Multiobjective Optimization For Performance-Based Seismic Design of Steel Moment Frame Structures

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS
Earthquake Engng Struct. Dyn. 2005; 34:289–306

Published online 15 December 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/eqe.426
Multiobjective optimization for performance-based seismic

design of steel moment frame structures
Min Liu1 , Scott A. Burns2; ∗; † and Y. K. Wen3

1 Department of Civil; Environmental; and Architectural Engineering; University of Colorado at Boulder;
Campus Box 428; Boulder; CO 80309; U.S.A.
2 Department of General Engineering; University of Illinois at Urbana-Champaign; 104 S. Mathews #117;
Urbana; IL 61801; U.S.A.

3 Department of Civil and Environmental Engineering; University of Illinois at Urbana-Champaign;
205 N. Mathews; Urbana; IL 61801; U.S.A.
SUMMARY
The performance-based seismic design of steel special moment-resisting frame (SMRF) structures is
formulated as a multiobjective optimization problem, in which conicting design criteria that respectively
reect the present capital investment and the future seismic risk are treated simultaneously as separate
objectives other than stringent constraints. Specically, the initial construction expenses are accounted
for by the steel material weight as well as by the number of dierent standard steel section types,
the latter roughly quantifying the degree of design complexity related additional construction cost; the
seismic risk is considered in terms of maximum interstory drift demands at two hazard levels with
exceedance probabilities being 50% and 2% in 50 years, respectively. The present formulation allows
structural engineers to nd an optimized design solution by explicitly striving for a desirable compromise
between the initial investment and seismic performance. Member sizing for code-compliant design of a
planar ve-story four-bay SMRF is presented as an application example using the proposed procedure
that is automated by a multiobjective genetic algorithm. Copyright ? 2004 John Wiley & Sons, Ltd.
KEY WORDS: performance-based seismic design; steel moment-resisting frame; optimal design; multi-
objective optimization
INTRODUCTION
In contrast to traditional prescriptive seismic code provisions that typically use empirical for-
mulations for structural design, emerging performance-based seismic design methodologies
∗ Correspondence to: Scott A. Burns, Department of General Engineering, University of Illinois at Urbana-Champaign,
104 S. Mathews #117, Urbana, IL 61801, U.S.A.
† E-mail: scottb@uiuc.edu
Contract=grant sponsor: National Science Foundation; contract=grant number: CMS 99-12559
Received 28 October 2003

Revised 22 April 2004
Copyright ? 2004 John Wiley & Sons, Ltd. Accepted 20 September 2004
290 M. LIU, S. A. BURNS AND Y. K. WEN
require explicit evaluation of seismic demands on structural systems at predened perfor-

