You are on page 1of 12

ARTICLE IN PRESS

Soil Dynamics and Earthquake Engineering 28 (2008) 593604


www.elsevier.com/locate/soildyn

Dynamic parameters of structures extracted from ambient vibration


measurements: An aid for the seismic vulnerability assessment of
existing buildings in moderate seismic hazard regions
Clotaire Michela,, Philippe Gueguena,b, Pierre-Yves Barda,b
a
Laboratoire de Geophysique Interne et Tectonophysique, Universite Joseph Fourier, CNRS, Grenoble, France
b
Laboratoire Central des Ponts-et-Chaussees, Paris, France

Received 23 October 2006; received in revised form 23 March 2007; accepted 3 October 2007

Abstract

During the past two decades, the use of ambient vibrations for modal analysis of structures has increased as compared to the
traditional techniques (forced vibrations). The frequency domain decomposition (FDD) method is nowadays widely used in modal
analysis because of its accuracy and simplicity. In this paper, we rst present the physical meaning of the FDD method to estimate the
modal parameters. We discuss then the process used for the evaluation of the building stiffness deduced from the modal shapes. The
models considered here are 1D lumped-mass beams and especially the shear beam. The analytical solution of the equations of motion
makes it possible to simulate the motion due to a weak to moderate earthquake and then the inter-storey drift knowing only the modal
parameters (modal model). This process is nally applied to a nine-storey reinforced concrete (RC) dwelling in Grenoble (France). We
successfully compared the building motion for an articial ground motion deduced from the model estimated using ambient vibrations
and recorded in the building. The stiffness of each storey and the inter-storey drift were also calculated.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Ambient vibrations; FDD method; Existing buildings; Modal model; Seismic vulnerability; Grenoble

1. Introduction modulus of the cracked or undamaged concrete) and of


the structural design (e.g. irregularity of the shear
Knowing the dynamic parameters of a structure (e.g. a resistance or soft storey). All large-scale vulnerability
building or a bridge) may be useful: (1) to calibrate its assessment methods (e.g. Refs. [13]) identify the stiffness
elastic properties for numerical modelling, (2) to detect the regularity of buildings as one of the main parameters
modication of its behaviour after retrotting or damage, controlling their seismic resistance. The stiffness is also the
(3) and, nally, to predict its behaviour under earthquakes. basic parameter needed to draw the capacity curve of
The linear dynamic behaviour can be fully described by the buildings, which is the basis of recent large-scale vulner-
modal parameters: resonance frequencies, modal shapes ability assessment methods (e.g. Refs. [1,3,4]). One of the
and damping ratios. These parameters mainly depend on greatest difculties within such methods is the lack of
the storey masses, which remain unchanged whatever the information on existing buildings. Within vulnerability
state of the structure, and the storey stiffnesses, which are assessment, we have to deal with questions on, for example,
inuenced by the structural modications such as reinfor- ageing, structural design, quality of materials and the
cing and damage. It can also be representative of the building state.
quality of the material (e.g. the equivalent Youngs Assessing the stiffness of each storey, and therefore the
modal shapes of the structure, is then a critical point for the
Corresponding author. Tel.: +33 4 76 82 80 71; fax: +33 4 76 82 81 01. evaluation of the dynamic properties of existing buildings
E-mail address: cmichel@obs.ujf-grenoble.fr (C. Michel). and hence to predict their seismic response and vulnerability.

0267-7261/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.soildyn.2007.10.002
ARTICLE IN PRESS
594 C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604

