Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Soil Dynamics and Earthquake Engineering 42 (2012) 219–228

Contents lists available at SciVerse ScienceDirect

Soil Dynamics and Earthquake Engineering


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

NEES integrated seismic risk assessment framework (NISRAF)


Sheng-Lin Lin n,1, Jian Li 2, Amr S. Elnashai 3, Billie F. Spencer, Jr. 4
Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA

a r t i c l e i n f o abstract

Article history: The paper presents an integrated and extensible framework for assessment of the impact of earth-
Received 17 March 2011 quakes on civil infrastructure systems, particularly buildings and bridges. The framework, referred to as
Received in revised form NEES Integrated Seismic Risk Assessment Framework (NISRAF), is developed with a focus on improving
23 May 2012
the reliability of earthquake assessment results. The components are structural fragility assessment
Accepted 4 June 2012
Available online 10 July 2012
using measured data and hybrid simulation, hazard characterization by free-field site response analysis,
and integrated impact assessment. The hazard and fragility components are refined by employing
nonlinear site response tools and model updating techniques, respectively. Several of these NISRAF
components are tailored to achieve seamless integration and to arrive at an operational system. The
novelty of the developed framework is primarily the integration of the various components of
earthquake impact assessment, which have not been deployed in such an application before. The
framework has been built and demonstrated via applications to a test bed in California. Earthquake
impact assessment results using the generated hazard map and fragility curves correlate well with field
reports, indicating the efficacy of the approach.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction uncertainties remain in their outcomes, not only because of their


inherent characteristics, but also the interactions between them.
Earthquakes can inflict severe damage when they hit urban For example, the derivation of fragility curves often requires that
areas. For instance, the 2008 Wenchuan earthquake caused a large number of simulations be performed. An accurate struc-
thousands of deaths and over $150 billion in economic losses tural model which closely represents the response of the real
[1], and the 2010 Haiti earthquake killed more than a quarter of structure is essential to derive high-fidelity fragility curves. For
million people and resulted in losses over $14 billion [2]. The level most approaches to fragility curve generation, either a simplified
of the devastation resulting from the recent Tohoku earthquake is structural model or a complicated numerical model is used
already enormous and still mounting. usually without being calibrated to measured response. Such
Since the 1960s, practitioners and researchers – through field methods, therefore, introduce significant and by-and-large
investigations after major earthquakes, along with theoretical and unquantifiable uncertainties in the derived fragility curves. More-
experimental studies – have significantly improved our under- over, the fragility curves heavily depend on input ground motions,
standing of earthquakes and strategies to mitigate their particularly when they are defined in terms of peak ground
impact. Examples of disciplinary developments promoting this acceleration (PGA) [3]. The ground motion in turn is influenced
understanding are: strong-motion measurements, system identi- by source, path, and site characterization, each of which is a
fication, model updating, structural performance evaluation formidable challenge in its own right. The realism of both model
through experimental and analytical simulations, and fragility and input is therefore essential to the accuracy and applicability
derivation, all of which have contributed to the development of of the ensuing fragility relationships.
today’s approaches to earthquake impact assessment. Earthquake impact assessment is the basis for emergency
The above-mentioned component studies have progressed planning, mitigation, response, and recovery. The assessed impact
considerably and produced mature research results; however, on civil infrastructure systems, national economies, and societal
activities, provides the foundation for developing plans that
adequately protect vulnerable communities. As mentioned pre-
n
Corresponding author. viously, significant progress has been made in earthquake impact
E-mail address: shenglin0110@gmail.com (S.-L. Lin). assessment, including consequence estimation methodology and
1
Research Associate.
2 development of software that provides decision-makers with a
Doctoral Candidate.
3
Bill and Elaine Hall Endowed Professor. tool to assess the impact [4]. Generally, such impact assessment
4
Nathan M. and Anne M. Newmark Endowed Chair in Civil Engineering. software is composed of three main components: (i) hazard, (ii)

0267-7261/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.soildyn.2012.06.005
220 S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228

fragility, and (iii) inventory. Among these, the inventory can be


improved through the development and application of survey
methods and technologies to reduce the associated uncertainty.
In contrast, unquantifiable uncertainty and inaccuracies in the
former two components (i.e., hazard and fragility) can result in
unreliable earthquake impact assessments, which cannot form a
viable basis for improving societal readiness.
This paper will present an integrated and extensible frame-
work, referred to as the NEES Integrated Seismic Risk Assessment
Framework (NISRAF), which is intended to reduce the above-
mentioned uncertainties and unreliability in earthquake impact
assessment. In subsequent sections, the development and imple-
mentation of this framework, as well as its components, will be
discussed. An example using a test bed in California will be
presented as well to illustrate the efficacy of the approach.

2. NEES integrated seismic risk assessment framework


(NISRAF)