mance levels so that the resulting designs are expected to achieve predictable performances
when subjected to future seismic events. Recent guidelines such as Vision 2000 [1], ATC-40
[2], and FEMA-273 [3] have outlined analysis procedures to evaluate seismic performance
of building structures and provide both qualitative and quantitative denitions for structural
and component performance levels. In terms of overall structural and nonstructural damage
levels, FEMA-350 [4] describes Immediate Occupancy (IO) and Collapse Prevention (CP) as
two performance levels paired with seismic hazard levels of exceedance probabilities being
50% and 2% in 50 years (briey denoted hereafter as 50=50 and 2=50 hazard levels), re-
spectively. The IO level implies very light damage with minor local yielding and negligible
residual drifts, while the CP level is associated with extensive inelastic distortion of structural
members with little residual strength and stiness. Predicted structural performance parame-
ters at each hazard level are then checked against respective threshold values that result from
a combination of numerical/experimental study and engineering judgment. Using these mul-
tilevel multicriteria performance-based design procedures, structural engineers are hopefully
able to directly and explicitly control the seismic performances of a design project, indicat-
ing that structural as well as nonstructural damage is expectedly reduced to an acceptable
level.
In addition to satisfying conventional code requirements, it is of particular interest to seek
a design solution with more economical use of resources. An optimized seismic design is
obtained when it can achieve balanced minimization of two general competing objective func-
tions: the present capital investment and future seismic risk. Multiple merit measures exist in
design practice and can be used to assess the quality of a design candidate. Structural material
usage is one such merit objective function. Besides, degree of design complexity may also be
used as another merit measure because it may aect the labor-related construction cost sig-
nicantly. Unlike stringent code specications that a valid design has to comply with, these
various merit objective functions are not restrictive in nature and their actual values are subject
to structural engineers’ choice. This is especially true for performance-based seismic design
where multiple seismic performance objectives need to be appropriately achieved. Acceptable
structural performance parameters recommended in established guidelines are best interpreted
as indicative of performance ranges that a structure may sustain when responding at dierent
performance levels. Structural engineers are expected to make judgmental decisions on what
these performance measures should actually be for code-compliant seismic designs.
Much of the existing research on seismic design optimization in the literature is single-
objective based with the structural material usage (weight or cost) as the mostly cited objec-
tive function while treating seismic risk related performance merits as constraints for checking
design validity only. When facing a single optimized design solution, structural engineers do
not have a broad view of how other alternative code-compliant designs behave in terms of
relevant merit objective functions under consideration. Rather than accepting or rejecting a
single optimized design solution resulting from material usage minimization, structural en-
gineers may be more interested in actively selecting a nal structural design from among
a group of design candidates that exhibit diverse characteristics in terms of dierent merit
measures. Therefore, a more natural way is to treat all relevant merit measures separately as
well as simultaneously in structural design optimization, which leads to the formulation of a
multiobjective optimal design problem and thus a distribution of design solutions that estab-
lishes tradeo among all selected conicting objectives in a Pareto optimal sense. Structural
Copyright ? 2004 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2005; 34:289–306
PERFORMANCE-BASED SEISMIC DESIGN OF SMRF STRUCTURES 291
engineers then compare these alternative designs and choose with much freedom the one that
compromises dierent competing merit aspects in the most preferred manner.
Research on the design optimization of civil structural systems within the framework of
performance-based seismic design has appeared only recently. Beck et al. [5] presented a
multi-criteria optimal design framework for performance-based design of structural systems,
using a decision theoretic approach based on aggregation of preference functions for the
multiple conicting design criteria. Li et al. [6] proposed a multiobjective and multilevel pro-
cedure for optimizing seismic steel frames; the total structural strain energy and total structural
weight were considered as two objective functions at the system level and member weight
was considered as a single objective function at the element level. Ganzerli et al. [7] min-
imized overall material cost for a simple reinforced concrete portal frame with performance
constraints on beam/column plastic rotations. Foley [8] summarized the state-of-the-art of the
performance-based design for building structures and discussed the application of structural op-
timization techniques in such a design framework. Liu et al. [9] considered life cycle costs in
multiobjective design optimization of seismic steel special moment-resisting frame (SMRF)
structures using a series of structural performance (damage) levels in terms of maximum
interstory drift ratios. Liu [10] systematically developed multiobjective optimization proce-
dures for the seismic design of SMRF structures. Using a portal steel frame as an example,
Alimoradi et al. [11] investigated the construction cost minimization problem with constraints
of condence levels at dierent hazard levels set forth in FEMA-350.
In this study, the multiobjective optimization techniques are applied to performance-based
seismic design of SMRF structures. The initial expenses are reected by the steel material
weight in conjunction with the number of dierent standard steel section types, which in
an approximate sense measures the additional construction cost due to the varied degree of
design complexity an SMRF bears. In compliance with FEMA-350, the seismic structural
performance is indicated by the maximum interstory drift demands at two hazard levels with
exceedance probabilities being 50% and 2% in 50 years, respectively. A multiobjective genetic
algorithm (GA) is used as the search engine to locate a group of optimized tradeo seismic
designs in the presence of these conicting design objectives. As a simple numerical example,
member sizing for code-compliant design of a planar ve-story four-bay SMRF with a given
layout is presented using the proposed procedure.
MULTIOBJECTIVE PERFORMANCE-BASED DESIGN OPTIMIZATION
An automated structural design optimization procedure typically comprises three general com-
ponents: (1) relevant constraints that dene the valid design space, (2) appropriate objective
functions based on which merits of dierent valid designs are assessed and compared, and (3)
suitable numerical algorithms that guide the search toward optimized solutions. In this sec-
tion, the relevant seismic steel design provisions, merit objective functions, the performance
evaluation procedure, and the optimization tool are discussed.
Design provisions
The equivalent lateral force procedure of the 2000 NEHRP Recommended Provisions for New
Buildings and Other Structures (FEMA-368 [12]) is used in this study. It permits an elastic
analysis for the seismic design of SMRF structures that are represented by simple linear
elastic models with centerline dimensions. The design response spectrum is based on 2/3 of
the 5%-damped maximum considered elastic response spectrum (at the 2=50 hazard level) with
proper site class eects adjustment. A structure designed in accordance with these provisions is
equipped with a resistance that is typically lower than what can keep the structure elastic when
subjected to design earthquakes. Inelastic structural responses are generally expected, which
are considered approximately through empirical factors provided in these provisions. For steel
SMRF, a factor of 8 is used for the elastic strength reduction, and drift ratios calculated from
the linear elastic analysis are amplied by a factor of 5.5 to account for inelastic response
and compared to a 2% threshold. Note that changes to 1997 NEHRP provisions (FEMA-
302 [13]) relevant to seismic steel SMRF design in this study include a new lower bound
for the seismic base shear coecient, a modied formula for the approximated fundamental
period, and larger coecients for the upper limit on the calculated period. The fundamental
period calculated from a rational dynamic analysis is used in this study to determine design
base shear level (with an upper bound on the calculated period) as well as to calculate the
nominal design drift ratios (without upper bound on the calculated period).
AISC load and resistance factor design (LRFD) specications for structural steel build-
ings [14] and AISC seismic provisions for structural steel buildings [15] dene relevant load
combinations, member strength requirements, and cross-section slenderness ratio limits for
steel SMRF designs. Specically, in addition to load combination scenarios prescribed in
AISC-LRFD, two more load combinations with amplied horizontal earthquake load eects
are needed in AISC seismic provisions to check axial strength of column members, which
are sensitive to the eects of structural overstrength. The column member strength is checked
by interaction equations of axial forces and exure. A strong-column–weak-beam mechanism
needs to be ensured for SMRF designs that are classied in seismic design category D and
higher. Appropriate width-thickness limits are applied for web and ange elements of beams
and column members, respectively.
Merit objective functions

