Mechanical Systems and Signal Processing 86 (2017) 286–307

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Operational modal analysis of a high-rise multi-function building

with dampers by a Bayesian approach
cross
⁎
Yanchun Ni, Xilin Lu , Wensheng Lu
State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, Shanghai 200092, China

A R T I C L E I N F O ABSTRACT

Keywords: The field non-destructive vibration test plays an important role in the area of structural health
Vibration test monitoring. It assists in monitoring the health status and reducing the risk caused by the poor
Modal identification performance of structures. As the most economic field test among the various vibration tests, the
Bayesian ambient vibration test is the most popular and is widely used to assess the physical condition of a
Shaking table test
structure under operational service. Based on the ambient vibration data, modal identification can
High-rise building
help provide significant previous study for model updating and damage detection during the service
life of a structure. It has been proved that modal identification works well in the investigation of the
dynamic performance of different kinds of structures. In this paper, the objective structure is a high-
rise multi-function office building. The whole building is composed of seven three-story structural
units. Each unit comprises one complete floor and two L shaped floors to form large spaces along the
vertical direction. There are 56 viscous dampers installed in the building to improve the energy
dissipation capacity. Due to the special feature of the structure, field vibration tests and further
modal identification were performed to investigate its dynamic performance. Twenty-nine setups
were designed to cover all the degrees of freedom of interest. About two years later, another field test
was carried out to measure the building for 48 h to investigate the performance variance and the
distribution of the modal parameters. A Fast Bayesian FFT method was employed to perform the
modal identification. This Bayesian method not only provides the most probable values of the modal
parameters but also assesses the associated posterior uncertainty analytically, which is especially
relevant in field vibration tests arising due to measurement noise, sensor alignment error, modelling
error, etc. A shaking table test was also implemented including cases with and without dampers,
which assists in investigating the effect of dampers. The modal parameters obtained from different
tests were investigated separately and then compared with each other.

1. Introduction

In the field of structural health monitoring, field non-destructive vibration tests and modal identification are important issues and
have received increasing attention in recent decades to investigate the structural performance and monitor the health status of
constructed structures for the purpose of reducing the risk caused by poor performance, especially for some significant buildings, such as
lifeline involved structures [1–5]. The happening of this phenomenon is due to the higher requirement of new structures than before in
both functional and height aspects. For the new style of structures, there is less experience about their performance. Modal identification
provides an effective manner to obtain the structural performance in reality. It is also an important procedure for the study of structural
control, condition assessment, damage detection, model updating, etc. [6–16]. The development has improved significantly through the

⁎
Corresponding author.
E-mail address: [email protected] (X. Lu).

http://dx.doi.org/10.1016/j.ymssp.2016.10.009
Received 21 April 2016; Received in revised form 17 September 2016; Accepted 8 October 2016
0888-3270/ © 2016 Published by Elsevier Ltd.
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

development of advanced modern computing capabilities and highly precise equipment [17]. The performance of diverse structures has
been investigated by modal identification in the last few decades, such as pre-stressed concrete bridges, footbridges, railway bridges,
highway bridge, mid-rise buildings, frame structures, beam structures, high-rise buildings, super high-rise buildings, gymnasiums,
reinforced concrete floors, etc. [6,8,18–24].
The vibration of the structure is mainly induced by the excitation of the surroundings or some special exciter correspondingly [25].
There are mainly three types of field vibration tests for carrying out and collecting vibration data for modal identification usually, including
forced vibration test, free vibration test, and ambient vibration test. These three kinds of tests do not compete with each other and in fact
can be seen as complementary with different implications for each other regarding both costs and benefits. For the ambient vibration test,
which is the most economic one, no artificial excitation is required and it can be performed with the structure under normal service. Thus it
is preferred in many situations to show the performance of a structure under low levels of environmental excitation and this kind of field
test causes no interference with the structure's operation. Based on the vibration data collected, modal identification can be performed to
obtain the values of the modal parameters, mainly including the natural frequencies, damping ratios and mode shapes [8,26–30]. Besides
the traditional identification of the most probable values of the modal parameters, the estimation of the accuracy of identified results is
becoming more and more important, which could assist to look into the field of modal identification [18,31,32]. Based on the identified
values together with the estimation variable of the accuracy, the distribution of the modal parameters can be obtained [33]. These
improvements are especially important for damage detection and structural health monitoring in the future step. A Fast Bayesian FFT
method has been developed recently to perform the ambient modal identification based on the collected acceleration response [27,34–37].
This method not only provides the most probable values of the modal parameters but also assesses the associated posterior uncertainty
analytically, which is especially relevant in field vibration tests arising due to measurement noise, sensor alignment error, modelling error,
etc.. It has been applied in many field structures, including cable-stayed bridges [38], pedestrian bridges [39], special configuration
structures [40], TV towers [41], super tall buildings [42,43], primary-secondary structures [44], coupled floor slab systems, etc. [23].
The structure investigated in this paper is a high-rise multi-function office building, which is a 22-storey building with 21 storeys
above the ground and a basement. The configuration of the building comprises seven structural units, and each unit covers three
storeys with a total of 56 viscous dampers installed. The L shaped floors are designed to make the whole building form some large
spaces for public usage and meet the architectural demand. Field vibration tests were performed based on the measured
accelerations on the target building. Multiple setups were used to cover all the degrees of freedom of interest, including two
staircases from 2/F to 21/F and two flats. About two year later, another field test was performed on the 21st floor of this building.
Forty-eight hours of data were collected and Bayesian modal identification was also performed. Besides the two groups of field tests,
a shaking table test was also implemented before the field test including the case with and without dampers. The Fast Bayesian FFT
method aforementioned was employed for the identification of the modal parameters and the associated posterior uncertainty. On
the basis of these data, several investigations have been carried out. At first, the effects of damper on the damping ratios of an
irregular tall building were studied. Secondly, the feasibility changing reference design in ambient vibration test is investigated to
obtain the global mode shapes of the whole structure by limited number of sensors. Thirdly, the distribution of modal parameters of
this building in the future is predicted using long-term data based on a probabilistic model. Fourthly, The modal parameters with
and without damper are investigated based on shaking table test data. Finally, comparison of modal parameters between field test
and shaking table test are performed to provide valuable information for the future design of shaking table test.

Fig. 1. MF building.

287
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

This paper is organized as follows. Section 2 gives a description of the targeted building. Section 3 presents three different tests related
to the structure. The first is a detailed field test covering two staircases and two floors, while the second field test measures a long term
performance of 48 h of the structure. The last one is the shaking table test performed before the field test. In Section 4, the Fast Bayesian
FFT method for ambient modal identification is introduced briefly. Section 5 presents the analyzed results obtained from different tests
and comparison between these tests is also presented and discussed. Section 6 summarizes this work and draws conclusions.

2. Description of the MF building

The test structure is a high-rise building situated in the campus of Tongji University to meet the demand for office and multi-
function facilities and was built in 2005. It has 22 storeys with 21 storeys above the ground and a basement. The weight is about
50,000 t. Fig. 1(a) shows the overview of the structure, whose dimension is 49m × 49m × 100m . The 21 storeys are separated into
seven three-storey structural units. Each unit is comprised of one complete floor and two L shaped floors with the main dimension of
16.2 m×16.2 m. The seven units rotate clockwise along the center of the building with 90° different from each other. The L shaped

Fig. 2. Damper installed.

