Dynamic Testing and Monitoring of Bridges - Paper

STRUCTURAL CONTROL AND HEALTH MONITORING
Struct. Control Health Monit. 2012

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/stc.1516
Recent perspectives in dynamic testing and monitoring of bridges
A. Cunha, E. Caetano*,†, F. Magalhães and C. Moutinho

Faculty of Engineering (FEUP), University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal
SUMMARY
The paper reviews some of the most recent aspects involved in the dynamic testing and continuous monitoring of bridges.
This includes a discussion of testing techniques, instrumentation, modal identification and damage detection. On
the basis of their experience, the authors described several case studies in which some of the most recent developments
have been used to accomplish different purposes: the Millau Viaduct, in France, the Pedro e Inês footbridge and
the Infante D. Henrique Bridge, in Portugal, and the Grande Ravine Viaduct, at the Reunion Island. Copyright ©
2012 John Wiley & Sons, Ltd.
Received 3 November 2011; Revised 27 June 2012; Accepted 28 June 2012
KEY WORDS: dynamic testing, monitoring, bridge testing, modal identification, damage detection, wind measurements
1. INTRODUCTION
From the establishment of experimental stress analysis in the 1940s up to the present time, experimental
dynamic analysis has increasingly been incorporated in civil engineering applications. The consolidation
of testing techniques, the improvement and diversified use of sensors and acquisition systems, the devel-
opment of modern algorithms for identification and detection of damage and the increased storage and
data transmission capacities have determined an increased understanding of the structural behaviour of
complex structures. In fact, the growing importance of the use of experimental dynamics has led to the
present trend to integrate it in management construction programs, aiming at supporting decisions related
with the implementation of rehabilitation/strengthening tasks upon detection of structural damage. It is
then relevant to discuss the most recent developments in the various aspects involving the dynamic testing
and monitoring of bridge structures and to understand how they can be used to solve different problems
related with bridge design, construction and management.
Having these aspects into consideration, this paper presents some recent perspectives that, according to
the authors’ experience, represent important trends in terms of types and techniques of dynamic testing
and monitoring, sensors and data acquisition systems, and modal identification. Considering the possible
use of continuous monitoring with the purpose of detecting damage, automated identification, correlation
of data with environmental and operational factors, and techniques to assess the health condition of a
bridge are also discussed. The description of a series of case studies based on the authors’ experience
and focusing on relevant bridge structures shows the importance of the use of some of these most recent
developments to accomplish different purposes in the study of bridges. The use of ambient vibration tests
based on an efficient nonwired instrumentation is illustrated with the case of a very long bridge, the Millau
Viaduct, in France. Natural frequencies, vibration modes and damping ratios were identified with very
high accuracy from a test conducted on two single measurement days. In the Pedro and Inês footbridge,
in Portugal, a continuous dynamic monitoring has successfully been in use to detect excessive vibrations
*Correspondence to: E. Caetano, Faculty of Engineering (FEUP), University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal.
†
E-mail: ecaetano@fe.up.pt
Copyright © 2012 John Wiley & Sons, Ltd.

A. CUNHA ET AL.
induced by pedestrians for a period of 5 years. The systematic measurement of accelerations using a
simple acquisition system and data transmission facilities allows the permanent characterization of
the behaviour of the footbridge and of the efficiency of a set of tuned mass dampers (TMDs) installed
in the deck to mitigate lateral and vertical vibrations. The pilot monitoring system installed at Infante D.
Henrique Bridge, in Porto, has been active since 2007, aiming at testing and implementing damage detec-
tion techniques. The systematic automatic identification of modal parameters has permitted correlations
with different environmental and operational factors and the construction of models that remove those
effects, allowing the detection of small damage. Finally, the continuous monitoring of the Grande Ravine
Viaduct was implemented in 2009 with the purpose of using prototype measurements to validate the
characterization of wind models and the assumptions and force coefficients used in design, based on
wind tunnel tests. This is actually possible, as the modal characterization of the bridge is very accurate,
allowing for a validation of a numerical model. Therefore, the correlation of the prototype measured
response with the response calculated on the basis of wind loads characterized on the basis of site measure-
ments will allow validation of coefficients extracted from wind tunnel measurements tests.
2. EXPERIMENTAL APPROACH
2.1. From forced vibration to ambient and free vibration testing
About 30 years ago, the experimental identification of relevant dynamic properties of large structures,
such as bridges [1,2] or dams [3,4], was carried out on the basis of conventional modal testing procedures
previously developed in the fields of mechanical and aeronautical engineering. Such tests involved the
estimation of a set of frequency response functions (FRFs) relating the applied force and corresponding
response at several pairs of points along the structure with enough high spatial and frequency reso-
lution and required the use of an instrumentation chain for structural excitation, data acquisition
and signal processing.
Although forced vibration tests may lead to very accurate modal estimates, they present a strong
drawback when dealing with large structures, which stems from the difficulty in exciting the most signifi-
cant modes of vibration in a low range of frequencies with sufficient energy and in a controlled manner.
Fortunately, recent technological developments in transducer and digitizer technology have made it
possible to accurately measure the very low levels of dynamic response induced by ambient excitation
such as wind or traffic. This has stimulated the performance of ambient vibration tests, which became
an alternative of great importance in the field of civil engineering, allowing accurate identification of
modal properties of large structures at construction, commissioning or rehabilitation stages without
interruption of normal traffic [5].
In this context, some pioneering studies have been developed still in the era of the analogue equipment
(e.g. [6–12]). However, the recent use of sensors specially suitable for ambient vibration measurements in
large structures, conjugated with high level digitizers with local storage and time synchronization capabil-
ities, made feasible the accurate modal testing of large structures in a short time and in a rather comfortable
way, as happened at Vasco da Gama Bridge [13], at Millau Viaduct [14] or at Humber Bridge [15].
In some applications, for instance related with the aerodynamic behaviour of slender structures, the
accurate identification of modal damping factors is a major problem in the identification process as a result
of the considerably larger scatter of modal damping estimates with regard to the corresponding natural
frequencies and modal shapes counterparts. This is also true because the viscous damping assumption
does not correspond exactly to real damping characteristics. Modal damping ratios increase gradually with
levels of oscillation. This concern has led frequently to the performance of free vibration tests, as
happened at several outstanding cable-stayed bridges (e.g. Normandy, Vasco da Gama or Millau Bridges).
At Vasco da Gama Bridge, the test was made suspending a barge with a mass of 60 t from an eccentric
point at the deck (Figure 1) [13], at one-third span upstream, through a cable, which was cut when the tide
became low and the wind speed inferior to 3 m/s, to avoid the influence of aerodynamic damping. At
Millau Viaduct, the test [14] consisted of the sudden rupture of a tensioned cable connected from the
ground to the deck. This operation was conducted first with a cable tensioned at 600 kN and then repeated
with a tensioning of the cable to 1000 kN. In both cases, the wind speed was low, oscillating between 2
and 5 m/s. It is worth noting that such type of impulsive excitation leads to a considerable increase of
Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2012;
DOI: 10.1002/stc
RECENT PERSPECTIVES IN DYNAMIC TESTING AND MONITORING OF BRIDGES
Figure 1. Free vibration tests at Vasco da Gama Bridge (left) and Millau Viaduct (right).
the level of vibration with regard to ambient excitation. For instance, the root mean square value of vertical
acceleration at the reference section at Vasco da Gama Bridge during the ambient vibration test was
0.017 m/s2, whereas the peak vertical acceleration during the free vibration test reached about 0.30 m/s2.
At Millau Viaduct, the measured value of maximum vertical acceleration was about 1 m/s2, and the
corresponding peak displacement (obtained by integration) was about 50 mm.
2.2. From dynamic testing to continuous dynamic monitoring

During several decades, experimental dynamics was essentially used to identify the most relevant dynamic
parameters of large civil structures at a certain stage of their lifetime, normally at the commissioning stage,
with the main purpose of establishing correlations with numerical predictions or in some cases developing
the updating of finite element models [8,16,13].
It was then considered that such tests could characterize the baseline condition of the structural behaviour,
allowing subsequent detection of structural changes. Therefore, ambient vibration tests became gradually
also more common before and after rehabilitation works [17,18], and even many research attempts have
been developed to detect early damage on the basis of variations of modal parameters estimates, despite
the adverse influence of several environmental factors [19–21].
However, the remarkable technological progress that recently occurred in the field of data acquisition
systems and information transmission through the Internet made it feasible a continuous dynamic
monitoring of the structural behaviour [22,23], which may complement other components of structural
monitoring, such as the monitoring of loads (e.g. wind and traffic loads), static behaviour and dura-
bility [24]. These systems can presently play a very important role in the understanding of the
structural behaviour either during the bridge construction or during the service lifetime, as will be
shown with the examples presented in this paper.
3. INSTRUMENTATION
3.1. From mechanical excitation devices to ambient vibration measurement systems
Conventional forced vibration bridge tests, normally involving the construction of FRFs, require the
use of appropriate dynamic excitation devices.
In small-size and medium-size structures, the excitation can be induced by an impulse hammer similar to
those currently used in mechanical engineering. This device has the advantage of providing a wide-band
input that is able to stimulate different modes of vibration. The main drawbacks are the relatively
low frequency resolution of the spectral estimates (which can preclude the accurate estimation of
modal damping factors) and the lack of energy to excite some relevant modes of vibration. Because
of this problem, some laboratories have built special impulse devices specifically designed to excite
bridges (Figure 2(a)). An alternative, also derived from mechanical engineering, is the use of large
electrodynamic shakers, which can apply a large variety of input signals (random, multisine, and
DOI: 10.1002/stc
A. CUNHA ET AL.
(a) (b) (c)

