content of the modal filter is used as the damage oped in previous works to a probabilistic approach, with the
detection feature. First, these authors show that if local dam- aim of performing a sensitivity analysis of the variables that
age exists, then peaks will appear in the FRF of the modal may result in errors on strain measurements. The experimen-
filter. If temperature changes are registered instead, the modal tal tests on several specimens made of different composite
filter is shifted but its shape remains unchanged. Thus, the materials allowed some conclusions to be drawn with respect
proposed approach allows the user to distinguish between to the thickness of the resin, the ESG length, deviation angle
local damage and global changes (due to the surrounding and the sensors bonding glue. It was observed that the ESG
environment, for example). For the case where the excitation length is the most important factor and that short strain
forces are unknown, it is suggested that the square root of the gauges are to be avoided when the surface of the FRP mate-
power spectral densities (PSD) be used instead of the FRFs, rial is very rough. One interesting conclusion stated by these
because the square root of the PSD has, in principle, the authors is that the proposed model allows evaluation of the
magnitude of the transfer function if the excitation force is Young’s modulus of the adhesive, which is very difficult to
white noise. It was demonstrated that this method allows the determine experimentally.
detection of low levels of damage, as well as being able to Qing et al. (2005) developed a hybrid piezoelectric/fibre
distinguish damage from temperature effects. optic (HyPFO) diagnostic system for quick non-destructive
Lynch et al. (2004) developed an active wireless sensing evaluation and long term health monitoring of aerospace
unit to command a PZT sensor surface mounted on an alumi- vehicles and structures. Piezoelectrics are used as actuators
num plate. Low energy Lamb waves are introduced in the and fibre gratings as sensors. One of the major advantages
plate and the unit collects the signals measured by a second mentioned is the excellent actuator/sensor decoupling, since
PZT sensor placed elsewhere on the plate surface. The poten- the transducers use different mechanisms for signal trans-
tial of this unit to locally perform system identification anal- mission (one is electrical and the other optical). Another
ysis is illustrated by calculating ARX models. advantage is that these transducers can be used for a range of

measurements; the fibre-optic sensors can be used for tem- 2.16. Other Methods
perature sensing, while the piezoelectric system can be used
for hydrogen sensing, and both can be used for acoustic Afolabi (1987*) observed that anti-resonance shifts could
emission measurements. be used to detect and locate damage, and showed the depend-
Wu et al. (2005) studied the correlation between the ence of anti-resonance frequencies on the measurement points.
mechanical and electrical properties of concrete beams Consequently, he observed that as the point of measurement
strengthened with hybrid carbon fibre-reinforced polymer gets closer to the location of the defect, fewer anti-resonances
(HCFRP) sheets. These have the simultaneous aims of rein- are shifted from their original values, until one reaches the
forcement and providing built-in self-structural health mon- location of the defect, when all the anti-resonances are as
itoring through the electrical conductivity and piezoresistivity they were in the undamaged state. Although this is an inter-
of the carbon fibres. The HCFRPs have self-diagnosis capa- esting observation for the beam model and for the simulation
bilities due to the relationship between strain/load and elec- used, one should not forget that stiffness relates to at least
trical resistance. These authors observed that different damage two co-ordinates, which means that the interpretation of the
stages, failure processes and ultimate failure modes of the results obtained in real structures may be more complicated.
HCFRP reinforced concrete structures can be identified Also, he only formulated relationships for direct FRF meas-
through the electrical resistance changes. urements, since this method may not work for transfer func-
Castellini et al. (2005) produced software for data acquisi- tions. The transfer FRF relationships, which are very simple
tion, control of a scanning Laser Doppler vibrometer (SLDV) to derive following the same lines as Afolabi (1987*), can be
and automatic measurement for the purpose of detecting found in Montalvão (2003). More recently Wahl et al. (1999)
delamination in composite plates, using as case study a dela- presented a work in which they discuss the significance of
minated composite panel made from 10 layers of aluminum antiresonances in experimental structural analysis for lightly
sheets and glass fibre-reinforced epoxy prepreg (GLARE) for damped linear systems. d’Ambrogio and Fregolent (2000)
Airbus Deutschland aeronautical applications. These authors observed that the distribution of antiresonances may be sig-
point out that the SLDV has several advantages: it does not nificantly altered by small changes in the structural model.
interfere with the structure, it has a high spatial resolution They use an updating technique that includes antiresonances
and sensitivity, it is adequate for high frequency analysis and in the definition of the output residual, showing that the use
it is suitable for use in operational conditions. The software of antiresonances extracted from point FRFs allows for
uses a multivariate projection method, Principal Component robust model updating procedures, mainly because antireso-
Analysis (PCA), to extract and display the systematic variation nances can be identified from experimental FRFs with much
in a data matrix. Also, statistical processing aids are used to less error than mode shapes. Bamnios et al. (2002) moni-
pre-process the measured data, reducing the amount of infor- tored the change of the first antiresonance as a function of
mation. Consequently, multivariate analysis methods, such the measuring location along the beam and proposed a pre-
as PCA, offer the possibility of eliminating systematic effects diction procedure for transverse open crack localization in
like noise. These researchers are continuing to search for beams under bending vibrations. Dilena and Morassi (2004)
optimal settings for the measurements, so that the number of used antiresonances to avoid the non-uniqueness of the
measurement points is reduced. Research is also continuing damage location problem, which may occur in symmetrical
in the field of composite materials excitation, because it is beams with a single open crack when only frequency data is
important that the excitation energy is constant and high for employed.
the entire frequency band. Zimmerman et al. (1995a*) discuss the implementation of
In a survey of current smart sensor technologies for monitor- the minimum rank perturbation theory (MRPT) based tech-
ing, Spencer et al. (2004) show that many efforts have been niques to identify both the location and extent of damage,
made to use remotely operated wireless systems, especially using data from the NASA 8-bay dynamic scale model truss
for damage detection in civil structures. Since traditional test-bed. The evaluation of damage location was included in
impedance-based methods require the use of bulky and expen- this study, as well as the selection of the number of vibration
sive analyzers, Grisso et al. (2005) present the initial steps in modes to measure, eigenvector and damage vector filtering
the development of an automatic wireless system for imped- to minimize the effects of measurement noise, filtering of
ance-based damage detection, which they claim is cheaper, dynamic residual decomposition among various property
more eco-efficient and has smaller dimensions. The final pur- matrices (mass, damping and/or stiffness) when multiple prop-
pose is to develop a sensor that can be permanently attached erty matrices are being updated and rank-estimation algo-
to the structure and provide meaningful data concerning struc- rithms. Zimmerman et al. (1995b*) extended the MRPT theory
tural integrity. Other features include utilizing energy from to determine matrix perturbations directly from FRF data,
the ambient environment, exciting the host structure with a discussing the benefits of this formulation. They concluded
high-frequency excitation, analyzing the results and wire- that the MRPT algorithm implementation using FRFs sim-
lessly providing the status of the structure to an end user. The plifies the analysis by avoiding the need to perform modal
hardware development of a prototype version of this sensor parameter identification. Another interesting conclusion is that,
is described in detail. This sensor has been built and devel- contrary to what would be intuitively expected, the regions
oped within the scope of the MEMS (micro electro-mechan- away from resonances and anti-resonances are those con-
ical systems) augmented structural sensor (MASSpatch) joint- taining the richest data. More recently, Zimmerman (2005)
venture. investigated the effect of measurement noise on damage
detection performance and sensitivity of the MRPT and
resulting stiffness perturbation matrices. It was concluded
(by performing a Monte-Carlo simulation) that the use of

linear sensitivity theory for estimation of the variance and

standard deviation of each element in the damage vector and
stiffness perturbation matrix is accurate. These standard devi-
ations were used to establish a threshold for damage existence
and, after that, for damage location and extent assessment.
Worden et al. (2000**) presented a study of a statistical
method for damage detection using outlier analysis. These
authors claim that the method not only allows for novelty
detection (deviation from normal condition), but also sug-
gests the optimal reduction of the dimension of the data set
without compromising the diagnosis. A discordant outlier in
a data set is an observation that appears inconsistent with the
rest of the data. The discordance test is based on the Maha- Figure 1. Schematic concept of time reversal based dam-
lanobis squared distance for multivariate data sets, where the age identification (in Sohn et al. (2005)).
threshold value to classify whether an observation is an out-
lier or not is determined using a Monte Carlo method. The
method was tested in four different situations: transmissibil-
ity data in a simulated 3 DOF system, gearbox experimental which does not rely on any past baseline signals for assess-
data, composite plate Lamb wave experimental data, and ment of damage in composite panels. The test setup consists
ball-bearing experimental data. In conclusion, they point out of a square CFRP plate with surface mounted PZT patches,
that for the 3-DOF numerical model the method was unable to which delamination is introduced by impact damage with
to detect any of the 1% stiffness reduction cases as outliers, varying velocities. A time reversal concept of modern acous-
though for higher reduction cases the method correctly iden- tics is adapted to Lamb wave guided propagation. According
tified the outliers. However, it should be noted that some to the time reversal concept, an input signal can be recon-
simplifications were made. Firstly, it was considered that structed at an excitation point A if an output signal recorded
only a single outlier is present, which considerably simplifies at point B is reemitted to the original source point A after
the problem. Also, when choosing the Mahalanobis squared being reversed in a time domain. This process is referred to
distance, it is implicit that the normal condition set has Gaus- as the time reversibility of waves, and can be better under-
sian statistics. In the case of, for example, multi-modal dis- stood with the aid of Figure 1.
tribution, these authors suggest other techniques such as the Damage detection is based on the fact that if a non-linear
Kernel density estimation. defect exists in the wave propagation path, time reversibility
Gutschmidt and Cornwell (2001) presented a false posi- will no longer hold. A damage index based on the comparison
tive damage indication method (FPDIM) and applied it to between the original input waveform and the reconstructed
experimental data for a plate that was progressively dam- signal is used to classify damage. A zero value means that time
aged. The purpose was to establish statistical bounds in order reversibility of Lamb waves is preserved and thus no non-linear
to distinguish damage from the natural variability associated defects are present. Once the damage index value exceeds a
with any measurement. The statistical bounds are defined threshold value, the state of the system is defined as damaged
using several sets of data from the baseline structure. The in a conventional sense. These authors employ the use of a con-
FPDIM was applied to two damage identification techniques secutive outlier analysis (Barnett and Lewis (1994)) to establish
that do not require a prior numerical model: the flexibility and the threshold value without requiring past reference data and
the strain-energy methods. Using both of these methods, stand- to address the issue of multiple outliers. By using this approach,
ard deviations σDI of the damage indices DIi in each element these authors successfully located and quantified damage in
i for the undamaged structure were determined. A prediction the test plate without relying on prior baseline signals.
interval is then calculated, within which it is expected that According to Ibrahim et al. (1997) and to Li et al. (2004) the
the next data point will fall. Any point outside that interval introduction of the random decrement (RD) technique is attrib-
represents a certain probability of failure. Eight damage cases uted to Cole (1968) (1973). Ibrahim et al. (1997) proposed the
were considered, and tests were carried out under the same vector random decrement (VRD) technique. Later, Asmus-
experimental and environmental conditions. It was observed sen et al. (1999) showed that the VRD can be considered a
that application of the FPDIM to results from the strain-energy generalization of the RD. Rodrigues et al. (2004) explored the
method eliminated false-positives. When it was applied to idea of estimating the spectral densities as the Fourier trans-
the flexibility method, however, false-positives still occurred; form of the RD functions for the application of frequency
i.e., the flexibility method successfully determined damage, domain output-only modal identification methods. Li et al.
but was unsuccessful in locating it. (2004) investigated the potential of the RD technique for
Within the statistical pattern recognition paradigm estab- damage detection in composite beams by introducing differ-
lished by Sohn et al. (2003), damage is in most cases identi- ent levels of delamination at the midplane of the beam at
fied by comparison between signals corresponding to two various locations. The initial formulation consists of obtain-
different structural states, one of them often corresponding ing the free-response signature of the beam by processing its
to the undamaged state. Still, the operational and environ- random response based on the global average. This can be
mental variability, if not considered, may cause the unpre- achieved by summing up many records of random responses
dictable occurrence of either false-negatives or false-positives. in previously defined periods of time with appropriate crite-
In order to address this issue in the context of continuous ria. This summation will result in a random decrement signa-
online monitoring, Sohn et al. (2005) proposed a technique ture which minimizes the random component of the response.

