You are on page 1of 15

SHM-Based Seismic Performance Assessment of

High-Rise Buildings under Long-Period Ground Motion


Rong-Pan Hu 1 and You-Lin Xu, F.ASCE 2

Abstract: Long-period ground motions from large-magnitude distant earthquakes can cause serious damage to high-rise buildings, even to
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

those located in intraplate urban areas. Real-time monitoring and postearthquake performance assessment of high-rise buildings are therefore
essential. Using an integrated optimal sensor placement and response reconstruction scheme, this paper proposes a determinstic real-time
warning system and a probabilistic postearthquake performance assessment method for high-rise buildings during and after long-period
earthquakes, respectively. Both the warning system and the assessment method make good use of the accurate and complete estimation
of all the key structural responses reconstructed from the incomplete measurement data from limited sensors installed in the building.
The warning system monitors the building structural behavior in real-time and issues various levels of warning whenever it detects structural
responses exceeding the preset safety thresholds. The assessment method evaluates the structural component integrity after an earthquake in a
probabilistic manner, incorporating the extreme value distribution of the reconstructed structural responses with the structural component
fragility functions to determine the damage probabilities of the structural components. The proposed warning system and assessment method
were applied to a superhigh-rise building as a case study. The results from the case study showed that the proposed warning system and
assessment method are feasible and effective and have good potential for real application in high-rise buildings. DOI: 10.1061/(ASCE)
ST.1943-541X.0002323. © 2019 American Society of Civil Engineers.

Introduction engineers, and such an assessment can take a long time to be com-
pleted, causing considerable financial losses to the owners, particu-
High-rise buildings, even those located in intraplate regions, may larly to those of commercial buildings (Naeim 2013). The USGS
be subject to long-period ground motions from large-magnitude has developed tools, such as ShakeMap and ShakeCast, to provide
distant earthquakes. Because the long-period waves attenuate rel- rapid estimates of the impact of an earthquake on the built environ-
atively slowly along the traveling path and can be amplified by the ment. For example, ShakeCast enables estimating the damage to
soft soil site, they could cause considerable seismic damages to bridges in California following an earthquake using the stored fra-
high-rise buildings due to resonance. The impact of long-period gility curves (Mangalathu 2017). HAZUS, proposed by FEMA,
ground motions on high-rise buildings was shown by the 1985 can estimate the damage to buildings using built-in fragility curves
Michoacán earthquake (Beck and Hall 1986) and the 2011 Tohoku (HAZUS-MH 2003). However, these tools are more suitable to
earthquake (Takewaki et al. 2011), during which excessive vibra- general structures such as continuous girder bridges and low-rise
tions of high-rise buildings and serious damage to their nonstruc- buildings. The seismic performance of high-rise buildings sub-
tural components were observed in Mexico City and Tokyo, jected to long-period ground motion cannot be exactly quantified
respectively. However, because of the complex nature of long- by using these tools. With the recent development of sensor tech-
period ground motions and the lack of reliable long-period seismic nology, structural health monitoring (SHM) systems have been
records, the structural behaviors of high-rise buildings subjected installed on some high-rise buildings to provide real-time mea-
to long-period ground motions have not been fully understood. surements of earthquake-induced ground motions and dynamic re-
Therefore, there is a great need to develop real-time monitoring sponses of building structures. However, how to make good use of
systems for high-rise buildings so that ground motions and struc- SHM systems for real-time and postearthquake performance as-
tural responses can be recorded, real-time warning signals can be sessment of high-rise buildings is not clear.
issued during earthquakes, and postearthquake performance assess- Celebi et al. (2004) proposed quantifying the structural condi-
ment can be conducted immediately after earthquakes. tion using the interstory drift ratio (IDR) calculated by differencing
Traditionally, the postearthquake performance assessment of the relative displacement between two adjacent floors computed
a high-rise building requires a detailed evaluation by licensed through double integration of accelerometer measurement. Ulusoy
et al. (2012) designed and implemented a structural health moni-
1
Ph.D. Candidate, Dept. of Civil and Environmental Engineering, toring and alerting system consisting of sensing and analysis
Hong Kong Polytechnic Univ., Hung Hom, Kowloon 999077, Hong Kong modules to monitor the hospital buildings in high-seismic-hazard
(corresponding author). ORCID: https://orcid.org/0000-0002-3293-2983. regions. Kaya and Safak (2015) developed a real-time data process-
Email: hu.rongpan@connect.polyu.hk ing and performance evaluation scheme which set three levels of
2
Professor, Dept. of Civil and Environmental Engineering, Hong Kong alarms in accordance with the interstory drift ratio thresholds of
Polytechnic Univ., Hung Hom, Kowloon 999077, Hong Kong. Email: three different damage states as defined in seismic design codes.
ceylxu@polyu.edu.hk
All the aforementioned works selected the interstory drift ratio as
Note. This manuscript was submitted on May 15, 2018; approved on
November 6, 2018; published online on March 19, 2019. Discussion period an indicator of the structural performance. The calculation of the
open until August 19, 2019; separate discussions must be submitted for interstory drift ratio requires the integration of the measured accel-
individual papers. This paper is part of the Journal of Structural Engineering, eration responses and the subtraction of the displacement responses
© ASCE, ISSN 0733-9445. of two adjacent floors. The integration and subtraction processes,

© ASCE 04019038-1 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 1. Flowchart of the RTW system and the PPPA method.

however, significantly amplify the errors caused by the measure- did not address the OSP-RR issue, they took into consideration the
ment noise in the records and thus introduce tremendous uncertain- uncertainty in the estimation process and gave an analytical solu-
ties (Safak and Hudnut 2006). Furthermore, to obtain a complete tion of the extreme value distribution (EVD) prediction for struc-
estimation of structural performance, large numbers of accelerom- tural responses in terms of the measured acceleration responses.
eters should be deployed on each floor of the building, which However, their work used the Poisson assumption in analyzing the
greatly increases the cost of implementing a SHM system or makes peaks-over-threshold (POT) rates of the nonstationary process, and
it impossible to install the SHM system. the results tend to be too conservative when the threshold is low.
To deal with such issues, the authors proposed an integrated This is particularly true for superhigh-rise buildings subjected to
optimal sensor placement and response reconstruction (OSP-RR) long-period ground motion because their responses exhibit signifi-
method based on Kalman filtering algorithm, which enables an cant differences in duration and frequency characteristics from
accurate and complete estimation of structural responses using in- the structures discussed by Tien et al. (2016). Therefore, this study
complete measurement data from limited sensors installed on a proposes a probabilistic postearthquake performance assessment
long-span bridge or a high-rise building (Xu et al. 2016; Hu et al. (PPPA) method for high-rise buildings subjected to long-period
2018). In this method, the state of the structure can be estimated ground motion based on the OSP-RR method and the refined ex-
based on the incomplete structural response measurements from treme value distribution (EVD) prediction method. The extreme
multitype sensors (including GPS, inclinometers, and accelerome- values of the reconstructed structural responses are incorporated
ters) using the Kalman filtering algorithm, through which the with the structural component fragility functions to determine
unmeasured structural responses can be reconstructed. At the same the damage probabilities of the structural components. Fig. 1 is
time, the optimal placement of the multitype sensors can be ob- a flowchart of the RTW system and the PPPA method proposed
tained by minimizing the estimation errors of structural responses in this study and which are discussed in detail in the subsequent
to a target value. This study developed a real-time warning (RTW) sections. The proposed RTW system and PPPA method is applied
to a superhigh-rise building to demonstrate their feasibility and
system based on the OSP-RR method which can be used to monitor
effectiveness.
the performance and integrity of the building in real-time and issue
warning massages whenever the RTW system detects the structural
responses exceeding the preset safety thresholds.
Although the RTW system is very useful to issue warning mes-
Optimal Sensor Placement and Response
Reconstruction
sages for avoiding fatality, it is too conservative to use the design
value as a safety threshold, which reflects the overall structural
performance, for component level performance evaluation of a Modeling High-Rise Building Structure System in
high-rise building. The postearthquake performance evaluation (or State-Space
offline assessment) is therefore essential to provide the building Postdisaster investigations demonstrated that long-period ground
owners with reliable information for making decisions on mainte- motions from large-magnitude distant earthquakes can cause exces-
nance, repair, and occupancy. Miranda (2006) and Naeim et al. sive vibrations of high-rise buildings and serious damage to their
(2006) developed an automated post-earthquake damage assess- nonstructural components (Beck and Hall 1986; Takewaki et al.
ment of instrumented buildings in a probabilistic way by incorpo- 2011), but almost no partial or complete collaspe of high-rise build-
rating the fragility curves, through which the damage level of ing structures was observed. This is mainly because the seismic
specific structural components can be determined. However, their design requirements for high-rise buildings are strict and the
work did not address how to obtain a complete and accurate esti- high-rise building structure systems have great redundancy. Minor
mation of building responses, and also did not consider the uncer- damage to local structural components and serious damage to non-
tainty in the structural response measurement. Tien et al. (2016) structural components may not change the global dynamic proper-
developed a structural response estimation scheme based on the ties and behaviors significantly. In this regard, a high-rise building
Kalman smoothing algorithm that can make use of all the available is modeled as a linear-elastic structure system in this study. The
measurement data after the earthquake event. Although their work equation of motion for a linear structure with n degrees of freedoms