The schematic representation and architecture given in


Figs. 1 and 2 illustrate the proposed framework and how its
components are combined to achieve the main goal of this
research. Free-field measurements (I1) along with nonlinear site
response analysis (SR) are used to generate the hazard map and
Fig. 2. Architecture of the proposed integrated framework.
ground motion records (AH). The measured and synthetic records
are then used as seismic inputs in hybrid simulation and fragility
analysis. Meanwhile, the structural model is calibrated with the
measured structural response (I2). Next, hybrid simulation (HS) is more accurate due to calibration with system identification
performed with the most critical component of the structural results from sensor measurements. As a result, the uncertainties
system tested in the laboratory and the remainder of the resulting from deriving fragility relationships can be greatly
structure simulated analytically. These simulations are conducted reduced through the use of more reliable representation of hazard
to derive the mean seismic intensity value (PGA, for example) of and more accurate structural models. Confident with seismic
the corresponding performance limit state. The fragility curves hazard from field measurements and fragility curves from more
(FA) of the structure are then generated using the hybrid simula- accurate models, NISRAF can improve upon earthquake impact
tion data and the dispersions from the literature. Finally, the assessment results with higher reliability. In the rest of this
derived fragility curves and hazard map are fed into an impact section, components, development, and implementation of
assessment tools, such as MAEviz [5] (IA) to evaluate the seismic NISRAF will be discussed.
losses.
Utilizing the proposed integrated framework, conducting loss 2.1. NISRAF components
assessment is more reliable, and opportunity for reducing the
uncertainty in these analyses is more straightforward. For exam- As discussed previously, the proposed framework integrates
ple, the reliability of probabilistic seismic hazard can be improved hybrid simulation with free-field and structure sensor measure-
through the use of free-field strong-motion measurements. Ana- ments, hazard characterization analysis, system identification-
lytical and hybrid (analytical–experimental) simulations can be based model updating technology, hybrid fragility analysis, and
earthquake impact assessment tools. To achieve a seamless
integration and to arrive at an operational system, these compo-
nents were refined and tailored to mesh together NISRAF. Below,
methodologies and technologies utilized in these components are
described.

2.1.1. Integrated hazard analysis


Owing to uncertainties from seismotectonics, earthquake
energy attenuation, and site conditions, accurately characterizing
the hazard is difficult. Many tools for seismic hazard analysis have
been developed over the past decades. Among these, Determinis-
tic Seismic Hazard Analysis (DSHA) and Probabilistic Seismic
Hazard Analysis (PSHA) are the most commonly used methods,
and both are generally implemented within earthquake impact
assessment packages [4]. Due to the probabilistic nature and
simplified assumptions for the local site effect, such as the use of
site coefficients Fa and Fv [6], uncertainties remain in the proce-
dures and results.
To reduce these uncertainties, an integrated hazard character-
Fig. 1. Schematic of the proposed integrated risk assessment framework. ization analysis approach, including site response analyses, is
S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228 221

proposed. First, the available natural records in the vicinity of the 2.1.3. System identification
site in question are directly investigated to evaluate the hazard Among the various system identification methods, the Eigen-
characterization. Synthetic records of different hazard levels are system Realization Algorithm (ERA) [10] is implemented in
then generated for further use in hybrid simulation and fragility NISRAF due to its wide application and good performance in
curve derivation. Among various artificial motion generation multi-input multi-output (MIMO) problems. The basic idea of ERA
methods, SIMQKE [7] is implemented in NISRAF due to its is to find a minimum realization of the system (a state-space
efficiency of computation and wide application since its release. representation with minimum dimensions) using the Singular
The basic concept of SIMQKE is iteratively to smooth the calcu- Value Decomposition (SVD) on the Hankel matrix built by the
lated spectrum and improve its matching with the specified Markov parameters (impulse response functions), so that the
spectrum. The calculated spectrum here is derived from a density modal properties can be extracted from the realized minimum
function based on the defined duration and response spectrum, state-space representation. First, for a system with the numbers
and modified via peak ground acceleration and intensity function. of input and output equal to p and q, respectively, the impulse
Procedures for generating synthetic ground motions include: response function Y is obtained by performing Inverse Fast Four-
(i) seismic parameters definition, (ii) response spectra and dura- ier Transform (IFFT) with transfer functions. Then, the following
tion specification, and (iii) intensity function definition. Either Hankel matrix is constructed:
user-supplied or ground motion prediction equations-based spec- 2 3
YðkÞ Yðk þ 1Þ Yðk þ2Þ    Yðk þ s1Þ
tra (the Next Generation Attenuation (NGA) models [8], for 6 Yðk þ1Þ 7
6 Yðk þ 2Þ Yðk þ3Þ    Yðk þ sÞ 7
example) are acceptable in NISRAF. By providing an intuitive 6 7
interface, NISRAF collects the above required information and H rs ðkÞ ¼ 6
6 Yðk þ2Þ Yðk þ 3Þ Yðk þ4Þ    Yðk þ s þ1Þ 7:
7
6^ ^ ^ & ^ 7
executes SIMQKE to generate motions on the bedrock. Finally, in 4 5
order to capture the complicated soil behavior, nonlinear site Yðk þr1Þ Yðk þ rÞ Yðk þr þ 1Þ    Yðk þ r þs2Þ
response analysis is incorporated in NISRAF. Currently, DEEPSOIL ð1Þ
[9], a 1-D site response analysis, is conducted in the console mode
to convolute the bedrock motions to the surface. The basic In which Y(k) is the impulse response function at time k, a
concept of DEEPSOIL is to perform nonlinear dynamic analyses block matrix consisting of p rows and q columns; r and s are the
on the defined soil column using the bedrock motions as input. number of block rows and block columns; Singular value decom-
These generated site specific synthetic motions are compatible to position is then carried out for Hrs(k), which yields:
be used with other components, such as hybrid simulation and Hrs ð0Þ ¼ P D Q T : ð2Þ
fragility analysis.
In addition to the seismic hazard analysis, synthetic ground where P and Q are left and right singular vectors. D is an (r  p) by
motion generation, and site response analysis, this hazard analy- (s  q) diagonal matrix with singular values on the diagonal.
sis component also provides a function to generate hazard map, If Y(k) is noise free, the number of non-zero singular values is
which is the exposure of the earthquake impact assessment. the dimension of the minimum realization of the system. How-
Similar to the procedures for synthetic ground motion generation, ever, due to noise in reality, D usually has full rank. The singular
seismic information is specified in the beginning. Additional values associated with noise are usually much smaller compared
information (e.g., the interested area and the cell size used in with those corresponding to real modes. By preserving the first N
map) particular to hazard map generation is then needed to be biggest singular values, by which noise modes are eliminated, the
defined. Afterward, the seismic intensity, such as PGA at the minimum realization of the state space system can be calculated as
center of each cell is calculated. Finally, the hazard map with a
A ¼ D-1=2 P T Hrs ð1ÞQ D-1=2
compatible format is produced.
The proposed hazard analysis method integrates seismic B ¼ D1=2 Q Eq
hazard analysis, synthetic ground motions, site response analysis, C ¼ ETp PD1=2 : ð3Þ
and hazard map generation. The use of instrumentation data
provides an opportunity to calibrate the hazard models, which, In the above equations, ETp ¼ ½Ip 0p    0p  and ETq ¼ ½Iq 0q    0q .
therefore, improve the reliability of the hazard characterization The above state space matrices are then transformed into modal
used in the further analyses, such as synthetic ground motion and coordinates by using the eigenvalues Z and the eigenvector
hazard map generation. Additionally, the incorporation of site matrix of A, which yields:
response analysis improves the accuracy in synthetic ground A0 ¼ C1 AC ¼ Z
motions, as well as in hazard maps. Moreover, the proposed
B0 ¼ C1 B ð4Þ
method simplifies previous tedious and complicated procedures 0
in each hazard model, solves the compatibility between them, C ¼ C C:
and provides an interactive interface for ease of use. Mode shapes are the corresponding columns of C0 . The modal
damping ratio, xi and damped natural frequencies,odi can be
calculated as
2.1.2. Model calibration
Although finite element (FE) model simulation provides a si ¼ xi oni 7iodi ¼ lnðzi Þ=Dt: ð5Þ
powerful way to understand the response of structures, well- qffiffiffiffiffiffiffiffiffiffiffiffi
2
constructed models may produce significant differences in some where oni ¼ odi = 1zi are undamped natural frequencies, and
dynamic response predictions, particularly when the structure Dt is the sampling interval.
behaves nonlinearly. This difference is due to uncertainties in the As mentioned previously, measurements are always contami-
materials, boundary conditions, and the contribution of the non- nated by noise in practice, resulting in nonzero values for all
structural elements in a structure. To overcome this drawback, a singular values in Eq. (2). Cutting of small singular values based
model calibration method – composed of system identification on judgment does not guarantee reliable identification results.
and model updating techniques – is proposed based on measured Therefore, the concept of stabilization diagram is employed to
laboratory or full-scale structural responses. The following para- more effectively filter out noise modes, based on an idea that
graphs provide an overview of these two techniques. genuine modes should always exist in the system when the
222 S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228

