Professional Documents
Culture Documents
www.elsevier.com/locate/engstruct
Received 24 April 2007; received in revised form 8 November 2007; accepted 9 November 2007
Available online 21 December 2007
Abstract
The paper presents the most relevant results obtained from the application of different output-only modal identification methods to data
collected in dynamic tests of the new Braga Sports Stadium suspended roof and the subsequent finite element model correlation analysis. This
experimental work allowed the validation of the numerical modelling of the suspended roof, which takes into account the geometric non-linear
structural behaviour and the progressive application of loads during the construction phase. Particular attention was given to identification of
modal damping ratios, due to the necessity of analysing the susceptibility of the suspended roof to buffeting effects. Modal damping ratios were
estimated from data provided by free, forced and ambient vibration tests, using an improved implementation of the Enhanced Frequency Domain
Method and the Stochastic Subspace Identification Methods: SSI-DATA and SSI-COV.
c 2007 Elsevier Ltd. All rights reserved.
Keywords: Suspended roof; Ambient vibration test; Free vibration tests; Output-only modal identification; Finite element model correlation
Fig. 1. Braga Municipal Sports Stadium (lateral view of the east stand and top view of the roof from the west side).
experimental validation of the numerical modelling of the direction, allowing relative tangential movements. A transversal
dynamic behaviour of the suspended roof, which takes into triangular truss is suspended from the inner border of each slab
account the geometric non-linear structural behaviour and the acting as a stiffness beam and simultaneously accommodating
progressive application of loads during the construction phase. the floodlights and loudspeakers.
The identification of the modal parameters was an The roof cables are anchored in two large beams at the
interesting challenge, as the flexibility of the roof is associated top of both stands — east and west. The east stand (Fig. 2,
with very low and closely spaced natural frequencies. Particular 3) is structurally formed by 50 m high concrete walls, whose
attention was given to identification of modal damping ratios, geometry was defined in order to minimize the unbalanced
due to the necessity of analysing the susceptibility of the moments at the level of the foundation, motivated by the
suspended roof to buffeting effects. These coefficients were combination of the gravitational action of the stand and the high
estimated using data provided by free, forced and ambient forces transmitted by the roof cables. In the west stand (Fig. 2),
vibration tests, and using an improved implementation of the concrete walls are anchored in the rock and the roof cables’
the Enhanced Frequency Domain Method and the Stochastic tension forces are transmitted to the foundation by prestressing
Subspace Identification Methods: SSI-DATA and SSI-COV. tendons embedded in the concrete.
The comparison of the estimates achieved by application of The outstanding characteristics of the structure and the need
output-only identification methods with the ones provided by for a tight control of the corresponding behaviour during the
data collected in artificial excitation tests was important to construction justified the installation of a monitoring system,
understand the capabilities of each approach. which comprehends static and dynamic components. The static
monitoring system was essential during the construction and
2. Description of the structure is based on a series of load cells installed in the cables’
anchorages, on embedded instrumentation of the concrete
The stadium was constructed on the slopes of Monte Castro, structure (strain gauges, tiltmeters and thermometers) and on
and developed as an amphitheatre over a wide rural landscape, instrumentation of the rock massifs and foundations, with
formed only by two rows of stands, on either side of the load cells installed in the anchors to the earth and in-place
pitch, and by a granite massif (Fig. 2). The most noticeable inclinometers. The dynamic monitoring system is important to
element of the stadium is its roof (Fig. 3), which is formed observe the response of the roof to the wind excitation and is
by pairs of full locked coil cables with diameters varying composed by 6 accelerometers, installed in the inner edges of
between 86 and 80 mm, spaced 3.75 m apart from each the concrete slabs, and by cells to measure the wind pressure at
other, supporting two concrete slabs over the two stands of various points on the underside and top of the roof slabs.
the stadium. The cables’ span is 202 m and the slabs’ length The processing of experimental data continuously collected
is 57.3 m, therefore the remaining 88.4 m of the central part by this long term dynamic monitoring system is beyond
are free. The rain water is drained from the roof along one the scope of the present paper. However, the installed
side only, the slope being achieved by a variation of the accelerometers were used to characterize the dynamic response
length of the cables. The concrete slabs have a thickness of of the roof during the free vibration tests described in Section 5,
0.245 m and are connected to the cables only in the normal whereas the ambient vibration test described in Section 3 was
1690 F. Magalhães et al. / Engineering Structures 30 (2008) 1688–1698
Fig. 3. Views of Braga Municipal Sports Stadium suspended roof (east stand, cables of the roof, slab of the east stand, detail of the connection between the cables
and the slab, detail of the triangular truss in the inner border of the east slab and west stand and slab).