© ASCE 04019038-2 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


2 3
(DOFs) subjected to seismic-induced ground motion can be ex- Φm 0
pressed in the state-space form 6 7
Cm ¼ 6
4 Ψm 0 7
5 ;

żðtÞ ¼ Ac zðtÞ þ Bc g̈ðtÞ ð1Þ −Φm ω20 −2Φm ξω 0 m×2s


2 3
0
6 7
where zðtÞ ∈ ℜ2s = state vector in the modal coordinate, where Dm ¼ 4 0 5 ð5Þ
superscript sð<nÞ = number of observable modes in real applica- −Φ Φ ML
m T
m×p
tion; Ac ∈ ℜ2s×2s = state matrix; Bc ∈ ℜ2s×p = input matrix; and
g̈ðtÞ ∈ ℜp = seismic excitation vector consisting of p ground
motion accelerations To discretize the continuous state space equation and to take the
modeling error and measurement noise into consideration, Eqs. (1)
    and (3) can be rewritten
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

qðtÞ 0 I
zðtÞ ¼ ; Ac ¼ ;
q̇ðtÞ 2s×1 −ω20 −2ξω0 2s×2s zkþ1 ¼ Azk þ Bg̈k þ wk
 
0
k ¼ C zk þ D g̈k þ v k
ym ð6Þ
m m
Bc ¼ ð2Þ
−ΦT ML 2s×p

where zk ¼ zðkΔtÞ ∈ ℜ2s = discrete-time state vector; A ¼


eAc Δt ∈ ℜ2s ; B ¼ ∫ Δt Ac τ 0 dτ 0 B ¼ A−1 ðeAc Δt − IÞB ∈ ℜ2s×p ;
where qðtÞ ∈ ℜs = modal coordinate vector; ω0 ∈ ℜs×s and 0 e c c c
2s
ξ ∈ ℜs×s = modal frequency matrix and the modal damping ratio k ∈ ℜ = measured responses; and w k ∈ ℜ = process noise and
ym m

matrix, respectively; Φ ∈ ℜn×s = mode shape matrix of selected vk ∈ ℜm = measurement noise, which are modeled as zero-mean
mode sets; M ∈ ℜn×n = mass matrix of the dynamic system; and white processes with constant variance matrixes Q ∈ ℜ2s×2s and
L ∈ ℜn×p = mapping matrix that relates the excitations to the R ¼ EðvvT Þ ∈ ℜm×m , respectively.
corresponding DOFs.
The observation function can be expressed as
Dynamic Response Reconstruction Based on Kalman
Filtering
yðtÞ ¼ Cm zðtÞ þ Dm g̈ðtÞ ð3Þ
The building structure system described by Eq. (6) is a dynamically
evolving system with noise-contaminated measurements and sys-
where yðtÞ ∈ ℜm = observation vector, m = number of sensor tem uncertainties, and it can be modeled graphically as a dynamic
measurements; and Cm ∈ ℜm×2s and Dm ∈ ℜm×p = output and Bayesian network (DBN) (Fig. 2).
transmission matrixes, respectively, which consist of the mode The DBN is composed of a series of Bayesian networks (BNs),
shapes at the DOFs with sensors. which represents the system at a discrete time step k (k ¼ 0;
In this study, according to the types of sensor measurements, 1; : : : ; N), where k ¼ 0 indicates the initial values (Fig. 2). The
the displacement, inclination and acceleration responses at differ- DBN shows the evolution of the system state z over time under
ent levels of a high-rise building are included in the observation the input stochastic process u (referred to as the ground accelera-
vector yðtÞ tion g̈ in this study) and the process noise w. The observation of z is
taken as the measurement vector y with measurement noise v. If the
2 3 DBN is a linear-Gaussian system in state space, one of the widely
Φm qðtÞ
used methods for finding its solution is the Kalman filtering.
6 7
yðtÞ ¼ ½ dðtÞ θðtÞ aðtÞ T ¼ 4 Ψm qðtÞ 5 ð4Þ The Kalman filtering approximates the filtering distribution by a
Φ q̈ðtÞ
m Gaussian distribution and results in the best linear estimator in the
m×1
sense of minimizing the mean-square error. Given the measurement
y 1∶k and external force g1∶k from the first step up to time step k, the
where Ψm and Φm = rotation and displacement modal shapes at estimated system state zk through the Kalman filtering follows a
measured DOFs, respectively; and dðtÞ, θðtÞ, and aðtÞ = vectors Gaussian distribution expressed as
representing the displacement, inclination, and acceleration re-
sponses, respectively pðzk jy 1∶k ; g1∶k Þ ¼ Nðẑk ; Pk Þ ð7Þ

w1 wk wk+1 wN

u0 uk-1 uk uN

z0 z1 zk zk+1 zN

v1 vk vk+1 vN

y1 yk yk+1 yN

Fig. 2. Dynamic system modeled as a DBN. (Adapted from Tien et al. 2016.)

© ASCE 04019038-3 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


δk ¼ y rk − y ek ¼ ðCe zk þ De g̈k Þ − ðCe ẑk þ De g̈k Þ ¼ Ce ðzk − ẑk Þ
ð15Þ

The covariance matrix of the estimation error can be ex-


pressed as
n o
Δk ¼ E½δk δTk  ¼ E ½Ce ðzk − ẑk Þ½Ce ðzk − ẑk ÞT ¼ Ce Pk CeT
ð16Þ

Matrix Cm can be an ill- conditioned matrix due to the different


magnitudes of the displacement, rotational angle, and acceleration
Fig. 3. Diagram of operation of Kalman filter algorithm. responses. To deal with this issue, a normalization method origi-
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

nally proposed by Zhu et al. (2013) is adopted, in which the stan-


dard deviation of the sensor noise is used to normalize the modal
shapes. The normalized Cm is given by
where Nð·Þ = Gaussian distribution; and ẑk and Pk = mean vector 2 3
and covariance matrix, respectively, of the Gaussian distribution Φm =σd 0
6 7
Nð·Þ (Chen 2003). C~ m ¼ 6 Ψm =σθ 0 7 ¼ ðRÞ−1=2 Cm ð17Þ
4 5
To obtain a steady-state solution, the Kalman filter (KF) algo-
rithm is performed in the following steps (Fig. 3). −Φ ω0 =σa −2Φ ξω0 =σa
m 2 m

