You are on page 1of 13

Engineering Structures 29 (2007) 3197–3209

www.elsevier.com/locate/engstruct

An analytical framework for seismic fragility analysis of RC


high-rise buildings
Jun Ji, Amr S. Elnashai, Daniel A. Kuchma ∗
Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, IL 61801, USA

Received 20 January 2007; received in revised form 24 July 2007; accepted 30 August 2007
Available online 13 November 2007

Abstract

An analytical framework and sample application are presented for the seismic fragility assessment of reinforced-concrete high-rise buildings.
Since no probabilistic fragility relationships exist for this class of structure, the work fills an important void in regional earthquake impact
assessment. The key element of the presented framework is the methodology for the development of a simple lumped-parameter model
representative of the complex high-rise building system. This model was created in the ZEUS–NL environment to enable computationally efficient
dynamic response history analyses of high-rise structures that were previously not possible and that can accurately account for the complex
behaviour and interactions predicted by more detailed analytical models. The parameters for this model were selected using genetic algorithms.
The development of a simple lumped-parameter model is presented for an existing high-rise structure with dual core walls and a reinforced
concrete frame. The accuracy of the individual components of this model is compared with the predictions of more detailed analytical models and
sample fragility curves are presented. The proposed framework is generally applicable for developing fragility relationships for high-rise building
structures with frames and cores or walls.
c 2007 Elsevier Ltd. All rights reserved.

Keywords: RC high-rise buildings; Seismic fragility; Lumped modelling; Genetic algorithm

1. Background and motivation from seismic action are often not separately considered. This
is often inappropriate as the wide frequency content in real
Urbanization has led to a dramatic increase in the ground motions can significantly excite both lower and higher
number and variety of high-rise building structures. The modes to a degree that is not excited by wind forces. This
seismic vulnerability of these high-rise infrastructures is
results in a complex seismic response and leads to demands
poorly understood and probabilistic assessment tools of their
in some parts of the structure that can be significantly larger
performance is lacking. Reinforced concrete (RC) is now the
then the lateral forces from wind actions Brownjohn [2]. In
principal structural material used to provide lateral resistance
addition, the imposed displacements in earthquakes may be
in high-rise structures. The tendency to use RC systems is
expected to continue due to the development of commercial very substantial since the standard earthquake displacement
high-strength concretes up to 170 MPa, the advent of spectrum peaks in the period range of 3–6 s (Bommer and
admixtures that can provide high-fluidity without segregation Elnashai [3]), corresponding to the fundamental modes of many
and advances in construction techniques in both pumping and RC high-rise structures.
formwork erection (commented by Ali [1]). For the reasons given above, there is the need for an
Due to the significance of wind forces on the lateral load improved understanding of the inelastic non-linear dynamic
demands in high-rise structures, the effects of lateral loads response of RC high-rise structures subjected to realistic
seismic records representative of near and far field earthquakes.
∗ Corresponding address: Department of Civil and Environmental Engineer- Moreover, and motivated by the increasing interest in obtaining
ing, University of Illinois at Urbana-Champaign, 2106 Newmark CE Lab, MC- more accurate assessments of earthquake losses, there is the
250, 205 N. Mathews Ave., Urbana, IL 61801-2352, USA. Tel.: +1 217 333
1571; fax: +1 217 265 8040. need for deriving fragility relationships for various forms of
E-mail address: kuchma@ad.uiuc.edu (D.A. Kuchma). high-rise structures from these analyses.

0141-0296/$ - see front matter c 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2007.08.026
3198 J. Ji et al. / Engineering Structures 29 (2007) 3197–3209

Fig. 1. Proposed fragility assessment framework.

Fragility is the conditional probability of attainment or influence the response of these complex and highly variable
exceedance of multiple damage states for a given intensity of structural systems and the predictions of their response using
ground excitation, as shown in Eq. (1). a range of realistic input motions. There is a dearth of fragility
relationships for high-rise buildings due to limited field data
P(fragility) = P [LS|IM = x] , and absence of efficient analytical approaches for conducting
(1)
P(LS) = P(C = D) dynamic response history analyses (DRHA) of these structures.
Of particular importance for predicting the response of RC
where C is the capacity, D is the demand, and IM represents the high-rise buildings to seismic actions is the complex behaviour
intensity measure of input ground seismic hazard with intensity of RC structural walls that can dominate the response of
level x. LS refers to the limit state. these structures to lateral loadings. Unfortunately, previous
There are multiple proposed fragility relationships for experimental studies on different configurations of complex
reinforced concrete structural systems that were developed walls, such as those done by Salonikios et al. [9], Thomsen
using different methodologies and parameters for represen- and Wallace [10] and Riva et al. [11], are limited either in
tation of seismic demand and damage. These relationships loading and boundary conditions or in the provided high-
are derived using a few different types of approaches, which resolution test data. One ongoing work by Lowes et al. [12]
are summarized by Rossetto and Elnashai [4] as follows: is considering complex wall configurations including C-shape
(i) Empirical Fragility Curves based on field data. These are to double C core sections and employing high-end technologies
derived through statistical analysis of how real buildings per- for high-resolution data acquisition; however, this project has
formed in past earthquakes. Examples of such fragilities are not yet provided any experimental data. Consequently, fragility
those proposed by Miyakoshi et al. [5] and Orsini [6]; (ii) An- relationships for high-rise RC buildings must at this time be
alytical Fragility Curves. This approach uses numerical tech- derived by purely analytical methods.
niques to simulate the behaviour of systems including variations The proposed framework for deriving analytically-based
in structural capacity and seismic demands. Studies done in fragility curves for RC high-rise buildings is presented
this category include those by Singhal and Kiremidjian [7] and in Fig. 1 which illustrates the key features such as
Mosalem et al. [8]; (iii) Judgemental Fragility Curves. These selection of representative building structures, appropriate
are fragility curves that are based partially or wholly on expert analytical modelling, uncertainty consideration, limit state
opinion. With this approach a wide range of structure types are definitions and numerical simulations. Ideally, these analytical
dealt with in the same manner treating the level of uncertainty assessments would be conducted by running DRHA using
uniformly; (iv) Hybrid Fragility Curves, which are constructed three-dimensional inelastic and geometrically non-linear finite
through a combination of more than one of the other three element models. This is not possible as these models are not
approaches. sufficiently mature to provide reliable predictions and the time
The derivation of fragility relationships for high-rise required to build and run the analyses is prohibitive by orders
structures requires a thorough evaluation of the factors that of magnitude. Hence, the success of an analysis framework is
J. Ji et al. / Engineering Structures 29 (2007) 3197–3209 3199