Figure 2. (a) Impulse device used by K.U. Leuven; (b) servo-hydraulic shaker used by Eidegenössische Materialprufungs-
und Forschungsanstalt (EMPA); (c) servo-hydraulic shakers used in the forced vibration tests of Tatara Bridge (http://www.
hsba.go.jp).
others) when duly controlled both in frequency and amplitude with the use of a signal generator and a
power amplifier. The shakers have the capacity to excite structures in a lower frequency range and
higher frequency resolution. The possibility of applying sinusoidal forces allows for the excitation
of the structure at resonance frequencies and, consequently, for a direct identification of mode shapes.
The controlled excitation of large civil engineering structures requires the use of heavy excitation
equipment. Although eccentric mass vibrators can be used for that purpose, the best option, in terms
of providing a wide-band excitation over the most interesting frequency range for large civil structures,
is the use of servo-hydraulic shakers (Figure 2(b)).
However, in very large, flexible structures such as cable-stayed or suspension bridges, the forced
excitation requires extremely heavy and expensive equipment (Figure 2(c)) usually not available in
most dynamic laboratories. At the same time, the response induced by ambient loads may be then
comparably large. If convenient assumptions are made with respect to the characterization of natural
excitation, mechanical excitation devices become unnecessary. The performance of ambient vibration
tests involves then the measurement of the structural response under ambient excitation by one or more
reference sensors at fixed positions and with a set of roving sensors at different measurement points
along the structure in different setups. The number of points used is conditioned by the spatial resolu-
tion needed to characterize appropriately the shape of the most relevant modes of vibration (according
to preliminary finite element modelling), whereas the reference points must be far enough from the
corresponding nodal points.
The success of ambient vibration measurements strongly depends on the type and characteristics of
equipment used. From this point of view, an appropriate choice of the accelerometers and digitizers is
especially important. The accelerometers must fulfil the following specifications: (i) frequency bandwidth
DC – 50 Hz; (ii) very low peak-to-peak noise (if possible, inferior to about 2 micro-g); (iii) high sensitivity
(at least 1 V/g); and (iv) low full scale range (+/ 0.5 g, lower or selectable). This means that some conflict
can occur if the measurement system must be also prepared to record possible seismic events, which may
require high full scale range (+/ 2.0 g or selectable) and a high dynamic range. The choice of the sensi-
tivity of the accelerometers must be conjugated with the characteristics of the digitizer (with at least 16
bit), so as to achieve a resolution of the digitization scale compatible with the accurate measurement of
very low levels of ambient structural response. For this purpose, it is convenient to be aware of the typical
order of magnitude of measured accelerations under ambient excitation that can be found in applications in
bridges and other special structures. As an example, it is referred that the long-term dynamic monitoring
system installed at Infante D. Henrique Bridge [25], over Douro river at Porto, has provided peak accelera-
tions varying in the interval 0.01–0.05 m/s2 and root mean square values in the range 0.0005–0.003 m/s2. It
should be mentioned that these values can be however considerably lower in other applications involving
rather stiff structures.
3.2. From conventional star data acquisition systems to decentralized and wireless solutions
The data acquisition systems traditionally used either, in a first instance, for forced vibration tests or, more
recently, for ambient or free vibration tests were always characterized by a ‘star’ system architecture.
DOI: 10.1002/stc
This means that a typical ambient or free vibration bridge testing system comprehended usually [26]
the following: (i) a set of sensors, commonly forced balance accelerometers; (ii) amplifier and filter
units, covering high gains and providing selectable low-pass filtering with low cut-off frequencies to
remove all unwanted higher frequencies from the signals; (iii) an analogue to digital converter capable
of digitizing the analogue signals, conveniently controlled by software so as to permit the acquisition
of very long records; and (iv) one computer coordinating the data acquisition and eventually a second
one to perform all on site data analysis, mode shape animation and printing.
Although such conventional ambient vibration systems have been used with success [10,27], they
present a strong drawback related with some lack of portability and with the necessity of developing
a rather hard and tedious setup, using many hundreds of meters of electric cables, which should be
tough, shielded and ensure a minimal signal loss and interference over large distances.
This important drawback of traditional ‘star’ system configurations motivated the development of more
comfortable and portable alternative decentralized systems, minimizing or even totally avoiding the use of
electrical cables. In this perspective, a pioneering work has been developed at Laboratório Nacional de
Engenharia Civil (Portugal) (LNEC) and Faculty of Engineering of the University of Porto (FEUP) [5],
applying a system composed by several independent triaxial seismographs conveniently synchronized
on the basis of GPS sensors (Figure 3). Additionally, important research efforts have been also developed
to achieve really efficient wireless systems [28], although the technological solutions currently available
still face some limitations in terms of accuracy, reliability and capabilities.
Concerning the more recent development and implementation of continuous dynamic monitoring
systems in bridges, a similar trend to decentralized system architectures has been also observed owing
to similar reasons. In that sense, several monitoring equipment suppliers present nowadays monitoring
solutions based on the use of several independent digitizers, duly synchronized by GPS and
interconnected by Ethernet cables. These digitizers allow the connection of a certain number of dynamic
channels, are usually equipped with a 24-bit analogue-to-digital converter and permit the simultaneous
telemetry of the acquired data to a central site and a link to a local recording unit [25]. Alternatively, flexible
monitoring solutions based on the use of local digitization close to each sensor and single cable digital data
transmission have been also developed and implemented with success [29].
3.3. From conventional sensors to noncontact measurement techniques

From the former [6,7] to present vibration tests [13–15], force balance accelerometers have been used and
established as the most robust and adequate sensors for bridge response measurements. The evolution of
electronics has led to a progressive improvement of the sensitivity and signal-to-noise ratio of these
devices so that accelerometers 165 dB in dynamic range and 2 micro-g peak-to-peak noise can presently
be found in the market.
Even though the use of conventional force balance accelerometers provides extremely high quality
response measurements, some difficulties can be encountered in practice with these or with other conven-
tional sensors, as piezoelectric accelerometers. Such is the case when a very large number of measurement
points are required or when the access to these points is limited. To exemplify, consider the identification
of cable force in cable-stayed bridges based on the indirect measurement of frequencies. Attaching an
accelerometer successively at every cable to obtain 10-min to 1-h records of acceleration is the usual
procedure, which implies the use of a portable data acquisition system and the measurement at a location
sufficiently distant from the anchorage. Ideally, the measurement point should be easily accessible in order
to render this procedure quick. However, when instrumenting short cables or cables with installed
Figure 3. Recorders, sensors and GPS antennae during cross-calibration measurement at Humber suspension bridge.
DOI: 10.1002/stc
A. CUNHA ET AL.
dampers, there may exist the need to use cranes or other elevating machines to access points with measurable
vibrations. In that circumstance, the use of noncontact devices may be of high interest. The laser sen-
sor, based on the Doppler effect, has been used to identify the force installed in the cables of the Vasco
da Gama Bridge [30]. Except for the fact that relative velocity records are obtained, data processing
techniques are identical to those used for conventional accelerometers. More recently, cable measure-
ments have been successfully performed using radar techniques and a microwave interferometer [31].
This system allows for the simultaneous instrumentation of various cables from a single location and
has proven to provide high accuracy results by comparison with conventional accelerometers [32].
Amongst the diverse systems available for noncontact measurements, reference is still made to
image-based techniques and to the measurement of midspan displacements based on a video camera
installed at one tower of a cable-stayed bridge and on algorithms for target movement description
[33]. A recent implementation requiring no targets has been used in the monitoring of cable vibra-
tions. Such system was developed to overcome the difficulty raised by the need of characterizing
the sources and patterns of cable vibrations in a cable-stayed bridge [34]. Mounting accelerometers
on the many vibrating cables would be extremely onerous, considering the high number of sensors
and the need to transport electrical signals to a central computer. Image-based sensors have many
limitations, in particular those referring to the accuracy and frequency resolution. But stay cables
vibrate with high amplitudes, and the most critical frequencies are in general within the reach of
modern video cameras. In the case referred in [34], data transmission limitations and difficulty in
the mounting of video cameras on the bridge motivated the observation of the cables from a building
located about 850 m from the bridge (Figure 4). The camera, connected to a laptop, performed
continuous measurements during several months, in the period 0900–1600. In those periods, with the
optical flow method used, velocity time series associated to a set of preselected points were formed,
which were subsequently integrated, rotated and scaled to provide the temporal variation of displace-
ments. Figure 4 shows the daily variation of maximum amplitudes observed in a set of stay cables in
records of 20 min, which allowed the detection of episodes of cable vibrations, characterizing the
corresponding duration and the correlation with direction and wind velocity, according to information
provided by a local meteorological station.
24/11/2007
1J
1.8
2J
Amplitude of vibration
1.6
3J
1.4
4J
1.2
4M
1
12M
0.8
0.6
0.4
0.2
0
8:24 9:36 10:48 12:00 13:12 14:24 15:36 16:48
Time
Figure 4. Application of vision system to the dynamic monitoring of the Guadiana Bridge. Location of the video
camera, selection of measurement points and envelope of response of six stay cables on 1 day, from 0900 to 1600.
DOI: 10.1002/stc
4. MODAL IDENTIFICATION
4.1. From experimental modal analysis to operational modal analysis
The first attempts to identify modal properties of large civil structures were based on well-established
methods of experimental modal analysis, previously developed and applied in the field of mechanical
engineering. These methods imply the simultaneous measurement of a controlled excitation and the
corresponding structural response.
In this context, there is a wide variety of input–output modal identification methods whose application
relies either on estimates of a set of FRFs or on the corresponding impulse response functions, which can
be obtained through the inverse Fourier transform. These methods are based on the fitting of theoretical
functions to the measured data, employing different optimization procedures and different levels of
simplification. Accordingly, they are usually classified according to the following criteria: (i) domain
of application (time or frequency); (ii) type of formulation (indirect or modal and direct); (iii) number
of modes analyzed (SDOF or MDOF); and (iv) number of inputs and type of estimates (SISO, single-input
multi-output, MIMO, and multi-input single-output).
Early methods of identification were developed for the frequency domain. The simplest SDOF formula-
tions (e.g. peak amplitude, curve-fit and inverse methods) fit measured data with a theoretical FRF of a
SDOF system in the vicinity of each resonant frequency, neglecting the contribution of resonant modes.
More sophisticated MDOF methods – rational fraction polynomial, complex exponential (CE) frequency
domain and polyreference frequency domain – involve the global fitting of the measured data by theoretical
FRFs in a wide range of frequencies.
Time-domain methods, which tend to provide the best results when a large frequency range or a large
number of modes exist in the data, were developed because of limitations in the frequency resolution
of spectral estimates and leakage errors in the estimates. The most widely known methods are either
indirect – CE, least-squares CE, polyreference CE, Ibrahim time domain (ITD), eigensystem realization
algorithm or direct autoregressive moving average.
The gradual development of all these methods, which are extensively described in [35], shows a trend to
automated systems of acquisition, analysis, processing and identification, instead of the initially developed
interactive codes. Beyond that, the best-performing methods have been implemented in robust modal
analysis software [36].
However, the increasing interest and popularity of ambient vibration testing of large bridges and
special civil structures led to a remarkable recent development of a new family of modal identification
methods under operational conditions, called operational modal analysis (OMA).
Ambient excitation usually provides multiple inputs and a wide-band frequency content thus stimulating
a significant number of vibration modes. For simplicity, output-only modal identification methods assume
that the excitation input is a zero-mean Gaussian white noise. This means that real excitation can be
expressed as the output of a suitable filter excited with white noise input.
There are two main groups of output-only modal identification methods – nonparametric methods,
essentially developed in the frequency domain, and parametric methods, most of them in the time domain.
Although already applied for some decades to the modal identification of buildings [37,9] and bridges [8,38],
it was only in the beginning of the 1990s that the basic frequency-domain method (peak picking) was con-
veniently implemented [26]. This methodology leads to estimates of operational mode shapes and is based
on the construction of average normalized power spectral densities and ambient response transfer functions
involving all the measurement points. The software for modal identification and visualization used at the
University of British Columbia and Eidegenössische Materialprufungs- und Forschungsanstalt
(EMPA) [26] was based on this procedure. The frequency-domain approach was subsequently improved
[39,40] by performing a single-value decomposition of the matrix of response spectra to obtain power
spectral densities of a set of SDOF systems. This method, frequency-domain decomposition (FDD),
was implemented by Brincker et al. [41] and enhanced [42] to extract modal damping factor estimates.
In this last approach (enhanced FDD (EFDD)), these estimates are obtained through inspection of the
decay of autocorrelation functions evaluated by performing the inverse Fourier transform of the SDOF
systems’ power spectral densities.
Time-domain parametric methods involve the choice of an appropriate mathematical model to idealize
the dynamic structural behaviour (usually time-discrete, state-space stochastic models, autoregressive
DOI: 10.1002/stc
A. CUNHA ET AL.
moving average vector or autoregressive vector) and the identification of the values of the modal
parameters so that the model fits the experimental data as well as possible, following some appropriate
criterion. These methods can be directly applied to discrete response time series or, alternatively, to
response correlation functions. The evaluation of these functions can be made on the basis of their defini-
tion using the FFT algorithm [43] or applying the random decrement method (RD) [44]. A peculiar aspect
of output-only modal identification based on the fitting of response correlation functions is the possibility
to use methods that stem from classical input–output identification based on impulse response functions.
Some of these methods are the ITD [45], the multiple reference ITD [46], the least-squares CE [47], the
polyreference CE [48] or the covariance-driven stochastic subspace identification (SSI-COV) [49,50].
An alternative method that allows direct application to the response time series is the data-driven SSI
[51]. Note that the RD technique usually associated with the application of time-domain methods such
as Ibrahim’s can also be the base for the application of frequency-domain methods (such as PP, FDD
or EFDD).This leads to free vibration responses from which power spectral densities can be evaluated
using the FFT algorithm [52], thus reducing noise effects (methods RD-PP, RD-FDD and RD-EFDD).
A new frequency-domain parametric method (p-LSCF) was recently introduced by LMS [53]. It
operates on spectra or half spectra (i.e. the Fourier transforms of the positive time lags of the correlation
functions), and its main advantage consists in yielding extremely clear stabilization diagrams, making an
automation of the parameter identification process rather straightforward and possibly enabling continuous
monitoring of structural dynamic properties. However, it is worth noting that the current operational modal
testing practice, where the modal estimates are obtained from output-only data, does not allow the evalua-
tion of modal scaling factors, the frequency content of the excitation not being controlled either. To
overcome these drawbacks, a combined deterministic–stochastic modal identification approach
(OMA with exogenous inputs) has been also recently proposed [54].
5. VIBRATION-BASED DAMAGE DETECTION