Fig. 3. Floor plan on the 16/F.

288
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

floors form large spaces along the vertical direction, which can be seen clearly in Fig. 1(b). The floors with large space are used as
conference rooms, multi-function rooms and public space for rest.
The structural system is a complicated steel frame with bracing system made of concrete filled steel square tubular columns and steel
beams with profiled steel concrete composite slabs. If the reader refers to Fig. 2, it is found that viscous dampers are used as support
around the main steel square tubular frame to form a new system. Fifty-six dampers are installed on the four outside shear frames on the
4th to 9th and 16th to 21st floors, with each damper group covering three floors. The dampers are used to reduce the vibration response
during an earthquake and enhance the seismic performance by improved energy dissipation capacity of the whole system. Another feather
of this building is the special construction configuration, which is code-exceeding along the vertical direction.
Due to the viscous dampers used and the special configurations of the structure itself, more attention should be paid to
investigating the effect of dampers and the structural performance of the structure. For these purposes, vibration tests and further
modal identification are necessary to obtain the first-hand real response in order to know more about the dynamic characteristics.

3. Field vibration test and shaking table test

In this section, the field test on the staircases and two floors will be presented first, followed by the field test covering 48 h, and
the last part is the shaking table test.

3.1. Field vibration test on staircases and floors

The aim was to measure two staircases through the two sides of the building and two flats (4/F and 16/F), which function as
public spaces for rest and communication. At the time of measurement, only four uniaxial sensors were available. Due to the limited
number of sensors, 29 setups were designed to cover all the degrees of freedom (DOFs) of interest. The first nine setups were
performed for Staircase 1 covering floors 2, 4, 6, 8, 11, 14, 16, 17, 19 and 21, while setups 10–18 covered the corresponding
locations in Staircase 2. Setups 19–23 were designed to cover the flat on 16/F, and the last six setups covered the flat on the 4/F. For
each flat, four corners were measured to represent the performance of that floor. Each location was measured bi-axially.
The alignment of sensor location is significant for the quality of the modal identification results through the mode shapes.
Inaccurate alignment will cause modelling error which cannot be reduced through the consequent analysis. Besides that, another
important issue is the determination of the sensor orientations. Since it is difficult to align the locations at different floors visually,
the wall of the building should be used, which is common in field measurement, since the surroundings are usually more complicated
than the laboratory and many decisions should be made in situ. The same distances were kept away from the sensor locations to the
wall nearby and the floor plan in Fig. 3 can help to determine the direction of the whole measurement. The identified results obtained

Table 1
Setup plan.

Setup/channel 1 2 3 4

1 19011 19012 21011 21012

2 19011 19012 17011 17012
3 14011 14012 17011 17012
4 14011 14012 16011 16012
5 14011 14012 11011 11012
6 08011 08012 11011 11012
7 04011 04012 08011 08012
8 04011 04012 02011 02012
9 04011 04012 06011 06012
10 19021 19022 21021 21022
11 19021 19022 17021 17022
12 14021 14022 17021 17022
13 14021 14022 16021 16022
14 14021 14022 11021 11022
15 08021 08022 11021 11022
16 04021 04022 08021 08022
17 04021 04022 06021 06022
18 04021 04022 02021 02022
19 16021 16022 162021 162022
20 16021 16022 163021 163022
21 16021 16022 164021 164022
22 16021 16022 165001 165002
23 16011 16012 165001 165002
24 04021 04022 401021 401022
25 04021 04022 400021 400022
26 04021 04022 404001 404002
27 04011 04012 404001 404002
28 04011 04012 406011 406012
29 04011 04012 408011 408012

289
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

in the following sections of the paper show that the method used was satisfactory for field vibration testing.
In the whole measurement plan, it was difficult to use only one fixed reference location providing common information for so many
storeys and two sides of the building due to the limitation of cable length. Thus in the transition process of different setups, the reference
locations were also changed accordingly. Although it increased the difficulty of the modal identification method, fortunately the method
used in this study could solve this problem well. With common information provided by reference sensors, a global mode shape covering all
the measured DOFs could be obtained. In the selection of reference locations, one key issue was to make them collect as much information
about different modes as possible; otherwise there may not be enough information to assemble the mode shapes of some modes. In this
measurement, locations 19, 14, 11 and 4 were chosen to be the reference locations in different setups to connect the whole stair from the 4/
F to 21/F. The other two references on the 16/F and 4/F were used to connect the other measurement locations of flats and in two different
staircases. The whole setup plan can be found in Table 1. In the name of locations, the first two numbers represent the floor number, while
the third and fourth numbers represent the location on that floor, and the last is for direction (1 represents X; 2 represents Y).
Four piezoelectric uniaxial accelerometers Lance LC0132T (Fig. 4) were used with a measurement frequency range of 0.05–
500 Hz and sensitivity approximated to be 50 V/g. The specifications of the sensor limited the measurements from −0.1 to 0.1 g. The
National Instrument (NI) data acquisition system was used to collect the acceleration response, and it could provide four channels as
shown in Fig. 5. The sampling frequency was set to be 2048 Hz, although for analysis purposes the data were later decimated by 32
to a sampling rate of 64 Hz. In each setup, 20 min of data were acquired. Before the measurement, all the sensors were put together
for the purpose of calibration. It took two whole days to finish the measurement.

3.2. Field test on one typical floor covering 48 h

About two years later, to continue monitoring this building, a field test was performed on the 21st floor. Two locations at the
corners of the building inside a weak current equipment room were measured as shown in Fig. 6. To reduce the noise from human
disturbance, the measurement was performed during a holiday. The same data acquisition system was used in the measurement as
the previous one. The sampling frequency was set to be 2048 Hz and later decimated by 32 to be 64 Hz for analysis.

Fig. 4. Accelerometer.

Fig. 5. NI data acquisition system.

290
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

Fig. 6. Measurement location for 48 h.

3.3. Shaking table test

Shaking table test is a well-known tool to investigate the seismic performance of structures, especially for newly designed special
structures [45–47]. According to the China codes for seismic design of high-rise steel and concrete structures, this structure is a
special irregular structure in both horizontal and vertical directions. Therefore, besides the field tests, a shaking table test was
implemented before the construction of the multi-function building to investigate the dynamic and seismic performance of the
structure. Since dampers were employed to enhance the energy dissipating performance of the whole structure, the shaking table test
was performed before and after the installation of the dampers respectively to investigate the effect of the damper. The results could
also be utilized for comparison with the identification results in the full-scale field vibration tests.
The shaking table used in Tongji University was imported from the US in 1983 with a dimension of 4 m by 4 m. The maximum weight
of the structural model should be less or equal to 25 t. It can provide simple harmonic vibration, impact, and earthquake excitation in X,
Y, and Z directions with a frequency range of 0.1–50 Hz. A 1/15 scale model was designed to carry out the shaking table test as shown in
Fig. 7. The total height was 6.7 m, including a base of 0.3 m and a main structure of 6.4 m. The total weight of the model was 19 t. After
the main structure was constructed, the dampers were installed and connected with hinge structures. Sixty-seven sensors were installed
on the model to monitor the structural response, including 35 accelerometers, eight displacement meters and 24 strain gauges. The sensor

Fig. 7. Shaking table test model.