Each record can be seen as representing a linear system tion of the COMAC instead of normal modes. The theoretical
response composed of the overlapping of a step response, an backgrounds of black box modeling and subspace identifica-
impulse response, a random response and random noise. The tion are described, as well as some of their limitations. Also,
step and impulse responses occur due to the initial displacement it is stated that the advantage of subspace identification is the
and initial velocity, respectively. When the various records are physical weighting of the vibration modes with respect to
summed up, the random components, including noise, and the excitations.
impulse responses, average to zero, thus leaving only the step Taylor and Zimmerman (2005) presented a non-model-
response of the system. It is shown that the RD signature is based method for detecting and locating structural damage
affected by both the delamination dimension and the sensor (by evaluating linear stiffness changes) using load dependent
location. This may mean that damage detection can only be Ritz vectors (Wilson et al. (1982)). They state that Ritz vec-
carried out successfully if a large database of RD signatures tors are more sensitive to localized structural damage than
is created from predicted defects or by the use of pattern rec- mode shape vectors, which is an advantage when one cannot
ognition algorithms, such as neural networks. Li et al. (2004) deal with a FEM of the structure (which can often occur in large
also note that the generation of such a database requires accu- civil structures) and because of the incompleteness problem.
rate modeling with appropriate treatment of structural damp- Also changes in mode shapes or shifts in the natural frequen-
ing, which they say plays an important role in the vibration cies are often poor damage locators without the application
of composite materials. of some model updating technique. Experiments were carried
Coppotelli et al. (2004) compared two different experimen- out on a steel bay welded frame structure and showed that
tal techniques for damage identification in composite plates. Ritz vectors performed better than mode shape vectors in
One approach is based on measurements of the dynamic dis- detecting low severity localized damage without the use of a
placement of the plate surface, which are evaluated through FEM.
the speckle fringe patterns resulting from electronic speckle Thien et al. (2005) presented some guidelines for the devel-
pattern interferometry (ESPI), a high resolution holographic opment of a low-cost active-sensing based diagnostic system
technique explored by Caponero et al. (2000) exploiting both for pipeline structures. Integration of impedance-based and
interference and diffraction in optics. The other approach Lamb wave propagation methods in a common SHM is
is based on experimental modal analysis techniques for the investigated. The impedance-based methods are used to detect
extraction of eigenproperties following the complex exponent- and locate damage at the pipeline connection joints, while
based method (Ewins (1984), Maia and Silva (1997a)). The Lamb wave propagation methods are used to identify cracks
damage identification procedure used for the first procedure and corrosion along the surface and through the thickness of
is based on the image correlation function (ICF), which is the piping. Both techniques make use of the electromechan-
the square root of the FDAC in the special case where the ical coupling effects of piezoelectric-based active sensors,
vectors involved are real and positive, whereas for the sec- which in this case are small and non-intrusive macro-fibre
ond case, they use the MAC. One of the two test-structures composite patches. The procedure requires measurements in
used to evaluate the techniques was a sandwich plate structure a reference state prior to damage. Minor defects can be
made from two carbon fibre skins and an aluminium honey- detected through operation at high frequencies.
comb core. This plate belongs to the Gamma-ray Large Area
Space Telescope (GLAST), an international space mission 3. Prognosis
intended to collect gamma-ray data from the universe. Its
launch is expected in 2007. Prognosis – or the prediction of a system’s lifetime, which
Peil et al. (2004) acknowledge that, in order to predict the corresponds to the last level of the classification of damage
remaining lifetime of a structure, the existing accumulated detection methods introduced by Rytter (1993*,**) – is a
damage must be first assessed. To accomplish this task on matter that is traditionally approached in terms of fracture
civil engineering structures such as bridges, they use either a mechanics and fatigue. However, considering the benefits
theoretical model of past actions, or an experimental method that such technology may bring to the security, economics and
using acoustic emissions to assess damage. In the latter case, resource management fields, the scientific community that
they take into account the entire transient signal, based on deals with vibration analysis is now beginning to take some
the assumption that different material damage processes lead interest in this area. Farrar et al. (2003**) published a report
to different transient signal forms, and the signals are assigned whose purposes were to define the technology involved in
to certain damage state classes. prognosis. This work approaches (among other topics) the
Lenzen (2005) presents a procedure for damage detection following: (i) a summary of the technologies and tools neces-
and localization using the deterministic and stochastic subspace sary to solve the problem of predicting a system’s lifetime;
method for black box identification, which was tested on a (ii) a summary of the state of the art in damage prognosis;
steel frame and two bridges. If it is not practical to set up (iii) a generic approach to the problem, considering, however,
physical equations for a system, black box modeling can be that until that time all the prognosis solutions were directed at
used to describe the input-output relations of the system, in specific cases; (iv) the limitations inherent in such technolo-
accordance with the theory of systems and the principle of gies; (v) a brief description of several applications in dam-
cause and effect, in which systems are generally formulated age prognosis.
as transfer functions. The black box models of both the intact
and damaged systems, which have to be identified first, are 4. Final Remarks
compared and used to detect and localize system variations.
This comparison is made by determining the so-called dynamic Unfortunately, the more than two hundred works referred
influence coefficients, which are then used for the evalua- to in this article represent only a very small fraction of the