The ratio between transverse and longitudinal frequencies cost and the efciency of operational modal analysis, the
can, for example, provide information on the relative ambient vibration method is well adapted to large-scale
stiffness in perpendicular directions that may expose a lack analysis for which a large set of buildings has to be
of stiffness and may help in structural design under- analysed. We rst describe the FDD method we used in
standing. Comparing the frequencies and mode shape this study, one method among a large set of other existing
before and after the shaking [57] allows the evaluation of methods. Then we propose the shear beam model, adapted
damage. This method requires knowledge of the initial to our recording layout to approximate the motion of a
state of the structure in such a way that the modication building. An analytical solution is proposed to estimate the
due to damage can be observed. For example, comparisons stiffness at each storey as a function of the modal shapes.
of frequencies among a group of identical buildings can We apply this theoretical work to a building located in
quantify the damage level of each building [8]. Based on the Grenoble (France). We estimate its modal properties using
integrity threshold concept [9], and with only limited ambient vibration tests and recordings of a ground motion
structural information, the integrity of the building under induced by a bridge demolition (BD) in the same area. We
seismic shaking can be estimated using the experimental then validate the modal model derived using modal
modal shape deduced from forced or ambient vibrations parameters estimated from AVs by comparing the simula-
(AVs). tion results of the bridge collapse with real data. Finally,
Ambient vibrations provide information about the we estimate the stiffness at each storey of this building and
modal parameters of a structure that we can extract using the inter-storey drift during the seismic motion and deduce
modal analysis methods. There are many different techni- some information about its seismic behaviour.
ques to identify the natural modes of a structure, which
can be divided into parametric and non-parametric 2. Modal parameters from AV recordings
methods. In the rst category, the parameters of the
considered model are updated to t the recorded data in In the 1930s, US Coast and Geodetic Survey rst
frequency or in time domain (see Ref. [10] for details). In undertook AV recordings in high-rise buildings and
the second category, the methods use only signal proces- bridges to determine their fundamental periods [12,13].
sing tools so that they are more user-friendly and easier to Until the late 1990s engineers preferred forced vibration
implement. Moreover, existing civil engineering structures tests (traditional modal analysis) because of the accuracy of
are often difcult to model relevantly so that it is necessary the corresponding system identication techniques. How-
to test these approaches before any more complicated ever, during the last two decades AV recordings are
method. For example, the Peak Picking (PP) method preferred because of their low cost. They are used to
consists in taking the frequency peaks of average spectra determine the behaviour of structures (frequencies and
for each sensors placed at different points. The frequency modal shapes) [14,15], to quantify the damage after
domain decomposition (FDD [11]) is an improvement of earthquakes [8,1618] and to assess the benet of retro-
the PP method. It consists of decomposing the power tting [19,20]. They also allow the calibration of numerical
spectral density (PSD) matrices into single-degrees-of- and analytical modelling (model updating) [15,21,22]. AVs
freedom systems by singular value decomposition. are produced by natural sources such as local atmospheric
The main goal of this paper is to demonstrate the utility conditions (e.g. the wind and the sea) or by human
of techniques using AV recordings to complement and activities (e.g. trafc and factories). The expected range of
improve seismic vulnerability investigations of existing acceleration values for ambient vibration tests is 107 to
buildings. AVs recorded in buildings are processed for 104 g. Therefore, there is no doubt that the low level of
estimating the dynamic parameters of the building based the shaking only gives relevant information on the
on modal parameter estimates (Fig. 1). Because of its low elastic behaviour of the structure. Many authors (e.g.
Refs. [1719,2326]) showed that nonlinearity affects
resonance frequencies when the level of loading increases.
This nonlinearity is due to the opening of micro-cracks in
the concrete during vibrations, which may decrease
temporarily, or permanently the stiffness, and then the
resonance frequencies, of the structure. Nevertheless,
recent studies showed this decrease may be low for weak
to moderate motions [17] and is usually less than 20%,
which conrms the interest of conducting AV recordings
within buildings.
Recording AVs at different points of a civil engineering
structure (e.g. a bridge, a building or a chimney) allows the
Fig. 1. Overview of this studys scope. We relate, in a simple way,
experimental modal parameters of a building (frequencies, modal shapes
determination of its modes of vibration through opera-
and damping) to ground-motion parameters (earthquake scenario) and tional modal analysis techniques [27]. The efciency of the
structural parameters (stiffness) for seismic vulnerability assessment. output-only modal analysis algorithms, the low cost of
ARTICLE IN PRESS
C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604 595

ambient vibration tests and the wide range of potential [Y] and [H] are r  r, m  m and m  r, respectively. In
applications are the main reasons why their use is wide- operational modal analysis, the usual assumption is that
spread today. the input is white noise, which means:
Practical issues for vibration recordings are, on one
X o C, (2)
hand, the layout of measurement arrays and, on the other
hand, the frequency and time length of recordings. The which is a constant matrix. The [H] matrix can be written in
number of recorded points depends on the height, a pole (lk)/residue ([Rk]) form as:
complexity and accessibility of the building. For example,
Xn
Rk  Rk 
in dwellings, only the stairs are usually available so that Ho , (3)
motions can only be recorded at one point per oor. The k1
jo  lk jo  l k
drawback is that it is impossible to discriminate torsion
modes in that case. The frequency of sampling depends on where j2 1.
the greatest frequency we want to estimate, practically Using Eqs. (2) and (3) and for low values of damping, it
can be shown that the term Rk CR k  dominates the
never more than 30 Hz for usual buildings. Moreover,
Brincker et al. [27] showed that in practice the time length expression of [Y] (Eq. (1)). In this case, this term becomes
of the recorded window should be at least equivalent to also proportional to the mode shape vector {fk}:
1000 periods of the structure in order to calculate an k  d k fFk gfFk gT ,
Rk CR (4)
accurate spectral estimate.
Contrary to forced vibrations and earthquakes, the where dk is a constant. Only a limited number of modes
excitation is transmitted by external forces spread along (typically no more than two) noted Sub(o) has energy at
the structure (wind), by the ground (seismic ambient noise) one particular angular frequency o. The PSD matrices of
and by internal forces (human occupancy). Hence, it is the outputs (Eq. (1)) then have the following form:
impossible to determine the input load. The techniques
X d k fFk gfFk gT d k fF
k gfF
k gT
used to determine the modal parameters of the considered Y o . (5)
system are called output-only or operational modal k2Subo
jo  lk jo  lk
analysis [10]. Many different techniques exist, the most
widely used and easiest one being the PP method. It In practice (see Fig. 2 for an illustrated case for synthetic
consists of calculating the Fourier transforms of short time data), the AVs recordings allow the estimation of the PSD
windows (several seconds) and picking the value of the matrices of the outputs Y^ oi at each known frequency
frequency peaks of the average spectrum. The amplitude of oi, i.e. the Fourier Transforms of the cross-correlation
the peak for each simultaneous recording gives a point of matrices. We used Welchs method [28] for this estimation,
the modal shape. The normalised shape is obtained by which means we took the average square spectrum of
dividing it by the value at the reference sensor. Theoreti- moving Hamming windows in the signal. The length of the
cally, it only gives the operational deection shapes, which windows is directly linked to the precision in the frequency
are not characteristics of the system but depends on the we require. The common number of points in the Fourier
input. Practically, PP can give accurate estimates of the Transform we used was 8192, which means windows of 20 s
modal shapes if the modes are well separated, that is often for a 200 Hz sampling frequency, corresponding to a
the case for regular tall buildings. frequency step of 0.024 Hz.
Many other methods in time or frequency domains have The PSD matrices are symmetrical and therefore can be
been developed under the theory of System Analysis. The diagonalised such as
easiest one is called frequency domain decomposition [11]
Y^ oi U i Si U
i T , (6)
and consists of decomposing the PSD matrices into single-
degrees-of-freedom systems by singular value decomposi- where [Ui] is the matrix of the singular vectors and [Si] is
tion. The initial description of the FDD method can be the diagonal matrix of the singular values. Close to a peak,
found in Ref. [11]. The PSD matrices are the Fourier if only one mode is dominating so that there is only one
transforms of the correlation matrices between all the term in Eq. (5), the rst singular vector is an estimate of the
simultaneously recorded signals so that no a priori model is modal shape. If two modes are dominating at this
supposed. The PSD matrices of the input (unknown) and frequency peak, which means two modes are close in
the output (recorded) signals, functions of the angular frequency, and if they are geometrically orthogonal, the
frequency o, can be respectively noted [X](o) and [Y](o). estimates of the corresponding modal shapes are the two
They are linked to the frequency response function matrix rst singular vectors. Moreover, the estimate of the two
[H](o) through the equation: close mode shapes is still unbiased. In conclusion, the FDD
method allows the accurate extraction of the mode shapes