dependent on the efficiency, appropriateness, and reliability of VecTor2 employs a rotating-angle smeared crack modeling
the selected simplified model of the RC high-rise structure. approach and implements both the Modified Compression Field
The primary technical contribution presented in this paper Theory (MCFT) by Vecchio and Collins [18] and Disturbed
is the suggested process for developing a simple lumped- Stress Field Model (DSFM) by Vecchio [19]. VecTor2 can
parameter model that can capture key aspects of the dynamic model concrete expansion and confinement, cyclic loading and
behaviour of RC high-rise structures including RC material hysteretic response, construction and loading chronology for
inelasticity, second-order geometrical non-linearity, and frame- repair applications, bond slip, crack shear slip deformations,
wall interactions. The selection of appropriate parameters reinforcement dowel action, reinforcement buckling and crack
was made possible using genetic algorithms. Following the allocation processes.
presentation of this model, its effectiveness and computational
efficiency is demonstrated in a case study for an existing 2.2. Lumped modelling approach
complex high-rise building. Sample fragility relationships are
then presented for three performance limit states, which are The proposed two-stage optimization procedure is now
defined as “Serviceability”, “Damage control” and “Collapse described for the selection and calibration of the lumped
prevention” in a companion paper by Ji et al. [13]. Whereas ZEUS–NL model that was selected for running the non-linear
this paper presents the framework and a reference application, DRHA from which example fragility relationships can then be
more extensive presentation of fragility relationships is given derived.
in the companion paper and in which the sensitivity analyses
are presented that are essential for developing final fragility 2.2.1. Stage I — Outer frame elimination and global system
relationships. simplification
The walls and core systems in RC high-rise buildings are
2. Analytical structural modelling usually dominant in resisting seismic loads especially at lower
building levels, while the frame components principally support
For typically complex wall-frame systems in RC high-rise gravity loads and contribute to the lateral resistance with frame-
buildings, the behaviour of the beams and columns can usually wall interactions (Ali [1]). It is therefore acceptable to replace
be adequately captured by fibre or multi-layer beam elements the frames in the dynamic analyses with equivalent non-linear
in which only a strength check is made for shear (Spacone boundary springs at the connection point of the wall and frame
et al. [14], Elnashai et al. [15]). For the walls, either continuum at each floor. This approach, as illustrated in Fig. 2, employs
analysis is required or the effects of shear must be handled a joint element containing three springs which simulate the
separately. It is essential to consider the contribution of the reaction forces [Fx , Fy , Mz ] for the outer frame (as within a
frame and the frame-wall interactions to obtain sufficiently plane) based on the displacements [u x , u y , θz ] at the wall joints.
accurate results from dynamic response history analyses.
A two-stage lumped modelling approach was used in 2.2.2. Stage II — Simplify the structural wall into lumped
this study to provide an efficient computational model for elements
conducting DRHA from which fragility relationships could be It is possible to simplify the model for the wall into lumped
derived. In Stage I, the outer frame was modelled as equivalent elements that capture the non-linear longitudinal behaviour
non-linear springs at the wall, while in Stage II the wall was across the width of the wall as well as its shear behaviour. Two
modelled using lumped elements. different simplified lumped models were used and compared
in this Stage II optimization procedure. The simple-vertical-
2.1. Analysis software platforms line-element model (SVLEM) proposed by Vulcano [20] is
shown in Fig. 3(a) where single linear elastic beam elements
The ZEUS–NL environment, developed by Elnashai are connected with nonlinear axial and rotational springs.
et al. [15], was adopted for running the analyses as: (i) it Ghobarah and Youssef [21] proposed a multiple-vertical-
can be used to model all elements of the frame FE model line-element model (MVLEM) as shown in Fig. 3(b). They
and utilizes non-linear axial and rotational springs as well as modelled wall boundary elements as elastic truss elements
Bernoulli–Euler beam elements that employ a fibred sectional which are connected with horizontal rigid beams supported by
approach in which realistic constitutive relationships are used four non-linear springs simulating the potential plastic hinge
for the response of structural concrete to cyclic loadings, and region. An elastic beam element is used to represent wall web
(ii) it is capable of conducting conventional pushover, adaptive region which is connected with the rigid bars at the center, and
pushover, Eigen-value analysis and dynamic response history then cut and linked with a spring k H for simulating the shear
analyses. behaviour.
It is inadequate to simply predict the behaviour of RC
walls using Bernoulli–Euler beam models as the behavior of 2.2.3. Genetic algorithm application for parametric study
walls is strongly influenced by their shear behaviour. The Although lumped-modelling approaches are conceptually
non-linear 2D continuum analysis tool VecTor2 (developed at simple, the selection of a suitable replacement structure that
University of Toronto, Vecchio [16], Wong and Vecchio [17]) properly considers the influence of the dominant parameters
was used to predict the detailed behaviour of the walls. on the non-linear response is not trivial. There is no explicit
3200 J. Ji et al. / Engineering Structures 29 (2007) 3197–3209

