You are on page 1of 10

Seismic Design Optimization of Steel Building Frameworks

Lei Xu, M.ASCE1; Yanglin Gong, M.ASCE2; and Donald E. Grierson, M.ASCE3

Abstract: This paper presents a multicriteria optimization method for the performance-based seismic design of steel building frameworks
under 共equivalent static兲 seismic loading. Minimizing structural cost 共interpreted as structural weight兲 is taken as one objective. The other
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

objective concerns minimizing earthquake damage which, since uniform postelastic ductility demand over all stories generally avoids
local weak-story collapse, is interpreted as providing a uniform interstory drift distribution over the height of the building. That is, the
overall objective for the design of a building framework is to have minimum structural weight and uniform plastic ductility demand while,
at the same time, meeting displacement and strength constraints corresponding to the various performance levels. Explicit forms of the
objective functions and constraints in terms of member sizing variables are formulated to enable computer solution for the optimization
model. The concepts are illustrated for three-story and nine-story steel building frame examples.
DOI: 10.1061/共ASCE兲0733-9445共2006兲132:2共277兲
CE Database subject headings: Steel frames; Seismic design; Optimization; Nonlinear analysis.

Introduction while the risk to life is low, repairs may be required before reoc-
cupancy can occur. The CP level is when a building has reached a
Experience in recent California earthquakes, including the 1989 state of impending partial or total collapse, where the structure
Loma Prieta and 1994 Northridge events, indicates that modern has suffered a significant loss of lateral strength and stiffness with
design code provisions in zones of high seismicity are relatively large permanent lateral deformation, but the major components of
reliable in avoiding life-threatening building damage. However, the gravity load carrying system should still continue to carry
the economic losses associated with these earthquakes were un- gravity load demands.
The above descriptions of structural performance are in quali-
acceptably large, as many structures were found with damage that
tative terms, and are only meaningful to provide owners with a
was too costly to repair. As a result, higher-performance levels
basis for selecting among building performance alternatives. For
with the ability to limit property and business interruption losses
design engineers, the descriptions are typically further quantified
need be developed and implemented. This new requirement
in terms of structural performance parameters, such as roof drift,
for better performance has led to the development of the interstory drift, plastic rotation, and strength capacity.
performance-based seismic design methodology. With the emergence of the performance-based design method-
A design performance level is a statement of the desired be- ology, there is a need to develop corresponding analytical meth-
havior of the building, should it experience an earthquake of ods. Among a number of analysis types, equivalent-static push-
specified severity. Building performance is a combination of the over analysis is often viewed as a viable and attractive method
performance of both structural and nonstructural components. because of its simplicity and ability to estimate component and
There are a number of structural performance levels 共or particular system deformation demands with acceptable accuracy without
damage states兲 defined in the literature; e.g., three such levels are the need to conduct full dynamic-response analysis 共FEMA 1997;
FEMA-273 共FEMA 1997兲: immediate occupancy 共IO兲, life safety Krawinkler and Seneviratna 1998; Chopra and Goel 2002兲. The
共LS兲, and collapse prevention 共CP兲. A structure at the IO level has essential feature of pushover analysis is a nonlinear procedure in
sustained minimal or no damage to its structural elements, and is which monotonically increasing lateral loads along with constant
safe to be reoccupied immediately following the earthquake. A gravity loads are applied to a framework until a control node
structure at the LS level has experienced extensive damage and, 共usually referred to as the building roof兲 sways to a predefined
“target” lateral displacement that corresponds to a performance
1
Associate Professor, Dept. of Civil Engineering, Univ. of Waterloo, level.
Waterloo, Ontario Canada N2L 3G1 共corresponding author兲. E-mail: Performance-based design is an iterative procedure, in which
lxu@uwaterloo.ca the initial design is modified repeatedly through the results of
2
Assistant Professor, Dept. of Civil Engineering, Lakehead Univ., analysis to meet code- or designer-specified requirements. The
Thunder Bay, Ontario Canada P7B 5E1. analysis should take into account both geometric and material
3
Professor, Dept. of Civil Engineering, Univ. of Waterloo, Waterloo, nonlinearity at multiple performance 共loading兲 levels, and can in-
Ontario Canada N2L 3G1. volve significant computational effort. In fact, the design optimi-
Note. Associate Editor: Christopher M. Foley. Discussion open until zation of building frameworks using pushover analysis to account
July 1, 2006. Separate discussions must be submitted for individual pa-
for the effects of seismic loading can be extremely computation-
pers. To extend the closing date by one month, a written request must be
filed with the ASCE Managing Editor. The manuscript for this paper was ally intensive, and much research is yet ongoing to establish the
submitted for review and possible publication on April 13, 2004; ap- means to efficiently incorporate pushover analysis together with
proved on May 3, 2005. This paper is part of the Journal of Structural optimal performance-based design. Ganzeri et al. 共2000兲 were the
Engineering, Vol. 132, No. 2, February 1, 2006. ©ASCE, ISSN 0733- first to incorporate pushover analysis and the performance-based
9445/2006/2-277–286/$25.00. design concept in the design optimization of reinforced concrete