Y o HoX oHoT , (1)
from AVs recorded simultaneously at several points of the
where T denotes transpose and H is complex conjugate. If r structure. Several sets of recordings can give the mode
is the number of inputs and m the number of simultaneous shapes of the whole structure since a reference sensor
recordings, at each angular frequency o, the sizes of [X], allows the normalisation of the shape.
ARTICLE IN PRESS
596 C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604

Fig. 2. Practical use of the frequency domain decomposition (FDD) method on synthetic data of a Gaussian random signal convolved with a sine function
of frequency 0.1 (simulation of a two-storey building) (simultaneous recordings) (a). The singular value decomposition (c) extracts redundant information
in the power spectral density (PSD) matrices Y(3  3) (b). The obtained modal shape is exactly the preassigned one (d).

Many peaks may appear in the rst singular value 1


computed from AVs, not only those corresponding to
structural modes. The Modal Assurance Criterion (MAC) 0.9 MAC>80%

[29] gives a quantitative value to compare two modal 0.8


shapes F1 and F2 through the following expression:
0.7
 H 2
F F 2  0.6
MAC value

MACF1 ; F2  H 1  H  , (7)


F1 F1 F2 F2 0.5 unused points
H
where denotes complex conjugate and transpose. The 0.4
MAC has been used in order to evaluate the damage by
0.3
comparing the initial and the nal mode shapes [5,6] as MAC
well as to compare modes obtained by different ways 0.2 1st Sing. Value
Selected Mode
(e.g. different methods or recordings). Moreover, compar- 0.1
ing the modal shape at a given frequency peak with the
modal shape at the surrounding frequencies can give the 0
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
extent of the bell representing the corresponding single
Frequency (Hz)
degree of freedom (Fig. 3). In practice, as long as the MAC
value between the shape at one frequency and the shape at Fig. 3. Validity range of the building mode extracted by FDD method
the peak is greater than 80%, we consider that the using a MAC value greater than 80%. The solid line is the normalised rst
frequency belongs to the mode. By this way, it is possible singular value of PSD matrices of ambient vibration recordings in a
building. The bold line (rst singular vectors with a MAC value greater
to determine the frequency bandwidth where each mode is than 80% except for two unused points) is one of the estimated mode of
dominating and to discriminate peaks that do not the structure. The crosses display the MAC values associated with this
correspond to a structural mode. mode.
ARTICLE IN PRESS
C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604 597