The mean vector and covariance matrix of the system state at


time step k are first predicted by where σd , σθ , and σa = standard deviations of the noise from the
GPS, inclinometer, and accelerometer, respectively.
ẑkjk−1 ¼ Aẑk−1 þ Bgk−1 ð8Þ The maximum and average estimation errors at all locations are
defined as

Pkjk−1 ¼ APk−1 AT þ Q ð9Þ ~


trðΔÞ
~
σ~ 2max ¼ maxðdiagðΔÞÞ; σ~ 2avg ¼ ð18Þ
n
where ẑkjk−1 ¼ E½zk jy k−1 ; gk−1  and Pkjk−1 ¼ Cov½ðzk − ẑkjk−1 Þ =
a priori estimations of the mean vector and covariance matrix, The optimization function and constraint condition can be
respectively, at time step k propagated from step k − 1. defined as
The Kalman gain matrix Kk , the updated estimation of mean ~
min traceðΔÞ ð19Þ
vector ẑk, and the covariance matrix Pk are then computed by
subject to
Kk ¼ Pkjk−1 CmT ½Cm Pkjk−1 CmT þ R−1 ð10Þ
σ~ 2max ≤ ½σ~ 2max  ð20Þ
ẑk ¼ ẑkjk−1 þ Kk ½y k − C ẑkjk−1 − D gk 
m m
ð11Þ where ½σ~ 2max  refers to the maximum estimation error.
Instead of exhaustive searching, the optimal sensor placement
Pk ¼ ½I − Kk Cm Pkjk−1 ð12Þ is achieved by a sequential selection process in which the sensor
candidates are deleted location by location until the preset estima-
where ẑk ¼ E½zk jy k ; gk  and Pk ¼ Cov½ðzk − ẑk Þ = a posteriori tion error criterion is achieved. The optimal locations and the cor-
estimation of the mean vector and covariance matrix, respectively, responding sensor numbers for the three types of sensors can be
at time step k given the measurement y k . determined simultaneously. Based on the optimal sensor placement
By Kalman filter terminology, the mean state vector ẑk obtained scheme, structural responses error can be can be reconstructed with
by an iterative algorithm of the aforementioned two processes is minimum estimation error using Eq. (13).
regarded as the minimum-variance unbiased state estimate. Thus,
the estimated responses y ek at the interested locations and the cor-
responding real responses y rk are respectively obtained by SHM-Based Real-Time Warning System

y ek ¼ Ce ẑk þ De gk ð13Þ Working Flow of RTW System


The SHM-based RTW system should be able to detect whether
y rk ¼ Ce zk þ De gk ð14Þ the building structure remains in a safe state and expedite alarms
immediately when the structural responses exceed the predefined
where subscripts e and r represent estimation and real, respectively. safety threshold, which is essential to prevent fatalities and eco-
nomic loss. A reliable and effective warning system requires a com-
plete and accurate estimation of structural responses at all key
Optimal Sensor Placement
locations. To achieve it, the real-time response reconstruction with
In the framework of Kalman filtering–based state estimation, the a well-designed sensor placement configuration that optimizes
optimal sensor placement can be achieved by selecting sensor types the response reconstruction results is required (section “Optimal
and locations that minimize the covariance of estimation error Sensor Placement and Response Reconstruction”). Proper safety
(mean-squared deviation from the mean). The error between the criteria should also be selected to reflect various levels of struc-
estimated and actual responses is defined as tural behavior and performance. In this study, the RTW system is

© ASCE 04019038-4 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


GPS Safety Threshold Values
Sensory system Accelerometer
Inclinometer For high-rise buildings, the interstory drift ratio is a significant in-
et. al dex for structural condition and damage extent (Celebi et al. 2004).
In the framework of performance-based earthquake engineering de-
sensors sign, the building structures are designed to different performance
Data Acquisition levels by selecting different criteria in terms of the interstory drift
Sub-stations
System ratios (Moehle and Deierlein 2004). In order to indicate the struc-
Control Center
tural performance levels and send out different levels of alarm,
proper safety criteria in accordance with different performance lev-
Data Displacement
els should be predefined for the RTW system. According to the
processing Chinese Seismic Design Code, there are three earthquake design
software levels (Wang 2008)
Acceleration
Data processing
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

1. The minor earthquake design level (63% exceeding probability


High- Inclination in 50-year service life, or a 50-year return period);
performance 2. The moderate earthquake design level (10% exceeding probabil-
computer et. al
ity in 50-year service life, or a 475-year return period); and
3. The major earthquake designs level (2%–3% exceeding prob-
Response reconstruction
ability in 50-year service life, or a 2,000-year return period).
Real-time warning For the minor earthquake design level, the maximum interstory
system Comparison with safety threshold drift ratio should not exceed 1/500 and the high-rise building struc-
ture remains intact without damage. Under a moderate earthquake,
Performance assessment the maximum interstory drift ratio is expected to be less than 1/200
but could be larger than 1/500. The dynamic properties of the high-
rise building structure remain almost the same as the linear elastic
Sending out alerts
Action and control state, and only minor damage to local structural components is
system Evacuating residence expected to occur. When subjected to a major earthquake, the high-
Trigger control system rise building structure may experience severe damage locally, but
major connection failure and collapse are not allowed. The inter-
Fig. 4. Flowchart of the SHM-based RTW system. story drift ratio limit in such a case is 1/100. The details of the de-
sign seismic hazard and corresponding structural behaviors are
described in Table 1, with reference to a region with seven-degree
embedded with a three-level alarm scheme which is designed in design seismic intensity (Building Industry Press 2010).
accordance with the three seismic performance levels stipulated in In accordance with the three different performance levels, the
seismic design code GB50011 (Building Industry Press 2010). The RTW system is designed with a three-level alarming system, and
working flowchart of the SHM-based RTW system is plotted in the details of the alarming system are described in Table 2. In ad-
Fig. 4 and can be described as follows: dition to the prevention of fatalities and economic loss, the alarm
1. The real-time measurement signals are collected by the sensors system together with the measured ground motion, if any, can be
and the data acquisition system to obtain the measurement data; used to assess the design quality of the high-rise building.
2. The measurement data are transmitted to the data processing
system for data preprocessing;
3. The preprocessed measurement data are used in conjunction SHM-Based Probabilistic Postevent Performance
with the response reconstruction method to provide the com- Assessment Method
plete response estimation of the building structure in the real-
time warning system; Although the RTW system is a direct and effective way to detect
4. The complete response estimation is compared with the prede- whether a high-rise building subjected to earthquake-induced
fined safety thresholds to find the possible damage whenever ground motion remains in the safe state or not, it is inadequate
the response estimation exceeds the threshold in the real-time to evaluate the structural integrity and damage extent at the struc-
warning system; and tural component level. The safety threshold adopted in the RTW
5. The warning signal is issued, and the active or semiactive con- system is a relatively rough indicator which reflects the overall
trol devices installed in the building, if any, are activiated. behavior of a high-rise building, and the safety threshold values

Table 1. Three levels of seismic hazards and corresponding structural performance


Seismic hazard level
Key parameter Minor earthquake Moderate earthquake Major earthquake
Return period 50 years 475 years 2,000 years
PGA (cm=s2 ) 35 100 200
IDR limitation <1=500 <1=200 <1=100
Damage feature No damage Minor damage Severe damage
Structural performance Intact; elastic; linear Primarily elastic; dynamic properties Severe damage; no connection
remain consistent with the elastic state fracture; no collapse
Note: PGA = peak ground acceleration.