system order is increased. In NISRAF, the natural frequencies are developed through field investigations after earthquakes are the
identified for systems with different orders, by varying the most realistic. However, this observation data is scarce and
number of retained singular values obtained in Eq. (2). For each clustered in the low damaged range. Judgmental fragility curves
particular order of the system, three mode accuracy indicators, are based on expert opinion and are therefore subjective. Unlike
namely, Modal Amplitude Coherence (MAC) [10], Extended Modal the empirical and judgmental fragility curves, analytical fragility
Amplitude Coherence (EMAC), and Modal Phase Colinearity (MPC) curves are more general; curves are allowed to be generated for
[11] are used to eliminate the noise modes. The retained modes different limit states and different structural types, although at a
are plotted in a stabilization diagram. A mode which is identified higher computation cost. Due to this limitation, most analytical
for at least five times is considered stable and reliable. Among fragility curves are generated either by simplified models or by
them, the one with highest EMAC value is then selected as the complicated models without calibration to the real structural
confirmed mode. response, which can result in uncertainties in these curves.
To reduce uncertainties, a hybrid fragility analysis method is
2.1.3.1. Model updating. Model updating aims to minimize the proposed, as shown in Fig. 3. In this hybrid approach, hybrid
discrepancies between the numerical and real model by adjusting simulation, with critical element tested in the laboratory and the
the stiffness and mass matrices. The objective function is formed rest simulated in the calibrated finite element model, is con-
as a linear combination of the natural frequency residuals and ducted to evaluate the structural response. By scaling ground
mode shape residuals, with different weighting factors for each motions, several hybrid simulation tests are performed to reach
residual. the target structural response. The PGA value of the associated
pffiffiffiffiffiffiffiffiffiffiffiffiffi ground motion is then assigned as the mean PGA value for the
Nf  2
X f ak f ek XNm
cos1 ð MAC k Þ current limit state. Here, the target structural response is defined
FðxÞ ¼ wf þ wm : ð6Þ
k¼1
f ek k¼1
ðp=2Þ for different limit states, such as interstory drift angle of 0.7% for
immediate occupancy limit state for steel moment frame build-
fak and fek denote the analytical and experimental natural
ing. Finally, with the mean PGA values and the dispersions from
frequencies; wf and wm are weighting factors applied to the
similar structures found in the literature, the fragility curves are
frequency residuals and mode shape residuals, respectively.
generated based on the lognormal distribution assumption [15].
MAC (Modal Assurance Criteria) is a measurement of mode
Dispersions suggested by published research papers are uti-
shape discrepancy and is defined as [12]:
lized to account for the uncertainty term in the proposed hybrid
T fragility derivation. Alternatively, dispersions based on analytical
ðfai fei Þ2
MAC i ¼ T T
: ð7Þ analysis with varying material properties, input motions, and
ðfai fai Þðfei fei Þ
others can be employed when experimental results are absent or
fai and fei are analytical and experimental mode shapes, insufficient. Furthermore, it is recommended that with the
respectively. MAC¼1 means fai and fei are perfectly matched; increase in experimental results, fragility relationships can con-
MAC¼0 means they are orthogonal. However, MAC is the square tinue being updated and improved using Bayesian methods or
of the inner product between the two mode shape vectors, and is other statistic approaches.
known to be insensitive to the change of mode shapes. Therefore, The proposed hybrid fragility analysis method, including the
the objective function for the mode shape residual is formed as incorporation of hybrid simulation into fragility analysis and
the normalized angle between the two mode shape vectors, as uncertainty consideration can be used to derive more reliable
shown in Eq. (6), which are much more sensitive to the changes of fragilities than those based on analysis alone. The incorporation of
mode shapes [13].
This proposed model calibration component takes advantage
of the finite element modeling capability of NISRAF, which allows
a structural model to be built within NISRAF, including geometry,
structural elements, nodal connectivity, material properties, etc.
Therefore, the candidate parameters are readily available and can
be selected as needed. The objective function is customized by
including different number of modes and different weighting
factors for frequencies and mode shapes. Sensitivity analysis is
conducted to investigate the appropriateness of the selected
parameters which can be adjusted accordingly if needed. Finally,
model updating result, including the changes of objective func-
tion, natural frequencies, mode shapes and the selected para-
meters, is summarized and displayed in tables. If satisfactory
result is obtained, NISRAF can accept it so that changes of
parameters will be made to the model. Otherwise model updating
process can be repeated with different setup, such as optimization
algorithm, objective function and candidate parameters, etc.