3. Stiffness deduced from modal parameters where fFj gT KfFj g=fFj gT MfFj g o2j are the eigenva-
lues, ({Fj}T[C]{Fj})/({Fj}T[M]{Fj}) 2xjoj and ({Fj}T[M]{1})/
3.1. Modelling of the system ({Fj}T[M]{Fj}) pj is the participation factor of mode j.
The analytical solution of these N independent equations
As the mass per storey can be considered as constant, the for yj(0) 0 and y0j 0 0 is known as Duhamel integral,
main parameter controlling the mode shape is the stiffness. given by
The mode shapes of the system (building) are unscaled Z
pj t
parameters, which are linked to the scaled physical 8j 2 1; N yj t U 00S t exj oj tt sino0 t  t dt,
parameters used in earthquake engineering (e.g. forces, o0 0
acceleration, Youngs modulus and inertia). For that (10)
reason, in order to derive the physical parameters of the
structure from the modal parameters, a model is necessary. with o0 2j o2j 1  x2j .
Based on the knowledge of the structural parameters of the Assuming a constant mass per oor equal to 1000 kg/m2
building, the modal parameters and the output parameters and knowing the geometry of recording array, it is
(i.e. stress and strain in the building) are deduced from possible to determine the elastic motion of the system
forward modelling. On the other hand, the a priori under moderate earthquakes from the knowledge of
structural parameters of the model have to be updated by the modal parameters determined using AVs, without
solving the inverse problem [21], based on the knowledge of any assumptions having to be made on the structural
the modal parameters. design.
Modelling techniques using the nite element method In order to evaluate the stiffness matrix from the modal
(e.g. Refs. [15,30]) and homogenisation [9] can be found in parameters, an assumption on the behaviour of the system
literature. The updating procedure can use only the rst must be made. Two models are generally employed for the
frequency to check if the others are well represented by the description of building behaviour [31]: a cantilever beam
model or all the measured modal parameters. In this paper, representing shear-wall RC buildings and a shear beam for
we use simple elastic lumped-mass modelling because of its RC-frame buildings.
low cost for a large set of buildings. Moreover, it does
not require detailed information on the building (e.g. the
plan, structural design and quality of materials), which
is usually not available for existing buildings. Since the 3.2. Stiffness matrix form
modal parameters are deduced using, at most, a few AV
recordings per oor, the elastic lumped-mass modelling is The shear beam model assumes that the motion at one
accurate enough. It amounts to neglecting the internal storey depends only on the motion of the storeys above and
deformation of the oors and in considering the deforma- below. The underlying hypothesis is that the oors are
tion between two storeys as linear. much stiffer than the walls. The stiffness matrix [K] can be
written as follows:
Considering displacement in only one direction,
Newtons second law for the displacement of the con- 2 3
k1 k2 k2 0   0
sidered degrees of freedom (DOF) {V(t)} can be written as 6 .. .. 7
follows: 6 k k2 k3 k3 . . 7
6 2 7
6 . . . .. 7
6 . . . 7
MfV 00 tg CfV 0 tg KfV tg fStg, (8) 6 0 k3 . . . . 7
K 6
6 .. .. .. .. ..
7
7
6 . . . . . 0 7
where [M] is the mass matrix (it is diagonal with the mass 6 7
6 . . 7
mi of each DOF i), [C] is the damping matrix, [K] is the 6 .. . . k 7
4 n1 kn1 k n kn 5
stiffness matrix and {S(t)} is the applied force. 0   0 kn kn
To solve this linear system, we simultaneously diagona-
lise matrices [K] and [M] and assume that [C] is also (11)
diagonal with the N damping ratios xj on the diagonal. The
N eigenvalues o2j the corresponding eigenvectors Fj and where ki is the stiffness of the storey i.
damping ratios xj are the modal parameters of the system. For the shear beam assumption, the characteristic ratio
Considering a building forced into vibration by an earth- between the harmonics and the fundamental frequency
quake, the displacement Vi(t) of each DOF i is the sum of is [31]:
the displacement of the ground US and the unknown
fk
relative displacement of the structure Ui. The solution of 8k 2 1; N 2k  1. (12)
Eq. (8) is of the form {U(t)} [F]{y(t)}. f1
The equation of motion can be rewritten: The equation of the eigenvalues KfFi g o2i MfFi g for
the shear beam model can be then inverted in order to
y00j t 2xj oj y0j t o2j yj t pj U 00S t, (9) evaluate the stiffness matrix [K].
ARTICLE IN PRESS
598 C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604

The corresponding linear system can be solved into the The building is a nine-storey reinforced concrete (RC)
analytical formula: structure with both frames and shear walls (Fig. 4). It was
built in 1939 and it is one of the most important classes of
P
n
ml Fli buildings found in the urban area of Grenoble. The
8j 2 1; N kj o2i
lj
, (13) concrete is poorly reinforced and it has been built without
Fji  Fj1i using a seismic building code. It is located within an urban
block, with two walls in contact with the surrounding
with F0i 0.
buildings, which are only separated by lled joints.
This analytical formula (Eq. (13)) allows the extraction
Following the European Macroseismic Scale (EMS98)
of the stiffness values at each oor from the modal shapes.
[32], it corresponds to the type Reinforced Concrete
Theoretically, only one mode is necessary to calculate
Frame Building which is one of the most vulnerable RC
the stiffness matrix under the shear hypothesis. Never-
classes. In this study, we used the Cityshark II acquisition
theless, this formula shows that when two close-together
system [33], a user-friendly seismological station dedicated
storeys have approximately the same deection for one
to AVs experiments. It allows the simultaneous recording
particular mode, i.e. FjiEF(j1)i, the stiffness values are
of 18 channels. Six Lennartz Le3D-5s velocimeters with a
highly sensitive to uncertainties in the modal shape
at response in the 0.250 Hz frequency band were used.
determination. Therefore, considering two modes reduce
these uncertainties.
4.1. First experiment: ambient vibrations

4. Example of stiffness estimation using experimental data For the rst experiment, AVs were recorded for 15 min
(Fig. 4), at a sample rate of 200 Hz, which is long enough
To illustrate the use of AV for the stiffness evaluation of with respect to the Brincker criterion (1000 periods) [27].
the building, two experiments have been carried out in a The sensors were oriented along the longitudinal direction
dwelling located in the centre of Grenoble (France). The of the building. Two series of recordings were performed in
AVs and the vibrations induced in this structure by the order to record one point per oor (Fig. 4). The sensors at
demolition of a close-by bridge allowed the extraction of the top of the building (8th and 9th oors) were kept as
the modal parameters of the building using the FDD references since they have the greatest amplitudes while the
method. We compared the modal parameters found in others were roving sensors.
both experiments and used them: Using the FDD method, the two rst bending modes in
each direction have been extracted (Fig. 5; Table 1): at 2.73
 to build a modal model that can be validated by and 7.71 Hz in the longitudinal direction and at 2.28 and
recordings of the bridges collapse; 8.69 Hz in the transverse direction. Moreover, despite the
 to estimate the stiffness of the building at each storey in low level of energy it is possible to identify the frequency of
each direction. the third bending mode in each direction but not the modal