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006 / 277

J. Struct. Eng. 2006.132:277-286.


structures. In their paper, the design objective to be minimized precisely defined. As the mathematical formulation of a meaning-
was formulated in terms of the material cost of the structure, and ful cost function in a truly broad context is virtually impossible to
was constrained by limiting the plastic rotations at the ends of the achieve for a structural framing system considered in isolation, as
structural members. However, only the IO performance level was herein, the cost of the members of the structure is alone taken to
explicitly considered in the design, while the LS and CP levels define the explicit cost objective function for this study 共while
were checked by a final pushover analysis. Further, constraints structural fabrication and erection costs are implicitly considered
were not explicitly formulated and no sensitivity information was in the design process through a member grouping technique,
provided. Foley and Schinler 共2003兲 presented a design method- illustrated by the design examples兲. Assuming that the cost of a
ology for partially and fully restrained steel frames using an evo- member is proportional to its material weight, the least-cost
lutionary algorithm. A nonlinear distributed-plasticity analysis design can be interpreted as the least-weight design of the struc-
technique was used to evaluate structural response under gravita- ture, and the cost objective function f 1 共also called the weight
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

tional and wind loads. The design objective to minimize structure objective兲 to be minimized can be formulated as


weight was written as a summation of column weights and modi-
n
fied beam weights. Design constraints included member end
rotation, interstory drift, vertical beam deflection, percentage of f 1共x兲 = 兺
j=1
␳L jA j Wmax 共1兲
cross-section yielding, flange and web slenderness, unbraced
length for columns, and shape constraints. Chan and Zou 共2004兲 where x = generic representation of the vector of design variables;
proposed an optimization algorithm for reinforced concrete struc- n = number of members; L j and A j = length and cross-section area
tures under earthquake loads. They decomposed the design pro- of the jth member, respectively; and ␳ = material density. The
cess into two single-criterion phases. The first phase involved an function f 1 is normalized by the maximum possible weight of the
elastic design optimization in which the cost of concrete is mini- frame Wmax, which is calculated as Wmax = ⌺␳L jAUj , where
mized subject to elastic spectral displacement constraints due to a AUj = upper-bound cross section area for member j.
minor earthquake. The second phase involved minimizing the In addition to minimizing the structure cost, minimizing the
cost of steel reinforcement subject to constraints on inelastic dis- damage to the building superstructure under earthquake loading is
placements. Pushover analysis was performed based on the as- another favorable 共perhaps more desirable兲 design objective.
sumption that the fundamental mode of vibration was the pre- Here, it is required to quantitatively formulate structural damage
dominate response. Though not directly concerning performance- in terms of structural response. One way to quantify the degree of
based design, other structural optimization studies involving damage to a building framework is to establish the relationship
nonlinear analysis can be found in works by Bhatti et al. 共1979兲, between interstory drift and damage. Interstory drift is the pri-
Bhatti and Pister 共1981兲, Haftka 共1989兲, Wu and Arora 共1987兲, mary parameter in evaluating structural performance 共FEMA
Jao and Arora 共1993兲, Xu and Grierson 共1993兲, and Xu et al. 2000兲, and is widely regarded as a major parameter characterizing
共1995兲. the extent of plastic deformation of a building. It is necessary to
This paper develops a computer-automated design method for only consider the plastic interstory drift distribution at extreme
the optimum design of rigid or fully restrained steel moment performance levels, such as the CP level 共see “Examples”兲, since
frameworks under 共equivalent static兲 seismic loading. A newly structural damage in the elastic range of structure response is of
developed pushover analysis technique 共Hasan et al. 2002; Gong minor consequence 共the same cannot be said concerning non-
2003兲 is employed to evaluate seismic demands at various perfor- structural damage, which is not considered by this study兲. It has
mance levels. Newly developed sensitivity analysis techniques been observed in many collapsed structures that deformation con-
共Gong 2003; Gong et al. 2005兲 are employed to establish gradient centration took place at a soft 共or weak兲 story under severe earth-
information that permits the design problem to be explicitly for- quake loading, which directly led to building failure. It is reason-
mulated in terms of member sizing variables. The generalized able to assume that the structure will undergo less damage if this
optimality criteria algorithm DUAL is used to carry out the deformation concentration can be mitigated; i.e., that less damage
computer-based design optimization procedure 共Fluery 1979兲. will occur if the structure exhibits a uniform interstory drift dis-
Three-story and nine-story moment frames are designed to illus- tribution when undergoing significant plasticity, since minimum
trate the developed methodology. plastic deformation concentration occurs when all the stories of a
building have the same interstory drift. Thus, the damage-
mitigating objective can be stated as pursuing a uniform interstory
Design Problem Formulation drift distribution since this is equivalent to achieving a uniform
ductility demand in all stories 共Chopra 2001兲. Since a linear story
drift distribution is equivalent to a uniform interstory drift distri-
Objective Criteria
bution, the structure damage function f 2 to be minimized can be
An objective criterion, often known as a cost or performance defined as