published works available on the topic. Other works are not Notes
mentioned for several different reasons, most often because
1. International Association of Structural Control.
they use approaches that were not covered in this article. For 2. American Society of Civil Engineers.
example, there are a number of recent articles in which the
excitation is applied using mathematical models of chaotic
time series (Chang et al. (2004), Fasel and Todd (2005) and References
Olson et al. (2005)), techniques dealing with model updating Adams, R. D., Cawley, P., Pye, C. J. and Stone, B. J, 1978*, “A Vibration
(Yu et al. (2005) and De Roeck et al. (2005)) and other Technique for Non-Destructively assessing the Integrity of Struc-
methods and applications (Cioara and Alampalli (2004), Liu tures”, Journal of Mechanical Engineering Science, Vol. 20, 93–100.
Afolabi, D, 1987*, “An Anti-Resonance Technique for Detecting Struc-
and Chelidze (2005), Moura (2005), Nejad et al. (2005) and th
tural Damage”, In Proceedings of the 5 International Modal Analy-
Law and Lu (2005)). sis Conference (IMAC V), London, England, UK, 491–495.
Currently, an immense number of techniques exist for the Aktan, A. E., Catbas, F. N., Grimmelsman, K. A. and Tsikos, C. J, 2000**,
identification and location of damage. Because all the tech- “Issues in Infrastructure Health Monitoring for Management”, Jour-
nal of Engineering Mechanics, Vol. 126, 711–724.
niques have their own advantages and disadvantages, there Alampalli, S, 2000, “Effects of Testing, Analysis, Damage and Environ-
is no general algorithm that allows the resolution of all kinds ment on Modal Parameters”, Mechanical Systems and Signal
of problems in all kinds of structures. Every technique tends Processing, Vol. 14, 63–74.
to have damage related sensitivities; i.e., a very sensitive Allemang, R. J. and Brown, D. L, 1982, “A Correlation Coefficient for
st
Modal Vector Analysis”, In Proceedings of the 1 International Modal
technique may produce false-positives, while a less sensi- Analysis Conference (IMAC I), Orlando, Florida, USA, 110–116.
tive technique may lead to false-negatives, the latter case Alleyne, D. N. and Cawley, P, 1992, “The Interaction of Lamb Waves with
being more problematic. Generally, only damage above a Defects”, IEEE Transactions on Ultrasonics, Ferroelectrics and
certain size can be detected. The quantification of damage Frequency Control, Vol. 39, 381–397.
Andersen, P., Kirkegaard, P. H. and Brincker, R, 1997, “Filtering Out
and prediction of the remaining lifetime are beyond any Environmental Effects in Damage Detection of Civil Engineering
doubt the most difficult issues, the latter deserving particu- th
Structures”, In Proceedings of the 15 International Modal Analysis
lar attention. Conference (IMAC XV), Orlando, Florida, USA, 905–911.
During the last few years a considerable investment in Asmussen, J. C., Brincker, R. and Ibrahim, S. R, 1999, “Statistical Theory
of the Vector Random Decrement Technique”, Journal of Sound and
instrumentation techniques has been observed, especially in Vibration, Vol. 226, 329–344.
those using smart technologies for monitoring, as well as in Bamnios, Y., Douka, E. and Trochidis, A, 2002, “Crack Identification in
the use of high frequency waves, such as Lamb waves, for Beam Structures using Mechanical Impedance”, Journal of Sound
the detection of localized damage. and Vibration, Vol. 256, 287–297.
Barnett, V. and Lewis, T, 1994, Outliers in Statistical Data, John Wiley &
Most of the traditional methods are based on the fact that Sons.
damage leads to appreciable reduction in the rigidity of a Basseville, M., Abdelghani, M. and Benveniste, A, 2000, “Subspace-based
structural element. However, one must not forget that a Fault Detection Algorithms for Vibration Monitoring”, Automatica,
reduction in stiffness does not necessarily mean that there is Vol. 36, 101–109.
Battipede, M., Ruotolo, R. and Surace, C, 2001, “Damage Detection of
a decrease in strength of the structural materials. This also Plate-like Structures”, In Proceedings of the 4th International Con-
represents one of the difficulties that prognosis has to deal ference on Damage Assessment of Structures (DAMAS 2001), Car-
with in the future. Also, some authors argue that in some diff, Wales, UK, 27–34.
specific applications, combining static and dynamic experi- Beadle, B. M., Hurlebaus, S., Jacobs, L. J. and Gaul, L, 2005, “Detection
and Localization of Small Notches in Plates using Lamb Waves”, In
mental procedures is a good idea, since static and dynamic rd
Proceedings of the 23 International Modal Analysis Conference
stiffness are different measurands. In structures made of com- (IMAX XXIII), Orlando, Florida, USA, paper no. 96.
posite materials there seems to be a tendency toward the use Bernal, D, 2000**, “Extracting Flexibility Matrices from State-Space
of damping as the damage feature, since damping variations – Realizations”, In Proceedings of the European COST F3 Conference
on System Identification and Structural Health Monitoring, Madrid,
associated with the dissipated energy – seem to be more sen- Spain, 127–135.
sitive to damage than the rigidity variations, especially where Bernal, D, 2002a, “Load Vectors for Damage Localization”, Journal of
delamination is concerned. However, considering the more Engineering Mechanics, Vol. 128, 7–14.
simple case of a crack, damping will only increase when the Bernal, D, 2002b, “Phase II of the ASCE Benchmark Study – Performance
st
of the DLV Approach”, In Proceedings of the 1 European Work-
crack is opening, since that is the point at which there is most shop on Structural Health Monitoring, Ecole Noemale Supérieure,
interaction between the faces of the crack. When the crack is Cachan, Paris, France, 161–168.
fully opened (or fully closed) there is no friction, and thus Bernal, D. and Gunes, B, 2002, “Damage Localization in Output-Only Sys-
damping remains at a low value. The evaluation of non-line- tems: A Flexibility Based Approach”, In Proceedings of the 20th
International Modal Analysis Conference (IMAC XX), Los Angeles,
arity also seems to be a promising technique for damage California, USA, 1185–1191.
detection in composite materials, since the delamination Berthelot, J. M. and Sefrani, Y, 2004, “Damping Analysis of Unidirec-
may cause non-linear effects through friction. tional Glass and Kevlar Fibre Composites”, Composites Science and
The use of statistical parameters has also deserved Technology, Vol. 64, 1261–1278.
Betti, R, 2005, “Dynamic Methods for Damage Detection in Structures”,
considerable attention, given the experimental errors and Advanced Course on Dynamic Methods for Damage Detection in
incompleteness, as well as the environmental and opera- Structures, CISM (Centre International des Sciences Méchaniques),
tional conditions, which may disguise the consequent damage Udine, Italy.
alterations. The emergence of statistical pattern recognition Boltezar, M., Strancar, B. and Kuhelj, A, 1998, “Identification of Trans-
verse Crack Location in Flexural Vibrations of Free-Free Beams”,
techniques that allow a reduction in the number of sensors Journal of Sound and Vibration, Vol. 211, 729–734.
needed still requires considerable further effort. Bonfiglioli, B., Strauss, A., Pascale, G. and Bergmeister, K, 2005, “Basic
Study of Monitoring on Fibre Reinforced Polymers: Theoretical and
Experimental Study”, Smart Materials and Structures, Vol. 14, 12–
23.

Brincker, R. and Andersen, P, 2003, “A Way of getting Scaled Mode ence on Modal Analysis (IOMAC 2005), Copenhagen, Denmark,
st
Shapes in Output Only Modal Testing”, In Proceedings of the 21 115–126.
International Modal Analysis Conference (IMAC XXI), Kissimmee, Deraemaeker, A. and Preumont, A, 2004a, “Modal Filters for Vibration
Florida, USA, paper no. 141. Based Damage Detection”, In Proceedings of the 2nd European
Brincker, R., Andersen, P., Kirkegaard, P. H. and Ulfkjaer, J. P, 1995a*, Workshop on Structural Health Monitoring, Forum am Deutschen
“Damage Detection in Laboratory Concrete Beams”, In Proceedings Museum, Munich, Germany, 844–851.
of the 13th International Modal Analysis Conference (IMAC XIII), Deraemaeker, A. and Preumont, A, 2004b, “Modal Filters for Vibration
Nashville, Tennessee, USA, 668–674. Based Damage Detection”, In Proceedings of the International Con-
Brincker, R., Kirkegaard, P. H., Andersen, P. and Martínez, M. E, 1995b*, ference on Noise and Vibration Engineering (ISMA 2004), Leuven,
“Damage Detection in an Offshore Structure”, In Proceedings of the Belgium, 501–514.
th
13 International Modal Analysis Conference (IMAC XIII), Nash- Dilena, M. and Morassi, A, 2004, “The use of Anti-Resonances for Crack
ville, Tennessee, USA, 661–667. Detection in Beams”, Journal of Sound and Vibration, Vol. 276,
Brincker, R., Ventura, C. E. and Andersen, P, 2003, “Why Output-Only 195–214.
Modal Testing is a Desirable Tool for a Wide Range of Practical Doebling, S. W., Farrar, C. F., Prime, M. B. and Shevits, D. W, 1996**,
st
Applications”, In Proceedings of the 21 International Modal Analy- “Damage Identification and Health Monitoring of Structural and
sis Conference (IMAC XXI), Kissimmee, Florida, USA, paper no. Mechanical Systems from Changes in their Vibration Characteris-
265. tics: A Literature Review”, Los Alamos National Laboratory, USA.
Caponero, M. A., Pasqua, P., Paolozzi, A. and Peroni, I, 2000, “Use of Doebling, S. W., Farrar, C. R. and Prime, M. B, 1998**, “A Summary
Holographic Interferometry and Electronic Speckle Pattern Interfer- Review of Vibration-Based Damage Identification Methods”, Shock
ometry for Measurements of Dynamic Displacements”, Mechanical and Vibration Digest, Vol. 30, 91–105.
Systems and Signal Processing, Vol. 14, 49–62. Ewins, D. J, 1984, Modal Testing: Theory and Practice, Research Studies
Carden, E. P. and Fanning, P, 2004, “Vibration Based Conditioning Moni- Press, Ltd.
toring: A Review”, Structural Health Monitoring, Vol. 3, 355–377. Farrar, C. R., Baker, W. E., Bell, T. M., Cone, K. M., Darling, T. W.,
,
Carvalho, R. J. S, 2003, “Resistência Residual de um Compósito de Matriz Duffey, T. A., Eklund, A. and Migliori, A, 1994* **, “Dynamic
Epóxídica reforçado por Fibras de Carbono após Reparação”, PhD Characterization and Damage Detection in the I-40 Bridge over the
Thesis, Instituto Superior Técnico, Technical University of Lisbon, Rio Grande”, Los Alamos National Laboratory, USA.
Portugal (in Portuguese). Farrar, C. R., Doebling, S. W. and Nix, D. A, 2001, “Vibration-based
Castellini, P., Willemann, D. P. and Revel, G. M, 2005, “Application of a Structural Damage Identification”, Philosophical Transactions of the
Laser Doppler Vibrometry for Structural Diagnostics on Composite Royal Society of London Series A-Mathematical Physical and Engi-
rd
Panels”, In Proceedings of the 23 International Modal Analysis neering Sciences, Vol. 359, 131–149.
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 371. Farrar, C. R., Sohn, H., Hemez, F. M., Anderson, M. C., Bement, M. T.,
Cawley, P, 1997**, “Long Range Inspection of Structures Using Low Fre- Cornwell, P. J., Doebling, S. W., Lieven, N., Robertson, A. N. and
nd
quency Ultrasound”, In Proceedings of the 2 International Confer- Schultze, J. F, 2003**, “Damage Prognosis: Current Status and
ence on Structural Damage Assessment using Advanced Signal Future Needs”, Los Alamos National Laboratory, USA.
Processing Procedures (DAMAS 97), Sheffield, England, UK, 1–17. Farrar, C. R., Sohn, H. and Robertson, A. N, 2004, “Applications of Non-
Cawley, P. and Adams, R. D, 1979*, “The Location of Defects in Struc- linear System Identification to Structural Health Monitoring”, In
nd
tures from Measurements of Natural Frequencies”, Journal of Strain Proceedings of the 2 European Workshop on Structural Health
Analysis for Engineering Design, Vol. 14, 49–57. Monitoring, Forum am Deutshcen Museum, Munich, Germany, 59–
Chang, L. Y., Erickson, K. A., Lee, K. G. and Todd, M. D, 2004, “Struc- 67.
tural Damage Detection using Chaotic Time Series Excitation”, In Fasel, T. R. and Todd, M. D, 2005, “Effects of Data Acquisition and Gen-
Proceedings of the 22nd International Modal Analysis Conference eration on Using Upconverted Chaotic Waves for Active Structural
rd
(IMAC XXII), Dearborn, Michigan, USA, paper no. 221. Health Monitoring”, In Proceedings of the 23 International Modal
Choi, S. and Stubbs, N, 2004, “Damage Identification in Structures using Analysis Conference (IMAC XXIII), Orlando, Florida, USA, paper
the Time-Domain Response”, Journal of Sound and Vibration, Vol. no. 210.
275, 577–590. Fox, C. H. J, 1992, “The Location of Defects in Structures: A Comparison
Cioara, T. G. and Alampalli, S, 2004, “Considerations on Damage Detec- on the use of Natural Frequency and Mode Shape Data”, In Proceed-
tion of a Structure using Vibration Test Data”, In Proceedings of the ings of the 10th International Modal Analysis Conference (IMAC X),
nd
22 International Modal Analysis Conference (IMAC XXII), Dear- San Diego, California, USA, 522–528.
born, Michigan, USA, paper no. 146. Fritzen, C. P. and Mengelkamp, G, 2005, “In-Situ Damage Detection and
Cole, H. A, 1968, “On-the-Line Analysis of Random Vibrations”, AIAA Localization in Stiffened Structures”, In Proceedings of the 23rd
Paper no. 68–288. International Modal Analysis Conference (IMAC XXIII), Orlando,
Cole, H. A, 1973, “On-line Failure Detection and Damping Measurement Florida, USA, paper no. 268.
of Aerospace Structures by Random Decrement Signatures”, NASA Fritzen, C. P., Mengelkamp, G. and Guemes, A, 2002, “A CFRP Plate with
CR-2205. Piezo-electric Actuators and Sensors as Self-Diagnosing Intelligent
Coppotelli, G., Di Conza, R., Mastroddi, F., Pascual, R. and Caponero, M. Structure”, In Proceedings of the International Conference on Noise
A, 2004, “Damage Identification in Composite Plates by Dynamic and Vibration Engineering (ISMA 2002), Leuven, Belgium, 185–191.
Displacement Measurements”, In Proceedings of the International Galvin, P. and Dominguez, J, 2005, “Modal Identification of a Pedestrian
rd
Conference on Noise and Vibration Engineering (ISMA 2004), Leu- Bridge by Output-Only Analysis”, In Proceedings of the 23 Inter-
ven, Belgium, 417–432. national Modal Analysis Conference (IMAC XXIII), Orlando, Flor-
Cornwell, P. J., Doebling, S. W. and Farrar, C. R, 1997, “Application of the ida, USA, paper no. 63.
Strain Energy Damage Detection Method to Plate-Like Structures”, Giurgiutiu, V., Buli, X. and Cuc, A, 2005, “Dual Use of Traveling and
In Proceedings of the 15th International Modal Analysis Conference Standing Lamb Waves for Structural Health Monitoring”, In Pro-
rd
(IMAC XV), Orlando, Florida, USA, 1312–1318. ceedings of the 23 International Modal Analysis Conference (IMAC
d’Ambrogio, W. and Fregolent, A, 2000, “The use of Antiresonances for XXIII), Orlando, Florida, USA, paper no. 361.
Robust Model Updating”, Journal of Sound and Vibration, Vol. 236, Golub, G. H. and Van Loan, C. F, 1983, “Matrix Computations”, North
227–243. Oxford Academic.
Davies, P. and Hammond, J. K, 1984, “A Comparison of Fourier and Para- Gomes, A. J. M. A. and Silva, J. M. M, 1992, “CRACAR – A Computer
metric Methods for Structural Systems Identification”, Journal of Program for Detection of Cracks in Structural Elements through the
Vibration, Acoustics, Stress and Reliability in Design, Vol. 106, 40– Variation of their Natural Frequencies”, Internal Report, Faculdade
48. de Ciências e Tecnologia, New University of Lisbon, Portugal (in
De Roeck, G. and Degrauwe, D, 2005, “Damage Identification of the Portuguese).
rd
Lanaye Cable-Stayed Bridge”, In Proceedings of the 23 Interna- Grisso, B. J., Martin, L. A. and Inman, D. J, 2005, “A Wireless Active
tional Modal Analysis Conference (IMAC XXIII), Orlando, Florida, Sensing System for Impedance-based Structural Health Monitoring”,
USA, paper no. 264. In Proceedings of the 23rd International Conference on Modal Anal-
De Roeck, G., Teughels, A. and Reynders, E, 2005, “Damage Identifica- ysis (IMAC XXIII), Orlando, Florida, USA, paper no. 236.
tion of Civil Engineering Structures Based on Operational Modal Grisso, B. J., Peairs, D. M. and Inman, D. J, 2004, “Impedance-based
Data”, In Proceedings of the 1st International Operational Confer- Health Monitoring of Composites”, In Proceedings of the 22nd Inter-