Because ambient excitation is always present, the techniques used for identification of modal parameters
only based on structural responses (OMA) can be applied to continuously process time series acquired by
a permanent installation of a set of accelerometers. This permits the tracking of the variation with time of
modal parameters. As structural deterioration or damage due to some extreme event imply a stiffness
reduction and consequently a decrease of natural frequencies, an accurate characterization of their varia-
tion with time can be used to detect structural problems in bridges, as proposed by different investigators
in the past years [55,56]. This is a core idea behind vibration-based health monitoring systems. These
systems may also be based on other dynamic features of the structure, as for instance the variability of
identified mode shapes, which is a significantly different issue, as characterized in [57], but this discussion
is only focused on the ones that analyze fluctuations of the natural frequencies.
The processing of data collected by dynamic monitoring systems with the aim of identifying structural
deficiencies comprehends not only the identification of modal parameters but also removal of the environ-
mental and operational effects on natural frequencies. Considering the purpose of identifying very small
structural changes in early phases of development, the minute effects of, for instance, the temperature or
the traffic intensity over a bridge on the natural frequencies have to be minimized. Therefore, vibration-based
damage detection systems have to include the three components illustrated in Figure 5: (i) automatic
Acceleration time series undamaged damaged

Natural frequencies
(ii)
…
(i)
(iii)
time time
(a) (b) (c)
Figure 5. Main processing steps of a vibration-based health monitoring system: (a) automatic identification of natural
frequencies; (b) removal of environmental and operational effects; (c) index for detection of damage.
DOI: 10.1002/stc
identification of modal parameters; (ii) removal of the environmental and operational effects on modal
parameters; and (iii) calculation of an index able to flag relevant frequency shifts.
5.1. From operational modal analysis to automated operational modal analysis

The standard algorithms for identification referred in the previous section require human intervention at
some stages of analysis. The evolution of health monitoring systems has implied research in order to
develop algorithms for automated extraction of accurate estimates of the structure modal parameters
from continuously recorded responses during normal operation conditions.
An automated version of the FDD method has been proposed by Brincker et al. [58] and used in a full
scale application described in [25]. This nonparametric method has some limitations than can however be
overcome by parametric algorithms, developed either in the time or frequency domains, such as the
SSI-COV and the p-LSCF.
One of the difficulties of parametric methods is the need to define the order of the model that better fits
the experimental data. This is usually solved by the fitting of several models with orders varying in an
interval previously fixed in a conservative way. The results provided by the adopted models are then
organized in stabilization diagrams. These permit the visual identification of the modes that appear in most
of the models with consistent frequency, mode shape and damping. Modes that only appear in some
models are considered spurious. In the context of structural health monitoring (SHM) applications, the
visual analysis of these diagrams has to be replaced by an algorithm. This can be carried out with the
use of hierarchical clustering algorithms as suggested in [59]. With this procedure followed, all the mode
estimates provided by the identification algorithm are compared using a similarity measure that depends
on the natural frequencies and on an index that measures the resemblance between mode shape estimates.
The hierarchical clustering technique is used to group the mode estimates associated with the same
physical vibration mode and permits the separation of numerical or noise estimates.
5.2. Removal of the influence of environmental and operational factors on modal estimates
After installation of the dynamic monitoring system and implementation of routines for the automatic
identification of modal parameters, it is necessary to study the time variation of the modal parameters
with the final goal of defining strategies for the automatic identification of abnormal values that might
be indicators of structural damage.
In a first stage, it is important to identify the ambient and operational factors with greater influence on
the natural frequencies, which may disturb the detection of structural damage based on the variability of
modal frequencies. In the case of bridges, obvious candidates are the temperatures measured on structural
elements, the vibration amplitude or the traffic intensity.
One possible approach to remove the effects of environmental and operational factors on natural
frequencies could be the establishment of models capable of representing the physical phenomena behind
the frequency changes [60], which would require an appropriate modelling of different physical aspects,
such as the dependency of structural materials elasticity modulus on temperature, the influence of expan-
sion or contraction movements driven by temperature on the support conditions, opening and closing of
cracks that may exist in concrete elements motivated by curvature variations due to differential tempera-
tures and freezing of the soil around the structure supports. However, this kind of approach is not adequate
in the context of SHM programs, because it would require the construction of a very complex model for
each new application, and even so, some effects would not be correctly represented.
Accordingly, instead of trying to understand the physics of the problem, it is possible to rely on black
box models, whose parameters are tuned using a large number of observations, to establish relations
between the natural frequencies and the factors that may influence them (input–output methods).
However, this approach imposes the measurement of the factors that may affect the modal frequencies
variability. As the selection of these factors is not always straightforward, it is possible to follow an
alternative approach, based on statistical tools that allow the correction of natural frequency estimates
without the necessity of measuring environmental and operational factors (output-only methods).
The input–output methods are often based on the development of linear regression analyses between
identified natural frequencies and environmental or operational factors measured by the monitoring
DOI: 10.1002/stc
A. CUNHA ET AL.
system. Classical statistical methods with different level of complexity are the multiple linear regres-
sion [61], the multivariate linear regression or dynamic regression models such as the autoregressive
with exogenous input model [62,49]. Nonlinear methods and methods based on the statistic learning
theory [63], such as the support vector machine or neural networks can also be used. Ni et al. [64],
Hua et al. [65], Zhou et al. [66], Ni et al. [67] and Zhou et al. [68] described this kind of methods
and illustrated their application to continuous monitoring data from the Ting Kau cable-stayed bridge.
The output-only methods often correspond to a set of methods that eliminate the effects of environmental
and operational factors on features extracted from data collected by a dynamic monitoring system through
the decomposition of a correlation or covariance matrix of the time variation of the structural features over
a reference period, as is the case of principal component analysis (PCA) or factor analysis [69,70]. These are
described and applied for instance in [71–73,69]. Alternatively, it is also possible to achieve the same goal by
the direct decomposition of time series with the extracted features or through the use of neural networks
[74,75].
Thus, some months of observation permit to have records associated with a large range of environmental
and operational conditions, which can be used for the establishment of multiple regression models where
the estimated natural frequencies are correlated with the other observed factors. Models based on past
observations are then used to eliminate the effect of the variables considered independent on the future esti-
mations of the natural frequencies. The influence of factors with consequences on the natural frequencies
that are not directly measured may be minimized through the use of PCA or factor analysis.
At this stage, the plots of the natural frequencies over time should be marked by very narrow range
around a mean value. So, the occurrence of any structural problem with consequences on the monitored
natural frequencies should be easily detected.
5.3. Damage detection

