Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Bull Earthquake Eng (2016) 14:2543–2563

DOI 10.1007/s10518-016-9916-5

ORIGINAL RESEARCH PAPER

Structural monitoring and earthquake early warning


systems for the AHEPA hospital in Thessaloniki

K. Pitilakis1 • S. Karapetrou1 • D. Bindi2 • M. Manakou1 •

B. Petrovic2 • Z. Roumelioti1 • T. Boxberger2 •


S. Parolai2

Received: 20 November 2015 / Accepted: 12 April 2016 / Published online: 26 April 2016
Ó Springer Science+Business Media Dordrecht 2016

Abstract This study aims at presenting the analyses of monitoring data that have been
used in the context of structural monitoring and Earthquake Early Warning (EEW) for a
hospital building in Thessaloniki. Permanent and temporary instrumentation arrays,
implemented under the responsibility of Aristotle University of Thessaloniki (SDGEE-
AUTH) in close cooperation with German Centre for Geosciences (GFZ) are presented.
The ambient noise data recorded at the temporarily installed networks are used for the
dynamic characterization of the building based on both vibrational and waveform
approaches. Moreover, long-term ambient noise recordings from the permanent array
installed within the hospital are used for the investigation of the daily and seasonal
wandering of the building resonance frequencies related to environmental effects. The
modal identification results are used in a comprehensive framework for the computation of
the up-to-date fragility curves representing the actual structural state considering aging
effects of the construction materials, possible pre-existing damages and changes in the
geometry and mass distribution. The building-specific fragility functions are integrated into
two independent EEW systems and rapid damage assessment approaches, namely the
PRESTo software and an onsite EEW algorithm on the instruments of the permanent array,
to provide the expected level of damage after strong ground shaking at the monitored
building. The implemented monitoring networks and the developed operational tools can
be used in the context of seismic risk mitigation and preparedness for structural safety
assessment under earthquake loading.

Keywords Building monitoring  Modal identification  Interferometry  Environmental


and operational variation

& S. Karapetrou
gkarapet@civil.auth.gr
1
Department of Civil Engineering, Research Unit of Geotechnical Earthquake Engineering and Soil
Dynamics, Aristotle University, Thessaloniki, Greece
2
Helmoholtz Centre Potsdam GFZ German Research Centre for Geosciences, Potsdam, Germany

123
2544 Bull Earthquake Eng (2016) 14:2543–2563

1 Introduction

Risk and vulnerability of urban sites to earthquake hazard lead to an emerging need for
developing operational frameworks that can be used by the authorities (e.g. civil protection
authorities, end users) in pre-crises situation to establish decision making procedures and
risk mitigation strategies. The collection of monitoring data is a prerequisite to enhance the
reliability in the safety assessment of structures. Real-time monitoring of civil structures
and infrastructures provides valuable information to assess the health and identify the
actual state and vulnerability of the associated systems. Furthermore, it allows monitoring
the evolution of the structure’s safety during the earthquake crisis while it constitutes the
key component for the development of an Earthquake Early Warning (EEW) system in the
context of real-time risk assessment and seismic safety risk evaluation in an urban scale.
The aim of the paper is to present monitoring schemes and operational tools that may be
used in the context of real time seismic risk mitigation and preparedness for site response
evaluation and structural safety assessment under earthquake loading. In the present study,
we assess the dynamic characteristics and the seismic performance of an eight-storey
hospital building (AHEPA) constructed in the seventies in Thessaloniki (Greece). This
building, belonging to one of the largest hospitals in Greece, has been selected in the
framework of European funded project REAKT (Strategies and Tools for Real Time
EArthquake RisK ReducTion, http://www.reaktproject.eu/) as test site for developing
structural monitoring techniques and an EEW system aiming to (a) identify the real
structural condition and potential pathology of the building, (b) evaluate the actual today
vulnerability of the building and (c) integrate these vulnerability algorithms in recently
developed EEW systems aiming to estimate the real time seismic vulnerability of the
building during an upcoming seismic event. To accomplish these goals the building has
been instrumented with permanent and temporary monitoring arrays in close cooperation
of Aristotle University of Thessaloniki (AUTH) with the German Centre for Geosciences
(GFZ), Helmholtz Centre in Potsdam. Noise measurements and strong ground motion data
collected from the temporary and permanent arrays are used for the dynamic structural
characterization and the derivation of building-specific fragility curves.

2 The AHEPA hospital structure

The AHEPA general hospital in Thessaloniki is one of the largest hospitals in northern
Greece, located in the campus of Aristotle University (Fig. 1). It is a major teaching and
research center and part of the National Healthcare System of Greece. The target building
hosts both administration and hospitalization activities. It was constructed in 1971 and is
considered representative of structures that have been designed according to the old 1959
Greek seismic code (Royal Decree on the Seismic Code for Building Structures 1959),
where the ductility and the dynamic features of the constructions are ignored. During the
Thessaloniki earthquake in 1978 (Moment magnitude, Mw = 6.5; Epicentral distance,
Repi = 26.7 km, Papazachos et al. 1979; Soufleris et al. 1982), which generally caused
extensive damages and the collapse of one high-rise residence structure, the hospital
building suffered only slight damage. It is an eight storey infilled structure and its special
feature is that it is composed of two adjacent tall building units that are connected with a
structural joint (Fig. 1). UNIT 1 covers a rectangular area of 29 m by 16 m while UNIT 2
has a trapezoidal cross section of 21 m by 27 m by 16 m. The total height of the building

123
Bull Earthquake Eng (2016) 14:2543–2563 2545

Fig. 1 The AHEPA hospital. Typical floor plan and middle floor of the hospital with the structural joint and
the two adjacent building units UNIT 1 and UNIT 2 (dimensions in m)

with respect to the foundation level is 28.6 m with a constant inter-storey height of 3.4 m
except for the second floor where the height increases to 4.8 m due to the presence of a
middle floor level which covers only a part of the typical floor plan. From the structural
point of view the building’s force resisting mechanism comprises longitudinal and exter-
nally transverse reinforced concrete moment resisting frames (Fig. 1). The foundation
system consists of simple footings of variable geometries without tie-beams combined
partially with a raft foundation. The foundation soil consists of stiff clay with an average
shear wave velocity Vs equal approximately to 400–450 m/s and can be characterized as
soil type B according to Eurocode 8 soil classification system (CEN 2004). Using the
SYNER-G taxonomy (Pitilakis et al. 2014) for reinforced concrete (RC) structures to
describe the typology of the hospital building, it may be considered typical of high-rise
infilled moment resisting frame buildings designed with low seismic code level. Table 1
summarizes the main characteristics of the two units, which are deduced from the available
design and construction blueprints, namely the total mass, the concrete (fc) and steel (fy)
strength, which are used for the structural analyses of the two building units.
AHEPA is monitored on a continuous basis since May 2012 by a permanent
accelerometric array, the SOSEWIN array (Bindi et al. 2015a), which comprises 13 triaxial
accelerometers (MEMS ADXL203 chip) installed on the basement, the 1st and 4th floor,
and the roof of the hospital (Fig. 2). The stations record with sampling rate 100 Hz, and
their timing is achieved through GPS antennas. Eight local and regional earthquakes with