再 冎
function, is expressed in terms of the design variables and serves ns−1 1/2
as a decision motivator. Two objective criteria concerning struc-
ture cost and structural damage are identified for this study. f 2共x兲 = 共1/ns兲 兺
s=1
关共␯sCP共x兲/⌬CP共x兲兲共H/Hs兲 − 1兴 2
共2兲
Minimum structural cost is perhaps the most favorable objec-
tive of a design and is universally adopted for many optimization where ns= number of stories; Hs and H = distances from the build-
problems. Some studies 共Walker 1977; Wen and Kang 1998兲 have ing ground level to story s and the roof, respectively; while ␯sCP共x兲
been conducted to develop a general building cost function. Un- and ⌬CP共x兲 = lateral translations of story s and the building roof at
fortunately, a complete description of the real cost of a building the CP performance level, respectively. In effect, f 2 defines the
before its construction is often nearly impossible because accurate coefficient of variation of the lateral translation distribution, since
cost data require information from many design disciplines, and ␯sCP / Hs and ⌬CP / H represent the story-drift ratio and the building
many factors that influence the cost are unpredictable and not mean-drift ratio, respectively. By definition, the value of f 2共x兲 is

278 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006

J. Struct. Eng. 2006.132:277-286.


not less than zero, and only for the extreme case of a perfectly placements of 0.7, 2.5, and 5% of the height of the building are
uniform interstory drift distribution is f 2共x兲 = 0. adopted by this study as the maximum allowable roof drift for the
IO, LS, and CP performance levels, respectively 共FEMA 1997兲.
The allowable interstory drift is also determined by the perfor-
Design Variables
mance level; e.g., the allowable interstory drift of a low building
For this study, the design variables are taken to be the cross sec- is taken as 1.25% of the height of the story at the IO level, and
tion sizes of the members. There are four basic cross-sectional 6.1% at the CP level 共FEMA 2000兲.
properties for each member, i.e., the area A, moment of inertia I, The interstory and roof drift constraints imposed at each
elastic modulus S 共associated with first yield兲, and plastic modu- performance level are expressed as
lus Z 共associated with full plasticity兲. These four properties for
commercially available standard steel sections can be related to- ␦s共x兲 艋 ¯␦ 共s = 1, ¯ ns兲 共6兲
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

gether through functional relationships which, for this study, are


as expressed as ¯
⌬共x兲 艋 ⌬ 共7兲
2
I = C 1A + C 2A + C 3 共3兲 where ␦s = interstory drift of story s 共i.e., ␦s = ␯s − ␯s−1, the
difference between the drift ␯s at story s and the drift ␯s−1 at story
S = C 4A + C 5 共4兲 s − 1兲; ¯␦ = allowable inter-story drift; and ⌬ and ⌬
¯ = roof drift and
allowable roof drift, respectively.
Z = fs S 共5兲 Sizing limitations imposed on member cross-section areas are
taken, for this study, as the lower and upper bound values AL and
where C1 – C5 = constants determined by regression analysis
AU of the section areas A for the type of commercially available
共Gong 2003兲. For a specified type and nominal depth of section,
standard steel members specified for the design, i.e.
instantaneous updating of the section properties I, S, and Z is
achieved through Eqs. 共3兲–共5兲 for a given section area A. That is, ALj 艋 A j 艋 AUj 共j = 1,2, . . . ,n兲 共8兲
having such relationships, the cross-section area A can be taken as
the only design variable, thereby reducing the number of design
variables significantly. Eqs. 共3兲–共5兲 play an important role in the Design Formulation
design automation process since pushover analysis results are
very sensitive to the sectional properties I, S, and Z, which there- From the foregoing, the optimization model can be formulated as
fore need to be estimated as accurately as possible. 共note that while interstory and roof drift constraints are applied
For this study, the ultimate purpose is to find the optimal de- for all three performance levels IO, LS, and CP, the damage
sign for steel frameworks using commercially available sections. objective function f 2 applies only for the CP level兲.
The design process is divided into two phases: Phase I, continu- Min
ous design stage and Phase II, discrete design stage. The design f共x兲 = f 1 + f 2