Vibration-based damage detection can be efficiently performed on the basis of the observation of the time
variability of modal frequency estimates. However, it is important to take into account that this variability
can be a consequence of three main factors: (i) uncertainties associated to the system identification proce-
dure used; (ii) effects of environmental (e.g. temperature) and operational (e.g. traffic intensity) factors;
and (iii) structural damage. The removal of the environmental and operational effects using one of the
methods mentioned in the previous section and some thousands of estimates during an initial significant
period (e.g. 1 year) can lead to much more stable estimates of natural frequencies. The order of magnitude
of the scatter of these corrected modal frequency estimates during an initial period of healthy structure
defines the order of magnitude of damage that can be reliably detected in the long term, based on the
observation of natural frequency shifts.
However, to avoid the analysis of multiple graphics (one for each frequency), the deviation of all the
monitored natural frequencies with regard to their reference values can be summarized in an index with
the use of the concept behind the control charts normally used in the context of industrial production
for quality control. A control chart typically consists of data plotted in time order and horizontal lines.
These are designated as control limits, indicating the amount of variation due to common causes. There-
fore, an observation outside the control region is understood as an out-of-control observation, that is, an
observation suggesting a special cause of variation. In the context of SHM, this cause of variation may
be associated with the occurrence of damage. In [76–78], several alternative multivariate control charts
are detailed and applied. A commonly used chart is the Shewhart or T2 [70].
6. DYNAMIC TESTING AND CONTINUOUS MONITORING: CASE STUDIES

6.1. Dynamic testing
Accurate modal identification of bridges through dynamic tests can be presently conducted with different
purposes, such as the following: (i) the calibration and validation of finite element models used in the
design of new structures; (ii) the confirmation or understanding of structural changes introduced by reha-
bilitation works; (iii) the evaluation of modal damping ratios, which may be responsible for high levels of
DOI: 10.1002/stc
vibration induced by wind or traffic loads; and (iv) the design, tuning and evaluation of efficiency of
vibration control devices.
In most of those situations, ambient and/or free vibration tests provide the simplest experimental
approach, as evidenced by a large number of case studies [5]. However, in some specific applications,
specially related with the implementation of vibration control devices, forced vibration tests still
remain the most reliable solution, as shown in [79,80], concerning the evaluation of the efficiency of
TMDs installed to control lateral lock-in problems at Pedro e Inês footbridge (Figure 6).
The potential of OMA techniques can be easily illustrated with the dynamic tests of Millau Viaduct.
This is a large cable-stayed road-bridge spanning the valley of the River Tarn near Millau in southern
France (Figure 7). Designed by the architect Norman Foster and the bridge engineer Michel Virlogeux,
it is the tallest vehicular bridge in the world, with one pylon’s top rising at 343 m above the tarn level.
The eight cable-stayed spans with a total length of 2460 m make this bridge also the longest cable-stayed
bridge in the world. The 32 m in width and 4.2 m in thickness deck has a continuous steel box section.
The six central spans have a length of 342 m, whereas the outer spans are 204 m long. The seven concrete
piers range in height from 77 to 245 m, supporting 87-m tall steel pylons.
The ambient vibration test was performed at the commissioning stage by the Laboratory of Vibrations
and Structural Monitoring (www.fe.up.pt/vibest) of FEUP under coordination of the Centre Scientifique et
Technique du Bâtiment, Nantes (CSTB), and was developed using four seismographs, including triaxial
accelerometers. These seismographs, duly synchronized using GPS sensors, were placed outside the deck.
With the purpose of identifying as many modes of vibration as possible, essentially of vertical and trans-
versal bending nature, two of those triaxial sensors were used as references (sections R1 and R2, Figure 8)
and kept in fixed positions, whereas the other two were successively placed in each one of the remaining
26 sections schematically indicated in Figure 8. The seismographs were programmed using a laptop to
acquire signals with a sampling rate of 100 Hz in periods of 960 s every 20 min, the last 4 min being used
Figure 6. Pedro e Inês footbridge, Portugal.
Figure 7. Millau Viaduct, France.
Figure 8. Instrumented sections of the deck in the ambient vibration test.
DOI: 10.1002/stc
A. CUNHA ET AL.
to change the position of the moving sensors for the subsequent measurement setup. The test was com-
pleted at 0730, and during this period, the wind speed was always very low and the traffic very sporadic,
as the bridge had not been open to traffic yet.
Preliminary frequency-domain analysis was developed in order to identify the most relevant natural
frequencies and to analyze the variation of the frequency content of the collected time series during the
13 setups. For each setup, spectral matrices were calculated for vertical and for lateral accelerations.
Then, a singular value decomposition of these matrices was performed.
Figure 9 shows the average singular values and colour maps with the variation of the first singular
values during the ambient vibration test, for both directions. These plots show the existence of a large
number of modes in the frequency range 0.1–1.0 Hz.
The experimental data obtained were subsequently processed by different alternative stochastic modal
identification methods, namely p-LSCF (PolyMax) and SSI-COV (with single or multiple analysis of all
test setups). Figure 10 shows the stabilization diagram provided by SSI-COV, for the simultaneous analysis
of all experimental data, which involved the singular value decomposition of a matrix with dimensions
10 400 400. All the natural frequencies of the viaduct can be identified from this very ‘clean’ diagram.
Concerning damping, it is well known that the estimates of modal damping ratios obtained from
ambient vibration tests have always a significantly higher scatter when compared with the vari-
ability of the corresponding natural frequencies and mode shapes. However, looking at Figure 11,
that resumes the results of all the applied techniques in terms of modal damping ratios estimates
(interval of variation defined by mean value and standard deviation), we can conclude that when
sophisticated parametric output-only modal identification methods (such as p-LSCF (PolyMax) or
SSI-COV) are used in Ambient Vibration Test (AVT) tests, the modal damping ratio estimates
obtained show in general a rather good agreement.
Vertical direction Lateral direction

12 12
10 10
Setup
Setup
8 8
6 6
4 4
2 2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Frequency (Hz) Frequency (Hz)
Figure 9. Average singular values of the spectra matrices and colour maps with the variation of the first singular
values during the ambient vibration test.
Model
order Stabilization Diagram All Poles Stable Freq. Stable Damp. MAC
100
90
80
70
60
50
40
30
20
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Frequency (Hz)
Figure 10. Stabilization diagram of the single analysis – covariance-driven stochastic subspace identification.
DOI: 10.1002/stc
Modal Damp. Ratios (%)

1.2 PolyMax SSI-COV, multiple SSI-COV, Single
0.9
0.6
0.3
0.0
0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75
Natural Frequencies (Hz)
Figure 11. Comparison of modal damping ratios estimates provided by the applied methods.
The accurate identification of a significant number of natural frequencies and modes of vibration
in the frequency range 0–1 Hz enabled the subsequent development of a finite element model corre-
lation analysis. Figures 12 and 13 superimpose 20 modes identified in that range (10 vertical and 10
transversal bending modes) using the SSI-COV method (single analysis) and the corresponding esti-
mates calculated at the design stage [81].
These figures show an excellent agreement between identified and calculated vertical modes,
the calculated frequencies being slightly higher than the identified ones. It is also possible to
observe an excellent agreement between the first transversal identified and calculated bending
modes, although not so good for modes of higher order. This fact stems probably from less
perfect modelling of the pier-foundation interaction, which plays a more significant role in the
shortest piers.
6.2. Continuous dynamic monitoring

Continuous dynamic monitoring systems have gained an increasing interest and are currently used with
different purposes, such as the following: (i) the safety checking of vibration serviceability limits; (ii)
the assessment of the structural health condition; or (iii) the development of specific research studies,
namely in the field of bridge aeorodynamics.
1 1
0.5
0.5
0
0
-0.5
-0.5 -1
-1 -1.5
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
Figure 12. Vertical bending modes in the range 0.2–0.7 Hz: identified versus calculated modal components.
DOI: 10.1002/stc
A. CUNHA ET AL.
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1.5
1
0.5
0.5
0 0
-0.5
-0.5
-1
-1 -1.5
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
1 1
0.5 0.5
0
0
-0.5
-0.5 -1
-1 -1.5
0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394 0 171 342 513 684 855 1026 1197 1368 1539 1710 1881 2052 2223 2394
Figure 13. Transversal bending modes in the range 0.1–0.6 Hz: identified versus calculated modal components.
a) Continuous dynamic monitoring for safety checking of vibration serviceability limits