Practical objective functions are essential for an optimal structural design procedure to pro-
duce improved designs that are viable in the real-world practice. In the performance-based
seismic design optimization, merit objective functions are dened such that they either ad-
dress the immediate economic concern or reect future seismic risks in terms of structural
performance parameters under predened seismic hazard levels. Weight minimization has been
the most widely used criterion in the structural optimization community over the decades. In
this study, the initial expenses are taken into account by two separate objective functions:
the steel material weight and the degree of design complexity in terms of the number of
dierent standard steel section types. Seismic structural performances are represented by the
maximum interstory drift ratio at any designated hazard level. It is taken as the largest value
of height-wise peak interstory drift ratios, which are the ratios of the transient peak story dis-
placements obtained at a hazard level to the respective story heights. These localized response
parameters reveal structural resistance against instability and collapse due to P-delta and thus
are an excellent measure of both structural and non-structural damage because of their close
relationship to plastic rotation demands on individual beam–column connection assemblies
(FEMA-350).
Figure 1. Static pushover analysis and seismic performance indices.
In addition, the peak roof drift ratio and system displacement ductility at designated seis-
mic hazard levels are selected as complementary performance indices but are not used in the
optimization process. As shown in Figure 1, the peak roof drift ratio is dened as the peak
lateral roof displacement at a particular hazard level, u , normalized by the building height
H. This parameter reects the height-wise average deformation severity and overall damage
to nonstructural elements. It also reects the degree of vulnerability of a building structure
to P-delta eects due to cumulative interior gravity loads above a story level acting on the
deected shape; a large peak roof drift ratio indicates possible loss of stability and poten-
tial collapse [16]. Based on a single-degree-of-freedom (SDOF) system analogy, the system
displacement ductility is dened as the ratio of the peak displacement u to the yield dis-
placement y of a control node (usually the roof node). In the context of the static pushover
analysis, the yield displacement corresponds to the intersection point of a bilinear idealization
of the original pushover curve (Figure 1). System ductility approximately measures the extent
of structural damage due to post-yield inelastic deformation in a global sense. A steel design
practice usually relies on component ductility to dissipate seismic energy imparted on the
structural system. On the other hand, sucient detailing is necessary to ensure enough defor-
mation capacity for the structural system in order to accommodate a large ductility demand,
which will in turn require additional construction eorts.
Static pushover analysis

The non-linear static procedure or the pushover analysis is used in this study for seismic
performance prediction. As schematically shown in Figure 1, the non-linear analytical model
of a building structure is statically subjected to a monotonically increasing lateral load pattern
(either predetermined or adaptive), which approximates the time-varying earthquake-induced
inertial force distribution over the height of the structure, until a target displacement level
of a controlling point (at roof level in this study) is reached. Seismic demands at this target
displacement are checked against predened criteria for performance evaluation. The lateral
load pattern used in this study follows the height-wise force distribution of seismic base shear
as described in 2000 NEHRP provisions. The peak roof displacement at a given seismic
hazard level is estimated using an empirical R––T relationship proposed by Nassar and
Krawinkler [17], where, as shown in Figure 1, R is the ratio of elastic force Vel to yield
force Vy , is the SDOF displacement ductility, and T is the ‘equivalent’ period of the SDOF
system. DRAIN-2DX [18] is used in this study to perform the pushover analysis with a
displacement control option.
Genetic algorithms
Compared to most of the traditional methods that are single-objective based and dicult to
handle discrete-valued design variables, GAs are problem-independent methods and they do
not require sensitivity information to guide the search process, which makes them especially
eective for solving the posed multiobjective design optimization for steel SMRF structures,
where design variables are discrete beam and column section types that are selected from a
catalog of commercially available standard steel wide-ange sections. More importantly, GA
can handle multiple competing objectives simultaneously, leading to a spread of optimized
tradeo design solutions by a single algorithm run. Selection, crossover, and mutation are
three basic operators in GA. In view of their salient advantages, GAs have been fruitfully
applied for structural optimization problems [19, 20].
In this study, each design is encoded as a string of pointers into two tables of standard
hot-rolled steel wide-ange sections typically used for column and beam members, respec-
tively [10]. The initial GA design population is generated by an exhaustive combination of
single beam and single column sections from the two section tables. Each of the following
generations consists of 1000 ospring designs and the non-dominated elitist designs from the
parent generation. Fitness of each design solution is determined by a non-dominated sorting
technique [21] plus a parameter-free niching strategy termed ‘crowding distance’ measure [22].
A constrained binary tournament selection scheme is used to handle constraint-violating solu-
tions [22]. A two-point crossover is applied with a probability of 50%, one crossover being in
each column and beam portion of the encoded string. Mutation is performed with a probability
of 30% by randomly perturbing one pointer anywhere in the string.
AN APPLICATION EXAMPLE
Example steel SMRF