problem is first treated as a continuous optimization problem; n
then, a dynamic rounding-off strategy 共Gong 2003兲 is employed
to size design variables in Phase II. It is assumed that the optimal
= ␻1 兺 ␳L jA j Wmax

再 冎
j=1
discrete solution is in the neighborhood of the optimal continuous
ns−1 1/2
solution. The dynamic rounding-off strategy is a modification of
the simple rounding-off method 共a common approach to select + ␻2 共1/ns兲 兺
s=1
关共␯sCP共x兲/⌬CP共x兲兲共H/Hs兲 − 1兴 2
共9兲
cross sections by rounding up all the sections found from the
continuous optimization solution to their nearest available dis- subject to
crete section size兲. While the simple rounding-off method is direct
and easy to implement, it often results in an infeasible design for ␦si共x兲 艋 ¯␦i 共s = 1, . . . ns; i = IO,LS,CP兲 共10兲
a problem having a large number of variables 共Arora 2000兲. It is
also unnecessary to increase all the variables to their upper dis- ¯i
⌬i共x兲 艋 ⌬ 共i = IO,LS,CP兲 共11兲
crete neighbors. Some of them could be decreased to their lower
neighbor. The dynamic rounding-off strategy rounds up only one ALj 艋 A j 艋 AUj 共j = 1,2, . . . ,n兲 关Phase I兴
or a few variables at a time. The selected variable共s兲 are fixed at
the rounded-up discrete value and the design problem is opti- or
mized again with a reduced set of variables. This process is re-
Aj 苸 aj 关Phase II兴 共12兲
peated until all variables are selected and fixed to discrete values.
where superscript i refers to the ith performance level;
f共x兲 = combined 共weight+ damage兲 objective function; a j = set
Design Constraints
of discrete-section areas available for design; and ␻1 and
The design variables are chosen to satisfy certain specified re- ␻2 = combination factors for the objectives f 1共x兲 and f 2共x兲 from
quirements called design constraints. According to the roles they Eqs. 共1兲 and 共2兲, respectively, which turn the design problem into
play in determining the design variables, the constraints can be a single criterion problem 共note that ␻1 and ␻2 are called “com-
classified as being primary or secondary. The primary constraints bination” factors instead of “weight” factors since their values do
considered in this paper serve to control building drift, including not reflect the relative importance of the two objectives兲.
roof drift and interstory drift. The roof drift is taken to be repre- The normalized structural weight f 1 and the coefficient of varia-
sentative of the lateral translation of the building as a whole, tion of lateral displacements f 2 are two different entities and, as
and the maximum allowable roof drift at each performance level such, the appropriate selection for the values of ␻1 and ␻2
is specified by a “target” displacement. For example, target dis- requires numerical experiments that account not only for how f 1

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006 / 279

J. Struct. Eng. 2006.132:277-286.


and f 2 are defined and normalized but also for how much the
gradients of the two objectives conflict with each other 共Gong
2003兲.
In Eqs. 共9兲–共12兲, only the f 1共x兲 term of Eq. 共9兲 is an explicit
function of the design variables, while the f 2共x兲 term of Eq. 共9兲
and all drift constraints are implicit functions of the design vari-
ables. In such a case, to enable computer solution of the design
optimization problem, it is necessary to use an approximation
technique to formulate the objective function f 2共x兲 and all drift
constraints explicitly in terms of design variables. High quality
approximations can be obtained by adopting the reciprocal
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