The first situation can be well illustrated by the continuous dynamic monitoring of Pedro e Inês
footbridge, in Coimbra, Portugal (Figure 6). This bridge is a slender structure 275 m in length and
4 m wide, except in the central square with dimensions of 8 8 m. The metallic arch spans 110 m and
rises 9 m and has a rectangular box cross section with 1.35 1.80 m. The deck has an L-shaped box
cross section, the top flange being formed by a composite steel-concrete slab 0.11 m thick. In the central
part of the bridge, each L-shaped box cross section and corresponding arch ‘meet’ to form a rectangular
box cross section 8 0.90 m. In the lateral spans, arch and deck generate a rectangular box cross
section 4 0.90 m. The significant slenderness of the bridge and the geometric characteristics lead to
a complex structural behaviour.
According to preliminary studies developed at design stage by the Laboratory of Vibrations and
Structural Monitoring (ViBest) of FEUP, this bridge would be prone to excessive lateral and vertical
human-induced vibrations, and so the structure was built with the necessary precautions in order to
enable the subsequent installation of several TMDs.
Close to the end of the bridge construction, ViBest/FEUP performed a series of dynamic tests [79]: (i) an
ambient vibration test that served to accurately identify natural frequencies and mode shapes; (ii) free vibra-
tion tests for very accurate identification of modal damping ratios, which consisted of sudden release of
masses suspended from different points of the deck chosen so as to stimulate the most relevant modes
of vibration; and (iii) crowd tests, conducted with the purpose of analyzing the lateral behaviour of the
footbridge and assessing the lock-in effect, by response measurement of the bridge to the action of a
continuous stream of pedestrians with a gradually increased number of persons to a maximum of 145.
The ambient vibration test permitted to identify a significant number of natural frequencies in the
range 0–4.5 Hz, several of them easily excited by pedestrians, namely 0.91, 2.05 and 2.88 Hz, with
dominant lateral component, and 1.54, 1.88, 1.95, 2.54, 3.36, 3.57, 3.83, 4.28 and 4.44 Hz, with
dominant vertical component. These results were the basis for the subsequent construction of a new
and more sophisticated shell finite element model, whose stiffening constants of the springs at the
foundations of the arches were iteratively adjusted to achieve a very good matching between calculated
and measured values. The excellent tuning of the finite element modelling was essential for the numerical
evaluation of modal masses and subsequent final design of required TMDs.
DOI: 10.1002/stc
On the other hand, free vibration tests permitted to identify, in particular, a modal damping ratio of
0.55% for the critical fundamental lateral mode (lower than the value of 1% assumed at design stage),
which enhanced the dynamic effects induced by pedestrians.
At last, the crowd tests indicated that the increase of acceleration with the number of pedestrians on the bridge
is not linear, but instead exhibits a ‘jump’ precisely for values of the number of pedestrians close to 70. This value
is in perfect agreement with the formula developed in the context of the studies of the Millennium Bridge [82], in
London, to estimate the critical number of pedestrians NL above, which significant lateral oscillations may occur,
which led to an estimate of NL = 73 pedestrians. Extreme values of acceleration of 1.2 m/s2 (about 12 times
higher the limit value suggested in the recent Services d’Éatudes Techniques des Routes et Autoroutes
(SETRA) [83] and Human- induced Vibrations of Steel Structures (HIVOSS) [84] guidelines) were
measured at the midspan section, corresponding to a dynamic displacement of 4 cm, occurring when
145 pedestrians were walking on the bridge.
The lively behaviour of this footbridge was particularly critical for lateral vibrations associated to the
fundamental lateral mode, which motivated a special concern with the design of the corresponding TMD
[80]. In this case, assuming the validity of the formula developed by Dallard et al., it was possible to fix
the number of pedestrians for which lock-in should not occur, which led to an estimate of damping that
the control solution should introduce. Following that procedure and the studies of Bachmann et al. [85],
the value of 6% was adopted as the value of damping ratio to achieve after installation of a TMD tuned
for this lateral mode of vibration. This implied the specification of a mass of 15 000 kg that was accom-
modated inside the bridge deck by means of six TMD units, each with a mass of 2500 kg.
The efficiency of this lateral vibration control device (Figure 14) was checked, in a first instance,
performing a forced vibration test applying a servo-hydraulic shaker at midspan and measuring the lateral
response of the bridge and of each of the TMD units. This test revealed that the activation of all TMD units
was achieved for accelerations below the comfort limit of about 0.1 m/s2, which is essential to avoid
lock-in phenomena. However, because of different friction characteristics in the corresponding rods,
inducing different damping properties, it was noticed that the equivalent damping introduced by the TMD is
lower than the required at the corresponding design. Therefore, and taking into account the complex char-
acter of the bridge and the scarce experience with mechanical control devices, the owner of the footbridge
required the permanent observation of its structural behaviour during a period of 5 years after construction.
In this context, the dynamic behaviour is being permanently monitored with remote control from the
ViBest/FEUP, by recording vertical and lateral accelerations in six points along the deck (Figure 15).
The monitoring system includes six piezoelectric accelerometers, a signal conditioner, an uninterruptable
power supply and a digital computer (Figure 14), stored inside one of the bridge abutments and transmitting
(a) (b) (c)

Figure 14. (a) Horizontal tuned mass damper installed at the midspan; (b) accelerometers in deck close to the
lateral tuned mass damper; (c) signal conditioner, digital computer and uninterruptable power supply.
3 5 4 6
1 2
30.50 60.00 110.00 64.00 6.00
Figure 15. Locations of the six accelerometers installed inside the deck box section for permanent monitoring.
DOI: 10.1002/stc
A. CUNHA ET AL.
measured records of vibration every 20 min through the Internet to a central computer at FEUP. The
signals are then organized in a main database that can be accessed at any time by post-processing tools
in LabView (National Instruments Corporation Austin, Texas). To satisfy the main objective of this
monitoring system, a website was developed, which allows the visualization of the time signals of
the six accelerometers and subsequently the monitoring of the vibration levels of the structure.
Figure 16(a) shows a plot with the maximum daily lateral accelerations measured during 3 years (from
June 2007 to May 2010), whereas Figure 16(b) shows the corresponding histogram. It could be observed
that the maximum lateral acceleration was 0.099 m/s2, whereas the maximum vertical acceleration was
0.849 m/s2. Considering the comfort limits recommended by SETRA [83] and HIVOSS [84] guidelines,
it can be concluded that all maximum lateral accelerations and nearly all vertical counterparts fall in the
range of maximum comfort levels, so no serviceability problem has occurred in this footbridge under
normal operational conditions during 3 years. This is a consequence of the implementation and tuning
of the passive control devices. Furthermore, it is worth noting that maximum lateral acceleration of about
0.1 m/s2 amplitude was just recorded before on the opening day, when a large number of pedestrians
crossed the bridge, corresponding to an amplitude of displacement of 3 mm.
b) Continuous dynamic monitoring for structural damage detection