291
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

instrumentation can be found in reference [48]. The following seismic waves were taken as excitations in the shaking table test, including
EI Centro wave (1940, USA), Pasadena wave (1952, USA) and Shanghai artificial wave (SHW2). Before and after the input of the seismic
wave, white noise excitations were applied to measure the structural response and identify the modal parameters mainly including the
natural frequencies, damping ratio, and mode shapes to track their values during different damage stages.

4. Fast Bayesian FFT Approach

A fast Bayesian FFT modal identification method was used to investigate the property of the structure. The main theory is
described briefly in the following [27].
The modal parameter set needs to be identified and is denoted by θ and the value will be identified from the measured
acceleration denoted by D at n measured DOFs, i.e., D = {xˆ ̈j ∈ R n : j = 1, ... , N}, in which N is the number of samples measured per
channel. Under the classical damping assumption, the measured acceleration is expressed as

xˆ ̈j = ẍj + εj = ∑ Φiηï (t j ) + εj
i (1)

where εj is the prediction error and the sum is all the contributing modes; Φi is the i-th mode shape confined to the measured DOFs;
ηi(t j ) is the i-th modal response satisfying the general uncoupled modal equation

ηï (t ) + 2ζiωiηi̇ (t ) + ωi2ηi(t ) = pi (t ) (2)

and ωi = 2πfi ; fi , ζi and pi are the natural frequency, damping ratio, and modal force of the i-th mode, respectively. The parameter set
θ mainly consists of the natural frequency, damping ratio, mode shape of the modes of interest, the power spectral density (PSD)
spectra of modal force and PSD of prediction error. In the Bayesian method, the data used are all the FFT values calculated based on
the time domain data collected, i.e., D = {Fk}, where the definition is
N
2Δt (k − 1)(j − 1)
Fk = ∑ xˆ ̈jexp[ − 2π i ] (k = 1, ... , N )
N j =1
N (3)
2
with i = − 1 and Δt the sampling interval; the FFT components correspond to frequency fk = (k − 1)/ NΔt at k = 2, 3, ... , Nq , where Nq
is the frequency index at the Nyquist frequency and equal to the integer part of N /2 + 1. In reality, only the FFT data Fk within a
selected resonance band containing the modes of interest, single mode or closely spaced modes based on different cases, are used for
-4
10

-5
10
[g/sqrt(Hz)]

-6
10

-7
10
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)

Fig. 8. Power spectral density spectra.

Table 2
Identification results of all setups in the field tests.

Mode MPV Uncertainty Posterior c.o.v.

f (Hz) ζ (%) S (1012g2/Hz ) Se (1012g2/Hz ) f (%) ζ (%) S (%) Se (%)

1 0.430 2.2 1.6 0.8 0.7 18.5 11.6 4.7

2 0.455 2.8 4.6 0.8 0.3 15.4 11.7 4.7
3 1.310 2.0 0.1 2.1 0.3 14.8 15.3 3.7
4 1.458 1.8 0.3 2.9 0.2 13.0 13.7 4.0
5 1.643 2.2 0.3 2.5 0.4 12.0 13.2 3.4
6 2.552 1.6 1.6 10.8 0.1 7.9 10.7 2.5
7 2.650 2.5 2.5 10.8 0.1 7.1 9.2 2.5
8 3.742 2.4 1.2 22.4 0.1 6.8 6.5 1.6
9 4.842 4.1 0.8 10.5 0.3 7.5 9.7 1.6
10 5.875 3.2 0.1 5.0 1.4 43.0 90.2 4.0

292
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

modal identification. The data used are defined by Zk = [ReFk ; ImFk ] ∈ R 2n as an augmented vector of the real and imaginary part of
Fk .
The ‘negative log-likelihood function’ (NLLF) is usually used during the optimization of the modal parameters to instead of the
original form. Some computational difficulties were met in the development process, such as the ill-conditioned minimization
process, the massively time-consuming computation due to the large number of modal parameters caused by the number of DOFs to
be identified simultaneously. To solve the problems, fast algorithms have been developed recently which allow the MPV of modal
parameters to be computed typically in only a few seconds together with the associated posterior covariance matrix. On the other
hand, the approach also develops the computation strategy for the posterior uncertainty without resorting to finite difference.
For general multiple mode cases, i.e., closely spaced modes, after complex derivation the final NLLF is in the form of

L (θ) = − nNf ln 2 + (n − m′)Nf ln Se + Se−1d + ∑ ln |detE′k| − Se−1∑ Fk*B′(Im′ − Se E′k−1)B′T Fk

k k (4)

in which Nf denotes the number of FFT ordinates in the selected frequency band; m′ denotes the dimension of the mode shape
subspace and it satisfies m′ ≤ min(n , m ), m denotes the number of the mode shapes in the selected band; Se is the PSD of prediction
error; d = ∑k (ReFkT ReFk + ImFkT ImFk ), which only relates to the measured data and could only be calculated once the data were
available; I m′ is the m′ × m′ identity matrix; B′ and α satisfy the equation Φ = B′α , where Φ is the mode shape matrix in the selected
band, B′ is a set of orthonormal ‘mode shape basis’ spanning the ‘mode shape subspace’ in its columns, α is the coordinates of each
mode shape with respect to the mode shape basis in its columns; and where

E′k = αHk αT + Se I m ′ (5)

m×m
with Hk ∈ C being the theoretical value of the spectral density matrix of the modal acceleration responses equal to

Hk = diag(hk ) S diag(h*k ) (6)

S is the PSD matrix of modal forces and diag(hk ) is a diagonal matrix with the i-th diagonal element equal to hik

hik = [(βik2 − 1) + i(2ζiβik )]−1 (7)

diag(h*k ) is the corresponding conjugate transpose and βik = fi /fk is the frequency ratio, fi is the i-th natural frequency, while fk is the
FFT frequency abscissa; ζi is the i-th damping ratio. The (i, j)-entry of Hk is
*
Hk (i , j ) = Sijhik hjk (8)

and Sij denotes the (i, j)-entry of S .

During the derivation of NLLF above, it was found that the final dimension of matrix computation reduces to m′ from the original n and
it is obviously much smaller than n . The other significant component influencing the computation process is the mode shape, which is
embedded in Eq. (4). The most probable basis minimizes the quadratic form of mode shape under orthonormal constraints. A strategy has
been developed to determine different parameters by an iterative procedure. The associated posterior uncertainties of the modal parameters
could also be assessed through the posterior covariance matrix simultaneously. It was determined to be the inverse of the Hessian matrix and
calculated analytically on the basis of Eq. (4). The value of uncertainty could be used to estimate the accuracy of the identified values of
modal parameters based on the model selected and data measured. For more details of the whole theory, please refer to [27,36,37].

5. Analysis results

In this section, the result of two field tests will be presented as the sequence of measurement, and then the results of the shaking
table test will be discussed. A comparison of different tests will also be made.