Fig. 2. Equivalent non-linear springs at wall joint.

(a) Single-vertical-line-element (b) Multiple-vertical-line-element (MVLEM) (After Ghobarah


(SVLEM) (After Vulcano [9]). and Youssef [10]).

Fig. 3. Macroscopic structural wall models.

Table 1
Standard versus genetic algorithms (adapted from Goldberg [12])

Requirement Standard algorithm Genetic algorithm


General process Generates a single point at each iteration Generates a population of points at each iteration
Deal with parameters themselves Deal with coding of parameter sets
The sequence of points approaches an optimal solution The population approaches an optimal solution

Seeking algorithm Use functional derivatives Use objective function (payoff information)
Selects the next point in the sequence by a deterministic Selects the next population by computations that
computation involve random probabilistic choices

analytical approach to derive this model from inelastic FEA cost-based structural optimizations of tall buildings using GA
and direct optimization techniques for parametric studies and Fuzzy Logic. However the GA applications in structural
are not appropriate since they all depend on explicit engineering are still in their early stages and this work by the
functions. To develop a suitable lumped model, a reliable authors is the first application of GA for the tuning of FE
and effective computational approach is needed in Stages model parameters with these software platforms and for use
I and II to implicitly optimize the essential parameters in modelling both global and local behaviour of RC high-rise
through consecutive FEA runs and progressively lead to buildings with a complex dual core wall-frame system.
the convergence of typical structural response to standard Conceptually different from standard algorithms, GA
data for calibrations. Genetic Algorithm (GA), a well-known repeatedly modifies a population of individual solutions called
implicit goal-seeking technique, was employed as a primary “generation”, as described in Table 1. In each generation, GA
methodology in this analytical approach. A brief introduction evaluates all individuals and selects those better performing or
to GA and its use in this study is now presented. other specific individuals randomly as parents and reproduce
The Genetic Algorithm, developed by John Holland children for the next generation using three methods: elite,
(Holland [22]) and his colleagues (Goldberg [23]), is a goal- cross-over and mutation. Over successive generations, the
seeking technique used for solving optimization problems, population “evolves” toward an optimal solution as shown in
based on natural selection. Since the appearance as an Fig. 4.
innovative subject, GA application has been expanded from The two-stage structural optimization procedure using GA is
social science, biology and computer science to the fields of illustrated in Fig. 5 in which the left column shows the overall
engineering topology and optimizations. For example, Adeli two-stage process, the middle column presents the structure of
and Sarma [24] have developed computational approaches for the two lumped modelling steps, and in the right column is
J. Ji et al. / Engineering Structures 29 (2007) 3197–3209 3201

Fig. 4. Structure of genetic algorithm (adapted from Goldberg [12]).

Fig. 5. Global lumped model derivations using genetic algorithm.

given a flow chart of the generic algorithm toolbox used in this Table 2
parametric study. Main features of sample building
MATLAB codes were developed and used to realize the Features Description
functionalities presented in Fig. 5. The described procedure is
Height (m) 184.000
illustrated in the following case study. Total stories 54
Regular storey height (m) 3.400
2.3. Case study: Development of lumped-parameter model for Irregular storey height (m) 4.488
sample building Core walls (exterior and interior size) 9.43×3.25 (8.48×2.55) (m)
9.33×3.15 (8.48×2.55) (m)
The framework was implemented for the development of 9.18×3.05 (8.48×2.55) (m)
a simple lumped-parameter model for a single frame (F4) of Concrete f 0 c (MPa) 60 (wall); 40 (slab)
Reinforcing bars f y (MPa) 421 (Grade 60)
an existing RC high-rise structure, the newly constructed high-
rise Tower C03 in the Jumerirah Beach development, Dubai,
United Arab Emirates as illustrated in Fig. 6. The primary frame-wall connection at each storey level. GA will be used
characteristics of this tower are given in Table 2. for the selection of model parameters. The assessment of
the suitability of the simplified model will be made by a
2.3.1. Simplification Stage I using genetic algorithm comparison of natural modes as well as the results from
The objective of the Stage I simplification is to replace pushover and DRHA. The complete frame model is shown in
the outer frame with non-linear springs at the point of the Fig. 7(a) and typical cross-sections in Fig. 7(b)–(d).
3202 J. Ji et al. / Engineering Structures 29 (2007) 3197–3209

Fig. 6. Sample building structure and half plane view.