Table 1 Main characteristics of the hospital building units


RC building Total mass (t) fc (MPa) fy (MPa)

UNIT 1 3804.0 14.0 220.0


UNIT 2 3144.0 14.0 220.0

123
2546 Bull Earthquake Eng (2016) 14:2543–2563

Fig. 2 Floor plans of the basement, 1st and 4th floors and the roof of the AHEPA hospital showing the
permanent SOSEWIN array

Table 2 Relevant parameters of the earthquakes with magnitude ML [ 2.5 recorded at the SOSEWIN
array during May 2012 until July 2015
No Date Time (GMT) Lat°N Long°E Depth (km) ML

1 12 May 2012 22:48:12.80 40.564 22.841 7.4 4.0


2 22 May 2012 00:00:33.20 42.601 23.063 7.0 6.0
3 11 Oct 2013 05:15:46.90 40.689 23.410 3.8 4.4
4 26 Oct 2013 17:07:14.30 40.227 23.102 8.1 3.6
5 6 Dec 2013 17:00:52.20 40.799 22.938 9.5 2.6
6 24 May 2014 09:25:02.10 40.286 25.375 12.8 6.2
7 16 Jul 2014 16:14:58.00 40.569 22.948 12.2 3.0
8 22 Aug 2014 04:27:54.00 39.935 23.431 13.5 5.0
9 26 Jul 2015 18:40:42.80 40.460 23.960 11.5 4.0

local magnitude ML [ 2.5 have been recorded from May 2012 until July 2015 in the
SOSEWIN array (Table 2). Ambient noise and earthquake data from the SOSEWIN array
are used for the dynamic characterization of the building, and for the testing in real-time of
the two independent EEW and rapid damage assessment approaches for Thessaloniki.

123
Bull Earthquake Eng (2016) 14:2543–2563 2547

3 Dynamic characterization

The dynamic behavior of a building can be determined using either vibrational or wave-
form approaches. With the vibrational methods this is achieved through the identification
of the modal parameters of the structure (Chopra 1996) whereas, with waveform
approaches, through the properties of seismic wave propagation within the building (e.g.
Nakata et al. 2013). Dynamic characterization of civil engineering structures plays a
significant role not only in the dynamic response prediction but also in structural health
monitoring and vibration control engineering where damage detection may be accom-
plished through the comparison between the structural modal properties extracted from the
response of the system under undamaged and damaged conditions. There are several
studies available in the literature, which investigate the variations in natural frequency due
to damage (e.g. Banks et al. 1996; Ray and Tian 1999 etc.) or use the changes in mode-
shapes to detect and locate damage along a building (e.g. Pandey et al. 1991; Pandey and
Biswas 1994). In this study the short term ambient noise measurements are used to perform
both vibrational (operational modal) and waveform (interferometric) analyses, in order to
gain insight on the dynamic behavior and seismic response of the hospital building units.
Structural health monitoring in this case is more related to the quantification of the severity
of damage through the vulnerability assessment based on monitoring data rather than the
identification and localization of damage along the building. Moreover real-time moni-
toring of civil structures and infrastructures provide valuable information to assess the
structural health and identify the actual state and vulnerability of the associated systems
(Guéguen et al. 2007; Michel et al. 2008, 2012). Furthermore, it allows monitoring the
evolution of the structure’s safety during the earthquake crisis while it constitutes the key
component for rapid damage assessment or the preparation of reliable damage scenarios.
Ambient noise measurements were collected on February 2013 from a temporary array
of 39 triaxial seismometers (Mark Products L4C-3D of 1 Hz coupled to 24-bit Earth Data
Logger, PR6-24) in the two building units. The scope of the temporary experiment was the
dynamic characterization of the hospital building in terms of eigenfrequencies and mode
shapes. Four stations were installed along the middle corridor of each floor near and far
from the structural joint of the building (Fig. 3). The instrumentation array was defined so

Fig. 3 Instrumentation (red squares) during the temporary array on February 2013. Section A–A0 and B–B0
are along the longitudinal and transversal direction of the hospital (see Fig. 1)

123
2548 Bull Earthquake Eng (2016) 14:2543–2563

as to capture the translational and torsional modes of the two building units. North
direction of the stations was placed parallel to the longitudinal structural direction of the
building. Ambient noise was recorded by all stations simultaneously for about 4 h with
500 Hz sampling rate and gain 10. The time synchronization among the stations was
established through GPS antennas.

3.1 Short term noise measurements

3.1.1 Operational modal analysis

The use of field monitoring data constitutes a significant tool for the representation of the
actual structural state, reducing uncertainties associated with the building modeling
properties as well as many non-physical parameters (aging effects, maintenance, etc.)
enhancing thus the reliability in the ‘‘real-time’’ risk assessment procedure. Building
monitoring data are used to identify the actual model properties of a structure based on
system identification and operational modal analysis (Reynders 2012). For the AHEPA
hospital case parametric (Stochastic Subspace Identification—SSI, Van Overschee and De
Moor 1991) and non-parametric methods (Frequency Domain Decomposition—FDD,
Brincker et al. 2000) are employed to extract the modal properties of the systems analyzed.
Table 3 summarizes the modal identification results in terms of eigenfrequencies and mode
shape types for the two adjacent building units separately (UNIT 1 and UNIT 2) as well as
for the entire hospital building analyzed as one taking into account the interaction of the
two building units due to their connection with the structural joint (BUILDING). Similar
orders and shape types of the modes are estimated for the different systems analyzed
(UNIT 1, UNIT 2, BUILDING) and the different identification methods applied. The mode
shapes corresponding to the first five identified frequencies are indicatively presented for
BUILDING (Fig. 4). Based on the identification results, the two building units cannot be
considered as an entire monolithic building, neglecting thus the influence of the structural
joint, as they do not present a common modal behavior. In fact the identified mode shapes
reveal a coupling of the two building units under low vibration (operational conditions).
Under strong ground motion however, due to the high nonlinearities which the systems are
expected to experience and taking into account the fact that the two units do not have
common foundation, the interaction between the two units may be completely different
affecting consequently their seismic response. As will be discussed in Sect. 4, the iden-
tified modal parameters are used to calibrate the modeling properties of the two building
units in order to represent their actual structural state. The modal identification analyses as