2.1.4. Hybrid fragility analysis


Fragility, or vulnerability, presents the probability of reaching
or exceeding a specific performance level under a specific seismic
hazard. Fragility curves relate the effects of seismic hazard to the
damage of the structures. Through the application of fragility
curves, losses from earthquake hazard can be estimated.
Generally, fragility curves are sub-divided into four categories
based on data sources, namely, empirical, judgmental, analytical,
and hybrid fragility curves [14]. Empirical fragility curves Fig. 3. Methodology and procedures for the hybrid fragility analysis.
S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228 223

hybrid simulation test provides a chance to capture real structural Instrumentation Program (CSMIP) in 1980 with 13 sensor chan-
responses, which therefore improves the reliability of the fragility nels. Several significant earthquakes were captured, including the
relationship. Meanwhile, a comprehensive consideration of 1987 Whittier earthquake, the 1991 Sierra Madre earthquake and
uncertainties increases the confidence when using the generated
fragility relationships.
1.4
In this section, refined approaches utilized in NISRAF compo-
1994 Northridge EQ
nents were discussed. All the components are developed separately Avg. GMs W/O SR
(modularized), which facilitates understanding and maintenance. 1.2 Avg. GMs W/ SR
Therefore, the methodologies and techniques discussed in this

Spectral Acceleration (g)


section can be updated or replaced with any latest research findings 1
and program techniques. With its accessible and extensible features,
NISRAF provides an integrated platform, which contributes to the
0.8
achievement of societal requirement for accurate evaluation of the
impact of earthquakes on the built environment.
0.6

2.2. Development and implementation of NISRAF 0.4

NISRAF was developed using MATLAB [16], combining the


0.2
above-proposed methodologies. Several components – instrumenta-
tion, hazard characterization, system identification, model updating,
hybrid simulation, hybrid fragility analysis and impact assessment 0
10-2 10-1 100 101
tools – have been tailored to build the framework and the commu-
nication between them. A pilot implementation of NISRAF using an Period (sec)
instrumented building in Burbank, California has been conducted to Fig. 5. Comparison of response spectra with and without site response analysis (SR).
demonstrate the feasibility of NISRAF. Other applications, such as
the regional impact assessment for Los Angeles area, the evaluation
of the seismic behavior of the Meloland Road Overcrossing (MRO) Map of PGA (g) for Northridge, CA, 1994 event
Bridge in California, and the seismic assessment for Abbottabad city 0.32
in Pakistan, are conducted currently to further demonstrate the 0.3
capabilities of NISRAF. NISRAF has been published and can be 34.3
downloaded from NEES website (/www.nees.orgS). Reference can 0.28
be made to Lin [17] for more detailed information about the 34.2
0.26
development and implementation of NISRAF. In the following
section, the pilot implementation on the Burbank building, which 34.1 0.24
Latitude

was utilized to demonstrate NISRAF from instrumentation, testing,


0.22
to loss assessment, will be described in detail. 34
0.2
33.9
0.18
3. Demonstration of NISRAF
33.8 0.16
An instrumented building in Burbank, California was selected
to illustrate the use of NISRAF using the 1994 Northridge earth- 33.7 0.14
quake as the scenario event. This building is a 6-story steel 0.12
moment resisting frame structure, in which the perimeter frames -119.4 -119.2 -119 -118.8 -118.6 -118.4 -118.2 -118
are the primary lateral load resisting system and the internal Longitude
frames only resist the gravity load, as shown in Fig. 4a. The
building was instrumented by the California Strong Motion Fig. 6. Hazard map generated through NISRAF.

Fig. 4. (a) Plan view of Burbank building; (b) 2-D analytical model.
224 S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228