The example structure used in this study is a regular planar ve-story four-bay steel SMRF
structure which constitutes one of the two identical north–south perimeter frames in a 100 ft
(30.48 m) by 150 ft (45.72 m) oce building xed at the base, with a height of 67 ft
(20.42 m), and assumed to be built in the Los Angeles area, California. The planar view
and elevation of this building structure can be found in Reference [9], where information
on nominal and expected steel strength, dead load, live load, and exterior wall and facade
intensities is also provided. Symmetric pairs of members about the vertical centerline of this
SMRF are of the identical section types, respectively. All beam members across the same
oor (or roof) level are grouped with the same section type, respectively. Column splices are
assumed to be located exactly at the third oor level, although in reality they usually appear
at a distance above the oor level. Therefore, there are in total eleven design variables: six for
Figure 2. Acceleration response spectra at: (a) 50=50; and (b) 2=50 hazard levels (thin line = individual
spectrum; thick line = target smooth spectrum).
columns and ve for beams. The labor cost needed for a column splice is assumed equivalent
to 500 lbs (2.2 kN) of Grade 50 steel [23].
Structural modeling
The main goal of the present study is to develop a general performance-based seismic design
optimization procedure for steel SMRF structures while considering multiple merit objective
functions, some of which are based on the evaluation of actual seismic performance of code-
conforming alternative designs. A simplied analytical steel frame model may be accepted
that accounts for major contributions of structural elements to seismic responses, in particu-
lar, to displacement-related responses. The following modeling issues are considered in the
present study: (a) a bare-frame model of the planar steel SMRF structure is created in the
DRAIN-2DX, where centerline beam–column elements (Type 2) without strength=stiness de-
terioration are used to model all beams and columns with a 3% strain hardening ratio for point
plastic hinges located at the element ends; (b) full restraints are assumed in the minor axis
directions of column members and only buckling strengths of columns in the major axes are
considered; compact sections are used and adequate lateral bracing is applied where appro-
priate; (c) P-delta eects due to interior gravity loads are modeled using a ctitious column
element technique [24]; (d) rigid beam-to-column connections are assumed, that is, hysteretic
connection behavior under cyclic loading is not modeled; and (e) panel zone deformation
(shear distortion) is neglected.
Seismic inputs
As shown in Figure 2, the target 5%-damped smooth elastic response spectra for the static
pushover analysis are used to generate two sets of ground motion records for the Los Angeles
area as part of the FEMA=SAC Steel Project [25]. The two sets are with soil prole D at
two hazard levels corresponding to 50-year exceedance probabilities of 50% and 2% (denoted
as 50=50 and 2=50), respectively. Because of the approximate nature of the static pushover
analysis, seismic performance of a code-compliant structural design may be further evaluated
by the more accurate time history analysis so that possible structural weakness that cannot
be captured by a simplied analysis procedure may be exposed and assessed. Sets of twenty
SAC ground motion records at 50=50 and 2=50 hazard levels, respectively, whose 5%-damped
median elastic response spectra match the above-mentioned target response spectra at 0.3,
1.0, 2.0, and 4.0 sec in a least-square sense, will be used later to perform the time history
analysis with DRAIN-2DX. Their individual response spectra are also plotted in Figure 2.
Viscous damping ratios are set, at the rst mode and at the 0.2 second, to be 4.0%, which
is linearly interpolated from 4.3% and 3.6% for typical three- and nine-story steel SMRF
buildings, respectively [24].
Distribution of optimized designs with respect to primary merit measures

The present GA-based multiobjective optimization procedure is now applied for seismic design
of the planar ve-story four-bay steel SMRF. DRAIN-2DX is used to obtain the fundamental
period, based on which the design base shear and nominal design drift ratios are determined
in accordance with 2000 NEHRP provisions. The multiple merit objective functions used in
the optimization are steel material weight, number of dierent section types, and maximum
interstory drift ratios at both 50=50 and 2=50 hazard levels. All of these four objective functions
are subject to simultaneous minimization and are hence referred to as primary merit measures.
A design process that simply complies with seismic code provisions does not explicitly
consider actual seismic performance or damage implication and is likely to end up with a nal
design solution based on reduction of initial expenses only. The essence of the performance-
based seismic design methodology is that realistic seismic structural behaviors are explicitly
considered during the design process. A total of 1560 optimized tradeo seismic designs are
obtained at the 400th generation. Note that it takes about 10 minutes for each generational
computation on a Dell Inspiron 8200 laptop. This set of designs exhibits a broad distribution
over the four primary merit measures, as depicted in Figure 3. The steel material weight stays
in a range of 128,160 lbs (570.0 kN) to 397,050 lbs (1765.9 kN); the number of dierent steel
section types, which varies from 2 to 10, indicates optimized designs with diverse degrees
of design complexity. FEMA-273 suggested deterministic thresholds for median maximum
interstory drift demands for dierent performance levels; for steel SMRF structures, 0.7% at
the IO level and 5% at the CP level are considered acceptable. From Figure 3, it is seen that
most of the designs satisfy the drift ratio limit at the IO level (paired with 50=50 hazard level)
and all designs satisfy the drift ratio limit at the CP level (paired with 2=50 hazard level).
Note that drift demands plotted in Figure 3 are obtained by the static pushover analysis.
FEMA-350 presented a probabilistic procedure that evaluates structural performance in terms
of condence levels for specied performance levels, taking into account the uncertainty
and randomness in both structural demand and capacity estimation as well as in seismic
excitations. Median interstory drift capacities are 2% and 10% for global behaviors at IO
and CP performance levels, respectively. Minimum recommended condence levels, in terms
of global behavior limited by interstory drifts, are 50% at IO level and 90% at CP level,
respectively. Calculation [10] shows that most of the present optimized structural designs are
associated with much higher condence levels of satisfying both IO and CP performance
levels than those recommended in FEMA-350, which is in agreement with Reference [26]
regarding post-Northridge steel SMRF designs complying with new NEHRP provisions.
Figure 3. Distribution of all optimized tradeo designs at the 400th generation with respect to each
primary merit measure (MIDR = Maximum Interstory Drift Ratio).
Distribution of optimized designs with respect to secondary merit measures

