Experimental Vibration Analysis for Civil Engineering Structures

307

Comparison of damping estimates from ambient and free

vibration tests in large structures
F. Magalhes, A. Cunha & E. Caetano
Faculty of Engineering of University of Porto (FEUP), Portugal

ABSTRACT: This paper presents a comparative study of modal damping identification from
ambient and free vibration tests developed in large structures, in order to evaluate the level of
accuracy that modern output-only modal identification techniques can provide. This study is
based on high quality databases of four outstanding Civil Structures: Vasco da Gama Bridge
(cable-stayed bridge in Portugal), Millau Viaduct (multi-span cable-stayed bridge in France),
Braga Sports Stadium suspended roof (Portuguese stadium) and a slender lively footbridge in
Coimbra (Portugal).

1 INTRODUCTION
Modal damping ratios are very difficult to predict at the design stage, but have a strong influence on the dynamic structural behaviour, the amplitude of vibrations at resonance being inversely proportional to these coefficients. Therefore, in structures subjected to relevant dynamic
excitations it is important to perform dynamic tests after construction to check the assumptions
adopted during the design and eventually implement solutions to reduce the vibrations.
The experimental identification of damping ratios can be based on forced vibration, free vibration or ambient vibration tests.
Forced vibration tests are accurate but require very heavy and expensive equipments, like
servo-hydraulic shakers or eccentric mass vibrators. Besides that, the results are always influenced by the uncontrolled ambient excitation and the estimated values are associated with levels
of vibration that are higher than the ones observed during the normal use of the structure.
Ambient vibration tests have the strong advantage of being very practical and economical,
as they use the freely available ambient excitation. Furthermore, the data is colleted during the
normal use of the structure and consequently the identified modal parameters are associated
with realistic vibration levels. However, although modern output-only modal identification
techniques, in conjunction with the use of appropriate vibration measurement equipment, can
provide nowadays very accurate estimates of natural frequencies and mode shapes, it is usually
observed that the corresponding damping estimates present a significant scatter. That is why it is
still very common to perform complementary free vibration tests, when the accurate identification of modal damping ratios is required.
Free vibration tests, based on the sudden release of a suspended mass or the sudden cut of a
tensioned cable, are less practical than ambient vibration tests, but considerably simpler than
forced vibration tests. In this type of tests the modal parameters are also derived from vibrations
levels that are higher than the service ones.
During the last years, the Laboratory of Vibrations and Monitoring (ViBest,
www.fe.up.pt/vibest ) of FEUP has been involved in the performance and /or signal processing
of a large number of dynamic tests in large Civil Engineering structures (e.g. Vasco da Gama

308

EVACES07

Bridge, Millau Viaduct, Braga Sports Stadium suspended roof, Coimbra footbridge), collecting
very complete and high quality databases that enable comparative studies of the application of
alternative methods of modal identification.
In this context, this paper presents a comparative study of modal damping identification from
ambient and free vibration tests developed on those structures, in order to evaluate the level of
accuracy that modern output-only modal identification techniques can provide.
2 DYNAMIC TESTS FOR DAMPING ESTIMATION
In this section, some practical aspects and the theoretical background of ambient vibration and
free vibration tests are briefly described.
2.1 Ambient Vibration Tests
During the ambient vibration tests, the accelerations of structures excited by ambient loads are
measured. Accordingly, the traffic over the bridges and the wind are welcome, to increase the
signal intensity of the measured time series. Still, the levels of excitation are generally low, so
the used accelerometers have to be very sensitive. Furthermore, the test of slender structures
demands the use of transducers that should present a linear response from very low frequencies.
Besides that, when the size of the structure is considerable, the use of wireless systems duly synchronized by GPS is advantageous. All the applications presented in the following sections were
based on the use of seismographs that contain internal tri-axial force-balance accelerometers
(linear response from 0 to 50Hz), a battery, an A/D converter with 18-bit that ensure a good
resolution, a memory card to store the time series and a GPS antenna to continuously update the
internal clock using the satellites, in order to have a perfect synchronization between acquisition
units, without using cables between them.
The recorded data is then processed using output-only identification tools. Nowadays, there
are several robust methods, working in time or frequency domains, which are already implemented in user-friendly software (LMSInternational 2005; SVS 1999-2004). A review of the
most commonly used methods in civil applications was presented by Cunha & Caetano (2005).
In the present work, the ambient data is analysed with Stochastic Subspace Identification
Methods based on correlations or directly on the time series, SSI-COV and SSI-DATA respectively. Throughout the description of the applications, reference is made to papers where the results of several other output-only identification algorithms are described.
The Stochastic Subspace Identification methods perform the identification of the modal parameters using a stochastic state space model that, in its discrete form and assuming the excitation as a white noise, is represented by the following equations:
xk +1 = A xk + wk
y k = C xk + vk