the 1994 Northridge earthquake. Anderson and Bertero [18] and steel, lumped mass, and effective width of concrete slab were selected
Fumal et al. [19] provide more detailed information about this as candidate parameters based on the sensitivity analysis. The
building and the site condition, respectively. optimization problem defined previously was solved by the Nelder–
In this section, site response analyses are performed, and hazard Mead method [21]. Table 1 tabulates the structural frequencies of the
characterization and surface ground motion records are generated for identified, original, and updated model. As can be seen, the errors
further use during the hybrid simulations and fragility analyses. between the identified and updated model were reduced to 1% and
Additionally, the finite element model is built, and the natural 5.78% for the first and second natural frequency, respectively. Mean-
frequencies and mode shapes are extracted. The numerical model is while, the second mode shape was improved which gave a value of
updated based on structural measurements through a sensitivity- 0.981 for the MAC. With this refined finite element model, hybrid
based model updating technique. Next, Fragility curves are derived simulation was then conducted to yield a seismic response prediction
using hybrid simulation results along with dispersions from research with higher accuracy.
on similar structures from the literature. Finally, impact assessment
results using the generated hazard map and fragility curves are
compared with field observations following the 1994 Northridge 3.3. Hybrid fragility analysis
earthquake.
The calibrated Burbank building model and ground motions from
hazard characterization analysis were used to demonstrate the
3.1. Hazard characterization extension of the hybrid simulation to fragility analysis as well as
the integration of hybrid simulation in earthquake impact assess-
One purpose of the proposed integrated hazard analysis is to ment. The calibrated 2-D structure model was divided into two sub-
improve the estimation of hazard. Defining seismic hazard from structures, namely, the column (the lower part of the left exterior
field measurements in the region of study improves the reliability column at the first floor) and the frame (the remaining structure). The
of hazard map as well as synthetic ground motions. Additionally, frame module was simulated using ZEUS-NL [22], while the column
the incorporation of site response analysis improves the accuracy module – replaced by a small scale aluminum specimen (Fig. 7) – was
both in synthetic ground motions and hazard maps. Fig. 5 shows tested in the laboratory. The tests were conducted via UI-SimCor [23],
the comparison of the response spectra between the 1994 North- the University of Illinois software platform for hybrid simulation.
ridge earthquake record and the average of synthetic records with Fig. 8a shows the laboratory experimental setup using the
and without site response analysis at the Burbank site. As can be NEES@UIUC (MUST-SIM) facility. The small-scale specimens were
seen the generated ground motions, including site response designed with careful consideration of similitude relationships
analysis, can capture the actual hazard characterization. [24,25]. Hybrid simulation test using the 1994 Northridge earth-
With the above demonstration, the hazard map under the quake record as input was first conducted as the proof test prior
Northridge earthquake and synthetic ground motions for different to the generation of the hybrid fragility curves. Good agreement,
hazard levels were generated using the proposed hazard compo- as shown in Fig. 8b, proves that the hybrid structural model
nent. With the specification of seismic mechanism and soil composed of the calibrated analytical part and the small-scale
condition, hazard map of PGA for the 1994 Northridge earthquake
in the Burbank area in terms of standard gravity (g) was
generated, as shown in Fig. 6. When generating synthetic ground
motions for different hazard levels (i.e., 10%, 5%, and 2% prob-
ability of exceedance in 50 years), the deaggregation results were
fed into the Campbell and Bozorgnia NGA model [20]. Response
spectra and duration based on the selected attenuation model and
seismic parameters, respectively, were then used to generate
synthetic ground motions. Subsequently, nonlinear site response
analysis was used to generate site specific ground motions.

3.2. Model calibration

Due to the fact that only the perimeter frames are part of the
lateral load resisting system, a 2-D frame was built to represent the
entire structure, as shown in Fig. 4b. Next, the ERA method was
performed on structural measurements during the 1994 Northridge
earthquake. The first and second bending modal frequencies were
identified as 0.72 Hz and 2.14 Hz, respectively. The corresponding
damping ratios were 3.37% and 6.71%. Based on the identified modal
information, model updating was performed. Young’s modulus of Fig. 7. Hybrid simulation with two sub-structures (column and frame).

Table 1
Identified, original and updated model information.

Mode number Identified with ERA Original finite element model Updated finite element model

Freq. (Hz) x (%) Frequency Frequency


Value (Hz) Error (%) MAC Value (Hz) Error (%) MAC

1 0.72 3.37 0.688  4.31 0.999 0.712  1.00 0.999


2 2.14 6.71 1.956  8.77 0.975 2.020  5.78 0.981
S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228 225

Roof Drift History


5
Measured
4 Hybrid Simulation
3

dispacement (in)
2
1
0
-1
-2
-3
-4
-5
0 5 10 15 20 25 30
Time (sec)

Fig. 8. (a) Hybrid simulation setup; (b) Proof test result.