20
Posterior uncertainties (%)

15

10

0
1 2 3 4
Different parameters

Fig. 9. Effect of sampling frequency to posterior uncertainties. (For interpretation of the references to color in this figure, the reader is referred to the web version of
this article).

293
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

0.430 Hz, 2.2 %

100

100

Z (m)
Z (m) 50
50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)

100 60

40

Y (m)
Z (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 10. Modal parameters of Mode 1.

0.455 Hz, 2.8 %

100

100
Z (m)
Z (m)

50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)

100 60

40
Y (m)
Z (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 11. Modal parameters of Mode 2.

294
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

5.1. Field vibration test

After checking the stationary property of the time history collected, the PSD spectra were plotted, which are shown in Fig. 8 and
discussed next. From the typical PSD spectra shown, there are approximately 10 clear peaks corresponding to 10 different modes below
10 Hz. The peaks also show that the signal to noise ratio of the collected data was high enough. Since the plan of the building is square, it was
expected that some closely spaced modes along two horizontal directions would appear. It is seen that there are three closely spaced modes
in the frequency band of 1–2 Hz, and two modes within the frequency band of 2–3 Hz. The closely spaced modes need to be identified
within the same frequency band due to the coinciding parts of each other in the frequency domain. It should be noted that the human
walking frequency ranges from 1.5 to 4.5 Hz. This means that the modes with natural frequencies in this range may have some excitation
from pedestrian walking. The PSD spectra also provide the frequency bands and initial guesses of the target modes for identification, which
were picked manually during the identification process. The modal identification results were identified from the measured structural
acceleration time histories collected under ambient excitation setup by setup. By defining a selection matrix, the relationship between the
global mode shape and the mode shape in one setup can be established using the information collected from the reference locations. One
objection function based on a least square method was formulated theoretically. By minimizing the objective function, the optimal values of
global mode shape can be determined, which covers all the measured DOFs. For the detail of the least square method, please refer to [49].
Table 2 shows the mean values of identified modal parameters among all the 29 setups. The first ten modes are identified. The natural
frequencies identified coincide well with the resonance peaks found in the PSD spectra. It is seen that the first mode is about 0.430 Hz, which
is consistent with the structural height. Comparing the results of other structures [23,39,42], the damping ratios identified for all the modes
were relatively larger (all larger than 1.5%). One possible reason is the effects of the 56 dampers installed on the whole structure, which
improves the energy dissipation capacity of this building. From the table, it can be seen that the PSD of modal force and the PSD of
prediction error are in the order of 10−12. The low value means that the environment of the measurement is quiet. This is consistent with the
fact that at the time of the measurement, it was a holiday and there were not so many people working in the building. The mean values of the
posterior coefficients of variation (COV=posterior standard derivation/MPV) of the modal parameters are also shown in this table. It is seen
that the posterior COV values of natural frequencies are quite small (less than 1%), indicating that the identification of this quantity is
accurate. The damping ratio, however, has a relatively larger posterior COV with values from a few to a few ten percentage points. For the
two quantities that reflect the effect from environments, i.e., the PSD of modal force and PSD of prediction error, the results show that the
former is less accurate than the latter. Since the measured data has been decimated to be 64 Hz from the original 2048 Hz, a comparison of
the identified results has been listed in Fig. 9. The circle in blue represents the results of posterior uncertainty calculated with sampling
frequency of 64 Hz, while the dots in red shows those of 2048 Hz. The two groups of results are consistent with each other. This means that
once the data is enough, the sampling frequency has little influence on the posterior uncertainties of different parameters.

1.31 Hz, 2 %
100

100
Z (m)
Z (m)

50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)
100 60

40
Z (m)

Y (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 12. Modal parameters of Mode 3.

295
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

1.458 Hz, 1.8 %

100

100

Z (m)
Z (m)
50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)
100 60

40

Y (m)
Z (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 13. Modal parameters of Mode 4.

Next, the global mode shapes are presented and discussed. Figs. 10 and 11 show the first two modes at 0.430 Hz with a damping ratio
of 2.2% and 0.455 Hz with a damping ratio of 2.8%, respectively. As mentioned, the values of natural frequencies and damping ratios
reported here are the mean of all setups. It can be seen that these modes are mainly the translational modes of the high-rise building from
the YZ and XZ views. In light of the similar configuration of the X and Y directions, there were no obvious strong and weak directions
observed. For these two modes, although both stairs were excited respectively, the motions were dominated by stair 2 along the Y
direction for the first mode and approximately 30° to the X axis for the second mode. From the mode shapes, it can be seen that the
locations on the 16/F have less deformation than others due to the influence of the flat on that floor, which may increase its stiffness.
The third mode is at 1.310 Hz with a damping ratio of 2.0% as shown in Fig. 12. From the XZ and YZ views, it is clear that it is the
second mode of the structure. The deformation of the building above and below the 16/F are in the opposite direction. Stair 1 mainly
moves along the X direction, while stair 2 moves along 45° to the X axis and then undergoes deformations along both X and Y
directions. Generally speaking, in this mode, the whole building moves along the upper-right corner. As shown in Figs. 13 and 14,
the fourth and fifth modes are all around 1.5 Hz, namely 1.458 Hz and 1.643 Hz with damping ratios of about 2%. The fourth mode
is the mode of the whole structure; it can be readily seen that all the measured locations including the stairs and the flats move
together along 45° to the X axis. About the fifth mode, the two stairs are deformed along the X and Y axis respectively and the whole
structure moves to the upper-right corner.
Mode 3 to Mode 5 is an interesting group of mode shapes. Mode 3 is dominated by the deformation of stair 1 moving in the X
direction, while in Mode 4, the two stairs move together and are consistent with each other. For Mode 5, the deformation of stair 2 is
along the X direction. From the PSD spectra, these three modes are closely spaced modes. Although their natural frequencies are close, the
mode shapes are quite different. It is not necessary for the mode shapes of closely spaced modes to be orthogonal to each other.
For the sixth and seventh modes, the dominating parts are all stair 2, as shown in Figs. 15 and 16, respectively. The former is
along the X axis and the other is mainly along the Y axis but with a small angle. Although the direction is different, the deformation
of the mode shapes is similar to each other. The two flats have only a slight deflection along the same direction with the stair, i.e., X
direction in mode 6 and the same angle with the Y axis in Mode 7. Mode 6 is at 2.552 Hz and the damping ratio is 1.6%, while Mode
7 has a natural frequency of 2.650 Hz and damping ratio of 2.5%. They are also closely spaced modes, and it is shown clearly in the
PSD spectra. The similar deformation and closeness of the natural frequencies may be attributed to the square plan of this building.
The eighth mode in Fig. 17 is at 3.742 Hz with a damping ratio of 2.4%. The mode shape of this mode only moves from floor 16–21
of stair 1. The measured locations below the 16/F almost remain unchanged. This may be a local mode of the stair. The ninth and
tenth modes in Figs. 18 and 19 are all the mode of stair 2 at 4.842 Hz with a damping ratio of 4.1% and 5.875 Hz with a damping
ratio of 3.2%, respectively. The motion of stair 2 is similar for these two modes in the X direction. However, for the Y direction, there
is almost no motion for the ninth mode, but obvious deformation can be found in the tenth mode. The two flats almost remain
unchanged for both modes.