Fig. 4. Placement of one of the seismographs used (a) and measurement points of the ambient vibration test: (b) 1st day, (c) 2nd day. (• reference points; ◦ points
measured with moving sensors).
performed with a portable dynamic measurement system from The test was performed in two days. On the first day,
FEUP, described in the next section. the measurements were done at the points of the west slab
represented in Fig. 4(b), using 13 setups, while on the second
3. Operational modal analysis day the measurements were carried out at the points of the west
and east slabs represented in Fig. 4(c), using 15 setups.
3.1. Ambient vibration test On the first day of measurements, two reference points were
considered (two recorders were permanently placed at points
The ambient vibration test consisted of the measurement of 1 and 7 during all setups). After a preliminary analysis of
the vertical acceleration at 42 points of the roof, using 3 strong the data, it was concluded that, for the frequency range of
motion recorders (Fig. 4(a), http://www.geosig.com). These interest (0–1 Hz), all the modes were detected by the reference
devices are constituted by very sensitive internal force balance sensor placed at point 7. Thus, it was decided to use just this
accelerometers with full-scale range from −0.5g to +0.5g, reference for the remaining measurement sections, in order
a peak to peak noise lower than 50 µg and linear behaviour to reduce the duration of the test. The test developed on the
from DC to 50 Hz, 18 bit analog-to-digital converters (to second day provided a set of new responses on the east slab
guarantee high resolution), batteries that enable autonomy for and more response measurements on the west side slab that
one day of tests, memories consisting of removable Compact allowed the improvement of the spatial resolution used in the
Flash cards that permit a fast download of the acquired data characterization of the mode shapes.
and external GPS sensors to ensure an accurate time, so that For each setup, time series of 16 min were collected with a
they can work independently and synchronously, avoiding the sampling frequency of 100 Hz (minimum sampling frequency
use of cables and minimizing the labour associated with the allowed by the acquisition system). Fig. 5 represents one of
preparation of the dynamic test. In the present test, the use the time series collected at point 7 (reference point). Fig. 6
of cables connecting equipment placed on both sides of the shows the variation of the standard deviation of the time series
suspended roof would make the test extremely complex. measured at point 7 during the 28 setups. The amplitude of
F. Magalhães et al. / Engineering Structures 30 (2008) 1688–1698 1691
Fig. 11. SSI-COV method stabilization diagrams of setup 4: (a) i = 50; (b) i = 100.
Table 2
Natural frequencies and modal damping ratios (SSI-COV method)
Table 3 Table 4
Natural frequencies and modal damping ratios (SSI-DATA method) Calculated vs identified (SSI-COV) natural frequencies
Mode f (Hz) Std. f (Hz) Damp. (%) Std. damp. (%) Mode Frequency (Hz)
Calculated Identified
1 0.275 0.0007 0.51 0.20
2 0.292 0.0005 0.48 0.22 1 0.277 0.276
3 0.521 0.0008 0.39 0.09 2 0.305 0.292
4 0.539 0.0027 0.33 0.22 3 0.520 0.521
5 0.558 0.0012 0.53 0.23 4 0.532 0.539
6 0.631 0.0012 0.47 0.25 5 0.574 0.558
7 0.654 0.0032 0.73 0.51 6 0.610 0.632
8 0.684 0.0012 0.30 0.15 7 0.673 0.655
9 0.701 0.0010 0.31 0.17 8 0.678 0.684
10 0.737 0.0010 0.29 0.14 9 0.712 0.702
11 0.875 0.0015 0.38 0.10 10 0.754 0.737
11 0.844 0.874
Table 5
Summary of all the identified modal damping coefficients (%)
Mode Free vib. filter Free vib. SSI-COV Harmonic excitation Ambient vibration
FDD SSI-COV SSI-DATA
1 – 0.29 0.28 0.58 0.50 0.51
2 – 0.37 0.27 0.52 0.42 0.48
3 0.28 0.32 0.22 0.47 0.44 0.39
4 0.25 0.22 – – 0.40 0.33
5 – 0.44 – – 0.47 0.53
6 0.34 0.36 0.43 0.35 0.54 0.47
7 – 0.29 – – 0.28 0.73
8 – 0.11 0.20 – 0.27 0.30
9 – 0.18 – – 0.26 0.31
10 0.20 0.18 – 0.25 0.26 0.29
11 – – – 0.36 0.41 0.38
Fig. 16. Free decays measured after the application of an impulse (a) and after harmonic excitation of the 2nd mode (b).