Table 3 Modal identification results for UNIT 1, UNIT 2 and BUILDING estimated using parametric and
non-parametric identification techniques
Mode Mode type UNIT 1 UNIT 2 BUILDING

FDD (Hz) SSI (Hz) FDD (Hz) SSI (Hz) FDD (Hz) SSI (Hz)

1 Coupled translational 1.65 1.65 1.65 1.65 1.65 1.65


2 Coupled translational 1.90 1.91 1.91 1.91 1.91 1.91
3 Torsional 2.33 2.33 2.35 2.33 2.35 2.33
4 1st longitudinal 3.50 3.47 3.58 3.52 3.58 3.51
5 2nd longitudinal 5.20 5.15 5.22 5.16 5.20 5.15

123
Bull Earthquake Eng (2016) 14:2543–2563 2549

Fig. 4 Mode shapes corresponding to the five first identified frequencies for BUILDING

well as the discussion on the results may be found in detail in Bindi et al. (2015a) and
Karapetrou et al. (2016).

3.1.2 Interferometric analysis

The building impulse response functions (IRF) have been determined following a wave-
form approach based on seismic interferometry. The impulse response function DðxÞ at a
certain level (in our case, the height within the building) is defined as
uðxÞ
DðxÞ ¼ ð1Þ
uref ðxÞ
in the frequency domain, with u(x) and uref(x) being the Fourier transforms of
u(t) (recordings at a certain height) and uref(t) (recording at the reference location),
respectively (Snieder and Safak 2006). x = 2pf is the angular frequency.
Since the impulse response function may be unstable because of zeros of the spectrum
uref(x), the so-called water-level regularization (e.g. Wiggins and Clayton 1976) is applied
DðxÞ ¼ F ðxÞuðxÞ ð2Þ

uref ðxÞ
F ðxÞ ¼ ð3Þ
uref ðxÞ 2 þ e

where e is the regularization parameter and was chosen in our study as 0.1 % of the
average spectral power after a detailed analysis of the parameter selection (Bindi et al.
2015a). The IRFs are calculated by applying a moving window with a duration of 32 s,
50 % overlapping and a cosine-tapered at both ends, to a 1 h noise recordings. For the
horizontal components that are shown in this study a bandpass filtering [0.1–20 Hz] is
applied. Figure 5 shows the IRFs obtained for the two horizontal components (left: lon-
gitudinal, right: transverse) of the motion of UNIT 2 using as reference the motion
recorded at the roof. The IRFs for the basement and the first two floors (green traces) show
more complex patterns than the IRFs for the upper floors (blue traces) where the up and
down-going pulses are clearly identified. The complex patterns of the lower floors are
attributed to the complex distribution of internal sources within these levels due to
entrances at the basement level and the first floor. The vertical component is not shown
here since the pulses of up and down-going waves are not resolved well due to the high
P-wave velocity.
The time lag between the propagation pulses at different floors is used to estimate the
shear wave velocity. In Fig. 6 the travel times between the up and downward propagating

123
2550 Bull Earthquake Eng (2016) 14:2543–2563

Fig. 5 IRFs obtained for the longitudinal (left) and transverse (right) components. The blue and dashed
grey lines indicate the average velocities obtained by the interferometry for the upper and lower part of the
building, respectively

Fig. 6 Travel times between the propagation pulses at different floors versus distance to the roof,
considering the longitudinal (left) and transverse (right) components. The slowness is determined using least
squares fit, grouping the results of all columns. The red symbols indicated the velocity estimates from the
third to the sixth floors

pulses are presented versus the distance of the considered sensor to the reference sensor for
the estimation of the slowness for the longitudinal (left) and transverse (right) directions.
The average velocities are estimated considering the recordings for all four locations
(Fig. 3) within the building equal to 200 (±6) m/s and 276 (±6) m/s for the transverse and
the longitudinal direction, respectively for a building layer from the third floor to the roof.
At the seventh floor the up-going and down-going pulses cannot be resolved (Fig. 5), hence
this floor is not considered for the estimation of the average velocity from the 3rd floor to

123
Bull Earthquake Eng (2016) 14:2543–2563 2551

the roof. The distribution of the shear wave velocity estimated from interferometric
analysis using the ambient noise recordings indicates that the structure is stiffer at the base
and along the transverse direction, with a change of velocity between the second and third
floors (Fig. 5). The increased stiffness at the base is expected due to the fact that the
basement is partially embedded and because the dimensions of the reinforced concrete
elements (columns) are progressively decreasing along its height. A detailed presentation
of the interferometric analysis and the discussion on the results may be found in Bindi et al.
(2015a).
Methods for damage detection and localization based on the seismic interferometry
(Ivanicovic et al. 2001; Todorovska and Trifunac 2008a, b; Todorovska 2009; Picozzi et al.
2011) are related to the estimation of the shear wave propagation velocities. The changes in
the total travel time due to variationsof the average velocity, are related to the changes in
the seismic behavior of the structure alone, i.e., to the fundamental frequency of the fixed-
base building (Todorovska and Trifunac 2008b; Todorovska 2009). The fundamental
frequencies obtained from the vibrational approaches are linked to the soil-building sys-
tem. Moreover, using the seismic interferometry has the advantage that damage can be
localized on a smaller scale. Local changes in the travel times between different floors or
layers of the building consisting of more floors are directly related to changes in the
velocity and hence, to changes of the stiffness (Todorovska and Trifunac 2008a).