Referring to the purpose of assessing the structure health condition, the demonstrator system
installed by ViBest/FEUP at Infante D. Henrique Bridge is described.
The Infante D. Henrique Bridge (Figure 17), over river Douro at Porto, Portugal, is composed of
two mutually interacting fundamental elements: a very rigid prestressed reinforced concrete box
beam, 4.50 m high, supported by a very flexible reinforced concrete arch 1.50 m thick. The arch
spans 280 m between abutments and rises 25 m until the crown, thus exhibiting a shallowness ratio
greater than 11:1. In the 70 m central segment, arch and deck join and define a box beam 6 m high.
(a) (b)
Figure 16. Maximum daily lateral accelerations recorded between June 2007 and May 2010: (a) time variation;
(b) histogram.
Figure 17. Infante D. Henrique Bridge, Porto.
DOI: 10.1002/stc
The arch has constant thickness, and the width increases linearly from 10 m in the central span up to
20 m at the abutments.
The dynamic monitoring system of the Infante D. Henrique Bridge, in operation since September
2007, is composed by two digitizers that receive the signals collected by 12 force balance
accelerometers installed inside the deck box girder according to the scheme of Figure 18. Three
sensors equip each of four sections, one to measure the lateral acceleration and two for the vertical
acceleration at the downstream and upstream sides. The data produced by the digitizers is continuously
transferred to FEUP in the form of American Standard Code for Information Interchange files with as
many columns as the number of sensors, containing acceleration time series with a length of 30 min
sampled at a rate of 50 Hz.
This dynamic monitoring system is complemented by an independent static monitoring system
(performing six acquisitions per hour) that was installed in the bridge during construction, compris-
ing strain gages, clinometers and temperature sensors. In particular, the temperature sensors
embedded in the concrete (Figure 18) were crucial to eliminate the influence of temperature in the
modal parameters.
The dynamic monitoring system produces two files with acceleration time series per hour,
which are then handled by the monitoring software package, DynaMo [70], which performs the
following tasks:
• downloading and archiving in a database of the original data file, for later testing of alternative
processing methodologies;
• preprocessing of the data to eliminate the offset and reduce the sampling frequency to 12.5 Hz
(the first 12 modes are below 5 Hz);
• automatic identification of modal parameters using three different identification algorithms:
FDD, SSI-COV and p-LSCF;
• removal of environmental and operational effects from the identified natural frequencies;
• updating of the control chart for the identification of abnormal values; and
• updating of the database with the results of the processing.
All the collected data and results can then be consulted using a graphical user interface (DynaMo Viewer)
that comprehends tools to create several types of plots for a given time interval.
To exemplify, Figure 19 presents the time evolution, from September 2007 to May 2010, of the
bridge’s first four natural frequencies associated with vertical bending modes. The influence of the
annual temperature fluctuations on the values of the natural frequencies is evident. The additional
observation of the influence of the amplitude of vibration on those frequencies led to the construction
of an optimized regression model that can minimize the effects of these variables on the modal
parameters. Figure 20 shows the narrow range of variation of the natural frequency estimates
Figure 18. Position of the accelerometers and temperature sensors.
DOI: 10.1002/stc
A. CUNHA ET AL.
Figure 19. Time evolution of the natural frequencies of the first four vertical bending modes.
0.79
Frequency (Hz)
0.785
0.78
0.775
0.77
0.79
Frequency (Hz)
0.785
0.78
0.775
0.77
Oct07 Jan08 Apr08 Jul08 Oct08 Jan09 Apr09 Jul09
Figure 20. Time evolution of the first natural frequency before and after minimization of environmental and
operational effects.
associated with the first lateral mode upon removal of the referred effects, using a multiple linear re-
gression model [86].
Despite the good results obtained with the regression model, a complementary study with PCA was
conducted to eliminate the influence of unmeasured factors on the natural frequencies. The first 12
natural frequencies of the bridge corrected by the two statistical tools are then used in the construction
of a multivariable control chart. In the chart presented in Figure 21(a), one point refers to a group of
48 observations, as the performance of averages over 48 observation periods permitted to increase the
500 500
400 400
300 300
T2
T2
200 200
100 100
0 0
Oct08 Jan09 Apr09 Jul09 Oct08 Jan09 Apr09 Jul09
a) b)
Figure 21. Control charts without (a) and with (b) the numerical simulation of damages.
DOI: 10.1002/stc
accuracy in the detection of small frequency changes. It can be seen that the majority of the values are
bellow the control line, defined taking into account observations associated with a reference period
during which the structure is assumed undamaged.
To test the ability of control charts to detect damage, several damage scenarios were simulated with
a numerical model of the bridge previously tuned. These were simulated in simplified form as bending
stiffness reduction of 10% over lengths of 5 m at selected regions along the bridge [70]. The damage
scenario that is analyzed here is a stiffness reduction of the deck at midspan. This led to frequency var-
iations in some modes of the order of 0.15%. These variations, quantified with the tuned numerical
model, were applied to the experimental data collected after March 2009. The control chart in Figure 21
(b), designated Shewhart or T2, demonstrates that after introduction of the frequency shifts due to the
simulated damage, the points of the control chart are outside the control area. Therefore, it is concluded
that it would be possible to automatically detect such a small damage in the bridge with the installed
monitoring system.
c) Continuous dynamic monitoring for wind engineering studies based on prototype measurements
Finally, to exemplify the use of continuous dynamic monitoring in the context of research studies,
reference is made to the monitoring of the aerodynamic behaviour of the Grande Ravine Viaduct, at the
Reunion Island. Conceived and designed by Alain Spielmann and the French design office SETEC TPI
[87], this girder bridge crosses a breach 320 m in width and 170 m in depth, consisting of a steel orthotropic
deck 4 m thick continuous over the 288 m length and supported by inclined cantilevered struts made of
prestressed concrete at angles of 20 to the horizontal (Figure 22).
The location in a zone frequently affected by typhoons and the characteristics of the bridge, for
which limited experience exists in terms of characterization of wind effects, motivated the installation
of a monitoring system by SETEC. Collaboration has been established with ViBest/FEUP and CSTB
with the purpose of processing and interpreting prototype data so as to validate the wind tunnel tests
conducted by CSTB at design stage [88] regarding the characterization of the wind load and of deck
force coefficients, as well as to verify design calculations.
Figure 23 displays a scheme of the instrumentation of the bridge installed by ADVITAM,
which comprehends four sonic anemometers located along the bridge deck (Figure 23(a)) and
one propeller anemometer outside the bridge close to one of the abutments, two rings with a total
Figure 22. Grande Ravine Viaduct, Reunion Island.
a) b)
Figure 23. Monitoring of Grande Ravine Bridge, location of (a) anemometers and accelerometers; (b) pressure cells.
DOI: 10.1002/stc
A. CUNHA ET AL.
of 14 pressure cells in sections close by the midspan separated by 12.5 m (Figure 23 (b)), six
temperature sensors, four of which located outside and two inside the deck, and six acceler-
ometers distributed by the three sections indicated in Figure 23(a), measuring in each section
vertical and lateral accelerations. All sensors are connected to a central acquisition system located
at one abutment. Besides the possibility to control remotely the monitoring system, data records
sampled at 40 Hz are continuously transmitted to a central computer located at FEUP for data processing
and interpretation.
Although the main purpose of the acquisition system is the monitoring of particular events of high
wind, data are permanently collected and stored. This allows the characterization of the wind in order
to build a design wind model from prototype data and compare with the design model. For this
purpose, diagrams and statistics of the mean and gust of winds, of wind incidences, turbulent intensi-
ties, spectra and coherences are formed from records with mean wind speed greater than 5 m/s. To
exemplify, Figure 24 shows wind rose diagrams of the mean wind and longitudinal turbulence intensity
of records with mean wind greater than 5 m/s collected during the month of June 2010.
From the point of view of the structure, the systematic identification of modal parameters allows
correlation with environmental and operational effects for possible removal of their influence. An
important step in this continuous identification was the establishment of a baseline model of the struc-
ture. This was achieved by processing data collected during the commissioning tests conducted by
CSTB, which included both ambient and forced vibration data. Table 1 resumes some of the most
important calculated and identified natural frequencies and damping ratios. Figure 25 shows some
identified modal configurations and the corresponding numerical counterparts. No updating of the
design numerical model developed by SETEC was performed, considering the good correlation
between calculated and identified parameters.
(U) 20100601 to 20100631 (Iu) 20100601 to 20100631
0º 0º
330º 30º 330º 30º
12 1
300º 9.6 60º 300º 0.8 60º
7.2 0.6
4.8 0.4
2.4 0.2
270º 90º 270º 90º
240º 120º 240º 120º
210º 150º 210º 150º