296
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

1.643 Hz, 2.2 %

100

100

Z (m)
Z (m)
50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)
100 60

40
Z (m)

Y (m)
50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 14. Modal parameters of Mode 5.

2.552 Hz, 1.6 %

100

100
Z (m)
Z (m)

50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)

100 60

40
Z (m)

Y (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 15. Modal parameters of Mode 6.

297
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

2.65 Hz, 2.5 %

100

100

Z (m)
Z (m)
50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)

100 60

40
Z (m)

Y (m)
50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 16. Modal parameters of Mode 7.

3.742 Hz, 2.4 %

100

100
Z (m)
Z (m)

50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m)
Y (m)
100 60

40
Z (m)

Y (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 17. Modal parameters of Mode 8.

298
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

4.842 Hz, 4.1 %

100

100

Z (m)
Z (m) 50
50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)
100 60

40
Z (m)

Y (m)
50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 18. Modal parameters of Mode 9.

5.875 Hz, 3.2 %

100

100
Z (m)
Z (m)

50 50

0 60
40
0 20 20 0
40 60 0
Y (m) 0 20 40 60
X (m) Y (m)
100 60

40
Z (m)

Y (m)

50
20

0
0
0 20 40 60 0 20 40 60
X (m) X (m)
Fig. 19. Modal parameters of Mode 10.

299
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

Natural frequency (Hz)

0.45
---
---
0.44 --- ---
--- --- --- ---
--- --- --- ---
-
-- -
-- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---
--- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---
0.43 -
-- -
-- --- ---
--- --- -
--

0.42
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

8
---
Damping ratio (%)

4 --- ---
--- ---
2 --- --- --- --- --- --- --- ---
--- --- --- ---
--- --- --- --- --- --- ------ --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---
--- --- --- --- ------ --- --- --- --- ------ --- ---
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
Setup
-11
x 10
Modal force PSD (g2/Hz)

1 ---

---
0.5 --- --- ---
--- --- ---
--- --- --- ---
--- --- --- ---
--- --- ---
------ --- ------ --- ---
0 --- --- --- ------ --- --- --- --- --- --- --- --- ---
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
Setup
Fig. 20. Error bar of Mode 1.

To investigate the variation of modal parameters across different setups, Figs. 20–22 show all the identified values of natural
frequencies, damping ratios, and modal force PSD corresponding to different setups for Mode 1 to Mode 3, respectively, where each
parameter is shown with a dot at the MPV and an error bar indicating ± 2 posterior standard deviations. It should be noted that there are
27 setups plotted in Mode 1 and 29 for Modes 2 and 3. This is because for Mode 1, the two setups measuring the 2/F of the building are not
included, since the signal to noise ratio was too small to identify this mode in these two setups. From these three figures, it is found that
generally the values of natural frequency vary with different setups slightly and the posterior uncertainty is small compared with its
variation across different setups. The variation of damping is a little larger but it is still stable. For the PSD of modal force, a larger variation
can be found for this quantity. This may be attributed to the fact that this parameter is used to reflect the excitation from the environment.
During the whole measurement across two days, there was some environmental change, although it was difficult for one to feel it.
For mode 3 in Fig. 22, it is seen that in setups 19 and 20, the posterior uncertainties are much larger compared with the other setups.
Checking the PSD of data in setup 20 as shown in Fig. 23, it is found that it is difficult to find a peak for the mode near 1.3 Hz, and the
signal noise ratio is too low to identify accurate modal parameters. This also demonstrates that the posterior uncertainty is very useful to
reveal the accuracy of the identified modal parameters. A similar phenomenon arose in other modes. Generally speaking, the identified
values among different setups are consistent with each other and reflect the accuracy of that setup simultaneously.

5.2. Field test for 48 h measurement

For the target building, 48-h data were collected for modal identification. In the last subsection, it was found that the natural
frequency of the first mode was around 0.45 Hz. Commonly, at least 600 cycles of data need to be used for modal identification under
ambient vibration conditions to obtain accurate identification results. However, if the data are too long, the stationary assumption
for the loading cannot be satisfied well. Taking into account the balance between the above two factors, the 48 h of data were
separated into 96 data sets with half an hour for each data set. Every data set underwent modal identification by the Fast Bayesian
FFT method individually.
Based on the results obtained in different data sets, a probabilistic model could be obtained to predict the distribution of modal
parameters in the future under a similar vibration environment [33]. In this model, the PDF of modal parameters can be given by:

1
Ns
1 ⎡ 1 ⎤
pΘ (θ|D ) = ∑ exp⎢ − (θ − θˆi )2 ⎥
Ns i =1 2πcˆi ⎣ 2cˆi ⎦ (9)

300
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

Natural frequency (Hz)

0.5 ---
---

0.48
--- --- --- ---
--- ---
0.46 --- --- --- --- --- --- --- ---
--- --- --- ---
--- --- --- --- --- --- ---
--- ---
--- --- --- ---
--- --- --- --- --- --- ------ --- --- --- --- --- ---
--- --- --- --- ---
--- --- ---
0.44 --- --- ---
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

8
Damping ratio (%)

6 ---
--- --- ---
--- --- --- --- ---
--- --- --- --- --- ---
4 --- ---
--- --- --- ---
--- --- --- --- --- --- --- --- --- ---
---
2 --- --- --- --- --- --- --- ---
--- --- --- --- --- --- ---
--- --- --- --- --- --- --- ---
--- --- ---
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Setup
-11
x 10
Modal force PSD (g2/Hz)

4
---
3
---
--- ---
2
---
--- --- ---
1 ---
--- --- --- --- ---
--- --- ------
--- --- --- --- --- --- ------ --- --- --- --- --- --- --- --- --- ---
0 --- --- ---
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Setup
Fig. 21. Error bar of Mode 2.

in which both the identified MPV and the associated posterior uncertainty of every modal parameter were used to develop the
probabilistic model, and Ns is the number of sets of data used; all the data sets are denoted by D = {D1, ... , D Ns}; the posterior MPV of
θ over different data sets is {θˆi: i = 1, ... , Ns}; the corresponding posterior variances are {cˆi : i = 1, ... , Ns}.
The mean {θˆ } and standard derivation {cˆ } of the distribution p (θ|D ) could be calculated based on the following two equations:
i i Θ
Ns
1 Ns
mean = ∑ θˆi = sample mean of{θˆi}i =1
Ns i =1 (10)