3.2 Long term noise measurements: Check the stability of the results

In the time period of February to April 2014 four more triaxial seismometers (Guralp
CMG-40T of 30 s coupled to DAS130-1 24-bit digitizers) were installed on different floors
of UNIT 1, in addition to the permanently installed broad-band sensor at the basement. The
stations were installed on the 2nd, 3rd, 4th floor and the roof of UNIT 1, almost at the same
position along the height of the building (Fig. 7). Recording was on a continuous basis with
200 Hz sampling rate, while the timing of each station was achieved through GPS antenna.
The aim of monitoring the building with broad-band stations was to check the stability of

Fig. 7 Locations of the broad-band sensors (red cylinder) installed within the AHEPA hospital (9858:
basement, 9085: 2nd floor, 985F: 3rd floor, 908E: 4th floor and 9086: roof). Black squares show the
permanent SOSEWIN array

123
2552 Bull Earthquake Eng (2016) 14:2543–2563

the results obtained from the temporary array of February 2013. Additional goals were to
observe possible daily oscillation of the building resonance frequencies by analyzing long-
term ambient noise data, and to compare the results with those obtained from the noise
recorded by the permanent SOSEWIN array (Figs. 2, 7). In particular, while the SOSE-
WIN network allows extending the analysis to some years of continuous recordings, the
lower sensitivity of these instruments limits the noise investigations to the assessment of
the variation with time of the fundamental frequency of resonance. A comparison (here not
shown) of the SOSEWIN results obtained for the same time period when the broad-band
sensors were installed, confirms their reliability.
Using ambient noise recordings from the station 9086 installed at the roof of the
building during the installation period, the power spectra density (PSD) was calculated
versus time (Fig. 8a). The power spectra were calculated using running windows of 15 min
length, with 50 % overlapping and tapering at both ends. Only the frequency range
(1.2–2.8 Hz) including the fundamental resonance frequencies of the building is shown.
The variation of the maximum fundamental frequency obtained from the PSD during the
3 months period, is approximately 5 %, as Fig. 8b shows. Actually, the fundamental fre-
quency is varying between 1.63 Hz and 1.75 Hz. The time–frequency image of the fun-
damental frequency variations shown in Fig. 8b are obtained by applying the S-transform
(Stockwell et al. 1996). The results presented in Fig. 9 show that the fundamental fre-
quency of the building varies with a frequency of 1 and 2 per day. Diurnal variations in the
frequencies can be associated both to thermal effects (e.g. expansion of the concrete) and to
the daily building usage cycle. Considering that the hospital is operative 24 h per day, with
heating and cooling system continuously operating, external (atmospheric) and internal
thermal effects appear to be the main source for the daily frequency variation. Indeed,
more long-term and reversible variations driven by temporal changes of the atmospheric
conditions have been recognized in several studies (e.g. Clinton et al. 2006; Todorovska
and Al 2006; Herak and Herak 2010; Mikael et al. 2013). For example, Clinton et al.
(2006) and Herak and Herak (2010) discussed the changes in the natural frequencies during
rain events, while other studies (e.g. Hua et al. 2007; Nayeri et al. 2008) discussed the

Fig. 8 a Transverse power spectra density (PSD) for the broad-band station installed at the roof (station
9086 in Fig. 7) of the building in decibel (dB) with respect to velocity [(m/s)2/Hz], for the period February
2014 till April 2014. The grey regions depict data lack, due to station malfunctions. b Variation of the
maximum fundamental frequency obtained from the PSD

123
Bull Earthquake Eng (2016) 14:2543–2563 2553

Fig. 9 Time–frequency analysis


of the variation of the building
fundamental frequency with time

impact of temperature variations on modal frequencies and damping (Herak and Herak
2010; Mikael et al. 2013). Figure 8 shows some variations occurring over the scale of few
days (see for example days after 100). The availability of the permanent SOSEWIN array
allows investigating the presence of monthly trends in the wandering of the fundamental
frequency.
In particular, the variations with time of the resonance frequencies along the longitu-
dinal direction of the building were investigated using approximately 2 years noise
recordings of the SOSEWIN station SB22C, located close to the broad-band station 9086 at
the roof of the hospital (Fig. 7). Several frequency oscillations are identified, in particular
seasonal, weekly and daily variations (Fig. 10a), as detailed by the time–frequency anal-
ysis (S-transform) performed for year 2014 (Fig. 10b). A long-term positive drift starts
around day 100 (i.e., early April), reaching a maximum excursion around day 200 (middle
of July) both in 2014 and 2015. After that, the trend inverts direction, restoring the original
average value around day 260 (middle of September), in 2014.During the summer period,
the variation of the fundamental frequency also shows components in the range
0.07–0.15 day-1, corresponding to periods of 2 and 1 week, respectively. The peak-to-
peak amplitude of the daily variation of the fundamental frequency is about 0.05 Hz; the
summer trend further modulates this variation, leading to a maximum peak-to-peak vari-
ation of about 0.15 Hz, corresponding to about 10 % of the average value. Seasonal trends
related to environmental effects have been observed in previous studies. For example,
Mikael et al. (2013) analyzed the changes in the fundamental frequency and damping of
four buildings in France. While the increase of the fundamental frequency with increasing
temperature and exposure to the sun observed for two out of the four analyzed buildings
was in agreement with other studies (e.g. Clinton et al. 2006; Herak and Herak 2010), the
results achieved for the other two buildings showed different trends (i.e., either a negative
correlation or a more complex behavior). These results led the authors to conclude that the
correlation between temperature and changes in the fundamental frequency could be
building dependent and additional data are needed to understand the physical parameters
explaining the observed trends. In the case of AHEPA, a RC building constructed over stiff
clay, changes in the fundamental frequency show a positive correlation with temperature
and exposure to sun, reaching the maximum value in July.

123
2554 Bull Earthquake Eng (2016) 14:2543–2563

Fig. 10 a Variation of the maximum fundamental frequency from June 2013 to October 2015. The red line,
obtained through polynomial fitting, describes the trend in the data. b Time–frequency analysis of the
variation of fundamental frequency with time considering year 2014

The wandering of the fundamental frequency of the structure, which depends on both
building elastic properties and on the soil-structure coupling, should be taken into account
when used as reference for damage identification or other identification analysis. Quan-
titatively correlation of the frequency variation with changes in the atmospheric signals is
still ongoing.
In order to validate the obtained results, another time frequency analysis, the Empirical
Mode Decomposition, EMD (Huang et al. 1998) has been applied to the ambient noise data
of the SOSEWIN station SB22C. This method allows decomposing any complicated signal
into a finite number of Intrinsic Mode Functions (IFMs). The method is adaptive and it is
applicable to linear, non-linear and non-stationary signals. Furthermore, each IMF can also
be analyzed in the time–frequency domain using the Hilbert transform. Figure 11 shows
that also with this approach a clear 1 day and 12 h periodicity in the variation of the

123
Bull Earthquake Eng (2016) 14:2543–2563 2555

Fig. 11 Time–frequency analysis of signal obtained using the Hilbert Huang method applied to ambient
noise data of the SOSEWIN station SB22C installed at the roof of the building

fundamental resonance frequency is observed as well as the long period (between 0.05 and
0.1 per day period) around days 130 and 230. This coincidence highlights that the results
are not artifacts of the proposed analysis, but they indicate real changes in the characteristic
of the building structure or the soil occurred in that period of time.