© ASCE 04019038-5 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Table 2. Details of three alarming levels
Alarm levels
Key parameter Level 1 Level 2 Level 3
IDR estimation >1=500 >1=200 >1=100
Damage feature No damage Minor damage Severe damage
Structural safety Immediate occupancy Occupancy after repair Collapse prevention

are too conservative for component-level performance evaluation. The KS algorithm is performed through a forward-backward
At present, the fragility curves of different structural components process in which the forward recursion uses the observations from
are commonly adopted as an indicator of the component-level dam- y 1 to y k and is known as the Kalman filter, whereas the backward
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

age state as suggested by Naeim et al. (2006). Furthermore, to avoid recursion works backward using the measurement from y N to y kþ1 .
time lags in the online operation, the RTW system does not con- The mean vector and covariance matrix of the state vector at
sider the uncertainties inherent in the structural state estimation pro- time step k þ 1 are first predicted by
cess, but uses only the mean value ẑk of the posterior distribution of
ẑkþ1jk ¼ Aẑk þ Bgk ð21Þ
the estimated response state.
To account for the uncertainties and to facilitate more reasonable
and detailed decision for the postevent repair and maintenance of a Pkþ1jk ¼ APk AT þ Q ð22Þ
building structure, a probabilistic postevent performance assess-
The smoother gain matrix is obtained by
ment (PPPA) method is also proposed in this study. For the post-
event assessment of the structural performance, all the measured Jk ¼ Pk AT ðPkþ1jk Þ−1 ð23Þ
information during the earthquake are available and a smoothing
algorithm [Kalman smoother (KS)] can be used to obtain a more The updated estimation of the mean state vector and covariance
accurate and stable state estimation. One of the well-recognized matrix of the state vector at time step k then becomes
schemes for structural performance assessment and structural reli-
ability evaluation during/after an extreme event is to calculate the ẑkjN ¼ ẑk þ Jk ðẑkþ1jN − ẑkþ1jk Þ; k ¼ N − 1; : : : ; 0 ð24Þ
probability of the structural response up-crossing a certain thresh-
old, or the so-called peaks-over-threshold rates. The estimated re- PkjN ¼ Pk þ Jk ðPkþ1jN − Pkþ1jk ÞJTk ð25Þ
sponses can be regarded as a time-varying mean nonstationary
process in which the posterior distribution parameters of the where Jk = smoother gain matrix; N = final time step; and ẑkjN =
dynamically evolving structural responses can be quantified. Thus, smoothed estimation of the mean state vector at the kth time step.
it becomes possible to predict the extreme value distribution (EVD) The posterior distribution of the estimated state vector based on
in terms of POT rates based on the Kalman smoother inference. the Kalman smoother, given the measurement y 1∶N and external
Tien et al. (2016) proposed an analytical solution of the ex- force g1∶N from the first step to time step k, can be expressed by
treme value distribution prediction for structural responses based on pðzk jy 1∶N ; g1∶N Þ ¼ NðẑkjN ; PkjN Þ ð26Þ
the measured acceleration. They adopted the Poisson assumption
in analyzing the POT rates of the nonstationary process, and the The corresponding covariance matrix for the estimated state
results tend to be too conservative when the threshold is low. To vector z ¼ ½ q q̇ T can be described by
overcome the potential conservativeness induced by the Poisson  
assumption, the Vanmarcke assumption is used in this study as an Pzz Pzż
Pk ¼ ¼ Covðzk ; zk Þ
alternative. The EVD predictions based on both the Poisson as- Pżz Pż ż k
sumption and Vanmarcke assumption are computed and compared  
Covðq; qÞ Covðq; q̇Þ
with the results of Monte Carlo (MC) simulation (MCS). Moreover, ¼ ð27Þ
the fragility curves of structural components under slight damage Covðq̇; qÞ Covðq̇; q̇Þ k
scenario are incorporated with the calculated EVD results to give
the corresponding probability of the damage state. To convert the state covariance to the response covariance, the
following transformation can be used:

Kalman Smoothing and Evolutionary Probability σ2XX ¼ ΦPzz ΦT


Density σ2Ẋ Ẋ ¼ ΦPż ż ΦT
Whereas the Kalman filter algorithm computes the posterior prob- ρẊX ¼ ΦPżz ΦT ð28Þ
ability of the system given past and present measurements, the
Kalman smoother algorithm allows, after an event, computing the where σ2XX , σ2Ẋ Ẋ , and ρẊX = time-dependent outputs from the
posterior distribution at time step k with respect to all measure- smoothing process.
ments ½y 1 ; y 2 ; : : : ; y k ; : : : ; y N  even after the time at which the state
is being evaluated. The KS algorithm has the following advantages
Extreme Value Prediction of Structural Response
over the KF algorithm (Kim and Sohn 2017):
• The influence of noise is minimized; In reliability analysis, the extreme value distribution of a structural
• The measurement accuracy particularly at the beginning of response is essential in order to determine the extreme structural
estimation is improved; and response. Therefore, the distribution of the estimated structural re-
• The parameters of the posterior distribution can be determined, sponses can be computed using the Bayesian inference results pre-
given all the observations ½y 1 ; y 2 ; : : : ; y k ; : : : ; y N . sented in the previous section. To address this issue, a theoretical

© ASCE 04019038-6 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


formula for calculating the probability of a non-zero-mean, nonsta- Thus
tionary process up-crossing a given threshold is introduced (Tien
1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
et al. 2016).
σ ðtÞ
Consider a nonstationary process ZðtÞ, and calculate the up- ηPX ðβ; tÞ ¼ 1 − ρ2XẊ ðtÞ Ẋ
crossing probability over a specified threshold α during an interval π σX ðtÞ
0 1
ð0; TÞ, where ZðtÞ can be obtained from the estimated state vector ẑ  2

through linear transformation. For example, ZðtÞ can be seen as the 1 x B ρXẊ ðtÞx C
× exp − 2 Ψ@qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi A ð35Þ
interstory drift ratio, which is regarded as an essential index for 2 σX ðtÞ 2
1 − ρXẊ ðtÞσX ðtÞ
structural performance assessment. The estimated interstory drift
ratio at a certain location can be obtained by
pffiffiffiffiffiffiffiffi
where ΨðξÞ ¼ expð−ξ 2 =2Þ þ 2πξ ΦðξÞ, where ΦðξÞ = cumula-
ZðtÞ ¼ SΘẑ ð29Þ tive normal distribution function.
For a high threshold, it is reasonable to assume that the occur-
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

where S = location matrix designating the interested location; Θ =


rence of up-crossings is independent, especially for a broadband
modal matrix for the interstory drift ratio; and ẑ = estimated mean
process (Michaelov et al. 2001). However, the Poisson approxima-
vector of the structural state vector.
tion is thought to be too conservative for low-threshold scenarios
Define Zmax ¼ max½jZðtÞj and let ηZ ðα; tÞ denote the transient
(Tien et al. 2016), in which the CDFs are not monotonically de-
up-crossing rate of ZðtÞ above the threshold α within the time in-
creasing for low thresholds. Tien et al. (2016) attributed this phe-
terval ðt; t þ dtÞ. The cumulative distribution function (CDF) of the
nomenon to the fact that the Poisson assumption does not consider
extreme value of the nonstationary process ZðtÞ is given by
the fact that once the threshold is exceeded, the process can stay
 Z  above the threshold for quite a long time without making additional
T
PðZmax > αÞ ¼ 1 − PðZmax ≤ αÞ ≅ 1 − exp − ηZ ðα; tÞdt up-crossings, especially for low thresholds. In this regard, the
0 Vanmarcke approximation, which assumes the up-crossings as a
ð30Þ two-state Markov process, is alternatively proposed in this study.
According to Michaelov et al. (2001) and Hu and Xu (2014), the
In general, ZðtÞ is not a zero-mean process, and both the mean Vanmarcke approximation to a general nonstationary process can
value and variance of ZðtÞ vary with time. It thus becomes an issue be expressed as the up-crossing rate of VðtÞ, which is the envelope
of calculating the CDF of a nonstationary process with a time- of the process XðtÞ as
varying mean to up-cross a fixed threshold. To facilitate the finding
of such a CDF, the following transformation is performed:
½1 − FV ðβ; tÞ2ηX ð0; tÞ
ηVX ðβ; tÞ ¼
XðtÞ ¼ ZðtÞ − μZ ðtÞ ð31Þ FV ðβ; tÞ
  