(a) Whole model. (b) Core wall (400 fibres). (c) Normal column (100 fibres). (d) Slab beam (250 fibres).

Fig. 7. Structural models in ZEUS–NL typical component cross-sections.

Table 3
Modal periods and mass participation factors

Mode 1 2 3 4 (vertical) 5
Period (s) 3.05323 0.81950 0.36427 0.32787 0.22872
Mass participation factor 0.5610 0.2637 0.0729 Neglected 0.0433
Sum of mass participation factor of listed 5 modes 0.9409

From the results of the ZEUS–NL complete model of the The first five modal periods and related MPFs are listed in
outer frame, the modal mass participation factors (MPF) can be Table 3 in which the sum of the modal MPFs of four modes
determined by Eq. (2): among the first five is 94%.
Because of this, it is reasonable to evaluate the lateral
{φi }T [M] {1}
MPF: Γi = (2) distributed loads for a pushover analysis from the 4 horizontal
{φi }T [M] {φi } modes as illustrated in Eqs. (3) and (4) and this leads to the final
where, for i-th mode φi — normalized i-th mode shape vector, lateral load shape that is shown in Fig. 8. The vertical mode 4
M represents the mass. is not considered as it has little effect on the lateral response
J. Ji et al. / Engineering Structures 29 (2007) 3197–3209 3203

Fig. 8. Distributed lateral loads following mixed modal shape.

which is the focus of this study.


Proportional load vector with single modal shape:
M j φ ji
Fi j = P . (3)
M j φ ji
j
Proportional load vector with mixed modal shape:
4
X
F j∗ = Fi j Γi (4)
i=1
where, i — selected mode, j — floor level.
Using this force distribution, a pushover analysis was
undertaken in ZEUS–NL from which the overall nodal
displacements and wall-frame interface forces were evaluated.
In order to obtain a reasonable starting point for all the spring
constants in GA parametric study, a non-linear least square
method was used to simulate the inelastic behaviour of the outer (a) Simplified model protocol.
frame resistant forces corresponding to wall deformations at
each floor as shown in Fig. 2. Thereby, the outer frames can
be replaced by springs at joints that connect the wall nodes to
fixed supports as shown in Fig. 9.
As discussed previously, the first step is to define the
population of individuals and fitness functions used in GA.
In ZEUS–NL, the main properties of each single tri-linear
spring, as shown in Fig. 9(b), are now defined where the initial
and two-stage yielding stiffness and corresponding limit state
displacements at interested DOFs (x, y, r z) are:
 
K 1x K 1y K 1r z K 2x K 2y K 2r z K 3x K 3y K 3r z j
(b) Equivalent tri-linear joint spring.
u 1x u 1y θ1z u 2x u 2y θ2z j
 
and
where, j refers to the floor level. The strain hardening and Fig. 9. Main features in simplified model stage I.
strength degradation behaviours can be simulated by the tri-
linear stiffness properties defined using joint elements in There are two ways to optimize the parameters using GA;
ZEUS–NL. one is to directly define the individual population as a group
3204 J. Ji et al. / Engineering Structures 29 (2007) 3197–3209

containing all parameters and the other is to simulate the spring where, n = 1, 2, . . . , 45. K is of dimension 45 by 1, and in
constants as functions of other properties. For the first method, each generation the population consists of 50 such individual
the protocol parameter vector of the population is: vectors.
The primary comparison between the lumped and more
C j = K 1x , K 1y , K 1r z , K 2x , K 2y , K 2r z , K 3x ,

complete model was of the deformed shapes under the
K 3y , K 3r z , u 1x u 1y θ1z u 2x u 2y θ2z j

(5) same loadings. On account of different influences on whole
building behavior from boundary conditions at different
j = 1, 2, . . . , 54.
heights, weighting factors are introduced in the deformation-
The initial values Ci are obtained through the post- based fitness function. The fitness function is defined as:
processing of the original ZEUS–NL analysis results in the  
 Dtj − Doj 
previous step. In the population, the independent variable
R= , (11)
vector size is determined by the total number of joint spring  j × Doj 
parameters which are 810 for the 54-storey tower. Each
individual vector is a set of 810 generated random numbers of where, j—Floor level = 1, 2, . . . , 54.
uniform distribution as follows: Dtj — Wall nodal displacement along X at ith storey, from
lumped model.
X = [Rand(L B , U B )]1×810 . (6) Doj — Wall nodal displacement along X at ith storey, from
(L B , U B ) are lower and upper bounds, set as (0.5, 2.0) here. original model.
So for each trial in GA, the individual joint parameter vector Both approaches set the maximum generations at 100.
K is computed as The second approach was finally selected due to its greater
computational efficiency. For the evolution, all three techniques
K = C1 C2 · · · C54 ,
 