Fig. 4. Left: the studied building; centre: the experimental scheme for the rst and second ambient vibration (AV) datasets; right: time history acceleration
deduced from AV velocity recorded at the eighth oor.
ARTICLE IN PRESS
C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604 599

Longitudinal cross-section Transverse cross-section


35 35
Modes
30 30
Long. 1

25 25 Long. 2

108 Trans. 1
2.73 Hz
First Singular Value of the

2.27 Hz 20 20
Spectral Density Matrix

Z (m)
Trans. 2
109
4.74 Hz 15 15
5.64 Hz 8.69 Hz
7.71 Hz
1010 10 Planar View
10
1

Y (norm.)
1011 5 5 0

1
1012 0 0
1 2 3 4 5 6 7 8 9 10 15 1 0 1 1 0 1 1 0 1
X (normalized) Y (normalized) X (norm.)
Frequency (Hz)

Fig. 5. Building frequencies selected in the rst singular values of the power spectral density matrices computed from AV recordings (left) and
corresponding modal shapes (right) extracted by the FDD method (bending modes only). For each identied mode, the bold bell corresponds to a MAC
value greater than 80%.

Table 1
Frequencies of the building modes extracted from AV using the FDD
technique

Mode Frequency (Hz) fk/f1

Transverse 1 2.28 1
Longitudinal 1 2.73 1
Torsion 1 4.74
Torsion 2 5.64
Transverse 2 8.69 3.8
Longitudinal 2 7.71 2.8
Longitudinal 3 1213 4.5
Transverse 3 15.5 6.8

shapes. That may be the same reason for the second


transverse mode showing a curious modal shape. For the
longitudinal direction, the ratio between harmonics and
fundamental frequencies t the theoretically assumed
sequence 1, 3, 5 for the shear beam model. In the transverse Fig. 6. Aerial location of the bridge demolished in July 2004 (A) and the
study-building (B).
direction, the ratios are 1, 3.8 and 6.8.
This ratio is a characteristic of the model and helps to
choose which model provides a closer t to the building. position of the shear walls and beams induces undoubtedly
Our building (Table 1) behaves therefore more in bending this kind of torsional motion, that could be conrmed
in the transverse direction than in the longitudinal one. using several recording points at the same oor.
This can also be seen in the experimental modal shapes
(Fig. 5) where the rst transverse modal shape looks like a 4.2. Second experiment: bridge demolition
cantilever beam. Nonetheless, we used the shear beam
model for the two directions of the building, knowing this In order to check the relevancy of the AVs for the
model represents only roughly the longitudinal direction. evaluation of the modal parameters, the second experiment
Since the stiffness is proportional to the square of the was focused on building motion evaluation for ground
frequency, the rst frequencies tell us that the longitudinal excitation. This motion was induced by the controlled
direction is 40% stiffer than the transverse one. demolition of a bridge located 40 m away from the building
The FDD method shows other modes that may (Fig. 6). The same instrument was employed for this second
correspond to torsional behaviour Table 1. The irregular experiment. Because only six Lennartz Le3D-5s sensors
ARTICLE IN PRESS
600 C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604

could be plugged into the CityShark II, we installed them The modal parameters extracted from AVs and the
in six of the nine storeys of the building (Fig. 9). Two induced ground motion match so well that we can conclude
hundred seconds of signal were kept and the sampling rate on one hand that the building responded totally elastically
chosen for these recordings was 100 Hz (Fig. 7).
The instantaneous collapse of the deck (source time
duration of about 5 ms) generated a vertical seismic x 104
motion with a frequency peak around 11 Hz (Fig. 8). 6
Ground floor Longitudinal
The horizontal peak ground acceleration (PGA) recorded Ground floor Transverse
in the ground oor of the building was 0.025 m/s2 5
(Table 2). This value is greater than the PGA recorded
during recent Alpine earthquakes by the French Accelero-
4
metric Network (http://www-rap.obs.ujf-grenoble.fr) in

Amplitude
the Grenoble basin. Nevertheless, it is 60 times lower than
the design acceleration values enforced in the present code 3
[34] for the Grenoble area (1.5 m/s2). The maximum
horizontal acceleration at the top of the building was 2
0.061 m/s2.
Although the basic assumption of white noise is
1
required, the FDD method was also used to determine
the modes of the structure for these explosion recordings.
The following results show that neglecting the white noise 0
assumption does not greatly affect the modal parameter 1 2 Y1 X1 4 5 6 7 8 9 10 15 20
determination, as already shown by Ventura [35]. Only the Frequency (Hz)
rst bending mode in each direction (Fig. 9) and the rst Fig. 8. Fourier amplitude spectra of the recordings of the bridge
supposed torsion mode can be found. The transverse mode demolition at the ground oor of the structure. The spectra are smoothed
exhibits a non-negligible torsion part in its modal shape using the function described in Ref. [38]. The modal frequencies of the
that was not seen in the AV recordings. This explains the structure deduced from ambient vibrations are labelled X1 and Y1.
low correlation observed between the two experiments
(Fig. 9). The rst bending and torsion modes closely match Table 2
for the two recordings with a difference of less than 1% in Maximum accelerations recorded for the bridge demolition (m/s2)
frequency (Table 3). The MAC values show a very high Direction Ground oor Top oor
similarity between the mode shapes obtained in the two
experiments (Table 3). Some points of the modal shape Longitudinal 0.025 0.051
Transverse 0.019 0.062
were linearly interpolated for the BD because only six
Vertical 0.046 0.094
points could be simultaneously recorded.