(1)

Identification of matrices A and C is performed from the correlation functions of the measured responses time series, in the case of the Covariance driven Stochastic Subspace (SSI-COV)
method, or directly from the time series, in the case of the Data driven Stochastic Subspace
(SSI-DATA). The algorithms of both methods are based on the properties of stochastic systems
(Overschee & Moor 1996).
After identification of the state space model, modal parameters are extracted from matrices A
and C (Peeters 2000). It is worth noting that the identification of the state space model requires
the definition of the order of the model. However, for real Civil Engineering Structures it is not
possible to predict the order of the model that better fits the experimental data and more realistically characterizes the dynamic behaviour of the structure. The most appropriate way to overcome this difficulty is to estimate the modal parameters using models with an order within an
interval previously defined in a conservative way. The identified modal parameters are then represented in a stabilization diagram. This diagram shows parameters that are stable for models of
increasing orders, and these are the ones with structural significance. The others are just associated with numerical modes, which are important to model the noise that exists always in measured data. In Figure 1, two stabilization diagrams of two applications that will be presented in

309

Experimental Vibration Analysis for Civil Engineering Structures

the following sections are shown: the Coimbra footbridge and the Braga Stadium roof. The easily identified vertical alignments of stable poles (+ in Fig.1a and in Fig. 1b) allow the identification of the modes.
Model order

Stabilization Diagram - setup4

40
35
30
25
20
15

all poles
stable freq.
stable damp.
MAC

10
5
0
0.2

a)

0.3

0.4

0.5
0.6
0.7
Frequency (Hz)

0.8

0.9

1.0

b)

Figure 1. a) Stabilization diagram of Coimbra footbridge (SSI-DATA method); b) Stabilization diagram

of Braga Stadium suspended roof (SSI-COV method)

In this work the SSI-COV method was applied using MatLab routines developed at the University of Porto (Magalhes 2004), whereas the Artemis software (SVS 1999-2004) was used
for application of the SSI-DATA method.
2.2 Free Vibration Tests
Free vibration tests for identification of modal damping ratios in Civil Engineering structures
can be performed in two different ways: measurement of the free response after application of a
sinusoidal load with an excitation frequency coincident with a natural frequency and measurement of the free structural response after the application of an impulse (or imposed displacement). The first type of tests is less practical, since the application of a sinusoidal load requires
the use of special equipment, it takes a long time in order to allow the individual excitation of
each relevant mode of vibration, and previous knowledge of the natural frequencies is needed.
On the other hand, the impulsive tests are very convenient, because they can rely on the sudden
release of a mass previously suspended from the structure and it is possible to identify the modal
damping ratios of several modes with just one test. Of course that the prior identification of the
mode shapes is useful, as the mass must be hanged from a point where the most important
modes of vibration have significant modal components.
The most traditional procedure to analyse the recorded data consists of applying a band-pass
filter to isolate the contributions of the most important modes. As this filtered data should only
contain the contribution of a single mode, the modal damping ratio is directly estimated by fitting an exponential decay to the relative maxima of the time series. This approach faces two major limitations: the difficulty to isolate the contribution of modes with very close natural frequencies and the not always realistic assumption that damping is proportional.
As alternative to this rather simple procedure, the output-only SSI-COV method can be used
to extract the modal parameters from measured free decays. The free decays observed after the
application of impulses or imposed displacements are proportional to the correlations of the responses associated with a white noise excitation. Consequently, these measured free decays can
be used as input of the SSI-COV method, taking the place of the correlation functions calculated
from the ambient responses. This approach overcomes the limitations of the traditional procedure, due to the fact that, in this method, a model with non-proportional damping is fitted to the
data, allowing the identification of closely spaced modes. Furthermore, with this technique, after
the identification of the modal properties, it is possible to decompose the measured free decays
in modal decays. For instance, Figure 2 represents one of the free decays measured at Millau
Viaduct (the application presented in next section) and its decomposition in modal decays. This
step is important to check the results and to evaluate the importance of each mode in the measured response. The damping estimates of the less excited modes are expected to be less reliable.