4 Seismic vulnerability assessment using monitoring data

The modal identification results presented in Sect. 3.1.1 are used to update and better
constrain the initial finite element model of the building units, which are based on the
design and construction documentation plans. In the absence of any structural geometry
modification since 1971 when the building was constructed, only the variation in the
material properties is taken into account. An eigenvalue sensitivity analysis of the elastic
numerical models is performed to identify the most sensitive parameters influencing the
structural modes of interest, which are used in the manual updating process to define the
optimal analytical models that reflect the experimental results. As sensitivity parameter for
the updating procedure, the compressive strength of the masonry infill is selected to take
into account the uncertainties of the material behavior as well as the possible heterogeneity
between the material properties of the different infill parts. Besides the masonry strength of
the infills, also the sensitivity in the concrete strength was investigated as well as the
combination of both parameters. The updating procedure using the concrete strength
showed that frequencies and mode shapes are not affected significantly. Thus only one
sensitivity parameter is used, namely the compressive strength of the infill panels, in order
to avoid complicated updating schemes and allow a better observation of the updating
procedure. Modal analyses for all the derived numerical modal models are performed in
OpenSees for the three dimensional elastic linear finite element models of the two adjacent
buildings separately (UNIT 1 and UNIT 2). Only one among them is considered as the best
model representing the measured dynamic response. The selection of the best model is
based on the evaluation of the Modal Assurance Criterion-MAC (Allemang and Brown

123
2556 Bull Earthquake Eng (2016) 14:2543–2563

1982). The computation of the MAC values and the correlation of the responses between
the experimental and numerical modal models are made at 18 nodes for each building unit
(2 nodes per floor at each unit corresponding to the sensor locations). A good correlation
between the two tested modes is considered to be achieved for MAC values [0.8. In
Table 4 the results of the applied updating methodology is presented indicatively for UNIT
1. It is seen that high MAC values are obtained for the structural modes of interest
indicating a very good correlation between the experimentally and numerically derived
mode shapes.
Three-dimensional incremental dynamic analyses (IDA) of the nonlinear initial and
updated models are performed for real ground motion accelerograms, which are selected
based on regional seismic hazard, to derive the seismic fragility curves. IDA is an
emerging analysis method which involves performing a series of nonlinear dynamic
analyses under a suite of multiply scaled ground motion records whose intensities should
be ideally selected to cover the whole range from elasticity to global dynamic instability
(Vamvatsikos and Cornell 2002). In particular the earthquake scenario consists of a set of
15 real ground motion records (selected from the European Strong motion database, http://
www.isesd.hi.is) referring to stiff soil conditions classified as soil type B according to EC8
(CEN, 2004) which is the soil category of the foundation soil according to a detailed
geotechnical survey performed in the site (Anastasiadis et al. 2001). The primary selection
criterion was the average acceleration spectra of the set to be of minimal ‘‘epsilon’’ (Baker
and Cornell 2005) at the period range of interest with respect to the regional acceleration
spectrum adopted from SHARE for a 475 year return period (http://portal.share-eu.org:

Table 4 Comparison of the updated finite element model of UNIT 1 with the initial model and the
experimental results (T period, f frequency)
Initial FEM T(s)/f(Hz) Mode shape of updated Mode shape of experimental MAC
FEM T(s)/f(Hz) model T(s)/f(Hz)

Coupled translational 0.96


T1 = 0.69 s/f1 = 1.46 Hz

T1 = 0.64 s/f1 = 1.56 Hz T1 = 0.61 s/f1 = 1.65 Hz


Coupled translational 0.94
T2 = 0.48 s/f2 = 2.06 Hz

T2 = 0.53 s/f2 = 1.89 Hz T2 = 0.52 s/f2 = 1.91 Hz


Torsional T3 = 0.37 0.97
s/f3 = 2.70 Hz

T3 = 0.37 s/f3 = 2.70 Hz T3 = 0.43 s/f3 = 2.33 Hz

123
Bull Earthquake Eng (2016) 14:2543–2563 2557

8080/jetspeed/portal/). A detailed description of the record selection as well as the


implemented earthquake recordings can be found in Karapetrou et al. (2016).
The initial models are based on the design and construction plans and reflect the as-built
state of the buildings, whereas the updated models have been derived using monitoring
data and represent the real structure as it is nowadays. In Fig. 12 an example comparative
plot of building-specific fragility curves is presented in terms of peak ground acceleration
(PGA) for the initial and updated finite element models of the monitored hospital building
units in the AHEPA complex (Karapetrou et al. 2016). PGA is selected as intensity
measure due to the fact that the derived curves are incorporated in an operational tool for
Earthquake Early Warning and rapid post-event damage assessment that will be used by
the civil protection authority of the hospital. In this context, PGA is considered appropriate
due to its simple computation in real-time and its efficient use by the authorities. However,
accordingly to the chosen strategy adopted in the onsite early warning systems which is
described continuously in the following Sect. 5 of the present paper, additional new curves
are being calculated linking the peak ground velocity (PGV) to the fragility of the struc-
ture. It is seen that the up-to-date curves present a shift to the left in comparison to the
initial ones for both damage states considered (IO: Immediate Occupancy, CP: Collapse
Prevention), indicating an increase in the structures’ vulnerability. Since this difference in
the fragility between the initial and updated models is not attributed to geometrical
modifications and as no significant damages have been reported for the specific building
during past earthquake events, the structural deterioration may be related to potential
degradation of the structural materials over time. More information regarding the updating
methodology and the seismic vulnerability assessment of the hospital units using the
ambient noise measurements can be found in Karapetrou et al. (2016).