variables 共Schmit and Farshi 1974兲,

x j = 1/A j 共13兲
and then employing first-order Taylor series to reformulate Eqs. Fig. 1. Flowchart of proposed design algorithm
共9兲–共12兲 as the explicit design optimization problem:
Min
Design Examples

f共x兲 = ␻1 兺
n

␳L j 冒 共x jWmax兲 Example One

冋 册
j=1
Consider the three-story, four-bay moment frame in Fig. 2 共Gupta
n CP
and Krawinkler 1999; Hasan et al. 2002兲. The framework consists
+ ␻2 共f 2共x兲兲0 + 兺
j=1
共df 2共x兲/dx j兲0共x j − x0j 兲 共14兲 of 27 members. Through member linking, the number of design
variables is reduced to 9, designated as A1 – A9 in Fig. 2. There-
subject to fore, from Eq. 共13兲, there are nine reciprocal design variables
x j = 1 / A j, j = 1 , . . . , 9.
n It remains to employ Eqs. 共3兲–共5兲 to express section properties
关␦si共x兲兴0 + 兺
j=1
关d␦si共x兲/dx j兴0共x j − x0j 兲 艋 ¯␦i I j, S j, and Z j in terms of A j for each of the j = 1 , . . . , 9 design
section groups. To this end, it is first necessary to predetermine
the type and nominal depth of the set of sections from which a
共s = 1, ¯ ns, i = IO,LS,CP兲 共15兲 member design is to be selected. While providing considerable a
priori knowledge to the design automation algorithm, such prede-
n termination is entirely consistent with conventional design prac-
关⌬i共x兲兴0 + 兺
j=1
¯i
关d⌬i共x兲/dx j兴0共x j − x0j 兲 艋 ⌬ 共i = IO,LS,CP兲 tice where a designer generally knows which type of section and
approximate nominal depth is used for a certain type of frame
member. For instance, it is well known that W14 sections are
共16兲
commonly used sections for columns of low to mid-rise steel
frameworks, while W24 to W33 sections are often used for beam
xLj 艋 x j 艋 xUj 共j = 1,2, ¯ ,n兲 关Phase I兴 共17兲 members. A designer normally will check the market for avail-
ability before selecting section types and nominal depths. If a
or reasonable section is not able to be chosen from among the pre-
determined sections to meet the design requirements, the prede-
xj 苸 Xj 共j = 1,2, ¯ ,n兲 关Phase II兴 termined sections are upgraded to a larger available nominal
depth and the member design process is repeated.
where superscript 0 represents values for the current design; xLj For this example, all the column design areas A1 – A6 are pre-
and xUj = lower and upper bounds on the reciprocal area x j, respec- determined to be chosen from among available W14 sections,
tively, where xLj = 1 / AUj and xUj = 1 / ALj ; X j = 1 / a j; and d␦ / dx j, while the beam areas A7, A8, and A9 are to be chosen from among
d⌬ / dx j, and df 2 / dx j = interstory drift, roof drift, and ductility de-
mand derivatives with respect to the design variable x j, respec-
tively. The determination of the foregoing displacement and duc-
tility derivatives, also known as sensitivity coefficients, has been
reported in a companion study by the writers 共Gong et al. 2005兲,
where it is noted that their calculation also involves taking into
account the sensitivity of the lateral seismic loading to changes in
the structure design over the iterative synthesis history.
A generalized optimality criteria algorithm, called the Dual
method 共Fleury 1979; Gong 2003兲, is employed by this study to
solve the optimization problem Eqs. 共14兲–共17兲 to find member
sizes for which structure weight and damage are minimized. The
overall design procedure is illustrated by the flowchart in Fig. 1.
See Gong 共2003兲 for full details of the numerical realization of
the design algorithm. Fig. 2. Example one: three-story four-bay moment frame

280 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006

J. Struct. Eng. 2006.132:277-286.


Table 1. Bounds on Member Sizes and applied to the structure according to the load pattern defined
Type and Cross-section by 共FEMA 1997兲
nominal depth Lower/upper area Ai

冉 冒兺 冊
of members section size 共mm2兲 ns
W14 W14⫻ 22 4,190 Ps = Vb GsHsk GmHmk 共s = 1,2, . . . ,ns兲 共20兲
W14⫻ 808 152,900 m=1
W24 W24⫻ 55 10,450 where W = seismic weight of the structure; g is gravitational
W24⫻ 492 92,900 acceleration; Ps = lateral load applied at story level s; Hs and
W30 W30⫻ 90 17,030 Hm = heights from the base of the building to story levels s and m,
W30⫻ 477 90,320 respectively; Gs and Gm = seismic weights for story levels s and
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