310

EVACES07

50
0
-50

Measured decay

20
0
-20

f = 0.261 Hz

20
0
-20

f = 0.293 Hz

20
0
-20

f = 0.389 Hz

20
0
-20

f = 0.435 Hz

20
0
-20

f = 0.551 Hz

20
0
-20

f = 0.605 Hz

20
0
-20

f = 0.709 Hz

20
0
-20

f = 0.751 Hz

20
0
-20

f = 0.821 Hz

20
0
-20

f = 0.839 Hz
0

50

100

150

200

250

300

tempo (s)

Figure 2. Measured free decay of Millau Viaduct decomposed in modal decays

3 MILLAU VIADUCT
The Millau Viaduct is a large cable-stayed road-bridge that spans the valley of the River Tarn
near Millau in southern France (Figure 3). Designed by the architect Norman Foster and the
bridge engineer Michel Virlogeux, it is the tallest vehicular bridge in the world, with one pier's
summit at 343m. The 8 cable-stayed spans with a total length of 2460m make this bridge also
the longest cable-stayed bridge in the world. The 32m wide and 42.4m deep deck has a steel box
section continuous over the 2460m of extension. The six central spans have a length of 342m,
while the outer spans are 204m long. The concrete piers range in height from 77 to 245m, with
supporting 90m tall steel towers.

Figure 3. Millau Viaduct (general view and suspended mass used in the free vibration test)

311

Experimental Vibration Analysis for Civil Engineering Structures

The outstanding nature of the bridge justified the development of dynamic tests based on the
application of impulsive loads and on the measurement of the subsequent free decay, which
were coordinated by the CSTB (Centre Scientifique et Technique du Btiment). As a complement, an independent ambient vibration test was performed using equipment provided by the
Faculty of Engineering of University of Porto (ViBest / FEUP).
The former tests consisted of the sudden rupture of a tensioned cable connected to the deck.
This operation was conducted first with a cable tensioned at 600kN and then repeated with a
tensioning of the cable to 1000kN. In both cases the wind speed was low, oscillating between 2
and 5m/s. The structural response was measured during 960s by three independent sets of accelerometers: 12 accelerometers of the continuous monitoring system installed by SITES, 9 accelerometers of the acquisition system controlled by CSTB and 2 seismographs from FEUP.
The ambient vibration test was performed with 4 seismographs dully synchronized by GPS
that scanned 28 sections of bridge with 13 setups of 16 minutes. The processing of the data collected during this test provided estimates of the viaducts natural frequencies and modal damping ratios and a very good characterization of the vibration mode shapes. In the frequency range
of analysis, 0 to 0.75Hz, 9 lateral and 11 vertical bending modes of the deck were identified.
A detailed description of both dynamic tests, of the applied identification methods and of the
results obtained is presented by Caetano et al. (2007). This paper is only focused on the identification of the modal damping ratios.
The time series collected during the ambient vibration test were processed by the SSI-COV
method using two different approaches. The more classical one, based on multiple analyses (one
stabilization diagram for each setup) and a more recent algorithm (Parloo 2003) based on a single analysis (one stabilization diagram for all data sets). The first one provides several estimates
for each modal parameter (one for each setup), whereas the second one gives a single estimate.
Figure 4 summarizes the results achieved with both algorithms. It shows the scatter of the estimates provided by the SSI-COV method (average standard deviation of 20%) when the data of
each setup is processed separately and illustrates the proximity of the generality of the estimates
provided by the single analysis with regard to the average values provided by the multiple
analyses.
Modal Damp. Ratios (%) .