180º 180º
a) b)
Figure 24. Monthly wind rose diagrams of (a) mean wind; (b) longitudinal turbulence intensity.
Table I. Summary of calculated and identified modal parameters at the Grande Ravine viaduct.
Identified damping Identified damping
Calculated Identified ratio with Ambient ratio with Forced Characteristics of
Mode no. frequency (Hz) frequency (Hz) vibration test (AVT) (%) vibration test (FVT) (%) vibration mode
1 0.791 0.773–0.775 0.34–1.27 0.3–0.53 First lateral
2 0.775 0.817–0.821 0.27–0.57 0.37–0.68 First vertical
3 1.056 1.134–1.141 0.87–1.20 0.73–1.53 Second vertical
4 1.756 1.633–1.636 0.20–0.53 0.34–0.56 Second lateral
DOI: 10.1002/stc
f = 0.773 – 0.775Hz (lateral) f = 0.791Hz
-140 -105 -70 -35 0 35 70 105 140
f = 0.817 – 0.821Hz (vertical) f = 0.775Hz
-140 -105 -70 -35 0 35 70 105 140
f = 1.134 – 1.141Hz (vertical) f = 1.056Hz
-140 -105 -70 -35 0 35 70 105 140
f = 1.633 – 1.636Hz (lateral) f = 1.756Hz
-140 -105 -70 -35 0 35 70 105 140
Figure 25. Comparison between identified (left) and calculated (right) vibration modes.
20100601 to 20100631
4.5
3.5
Frequency (Hz)
2.5
1.5
0.5
06/01 06/03 06/05 06/07 06/09 06/11 06/13 06/15 06/17 06/19 06/21 06/23 06/25 06/27 06/29 07/01
Time
Figure 26. Variation of modal frequencies identified during the month of June 2010.
Results from the systematic identification of modal parameters are shown in Figures 26 and 27.
These figures represent the variation of natural frequencies with time during the period of 1 month
and the range of variation of the corresponding damping parameters during the same period. The
purpose of this ongoing work is now to correlate this data with temperature, amplitude of vibration
and wind intensity for possible removal of temperature effects and correlation of the measured
response with wind parameters.
DOI: 10.1002/stc
A. CUNHA ET AL.
20100601 to 20100631
4.5
4
3.5
Damping [%]
3
2.5
2
1.5
1
0.5
0
1 1.5 2 2.5 3 3.5 4
Frequency [Hz]
Figure 27. Range of variation of modal damping ratios identified during June 2010.
7. CONCLUSIONS
On the basis of their own experience, the authors described in this paper some important trends of
evolution of dynamic testing and continuous monitoring in bridge applications. In particular, the
following are shown:
• The experimental approaches followed for dynamic testing have evolved in general from forced
vibration tests, inspired in classical procedures from mechanical engineering, to ambient and free
vibration tests, which became the most practical and efficient techniques more and more used for
the accurate identification of modal properties of bridges at construction, commissioning or
rehabilitation stages. Still, in some cases where the efficiency of implemented vibration control
devices must be evaluated, forced vibration tests still play an important role;
• This evolution was enabled by significant technological progress related with accelerometers
(more suited for measurement in low frequency ranges, more sensible and with lower levels of
internal noise) and data acquisition systems (with higher and higher digitizers’ resolution and
allowing decentralized and wireless solutions).
• The development of different kinds of noncontact measurement techniques, based on laser, radar or
video technologies, open new perspectives for easy and efficient applications of dynamic testing in
different cases;
• The progress in the field of data acquisition and transmission through the Internet made it also
feasible a continuous dynamic monitoring of the structural behaviour, which may complement
other components of structural monitoring, such as the monitoring of loads, static behaviour
and durability;
• In parallel with this technological progress, a remarkable development of powerful methods of
modal identification under operational conditions occurred, using both nonparametric and
parametric methods, in frequency or time domain, and their automated implementation in
continuous dynamic monitoring was recently successfully achieved, often using hierarchical
clustering algorithms for the automatic and objective interpretation of stabilization diagrams
obtained by application of the SSI or Least- Square Complex Frequency Domain (L-SCFD) methods;
• The results obtained with the ambient and free vibration tests of the Millau Viaduct clearly
evidence not only the potential of the use of a set of independent seismographs duly syn-
chronized by GPS but also the high accuracy of the OMA methods for modal identification
of large bridges;
• A key aspect in vibration-based damage detection using continuous dynamic monitoring data is the
removal of the influence of environmental and operational factors, such as temperature of traffic
intensity, on the modal frequency estimates, which can be presently achieved with good efficiency
using both input–ouput and output-only statistical methods. This aspect was conveniently illus-
trated with the Infante D. Henrique Bridge case study, where such correction led to very stable esti-
mates of natural frequencies, and the detection of very low levels of simulated damage was
achieved on the basis of the construction of appropriate control charts.
DOI: 10.1002/stc
• Continuous dynamic monitoring can also be used for safety checking of vibration serviceability
limits, as it was clearly demonstrated with the Pedro e Inês footbridge case study, where the
continuous monitoring during the last 5 years has shown the good efficiency of the vibration
control devices (TMDs) implemented to mitigate excessive lateral and vertical human-induced
vibrations;
• At last, continuous dynamic monitoring has also been recently used in the development of wind
engineering investigations based on prototype measurements, with the purpose of validating
wind tunnel tests conducted at design stage regarding the characterization of the wind load and
of deck force coefficients, as well to verify design calculations, as it was illustrated with the
Grande Ravine case study.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the Portuguese Foundation for Science and Technology (FCT) for all the support
received for development of research in the field of dynamic testing and continuous monitoring of bridges. They
acknowledge also the fruitful collaboration with Xavier Cespedes, from SETEC, and Olivier Flamand, from CSTB
in the context of the studies developed by the PhD student Fernando Bastos on the Grande Ravine Viaduct.
REFERENCES
1. Cantieni R, Deger Y, Pietrzko S. Large structure investigation with dynamic methods: the bridge on the river Aare at Aarburg.
Prestressed Concrete in Switzerland, Report of the Swiss FIP Group to the 12th FIP Congress, Washington D.C., 1994.
2. Pietrzko S, Cantieni R, Deger Y. Modal testing of a steel/concrete composite bridge with a servo-hydraulic shaker. Proc. 14th
International Modal Analysis Conference (IMAC), Dearbon, Michigan, 1996; 91-98.
3. Keightley W, Housner G, Hudson D. Vibration tests of the Encino Dam in take tower. Report. California Institute of
Technology, 1961.
4. Cantieni R. Assessing a dam’s structural properties using forced vibration testing. Proc. IABSE International Conference on
Safety, Risk and Reliability - Trends in Engineering, Malta, 2001.
5. Cunha A, Caetano E, Magalhães F. Output-only dynamic testing of bridges and special structures. Structural Concrete.
Journal of FIB 2007; 8(2):67–85.
6. Carder D. Observed vibrations of buildings. Bull. Seismological Society of America 1936; 26(4):245–277.
7. Vincent G. Golden Gate Bridge vibration study. ASCE Journal of the Structural Division 1958; 4:ST6.
8. McLamore V, Hart G, Stubbs I. Ambient vibration of two suspension bridges. ASCE Journal of the Structural Division
1971; 97(ST10):2567–2582.
9. Trifunac M. Comparisons between ambient and forced vibration experiments. Earthquake Engineering and Structural
Dynamics 1972; 1:133–150.
10. Brownjohn JM, Dumanoglu AA, Severn RT, Blakeborough A. Ambient vibration survey of the Bosporous suspension
bridge. Earthquake Engineering and Structural Dynamics 1989; 18:263–283.
11. Brownjohn JMW, Dumanoglu AA, Severn RT. Ambient vibration survey of the Fatih Sultan Mehmet (Second Bosporus)
Suspension Bridge. Earthquake Engineering and Structural Dynamics 1992; 21:907–924.
12. James GH, Carne TG, Lauffer JP, Nord AR. Modal testing using natural excitation. Proceedings of the 10th International
Modal Analysis Conference, San Diego, California, USA, 1992; 1209-1216.
13. Cunha A, Caetano E, Delgado R. Dynamic tests on a large cable-stayed bridge. An efficient approach. Journal of Bridge
Engineering, ASCE 2001; 6(1):54–62.
14. Flamand O, Grillaud G. Dynamic testing of the Millau Viaduct. In Proc. Third Int. Conf. on Bridge Maintenance, Safety and
Management, Porto, Portugal, 2006.
15. Brownjohn JMW, Magalhães F, Caetano E, Cunha A. Ambient vibration re-testing and operational modal analysis of the
Humber Bridge. Engineering Structures 2010; 32(8):2003–2018.
16. Bietry J, Jan P. Essais Dynamiques du Pont de Normandie. Report of “Mission du Pont de Normandie”. EN-D 95.5 C, 1995.
17. Catbas FN, Grimmelsman KA, Aktan AE. Structural identification of the Commodore Barry Bridge. Proc. SPIE, 3995,
2000:84–97.
18. Brownjohn JMW, Moyo P, Omenzetter P, Lu Y. Assessment of highway bridge upgrading by dynamic testing and finite
element model updating. ASCE Journal of Bridge Engineering 2003; 8:162–172.
19. Farrar C, Doebling S, Cornwell P, Straser E. Variability of modal parameters on the Alamosa Canyon Bridge. Proc. of XV
IMAC Conference, 1997.
20. Alampalli S. Influence of in-service environment on modal parameters. Proc. of 16th Int. Modal Analysis Conference, Santa
Barbara, USA, 1998.
21. Kramer C, de Smet CAM, De Roeck G. Z24 Bridge damage detection tests. Proceedings of IMAC XVII, Kissimmee,
Florida, 1999; 1023–1029.
22. Wong KY. Instrumentation and health monitoring of cable-supported bridges. Structural Control and Health Monitoring
2004; 11:91–124.
23. Ko JM, Ni YQ. Technology developments in structural health monitoring of large-scale bridges. Engineering Structures
2005; 27(12):1715–1725.
24. FIB. FIB bulletin 22: monitoring and safety evaluation of existing concrete structures. 2003.
25. Magalhães F, Cunha Á, Caetano E. Dynamic monitoring of a long span arch bridge. Engineering Structures 2008;
30(11):3034–3044.
DOI: 10.1002/stc
A. CUNHA ET AL.
26. Felber A. Development of a hybrid bridge evaluation system, Ph. D. Thesis, University of British Columbia, Canada, 1993.
27. Felber AJ, Cantieni R. Introduction of a new ambient vibration testing system. Description of the system and seven bridge
tests. Report No. 156’521, EMPA, Dübendorf, 1996.
28. Feltrin G, Meyer J, Bischoff R, Motavalli M. Long-term monitoring of cable stays with a wireless sensor network. Structure
and Infrastructure Engineering 2009; 6(5):535–548.
29. Oliveira S, Mendes P, Garrett A, Costa O, Reis J. Long-term dynamic monitoring systems for the safety control of large
concrete dams. The case of Cabril dam, Portugal. Proceedings of the 6th International Conference on Dam Engineering,
Lisbon, Portugal, 2011.
30. Cunha A, Caetano E. Dynamic measurements on stay cables of cable-stayed bridges using an interferometry laser system.
Experimental Techniques, SEM 1999; 23(3):38–43.
31. Gentile C, Bernardini G. An interferometric radar for non-contact measurement of deflections on civil engineering structures:
laboratory and full-scale tests. Structure and Infrastructure Engineering 2009; 6(5):521–534.
32. Gentile C. Application of radar technology to deflection measurement and dynamic testing of bridges. InTech, January,
2010.
33. Macdonald J, Dagless E, Thomas B, Taylor C. Dynamic measurements of the Second Severn Crossing. Proc. Instn. Civil
Engineers, vol. 123, 1997; 241–248. Paper no. 11483.
34. Caetano E, Silva S, Bateira J. A vision system for vibration monitoring of civil engineering structures. Experimental
Techniques, SEM 2011; 35:74–82.
35. Maia N et al. Theoretical and Experimental Modal Analysis. Research Studies Press: UK, 1997.
36. Han M-C, Wicks AL. On the application of Forsythe orthogonal polynomials for global modal parameter estimation. Proc.
7th Int. Modal Analysis Conference, 1989.
37. Crawford R, Ward H. Determination of the natural period of buildings. Bulletin of the Seismological Society of America
1964; 54(6):1743–1756.
38. Abdel-Ghaffar AM. Vibration studies and tests of a suspension bridge. Earthquake Engineering and Structural Dynamics
1978; 6:473–496.
39. Prevosto M. Algorithmes d’Identification des Caractéristiques Vibratoires de Structures Mécaniques Complexes. Ph.D.
Thesis, Univ. de Rennes I, France, 1982.
40. Corrêa MR, Campos Costa A. Ensaios Dinâmicos da Ponte sobre o Rio Arade. Pontes Atirantadas do Guadiana e do Arade
(in Portuguese), ed. by LNEC, 1992.
41. Brincker R, Zhang L, Andersen P. Modal identification from ambient responses using frequency domain decomposition.
Proc. 18th Int. Modal Analysis Conference, Kissimmee, FL, 2001.
42. Brincker R, Ventura C, Andersen P. Damping estimation by frequency domain decomposition. Proc. 19th Int. Modal
Analysis Conference, San Antonio, Texas, USA, 2000.
43. Brincker R, Krenk S, Kirkegaard PH, Rytter A. Identification of the dynamical properties from correlation function
estimates, Bygningsstatiske Meddelelser. Danish Society for Structural Science and Engineering 1992; 63(1):1–38.
44. Asmussen JC. Modal analysis based on the random decrement technique – application to civil engineering structures. Ph.D.
Thesis, Univ. Aalborg, Denmark, 1997.
45. Ibrahim SR, Mikulcik EC. A method for the direct identification of vibration parameters from the free response. The Shock
and Vibration Bulletin 1977; 47(4):183–198.
46. Fukuzono K. Investigation of multiple-reference Ibrahim time domain modal parameter estimation technique, M.Sc. Thesis,
Univ. Cincinnati, OH, USA, 1986.
47. Brown DL, Allemang RJ, Zimmerman R, Mergeay M. Parameter estimation techniques for modal analysis. SAE Technical
Paper Series, No. 790221, 1979.
48. Vold H, Kundrat J, Rocklin GT, Russell R. A multi-input modal estimation algorithm for mini-computers. SAE Technical
Paper Series, No. 820194, 1982.
49. Peeters B. System identification and damage detection in civil engineering. Ph.D. Thesis, K. U. Leuven, Belgium, 2000.
50. Peeters B, De Roeck G. Reference-based stochastic subspace identification for output-only modal analysis. Mechanical
Systems and Signal Processing 1999; 13(6):855–878.
51. Van Overschee P, DeMoor B. Subspace Identification for Linear Systems – Theory, Implementation, Applications. Kluwer
Academic Publishers: The Netherlands, 1996.
52. Rodrigues J, Brincker R, Andersen P. Improvement of frequency domain output-only modal identification from the application
of the random decrement technique. Proc. 23rd Int. Modal Analysis Conference, Deaborn, MI, USA, 2004.
53. Peeters B, Vanhollebeke F, Van der Auweraer H. Operational PolyMAX for estimating the dynamic properties of a stadium
structure during a football game. Proc. 23rd Int. Modal Analysis Conference, Orlando, FL, USA, 2005.
54. Reynders E, De Roeck G. Reference-based combined deterministic-stochastic subspace identification for experimental and
operational modal analysis. Mechanical Systems and Signal Processing 2008; 22(3):617–637.
55. Ni YQ, Zhou HF, Chan KC, Ko JM. Modal flexibility analysis of cable-stayed Ting Kau Bridge for damage identification.
Computer-Aided Civil and Infrastructure Engineering 2008; 23(3):223–236.
56. Ko JM, Ni YQ, Zhou HF, Wang JY, Zhou XT. Investigation concerning structural health monitoring of an instrumented
cable-stayed bridge. Structure and Infrastructure Engineering 2009; 5(6):497–513.
57. Ni YQ, Fan KQ, Zheng G, Ko JM. Automatic modal identification and variability in measured modal vectors of a cable-stayed
bridge. Structural Engineering and Mechanics 2005a; 19(2):123–139.
58. Brincker R, Andersen P, Jacobsen N-J. Automated frequency domain decomposition for operational modal analysis. Proc.
IMAC-XXV, Orlando, USA, 2007.
59. Magalhães F, Cunha Á, Caetano E. Online automatic identification of the modal parameters of a long span arch bridge.
Mechanical Systems and Signal Processing 2009; 23(2):316–329.
60. Rohrmann RG, Baessler M, Said S, Schmid W, Rücker WF. Structural causes of temperature affected modal data of civil
structures obtained by long time monitoring. Proc. of IMAC-XVIII, International Modal Analysis Conference, San Antonio,
Texas, USA, 2000.
61. Hair J, Anderson R, Tatham R, Black W. Multivariate Data Analysis. Prentice Hall: New Jersey, 1998.
62. Ljung L. System Identification: Theory for the User. Prentice-Hall: New Jersey, 1999.
63. Vapnik VN. An overview of statistical learning theory. IEEE Transaction on Neural Networks 1999; 10(5):988–999.
DOI: 10.1002/stc
64. Ni YQ, Hua XG, Fan KQ, Ko JM. Correlating modal properties with temperature using long-term monitoring data and
support vector machine technique. Engineering Structures 2005b; 27:1762–1773.
65. Hua XG, Ni YQ, Ko JM, Wong KY. Modeling of temperature-frequency correlation using combined principal component
analysis and support vector regression technique. Journal of Computing in Civil Engineering, ASCE 2007; 21(2):122–135.
66. Zhou HF, Ni YQ, Ko JM, Wong KY. Modeling of wind and temperature effects on modal frequencies and analysis of
relative strength of effect. Wind and Structures 2008; 11(1):35–50.
67. Ni YQ, Zhou HF, Ko JM. Generalization capability of neural network models for temperature-frequency correlation using
monitoring data. Journal of Structural Engineering, ASCE 2009; 135(10):1290–1300.
68. Zhou HF, Ni YQ, Ko JM. Eliminating temperature effect in vibration-based structural damage detection. Journal of
Engineering Mechanics, ASCE 2011a; 137(12):785–796.
69. Deraemaeker A, Reynders E, De Roeck G, Kullaa J. Vibration-based structural health monitoring using output-only
measurements under changing environment. Mechanical Systems and Signal Processing 2008; 22(1):34–56.
70. Magalhães F. Operational modal analysis for testing and monitoring of bridges and special structures. PhD Thesis, Faculty of
Engineering of the University of Porto, 2010.
71. Kulla J. Structural health monitoring of a crane in variable configurations. Proc. of ISMA, International Conference on Noise
and Vibration Engineering, Leuven, Belgium, 2004.
72. Yan A-M, Kerschen G, De Boe P, Golinval J-C. Structural damage diagnosis under varying environmental conditions – part
I: a linear analysis. Mechanical Systems and Signal Processing 2005a; 19(4):847–864.
73. Yan A-M, Kerschen G, De Boe P, Golinval J-C. Structural damage diagnosis under varying environmental conditions – part
II: local PCA for non-linear cases. Mechanical Systems and Signal Processing 2005b; 19(4):865–880.
74. Sohn H, Worden K, Farrar CR. Statistical damage classification under changing environmental and operational conditions.
Journal of Intelligent Material Systems and Structures 2003; 13(9):561–574.
75. Zhou HF, Ni YQ, Ko JM. Structural damage alarming using auto-associative neural network technique: exploration of
environment-tolerant capacity and setup of alarming threshold. Mechanical Systems and Signal Processing 2011b;
25(5):1508–1526.
76. Johnson RA, Wichern DW. Applied Multivariate Statistical Analysis. Prentice Hall: New Jersey, 1992.
77. Montgomery D. Introduction to Statistical Quality Control. John Wiley & Sons: New York, 2005.
78. Kullaa J. Damage detection of the Z24 bridge using control charts. Mechanical Systems and Signal Processing 2003;
17(1):163–170.
79. Caetano E, Cunha A, Magalhães F, Moutinho C. Studies for controlling human-induced vibration of the Pedro e Inês
footbridge, Portugal. Part 1: assessment of dynamic behaviour. Engineering Structures 2010a; 32:1069–1081.
80. Caetano E, Cunha A, Moutinho C, Magalhães F. Studies for controlling human-induced vibration of the Pedro e Inês
footbridge, Portugal. Part 2: implementation of tuned mass dampers. Engineering Structures 2010b; 32:1082–1091.
81. Flamand O, Grillaud G. Identification modale du Viaduc de Millau. Research Report EN-CAPE 05.007 C – V0. CSTB, 2005.
82. Dallard P et al. The London Millenium Footbridge. The Structural Engineer 2001; 79(22):17–33.
83. SETRA/AFGC/AFGC. Passerelles piétonnes – Evaluation du comportement vibratoire sous l’action des piétons
(Footbridges – Assessment of dynamic behaviour under the action of pedestrians), Guidelines, Sétra, 2006.
84. HIVOSS. Design of footbridges: guideline. 2008. http://www.stb.rwth-aachen.de/projekte/2007/HIVOSS/download.php
85. Bachmann H et al. Vibration Problems in Structures: Practical Guidelines. Birkhäuser Verlag: Basel, 1995; 234.
86. Magalhães F, Cunha A, Caetano E. Vibration based structural health monitoring of an arch bridge: from automated OMA to
damage detection. Mechanical Systems and Signal ProcessingSpecial issue on interdisciplinary and integration aspects in
structural health monitoring 2012; 28:212–228.
87. Croiset JE, Ryckaert J, Spielmann A, Viel G. Viaduc de la Grande Ravine, Travaux n 823, 01/10/05, 110–116, 2005.
88. SETEC. Viaduc de la Grande Ravine: Études de vent réalisées par le CSTB, Dossier de Consultation des Entreprises,
pièce B4, 2004.
DOI: 10.1002/stc