0.06
Top floor 0.06
0.04 Top floor
Acclration (m/s2)

Acceleration (m/s2)

0.04
0.02
0.02
0
0
0.02
0.02
0.04
0.04
0.06
Ground floor 0.06
0.04 Ground floor
Acceleration (m/s2)

Acceleration (m/s2)

0.04
0.02 0.02
0 0
0.02 0.02
0.04 0.04
0.06 0.06
0 2 4 6 8 10 0 2 4 6 8 10
Time (s) Time (s)

Fig. 7. Accelerations produced by the bridge demolition in the structure: time histories in longitudinal (left) and transverse (right) directions at the ground
and top oors.
ARTICLE IN PRESS
C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604 601

Longitudinal cross-section Transverse cross-section Modes


35 35 Long. 1 AV

Long. 1 BD
30 30
Trans. 1 AV
25 25
Trans. 1 BD

20 20
Z (m)

15 15

10 10 Planar vue
1

Y (norm.)
5 5
0.5

0 0 0
0 0.5 1 0 0.5 1 0 0.5 1
X (normalized) Y (normalized) X (norm.)

Fig. 9. Comparison of the rst modal shapes extracted from AV and BD (left) and experimental scheme during the BD (right). The storeys that were not
measured are linearly interpolated.

Table 3 The displacements computed and recorded at the top of


Comparison of frequencies and modal shapes (MAC value) extracted the building are displayed in Fig. 10. We observe a good t
from ambient vibrations (AV) and bridge demolition (BD) experiments
between the data and the modelling. Even if the complexity
Mode f (Hz) Comparison of the building behaviour is not integrated in the model,
the simulation reproduces the level of amplitudes as well as
AV BD Frequency MAC value the duration, the frequency and the phase of the building
variations (%) (%)
motion. It is also possible to compute the inter-storey drift
Transverse 1 2.28 2.27 0.4 85 by difference between the time history displacements at
Longitudinal 1 2.73 2.75 0.7 99 consecutive oors. The maximum drift (Fig. 11) has been
Torsion 1 4.74 4.77 0.6 97 calculated in this way using the modal model and the
Transverse 2 8.69 8.46 2.6 20
recordings for the BD. An average of the model in non-
recorded storeys has been performed to compare the same
quantities. The model reproduces the magnitude of the
to the ground motion and on the other hand that both maximum drift but in the longitudinal direction, the drift at
modal identications were reliable, even if in the second the upper storeys and especially the fourth is under-
case the assumption of white noise input is not valid and estimated.
the length of the signal limited. Nevertheless, these comparisons conrm the relevancy
of the modal parameters of buildings extracted from AVs.
4.3. Simulation of the buildings response to earthquakes and Hence, once the modal parameters are known, it is possible
the calculation of stiffness to evaluate the ability of the building to resist (or not) the
seismic loading corresponding to any a priori ground
In order to validate the aforementioned methodology, motion scenario. If we suppose that the maximum drift
we rst veried if the motion of the building when forced sustained by concrete shear walls controlled by shear
into vibration by the BD would be well predicted by AV without any damage is 4  103 [37], we are able to
modal modelling. Once the experimental mode shapes are determine for a chosen ground motion if the building
dened, and by following the Section 3.1, the building reaches this threshold or not. As expected, in the case of
motion and its behaviour under earthquakes are simulated the bridge demolition experiment (Fig. 11), the maximum
using Eq. (10) and the experimental modal parameters drift remains very low (around 6  106). It is quite
extracted from the AVs. In this study, only the two rst similar in the longitudinal and transverse directions.
bending modes in each direction were used. We estimated The model outlines that the third oor is subjected to a
the corresponding damping ratio using the random greater drift in the transverse direction but does not show
decrement method [36] as 4% for all modes except for similar behaviour in the fourth storey in the longitudinal
the rst longitudinal mode (where 2% is found). direction.
ARTICLE IN PRESS
602 C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604

x 105 x 107

5 Longitudinal Simulated
Simulated 6
Displacement (m)

Recorded Recorded

Amplitude
4
0

2
Longitudinal
5
0

x 105 x 107

5 Transverse
6
Displacement (m)

Amplitude
4
0

2
Transverse
5
0
0 2 4 6 8 10 1 2 3 4 5 6 7 8 9 10
Time (s) Frequency (Hz)

Fig. 10. Comparison of time histories (left) and spectra (right) at roof level between the observations and the modelling of the building response to the
bridge demolition.