13 Setups
Average of the 13 setups
Single analysis

1.5
1
0.5
0
0.15

0.2

0.25

0.3

0.35

0.4
0.45
0.5
Natural Frequencies (Hz)

0.55

0.6

0.65

0.7

Figure 4. Modal damping ratios provided by the SSI-COV method applied to the ambient vibration data.

The data recorded by the three data acquisition systems during the free vibration tests was
processed by the SSI-COV method, adapted to deal with free decays, and by the application of
band-pass filters. A good agreement was found between results obtained from the different data
sets and identification methods. In this paper, emphasis will be placed in the results provided by
the SSI-COV method applied to the data colleted by the monitoring system installed by SITES.
This identification algorithm confirmed the modal damping coefficients identified by application of band-pass filters and made possible the identification of the damping ratios of a higher
number of modes.
Figure 5 compares the estimates given by the SSI-COV method applied independently to the
data of the 13 setups of the ambient vibration test (represented by the average value and by an
interval of variation equal to twice the observed standard deviation) with the estimates provided
by the SSI-COV method applied to the free decays produced by the two different impulsive
loads (Imp1: 600kN, Imp2: 1000kN cable tension), which are naturally associated with higher

312

EVACES07

Modal Damp. R. (%)

amplitudes of vibration. Of course that the application of the impulses didnt allow the identification of all the modes represented in Figure 4, because these loads only excited the vertical
modes with non-zero modal ordinates at the point where the impulse was applied. The comparison shows very small differences in the modes between 0.2 and 0.3 Hz. Significant differences
are observed in the third mode represented on the graphic. These can be explained by the dependence of the estimates of the modal damping ratio of this mode with the amplitude of oscillations, as it is observed a coherent increase of the damping ratio with the amplitudes of vibration of the used time segments. For the other modes the results present a reasonable consistency,
except for the last mode using imp2. The stabilization diagram used to obtain this last estimate
did not present a clear alignment of stable poles for this mode, and so a reliable result was not
expected.
AVT, SSI-COV

1.20

FVT, Imp1
FVT, Imp2

0.80
0.40
0.00
0.20

0.30

0.40

0.50
Natural Frequencies (Hz)

0.60

0.70

Figure 5. Modal damping ratios obtained from the Ambient Vibration Test (AVT) and from the Free Vibration test (FVT). Imp1: 600kN; Imp2: 1000kN cable tension.

4 COIMBRA FOOTBRIDGE
The Pedro e Ins footbridge is a new landmark that links both banks of Mondego River at the
city of Coimbra, in Portugal. The bridge has a total length of 275m and is formed by a central
parabolic arch that spans 110m and rises 9m and two half parabolic arches in steel supporting
with total continuity a composite steel-concrete deck (Figure 6).

Figure 6. Coimbra Footbridge (general view and hanging of the mass for the application of the impulse)

The preliminary dynamic studies of this bridge indicated that it would be prone to vibrations
induced by pedestrians, requiring control devices. For the correct final design of the Tuned
Mass Dampers (TMDs) it was crucial to have a good characterization of the dynamic behaviour
of the bridge. Therefore, after bridge construction and before the installation of the control devices, dynamic tests were performed. These included an ambient vibration test and several free
vibration tests. A detailed description of all the tests, processing procedures and results can be
found in (Magalhes et al. 2007). This paper is focused in the estimation of the modal damping
ratios.
The ambient vibration test comprehended 19 setups that allowed the measurement of acceleration time series with 16 minutes at 20 sections. The colleted data was processed by the Enhanced Frequency Domain Decomposition method and by the Data driven Stochastic Subspace