Kn = K n · Xn (7) including Elite (survival selection), Crossover and Mutation,
were coded in the MATLAB toolbox and employed to enhance
where, n = 1, 2, . . . , 810. K is of dimension 810 by 1, and in performance of the genetic algorithm searching.
each generation the population consists of N such individual For Elite procedure, 20% of all parents with lower fitness
vectors where N is the population size. values were selected directly for the next generation:
For the second method, only selected parameters are
subjected to optimization. This is encouraged by the Xm+1 = Xm , m — Generation number. (12)
observation that the stiffness values are decreasing with For Crossover procedure, two methods were used: one was
increasing height. Thereby, the spring parameters C j can be the partial crossover method as expressed in Eq. (13) and the
assumed to follow certain functional trends, approximately as other one was the complete crossover method as described
below: in Eq. (14). Crossover was the major reproduction method

 (C1 )k · j ak , k = 1, 2, . . . , 15 for j ≤ 25 generating 70% of the population of each generation
  bk
j (1) (1) (1)
 h
Xm+1 = X1 , X2 , . . . , αl Xl

C j k = (C1 )k · 25 · a k , k = 1, 2, . . . , 15 (8)

 25
(2) (2)

for j ≥ 25.
i
+ (1 − αl )Xl , . . . , Xn−1 , Xn(2) (13)
Since the building configuration changes at the 25th floor
as shown in Fig. 6, the stiffness variations are different for where, αl — Preset or random number, ∈ [0 1] , 1 < l < n, n
is variable vector length, m — Generation number
the structure below and above 25th floor, thus the functional
expressions for C j need to change accordingly. Post-processing (1)
Xm+1 = αXm (2)
+ (1 − α)Xm (14)
of the ZEUS–NL analysis is again applied to obtain the initial
values. Only C1 is needed here, but additional estimations of where, α — Preset value or random number, ∈ [0 1]; m —
a and b for each parameter in C1 are also desired for the Generation number.
optimization using genetic algorithms. The final number of For the Mutation procedure, random scaling or shrinking
independent variables in the vector is 45. Each individual is a was imposed on selected individuals according to fitness
set of 45 generated random numbers with a uniform distribution performance level as shown in Eq. (15). The mutation rate is
as follows: 10%.

X = [Rand(L B , U B )]1×45 (Xm+1 )n = (Xm )n · β m n , n = 1, 2, . . . , 45



(9) (15)

where, (L B , U B ) are lower and upper bounds, set as (0.5, 2.0) where, β m — Random
 vector
 generated from normal
here. distribution Norm 1, 1 − rnp , rnp — Ranking of parent
So for each trial in GA, the individual joint parameter vector fitness performance level in precentage, m—Generation
K is computed as: number, np—Parent number within [1, N ].
Lumped model Stage I was obtained as expected through the
K = C1 a b ,
 
Kn = K n · Xn (10) GA parametric study. Series of analyses with derived lumped
J. Ji et al. / Engineering Structures 29 (2007) 3197–3209 3205

Fig. 10. Pushover response comparisons between original and lumped models.

Table 4 slight adjustments on epistemic errors in uncertainty modelling


Modal analyses comparisons between original and simplified models for fragility assessment.
Mode Modal period (s) Similar comparisons were also made using DRHA as shown
Original model Lumped model Stage I in Fig. 11(b) and (c); the sample ground motion record from
Value Error (%) Kocaeli Earthquake (1999, Turkey, Duzce station) is given in
1 3.05323 3.08966 1.2 Fig. 11(a).
2 0.81950 0.80381 −1.9 At intermediate and top height levels, relative displacement
3 0.36427 0.34249 −6.0 time histories from the lumped model are quite close to those
4 0.32787 0.28195 −14.0
from the original model. That means that the lumped model
5 0.22872 0.19274 −15.7
replicates well the predicted seismic behaviour of the selected
building.
models were performed for both modal and pushover analyses
2.3.2. Simplification Stage II using genetic algorithm
and comparisons with original model results were made for
modal analyses (Table 4) and typical pushover curves (Fig. 10). The objective of this simplification is to produce a simple
lumped-parameter model that will provide a similar prediction
By these comparisons, it was observed that the relative
of the pushover response as that which would be predicted from
error for first three significant modal periods, for which the a 2D continuum analysis for the core wall panels. The program
sum of MPF at 90%, was less than 10% between these VecTor2 is used to provide the continuum analysis predictions.
two models. It was also observed that the load-deformation Both the SVLEM and MVLEM models, as were described
pushover responses from the lumped model matched well those in Fig. 3, were considered for developing the lumped model
from the original model. These comparisons thereby indicate in ZEUS–NL. The GA process was employed for parameter
that the derived lumped model was able to provide accurate value selection. In addition, the “hsv” type joint was applied to
predictions of natural modes and static pushover behaviours. consider the axial load-shear interaction effects in ZEUS–NL
There were some expected errors including shorter periods for (Lee and Elnashai [25]). For all wall panels of the same
higher modes and larger resistant forces in the highly inelastic size, only one parametric study for lumped modelling was
range. These errors are probably due to the computational joint required. The core wall panel prototypes used two typical
models used in ZEUS–NL that require non-negative tangent wall dimensions at lower and higher storey levels as shown in
stiffness values and thereby make it difficult to detect actual Fig. 12(a); total storey forces imposed on the wall panel are
strain softening of the substitute outer frame. Hence, the discretized into equivalent nodal forces at the top as shown in
lumped model is a little stiffer than the complete frame model in Fig. 12(b) and (c) which also present the FE mesh used in the
some cases, especially for higher modes and large deformation VecTor2 analyses.
ranges where the RC outer frame members are more likely to The axial loads at different levels, as shown in Table 5
get damaged. Such errors may be taken into consideration by and used in the VecTor2 pushover analyses, were imposed
3206 J. Ji et al. / Engineering Structures 29 (2007) 3197–3209