In this section, fragility relationships for a steel moment


Table 2
Interstory drift angle, dispersion and mean PGA value of different performance
resisting frame building in California were developed, which
levels. demonstrates the methodologies and procedures of the proposed
hybrid fragility analysis method, including the incorporation of
Performance Immediate Life Collapse hybrid simulation into fragility analysis and uncertainty consid-
level occupancy safety prevention
eration. The incorporation of model calibration and hybrid simu-
Interstory drift 0.7 2.5 5.0 lation test provides a chance to capture real structural responses,
angle (%)* which therefore improves the reliability of the fragility relation-
Dispersiona 0.311 0.328 0.346 ship. However, most existing buildings are neither instrumented
Mean PGA (g)b 0.545 1.627 2.777 nor tested in the laboratory. When conducting loss assessment on
n
Ref. [26].
these buildings, NISRAF has the capability to ingest existing
a
Refs. [27,28]. fragility relationships and continue the following loss estimation.
b
Hybrid simulation results. In the following, another set of fragility relationships of the
investigated Burbank building were developed via the conven-
part is representative. With this confidence in the hybrid model tional approach. In this approach, fragility curves were generated
and experimental setting, hybrid simulation tests under different using nonlinear time history analyses based on the untuned 2-D
synthetic ground motions (10%, 5%, 2% probability of exceedance finite element model. Latin Hypercube Sampling technique was
in 50 years for immediate occupancy (IO), life safety (LS) and used to generate 10 building models with different material
collapse prevention (CP) performance level, respectively) were properties. Concrete and steel strength were investigated; both
used to derive the relevant mean PGA values. Step-by-step were assumed as normal distribution with mean and dispersion
procedures shown above in Fig. 3 are detailed below. First, the values based on design documents and previous studies [29,30],
ground motion of 10% probability of exceedance in 50 years was respectively. Site specific synthetic ground motions generated via
selected as the seismic input for hybrid simulation test to derive NISRAF hazard component were included in the ground motion
the mean PGA value for IO limit state. After performing hybrid database. During the analytical simulation, for each building
simulation test, interstory drift angle (ISDA) was calculated based sample, the ground motion is randomly selected, and then scaled
on test results. Comparison of ISDA between the calculated one from 0.1 g to 1.0 g with 0.1 g increment. Therefore, a total of 100
and the target one (0.7% ISDA for IO performance limit state [26], GM-building samples were evaluated and the results were then
as listed in Table 2) was then made. Next, Hybrid simulation test regressed to get the fragility relationships. A comparison between
was resumed (replaced with new specimen if nonlinear behavior the fragility curves from the integrated NISRAF and conventional
occurs in previous test) with seismic input multiplied by a scale approach is presented in Fig. 10. As shown in the figure, the
factor, if the ISDA difference exceeded criterion ( 75%, for exam- fragility curves from conventional approach are more vulnerable
ple). The scale factor here was the target ISDA divided by the and have larger uncertainties.
calculated ISDA. Iterations of above procedures continued till the
criterion was met. Once the calculated ISDA matched the target 3.4. Loss assessment
ISDA, PGA value of current (scaled) record was assigned as the
mean PGA value for IO limit state. Similar procedures were Finally, the generated hazard map and fragility curves (both
applied to derive the mean PGA values for LS and CP limit states NISRAF and conventional sets) were ingested into impact assess-
using ground motions of 5% and 2% probability of exceedance in ment tools to calculate seismic losses. MAEviz, an earthquake
50 years, respectively. Fig. 9a shows the number of hybrid consequences assessment package, which follows the Conse-
simulation tests conducted in order to derive the mean PGA quence-based Risk Management (CRM) paradigm, developed by
values. Using these mean PGA values, along with dispersions from the MAE Center and the National Center for Supercomputing
previous studies for similar structures (Table 2), fragility curves Applications (NCSA) at the University of Illinois, was utilized in
were generated based on lognormal distribution assumption this application. Using the generated hazard map and fragility
(Fig. 9b). curves, along with inventory information (the structural type,
In this study, due to the limited number of tests, dispersions material, and others), MAEviz calculated structural damage on the
(including different building height, beam-column connection Burbank building due to the 1994 Northridge earthquake.
type, analysis procedure, and local and global failures) suggested To further evaluate the results, another assessment using MAEviz
by published research papers [27,28] were utilized to account for deterministic hazard and default fragility curves was also con-
the uncertainty term when deriving fragility relationships. ducted. Table 3 lists the probability of exceeding the three defined
226 S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228

Probability of Limit State Exceedance


1
0.05
0.8
Interstory Drift Angle

0.6

0.025
0.4

Immediate Occupancy 0.2 Immediate Occupancy


0.007 Life Safety Life Safety
Collapse Prevention Collapse Prevention
0
0 1 2 3 4 5 0 0.5 1 1.5 2 2.5 3
Number of Hybrid Simulation PGA (g)

Fig. 9. (a)Number of hybrid simulation tests; (b) Fragility curves generated by NISRAF.

Table 4
1 Probability of four structural damage states.
Probability of Limit State Exceedance

cy
pan

Insignificant (%) Moderate (%) Heavy (%) Complete (%)


cc u

0.8
O

NISRAF 85 15 0 0
ate

Conventional 60 39 1 0
e di

MAEviz default 62 37 1 0
Imm

0.6
ty
fe
Sa

Discrete probabilities of damage states may then be computed


fe

0.4 by Eq. (9) in order to calculate building structural damage. Note


Li

ti on
en that the definition damage states implemented in NISRAF are
ev
e Pr different from those defined in the FEMA document. Light,
0.2 ps
lla NISRAF moderate, and severe are used to describe the overall damage
Co
Conventional for IO, LS, and CP performance levels, as tabulated in Table C1–2 of
MAEviz
FEMA-356.
0
0 0.5 1 1.5 2 2.5 3
PðDS ¼ IÞ ¼ 1PðDS 4IÞ ¼ 1PðLSIO Þ
PGA (g)
PðDS ¼ MÞ ¼ PðDS 4 IÞPðDS 4MÞ ¼ PðLSIO ÞPðLSLS Þ
Fig. 10. Comparison of fragility curves via different approaches.
PðDS ¼ HÞ ¼ PðDS 4 MÞPðDS 4HÞ ¼ PðLSLS ÞPðLSCP Þ

Table 3 PðDS ¼ CÞ ¼ PðDS 4 HÞ ¼ PðLSCP Þ: ð9Þ


Probability of exceeding three performance limit states.
Table 4 tabulates the estimated probability of four structural
Immediate Life safety Collapse damage states for three different cases. As can be seen, assessment
occupancy (%) (%) prevention (%) results through NISRAF portray less damage than the results from
NISRAF 15 0 0
conventional and MAEviz default cases. Unlike the result from
Conventional 40 1 0 NISRAF, which shows only slight damage, more structural damage
MAEviz default 38 1 0 were estimated both in Conventional and MAEviz default cases.
Table 5 lists the difference in the hazard and fragility components
utilized in loss assessment among these three cases, which fully
explains the varying results. First of all, while under the same hazard
performance limit states for three different cases under the input, the Conventional case predicted more damage than NISRAF
generated hazard (based on the 1994 Northridge earthquake due to its more vulnerable fragility curve, as has been shown in
event). previous section. On the other hand, higher hazard was predicted in
As described previously, the thresholds for the three perfor- the NISRAF and Conventional cases (PGA¼0.4 g) than the MAEviz
mance limit states (i.e., IO, LS, and CP) defined in FEMA-356: default one (PGA¼0.23 g) since the former included site response
Prestandard and Commentary for the Seismic Rehabilitation of analysis. However, the more vulnerable fragility curves in the MAEviz
Buildings [26] were utilized to calculate the probabilities of default case, as shown in Fig. 10 resulted in more damage than the
exceeding these limit states. Based on the usage of three limit NISRAF case. The MAEviz default fragility curves are specific to
states, consistently four structural damage states (DS): insignif- buildings in Central and East United States; therefore, they represent
icant (I), moderate (M), heavy (H), and complete (C) are defined in higher vulnerability than west coast curves (the NISRAF and Conven-
MAEviz [31], therefore tional case). A post-earthquake report made by Applied Technology
Council (ATC) [32] claimed that insignificant damage was observed to
PðDS 4 IÞ ¼ PðLSIO Þ this building from the Northridge earthquake, as shown in Table 6.
PðDS 4 MÞ ¼ PðLSLS Þ Therefore, the structural damage (85% insignificant and 15% moder-
PðDS 4 HÞ ¼ PðLSCP Þ: ð8Þ ate) estimated via NISRAF correlated well with the field observation.
S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228 227