5 EEW and rapid damage assessment

The building-specific research described in the previous sections, leading to the updated
fragility curves of Fig. 12, is integrated into two independent EEW systems and rapid
damage assessment approaches that are currently being tested in real-time in Thessaloniki.
The first approach includes the PRESTo software (PRobabilistic and Evolutionary early
warning SysTem, http://www.prestoews.org) (e.g. Satriano et al. 2008; Lancieri and Zollo
2008; Zollo et al. 2010; Satriano et al. 2011), which covers both a regional and an on-site

Fig. 12 Comparative plot of the ‘‘building-specific’’ fragility curves derived for the initial and updated
models of UNIT 1 and UNIT 2 (adopted damage states: IO immediate occupancy, CP collapse prevention)

123
2558 Bull Earthquake Eng (2016) 14:2543–2563

approach to EEW. The second approach considers an additional onsite EEW algorithm,
namely SOSEWIN algorithm, implemented on stations of the permanent SOSEWIN array
(Fig. 2). Both EEW systems are combined with building-specific probabilistic damage
assessment procedures, to provide the expected level of damage after strong ground
shaking at the monitored building.
PRESTo, as currently implemented in Thessaloniki, is based on the real-time data of the
SOSEWIN array, 17 permanent accelerometric stations of SDGEE-AUTH located in and
around Thessaloniki (e.g. Roumelioti et al. 2015) and 6 more permanent accelerometric
stations of the Institute of Geodynamics of the National Observatory of Athens (http://
www.gein.noa.gr) located in the broader Thessaloniki area (Fig. 13). This EEW system has
been in testing operation since June 2004 and has already detected, quite successfully, tens
of small-magnitude (M \ 3) events, mostly from the area of the EUROSEISTEST array
(Fig. 13). However, no earthquakes of considerable magnitude have occurred during the so
far real-time operation of the system. PRESTo uses its fast estimations of the location and
magnitude of an ongoing earthquake in combination with empirical ground motion pre-
diction equations to estimate the peak amplitude of imminent ground motion at sites further
away from the earthquake epicenter. In the case of the AHEPA hospital, the Peak Ground
Acceleration (PGA) predicted for the location of the building is used to trigger an algo-
rithm for building-specific probabilistic damage assessment. More specifically, the PGA
provided by PRESTo is used as input to the following relation (Kappos et al. 2006) that is
being used to assess the conditional probability, P, of being in or exceeding certain damage
state dsi (in our case the pre-defined levels of Immediate Occupancy, IO and Collapse
Prevention, CP):
    
dsi 1 PGA
P ds  ¼U ln ð4Þ
PGA bdsi  dsi
PGA;

where PGA;  dsi is the median PGA value at which the building reaches the damage state
dsi , bdsi is the standard deviation of ln (PG) and U the standard normal cumulative
 dsi and bdsi define the building-specific vul-
distribution function. The quantities PGA;
nerability function that has been pre-computed as described in the previous section. The
methodology has not been tested in real-time yet, but playback tests using the data from the
1978 earthquake (the only event that has caused some damage to the studied AHEPA

Fig. 13 Real-time stations


providing data to the PRESTo
EEW software currently under
testing in Thessaloniki

123
Bull Earthquake Eng (2016) 14:2543–2563 2559

building in the past), led to successful predictions of the observed level of damage
(Roumelioti et al. 2015).
An independent procedure for EEW and rapid damage assessment is fully implemented
in each station of the permanent SOSEWIN array (Parolai et al. 2015; Bindi et al. 2015b)
and is also under real-time testing in Thessaloniki. The main steps of the pertinent
SOSEWIN algorithm, labeled from A to G, are sketched in Fig. 14. The first step (A) in-
cludes the filtering of the data streams using a 4th order recursive Butterworth band-pass
filter (typically in the range 0.075–25 Hz). In the subsequent step (B), an event detection
algorithm based on short-term average/long-term average (STA/LTA) is applied. When the
ratio exceeds a predefined threshold, a trigger condition is declared. Once triggered, the
node computes several parameters over a window with pre-defined duration (typically 3 s)
(C in Fig. 14). The most relevant parameters are the maximum acceleration, velocity (D in
Fig. 14), displacement (E in Fig. 14) and the predominant period. If selected, the
parameters measured over the early P-wave arrivals are used to predict some parameters
for the S-wave, by using empirical relations (F in Fig. 14). For example, the PGV for
S-waves can be predicted from the Peak Ground Displacement (PGD) over P-wave using
the Zollo et al. (2010) empirical relationship. If the fragility curves, computed considering
the PGV as demand parameter,are uploaded in the system, then the probability of damages
can be predicted in early-warning time. After the trigger, the algorithm produces messages
at regular time (e.g. 1 s) including the peak ground motion parameters and the probability
of damages accordingly to the fragility curves uploaded in the system for the specific
monitored building, considered the PGA as demand parameter (G in Fig. 14). As no strong

Fig. 14 On-site algorithm, namely SOSEWIN algorithm, for decentralized damage assessment. Labels
from A to G are described in the text

123
2560 Bull Earthquake Eng (2016) 14:2543–2563

earthquake has occurred since the installation of the permanent SOSEWIN array, the
system remains under continuous testing. More information about the implemented pro-
cedure can be found in Parolai et al. (2015), Bindi et al. (2015b), (2016, same Special
issue).

6 Discussion and conclusions

In the present paper, the dynamic response of an eight-story RC building, belonging to the
AHEPA general hospital complex in Thessaloniki (Greece), has been investigated and
evaluated using short- and long-term ambient noise measurements. The target building was
selected within the framework of European funded REAKT project in order to implement a
real-time permanent monitoring system to evaluate its risk for various earthquake scenarios
and generate real-time risk estimates. Short-term ambient vibration measurements were
used to perform operational modal analysis and to extract the natural frequencies and mode
shapes for the two adjacent buildings first separately, and then for the entire building
analyzed as one single structure taking into account the interaction of the two building
units. Interferometric analyses were also conducted in order to investigate the shear wave
propagation in the building. Ambient vibration measurements in combination with inter-
ferometry analysis of the wave propagation from the same ambient noise recording within
the building skeleton can be used to yield more reliable models with respect to their real
condition. Furthermore, long-term noise measurements were recorded to observe possible
daily oscillation of the building resonance frequencies analyzing ambient noise data
recorded on different monitoring arrays. Further analyses are required however to quan-
titatively correlate the fundamental frequency variations to potential seasonal and daily
thermal effects induced by the operational conditions of the hospital and with changes in
the atmospheric signals.
The modal identification results were used to derive building-specific fragility curves
that reflect the actual structural state of the two building units in terms of age and building-
specific characteristics. The applied updating procedure is a critical issue as the aim is to
capture the measured response representing at the same time realistically the actual
structural state. For this reason the use of different sources of information (structural design
drawings, visual inspections, geometric survey, and non-destructive tests) is fundamental
for the realistic calibration of the finite element model of the building. In the present
research a manual updating scheme is applied considering only a limited number of
parameters allowing thus a good observation of the process in order to gain complete
insight on the effects of the sensitivity parameters on the structural behavior. A more
advanced updating methodology however could be further developed allowing an
automation of the procedure investigating combinations of several sensitivity parameters
applying different optimization criteria. The methodology proposed for the seismic vul-
nerability assessment of RC buildings using the field monitoring data, should be extended
for ‘‘real-time’’ risk assessment and post-seismic fragility updating. As an example, by
exploiting the computing power of the strong motion sensing units installed in AHEPA, the
updated building-specific fragility curves can be incorporated in a building-customized
alerting procedure suitable for performing an automatic building tagging (Bindi et al.
2015b; Parolai et al. 2015). In this context, the use of field monitoring data will contribute
in reducing the uncertainties associated with the risk assessment procedure improving