⎧1
⎪
Ns
1
Ns
2 1
Ns ⎫1/2
⎪ N 1/2
std = ⎨ ∑ cˆi + ∑ θˆi −( ∑ θˆi )2⎬ s + sample variance of θˆ s
= {sample mean of{cˆi}iN=1 { i}i =1}
⎩ Ns ⎭
⎪ ⎪

i =1
Ns i =1
Ns i =1 (11)
It is readily found from the above equation that the theoretical model accounts for both the Bayesian and frequentist variability of
the modal parameters among different data sets and Eq. (9) is a mixture of Gaussian PDFs, which is not necessary to be Gaussian. One
advantage of utilizing the predicted PDF of modal properties is to investigate whether the variation of the modal parameter in the
future tests is in a reasonable range or not based on the current health status of modal parameters and their PDF. This provides a novel
index for observing the change of modal parameters instead of just using one optimal value, which has the potential to be used in
decision making of damage detection.
Figs. 24–26 show the PDF of the modal parameters of Modes 1–3 based on the above probabilistic model. It is seen from the
figures that the distribution of natural frequency and damping ratio appear to be approximately Gaussian. However, this is not true
for the PSD of modal force and PSD of prediction error. This may be attributed to the different physical means of these two
parameters from natural frequency and damping ratio. The PSD of modal force and PSD of prediction error correspond to the
environment excitation and noise disturbance, respectively. Thus it is normal for these two parameters to be more sensitive to the
environment. However, natural frequency and damping ratios, which are both the properties of the structure, should be more stable
than the other two parameters. This means that the predicted distribution of the natural frequency and damping ratios should be
more accurate than the PSD of modal force and PSD of prediction error. This is because that the complicated operational
environment will influence the prediction of these two parameters significantly due to their physical meanings and basic
assumptions. Thus, predicted distribution did not necessarily agree well with those obtained from the real case.
The mean and c.o.v. of the modal parameter distribution are all shown in the title of each subfigure. For Mode 1, the mean natural
frequency of 96 data sets is 0.432 Hz, and the c.o.v. is 0.46%. For the damping ratio, the MPV is 1.03% with the c.o.v. equal to 25%.

301
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

Natural frequency (Hz)

1.35
------ ------ --- --- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ --- --- --- --- ------ --- ---
1.3 --- --- --- ------ --- --- --- --- ------ --- ---
---
1.25

1.2 ---
---
1.15
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

30
Damping ratio (%)

20 ---

---
10

------ ------ ------ ------ ------ ------ ------ ------ ------ --- ------ --- --- --- ------ --- --- ------ --- --- ------ ------ ------ ------ ------ --- --- ------ ---
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Setup
-13
x 10
Modal force PSD (g2/Hz)

6 ---
--- ---
---
4 ---
--- --- --- ---
--- ---
2 --- --- --- --- --- ---
---
---
------ --- --- ----- ------ --- ------
---
------ --- --- ---
------ --- ---
--- ---
--- - ------ --- --- --- --- ---
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Setup
Fig. 22. Error bar of Mode 3.

-4
10

-5
10
[g/sqrt(Hz)]

-6
10

-7
10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Frequency (Hz)

Fig. 23. PSD of setup 20.

Compared with the results in Table 2, it is found that the MPV of the first three modes are all changed slightly with the largest difference
being 0.01 Hz at Mode 2. It should be noted that the 48 h of data were collected approximately two years after the field test for the staircases
and two floors. During these two years, the modal parameters may have undergone a little change which may cause the slight difference of
the MPV of natural frequencies in the first three modes. It is meaningful to compare the posterior c.o.v. identified by the modal identification
method directly based on one data set and the results calculated based on the above probabilistic model. Generally speaking, the calculated
results for the probabilistic model are bigger than the posterior uncertainty identified from previous field tests. This phenomenon fits for the
probabilistic model used in theory, which considers not only the posterior uncertainty of modal parameters from a Bayesian point of view
(the case in single field tests), but also the variability of modal parameters among different data sets from the frequentist aspect.

302
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

Mean: 0.432; c.o.v.: 0.46% Mean: 0.0103; c.o.v.: 25%

400 200

200 100

0 0
0.42 0.43 0.44 0 0.01 0.02
Natural frequency[Hz] Damping ratio[%]

Mean: 0.302; c.o.v.: 65% Mean: 1.612; c.o.v.: 200%

4 0.4

2 0.2

0 0
0 0.5 1 0 5 10
2 2
Modal force PSD[(μg) /Hz] Prediction error PSD[(μg) /Hz]
Fig. 24. PDF of the modal parameters, Mode 1.

Mean: 0.446; c.o.v.: 0.53% Mean: 0.0116; c.o.v.: 27%

400 200

200 100

0 0
0.435 0.445 0.455 0 0.01 0.02
Natural frequency[Hz] Damping ratio[%]
Mean: 0.477; c.o.v.: 75% Mean: 1.612; c.o.v.: 200%
2 0.4

1 0.2

0 0
0 1 2 0 5 10

Modal force PSD[(μg)2/Hz] Prediction error PSD[(μg)2/Hz]

Fig. 25. PDF of the modal parameters, Mode 2.

5.3. Shaking table test

In the shaking table test, recall that white noise excitation was performed before and after the installation of the dampers for the
purpose of investigating the influence of the damper on the dynamic performance of the structure. Figs. 27 and 28 show the modal
parameters obtained from the data collected before and after the damper installation for the natural frequency and damping ratio,
respectively [48]. It can be seen from the figures that the damper has less influence on the natural frequency of the structure, especially for
the first four modes, where the natural frequencies are almost the same in the two different cases. The small increase of the frequency may
be caused by the increase of the stiffness of this structure due to the installation of the damper. The main difference lies in the damping
ratio. For the fundamental mode of the structure, the effect of the damper is very clear and it enhances the damping up to 120%. The most
effective mode is the 6th mode with the damping ratios increasing up to 140%. It is also interesting to find that the damping ratios tend to

303
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

Mean: 1.31; c.o.v.: 0.39% Mean: 0.0125; c.o.v.: 24%

200 200

100 100

0 0
1.3 1.31 1.32 1.33 0 0.01 0.02
Natural frequency[Hz] Damping ratio[%]
Mean: 0.2303; c.o.v.: 48% Mean: 7.9; c.o.v.: 34%
4 0.2

2 0.1

0 0
0 0.5 1 0 10 20

Modal force PSD[(μg)2/Hz] Prediction error PSD[(μg)2/Hz]

Fig. 26. PDF of the modal parameters, Mode 3.

decrease with an increase in the mode number, indicating that the energy dispersion capability decreases from the lower to higher modes.
The results have also been compared with those in the field test shown in Table 3, where the natural frequencies obtained in the
shaking table test have been converted to the natural frequency of the prototype. It is found that the results from the field are almost
two times those in the shaking table test for all modes and there is a proportion between the two kinds of results. However, the mode
shapes of the shaking table test are consistent with the field one, which demonstrates that the shaking table could provide evidence
for the performance of the real structure. The damping ratios are not compared with each other directly since it is not necessary to be
the same due to the complexity of this quantity.

6. Conclusions

A high-rise multi-function building was investigated to determine its dynamic performance for future structural health
monitoring (SHM). An ambient vibration test with 29 setups was performed to acquire the first-hand vibration data of all the DOFs
of interest. Based on a recently developed fast Bayesian FFT modal identification method, the modal parameters were obtained for
the first ten modes below 10 Hz. Some modes were dominated by the motion of the two stairs, while some modes were the mode of
the whole structure. Due to the square plan and similar configuration along the X and Y directions of the building, closely spaced
modes were observed. In this work, although the design of the reference locations in the whole measurement is quite complicated, it
is still possible to obtain reasonable global mode shapes. Similar design is quite useful when the number of sensors is quite limited
but a larger number of locations desired to be measured. In general, the identified results from the field test are reasonable and could
represent the dynamic performance of the structure well. The posterior uncertainties of modal parameters were also calculated by
the Fast Bayesian FFT method. It was found that the posterior c.o.v. of natural frequency was less than 1% while the damping ratio

25
No damper
With damper
20
Frequency (Hz)

15

10

0
0 2 4 6 8 10 12
Mode Number

Fig. 27. Identified natural frequencies before and after the installation of dampers.