ηV ðβ; tÞ
· 1 − exp − ð36Þ
βðtÞ ¼ α − μZ ðtÞ ð32Þ ½1 − FV ðβ; tÞ2ηX ð0; tÞ

where μz ðtÞ = time-varying mean of the non-stationary pro-


cess ZðtÞ. where superscript V denotes the Vanmarcke approximation; ηX and
As a result, XðtÞ becomes a zero-mean process with σX ðtÞ ¼ ηV = unconditional transient up-crossing rates of XðtÞ and VðtÞ,
σZ ðtÞ and the fixed threshold α becomes a time-varying threshold respectively; and FV = transient CDF of VðtÞ. The joint PDF of
βðtÞ. Therefore, the probability of the random variable ZðtÞ exceed- the envelop function VðtÞ and its derivative V̇ðtÞ is
ing a threshold α can be obtained by determining the probability of
the random variable XðtÞ up-crossing a threshold βðtÞ, which can 1
be expressed as ηZ ðα; tÞ ¼ ηX ðβ; tÞ. f XẊ ðβ; ẋ; tÞ ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ρ2 ðtÞ
To determine the transient up-crossing rate, the simplest choice 2πσX ðtÞσẊ ðtÞqX ðtÞ 1 − qX2ẊðtÞ
X
is to assume that the up-crossings of a random process above a 2
given level occur independently and thus constitute a Poisson pro-  2
6 1 β
cess. The resulting Poisson approximation is then equal to the ex- · exp4− ρ2XẊ ðtÞ

2
pected unconditional transient up-crossing rate, which is given by 2 1 − q2 ðtÞ σ X ðtÞ
Z ∞
X
3
ηX ðβ; tÞ ¼
P
ðẋ − β̇ÞfXẊ ðβ; ẋ; tÞdẋ ð33Þ 
β ẋρXẊ ðtÞ ẋ2 7
β̇ −2 þ 5 ð37Þ
σX ðtÞσẊ ðtÞqX ðtÞ σ2Ẋ ðtÞq2X ðtÞ
where superscript p denotes the Poisson approximation; and
f XẊ ðβ; ẋ; tÞ = joint probability density function (PDF) of XðtÞ and
ẊðtÞ, which can be obtained by where qX ðtÞ is the transient bandwidth factor, defined as
1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
f XẊ ðβ; ẋ; tÞ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi R ∞ 
2
2πσX ðtÞσẊ ðtÞ 1 − ρ2XẊ ðtÞ 0 ωAðω; tÞ dω
qX ðtÞ ¼ 1 − ð38Þ
 σX ðtÞσẊ ðtÞ
1
· exp −
2ð1 − ρ2XẊ ðtÞÞ
 2  Using Eqs. (36) and (37), the formula for the Vanmarcke
β β ẋ ẋ2
× − 2ρ ðtÞ þ ð34Þ approximation originally developed by Michaelov et al. (2001)
σ2X ðtÞ X Ẋ
σX ðtÞσẊ ðtÞ σ2Ẋ ðtÞ can be extended for the crossing rate of VðtÞ as

© ASCE 04019038-7 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Z
1 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi σ ðtÞ ∞
ηVX ðβ; tÞ ¼ 1 − ρ2XẊ ðtÞ Ẋ Fdm ¼ PðDM ≥ dmÞ ¼ pðDM ≥ dmjαIDR Þ
π σX ðtÞ 0
 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  
pffiffiπ β q2X ðtÞ−ρ2XẊ ðtÞ ρXẊ ðtÞβ · PðZmax > αIDR ÞdαIDR ð41Þ
1 − exp − 2 σX ðtÞ 2
1−ρXẊ ðtÞ
Ψ p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
q2X ðtÞ−ρ2XẊ ðtÞσX ðtÞ
× h 2 i
exp 2σβ2 ðtÞ − 1 where pðDM ≥ dmjαIDR Þ = probability density function (PDF) of
X
damage given the IDR αIDR , which can be obtained from the CDF
ð39Þ formula in Eq. (40); and PðZmax > αIDR Þ = CDF of the estimated
pffiffiffiffiffiffiffiffi responses up-crossing the threshold αIDR .
where ΨðξÞ ¼ expð−ξ 2 =2Þ þ 2πξ ΦðξÞ, where ΦðξÞ = cumula-
tive normal distribution function.
Substituting the obtained up-crossing rate defined in Eq. (35) or Case Study
Eq. (39) into Eq. (30), the CDF of extreme value of the Poisson or
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Vanmarcke approximation of a random process can be determined. A high-rise building with a structural height of 580 m was used as a
case study to examine the feasibility and accuracy of the RTW sys-
tem proposed in section “SHM-Based Real-Time Warning System”
Definition of Fragility Function and Probabilistic and the PPPA method proposed in section “SHM-Based Probabi-
Damage State listic Postevent Performance Assessment Method.” Because both
In the framework of PPPA, the damage state of the structural the RTW system and the PPPA method are built on the multi-type
component is represented in a probabilistic way by adopting the sensor placement and response reconstruction method presented in
fragility function, which describes the conditional probability of section “Optimal Sensor Placement and Response Reconstruction,”
a structure component being or exceeding a specific damage state, the building model, the long-period ground motion, the optimal
such as slight, moderate, or severe damage, given the intensity of multitype sensor placement, and the response reconstruction of the
an earthquake. Unlike deterministic values recommended in high-rise building are discussed first.
FEMA-356 (FEMA 2000), fragility functions can consider the
structural, material, and geometric uncertainties and also can take
Building Model and Optimal Sensor Placement
into account the uncertainty of the structural motions that trigger
different levels of damage. In particular, a fragility function sup- The high-rise building had a megacolumn–core tube–outrigger
plies the probability that the structure component reaches or ex- truss structural system, and the building was divided into eight
ceeds a particular damage level. Available experimental data on zones along the height by the two-story outrigger truss. The core-
various types of structural components permit the development wall tube shrank gradually from the bottom to the top, with
of fragility functions. According to the research by Cornell et al. the cross section varying from 5,300 × 3,700 mm to 2,400 ×
(2002), the fragility functions for many structural components can 1,900 mm. Because it is still difficult to apply the Kalman filter
be modeled by a lognormal distribution and smoother algorithms to a structural system with a large num-
ber of DOFs, and to facilitate implementing the proposed method,
Fdm ðEDP ¼ IDRÞ ¼ P½DM ≥ dmjIDR a simplified two-dimensional (2D) finite-element (FE) model
  developed by Lu et al. (2014) for the high-rise building, was
ln dm − ln ηIDR adopted in this study. The 2D planar FE model consisted of
¼1−Φ ð40Þ
σln IDR 2,382 elements and 2,416 nodes with a total of 7,248 DOFs
(Fig. 5). The dynamic characteristics and modal information in
where Fdm ðEDPÞ = fragility function for damage state DM, and is translational direction of simplified 2D model are the same with
defined as the probability that the component reaches or exceeds those of the three-dimensional (3D) model. Thus, the 2D model
a given damage state dm given a particular engineering demand was adopted as a case study to implement the optimal sensor
parameter (EDP) value; ηIDR and σln IDR are the median and loga- placement and reconstruction the structural responses of the entire
rithmic standard deviation of DM, and these two parameters can be building.
determined from the test data; and Φð·Þ = cumulative standard nor- Inclinometers and accelerometers are usually installed on the
mal distribution. columns and core tube walls, which are regarded as candidate sen-
The fragility function serves as an effective tool for postearth- sor locations. Considering the symmetry of the 2D model about
quake structural performance assessment. Once the EDP is ob- the vertical axis, sensors were supposed to be placed on half of
tained by the response reconstruction based on the imcomplete the model. A total of 472 candidate sensor locations were selected,
sensor measurement, damage to specific components can be esti- including 236 for placing accelerometers and 236 for placing incli-
mated by using corresponding fragility functions. In this study, the nometers. Given the limited number of GPS, we only consider one
estimated extreme value distribution of the interstory drift ratio is GPS in this study, and the optimal location for the GPS was deter-
incorporated with the fragility curve to achieve a fast assessment mined together with those of the other sensors.
informing the probability of the structural component experiencing Fig. 6 shows the maximum and average estimation errors as the
what level of damage. Only slight and local damage scenarios are number of deleted sensors increased; 427 sensor location candi-
considered in this study such that the structural responses can be dates were deleted when the normalized maximum reconstruction
regarded as linear elastic, because the KF/KS-based Bayesian filter- error reached the predesignated threshold value of 1.0, indicating
ing method assumes the dynamic evolving system to be a linear that at least 46 sensors should be selected in this case (subtracting
Gaussian process. As mentioned previously, it is reasonable to the number of deleted sensors from the number of total candidate
make such an assumption because strict seismic design is required sensors). Finally, 1 GPS, 19 inclinometers, and 26 accelerometers
for the high-rise buildings and the high-rise building structure sys- were selected, and the corresponding optimal placement scheme for
tems have great redundancy. The joint probability of the damage each sensor is illustrated in Fig. 7. The optimized location for plac-
state can be rewritten ing the one GPS is on top of the building.