Dynamic Testing and Monitoring of Bridges - Paper

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

Dynamic Testing and Monitoring of Bridges - Paper

Uploaded by

Copyright:

Available Formats

STRUCTURAL CONTROL AND HEALTH MONITORING

Struct. Control Health Monit. 2012

Recent perspectives in dynamic testing and monitoring of bridges

A. Cunha, E. Caetano*,†, F. Magalhães and C. Moutinho

Received 3 November 2011; Revised 27 June 2012; Accepted 28 June 2012

Copyright © 2012 John Wiley & Sons, Ltd.

2.2. From dynamic testing to continuous dynamic monitoring

(a) (b) (c)

3.3. From conventional sensors to noncontact measurement techniques

5. VIBRATION-BASED DAMAGE DETECTION

Acceleration time series undamaged damaged

5.1. From operational modal analysis to automated operational modal analysis

5.3. Damage detection

6. DYNAMIC TESTING AND CONTINUOUS MONITORING: CASE STUDIES

Figure 6. Pedro e Inês footbridge, Portugal.

Figure 7. Millau Viaduct, France.

Figure 8. Instrumented sections of the deck in the ambient vibration test.

Vertical direction Lateral direction

Modal Damp. Ratios (%)

6.2. Continuous dynamic monitoring

a) Continuous dynamic monitoring for safety checking of vibration serviceability limits

(a) (b) (c)

30.50 60.00 110.00 64.00 6.00

b) Continuous dynamic monitoring for structural damage detection

Figure 17. Infante D. Henrique Bridge, Porto.

Figure 18. Position of the accelerometers and temperature sensors.

Figure 22. Grande Ravine Viaduct, Reunion Island.

(U) 20100601 to 20100631 (Iu) 20100601 to 20100631

240º 120º 240º 120º

210º 150º 210º 150º

f = 0.773 – 0.775Hz (lateral) f = 0.791Hz

-140 -105 -70 -35 0 35 70 105 140

f = 0.817 – 0.821Hz (vertical) f = 0.775Hz

-140 -105 -70 -35 0 35 70 105 140

f = 1.134 – 1.141Hz (vertical) f = 1.056Hz

-140 -105 -70 -35 0 35 70 105 140

f = 1.633 – 1.636Hz (lateral) f = 1.756Hz

-140 -105 -70 -35 0 35 70 105 140

You might also like