313

Experimental Vibration Analysis for Civil Engineering Structures

Identification method (Figure 1a). Both methods provided estimates of natural frequencies and
modal damping ratios for 13 modes in the frequency range 0-5 Hz. In Figure 8 the results of the
SSI-DATA in terms of modal damping ratios are summarized by the average value and by an
interval of variation with an amplitude of twice the standard deviation (average standard deviation of 45%).
As the identification of accurate modal damping ratios was of utmost importance for the correct tuning of the control devices, a total of 10 free vibration tests were performed: 3 vertical
and 3 lateral impulses applied at mid-span (Figure 7), 2 vertical impulses applied in the deck
span over one of the supports in the river (Figure 6) and 2 vertical impulses in a lateral span.
Acc.(g)
0.01

Lateral Acceleration, Section 10

0.00
-0.01
0

15

30

45

60 75 90 105 120 135 150

Time (s)

Figure 7. Free vibration response measured at Coimbra bridge after the application of a lateral impulse
and auxiliary structure used for its application.

The first eight impulses were applied to excite specific modes (the mass was suspended at the
anti-node of the mode). So, it was easy to isolate the contribution of that mode in the measured
free decays, by the application of low order band-pass filters. The results of the procedure are
represented in Figure 8 by the red dots. A total of 8 values are represented: three estimates for
the first and second modes and two estimates for the 10th mode. It is evident the good agreement
between these estimates and the average values from the ambient vibration test.
The last two impulses, applied in a lateral span, excited several modes and so the use of filters became inappropriate. Thus, the colleted decays were processed by the SSI-COV method,
providing the two groups of estimates represented also in Figure 8 by the green marks. These
estimates confirmed the ones provided by the other free decays for the 2nd and 10th mode and it
was possible to obtain estimates for more 5 modes that agree well with the results of the ambient vibration test.
Modal Damp. Ratios (%)

AVT, SSI-DATA
FVT, Filters
FVT, SSI-COV

0
0.0

0.5

1.0

1.5

2.0
2.5
3.0
Natural Frequencies (Hz)

3.5

4.0

4.5

5.0

Figure 8. Modal damping ratios obtained from the Ambient Vibration Test (AVT) and from the Free Vibration Test (FVT).

5 VASCO DA GAMA BRIDGE

The Vasco da Gama Bridge in Lisbon crosses the Tagus River over a total length of 12 Km.
This large structure includes a cable-stayed component (Figure 9) with a main span of 420 m
and 3 lateral spans on each side (62m + 70.6m + 72m), resulting in a total length of 829.2m.The
height of the two H-shaped towers is 147 m.

314

EVACES07

Figure 9. Vasco da Gama Bridge (general view and barge suspended for the application of the impulse)

Modal damp. Ratios (%)

At the end of construction dynamic tests were performed comprising an ambient vibration
test and a free vibration test. During the ambient vibration test the main excitation source was
wind, which suffered speed fluctuations between 1m/s and 22m/s. The free vibration test consisted in the measurement of the bridge responses at 3 sections after the sudden release of a
60ton mass (Figure 9) previously suspended from the bridge deck. To minimize the effect of
undesired aerodynamic damping, this test was performed on a day with low wind speeds (<2.5
m/s).
The identified modal parameters showed that the first mode shape is transversal (BT1) and
has a natural frequency of 0.30 Hz. The first vertical bending (BV1) mode has a natural frequency of 0.34Hz, whereas the first torsion mode (T1) has a natural frequency of 0.47 Hz. Further information about the dynamic tests, the procedures used to process the collected data and
the obtained results are presented in (Cunha et al. 2001) and (Peeters et al. 2002).
In this paper, only the results related with the modal damping coefficients estimates are detailed. Figure 10 shows the modal damping ratios estimated using the data colleted during the
ambient vibration test (AVT) and during the free vibration test (FVT). Like in Coimbra Bridge
the results of the AVT have a significant dispersion (average standard deviation of 36 %) and
the estimates of both procedures of analysis applied to the free decays (filters and SSI-COV) are
close to the average values of the AVT.
AVT, SSI-COV

2.5

FVT, Filters

FVT, SSI-COV
1.5
1
0.5
0
BT1

BV1

BV2

T1

Mode

BV3

T2

BV4

BV5

Figure 10. Modal damping ratios obtained from the Ambient Vibration Test (AVT) and from the Free Vibration Test (FVT).