In contrast to primary merit measures, parameters that assess other merit aspects of alternative
designs but are not used in formulating the present optimization problem are called secondary
merit measures in this study, including (a) the maximum nominal design drift ratio that is
calculated by the codied equivalent lateral force procedure, (b) the system yield coecient
Sy that is the ratio of the system yield force Vy to the participating building seismic weight
W (Figure 1), and (c) the peak roof drift ratio and system ductility at 50=50 and 2=50 hazard
levels, respectively.
Figure 4 shows the dispersion of all 1560 designs with respect to each secondary merit mea-
sure. It is observed that these designs have maximum nominal design drift ratios well below
the 2% threshold per 2000 NEHRP provisions for a codied steel SMRF design investigated
in this study, indicating many conservative designs are present in the optimized design popu-
lation. It is known that drift ratios other than strength requirements usually control the design
of seismic structures in practice. Conventionally, structural designs with very low nominal
drift ratios are often revised or discarded in order to achieve better economy. In this study,
these conservative design solutions are selectively retained in the present multiobjective opti-
mization for two reasons. First, they increase diversity in the population of alternative designs
Figure 4. Distribution of all optimized tradeo designs at the 400th generation

with respect to each secondary merit measure.
during the optimization process, which makes it more likely for designs to evolve toward
better ones; and second, conservative designs in the nal GA generation provide additional
design solutions for structural engineers to compare with other traditional designs in order to
obtain a structural design with a preferred balance among dierent conicting objectives.
The system yield coecient Sy obtained from the static pushover analysis indicates ac-
tual structural lateral strength against system yielding and can roughly measure the system
overstrength ratio, which is dened as the ratio of ultimate strength to the codied nominal
design strength, if additional strength due to any positive system strain hardening is ignored.
The value of Sy ranges from 0.20 to 0.81 for all 1560 optimized designs, as observed in
Figure 4. The codied nominal design base shear coecient for the present ve-story frame,
using the rationally computed fundamental period with a codied upper bound, is between
0.092 and 0.141 after being increased by 5% to consider eects of accidental torsion, from
which one obtains an estimate of system overstrength ratio varying between 2.23 and 5.74.
The overstrength ratio of a typical steel SMRF structure is approximately between 3.3 and
4.5 [26], which largely results from satisfying stringent drift limitations. In this study, the
fact that much stronger than code-required designs exist with nominal design drift ratios far
below the 2% threshold explains the presence of these large overstrength ratios.
It is noted, however, that structural designs with high yield strength levels usually incur
large roof and oor accelerations that decrease occupant comfort during mild excitations and
imply potential damage to mounted nonstructural systems when subjected to strong ground
motions; the associated large base shears also increase internal forces in column members at
the base, which imposes diculty on foundation design. All these concerns should be taken
into due consideration when selecting the desirable alternative designs.
As shown in Figure 4, system displacement ductility measures at the 50=50 hazard level
are all equal to unity. This indicates that maximum roof displacement is less than the yield
displacement for each optimized design and therefore it remains elastic at the 50=50 hazard
level. At the 2=50 hazard level, system ductility values exceed unity and are more scattered,
which implies that these structures encounter varied inelastic damage severities.
Use of the maximum interstory drift ratio as the sole deformation measure in the opti-
mization may lose information on height-wise drift variation. It may be desirable to design a
building structure that has relatively uniform deformation demands over the height in order
to avoid soft-story mechanisms where drift demands are concentrated in one or only a few
stories. Since the peak roof drift ratio describes the average height-wise drift demand, a ‘drift
uniformity ratio’, which is dened as the ratio of maximum interstory drift ratio to the peak
roof drift ratio, may be used to roughly address the severity of drift concentration and hence
to help selection of design solutions with more desirable deformation patterns. Plotted in
Figure 5 are such uniformity measures for all 1560 optimized designs at 50=50 and 2=50 seis-
mic hazard levels, respectively. The fact that a majority of drift uniformity ratios are close to
unity indicates that drift demands are satisfactorily evenly distributed over the building height
for the present structural design of ve-story four-bay steel SMRFs.
Optimized designs with the same number of dierent section types