Table 5
Designed capacities and applied load information
Design No (kN) 255 658
Compressive capacity Cmax (kN) 620 000
Tensile capacity Tmax (kN) 80 000
Applied loads 1 2 3 4 5 6
Axial load 0 0.5N o 1.0N o 1.5N o 0.075Tmax 0.225Tmax
Horizontal displacement (mm) 20

Such joints in ZEUS–NL have a similar tri-linear nature as


previous except the shear stiffness represented by ‘hsv’model,
which is defined as “Hysteretic Shear model under axial force
variation”, requests series of tri-linear shear response curves
with stiffness values and related displacements under different
axial loads including four compressions and two tensions
(in reference to ZEUS–NL version 1.7 manual by Elnashai
(a) Strong motion. et al. [29]).
For the SVLEM model in ZEUS–NL, only one joint element
is needed for each wall, with parameters at x, y and rotation
about z, and there are only three types of wall sections.
Therefore, even considering direct GA optimizations on all
joint element parameters, the population size of each generation
is small enough not to significantly increase the computational
cost. Initial elastic and the two-stage yielding stiffness values
and corresponding limit state displacements are defined as

K W =
 hx h ii
1 , 2 ,..., K6 6 6

1 1
  2 2

K 1x K 2x K 3x K 1x K 2x K 3x K 2x K 3x


 1x


 1×18
 W
(b) 25th floor. U x =



 u1 u 12x u 13x , u 21x u 22x u 23x , . . . , u 61x u 62x u 63x 1×18
      
1x
(16)
K W = K 1yT ,
   

 y K 2yT K 3yT K 1yC K 2yC K 3yC 1×6

W u 2yT , u 1yC
    
u u 2yC 1×4
Uy =  1yT




KW = K

K 2r z K 3r z 1×3

 Rz  1r z



W
= θ1r z θ2r z 1×2

URz

where, values with T and C for the y direction indicate different


responses under tension and compression. The six rows for the
x direction present the data sets for the six axial load levels.
The protocol parameter vector of the population is
h i
CW = KxW , K yW , K WRz , U W
x , U W
y , U Rz .
W
(17)
(c) Roof.
The initial values for CW are obtained from the VecTor2
Fig. 11. DRHA comparisons between original and lumped models. analysis results in the previous step. In this situation, the size
of the independent variable vector within each population is
51. As done for Stage I, each individual vector is a set of 51
as the initial load for the different analyses as required by generated random numbers which have a uniform distribution
the non-linear hsv joints in ZEUS–NL. Horizontal loads were as follows:
incrementally applied in displacement control. All loads were
XW = [Rand (L B , U B )]1×51 (18)
equivalent to uniformly distributed nodal loads along the floor
at the top of the wall panel. (L B , U B ) are lower and upper bounds, set as (0.5, 2.0) here. So
In the VecTor2 model, the concrete compressive stress–strain for each trial in GA, the individual joint parameter vector K is
curve by Popovics [26] was used for normal strength concrete computed as:
and Modified Popovics curve by Collins and Porasz [27] for
Kn = CnW · XnW n = 1, 2, . . . , 51. (19)
high-strength concrete was used for both pre-peak and post-
peak concrete behaviour. The effect of confinement stresses fol- A similar procedure is applied for MVLEM model, except
lows the suggestions by Kupfer et al. [28]. now for a total of five joint elements including four vertical
J. Ji et al. / Engineering Structures 29 (2007) 3197–3209 3207

(a) Wall prototype. (b) Equivalent nodal loads. (c) FEM model in VecTor2.

Fig. 12. Discrete FEM model of core wall panel.

(a) SVLEM. (b) MVLEM.