© ASCE 04019038-8 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Simplified model of the high-rise building.

Fig. 7. Sensor placement locations for each type of sensor.

A synthetic accelerogram for a M ¼ 8, R ¼ 400 km earth-


quake scenario was generated based on the ground-motion sim-
ulation scheme, where M is the earthquake magnitude and R
is the distance from the earthquake source to the site of the
building. A typical soil site condition in Shanghai was selected,
and the soil effect was calculated by program SHAKE91 (Idriss
and Sun 1992). In addition, three different seismic records, from
the 1971 San Fernando Earthquake, the 1985 Michoacán Earth-
quake, and the 2011 Tohoku Earthquake, were also selected as
input ground motions in this study to investigate how the differ-
Fig. 6. Number of sensors and corresponding estimation error. ent frequency characteristics of ground motions influence the
seismic responses and damage extent of the high-rise building.
Detailed information of the three selected seismic records are
listed in Table 3.
Input Ground Motions The peak ground accelerations (PGAs) of the selected ground
motions were scaled to 0.1 g (1 m=s2 ), which coincides with the
For superhigh-rise buildings, their first natural period can be as moderate earthquake design level in the Chinese Seismic Design
long as 6–10 s, which is far beyond the specification in seismic Code (Building Industry Press 2010), and the accelerograms are
design codes. Because most superhigh-rise buildings, including shown in Figs. 8–11.
Shanghai Tower, are located in low-to-moderate seismicity zones The Fourier amplitudes of the four scaled accelerograms are
where the seismic records are relatively rare, it is difficult to obtain illustrated in Fig. 12. Ground motions from the simulation, the
enough representative records of long-period ground motions. 1985 Michoacán Earthquake, and the 2011 Tohoku Earthquake had
Thus, a stochastic simulation scheme based on a seismological abundant long-period components, whereas the 1971 San Fernando
model is proposed to generate regional representative long-period Earthquake had fewer long-period waves. In addition, the Fourier
ground motion time histories (Hu et al. 2018), in which the param- amplitudes of the simulated ground motion were larger than those
eters used in the seismological model are determined from local of the three recorded ground motions within the 0.1–1 Hz fre-
geological investigation and limited seismic records. Once the tec- quency range which covered the first four modes of the building
tonic and geological parameters in the seismological model are de- model.
termined, it only requires specifying the earthquake magnitude and The natural frequencies of the first four modes of the building
hypocentral distance in the stochastic simulation method to gener- model and corresponding Fourier amplitudes of the four selected
ate a region-specific accelerogram. ground motions are listed in Table 4. The Fourier amplitudes of

© ASCE 04019038-9 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Table 3. Detailed information of three seismic records
Earthquake scenario Date of event Moment magnitude Station Hypocentral distance
San Fernando Earthquake February 9, 1971 6.6 Vernon, California CMD Building 49.9 km
Michoacán Earthquake Sepetmber 19, 1985 8 Villita, Mexico Guerrero Array 47.8 km
Tohoku Earthquake March 11, 201 9 Shinjuku, Japan 375 km
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 8. Accelerogram of the simulated ground motion. Fig. 10. Accelerogram of the 1985 Michoacán Earthquake.

Fig. 9. Accelerogram of the 1971 San Fernando Earthquake. Fig. 11. Accelerogram of the 2011 Tohoku Earthquake.

the simulated earthquake were larger than those of the other effectiveness and accuracy of the response reconstruction methods.
three ground motions at each of the four natural frequencies, Fig. 14 shows the close-up view of the peak interstory drift ratio
and the Michoacán earthquake had abundant frequencies of obtained from Fig. 13. The estimated peak interstory drift ratio us-
Modes 1, 3, and 4. ing the Kalman smoother was closer to the real ratio than was that
using the Kalman filter.
Response Reconstruction
Implementation of RTWS
The actual seismic responses of the 2D building model subjected to
the selected input ground motions were computed. The so-called The maximum IDR responses of the 2D model subjected to the four
measured responses were then obtained by adding 5% root-mean- ground motions are illustrated in Fig. 15. The maximum IDR re-
squared randomly generated noise to the computed responses at the sponses under all four seismic excitations were below the IDR lim-
sensor locations. The KF-based response reconstruction scheme its for minor damage specified by the seismic design code, which
was used to obtain the complete estimation of structural responses shows that the high-rise building was well designed. The maximum
based on the measured responses. The complete estimation based IDR responses in Zones 1–6 almost lay within the no damage area,
on the Kalman smoother algorithm was also obtained for compari- whereas the IDR responses in the upper zones (Zones 7–8) lay be-
son. The accuracy of the proposed optimal sensor placement and tween the no damage and minor damage areas. The maximum IDR
response reconstruction method was validated though the compari- responses along the whole structure occurred in Zone 8, which is
son of the estimated (reconstructed) responses with the real (com- consistent with the nonlinear finite element analysis results by
puted) responses. Jiang et al. (2011). Furthermore, comparing the IDR responses
The time histories of interstory drift ratio of the 110th story in of the four scenarios suggests that the long-period components
Zone 8 are illustrated in Fig. 13. The estimated response obtained have a significant impact on the seismic response of the high-rise
by both the KF-based and KS-based response reconstruction building. The different IDR responses are attributed to the differ-
methods matched well with the real response, thus verifying the ent frequency characteristics of the four ground motion excitations.

© ASCE 04019038-10 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 12. Fourier amplitudes of the four input ground motions.