national Modal Analysis Conference (IMAC XXII), Dearborn, Michi- Johnson, T. J., Brown, R. L., Adams, D. E. and Schiefer, M, 2004, “Dis-
gan, USA, paper no. 274. tributed Structural Health Monitoring with a Smart Sensor Array”,
Guillaume, P., Hermans, L. and Van der Auweraer, H, 1999, “Maximum Mechanical Systems and Signal Processing, Vol. 18, 555–572.
Likelihood Identification of Modal Parameters from Operational Kessler, S. S., Spearing, S. M. and Atalla, M. J, 2002, “In-Situ Damage
th
Data”, In Proceedings of the 17 International Modal Analysis Con- Detection of Composites Structures using Lamb Waves Methods”,
st
ference (IMAC XVII), Kissimmee, Florida, USA, 1887–1893. In Proceedings of the 1 European Workshop on Structural Health
Guillaume, P., Parloo, E., Verboven, P. and De Sitter, G, 2002, “An Monitoring, Ecole Noemale Supérieure, Cachan, Paris, France, 374–
Inverse Method for the Identification of Localized Excitation 381.
th
Sources”, In Proceedings of the 20 International Modal Analysis Keye, S., Rose, M. and Sachau, D, 2001, “Localizing Delamination Dam-
Conference (IMAC XX), Los Angeles, California, USA, 1382–1388. ages in Aircraft Panels from Modal Damping Parameters”, In Pro-
th
Guillaume, P., Verboven, P. and Vanlanduit, S, 1998, “Frequency-Domain ceedings of the 19 International Modal Analysis Conference (IMAC
Maximum Likelihood Estimation of Modal Parameters with Confi- XIX), Kissimmee, Florida, USA, 412–417.
dence Intervals”, In Proceedings of the International Conference on Kim, B. H., Stubbs, N. and Park, T, 2005, “Flexural Damage Index Equa-
Noise and Vibration Engineering (ISMA 23), Leuven, Belgium, 359– tions of a Plate”, Journal of Sound and Vibration, Vol. 283, 341–
366. 368.
Gutschmidt, S. and Cornwell, P. J, 2001, “Statistical Confidence Bounds Kim, B. S., Yoo, S. H. and Yeo, G. H, 2005, “Charaterization of Crack
th
for Structural Health Monitoring”, In Proceedings of the 19 Inter- Detection on Gusset Plates using Strain Mode Shapes”, In Proceed-
rd
national Modal Analysis Conference (IMAC XIX), Kissimmee, Flor- ings of the 23 International Modal Analysis Conference (IMAC
ida, USA, 193–198. XXIII), Orlando, Florida, USA, paper no. 150.
Hamey, C. S., Lestari, W., Qiao, P. and Song, G, 2004, “Experimental Kim, J. H., Jeon, H. S. and Lee, C. W, 1992*, “Application of the Modal
Damage Identification of Carbon/Epoxy Composite Beams using Assurance Criteria for Detecting and Locating Structural Faults”, In
Curvature Mode Shapes”, Structural Health Monitoring, Vol. 3, Proceedings of the 10th International Modal Analysis Conference
333–353. (IMAC X), San Diego, California, USA, 536–540.
Hanagud, S. and Luo, H, 1997**, “Damage Detection and Health Monitor- Kisa, M, 2004, “Free Vibration Analysis of a Cantilever Composite Beam
ing based on Structural Dynamics”, Structural Health Monitoring, with Multiple Cracks”, Composites Science and Technology, Vol.
Current Status and Perspectives, Stanford University, Palo Alto, 64, 1391–1402.
California, USA, 715–726. Klein, L. A, 1999**, “Sensor and Data Fusion: Concepts and Application”,
Hatem, T. M., Foutouh, M. N. A. and Negm, H. M, 2004, “Application of SPIE Press.
Genetic Algorithms and Neural Networks to Health Monitoring of Konstantinidis, G., Wilcox, P. and Drinkwater, B, 2005, “Damage Detec-
nd
Composite Structures”, In Proceedings of the 2 European Work- tion using a Distributed Array of Guided Wave Sensors”, In Pro-
shop on Structural Health Monitoring, Forum am Deutschen ceedings of the 23rd International Modal Analysis Conference (IMAC
Museum, Munich, Germany, 616–623. XXIII), Orlando, Florida, USA, paper no. 265.
He, J. and Ewins, D. J, 1985*, “Analytical Stiffness Matrix Correlation Krawczuk, M., Ostachowicz, W. and Kawiecki, G, 2000**, “Detection of
using Measured Vibration Modes”, Modal Analysis: The Interna- Delaminations in Cantilevered Beams using Soft Computing Meth-
tional Journal of Analytical and Experimental Modal Analysis, Vol. ods”, In Proceedings of the European COST F3 Conference on Sys-
1(3), 9–14. tem Identification and Structural Health Monitoring, Madrid, Spain,
Hearn, G. and Testa, R. B, 1991*, “Modal Analysis for Damage Detection 243–252.
in Structures”, Journal of Structural Engineering, Vol. 117, 3042– Kullaa, 2002, “Elimination of Environmental Influences from Damage-
3062. Sensitive Features in a Structural Health Monitoring System”, In
st
Hera, A., Shinde, A. and Hou, Z, 2005, “Issues in Tracking Instantaneous Proceedings of the 1 European Workshop on Structural Health Mon-
Modal Parameters for Structural Health Monitoring using Wavelet itoring, Ecole Noemale Supérieure, Cachan, Paris, France, 742–749.
rd
Approach”, In Proceedings of the 23 International Modal Analysis Kyriazoglou, C., Le Page, B. H. and Guild, F. J, 2004, “Vibration Damping
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 338. for Crack Detection in Composite Laminates”, Composites, Part A:
Heylen, W., Lammens, S. and Sas, P, 1998, “Modal Analysis Theory and Applied Sciences and Manufacturing, Vol. 35, 945–953.
Testing”, K. U. Leuven – PMA, Belgium, Section A.6. Lamb, H, 1917, “On Waves in an Elastic Plate”, Proceedings of the Royal
Hiroshi, H., Hassan, M. S. A., Hatsukade, Y., Wakayama, S., Suemasu, H. Society of London, Containing Papers of a Mathematical and Physical
and Kasai, N, 2005, “Damage Detection of C/C Composites using Character, Vol. 93, 293–312.
ESPI and SQUID Techniques”, Composites Science and Technol- Law, S. S. and Lu, Z. R, 2005, “Crack Identification in Beam from
ogy, Vol. 65, 1098–1106. Dynamic Responses”, Journal of Sound and Vibration, Vol. 285, 967–
Ho, Y. K. and Ewins, D. J, 2000**, “On the Structural Damage Identifica- 987.
tion with Mode Shapes”, In Proceedings of the European COST F3 Lazarov, B. and Trendafilova, I, 2004, “An Investigation on Vibration-
nd
Conference on System Identification and Structural Health Monitor- Based Damage Diagnosis in Thin Plates”, In Proceedings of the 2
ing Madrid, Spain, 677–686. European Workshop on Structural Health Monitoring, Forum am
Hunt, D. L, 1992, “Application of an Enhanced Coordinate Modal Assur- Deutschen Museum, Munich, Germany, 76–82.
th
ance Criterion”, In Proceedings of the 10 International Modal Le Page, B. H., Guild, F. J., Ogin, S. L. and Smith, P. A, 2004, “Finite Ele-
Analysis Conference (IMAC X), San Diego, California, USA, 66–71. ment Simulation of Woven Fabric Composites”, Composites: Part
Ibrahim, S. R., Asmussen, J. C. and Brincker, R, 1997, “Vector Triggering A, Vol. 35, 861–872.
Random Decrement Technique for Higher Identification Accuracy”, Lee, B. C. and Staszewski, W. J, 2002, “Modelling of Acousto-ultrasonic
In Proceedings of the 15th International Modal Analysis Conference Wave Interaction with Defects in Metallic Structures”, In Proceed-
(IMAC XV), Orlando, Florida, USA, 502–509. ings of the International Conference on Noise and Vibration Engi-
Jacob, P. J., Desforges, M. J. and Ball, A. D, 1997**, “Analysis of Suitable neering (ISMA 2002), Leuven, Belgium, 319–327.
Wavelet Coefficients for Identification of the Simulated Failure of Lenzen, A, 2005, “Identification of Mechanical Systems by Vibration
nd
Composite Materials”, In Proceedings of the 2 International Con- Analysis for Health Monitoring and Damage Detection”, In Pro-
st
ference on Structural Damage Assessment using Advanced Signal ceedings of the 1 International Operational Modal Analysis Confer-
Processing Procedures (DAMAS 97), Sheffield, England, UK, 31–40. ence (IOMAC 2005), Copenhagen, Denmark, 155–166.
Jauregui, D. V. and Farrar, C. R, 1986a*, “Damage Identification Algo- Leontaritis, I. J. and Billings, S. A, 1985a, “Input Output Parametric Mod-
rithms Applied to Numerical Modal Data from a Bridge”, In Pro- els for Non-Linear Systems.1. Deterministic Non-Linear Systems”,
ceedings of the 14th International Modal Analysis Conference (IMAC International Journal of Control, Vol. 41, 303–328.
XIV), Dearborn, Michigan, USA, 119–125. Leontaritis, I. J. and Billings, S. A, 1985b, “Input Output Parametric Mod-
Jauregui, D. V. and Farrar, C. R, 1986b*, “Comparison of Damage Identi- els for Non-Linear Systems.2. Stochastic Non-Linear Systems”,
fication Algorithms on Experimental Modal Data from a Bridge”, In International Journal of Control, Vol. 41, 329–344.
th
Proceedings of the 14 International Modal Analysis Conference Li, B., Chen, X. F., Ma, J. X. and He, Z. J, 2005, “Detection of Crack Loca-
(IMAC XIV), Dearborn, Michigan, USA, 1423–1429. tion and Size in Structures using Wavelet Finite Element Methods”,
Johnson, E. A., Lam, H. F., Katafygiotis, L. S. and Beck, J. L, 2004, “The Journal of Sound and Vibration, Vol. 285, 767–782.
Phase I IASC-ASCE Structural Health Monitoring Benchmark Prob- Li, H. C. H., Weis, M., Herszberg, I. and Mouritz, A. P, 2004, “Damage
lem using Simulated Data”, ASCE Journal of Engineering Mechan- Detection in a Fiber Reinforced Composite Beam using Random
ics, Vol. 130, 3–15. Decrement Signatures”, In Proceedings of the 2nd European Work-