Table 5
Comparison of analysis components among NISRAF, Conventional and MAEviz default.

NISRAF Conventional MAEviz default

Hazard Deterministic hazard Deterministic hazard


(i) Northridge earthquake mechanism (i) Northridge earthquake mechanism
(ii) Campbell and Bozorgnia NGA (ii) Campbell and Bozorgina NGA
(iii) 1-D site response analysis (iii) Fa and Fv site coefficient

Fragility Hybrid fragility Analytical fragility Analytical fragility

(i) Calibrated finite element model (i) Untuned 2-D finite element model (i) Single degree of freedom Model
(ii) Hybrid simulation using California (ii) Nonlinear time history analysis using California (ii) Nonlinear time history analysis using CEUS
synthetic ground Motions synthetic ground motions synthetic ground motions

Table 6
ATC-38 on-site report after Northridge earthquake, 1994.

Building ID number Model building Number of stories Design date General damage Structural damage Nonstructural
type state state damage state

y y y y y y y
CDMG370-MF-09 S1 6 1975 I 1 2
y y y y y y y

I ¼Insignificant, M ¼Moderate.
1¼ None (0% damage), 2¼ Slight (0–1%damage), 3 ¼Light (1–10% damage).
ATC-38, Table A-6 Building Damage Summary for Station CDMG 24370.

In addition to structural damage, loss assessment in NISRAF demonstrated not only the successful development of the integrated
also includes the estimation of nonstructural damage, economic framework for loss assessment, but also the achievement of the
cost, retrofit cost, and others. However, because of the lack of cost objectives of this study. NISRAF therefore contributes to the state-of-
information of the investigated building as well as no monetary the-art in using hybrid simulation in probabilistic fragility analysis,
information recorded in field reports, only structural damage was and proposes a refined method for hazard characterization from free-
presented and compared in the study. field measurement. The good agreement between loss assessment via
In this section, structural damage on the test bed in California, NISRAF and the post-earthquake observation approved the improve-
a 6-story steel building, was carried out to demonstrate the ments in the proposed hazard and fragility analysis using free-field
framework and its components. The building example demon- measurement and hybrid simulation with calibration, respectively.
strated not only the seamlessly-integrated, extensible, and trans- The demonstration utilizes a steel building which underwent slight
parent framework, but also showed that all the elements required damage during the Northridge earthquake. However, more applica-
for an more accurate impact assessment can be performed in a tions are being conducted to further demonstrate the feasibility of
unique software platform. Consequently, the impact assessment NISRAF, such as regional impact assessment for Los Angeles County
results showed good agreement with the post-earthquake report. and Abbottabad city in Pakistan, and seismic evaluation of a bridge in
Meanwhile, it also established that the proposed integrated California. Their results and findings will be published when available.
methods – the hazard characterization analysis, the model cali- In addition, by providing this integrated and unique platform,
bration, and the hybrid fragility analysis – were more reliable. other research is currently underway to extend and improve the
Although NISRAF results achieved good correlation with the filed capabilities of NISRAF, which include (i) providing a selection of
observation as described above, there is still considerable uncer- methodologies to evaluate component and system uncertainty, and
tainty in the procedures for loss assessment. One example is the (ii) incorporating other structural analysis platforms and site response
prediction (including description and thresholds) of building per- analysis tools.
formance limit states. As stated in FEMA-356, the performance The final but the most important effort of this study is that the
description in Table C1–2 is estimates rather than precise predic- first time that the integration of all components of impact
tions. Therefore, discrepancy between estimation and observed assessment has been achieved. With the successful development,
damage must be expected if there is another earthquake in the NISRAF not only provides a tool for linking, integration and ease
same region. Nonetheless, NISRAF provides a platform for refining of use, but also hopes to stimulate the sub-communities of
impact assessments as more data and knowledge become available. researchers (i.e., geotechnical and structural earthquake engi-
neers, impact assessment experts, seismologists, and structural
control researchers) to investigate the problems at the interfaces
4. Concluding remarks between them to pursue the ultimate goal of accurate and reliable
earthquake impact assessment.
NISRAF serves as a user-friendly framework through which impact
assessment can be efficiently and reliably performed by combining
hazard (exposure) and fragility (sensitivity), to provide assessment of Acknowledgments
impact of earthquakes on the built environment. It differs fundamen-
tally from impact assessment software such as HAZUS [33] in that This research was supported by National Science Foundation,
NISRAF is an analysis as well as an integration environment. The pilot Grant Number 0724172 (NEES Simulation Development), Dr. Joy
implementation illustrated in the Demonstration of NISRAF section M. Pauschke, Program Director.
228 S.-L. Lin et al. / Soil Dynamics and Earthquake Engineering 42 (2012) 219–228