8 8
L mod Longitudinal from 1st mode
Longitudinal from 2nd mode
7 T mod 7 Transverse from 1st mode
L rec
T rec 6
6
5
5
Story
Story

4
4
3
3
2

2 1

1 0
1 2 3 4 5 6 0 1 2 3 4 5
Inter-storey drift (m/m) x 106 Stiffness (N/m) x109

Fig. 11. Comparison of the drift envelope at each available storey Fig. 12. Stiffness of the building calculated from the AV modal shapes.
modelled and recorded in the structure during the bridge demolition. Solid The second transverse mode gives non-physical results.
and dashed lines are the longitudinal and transverse motion, respectively.
Thin and bold lines are for recordings and modelling, respectively.
reects the sensitivity to the error at the anti-node of the
second mode, when the deection is lower. For the last
Assuming a shear model for this building, as suggested oor, considering the rst mode shows the same impreci-
by the ratio f2/f1E3, we can deduce the stiffness at each sion in the stiffness evaluation. This illustrates the necessity
oor following Eq. (13). The results are displayed in to simultaneously consider more than one mode in order to
Fig. 12. The stiffness at each storey is between 20% and increase the quality of the stiffness evaluation. It shows
50% larger for the longitudinal direction than for the also the necessity in estimating the uncertainty of the
transverse one. In the longitudinal direction, because of the extracted modal shapes. It should be noticed that even if
quality of the modes extracted from the AVs, two modes the amplitudes vary slightly, these three independent
were available to determine the stiffness. In this direction, stiffness shapes show the same variations along all the
the only difference between the two modes appears for the storeys. This assures that these stiffnesses are valid, for
third storey. As mentioned in Section 3.2, this difference example the third storey is softer than the second and the
ARTICLE IN PRESS
C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604 603

fourth. This can be the reason for the greater strain [2] GNDT. Instruzioni per la Compilazione de lla Sceda di Relivamento
recorded (and modelled) for this storey (Fig. 11). Esposizione e Vulnerabilita Sismica Degli Edici. Italy: Gruppo
From these results, a linear trend of stiffness, decreasing Nazionale per la Difesa dai Terremoti, Regione Emilia Romagna y
Regione Toscana; 1986 [in Italian].
with height, in both directions is evident for this building. [3] RISK-UE. An advanced approach to earthquake risk scenarios with
It is not a very irregular conguration regarding seismic applications to different European towns. European Project EVK4-
vulnerability. However, the third storey has been identied CT-2000-00014, 2003.
as a soft storey. In case of an earthquake, the building [4] Lang K, Bachmann H. On the seismic vulnerability of existing
must resist a larger drift at this storey making it more buildings: a case study of the City of Basel. Earthquake Spectra
2004;20(1):4366.
vulnerable than other buildings of the same RC class. [5] West WM. Illustration of the use of modal assurance criterion to
detect structural chanes in an orbiter test specimen. In: Air Force
5. Conclusions conference on aircraft structural integrity; 1984. p. 16.
[6] Salane HG, Baldwin JW. Identication of modal properties of
bridges. J Struct Eng 1990;116(7):200821.
Herein we demonstrated the utility of a set of simple [7] Hans S, Boutin C, Ibraim E, Roussillon P. In situ experiments and
techniques using AV recordings to complement seismic seismic analysis of existing buildingsPart I: Experimental investiga-
vulnerability assessment investigations of existing buildings. tions. Earthquake Eng Struct Dyn 2005;34(12):151329.
Decomposing the motion of a building into simple modes [8] Dunand F, Ait Meziane Y, Gueguen P, Chatelain J-L, Guillier B, Ben
(bending and torsion) is the rst step in the assessment of its Salem R, et al. Utilisation du bruit de fond pour lanalyse des
dommages des batiments de Boumerdes suite au seisme du 21 mai
behaviour under an earthquake. We computed, therefore, 2003. Mem Serv Geol Alg 2004;12:17791 [in French].
the modal parameters using the FDD method, which is easy [9] Boutin C, Hans S, Ibraim E, Roussillon P. In Situ experiments and
to perform and reliable even in case of close modes. The seismic analysis of existing buildingsPart II: Seismic integrity
experiments showed that the FDD is able to extract reliable threshold. Earthquake Eng Struct Dyn 2005;34(12):153146.
information (frequency, modal shapes and eventually damp- [10] Peeters B. System identication and damage detection in civil
engineering. PhD thesis, Katholieke Universiteit Leuven, 2000.
ing) even if the basic assumption of white noise is not [11] Brincker R, Zhang L, Andersen P. Modal identication of output
fullled. The values of the measured frequencies are directly only systems using frequency domain decomposition. Smart Mater
linked to the stiffness of the building so that we can assess the Struct 2001;10:4415.
average stiffness ratio between the longitudinal and the [12] Carder DS. Observed vibration of buildings. Bull Seismol Soc Am
transverse directions. For vulnerability assessment this is not 1936;26:24577.
[13] Carder DS. Observed vibration of bridges. Bull Seismol Soc Am
sufcient so we used a simple lumped-mass model with an 1937;27(4):26789.
assumption on the stiffness matrix (shear beam) in order to [14] Brownjohn J. Ambient vibration studies for system identication of
calculate the stiffness at each storey. The choice of this tall buildings. Earthquake Eng Struct Dyn 2003;32(1):7195.
stiffness model is based on the sequence of the resonance [15] Ventura C, Liam Finn W-D, Lord JF, Fujita N. Dynamic
frequencies of the studied building. The inversion of the characteristics of a base isolated building from ambient vibration
measurement and low level earthquake shaking. Soil Dyn Earthquake
stiffness matrix is analytical in this case and two modes are
Eng 2003;23:31322.
practically needed for a reliable stiffness determination. [16] Clinton J, Bradford SC, Heaton T, Favela J. The observed wander of
Having the stiffness at each storey, we can identify soft the natural frequencies in a structure. Bull Seismol Soc Am
storeys or bad structural congurations, which increase the 2006;96(1):23757.
vulnerability of the building to earthquakes. We showed, for [17] Dunand F, Gueguen P, Bard P-Y, Rodgers J, Celebi M. Comparison
of the dynamic behaviour extracted from weak, moderate and strong
example, that the studied building displayed an irregularity
motion recorded in buildings. In: Proceedings of the rst European
in the third storey, which makes it more vulnerable to conference on earthquake engineering and seismology, 38 Septem-
earthquakes than other buildings of the same type. The ber 2006, Geneva, Switzerland, paper #1021.
presented method is therefore able to identify the most [18] Todorovska M, Trifunac MD, Hai TY. Variations of apparent
vulnerable buildings within a building class. The modal building frequencieslessons from full-scale earthquake observa-
tions. In: Proceedings of the rst European conference on earthquake
model, free of any assumptions on the stiffness behaviour of
engineering and seismology, 38 September 2006, Geneva, Switzer-
the building, allows the computation of its elastic motion land, paper #1547.
subjected to weak to moderate earthquakes and then the [19] Celebi M. Before and after retrotresponse of a building during
inter-storey drift. Using directly this drift or by applying the ambient and strong motions. J Wind Eng Ind Aerodyn 1998;77 &
integrity threshold concept developed by Boutin et al. [9] for 78:25968.
[20] Gentile C, Martinez Y, Cabrera F. Dynamic investigation of a
reinforced concrete, one can decide whether the building
repaired cable-stayed bridge. Earthquake Eng Struct Dyn 1997;26:
reaches the inelastic range, i.e. the beginning of permanent 4159.
damage. Several earthquake scenarios can also be applied in [21] Teughels A. Inverse modeling of civil engineering structures based on
order to identify for what acceleration levels the building operational modal data. PhD thesis, Katholieke Universiteit of
stays within its elastic domain. Leuven, December 2003.
[22] Ventura C, Ding Y. Linear and nonlinear seismic response of a
52-storey steel frame building. Struct Des Tall Build 2000;9:2545.
References [23] Celebi M. Dynamic characteristics of ve tall buildings during strong
and low-amplitude motions. Struct Des Tall Build 1993;2:115.
[1] FEMA. HAZUS earthquake loss estimation methodology. Washing- [24] Celebi M. Seismic response of two adjacent buildings. I: Data and
ton, DC: Federal Emergency Management Agency; 1999. analysis. J Struct Eng 1993;119(8).
ARTICLE IN PRESS
604 C. Michel et al. / Soil Dynamics and Earthquake Engineering 28 (2008) 593604