Table 4. Fourier amplitudes of four ground motions for first four modes “Kalman Smoothing and Evolutionary Probability Density.” The
Fourier amplitude (m=s) samples generated by this Monte Carlo method yielded empirical
EVD of the responses which were used to validate the analytical
First Second Third Fourth function of the EVD prediction proposed in section “Extreme Value
mode mode mode mode
Prediction of Structural Response.” Fig. 16 shows the Monte Carlo
Earthquake scenario (0.11 Hz) (0.3 Hz) (0.63 Hz) (1 Hz)
simulation based on the posterior PDF of the estimated state using
Simulated earthuqake 0.71 1.12 1.35 1.02 the Kalman smoother.
San Fernando Earthquake 0.18 0.32 0.73 0.78 The CDFs of Zmax > α were calculated based on the Poisson
Michoacán Earthquake 0.62 0.32 0.8 0.47
and Vanmarcke assumptions, respectively, and then compared with
Tohoku Earthquake 0.32 0.33 0.65 0.78
the CDF obtained by the Monte Carlo simulation of 500 samples.
The comparison showed a good match between the analytical and
simulated extreme value distribution of the target response in the
In conclusion, the alert signal for minor damage would be issued high-threshold range (IDR > 2.5 × 10−3 ), whereas in the middle-
for the building subjected to the simulated ground motion and the threshold range (1.0 × 10−3 < IDR < 2.5 × 10−3 ), the results based
1985 Michoacán Earthquake, and the signal for no damage would on the Poisson assumption were relatively conservative compared
be issued for the building subjected to the 1971 San Fernando with the results based on the Vanmarcke assumption and the MC
Earthquake and the 2011 Tohoku Earthquake. simulation. The CDF of the extreme IDR responses based on the
Poisson assumption, the Vanmarcke assumption, and the MC sim-
ulation for the 110th story are shown in Fig. 17.
Implementation of PPPA With the CDF of the extreme IDR responses of the building, the
The maximum IDR responses of the building structure subjected damage state probablity of structural components of the building
to the 1971 San Fernando Earthquake and the 2011 Tohoku can be calculated using Eq. (41). According to the design detail
Earthquake did not exceed the threshold for minor damage, which of the high-rise building, the shear wall layout for Zone 8 is shown
indicates that the building structure remained intact, without dam- in Fig. 18. The sizes and reinforcement details for each type of
age. Thus, the PPPA was applied only to the cases of the simulated shear walls are listed in Tables 5 and 6 respectively. And the cor-
earthquake and the 1985 Michoacán Earthquake to evaluate the responding fragility curves as shown in Fig. 19 are given by the
corresponding damage of structural components of the building. To Performance Assessment Calculation Tool (PACT) FEMA
implement the PPPA method, the PDF/CDF of the extreme re- (2012) according to the sizes and reinforcement ratios.
sponse of the building was determined based on KS inference. The The damage probabilities (probability of the selected structural
system state trajectories can be easily sampled from the DBN component experiencing the defined damage) based on the EVD of
shown in Fig. 2 by using the forward filtering backward sampling the estimated responses (Tables 7 and 8, EVD of estimation) for
approach (Tien et al. 2016). The Monte Carlo method was also used each type of shear wall in the 110th story were calculated using
to generate samples corresponding to the posterior PDF inferred Eq. (41). For comparison, the damage probabilities were also
from DBN by the Kalman smoother formulas proposed in section calculated based on the mean value of the estimated responses

© ASCE 04019038-11 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

(c) (d)

Fig. 13. Comparison of real and estimated IDR responses at 110th story subjected to (a) simulated ground motion; (b) 1971 San Fernando
Earthquake; (c) 1985 Michoacán Earthquake; and (d) 2011 Tohoku Earthquake.

(a) (b)

(c) (d)

Fig. 14. Close-up view of the real and estimated peak IDR responses at 110th story subjected to (a) simulated ground motion; (b) 1971 San Fernando
Earthquake; (c) 1985 Michoacán Earthquake; and (d) 2011 Tohoku Earthquake.

© ASCE 04019038-12 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


(Tables 7 and 8, Mean estimation) by multiplying the mean esti- Compared with considering only the mean value of the estimated
mation by the corresponding damage probability defined in the fra- response, the extreme value accounts for all the possibilities that
gility curves. It can be concluded from the results that the damage the estimated response exceeds a certain threshold. Thus, by in-
probability of shear walls in the high-rise building subjected to such corporating the fragility curves of the various types of shear walls
a moderate earthquake is low (Tables 7 and 8). The extreme value with the extreme values, the refined damage probability predic-
was used to consider the uncertainty in the response estimation due tion can yield safer and more reliable postearthquake assessment
to the filter used, which was not addressed properly previously. (Tables 7 and 8).
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Fig. 15. Profile of maximum IDR responses under four ground


excitations. Fig. 18. Shear wall layouts of Zone 8.

(a) (b)

Fig. 16. Monte Carlo simulation based on the KS estimated IDR responses under (a) simulated ground motion; and (b) 1985 Michoacán Earthquake.

(a) (b)

Fig. 17. CDF of the extreme IDR responses at 110th story under (a) simulated ground motion; and (b) 1985 Michoacán Earthquake.

© ASCE 04019038-13 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Table 5. Sizes of shear walls in Zone 8 and corresponding fragility curves Table 8. Damage probability of shear walls in 110th story subjected to
in Performance Assessment Calculation Tool (PACT) 1985 Michoacán Earthquake
Thickness Length Height Fragility curve Damage probability
No. (mm) (mm) (mm) number in PACT
Component Mean EVD of
FE2 600 3,500 4,500 B1044.021 type Fragility curve estimation (%) estimation (%)
FE3 600 3,500 4,500
B1044.021 Slight damage 1.3 8.7
FN1 600 3,500 4,500
B1044.103 Slight damage 1.5 6.3
FN2 600 3,500 4,500
B1044.102 Slight damage 1.8 6.5
FS2 600 3,500 4,500
FS3 600 3,500 4,500
FW2 600 3,500 4,500
FW3 600 3,500 4,500
WE1 500 13,230 4,500 B1044.103
Conclusions
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

WE2 500 13,230 4,500


WW1 500 13,230 4,500
For the high-rise building concerned, the impact of long-period
WW2 500 13,230 4,500 ground motion was demonstrated by its interstory drift responses
WS2A 500 12,600 4,500 under four different seismic excitations. Based on the optimized
WN2A 500 12,600 4,500 sensor placement configuration, the KF-based response estimation
WN1 500 6,580 4,500 B1044.102 achieved a good match with the actual responses and thus demon-
WN2 500 6,580 4,500 strated the effectiveness of the proposed integrated optimal sensor
WS1 500 6,580 4,500 placement and response reconstruction scheme. According to the
WS2 500 6,580 4,500 predefined safety threshold, the RTW system detected that only the
top zone of the high-rise building experienced minimum damage
and sent out Level 1 alarms in the top zone. Then the PPPA scheme
was applied to obtain detailed damage information of the structural
Table 6. Reinforcement ratio for shear wall panels in Zone 8 components in the top zone. The estimated responses based on
the Kalman smoother matched better with the actual responses
Vertical Horizontal than did those based on Kalman filter. The EVD prediction based
reinforcement reinforcement
on the Vanmarke’s assumption achieved a good match with the
Wall type Wall panels (%) (%)
Monte Carlo simulation by avoiding the conservativeness of
Flange FS2, FS3, FN1, FN2, FW2, 1.0 0.6 the Poisson assumption. By incorporating the fragility curves of
wall FW3, FE2, FE3 the various types of shear walls with the EVD, the refined damage
Web WN2A, WS2A 1.0 1.0 probability prediction yields safer and more reliable postearthquake
wall WN1, WS2, WN2, WS2, 0.5 assessment compared with considering only the mean value of the
WW1, WE1, WW2, WE2 estimation. The SHM-based RTW system and PPPA method show
great potential for expediting the emergency response to prevent
further losses and injuries during earthquakes and enabling better
decision making for the postearthquake maintenance and inspec-
tion of the target building structure.