6 BRAGA STADIUM SUSPENDED ROOF

The Braga Stadium was one of the Stadia constructed in Portugal for the 2004 European Championship (Figure 11). The most outstanding element of the structure is its very flexible suspended roof, which is formed by pairs of full locked coil cables spaced 3.75 m apart from each
other, supporting two concrete slabs over the stands of the stadium. The cables span is 202m
and the slab length is 57.3m, therefore the remaining 88.4m of the central part are free. The
slabs have a thickness of 0.245m.

315

Experimental Vibration Analysis for Civil Engineering Structures

Figure 11. Braga Stadium (view of the east side and records of measured decays in the roof after the application of a sinusoidal resonant load and after the application of an impulse).

The necessity of analysing the susceptibility of the suspended roof to buffeting effects required the accurate experimental identification of the modal damping ratios. Taking that in
mind, three different types of dynamic tests were performed: an ambient vibration test that allowed the identification of the natural frequencies, modes shapes and modal damping coefficients, forced vibration tests with the application of sinusoidal loads and free vibration tests for
the identification of the modal damping ratios. In the reference (Magalhes et al. 2006) the dynamic tests and the obtained results are extensively described.
The roof structure is characterized by low natural frequencies (the first ten modes have natural frequencies between 0.28Hz and 0.74Hz) and has pairs of modes with closely spaced natural
frequencies, which constitutes a significant challenge for the identification.
Figure 12 summarizes the results, in terms of modal damping ratios, of all the used methodologies. Like in the previous applications, the estimates provided by SSI-COV method applied to the ambient vibration data have some dispersion (standard deviations with an an average value of 20%, corresponding to the intervals of variation shown with blue marks). The use
of filters allowed the identification of the modal damping ratios of modes 3, 4, 6 and 10 (FVT,
Filters). For the others, it was not possible to have results owing to the closeness of the natural
frequencies or due to lack of excitation. The application of the SSI-COV method to the same
data (FVT, SSI-COV) made possible the identification of additional damping ratios and, when
the comparison is possible, the results of both techniques (FVT, filter and FVT, SSI-COV) are
almost coincident.
AVT, SSI-COV
FVT, Filters
FVT, SSI-COV
Sin. Decay

damp. (%)

1
0.8
0.6
0.4
0.2
0
1

5
6
Mode

10

Figure 12. Modal damping ratios obtained from the Ambient Vibration Test (AVT), from the Free Vibration Test (FVT) and from the tests with sinusoidal excitation (Sin. Decay).

The forced vibration tests with the application of sinusoidal loads were performed to excite
specifically 5 modes. The identified damping ratios are similar to the ones provided by the impulsive tests. The comparison between modal damping ratios identified using artificial and ambient excitation shows some differences, especially for the first mode. However all the values
are within the interval of variation of the AVT estimates.
7 CONCLUSIONS
The four presented applications have shown that the estimates of the modal damping ratios provided by the processing of the ambient vibration data have a significant dispersion. It was possible to observe fluctuations that motivated average standard deviations (average of the standard
deviations related with all the identified modes) varying from 20% (Braga Stadium Roof) to
45% (Coimbra Bridge). In (Peeters 2000) a simulation study was performed to test several output-only identification methods and it is observed that the SSI-COV method shows a maximum