Numerous design solutions exist for a particular number of dierent steel section types. For
illustration purposes, Figure 6 plots all 199 optimized designs with ve dierent section types,
which form a subset of the 1560 alternative designs. There are clear tradeos between the
steel material weight and maximum interstory drift demands. Note that the objective functions
of maximum interstory drift ratios at 50=50 and 2=50 hazard levels are complementary rather
than competing, since they both describe seismic demands on structures. A structural engineer
now has much exibility in choosing a preferred design other than the minimum weight
design with ve section types. For example, another alternative design with a material weight
of 150,504 lbs (669.4 kN) is selected, which is about 11.7% heavier than the design with a
Figure 5. Drift uniformity ratios of all optimized tradeo designs at the 400th generation.
Figure 6. Optimized tradeo designs with ve section types at the 400th generation (circle =
the minimum weight design; square = an alternative design).
minimum weight of 134,786 lbs (599.5 kN) and ve dierent section types. As a tradeo, the
maximum interstory drift demands at 50=50 and 2=50 hazard levels are reduced by 20.2%
and 22.1%, respectively. Whether or not this particular tradeo is benecial will be judged
by experienced engineers.
Table I. Member sizes for designs of steel material weight close to 150 kips with lowest maximum
interstory drift demands and varied section type numbers.
Number of
section types 3 4 5 6 7 8 9a
Section Group ID
C1 W14X342 W14X342 W14X342 W14X257 W14X257 W14X257 W14X257
B1 W30X99 W33X118 W33X118 W33X130 W33X130 W33X130 W33X130
a
The rightmost column contains the minimum weight design among all 1560 solutions.
Optimized designs with close material weights

Another design situation is considered where the steel material weight assumes a relatively
xed value, say, 150; 000 ± 2000 lbs (667:1 ± 8:9 kN). For a given section type number, the
design solution with the lowest maximum interstory drift ratio at the 2=50 hazard level is
identied in the 1560 alternative designs. Six structural designs are thus obtained with member
sizes and detailed information in Tables I and II, respectively. The meaning of section group
ID can be found in Reference [27]. The best compromise design could be, for example, the
one with four or ve dierent section types.
Time history analysis of two alternative designs

Two alternative designs are selected herein for detailed time history analysis. Design I is
the one with the overall minimum material weight of 128,160 lbs (570.0 kN) with nine dif-
ferent section types (refer to Table I), and Design II has a steel material weight close to
150 kips (specically 151,796 lbs or 675.1 kN) with the lowest maximum interstory drift
demands among all optimized designs of four dierent section types (refer to Tables I
and II).
For each design, the time history analysis is performed using twenty SAC ground motion
records at the 50=50 and 2=50 hazard levels, respectively. The peak absolute interstory drift
ratio for each story from each single time history analysis is selected and, as a result, there are
in total twenty peak interstory drift ratios for each story. Assuming a lognormal distribution of
each interstory drift demand, the sample median (50th percentile) drift demand is calculated
as the exponent of sample mean ofthe natural logarithm of drift demands from all time
history analyses, i.e. Dmedian = exp[( Ni=1 ln Di )=N ]; the 84th and 95th percentile sample drift
demands are obtained by multiplying the sample median value by the exponent of one and
two times sample standard deviation of the natural logarithm of drift demands from all time
historyanalyses, respectively, i.e. D84th = Dmedian exp() and D95th = Dmedian exp(2) with
= [( Ni=1 (ln Di − ln Dmedian )2 )=(N − 1)]1=2 .
302
Table II. Merit measures for designs of steel material weight close to 150 kips with lowest maximum interstory drift
demands and varied section type numbers.
50=50 hazard level 2=50 hazard level
Copyright ? 2004 John Wiley & Sons, Ltd.

Max.
Number Minimum nominal Max. Max.
of material System design interstory Peak roof Drift interstory Peak roof Drift
section weight yield drift drift ratio drift ratio uniformity System drift ratio drift ratio uniformity System
types (lbs) coe. ratio (%) (%) (%) ratio ducility (%) (%) ratio ductility
3 150,190 0.235 1.73 0.79 0.59 1.34 1.00 3.35 2.25 1.49 2.83
4 151,796 0.281 1.56 0.63 0.58 1.09 1.00 2.72 2.26 1.20 2.58
5 151,596 0.279 1.48 0.63 0.56 1.13 1.00 2.58 2.18 1.18 2.67
6 148,344 0.294 1.57 0.63 0.58 1.09 1.00 2.60 2.25 1.16 2.53
7 151,888 0.293 1.44 0.61 0.56 1.09 1.00 2.45 2.17 1.13 2.62
8 151,236 0.308 1.57 0.60 0.57 1.05 1.00 2.44 2.22 1.10 2.49
M. LIU, S. A. BURNS AND Y. K. WEN
9a 128,160 0.228 1.92 0.79 0.64 1.23 1.00 3.21 2.46 1.30 2.76
a
The bottom row is for the minimum weight design among all 1560 solutions.
Earthquake Engng Struct. Dyn. 2005; 34:289–306