Fig. 13. Pushover comparisons for SVLEM and MVLEM with VecTor2 model.

springs at the bottom and one horizontal spring at the lower part continuum FEM analysis. The SVLEM model does not provide
of the wall panel. As in Stage I, the three techniques of Elite quite as close a fit particularly around the ultimate storey
(survival selection), Crossover and Mutation, were employed shear but even this error is acceptably small. The comparisons
for generating populations using the genetic algorithm toolbox, illustrate that using GA optimization in parametric studies, both
as described by Eqs. (12)–(15). In the ZEUS analysis for SVLEM and MVLEM can provide lumped models for wall
fragility assessment, nodal displacements at the centre of the panels that are reliable and sufficiently accurate for capturing
wall panel at each floor level were required. Therefore, the main the inelastic behaviours including cracking, steel yielding and
comparison between the optimized lumped model and the RC concrete crushing.
continuum model were control point displacements. Because
the floor slab was treated as rigid in the ZEUS–NL model, while 2.3.3. Evaluation and comments
not in VecTor2, the nodal rotation at control points is excluded Through the implementation of the above two-stage model
from the comparisons. It follows that the fitness function is optimization procedure, a comparatively simple lumped-
defined as parameter model is created that consists of beam elements, rigid
bars and non-linear springs. The simplified model is sufficiently
Dt − Do
R= (20) accurate for evaluating nodal displacements, global internal
|Do | forces and at the same time accounting for shear deformations
where, Dt — Wall nodal displacement at X and Y direction within the structural walls. The required computational time
computed from lumped model, Do — Wall nodal displacement for completing a DRHA is reduced to a small fraction of
at X and Y direction obtained from VecTor2 results. what it would have been for an analysis of the original
The GA optimizations for the parametric study successfully model. Table 6 compares the details and computational times
help to finalize the lumped model for Stage II. A comparison between the original whole building frame model and the
of the pushover prediction under designed dead loads for wall final lumped model in ZEUS–NL. Though it took around one
section types is presented in Fig. 13. week to actually complete the lumped modelling process for
The static pushover response from both the SVLEM and this specific building, the computational saving of next step
MVLEM are close to the response from the VecTor2 detailed DRHA runtime is much more significant compared to this
3208 J. Ji et al. / Engineering Structures 29 (2007) 3197–3209

Table 6
Variation from original model to final lumped model

Variation From original to lumped models

Model geometry

Fig. 14. Sample fragility curves based on SA (T = 1 s).


Node number 876 → 472
Element number 1910 → 632
ZEUS–NL. The probability cumulative function of Log-normal
DRHA runtime 10–15 → 1
distribution was employed to obtain a comprehensive functional
fragility relationship. Fig. 14 presents fragility examples of
cost. The total estimated runtime of DRHA and post-process the reference building based on the intensity measure (IM) of
for the reference structure was about 3 h per 1000 time steps spectral acceleration (SA) at a natural period of T = 1.0 s.
on a Pentium 4 CPU running at 2.66 GHz and with 1 GB
of RAM. This would have required 1800 h (2.5 months) to 4. Conclusions
consider 600 ground motion records with an average of 1000 In this study, an analytical framework was presented for
time steps. Using the same computer, the derived lumped model making fragility assessments of RC high-rise buildings and its
approach resulted in substantial time savings which creates use was demonstrated in a reference application. The principal
greater flexibility for making fragility assessments. contribution presented in this paper is the development
of simplified lumped-parameter analytical models that were
3. Fragility assessment created to run in the ZEUS–NL environment and that consisted
of beam elements, rigid bars and non-linear springs. This model
The analytical framework described in this paper is used to was derived through a two-stage optimization procedure that
develop sample fragility curves for Tower C03 of the Jumeirah employed Genetic Algorithms to capture the predicted inelastic
Beach development in Dubai. The fragility assessment is more non-linear response from more detailed and refined analytical
fully described in another paper by Ji et al. [13], including models. This simplified model is capable of representing the
evaluation and discussion of the effect of uncertainties in interactions between core wall and external frame members
material properties, seismic action, and numerical modelling. by providing sets of non-linear interface springs at X , Y and
Herein, a sample fragility curve assessment is briefly introduced Rz directions at each storey over the height of the wall. The
to illustrate the basic ideas and main features. complex behaviour of the wall, including its flexural, shear and
Three categories of natural records were used in this axial response and their interactions, were also accounted for.
investigation to account for variation in magnitude as well as The ZEUS–NL program was also used in the determination of
the distance from fault and soil conditions. Hence the effects the spring stiffness values to represent the outer frame (Stage I
of uncertainties in ground excitation are taken into account optimization) while the VecTor2 non-linear continuum analysis
when performing numerical simulations with seismic load input tool was used in the selection of the lumped-wall model (Stage
consisting of records varying in all three categories. II optimization). The use of the simplified model resulted
Considering their unique structural configurations and in very significant reductions in analysis times that render it
seismic behaviours new performance limit states are proposed practical to develop probabilistic fragility curves for complex
for RC high-rise buildings under earthquake strikes, correlating high-rise structures, where none have existed in the literature.
global system response and local resisting component By the demonstrated functionality of the proposed frame-
behaviour together, as defined by: (1) Serviceability; (2) work, it is possible to extend the approach to the assessment of
Damage Control; (3) Collapse Prevention. The quantitative the seismic response of other types of RC high-rise buildings.
criteria corresponding to the definitions were derived from the Uncertainties of this proposed approach could be reduced by
results of the pushover analyses based on the knowledge of both further model validation with field experiences and experimen-
global structural response and local damage patterns. tal test data.
Numerical simulations were conducted using the derived Acknowledgements
lumped-parameter structural model and 30 ground motion
records for three defined limit states. This involved the use This study is a product of project EE-1 ‘Vulnerability
of a developed code combining pre- and post-processors for Functions’ of the Mid-America Earthquake Center. The MAE
J. Ji et al. / Engineering Structures 29 (2007) 3197–3209 3209