304
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

12
No damper
With damper
10

Damping ratio (%)

8

2
0 2 4 6 8 10 12
Mode Number

Fig. 28. Identified damping ratios before and after the installation of dampers.

Table 3
Comparison between field and shaking table test.

Mode number Field vibration test f (Hz) Shaking table test f (Hz) Direction

1 0.430 0.243 TX1

2 0.455 0.243 TY1
3 1.310 /
4 1.458 0.728 TX2
5 1.643 0.777 TY2
6 2.552 1.456 TX3
7 2.650 1.505 TY3
8 3.742 /
9 4.842 2.330 TX4
10 5.875 /

had a larger posterior c.o.v. with the order of magnitude of a few ten percentage points. Across different setups, the modal
parameters were found to have some variation. This is reasonable since the field test in different setups was carried out at different
times and in different locations. At the same time, uncertainty can also exist as studied before. It is possible for the modal parameters
to have undergone some change. The damping ratios of all the modes identified tend to be larger than other structures. This may be
attributed to the effects of the dampers installed on the building.
A shaking table test was also performed for the target building before construction, and the cases with and without dampers were
all investigated to study the performance of the dampers installed. The results show that the dampers can improve the damping of
the structure of most of the identified modes, which means better energy dissipating ability could be obtained. Through the
comparison of the two kinds of tests, it could be found that the natural frequencies in the field are all almost two times those in the
shaking table test, which means that the ratio of stiffness over mass in the shaking table model is smaller than in reality. However,
the consistency of the mode shape means that the results from the shaking table test could well provide a reference to the
performance of the structure after construction.
For the sake of investigating the distribution of the modal parameters to provide a prediction for the future performance
assessment of this structure, a 48-h measurement was performed and analyzed. The results show that the distribution of natural
frequency and damping ratio appear to be approximately Gaussian and it is not true for the PSD of modal force and prediction error
since they are sensitive to the environment changing. It was found that the posterior c.o.v. identified for the probabilistic model was
generally larger than those directly calculated based on one data set, i.e., in the single field test. This is because in the distribution
model of the modal parameters the uncertainty contains both the posterior uncertainty of modal parameters from a Bayesian point
of view and the variability of modal parameters among the group of data sets from the frequentist perspective. On the other hand,
since 48 h of data were collected two years after the field test, the results of modal parameters may have undergone a little change
compared with the previous ones.

Acknowledgement

The authors would like to thank the support of the National Natural Science Foundation of China through Grant 51261120377,
91315301-4, 51508413 and 51508407, Shanghai Pujiang Program (Grant No. 15PJ1408600) and Fundamental Research Funds for the
Central Universities (Grant No.: 20161143), China. Thanks are also due to Dr. Chun-Guang Meng, Prof. Bin Zhao from Tongji University
and Prof. Jie-Min Ding from Tongji Architectural Design (Group) Co., Ltd. for performing the shaking table test. The authors would like to
acknowledge Prof. Siu-Kui Au, Chair Professor at the University of Liverpool for sharing the software for modal identification.