shop on Structural Health Monitoring, Forum am Deutschen Museum, Montalvão, D., Ribeiro, A. M. R., Maia, N. M. M. and Silva, J. M. M, 2004a,
Munich, Germany, 1091–1098. “Estimation of the Rotational Terms of the Dynamic Response
Liang, C., Sun, F. P. and Rogers, C. A, 1994, “An Impedance Method for Matrix”, Shock and Vibration, Vol. 11, 333–350.
Dynamic Analysis of Active Material System”, Journal of Vibration Montalvão, D., Ribeiro, A. M. R., Maia, N. M. M. and Silva, J. M. M,
and Acoustics, Vol. 116, 121–128. 2004b, “On the Estimation of Frequency Response Functions”, In
Lieven, N. A. J. and Ewins, D. J, 1988, “Spatial Correlation of Mode Proceedings of the International Conference on Noise and Vibration
Shapes: The Coordinate Modal Assurance Criterion (COMAC)”, In Engineering (ISMA 2004), Leuven, Belgium, 2771–2785.
th
Proceedings of the 6 International Modal Analysis Conference (IMAX Morais, W. A., Monteiro, S. N. and d’Almeida, J. R. M, 2005, “Effect of the
VI), Kissimmee, Florida, USA, 690–695. Laminate Thickness on the Composite Strength to Repeated Low
Lifshitz, J. M. and Rotem, A, 1969*, “Determination of Reinforcement Energy Impacts”, Composite Structures, Vol. 70, 223–228.
Unbonding of Composites by a Vibration Technique”, Journal of Moura Jr., J. R. V, 2005, “Damage Detection Techniques for Aeronautic
rd
Composite Materials, Vol. 3, 412–423. Structures”, In Proceedings of the 23 International Modal Analysis
Lin, M, 1999**, “Development of SMART Layer for Built-In Structural Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 101.
Diagnostics”, Structural Health Monitoring 2000, Stanford Univer- Moura Jr., J. R. V. and Steffen Jr., V, 2004, “Impedance-based Health
nd
sity, Palo Alto, California, USA, 603–611. Monitoring: Frequency Band Evaluation”, In Proceedings of the 22
Lin, S., Yang, J. N. and Zhou, L, 2005, “Damage Identification of a Bench- International Modal Analysis Conference (IMAC XXII), Dearborn,
mark Building for Structural Health Monitoring”, Smart Materials Michigan, USA, paper no. 161.
and Structures, Vol. 14, 162–169. Nayfeh, S. A, 2004, “Damping of Flexural Vibration in the Plane of Lami-
Liu, M. and Chelidze, D, 2005, “Flow Variance Method for Damage Iden- nation of Elastic-viscoelastic Sandwich Beams”, Journal of Sound
rd
tification”, In Proceedings of the 23 International Modal Analysis and Vibration, Vol. 276, 689–711.
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 188. Nejad, F. B., Rahai, A. and Esfandiari, A, 2005, “A Constrained Structural
Lu, Y. and Gao, F, 2005, “A Novel Time-Domain Auto-Regressive Model Damage Detection Method Using Static Noisy Data and Natural Fre-
rd
for Structural Damage Diagnosis”, Journal of Sound and Vibration, quencies”, In Proceedings of the 23 International Modal Analysis
Vol. 283, 1031–1049. Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 127.
Lucero, J. and Taha, M. M. R, 2005, “A Wavelet-Aided Fuzzy Damage Nichols, J. M., Moniz, L., Seaver, M. and Trickey, S. T, 2005, “Use of
Detection Algorithm for Structural Health Monitoring”, In Proceed- Holder Exponents and Fiber Optic Sensing for Detecting Damage in
rd rd
ings of the 23 International Modal Analysis Conference (IMAC an Experimental Plate Structure”, In Proceedings of the 23 Interna-
XXIII), Orlando, Florida, USA, paper no. 78. tional Modal Analysis Conference (IMAC XXIII), Orlando, Florida,
Luo, H. and Hanagud, S, 1997**, “Dynamic Learning Rate Neural Net- USA, paper no. 197.
work Training and Composite Structural Damage Detection”, AIAA Nieuwenhuis, J. H., Neumann, J. J., Greve, D. W. and Oppenheim, I. J,
Journal, Vol. 35, 1522–1527. 2005, “Simulation and Testing of Transducers for Lamb Wave Gen-
rd
Lynch, J. P., Sundararajan, A., Law, K. H., Sohn, H. and Farrar, C. R, eration”, In Proceedings of the 23 International Modal Analysis
2004, “Piezoelectric Structural Excitation using a Wireless Active Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 216.
nd
Sensing Unit”, In Proceedings of the 22 International Modal Analysis Olson, C. C., Overbey, L. A. and Todd, M. D, 2005, “A Comparison of
Conference (IMAC XXII), Dearborn, Michigan, USA, paper no. 157. State-Space Attractor Features in Structural Health Monitoring”, In
rd
Maheri, M. R. and Adams, R. D, 2003, “Modal Vibration Damping of Ani- Proceedings of the 23 International Modal Analysis Conference
sotropic FRP Laminates using the Rayleigh-Ritz Energy Minimiza- (IMAC XXIII), Orlando, Florida, USA, paper no. 209.
tion Scheme”, Journal of Sound and Vibration, Vol. 259, 17–29. Ostachowicz, W. and Zak, A, 2004, “Vibration of a Laminated Beam with
Maia, N. M. M, 2001, “Detecção de Dano Estrutural por Análise Dinâmica”, Delamination including Contact Effects”, Shock and Vibration, Vol.
Habilitation Lecture, Instituto Superior Técnico, Technical Univer- 11, 157–171.
sity of Lisbon, Portugal (in Portuguese). Pandey, A. K., Biswas, M. and Samman, M. M, 1991*, “Damage Detection
Maia, N. M. M. and Silva, J. M. M, 1997, “Theoretical and Experimental from Changes in Curvature Mode Shapes”, Journal of Sound and
Modal Analysis”, Research Studies Press, Ltd. Vibration, Vol. 145, 321–332.
Maia, N. M. M., Silva, J. M. M., Almas, E. A. M. and Sampaio, R. P. C, Pandit, S. W. and Metha, N. P, 1985, “Data Dependent System Approach
2003, “Damage Detection in Structures: from Mode Shape to Fre- to Modal Analysis Via State Space”, ASME paper No. 85-WA/DSC-1.
quency Response Function Methods”, Mechanical Systems and Sig- Park, G., Cudney, H. and Inman, D. J, 1999**, “Impedance-Based Health
nal Processing, Vol. 17, 489–498. Monitoring Technique for Civil Structures”, Structural Health Moni-
Maia, N. M. M., Silva, J. M. M. and Sampaio, R. P. C, 1997, “Localization toring 2000, Stanford University, Palo Alto, California, USA, 523–532.
of Damage using Curvature of the Frequency-Response-Functions”, Park, G., Rutherford, A. C., Sohn, H. and Farrar, C. R, 2005, “An Outlier
In Proceedings of the 15th International Modal Analysis Conference Analysis Framework for Impedance-based Structural Health Moni-
(IMAC XV), Orlando, Florida, USA, 942–946. toring”, Journal of Sound and Vibration, Vol. 286, 229–250.
Mares, C., Ruotolo, R. and Surace, C, 1999**, “Using Transmissibility Park, G., Sohn, H., Farrar, C. R. and Inman, D. J, 2003, “Overview of Pie-
Data to Assess Structural Damage”, In Proceedings of the 3rd Inter- zoelectric Impedance-Based Health Monitoring and Path-Forward”,
national Conference on Damage Assessment of Structures (DAMAS Shock and Vibration Digest, Vol. 35, 451–463.
99), Dublin, Ireland, 236–245. Park, J. M., Kong, J. W., Kim, D. S. and Yoon, D. J, 2005, “Nondestructive
Matthews, F. L, 1999, “Damage in Fibre Reinforced Plastics; Its Nature, Damage Detection and Interfacial Evaluation of Single-Fibers/
rd
Consequences and Detection”, In Proceedings of the 3 Interna- Epoxy Composites using PZT, PVDF and P(VDF-TrFE) Copolymer
tional Conference on Damage Assessment of Structures (DAMAS Sensors”, Composites Science and Technology, Vol. 65, 241–256.
99), Dublin, Ireland, 1–16. Park, N. G. and Park, Y. S, 2003, “Damage Detection using Spatially
Messina, A., Jones, I. A. and Williams, E. J, 1992, “Damage Detection and Incomplete Frequency Response Functions”, Mechanical Systems
Localization using Natural Frequency Changes”, In Proceedings of and Signal Processing, Vol. 17, 519–532.
the 1st Conference on Identification, Cambridge, England, UK, 67–76. Park, Y. S., Park, H. S. and Lee, S. S, 1988*, “Weighted-Error-Matrix
Messina, A., Jones, I. A. and Williams, E. J, 1996, “Damage Detection and Application to Detect Stiffness Damage by Dynamic-Characteristic
Localization using Natural Frequency Changes”, In Proceedings of Measurement”, Modal Analysis: The International Journal of Ana-
the Conference on Identification in Engineering Systems, Swansea, lytical and Experimental Modal Analysis, Vol. 3, 101–107.
Wales, UK, 67–76. Parloo, E., Guillaume, P. and Van Overmeire, M, 2003, “Damage Assess-
Messina, A., Williams, E. J. and Contursi, T, 1998**, “Structural Damage ment Using Mode Shape Sensitivities”, Mechanical Systems and Sig-
Detection by a Sensitivity and Statistical-Based Method”, Journal of nal Processing, Vol. 17, 499–518.
Sound and Vibration, Vol. 216, 791–808. Parloo, E., Vanlanduit, S., Guillaume, P. and Verboven, P, 2002b, “Increased
Mevel, L. and Goursat, M, 2004, “A Complete Scilab Toolbox for Output- Reliability of Reference-based Damage Identification Techniques by
Only Identification”, In Proceedings of the 22nd International Modal using Output-only Data”, In Proceedings of the International Con-
Analysis Conference (IMAC XXII), Dearborn, Michigan, USA, paper ference on Structural Dynamics Modelling, Test, Analysis, Correla-
no. 60. tion and Validation, Funchal, Madeira, Portugal, 137–146.
Montalvão, D, 2003, “Determination of Rotational Terms of the Dynamic Parloo, E., Vanlanduit, S., Guillaume, P. and Verboven, P, 2004, “Increased
Response by means of Modal Analysis Techniques”, MSc Thesis, Reliability of Reference-based Damage Identification Techniques by
Instituto Superior Técnico, Technical University of Lisbon, Portugal using Output-only Data”, Journal of Sound and Vibration, Vol. 270,
(in Portuguese). 813–832.