References [18] Anderson JC, Bertero VV. Seismic performance of an instrumented six-story
steel building. Report UCB/EERC-91/11. Berkeley, CA: University of California;
1991.
[1] USGS. Magnitude 7.9-Eastern Sichuan, China. /http://earthquake.usgs.gov/
[19] Fumal TE, Gibbs JF, Roth EF In-situ measurements of seismic velocity at 19
earthquakes/eqinthenews/2008/us2008ryan/#summaryS; 2008.
locations in the Los Angeles, California region. SMIP geotechnical Report No.
[2] USGS. Magnitude 7.0-Haiti Region. /http://earthquake.usgs.gov/earthquakes/eq-
131. U.S. Geological Survey; 1979.
inthenews/2010/us2010rja6/#summaryS; 2010.
[20] Campbell KW, Bozorgnia Y NGA. Ground motion model for the geometric
[3] Kwon OS, Elnashai AS. The effect of material and ground motion uncertainty
on the seismic vulnerability curves of RC structure. Engineering Structures mean horizontal component of PGA, PGV, PGD and 5% damped linear elastic
2006;28(2):289–303. response spectra for periods ranging from 0.01 s to 10 s. Earthquake Spectra
[4] Molina S, Lang DH, Lindholm CD SELENA. – An open-source tool for seismic 2008;24(1):139–72.
risk and loss assessment using a logic tree computation procedure. Compu- [21] Nelder JA, Mead R. A simplex method for function minimization. Computer
ters & Geosciences 2010;36:257–69. Journal 1965;7:308–13.
[5] Center MAE. Mid-America earthquake center seismic loss assessment system, [22] Elnashai AS, Papanikolaou V, Lee D. Zeus NL—a system for inelastic analysis
MAEviz v3.1.1. Urbana, IL: University of Illinois at Urbana-Champaign; 2007 of structures. Urbana, IL: Mid-America Earthquake Center, University of
Mid-America Earthquake Center. Illinois at Urbana-Champaign; 2002 Department of Civil and Environmental
[6] FEMA P-750. NEHRP Recommended seismic provisions for new buildings and Engineering.
other structures. Washington, DC: Federal Emergency Management Agency; [23] Kwon OS, Nakata N, Park KS, Elnashai AS, Spencer BJ. User manual and
2009. examples for UI-SIMCOR v2.6. Urbana, IL: University of Illinois at Urbana-
[7] Gasparini DA, Vanmarcke EH Simulated earthquake motions compatible with Champaign; 2007 Department of Civil and Environmental Engineering.
prescribed response spectra. Evaluation of seismic safety of buildings Report [24] Harris HG, Sabnis GM. Structural modeling and experimental techniques.
No.2. Massachusetts Institute of Technology; 1976. Florida: CRC Press; 1999.
[8] Power M, Chiou B, Abrahamson N, Bozorgnia Y, Shantz T, Roblee C. An [25] Holub C. Interaction of variable axial load and shear effects in RC bridges.
overview of the NGA project. Earthquake Spectra 2008;24(1):3–22. Urbana, IL: University of Illinois at Urbana-Champaign; 2009 Department of
[9] Hashash Y, Groholski DR, Phillips CA, Park D DEEPSOIL. V3.5beta, user manual Civil and Environmental Engineering.
and tutorial. Urbana, IL: University of Illinois at Urbana-Champaign; 2009 [26] FEMA-356. Prestandard and commentary for the seismic rehabilitation of
Department of Civil and Environmental Engineering. building. Washington, DC: Federal Emergency Management Agency; 2000.
[10] Juang JN, Pappa RS. An eigensystem realization algorithm for modal para- [27] FEMA-350. Recommended seismic design criteria for new steel moment-
meter identification and model reduction. Journal of Guid Control Dynamics
frame buildings. Washington, DC: Federal Emergency Management Agency;
1985;8(5):620–7.
2000.
[11] Pappa RS, Elliott KB. Consistent-mode indicator for the eigensystem realiza-
[28] Cornell CA, Jalayer F, Hamburger RO, Foutch DA. Probabilistic basis for 2000
tion algorithm. Journal of Guid Control Dynamics 1993;16(5):852–8.
SAC federal emergency management agency steel moment frame guidelines.
[12] Allemang RJ, Brown DL A Correction coefficient for modal vector analysis.
Proceedings of the international modal analysis conference, Orlando, FL, Journal of Structural Engineering 2002;128(4):526–33.
1982. [29] Barlett FM, MacGregor JG. Statistical analysis of the compressive strength of
[13] Jang S, Li J, Spencer, BF. Corrosion estimation of a historic truss bridge using concrete in structures. ACI Materials Journal 1996;93(2):158–68.
model updating. Journal of Bridge Engineering. /http://ascelibrary.org/doi/ [30] Mirza SA, MacGregor JG. Variability of mechanical properties of reinforcing
abs/10.1061/%28ASCE%29BE.1943-5592.0000403S; 2012, in press. bars. Journal of Structural Division 1979;105(ST5).
[14] Rossetto T, Elnashai A. Derivation of vulnerability functions for European- [31] Bai JW, Hueste MD, Gardoni P. Probabilistic assessment of structural damage
type RC structures based on observational data. Engineering Structures due to earthquakes for buildings in Mid-America. Journal of Structural
2003;25(10):1241–63. Engineering 2009;135(10):1155.
[15] Shinozuka M, Feng MQ, Lee J, Naganuma T. Statistical analysis of fragility [32] ATC-38. Database on the performance of structures near strong-motion
curves. Journal of Engineering Mechanics-ASCE 2000;126(12):1224–31. recordings: 1994 Northridge, California, earthquake. Redwood City, CA:
[16] The MathWorks Inc. MATLAB version 7.8.0. Natick, MA; 2009. Applied Technology Council; 2001.
[17] Lin SL. An integrated earthquake impact assessment system. Urbana, IL: [33] Fema HAZUS-MH. Technical manual, FEMA. Washington, DC: Federal Emergency
University of Illinois at Urbana-Champaign; 2010 Department of Civil and Management Agency; 2006.
Environmental Engineering.

You might also like