305
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

References

[1] J.M.W. Brownjohn, P. Moyo, P. Omenzetter, S. Chakraborty, Lessons from monitoring the performance of highway bridge, Struct. Control Health Monit. 12
(2005) 227–244.
[2] P.C. Chang, A. Flatau, S.C. Liu, Health monitoring of civil infrastructure, Struct. Health Monit. 2 (3) (2003) 257–267.
[3] J.M. Ko, Y.Q. Ni, Technology developments in structural health monitoring of large-scale bridges, Eng. Struct. 27 (12) (2005) 1715–1725.
[4] Y.Q. Ni, Y. Xia, W. Lin, W.H. Chen, J.M. Ko, SHM benchmark for high-rise structures: a reduced-order finite element model and field measurement data, Smart
Struct. Syst. 10 (4) (2012) 411–426.
[5] H. Sohn, C.R. Farrar, F.M. Hemez, D.D. Shunk, D.W. Stinemates, B.R. Nadler, A review of structural health monitoring literature: 1996–2001, Los Alamos
National Laboratory Report, 2003, LA-13976-MS
[6] C.A. Perez-Ramireza, J.P. Amezquita-Sancheza, H. Adelib, M. Valtierra-Rodrigueza, D. Camarena-Martinezc, R.J. Romero-Troncoso, New methodology for
modal parameters identification of smart civil structures using ambient vibrations and synchrosqueezed wavelet transform, Eng. Appl. Artif. Intell. 48 (2016)
1–12.
[7] E.N. Chatzi, C. Fuggini, Online correction of drift in structural identification using artificial white noise observations and an unscented Kalman Filter, Smart
Struct. Syst. 16 (2) (2015) 295–328.
[8] C. Costaa, D. Ribeirob, P. Jorgec, R. Silvac, R. Calçadac, A. Arêde, Calibration of the numerical model of a short-span masonry railway bridge based on
experimental modal parameters, Procedia Eng. 114 (2015) 846–853.
[9] M. Stache, M. Guettler, S. Marburg, A precise non-destructive damage identification technique of long and slender structures based on modal data, J. Sound
Vib. 365 (2016) 89–101.
[10] M.I. Friswell, J.E. Mottershead, Inverse methods in structural health monitoring, Key Eng. Mater. 204 (2001) 201–210.
[11] A. Alvandi, C. Cremona, Assessment of vibration-based damage identification techniques, J. Sound Vib. 292 (2006) 179–202.
[12] L.S. Katafygiotis, K.V. Yuen, Bayesian spectral density approach for modal updating using ambient data, Earthq. Eng. Struct. Dyn. 30 (2001) 1103–1123.
[13] N. Dervilis, K. Worden, E.J. Cross, On robust regression analysis as a means of exploring environmental and operational conditions for SHM data, J. Sound Vib.
347 (2015) 279–296.
[14] M. Spiridonakos, E. Chatzi, B. Sudret, Polynomial chaos expansion models for the monitoring of structures under operational variability, J. Risk Uncertain. Eng.
Syst. Part A: Civ. Eng. (2016). http://dx.doi.org/10.1061/ AJRUA6.0000872 [B4016003].
[15] J. Kullaa, Distinguishing between sensor fault, structural damage, and environmental or operational effects in structural health monitoring, Mech. Syst. Signal
Process. 25 (8) (2011) 2976–2989.
[16] B. Jaishi, H.J. Kim, M.K. Kim, W.X. Ren, S.H. Lee, Finite element model updating of concrete-filled steel tubular arch bridge under operational condition using
modal flexibility, Mech. Syst. Signal Process. 21 (6) (2007) 2406–2426.
[17] S. Cho, B.F. Spencer, Sensor attitude correction of wireless sensor network for acceleration-based monitoring of civil structures, Comput.-Aided Civ. Infrastruct.
Eng. 30 (2015) 859–871.
[18] E. Reynders, K. Maes, G. Lombaert, G. De Roeck, Uncertainty quantification in operational modal analysis with stochastic subspace identification: validation
and applications, Mech. Syst. Signal Process. 66–67 (2016) 13–30.
[19] H. Li, J. Ou, X. Zhao, W. Zhou, H. Li, Z. Zhou, Y. Yang, Structural health monitoring system for the Shandong Binzhou Yellow River highway bridg, Comput.-
Aided Civ. Infrastruct. Eng. 21 (4) (2006) 306–317.
[20] F. Ubertini, C. Gentile, A.L. Materazzi, Automated modal identification in operational conditions and its application to bridges, Eng. Struct. 46 (1) (2013)
264–278.
[21] J.A. Gonilhaa, J.R. Correiaa, F.A. Brancoa, E. Caetanob, Á. Cunha, Modal identification of a GFRP-concrete hybrid footbridge prototype: experimental tests and
analytical and numerical simulations, Compos. Struct. 106 (2013) 724–733.
[22] W.X. Shi, J.Z. Shan, X.L. Lu, Modal identification of Shanghai World Financial Center both from free and ambient vibration response, Eng. Struct. 36 (2012)
14–26.
[23] S.K. Au, Y.C. Ni, F.L. Zhang, H.F. Lam, Full-scale dynamic testing and modal identification of a coupled floor slab system, Eng. Struct. 37 (2012) 167–178.
[24] J. Grosela, W. Sawicki, W. Pakos, Application of classical and operational modal analysis for examination of engineering structures, Procedia Eng. 91 (2014)
136–141.
[25] S.K. Au, Y.C. Ni, Fast Bayesian modal identification of structures using known single-input forced vibration data, Struct. Control Health Monit. 21 (3) (2014)
381–402.
[26] K.V. Yuen, S.C. Kuok, Ambient in terference in long-term monitoring of buildings, Eng. Struct. 32 (2010) 2379–2386.
[27] S.K. Au, F.L. Zhang, Y.C. Ni, Bayesian operational modal analysis: theory, computation, practice, Comput. Struct. 126 (2013) 3–15.
[28] W.J. Yan, L.S. Katafygiotis, A two-stage fast Bayesian spectral density approach for ambient modal analysis. Part I: most probable values and posterior
uncertainty, Mech. Syst. Signal Process. 54 (2015) 139–155.
[29] W.J. Yan, L.S. Katafygiotis, A two-stage fast Bayesian spectral density approach for ambient modal analysis. Part II: mode shape assembly and case studies,
Mech. Syst. Signal Process. 54 (2015) 156–171.
[30] K.V. Yuen, L.S. Katafygiotis, Bayesian Fast Fourier Transform approach for modal updating using ambient data, Adv. Struct. Eng. 6 (2) (2003) 81–95.
[31] K. Sepahvand, S. Marburg, Identification of composite uncertain material parameters from experimental modal data, Probab. Eng. Mech. 37 (2014) 148–153.
[32] R. Pintelon, P. Guillaume, J. Schoukens, Uncertainty calculation in (operational) modal analysis, Mech. Syst. Signal Process. 21 (6) (2007) 2359–2373.
[33] F.L. Zhang, S.K. Au, A probabilistic model for modal properties based on operational modal analysis, ASCE-ASME J. Risk Uncertain. Eng. Syst. Part A: Civ. Eng.
(2015). http://dx.doi.org/10.1061/AJRUA6.0000843.
[34] S.K. Au, Fast Bayesian FFT method for ambient modal identification with separated modes, J. Eng. Mech. ASCE 137 (2011) 214–226.
[35] F.L. Zhang, S.K. Au, Erratum for Fast Bayesian FFT method for ambient modal identification with separated modes by Siu-Kui Au, J. Eng. Mech. ASCE 139
(2013) (545-545).
[36] S.K. Au, Fast Bayesian ambient modal identification in the frequency domain, part I: posterior most probable value, Mech. Syst. Signal Process. 26 (2012)
60–75.
[37] S.K. Au, Fast Bayesian ambient modal identification in the frequency domain, part II: posterior uncertainty, Mech. Syst. Signal Process. 26 (2012) 76–90.
[38] F.L. Zhang, Y.Q. Ni, Y.C. Ni, Mode identifiability of a cable-stayed bridge based on a Bayesian method, Smart Struct. Syst. (2016). http://dx.doi.org/10.12989/
sss.2016.17.3.000.
[39] Y.C. Ni, F.L. Zhang, H.F. Lam, Series of full-scale field vibration tests and Bayesian modal identification of a pedestrian bridge, J. Bridge Eng. ASCE (2016).
http://dx.doi.org/10.1061/(ASCE)BE.1943- 5592.
[40] Y.C. Ni, X.L. Lu, W.S. Lu, Field dynamic test and Bayesian modal identification of a special structure-the Palms Together Dagoba, Struct. Control Health Monit.
(2016). http://dx.doi.org/10.1002/stc.1816.
[41] F.L. Zhang, Y.Q. Ni, Y.C. Ni, Y.W. Wang, Operational modal analysis of Canton Tower by a fast frequency domain Bayesian method, Smart Struct. Syst. 17 (2)
(2016) 209–230.
[42] F.L. Zhang, H.B. Xiong, W.X. Shi, X. Ou, Structural health monitoring of a super tall building during different stages using a Bayesian approach, Struct. Control
Health Monit. (2016). http://dx.doi.org/10.1002/stc.1840.
[43] S.K. Au, F.L. Zhang, P. To, Field observations on modal properties of two super tall buildings under strong wind, J. Wind Eng. Ind. Aerodyn. 101 (2012) 12–23.
[44] S.K. Au, F.L. Zhang, Ambient modal identification of a primary-secondary structure by fast Bayesian FFT method, Mech. Syst. Signal Process. 28 (2012)
280–296.
[45] X.L. Lu, Y. Zou, W.S. Lu, B. Zhao, Shaking table model test on Shanghai World Financial Center Tower, Earthq. Eng. Struct. Dyn. 36 (2007) 439–457.
[46] D. Benedetti, P. Carydis, P. Pezzoli, Shaking table tests on 24 simple masonry buildings, Earthq. Eng. Struct. Dyn. 27 (1998) 67–90.

306
Y. Ni et al. Mechanical Systems and Signal Processing 86 (2017) 286–307

[47] M. Dolce, D. Cardone, F.C. Ponzo, C. Valente, Shaking table tests on reinforced concrete frames without and with passive control systems, Earthq. Eng. Struct.
Dyn. 34 (2005) 1687–1717.
[48] C.G. Meng, J.M. Ding, X.L. Lu, W.S. Lu, B. Zhao, Shaking table experimental study of a high rise building of concrete filled square steel tubular frame structure
with dampers, Struct. Eng. 21 (5) (2005) 57–62.
[49] S.K. Au, Assembling mode shapes by least squares, Mech. Syst. Signal Process. 25 (1) (2011) 163–179.

307