Parloo, E., Verboven, P., Guillaume, P. and Van Overmeire, M, 2002a, “Sen- European Workshop on Structural Health Monitoring, Forum am
sitivity-based Operational Mode Shape Normalization”, Mechanical Deutschen Museum, Munich, Germany, 757–766.
Systems and Signal Processing, Vol. 16, 757–767. Sampaio, R. P. C., Maia, N. M. M. and Silva, J. M. M, 1999, “Damage
Pascual, R., Golinval, J. C. and Razeto, M, 1997, “A Frequency Domain Detection using the Frequency-Response-Function Curvature Method”,
Correlation Technique for Model Correlation and Updating”, In Pro- Journal of Sound and Vibration, Vol. 226, 1029–1042.
th
ceedings of the 15 International Modal Analysis Conference (IMAC Sampaio, R. P. C., Maia, N. M. M. and Silva, J. M. M, 2003, “The Fre-
XV), Orlando, Florida, USA, 587–592. quency Domain Assurance Criterion as a Tool for Damage Identifica-
th
Patil, D. P. and Maiti, S. K, 2005, “Experimental Verification of a Method tion”, In Proceedings of the 5 International Conference on Damage
of Detection of Multiple Cracks in Beams based on Frequency Meas- Assessment of Structures (DAMAS 2003), Southampton, England, UK,
urements”, Journal of Sound and Vibration, Vol. 281, 439–451. 69–76.
Peairs, D. M., Park, G. and Inman, D. J, 2005, “Impedance-based SHM Sampaio, R. P. C., Maia, N. M. M., Silva, J. M. M. and Ribeiro, A. M. R,
Modeling using Spectral Elements and Circuit Analysis”, In Pro- 2000, “On the use of Transmissibility for Damage Detection and
ceedings of the 23rd International Modal Analysis Conference (IMAC Location”, In Proceedings of the European COST F3 Conference on
XXIII), Orlando, Florida, USA, paper no. 232. System Identification and Structural Health Monitoring, Madrid,
Peeters, B. and De Roeck, G, 1999, “Reference-based Stochastic Subspace Spain, 363–376.
Identification for Output-only Modal Analysis”, Mechanical Systems Sanders, D. R., Kim, Y. I. and Stubbs, N, 1992*, “Nondestructive Evalua-
and Signal Processing, Vol. 13, 855–878. tion of Damage in Composite Structures Using Modal Parameters”,
Peeters, B. and De Roeck, G, 2000, “One Year Monitoring of the Z24- Experimental Mechanics, Vol. 32, 240–251.
Bridge: Environmental Influences versus Damage Events”, In Pro- Santos, J. V. A., Soares, C. M. M., Soares, C. A. M. and Maia, N. M. M,
th
ceedings of the 18 International Modal Analysis Conference (IMAC 2003a, “Análise de Erros na Identificação do Dano em Estruturas
XVIII), San Antonio, Texas, USA, 1570–1576. Laminadas”, In VII Congresso de Mecânica Aplicada e Computacional,
Peil, U., Mehdianpour, M., Frenz, M. and Weilert, K, 2004, “Determina- Universidade de Évora, Portugal (in Portuguese), 95–103.
tion of the Inherent Damage of Older Structures”, In Proceedings of Santos, J. V. A., Soares, C. M. M., Soares, C. A. M. and Maia, N. M. M,
nd
the 2 European Workshop on Structural Health Monitoring, Forum 2003b, “Structural Damage Identification: Influence of Model Incom-
am Deutschen Museum, Munich, Germany, 243–250. pleteness and Errors”, Composite Structures, Vol. 62, 303–313.
Penny, J. E. T., Wilson, D. A. L. and Friswell, M. I, 1993*, “Damage Loca- Santos, J. V. A., Soares, C. M. M., Soares, C. A. M. and Maia, N. M. M,
th
tion in Structures using Vibration Data”, In Proceedings of the 11 2004, “Modelos de Identificação do Dano com base nas Característi-
International Modal Analysis Conference (IMAC XI), Kissimmee, cas Dinâmicas de Estruturas”, Métodos Computacionais em Engen-
Florida, USA, 861–867. haria (in Portuguese).
Pirner, M. and Urushadze, S, 2004, “Dynamic Response as a Tool for Santos, J. V. A., Soares, C. M. M., Soares, C. A. M. and Maia, N. M. M,
Damage Identification”, International Applied Mechanics, Vol. 40, 2005, “Structural Damage Identification in Composite Structures using
487–505. FRF Data”, Composite Structures, Vol. 67, 239–249.
Qin, Q. and Zhang, W, 1998, “Damage Detection of Suspension Bridges”, Santos, J. V. A., Soares, C. M. M., Soares, C. A. M. and Pina, H. L. G,
th
In Proceedings of the 16 International Modal Analysis Conference 1999, “Damage Identification of Composite Structures: a Numerical
(IMAC XVI), Santa Barbara, California, USA, 945–951. Model”, Mechanics of Composite Materials and Structures, Vol. 6,
Qing, X., Kumar, A., Zhang, C., Gonzalez, I. F., Guo, G. and Chang, F. K, 363–379.
2005, “A Hybrid Piezoelectric/Fiber Optic Diagnostic System for Santos, J. V. A., Soares, C. M. M., Soares, C. A. M. and Pina, H. L. G,
Structural Health Monitoring”, Smart Materials and Structures, Vol. 2000, “Development of a Numerical Model for the Damage Identifi-
14, 98–103. cation on Composite Plate Structures”, Composite Structures, Vol.
Raghavan, A. and Cesnik, C. E. S, 2005, “Analytical Models for Lamb- 48, 59–65.
Wave Based Structural Health Monitoring”, In Proceedings of the Saravanos, D. A., Birman, V. and Hopkins, D. A, 1994*, “Detection of
rd
23 International Conference on Modal Analysis (IMAC XXIII), Delaminations in Composite Beams using Piezoelectric Sensors”, In
st
Orlando, Florida, USA, paper no. 289. Proceedings of the 31 AIAA/ASME/ASCE/AHS/ASC Structures, Struc-
Randall, R. B, 2002, “State of the Art in Monitoring Rotating Machinery”, tural Dynamics and Materials Conference, 181–191.
In Proceedings of the International Conference on Noise and Vibra- Savov, K. and Wenzel, H, 2005, “System Identification and Damage Detec-
tion Engineering (ISMA 2002), Leuven, Belgium, 1457–1477. tion using Wavelet Analysis: Applications in Frame Structures”, In
Ratcliffe, C. P, 2000, “A Frequency and Curvature based Experimental Proceedings of the 1st International Operational Modal Analysis
Method for Locating Damage in Structures”, Journal of Vibration Conference (IOMAC 2005), Copenhagen, Denmark, 511–518.
and Acoustics, Vol. 122, 324–329. Shi, Z. Y. and Law, S. S, 2005a, “Parameter Identification-Based Damage
Ratcliffe, C. P. and Bagaria, W. J, 1998, “A Vibration Technique for Detection for Nonlinear Systems”, In Proceedings of the 1st Interna-
Locating Delamination in a Composite Beam”, AIAA Journal, Vol. tional Operational Modal Analysis Conference (IOMAC 2005), Copen-
36, 1074–1077. hagen, Denmark, 503–510.
Ren, Y. and Beards, C. F, 1995, “Identification of Joint Properties of a Shi, Z. Y. and Law, S. S, 2005b, “Parameter Identification-Based Damage
Structure using FRF Data”, Journal of Sound and Vibration, Vol. Detection for Linear Time-Varying Systems”, In Proceedings of the
st
186, 567–587. 1 International Operational Modal Analysis Conference (IOMAC
Ricci, F., Banerjee, S. and Mal, A. K, 2004, “Health Monitoring of Com- 2005), Copenhagen, Denmark, 527–534.
posite Structures using Wave Propagation Data”, In Proceedings of Shu, D. and Della, C. N, 2004, “Vibrations of Multiple Delaminated
nd
the 2 European Workshop on Structural Health Monitoring, Forum Beams”, Composite Structures, Vol. 64, 467–477.
am Deutschen Museum, Munich, Germany, 1035–1042. Sidhu, J. and Ewins, D. J, 1984, “Correlation of Finite Element and Modal
nd
Rodrigues, J., Brincker, R. and Andersen, P, 2004, “Improvement of Fre- Test Studies of a Practical Structure”, In Proceedings of the 2 Inter-
quency Domain Output-Only Modal Identification from the Applica- national Modal Analysis Conference (IMAC II), Orlando, Florida,
tion of the Random Decrement Technique”, In Proceedings of the USA, 185–192.
nd
22 International Modal Analysis Conference (IMAC XXII), Dear- Silva, A. J. P. F, 2001, “Propagação de Delaminagem em Compósitos de
born, Michigan, USA, paper no. 92. Matriz Polimérica”, PhD Thesis, Instituto Superior Técnico, Techni-
Ruotolo, R. and Surace, C, 1997**, “Damage Detection Using Singular cal University of Lisbon, Portugal (in Portuguese).
nd
Value Decomposition”, In Proceedings of the 2 International Confer- Silva, J. M. M. and Gomes, A. J. M. A, 1990*, “Experimental Dynamic
ence on Structural Damage Assessment using Advanced Signal Process- Analysis of Cracked Free-Free Beams”, Experimental Mechanics, Vol.
ing Procedures (DAMAS 97), Sheffield, England, UK, 87–96. 30, 20–25.
,
Rytter, A, 1993* **, “Vibration Based Inspection of Civil Engineering Silva, J. M. M. and Gomes, A. J. M. A, 1991, “Crack Modelling Using
Structures”, PhD Thesis, Aalborg University, Denmark. Torsional Springs”, Actas das Quartas Jornadas de Fractura, Socie-
Safak, E, 1991, “Identification of Linear Structures using Discrete-Time dade Portuguesa de Materiais, Instituto Superior Técnico, Technical
Filters”, Journal of Structural Engineering, Vol. 117, 3064–3085. University of Lisbon, Portugal (in Portuguese).
Salawu, O. S. and Williams, C, 1994*, “Damage Location using Vibration Silva, J. M. M. and Gomes, A. J. M. A, 1994*, “Crack Identification of
Mode Shapes”, In Proceedings of the 12th International Modal Anal- Simple Structural Elements through the use of Natural Frequency
th
ysis Conference (IMAC XII), Honolulu, Hawaii, USA, 933–939. Variations: The Inverse Problem”, In Proceedings of the 12 Inter-
Sampaio, R. P. C. and Maia, N. M. M, 2004, “On the Detection and Rela- national Modal Analysis Conference (IMAC XII), Honolulu, Hawaii,
tive Damage Quantification Indicator”, In Proceedings of the 2nd USA, 1728–1735.