The comparison between modal damping coefficients Stadium suspended roof, on the basis of the application
identified using artificial and ambient excitation shows the of different output-only modal identification techniques, and
existence of satisfactory correlation. However, one can notice the subsequent validation of the finite element modelling
that relative differences tend to increase at lower frequencies. previously developed to investigate the static and dynamic
In effect, the difficulty in identifying modal damping ratios behaviour of this complex structure.
is well known, since they are dependent on the amplitude It is shown that, despite the low levels of signal captured
of vibration and also on the wind characteristics, which can during the ambient vibration test and the existence of a
introduce a significant component of aerodynamic damping, large number of modes of vibration in the frequency range
as experimentally observed in [16]. The very low damping 0–1 Hz, the available stochastic modal identification methods
values of this structure make the comparison even more difficult can provide very accurate estimates of natural frequencies
because very small differences are expressed by significant and mode shapes, which present an excellent correlation with
relative errors. the corresponding calculated values. This agreement validates
As for the quality of the estimates provided by the various the sophisticated finite element modelling, which takes into
identification methods, it is relevant to stress that in the present account the geometric non-linear structural behaviour and the
case, the results provided by the FDD method are close to the progressive application of the loads.
ones obtained by the SSI methods, as very long time series Due to the major importance of damping in the aerodynamic
were used and an alternative procedure to estimate the auto- behaviour of the suspended roof, special attention was
correlation function was introduced. It’s still worth mentioning dedicated to the estimation of modal damping ratios either
that the application of the standard EFDD method, using from free vibration or ambient vibration tests. Results achieved
independently the time series of each setup (with 16 min), led to show that, even in this challenging structure, with extremely
values of modal damping ratios for the first modes of about 1%. low natural frequencies and damping factors and with closely
spaced modes, operational modal analysis can provide reliable
6. Conclusions estimates of the order of magnitude of modal damping ratios
(absolute differences inferior to 0.25%), constituting therefore
The paper describes the experimental assessment of the an interesting alternative to the more costly, but certainly more
most relevant dynamic properties of the new Braga Sports accurate, procedures based on artificial excitation. An essential
1698 F. Magalhães et al. / Engineering Structures 30 (2008) 1688–1698
aspect in the application of the operational modal analysis is [6] Brincker R, Ventura C, Andersen P. Damping estimation by frequency
the use of very long time records, preferably collected over a domain decomposition. IMAC XIX. 2001.
[7] Bendat J, Piersol A. Engineering applications of correlation and spectral
period of very low and steady wind velocity condition. Even
analysis. USA: John wiley & Sons; 1980.
so, further research is still needed to improve the potential of [8] Brincker Rune, Krenk Steen, Kirkegaard Poul, Rytter Anders. Identifica-
operational modal analysis to estimate modal damping ratios. tion of dynamical properties from correlation function estimates. Bygn-
ingsstatiske Meddelelser 1992;63(1):38.
Acknowledgments [9] Allemang RJ, Brown DL. A correlation coefficient for modal vector
analysis. In: IMAC I. 1982.
The Ph.D. Scholarship (SFRH/BD/24423/2005) provided by [10] Van Overschee Peter, De Moor Bart. Subspace identification for linear
systems. Leuven (Belgium): Kluwer Academic Publishers; 1996.
the Portuguese Foundation for Science and Technology (FCT)
[11] Magalhães F. Stochastic modal identification for the validation of
to the first author is acknowledged. numerical models. Master thesis. Porto: University of Porto; 2004 (in
Portuguese).
References [12] Peeters Bart. System identification and damage detection in civil
engineering. Ph.D thesis. Leuven: Katholieke Universiteit Leuven; 2000.
[1] Furtado Rui, Quinaz Carlos, Bastos Renato. The new Braga Municipal [13] Peeters B, De Roeck G. Reference based stochastic subspace
Stadium, Braga, Portugal. Structural Engineering International 2005; identification in civil engineering. Inverse Problems in Engineering 2000;
15(2). 8(1):47–74.
[2] Caetano E, Cunha A. Numerical modeling of the structural behaviour of [14] Caetano E, Cunha A, Magalhães F, Furtado Rui. Numerical and
the new Braga Stadium roof, Technical report FEUP/VIBEST. 2001. experimental studies of Braga Sports Stadium suspended roof. Bordeaux
[3] Magalhães F, Caetano E, Cunha A. Operational modal analysis of the (France): EVACES; 2005.
braga sports stadium suspended roof. In: IMAC XXIV. 2006. [15] Magalhães F, Caetano E, Cunha A. Experimental identification of modal
[4] SVS - ARTEMIS Extractor pro, release 3.41. Structural vibration damping ratios from the new Braga Stadium roof. Technical report. 2004.
solutions, Aalborg (Denmark); 1999–2004. [16] Macdonald JHG, Daniell WE. Variation of modal parameters of a cable-
[5] Brincker R, Zhang L, Andersen P. Modal identification from ambient stayed bridge identified from ambient vibration measurements and FE
responses using frequency domain decomposition. In: IMAC XVIII. 2000. modelling. Engineering Structures 2005;27(12):1916–30.