123
Bull Earthquake Eng (2016) 14:2543–2563 2561

seismic safety and allowing the development of robust real time assessment tools and
appropriate risk mitigation strategies.
The building-specific fragility curves have been integrated into two independent EEW
and rapid damage assessment approaches that have been recently established for the
Thessaloniki area and are currently under testing. Thorough validation and testing of the
methodologies and algorithms that were described in the present paper require a consid-
erable amount of data. Such data can only be acquired by long-running networks. Our
intention is to continue the effort to produce EEW and probabilistic damage assessment for
selected buildings in Thessaloniki. Future research orientation includes the increase of
monitoring instruments density at the AHEPA building, the expansion of the regional
strong motion network by adding new stations at the periphery of the city and the
instrumentation of more buildings, especially inside the campus of the Aristotle University
of Thessaloniki. Local experiments with broadband instruments toward studying in more
detail the characteristics of the monitored AHEPA building, the geotechnical properties of
its foundation soil and the soil-foundation—building interaction have already taken place
and pertinent results are expected to be published soon.
Overall the present study, conducted in the framework of research project REAKT,
highlights the use of field monitoring data in reducing the uncertainties associated with the
risk assessment procedure improving seismic safety and allowing the development of
robust real time assessment tools and appropriate risk mitigation strategies.

Acknowledgments The work reported in this paper was carried out in the framework of the REAKT (http://
www.reaktproject.eu/) project, funded by the European Commission, FP7-282862. The seismic instruments
used for the noise measurements performed in AHEPA have been provided by the Geophysical Instrument
Pool Potsdam (GIPP) and the Aristotle University of Thessaloniki (SDGEE-AUTH). The temporary
instrumentation array installed on February 2013 was possible thanks to the enthusiastic collaboration of the
PhD students and part of the permanent staff from SDGEE-AUTH. The authors are also very grateful to
Claus Milkereit and Regina Milkereit for their help in the installation of the permanent SOSEWIN array.
The staff of AHEPA hospital is acknowledged for their kind support during the experiment. Finally the
authors would like to thank the reviewers for their comments and the guest editors for the special issue.

References
Allemang RJ, Brown DL (1982) A correlation coefficient for modal vector analysis. In: Proceedings of the
1st international modal analysis conference, vol 1, pp 110–116. SEM, Orlando
Anastasiadis A, Raptakis D, Pitilakis K (2001) Thessaloniki’s detailed microzoning: subsurface structure as
basis for site response analysis. Pure Appl Geophys 158:2597–2633
Baker JW, Cornell CA (2005) A vector-valued ground motion intensity measure considering of spectral
acceleration and epsilon, earthquake engineering & structural dynamics. Stanford University, USA
Banks HT, Inman DJ, Leo DJ, Wang Y (1996) An experimentally validated damage detection theory in
smart structures. J Sound Vib 191(5):859–880
Bindi D, Petrovic B, Karapetrou S, Manakou M, Boxberger T, Raptakis D, Pitilakis KD, Parolai S (2015a)
Seismic response of an 8-story RC-building from ambient vibration analysis. Bull Earthq Eng
13(7):2095–2120
Bindi D, Boxberger T, Orunbaev S, Pilz M, Stankiewicz J, Pittore M, Iervolino I, Ellguth E, Parolai S
(2015b) On-site early-warning system for Bishkek (Kyrgyzstan). Ann Geophys 58(1):1–8
Bindi D, Iervolino I, Parolai S (2016) On-site structure-specific real-time risk assessment: perspectives from
the REAKT project. Bull Earthq Eng. doi:10.1007/s10518-016-9889-4
Brincker R, Zhang L, Andersen P (2000) Modal identification from ambient responses using frequency
domain decomposition. In: Proceedings of the 18th international modalanalysis conference, San
Antonio, Texas
CEN (2004) Eurocode 8: design of structures for earthquake resistance—part 1: general rules, seismic
actions and rules for buildings. European Committee for Standardization, Brussels