[25] Celebi M. Comparison of damping in buildings under low [32] Gruenthal G. European Macroseismic Scale 1998, vol. 15. Luxem-
amplitude and strong motions. J Wind Eng Ind Aerodyn 1996;59: bourg: Cahiers du Centre Europeen de Geodynamique et de
30923. Seismologie; 1998.
[26] Celebi M, Safak E. Seismic response of transamerica building. I: Data [33] Chatelain JL, Gueguen P, Guillier B, Frechet J, Bondoux F, Sarrault
and preliminary analysis. J Struct Eng 1991;117(8). J, et al. Cityshark: a user-friendly instrument dedicated to ambient
[27] Brincker R, Ventura C, Andersen P. Why output-only modal testing noise (microtremor) recording for site and building response studies.
is a desirable tool for a wide range of practical applications. In: 21st Seismol Res Lett 2000;71(6):698703.
international modal analysis conference (IMAC), Kissimmee, FL, [34] DTU Re`gles PS92. French seismic building code, 1995 [in French].
2003. [35] Ventura C, Laverick B, Brincker R, Andersen P. Comparison of
[28] Welch PD. The use of Fast Fourier Transform for the estimation dynamic characteristics of two instrumented tall buildings. In: 21st
of power spectra: method based on time averaging over short, international modal analysis conference (IMAC), Copenhague,
modied periodograms. IEEE Trans Audio Electroacoust 1967; Denmark, 2003.
AU-15:703. [36] Brincker R, Krenk S, Jensen JL. Estimation of correlation functions
[29] Allemang R-J, Brown D-L. A correlation coefcient for modal vector by the random decrement technique. In: 9th international modal
analysis. In: 1st international modal analysis conference (IMAC), analysis conference and exhibit, Florence, Italy, 1991.
Orlando, FL, 1982. [37] Federal Emergency Management Agency. Prestandard and commen-
[30] Pan T-C, Brownjohn J, You X-T. Correlating measured and tary for the seismic rehabilitation of buildings (fema356). Technical
simulated dynamic responses of a tall building to long-distance report November, FEMA ASCE, 2000.
earthquakes. Earthquake Eng Struct Dyn 2004;33:61132. [38] Konno K, Ohmachi T. Ground motion characteristics estimated
[31] Clough RW, Penzien J. Dynamics of structures. New York: from spectral ratio between horizontal and vertical components of
McGraw-Hill; 1993. microtremor. Bull Seismol Soc Am 1998;88(1):22841.

You might also like