Simmers Jr., G., Sodano, H., Park, G. and Inman, D. J, 2005, “Impedance- Toyama, N. and Okabe, T, 2004, “Effects of Tensile Strain and Transverse
based Structural Health Monitoring to Detect Corrosion”, In Pro- Cracks on Lamb-Wave Velocity in Cross-Ply FRP Laminates”, Jour-
rd
ceedings of the 23 International Modal Analysis Conference (IMAC nal of Materials Science, Vol. 39, 7365–7367.
XXIII), Orlando, Florida, USA, paper no. 56. Tracy, J. J. and Pardoen, G. C, 1989*, “Effect of Delamination on the Natural
Sohn, H. and Farrar, C. R, 2001, “Damage Diagnosis using Time Series Frequencies of Composite Laminates”, Journal of Composite Materi-
Analysis of Vibration Signals”, Smart Materials and Structures, Vol. als, Vol. 23, 1200–1215.
10, 446–451. Uhl, T. and Mendrok, K, 2004, “Overview of Modal Model based Damage
Sohn, H., Farrar, C. R., Hemez, F. M., Shunk, D. D., Stinemates, D. W. and Detection Methods”, In Proceedings of the International Conference
Nadler, B. R, 2003, “A Review of Structural Health Monitoring Lit- on Noise an Vibration Engineering (ISMA 2004), Leuven, Belgium,
erature: 1996–2001”, Los Alamos National Laboratory, USA. 361–375.
Sohn, H., Farrar, C. R., Hunter, N. F. and Worden, K, 2001**, “Structural Vanhonacker, P, 1980, “Differential and Difference Sensitivities of Natural
Health Monitoring using Statistical Pattern Recognition Techniques”, Frequencies and Mode Shapes of Mechanical Structures”, AIAA
Journal of Dynamic Systems Measurement and Control-Transac- Journal, Vol. 18, 1511–1514.
tions of the ASME, Vol. 123, 706–711. Vanlanduit, S., Parloo, E., Cauberghe, B., Guillaume, P. and Verboven, P,
Sohn, H., Park, G., Wait, J. R., Limback, N. P. and Farrar, C. R, 2004b, 2005, “A Robust Singular Value Decomposition for Damage Detection
“Wavelet-based Signal Processing for Detecting Delamination in under Changing Operating Conditions and Structural Uncertainties”,
Composite Plates”, Smart Materials and Structures, Vol. 13, 153–160. Journal of Sound and Vibration, Vol. 284, 1033–1050.
Sohn, H., Park, H., Law, K. H. and Farrar, C. R, 2005, “Instantaneous Vecchio, A. and Van der Auweraer, H, 2001, “An Experimental Validation
Online Monitoring of Unmanned Aerial Vehicles without Baseline of a Model-Based Approach in Damage Detection and Validation”,
rd rd
Signals”, In Proceedings of the 23 International Modal Analysis In Proceedings of the 3 International Workshop on Structural Health
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 259. Monitoring, Stanford University, Palo Alto, California, USA, 957–966.
Sohn, H., Wait, J. R., Park, G. and Farrar, C. R, 2004a, “Multi-Scale Struc- Vecchio, A., Van der Auweraer, H., Ferraro, G., Garesci, F. and Petrone, F,
tural Health Monitoring for Composite Structures”, In Proceedings 2002, “An Experimental Validation of a Model-Based Approach in
nd st
of the 2 European Workshop on Structural Health Monitoring, Forum Damage Detection and Localization”, In Proceedings of the 1 Euro-
am Deutschen Museum, Munich, Germany, 721–729. pean Workshop on Structural Health Monitoring, Ecole Noemale
Spencer Jr., B. F., Sandoval, M. E. R. and Kurata, N, 2004, “Smart Sensing Supérieure, Cachan, Paris, France, 245–252.
Technology: Opportunities and Challenges”, Journal of Structural Vescovo, D. D. and Fregolent, A, 2005, “Assessment of Fresco Detach-
Control and Health Monitoring, Vol. 11, 349–368. ments through a Non-Invasive Acoustic Method”, Journal of Sound
Steenackers, G. and Guillaume, P, 2005, “Structural Health Monitoring of and Vibration, Vol. 285, 1015–1031.
the Z-24 Bridge in Presence of Environmental Changes using Modal Wahab, M. M. and De Roeck, G, 1999, “Damage Detection in Bridges using
Analysis”, In Proceedings of the 23rd International Modal Analysis Modal Curvatures: Application to a Real Damage Scenario”, Journal
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 297. of Sound and Vibration, Vol. 226, 217–235.
Stubbs, N., Kim, J. T. and Farrar, C. R, 1995*, “Field Verification of a Non- Wahl, F., Schmidt, G. and Forrai, L, 1999, “On the Significance of Antires-
destructive Damage Localization and Severity Estimation Algorithm”, onance Frequencies in Experimental Structural Analysis”, Journal of
In Proceedings of the 13th International Modal Analysis Conference Sound and Vibration, Vol. 219, 379–394.
(IMAC XIII), Nashville, Tennessee, USA, 210–218. Wang, K., Inman, D. J. and Farrar, C. R, 2005, “Modelling and Analysis of
Stubbs, N. and Osegueda, R, 1990a*, “Global Non-Destructive Damage a Cracked Composite Cantilever Beam Vibrating in Coupled Bending
Evaluation in Solids”, Modal Analysis: The International Journal of and Torsion”, Journal of Sound and Vibration, Vol. 284, 23–49.
Analytical and Experimental Modal Analysis, Vol. 5, 67–79. Wei, Z., Yam, L. H. and Cheng, L, 2005, “NARMAX Model Representation
Stubbs, N. and Osegueda, R, 1990b*, “Global Damage Detection in Solids– and its Application to Damage Detection for Multi-layer Compos-
Experimental Verification”, Modal Analysis: The International Jour- ites”, Composite Structures, Vol. 68, 109–117.
nal of Analytical and Experimental Modal Analysis, Vol. 5, 81–97. West, W. M, 1984*, “Illustration of the Use of Modal Assurance Criterion to
Su, Z., Ye, L. and Bu, X, 2002, “Evaluation of Delamination in Laminated Detect Structural Changes in an Orbiter Test Specimen”, In Proceedings
Composites based on Lamb Waves Methods: FEM Simulation and of the Air Force Conference on Aircraft Structural Integrity, 1–6.
st
Experimental Verification”, In Proceedings of the 1 European Work- Wilson, E. L., Yuan, M. W. and Dicken, J. M, 1982, “Dynamic Analysis by
shop on Structural Health Monitoring, Ecole Noemale Supérieure, Direct Superposition of Ritz Vectors”, Earthquake Engineering and
Cachan, Paris, France, 328–335. Structural Dynamics, Vol. 10, 813–821.
Sundararaman, S., Adams, D. E. and Rigas, E. J, 2005, “Characterizing Woon, C. E. and Mitchell, L. D, 1996a, “Variations in Structural Dynamics
Damage in Plates through Beamforming with Sensor Arrays”, In Characteristics caused by Changes in Ambient Temperature: I. Exper-
Proceedings of the 23rd International Modal Analysis Conference imental”, In Proceedings of the 14th International Modal Analysis
(IMAC XXIII), Orlando, Florida, USA, paper no. 249. Conference (IMAC XIV), Dearborn, Michigan, USA, 963–971.
Sundaresan, M. J., Pai, P. F., Ghoshal, A., Schulz, M. J., Ferguson, F. and Woon, C. E. and Mitchell, L. D, 1996b, “Variations in Structural Dynamics
Chung, J. H, (2001a), “Methods of Distributed Sensing for Health Characteristics caused by Changes in Ambient Temperature: II. Ana-
th
Monitoring of Composite Material Structures”, Composites: Part A, lytical”, In Proceedings of the 14 International Modal Analysis Con-
Vol. 32, 1357–1374. ference (IMAC XIV), Dearborn, Michigan, USA, 972–980.
Sundaresan, M. J., Grandhi, G., Schulz, M. J. and Kemerling, J. C, 2004, Worden, K., Manson, G. and Fieller, N. R. J, 2000**, “Damage Detection
“Monitoring Damage Growth in a Composite Plate using a Continuous using Outlier Analysis”, Journal of Sound and Vibration, Vol. 229,
nd
Sensor”, In Proceedings of the 22 International Modal Analysis 647–667.
Conference (IMAC XXII), Dearborn, Michigan, USA, paper no. 369. Wowk, V, 1991, Machinery Vibration: Measurement and Analysis, McGraw-
Sundaresan, M. J., Ghoshal, A. and Schulz, M. J, (2001b), “A Continuous Hill.
Sensor to Measure Acoustic Waves in Plates”, Journal of Intelligent Wu, Z. S., Yang, C. Q., Harada, T. and Ye, L. P, 2005, “Self-Diagnosis of
Material Systems and Structures, Vol. 12, 41–56. Structures Strengthened with Hybrid Carbon-Fiber Reinforced Poly-
Taylor, S. and Zimmerman, D. C, 2005, “Damage Detection in a Cargo mer Sheets”, Smart Materials and Structures, Vol. 14, 39–51.
rd
Bay Frame using Ritz Vectors”, In Proceedings of the 23 Interna- Yam, L. H., Wei, Z. and Cheng, L, 2004, “Nondestructive Detection of
tional Modal Analysis Conference (IMAC XXIII), Orlando, Florida, Internal Delamination by Vibration-based Method for Composite
USA, paper no. 355. Plates”, Journal of Composite Materials, Vol. 38, 2183–2198.
Testa, R. B, 2005, “Characteristics and Detection of Damage and Fatigue Yan, A. M., Golinval, J. C., Peeters, B. and De Roeck, G, 2004, “A Com-
Cracks”, Advanced Course on Dynamic Methods for Damage Detec- parative Study on Damage Detection of Z24-bridge: One-year Moni-
tion in Structures, CISM (Centre International des Sciences Mécha- toring with Varying Environmental Conditions”, In Proceedings of
nd
niques), Udine, Italy. the 2 European Workshop on Structural Health Monitoring Forum
Thien, A. B., Chiamori, H. C., Ching, J. T., Wait, J. R. and Park, G, 2005, am Deutschen Museum, Munich, Germany, 791–799.
“Piezoelectric Active Sensing for Damage Detection in Pipeline Yan, A. M., Kerschen, G., De Boe, P. and Golinval, J. C, 2005a, “Struc-
rd
Structures”, In Proceedings of the 23 International Modal Analysis tural Damage Diagnosis under Varying Environmental Conditions –
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 323. Part I: A Linear Analysis”, Mechanical Systems and Signal Process-
Tippetts, T. and Hemez, F. M, 2005, “Non-Linear Models of Composite ing, Vol. 19, 847–864.
rd
Laminates”, In Proceedings of the 23 International Modal Analysis Yan, A. M., Kerschen, G., De Boe, P. and Golinval, J. C, 2005b, “Struc-
Conference (IMAC XXIII), Orlando, Florida, USA, paper no. 10. tural Damage Diagnosis under Varying Environmental Conditions –