316

EVACES07

standard deviation of 23% with regard to damping estimation. As it was expected, in real applications the standard deviations are higher, owing to the variation of damping with the amplitude
of vibrations and the effect of wind, that induces aerodynamic damping. These aspects are not
taken into account in the numerical simulations.
Besides that, in all the analysed databases almost all the estimates of modal damping ratios
obtained from the free decays are within the interval defined by the average values +/- the standard deviation of the estimates of the ambient vibration tests, which leads to an increasing confidence in the results provided by ambient vibration testing.
In the analyses of the free decays, it was shown that the SSI-COV method overcomes the
limitations of the procedure based on the application of band-pass filters. Furthermore, it was
presented the possibility of decomposing the free decays in modal decays, even in the presence
of closely spaced modes.
The results taken from the free decays are more accurate, because the noise influence is
smaller (higher amplitudes), there is no need to make assumptions about the loading (in ambient
vibration tests it is assumed a stationary with noise) and the influence of wind can be reduced if
the impulse is applied with low wind velocities. However, these tests are more difficult to perform than the ambient vibration tests and the estimated values are associated with amplitudes
that are higher than the amplitudes observed during the normal use of the structure.
The ambient vibration tests need long periods of acquisition to provide reliable results and so
it is more difficult to ensure low wind velocities during the test to reduce the effect of aerodynamic damping. Anyway, they provide reasonable estimates of the modal damping coefficients
with very economical and practical procedures. It is important to develop further research in order to understand all the reasons for the observed dispersion and to find a process to reduce it.
The data colleted by permanent dynamic monitoring systems can be useful in this context.
ACKNOWLEDGEMENTS
The authors acknowledge all the supports provided by the Portuguese Foundation for Science
and Technology (FCT) for the development of research in the area of Structural Identification
and Monitoring at ViBest / CEC, in particular the Ph.D. Scholarship (SFRH/BD/24423/2005)
provided to the first author.
REFERENCES
Caetano, E.; Magalhes, F.; Cunha, A. (2007) - Comparison of Stochastic Identification Methods applied
to the natural response of Millau Viaduct, EVACES, Porto, Portugal.
Cunha, A.; Caetano, E. (2005) - From Input-Output to Output-Only Modal Identification of Civil Engineering Structures, Keynote Lecture, IOMAC, Denmark.
Cunha, A.; Caetano, E.; Delgado, R. (2001) - Dynamic Tests on a Large Cable-Stayed Bridge. An Efficient Approach, Journal of Bridge Engineering, ASCE, 6(1), 54-62.
LMSInternational (2005). - LMS Test.Lab. Leuvem, Belgium.
Magalhes, F. (2004) - Stochastic Modal Identification for the Validation of Numerical Models, Master
Thesis (in Portuguese), University of Porto, Porto.
Magalhes, F.; Caetano, E.; Cunha, A. (2006) - Operational Modal Analysis of the Braga Sports Stadium
Suspended Roof, IMAC XXIV, St. Louis, Missouri, USA.
Magalhes, F.; Caetano, E.; Cunha, A. (2007) - Dynamic Testing of the New Coimbra Footbridge before
Implementation of Control Devices, IMAC XXIV, Orlando, Florida, USA.
Overschee, Peter Van; Moor, Bart De (1996) - Subspace Identification for Linear Systems, Kluwer Academic Publishers, Leuven, Belgium.
Parloo, Eli (2003) - Application of frequency-domain system identification techniques in the field of operational modal analysis, PhD Thesis, Vrije Universiteit Brussel, Brussels.
Peeters, B.; Roeck, G. De; Caetano, E.; Cunha, A. (2002) - Dynamic Study of the Vasco da Gama Bridge,
ISMA 2002 International Conference on Noise and Vibration Engineering, Leuven, Belgium.
Peeters, Bart (2000) - System Identification and Damage Detection in Civil Engineering, PhD Thesis,
Katholieke Universiteit Leuven, Leuven.
SVS (1999-2004). - ARTeMIS Extractor Pro, Release 3.41. Structural Vibration Solutions, Aalborg,
Denmark.