Center is an Engineering Research Centre funded by the [13] Ji J, Elnashai AS, Kuchma DA. Seismic fragility relationships of
National Science Foundation under cooperative agreement reinforced concrete high-rise buildings. The Structural Design of Tall and
Special Buildings. Published Online: 4 Oct 2007, doi:10.1002/tal.408.
reference EEC 97-01785.
[14] Spacone E, Filippou FC, Taucer FF. Fibre beam–column model for
nonlinear analysis of RC frames: Formulation. Earthquake Engineering
References and Structural Dynamics 1996;25:711–25.
[15] Elnashai AS, Papanikolaou V, Lee DH. Zeus-NL — A system for inelastic
[1] Ali MM. Evolution of concrete skyscrapers: From Ingalls to Jinmao. analysis of structures. 2002.
Electronic Journal of Structural Engineering 2001. [16] Vecchio FJ. Reinforced concrete membrane element formulations. Journal
[2] Brownjohn JMW. Lateral loading and response for a tall building in the of Structural Engineering 1990;116(3).
non-seismic doldrums. Engineering Structures 2005;27(12):1801–12. [17] Wong PS, Vecchio FJ. Vector2 & formworks user’s manual. 2002.
[3] Bommer JJ, Elnashai AS. Displacement spectra for seismic design. [18] Vecchio FJ, Collins MP. The modified compression field theory for
reinforced concrete elements subjected to shear. ACI Structural Journal
Journal of Earthquake Engineering 1999;3:1–32.
1986;83(2):219–31.
[4] Rossetto T, Elnashai A. Derivation of vulnerability functions for
[19] Vecchio FJ. Disturbed stress field model for reinforced concrete:
European-type RC structures based on observational data. Engineering
Formulation. ASCE Journal of Structural Engineering 2000;126(8):
Structures 2003;25(10):1241–63.
1070–7.
[5] Miyakoshi J, Hayashi Y, Tamura K, Fukuwa N. Damage ratio functions
[20] Vulcano A. Macroscopic modeling for nonlinear analysis of RC structural
of buildings using damage data of the 1995 Hyogoken–Nanbu earthquake.
walls. Nonlinear Seismic Analysis of RC Buildings 1992;181–90.
1997.
[21] Ghobarah A, Youssef M. Modeling of reinforced concrete structural walls.
[6] Orsini G. A model for buildings’ vulnerability assessment using the
Engineering Structures 1999;21(10):912–23.
parameterless scale of seismic intensity (PSI). Earthquake Spectra 1999;
[22] Holland JH. Adaptation in natural and artificial systems: An introductory
15(3):463–83. analysis with applications to biology, control, and artificial intelligence.
[7] Singhal A, Kiremidjian AS. A method for earthquake motion-damage 1975.
relationships with application to reinforced concrete frames. NCEER-97- [23] Goldberg DE. Genetic algorithms in search, optimization and machine
0008 1997. p. 200. learning. 1989.
[8] Mosalem KM, Ayala G, White RN, Roth C. Seismic fragility of LRC [24] Adeli H, Sarma KC. Cost optimization of structures: Fuzzy logic, genetic
frames with and without masonry infill walls. Journal of Earthquake algorithms, and parallel computing. 2006.
Engineering 1997;1(4):693–719. [25] Lee DH, Elnashai AS. Inelastic seismic analysis of RC bridge piers
[9] Salonikios TN, Kappos AJ, Tegos IA, Penelis GG. Cyclic load behavior including flexure–shear–axial interaction. Structural Engineering and
of low-slenderness reinforced concrete walls: Failure modes, strength and Mechanics 2002;13(3):241–60.
deformation analysis, and design implications. ACI Structural Journal [26] Popovics S. A numerical approach to the complete stress–strain curve of
2000;97(1):132–42. concrete. Cement and Concrete Research 1973;3(5):583–99.
[10] Thomsen JH, Wallace JW. Displacement-based design of RC structural [27] Collins MP, Porasz A. Shear design for high strength concrete. In:
walls: An experimental investigation of walls with rectangular and T- Proceeding of workshop on design aspects of high strength concrete.
shaped cross-sections. Report # CU/CEE-95/06. 1995. p. 353. 1989. p. 77–83.
[11] Riva P, Meda A, Giuriani E. Cyclic behaviour of a full scale RC structural [28] Kupfer HB, Hilsdorf HK, Rusch H. Behavior of concrete under biaxial
wall. Engineering Structures 2003;835–45. stress. ACI Journal 1969;87(2):656–66.
[12] Lowes L, Lehman D, Kuchma DA, Zhang J. Seismic behavior, analysis [29] Elnashai AS, Papanikolaou V, Lee DH. ZEUS–NL user manual version
and design of complex wall systems: Project summary. 2003. 1.7. 2006.

You might also like