Part II: Local PCA for Non-Linear Cases”, Mechanical Systems and Sensors, Part I: The Indicator of Correlation Criteria”, In Proceed-
th
Signal Processing, Vol. 19, 865–880. ings of the 5 International Conference on Damage Assessment of
Yang, J. N., Lei, Y., Pan, S. and Huang, N, 2003, “Identification of Linear Structures (DAMAS 2003), Southampton, England, UK, 131–140.
Structures Based on Hilbert-Huang Transform Part II: Complex Zang, C., Friswell, M. I. and Imregun, M, 2003b, “Structural Health Moni-
Modes”, Earthquake Engineering and Structural Dynamics, Vol. 32, toring and Damage Assessment using Measured FRFs from Multiple
1443–1467. Sensors, Part II: Decision Making with RBF Networks”, In Proceed-
Yoon, M. K., Heider, D., Gillespie, J. W., Ratcliffe, C. P. and Crane, R. M, ings of the 5th International Conference on Damage Assessment of
2005, “Local Damage Detection using the Two-dimensional Gapped Structures (DAMAS 2003), Southampton, England, UK, 141–148.
Smoothing Method”, Journal of Sound and Vibration, Vol. 279, 119– Zhang, R., King, R., Olson, L. and Xu, Y. L, 2005, “Dynamic Response of
139. the Trinity River Relief Bridge to Controlled Pile Damage: Modelling
Yu, L., Yin, T., Zhu, H. P. and Xu, D. Y, 2005, “An Improved Genetic and Experimental Data Analysis Comparing Fourier and Hilbert-
Algorithm for Structural Damage Detection”, In Proceedings of the Huang Techniques.” Journal of Sound and Vibration, Vol. 285, 1049–
1st International Operational Modal Analysis Conference (IOMAC 1070.
2005), Copenhagen, Denmark, 361–368. Zheng, S., Wang, X. and Liu, L, 2004, “Damage Detection in Composite
Zabel, V, 2004, “A Wavelet-Based Damage Detection Indicator for Rein- Materials Based upon the Computational Mechanics and Neural Net-
forced Concrete Structures”, In Proceedings of the 22nd International works”, In Proceedings of the 2nd European Workshop on Structural
Modal Analysis Conference (IMAC XXII), Dearborn, Michigan, USA, Health Monitoring, Forum am Deutschen Museum, Munich, Ger-
paper no. 70. many, 609–615.
Zabel, V, 2005a, “A Wavelet-based Approach for Damage Detection on Zhu, H. P. and Xu, Y. L, 2005, “Damage Detection of Mono-Coupled Peri-
rd
Civil Engineering Structures”, In Proceedings of the 23 Interna- odic Structures based on Sensitivity Analysis of Modal Parameters”,
tional Modal Analysis Conference (IMAC XXIII), Orlando, Florida, Journal of Sound and Vibration, Vol. 285, 363–390.
USA, paper no. 178. Zimmerman, D. C, 2005, “Statistical Confidence using Minimum Rank
Zabel, V, 2005b, “Application of Wavelet Decompositions’ Energy Com- Perturbation Theory”, Mechanical Systems and Signal Processing,
st
ponents to Damage Detection”, In Proceedings of the 1 International Vol. article in press (available online).
Operational Modal Analysis Conference (IOMAC 2005), Copenha- Zimmerman, D. C., Kaouk, M. and Simmermacher, T, 1995a*, “On the
gen, Denmark, 519–526. Role of Engineering Insight and Judgement Structural Damage Detec-
th
Zak, A., Krawczuk, M. and Ostachowicz, W, 1999**, “Vibration of a Lam- tion”, In Proceedings of the 13 International Modal Analysis Con-
inated Composite Plate with Closing Delamination”, In Proceedings ference (IMAC XIII), Nashville, Tennessee, USA, 414–420.
rd
of the 3 International Conference on Damage Assessment of Struc- Zimmerman, D. C., Simmermacher, T. and Kaouk, M, 1995b*, “Structural
tures (DAMAS 99), Dublin, Ireland, 17–26. Damage Detection using Frequency Response Functions”, In Proceed-
Zang, C., Friswell, M. I. and Imregun, M, 2003a, “Structural Health Moni- ings of the 13th International Modal Analysis Conference (IMAC
toring and Damage Assessment using Measured FRFs from Multiple XIII), Nashville, Tennessee, USA, 179–185.


A Review of Vibration-Based Structural Health Monitoring

Uploaded by

Copyright:

Available Formats

You might also like

A Review of Vibration-Based Structural Health Monitoring

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Review of Vibration-Based Structural Health Monitoring

Uploaded by

Copyright:

Available Formats

The Shock and Vibration Digest

A Review of Vibration-based Structural Health Monitoring with Special Emphasis on Composite

The online version of this article can be found at:

Email Alerts: http://svd.sagepub.com/cgi/alerts

Citations (this article cites 99 articles hosted on the

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

A Review of Vibration-based Structural Health

D. Montalvão, N.M.M. Maia and A.M.R. Ribeiro

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Messina et al. (1996) showed that it is possible to locate ∑ αij ( ω 2 )αij* ( ω 1 )

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Another class of damage detection methods is based on  NT NT

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

cost function. First, a genetic search is applied to the cost ∆U

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Zabel (2004, 2005a, b) applied a wavelet-based damage 2.13. Hilbert Transform

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

linear sensitivity theory for estimation of the variance and

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

Downloaded from http://svd.sagepub.com at PENNSYLVANIA STATE UNIV on April 14, 2008

You might also like