Acknowledgments

The work in this paper is financially supported by the Hong Kong


Polytechnic University as part of the group project “Fundamentals
of Earthquake Engineering for Hong Kong.” The first author ap-
preciates the Hong Kong Research Grants Council (RGC) for
providing the Hong Kong PhD Fellowship and the Faculty of
Construction Engineering for providing the one-year top-up stu-
dentship. The authors sincerely thank Prof. X. Z. Lu from Tsinghua
University and Dr. X. Lu from Beijing Jiaotong University for
Fig. 19. Fragility curves for shear walls in Zone 8 (slight damage). sharing knowledge about the nonlinear behavior of the FE model
of the high-rise building and providing the simplified 2D FE model.
Great appreciation goes to Dr. Tien from the Georgia Institute
of Technology for the discussion of the analytical extreme value
distribution function.
Table 7. Damage probability of shear walls in 110th story subjected to
simulated ground motions
Damage probability References
Component Mean EVD of
type Fragility curve estimation (%) estimation (%) Beck, J. L., and J. F. Hall. 1986. “Factors contributing to the catastrophe in
Mexico City during the earthquake of September 19, 1985.” Geophys.
B1044.021 Slight damage 1.5 9.0 Res. Lett. 13 (6): 593–596. https://doi.org/10.1029/GL013i006p00593.
B1044.102 Slight damage 1.7 6.5
Building Industry Press. 2010. Code for seismic design of buildings.
B1044.103 Slight damage 2.1 7.1
[In Chinese.] GB50011. Beijing: Building Industry Press.

© ASCE 04019038-14 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038


Celebi, M., A. Sanli, M. Sinclair, S. Gallant, and D. Radulescu. 2004. Michaelov, G., L. D. Lutes, and S. Sarkani. 2001. “Extreme value of re-
“Real-time seismic monitoring needs of a building owner—and the sponse to nonstationary excitation.” J. Eng. Mech. 127 (4): 352–363.
solution: A cooperative effort.” Earthquake Spectra 20 (2): 333–346. https://doi.org/10.1061/(ASCE)0733-9399(2001)127:4(352).
https://doi.org/10.1193/1.1735987. Miranda, E. 2006. “Use of probability-based measures for automated dam-
Chen, Z. 2003. “Bayesian filtering: From Kalman filters to particle filters, age assessment.” J. Struct. Des. Tall Build. 15 (1): 35–50. https://doi.org
and beyond.” Statistics 37 (1): 1–24. https://doi.org/10.1080/023318 /10.1002/tal.342.
80309257. Moehle, J., and G. G. Deierlein. 2004. “A framework methodology for
Cornell, C., F. Jalayer, R. Hamburger, and D. Foutch. 2002. “Probabilistic performance-based earthquake engineering.” In Proc., 13th World
basis for 2000 SAC Federal Emergency Management Agency steel mo- Conf. on Earthquake Engineering. Vancouver, Canada: Canadian
ment frame guidelines.” J. Struct. Eng. 128 (4): 526–533. https://doi.org Association for Earthquake Engineering.
/10.1061/(ASCE)0733-9445(2002)128:4(526). Naeim, F. 2013. “Real-time damage detection and performance evalua-
FEMA. 2000. Prestandard and commentary for the seismic rehabilitation tion for buildings.” In Earthquakes and Health Monitoring of Civil
of buildings. Rep. FEMA-356. Washington, DC: FEMA. Structures, 167–196. New York: Springer.
FEMA. 2012. Seismic performance assessment of buildings volume 3— Naeim, F., S. Hagie, A. Alimoradi, and E. Miranda. 2006. “Automated
Downloaded from ascelibrary.org by Louisiana Dept Trans & Dev on 06/19/19. Copyright ASCE. For personal use only; all rights reserved.

Performance assessment calculation tool (PACT) version 2.9.65. post-earthquake damage assessment of instrumented buildings.” In
Rep. FEMA P-58-3.1. Washington, DC: FEMA. Vol. 66 of Advances in earthquake engineering for urban risk reduction
HAZUS-MH. 2003. Multi-hazard loss estimation methodology: Earth- NATO science series IV: Earth and environmental sciences, 117–134.
quake model. Washington DC: Dept. of Homeland Security, FEMA. Dordrecht, Netherlands: Springer.
Hu, L., and Y. L. Xu. 2014. “Extreme value of typhoon-induced Safak, E., and K. Hudnut. 2006. “Real-time structural monitoring and
non-stationary buffeting response of long-span bridges.” Probab. damage detection by acceleration and GPS sensors.” In Proc., 8th US
Eng. Mech. 36 (7): 19–27. https://doi.org/10.1016/j.probengmech National Conf. on Earthquake Engineering. San Francisco.
.2014.02.002. Takewaki, I., S. Murakami, K. Fujita, S. Yoshitomi, and M. Tsuji. 2011.
Hu, R. P., Y. L. Xu, and X. Zhao. 2018. “Long-period ground motion sim- “The 2011 off the Pacific coast of Tohoku earthquake and response
ulation and its impact on seismic response of high-rise buildings.” of high-rise buildings under long-period ground motions.” Soil Dyn.
J. Earthquake Eng. 22 (7): 1285–1315. https://doi.org/10.1080 Earthquake Eng. 31 (11): 1511–1528. https://doi.org/10.1016/j.soildyn
/13632469.2017.1286617. .2011.06.001.
Idriss, I. M., and J. I. Sun. 1992. User’s manual for SHAKE91: A computer Tien, I., M. Pozzi, and A. Der Kiureghian. 2016. “Probabilistic framework
program user’s manual. Davis, CA: Center for Geotechnical Modeling. for assessing maximum structural response based on sensor measure-
Jiang, H., L. He, X. Lu, J. Ding, and X. Zhao. 2011. “Analysis of seismic ments.” Struct. Saf. 61: 43–56. https://doi.org/10.1016/j.strusafe.2016
performance and shaking table tests of the Shanghai Tower.” J. Build. .03.003.
Struct. 32 (11): 55–63. Ulusoy, H., E. Kalkan, J. Fletcher, P. Friberg, W. Leith, and K. Banga.
Kaya, Y., and E. Safak. 2015. “Real-time analysis and interpretation of 2012. “Design and Implementation of a structural health monitoring
continuous data from structural health monitoring (SHM) systems.” and alerting system for hospital buildings in the United States.” In
Bull. Earthquake Eng. 13 (3): 917–934. https://doi.org/10.1007/s10518 Proc., 15th World Conf. on Earthquake Engineering. Lisbon, Portugal:
-014-9642-9. Sociedade Portuguesa de Engenharia Sismica.
Kim, K., and H. Sohn. 2017. “Dynamic displacement estimation by fusing Wang, Y. 2008. “A new round of updation of seismic design code of
LDV and LiDAR measurements via smoothing based Kalman filtering.” China.” In Proc., 14th World Conf. on Earthquake Engineering.
Mech. Syst. Signal Process. 82: 339–355. https://doi.org/10.1016/j Beijing: Chinese Association for Earthquake Engineering.
.ymssp.2016.05.027. Xu, Y. L., X. H. Zhang, S. Zhu, and S. Zhan. 2016. “Multi-type sensor
Lu, X., X. Lu, H. Sezen, and L. Ye. 2014. “Development of a simplified placement and response reconstruction for structural health monitoring
model and seismic energy dissipation in a super-tall building.” Eng. of long-span suspension bridges.” Sci. Bull. 61 (4): 313–329. https://doi
Struct. 67: 109–122. https://doi.org/10.1016/j.engstruct.2014.02.017. .org/10.1007/s11434-016-1000-7.
Mangalathu, S. 2017. “Performance based grouping and fragility analysis Zhu, S., X. H. Zhang, Y. L. Xu, and S. Zhan. 2013. “Multi-type sensor
of box girder bridges in California.” Ph.D. thesis, Georgia Institute of placement for multi-scale response reconstruction.” Adv. Struct. Eng.
Technology. 16 (10): 1779–1797. https://doi.org/10.1260/1369-4332.16.10.1779.

© ASCE 04019038-15 J. Struct. Eng.

J. Struct. Eng., 2019, 145(6): 04019038

You might also like