Figure 7. Performance of two alternative designs: (a) nominal design drift ratio proles; and
(b) normalized static pushover curves.
Figure 8. Peak interstory drift demand proles at dierent hazard levels for two alternative
designs by time history analysis.
Height-wise nominal design drift ratio proles of these two alternative designs based on the
codied elastic analysis are plotted in Figure 7(a) and the normalized static pushover curves
are plotted in Figure 7(b). Figure 8 shows the median, 84th percentile, and 95th percentile
of peak interstory drift ratio demand proles based on results of time history analysis. It is
observed that Design II consistently has a higher seismic capacity than Design I. In terms of
median maximum interstory drift demands at 50=50 and 2=50 hazard levels, for example, the
values for Design II are 0.88% and 3.15%, respectively, which are 21.4% and 20.5% less than
the median maximum interstory drift demands of 1.12% and 3.96% for Design I, respectively.
These observations imply that Design II would perform better seismically and therefore would
incur less potential seismic damage than Design I. Combining the knowledge of accurately
predicted structural performances with the fact that Design II is 23,636 lbs (105.1 kN) or 18.4%
heavier than Design I while Design I has ve more section types than Design II, a structural
engineer has the basis to make a judgmental decision on the choice of the nal frame design.
CONCLUSIONS
Optimal seismic design procedures existing in the literature are usually single-objective based
with structural material usage as the sole objective function while imposing constraints from
code specications as well as structural performance consideration. In contrast to stringent