123
2562 Bull Earthquake Eng (2016) 14:2543–2563

Chopra AK (1996) Modal analysis of linear dynamic systems: physical interpretation. J Struct Eng
122:517–527
Clinton JF, Bradford SC, Heaton TH, Favela J (2006) The observed wander of the natural frequencies in a
structure. Bull Seismol Soc Am 96(1):237–257
Guéguen P, Michel C, Le Corre L (2007) A simplified approach for vulnerability assessment in moderate-to-
low seismic hazard regions: application to Grenoble (France). Bull Earthq Eng 5:467–490
Herak M, Herak D (2010) Continuous monitoring of dynamic parameters of the DGFSM building (Zagreb,
Croatia). Bull Earthq Eng 8:657–669
Hua XG, Ni YQ, Ko JM, Wong K (2007) Modelling of temperature–frequency correlation using combined
principal component analysis and support vector regression technique. J Comput Civil Eng
21(2):122–135
Huang NE, Shen Z, Long SR, Wu MC, Shih EH, Zheng Q, Tung CC, Liu HH (1998) The empirical mode
decomposition method and the Hilbert spectrum for non-stationary time series analysis. Proc R Soc
Lond A Math Phys Eng Sci 454:903–995
Ivanović SS, Trifunac MD, Todorovska MI (2001) On identification of damage in structures via wave travel
times. In: M. Erdik, M. Celebi, V. Mihailov, and N. Apaydin (eds) Proceedings of NATO advanced
research workshop on strong-motion instrumentation for civil engineering structures, June 2–5, 1999,
Istanbul, Turkey. Kluwer Academic Publishers, p 21
Karapetrou S, Manakou M, Bindi D, Petrovic B, Pitilakis K (2016) ‘‘Time-building specific’’ seismic
vulnerability assessment of a hospital RC building using field monitoring data. Eng Struct
112:114–132
Kappos AJ, Panagopoulos G, Panagiotopoulos Ch, Penelis G (2006) A hybrid method for the vulnerability
assessment of R/C and URM buildings. Bull Earthq Eng 4:391–413. doi:10.1007/s10518-006-9023-0
Lancieri M, Zollo A (2008) Bayesian approach to the real-time estimation of magnitude from the early P and
S wave displacement peaks. J Geophys Res. doi:10.1029/2007JB005
Michel C, Guéguen P, Bard PY (2008) Dynamic parameters of structures extracted from ambient vibration
measurements: an aid for the seismic vulnerability assessment of existing buildings in moderate
seismic hazard regions. Soil Dyn Earthq Eng 28(8):593–604
Michel C, Guéguen P, Lestuzzi P, Bard P-Y (2010) Comparison between seismic vulnerability models and
experimental dynamic properties of existing buildings in France’’. Bull Earthq Eng 8(6):1295–1307
Michel C, Guéguen P, Causse M (2012) Seismic vulnerability assessment to slight damage based on
experimental modal parameters. Earthq Eng Struct Dyn 41:81–98
Mikael A, Gueguen P, Bard P-Y, Roux P, Langlais M (2013) The analysis of long-term frequency and
damping wandering in buildings using random decrement technique. Bull Seismol Soc Am
103:236–246
Nakata N, Snieder R, Kuroda S, Ito S, Aizawa T, Kunimi T (2013) Monitoring a building using decon-
volution interferometry, I: earthquake-data analysis. Bull Seismol Soc Am 103(1662):1678
Nayeri RD, Masri SF, Ghanem RG, Nigbor RL (2008) A novel approach for the structural identification and
monitoring of a full-scale 17-story building based on ambient vibration measurements. Smart Mater
Struct 17(2):1–19
Pandey AK, Biswas M, Samman MM (1991) Damage detection from changes in curvature mode shapes.
J Sound Vib 145(2):321–332
Pandey AK, Biswas M (1994) Damage detection in structures using changes in flexibility. J Sound Vib
169(1):3–17
Papazachos B, Mountrakis D, Psilovikos A, Leventakis G (1979) Surface fault traces and fault plane
solutions of the May–June 1978 shocks in the Thessaloniki area, North Greece. Tectonophysics
53:171–183
Parolai S, Bindi D, Boxberger T, Milkereit C, Fleming K, Pittore M (2015) On-site early warning and rapid
damage forecasting using single stations: outcomes from the REAKT project. Seismol Res Lett
86(5):1393–1404
Picozzi M, Parolai S, Mucciarelli M, Milkereit C, Bindi D, Ditommaso R, Vona M, Gallipoli MR, Zschau J
(2011) Interferometric analysis of strong ground motion for structural health monitoring: the example
of the L’Aquila, Italy, seismic sequence of 2009. Bull Seismol Soc Am 101(2):635–651
Pitilakis K, Crowley H, Kaynia A (2014) SYNER-G: typology definition and fragility functions for physical
elements at seismic risk. Buildings, lifelines, transportation networks and critical facilities. Geotech
Geol Earthq Eng Ser. doi:10.1007/978-94-007-7872-6
Ray LR, Tian L (1999) Damage detection in smart structures through sensitivity-enhancing feedback
control. In: 1999 Symposium on smart structures and materials international society for optics and
photonics, pp 314–324

123
Bull Earthquake Eng (2016) 14:2543–2563 2563

Reynders E (2012) System identification methods for (operational) modal analysis: review and comparison.
Arch Comput Methods Eng 19:51–124
Royal Decree on the Seismic Code for Building Structures (1959) Government’s Gazette, Issue A, No. 36,
February 19, 1959, Greece (in Greek)
Roumelioti Z, Karapetrou S, Manakou M, Pitilakis K, Raptakis D, Bindi D, Boxberger T (2015) The
contribution of EUROSEISTEST and building monitoring arrays in earthquake early warning and
rapid damage assessment in Thessaloniki. In: Proceedings of 6th ICEGE, 1–4 November 2015,
Christchurch, New Zealand
Satriano C, Lomax A, Zollo A (2008) Real-time evolutionary earthquake location for seismic early warning.
Bull Seismol Soc Am 98(3):1482–1494
Satriano C, Elia L, Martino C, Lancieri M, Zollo A, Iannaccone G (2011) PRESTo, the earthquake early
warning system for Southern Italy: concepts, capabilities and future perspectives. Soil Dyn Earthq Eng
31(2):137–153
Snieder R, Safak E (2006) Extracting the building response using interferometry: theory and applications to
the Millikan library in Pasadena, California. Bull Seismol Soc Am 96:586–598
Soufleris C, Jackson JA, King GCP, Spencer CP, Scholz CH (1982) The 1978 earthquake sequence near
Thessaloniki (northern Greece). Geophys J R Astonom Soc 68:429–458
Stockwell RG, Mansinha L, Love RP (1996) Localization of the complex spectrum: the S transform. IEEE
Trans Signal Process 44:998–1001
Todorovska MI (2009) Soil-structure system identification of Millikan Library North-South response during
four earthquakes (1970–2002): what caused the observed wandering of the system frequencies? Bull
Seismol Soc Am 99:626–635. doi:10.1785/0120080333
Todorovska MI, Al Rjoub Y (2006) Effects of rainfall on soil-structure system frequency: example based on
poroelasticity and a comparison with full-scale measurements. Soil Dyn Earthq Eng 26(7–8):708–717
Todorovska MI, Trifunac MD (2008a) Earthquake damage detection in the imperial county services
building III: analysis of wave travel times via impulse response functions. Soil Dyn Earthq Eng
28(5):387–404
Todorovska MI, Trifunac MD (2008b) Impulse response analysis of the Van Nuys 7-storey hotel during 11
earthquakes and earthquake damage detection. Struct Control Health Monit 15:90–116. doi:10.1002/
stc.208
Vamvatsikos D, Cornell CA (2002) Incremental dynamic analysis. Earthq Eng Struct Dyn 31:491–514
Van Overschee P, De Moor B (1991) Subspace Algorithm for the stochastic identification problem. In:
Proceedings of the 30th IEEE conference on decision and control, Brighton, UK, December,
pp 1321–1326
Wiggins RW, Clayton RA (1976) Source shape estimation and deconvolution of teleseismic body waves.
Geophys J R Astron Soc 47:151–177
Zollo A, Amoroso O, Lancieri M, Wu Y-M, Kanamori H (2010) A threshold-based earthquake early
warning using dense accelerometer networks. Geophys J Int 183(2):963–974

123

You might also like