constraints in traditional prescriptive seismic codes, acceptable performance parameters rec-
ommended in recent performance-based seismic design guidelines are best interpreted as in-
dicative of performance ranges that a structure may sustain when responding at dierent
performance levels. For performance-based seismic design optimization, a natural approach is
therefore to consider structural performance parameters and other applicable merit measures
as objective functions instead of constraints.
An automated procedure has been presented in this paper that combines performance-based
seismic design methodology and genetic algorithms (GAs) for optimized member sizing of
steel special moment resisting frames in accordance with the 2000 NEHRP seismic structural
design criteria, AISC-LRFD steel design criteria, and AISC seismic design criteria. Merits
of a code-compliant design are assessed by multiple objective functions, which reect either
initial expenses in terms of the steel material weight and the number of dierent steel section
types or future seismic risks in terms of interstory drift demands at selected seismic hazard
levels.
By treating all relevant objectives simultaneously as well as separately, the present GA-
based multiobjective design optimization procedure can produce a wide distribution of alter-
native designs that establishes optimized tradeo among these merit objectives. Consequently,
structural engineers have a much broader vision of the entire optimized valid design space
than a single structural design that is obtained from traditional single-objective based structural
optimization. Through an explicit tradeo analysis of design candidates that are preliminar-
ily selected from the optimized design pool, engineers can conveniently determine the nal
compromise design that has desirable seismic behavior with balanced initial expenses.
This study has focused on the development of general multiobjective optimization proce-
dures for performance-based seismic structural design. A simple centerline structural model
was used for seismic performance evaluation. A more rened structural model incorporating,
for example, panel zone deformation, realistic connection representation, strength and stiness
participation of the gravity frames, and consideration of realistic design requirement will in-
crease the accuracy in predicting seismic structural performance and hence improve the quality
of the nal optimized tradeo designs. In addition, large-scale structural optimization consid-
ering the eects of three-dimensional responses may become necessary [28]. The proposed
general procedure can also be applied in a more complex design optimization environment.
ACKNOWLEDGEMENTS
This material is based on work funded by the National Science Foundation under Award No. CMS
99-12559. This nancial support is gratefully acknowledged.
REFERENCES
1. Structural Engineers Association of California. Performance-based Seismic Engineering of Buildings, Vision
2000 Report, Sacramento, CA, 1995.
2. Applied Technology Council. Seismic Evaluation and Retrot of Concrete Buildings, ATC-40, Redwood City,
CA, 1996.
3. Building Seismic Safety Council. NEHRP Guidelines for the Seismic Rehabilitation of Building, FEMA-273,
Federal Emergency Management Agency, Washington, DC, 1997.
4. Building Seismic Safety Council. NEHRP Recommended Seismic Design Criteria for New Steel Moment-
frame Buildings, FEMA-350, Federal Emergency Management Agency, Washington, DC, 2000.
5. Beck JL, Chan E, Irfanoglu A, Papadimitriou C. Multi-criteria optimal structural design under uncertainty.
Earthquake Engineering and Structural Dynamics 1999; 28:741 – 761.
6. Li G, Zhou R, Duan L, Chen W-F. Multiobjective and multilevel optimization for steel frames. Engineering
Structures 1999; 21(6):519 – 529.
7. Ganzerli S, Pantelides CP, Reaveley LD. Performance-based design using structural optimization. Earthquake
Engineering and Structural Engineering 2000; 29:1677 – 1690.
8. Foley CM. Optimized performance-based design for buildings. In Recent Advances in Optimal Structural
Design, Burns SA (ed.). ASCE, 2002 (Chapter 8).
9. Liu M, Burns SA, Wen YK. Optimal seismic design of steel frame buildings based on life cycle cost
considerations. Earthquake Engineering and Structural Dynamics 2003; 32:1313 – 1332. DOI: 10.1002/eqe.273.
10. Liu M. Development of multiobjective optimization procedures for seismic design of steel moment frame
structures. Ph.D. Thesis, Department of Civil and Environmental Engineering, University of Illinois at
Urbana-Champaign, Urbana, IL, 2003.
11. Alimoradi A, Pezeshk S, Foley CM. Automated performance-based design of steel frames. ASCE Structures
Congress, Nashville, TN, 22–26 May 2004.
12. Building Seismic Safety Council. NEHRP Recommended Provisions for Seismic Regulations for New Buildings
and Other Structures, FEMA-368, Federal Emergency Management Agency, Washington, DC, 2002.
13. Building Seismic Safety Council. NEHRP Recommended Provisions for Seismic Regulations for New Buildings
and Other Structures, FEMA-302, Federal Emergency Management Agency, Washington, DC, 1998.
14. American Institute of Steel Construction. Manual of Steel Construction—Load and Resistance Factor Design,
2nd edn. Chicago, IL, 1994.
15. American Institute of Steel Construction. Seismic Provisions for Structural Steel Buildings. Chicago, IL, 2002.
16. Aschheim MA, Black EF. Yield point spectra for seismic design and rehabilitation. Earthquake Spectra 2000;
16(2):317 – 335.
17. Nassar A, Krawinkler H. Seismic demands for SDOF and MDOF systems, John A. Blume Earthquake
Engineering Center, Report No. 95, Department of Civil Engineering, Stanford University, CA, 1991.
18. Prakash V, Powell GH, Campbell, S. DRAIN-2DX base program description and user guide. Report No.
UCB/SEMM-93/17, Version 1.10, Department of Civil and Environmental Engineering, University of California,
Berkeley, CA, 1993.
19. Adeli H, Cheng N-T. Integrated genetic algorithm for optimization of space structures. Journal of Aerospace
Engineering (ASCE) 1993; 6(4):315 – 328.
20. Cheng FY, Li D. Multiobjective optimization design with Pareto genetic algorithm. Journal of Structural
Engineering (ASCE) 1997; 123(9):1252 – 1261.
21. Goldberg DE. Genetic Algorithms in Search, Optimization and Machine Learning. Addison-Wesley: Reading,
MA, 1989.
22. Deb K. Multi-objective Optimization Using Evolutionary Algorithms. Wiley: Chichester, 2001.
23. Carter CJ, Murray TM, Thornton WA. Economy in steel. Modern Steel Construction 2000; 40(4):39 – 48.
24. Foutch DA, Yun S-Y. Modeling of steel moment frames for seismic loads. Journal of Constructional Steel
Research 2002; 58:529 – 564.
25. Somerville P, Smith N, Puntamurthula S, Sun J. Development of ground motion time histories for phase 2
of the FEMA=SAC steel project. SAC Background Document SAC/BD-97/04, SAC Joint Venture, Richmond,
CA, 1997.
26. Lee K-H, Foutch DA. Performance prediction and evaluation of special moment frame buildings for seismic
loads. SAC Background Document SAC/BD-00/25, SAC Joint Venture, Richmond, CA, 2000.
27. Liu M, Wen YK, Burns SA. Life cycle cost oriented seismic design optimization of steel moment frame
structures with risk-taking preference. Engineering Structures 2004; 26(10):1407–1421.
28. Adeli H, Park HS. Neurocomputing for Design Automation. CRC Press: Boca Raton, FL, 1998.

Multiobjective Optimization For Performance-Based Seismic Design of Steel Moment Frame Structures

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multiobjective Optimization For Performance-Based Seismic Design of Steel Moment Frame Structures

Uploaded by

Copyright:

Available Formats

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2005; 34:289–306

Multiobjective optimization for performance-based seismic

Min Liu1 , Scott A. Burns2; ∗; † and Y. K. Wen3

Urbana; IL 61801; U.S.A.

205 N. Mathews; Urbana; IL 61801; U.S.A.

Contract=grant sponsor: National Science Foundation; contract=grant number: CMS 99-12559

Received 28 October 2003

require explicit evaluation of seismic demands on structural systems at predened perfor-

MULTIOBJECTIVE PERFORMANCE-BASED DESIGN OPTIMIZATION

Merit objective functions

Figure 1. Static pushover analysis and seismic performance indices.

Static pushover analysis

Example steel SMRF

Distribution of optimized designs with respect to primary merit measures

Distribution of optimized designs with respect to secondary merit measures

Figure 4. Distribution of all optimized tradeo designs at the 400th generation

Optimized designs with the same number of dierent section types

Optimized designs with close material weights

Time history analysis of two alternative designs

Copyright ? 2004 John Wiley & Sons, Ltd.

Earthquake Engng Struct. Dyn. 2005; 34:289–306

code specications as well as structural performance consideration. In contrast to stringent

You might also like