W33 W33⫻ 118 22,390 m, respectively; and k = exponent whose value depends on the
W33⫻ 354 67,100 period of the building 共k = 2 for this example兲.
W36 W36⫻ 136 25,610 The constraint Eq. 共15兲 includes nine interstory drift con-
W36⫻ 848 160,640 straints 共i.e., three constraints for each of the IO, LS, and CP
performance levels兲, while Eq. 共16兲 includes three roof drift
constraints 共i.e., one constraint for each performance level兲. The
W33, W30, and W24 sections, respectively 共AISC 1994兲. The allowable interstory drifts ¯␦ are taken to be 0.012h, 0.031h, and
corresponding upper-bound section areas given in Table 1, having 0.061h for the IO, LS, and CP levels, respectively, where h is the
material density ␳ = 7,850 kg/ m3 are used to calculate the maxi- height of the story under consideration, 共FEMA 2000兲. The allow-
mum weight of the structure as Wmax = 143,136 kg which, in turn, able roof drifts for the IO, LS, and CP levels are taken to be
is used to normalize the weight objective function f 1 关see Eq. 共1兲兴. 0.007H, 0.025H, and 0.05H, respectively 共FEMA 1997兲. The ini-
Three acceleration design spectra, which represent three dif- tial design point is taken to be defined by the maximum available
ferent earthquake levels corresponding to 20, 10, and 2% prob- discrete-section sizes for the design variables, i.e., A01 – A06 are
ability of exceedance in a 50 year period, are taken as the basis defined by a W14⫻ 808 section, while A07, A08, and A09 are defined
for calculating the 共equivalent static兲 seismic loading for the three by W33⫻ 354, W30⫻ 477, and W24⫻ 492 sections, respectively
performance levels IO, LS, and CP, respectively. Without loss of 共see Table 1兲.
generality, the calculation of the spectral acceleration Sia for each For pushover analysis, each beam member is modeled by two
design spectra i can be expressed as 共FEMA 1997兲 beam elements by considering the midspan point as a potential
plastic hinge location. 关Actually, it was subsequently found that
one element per beam would have been adequate for this 共and the

冦 冧
Ssi共0.4 + 3T/T io兲 0 ⬍ T 艋 0.2T i0 next兲 design example since the applied gravity loads were rela-
Sia = Ssi 0.2T i0 ⬍ T 艋 T i0 共i = IO,LS,CP兲 tively small such that no plastic hinges were formed at the mid-
span of the beams.兴 Each column member is represented by a
Si1/T T⬎T i
0 single beam–column element since axial forces and, hence, P-␦
共18兲 effects are not significant for this 共and the next兲 framework ex-
ample 关when the axial force in a member is less than 40% of the
where T = elastic fundamental period of the structure computed
Euler’s buckling load, modeling a column with but one element
from structural analysis; Ti0 = period at which the constant accel-
will produce reasonably accurate results 共Chan 2001兲兴.
eration and constant velocity regions of the response spectrum
Second-order and combined bending moment and axial force
intersect for the design earthquake associated with performance
effects are accounted for the columns alone 共Hasan et al. 2002兲.
level i; and Ssi and Si1 = corresponding short-period and 1 s period
The expected yield strength of steel material used for column
response acceleration parameters, respectively 共see Table 2兲. As
members is ␴ye = 397 MPa, while ␴ye = 339 MPa for beam mem-
shown by a companion study 共Gong et al. 2005兲, the period
bers 共Gupta and Krawinkler 1999; Hasan et al. 2002兲. For this
T changes during the design iterative process and the sensitivity
example, the objective function combination factors in Eq. 共14兲
of T to changes in the design variables, dT / dx j, is evaluated so as
are taken to be ␻1 = 0.95 and ␻2 = 0.05 共Gong 2003兲.
to facilitate determination of the corresponding sensitivities of
The results of the performance-based design optimization pro-
structural responses.
cess are summarized in Tables 3 and 4. It is observed that the
Having Sia from Eq. 共18兲, the corresponding base shears Vb for
normalized weight of the structure f 1 is decreased from 1.00 at
the IO, LS, and CP levels are then found as
the first cycle to 0.311 at the tenth cycle for the continuous design
stage. At design cycle 10, both the weight objective f 1 and the
Vib = W共Sia/g兲 共i = IO,LS,CP兲 共19兲
damage objective f 2 converge to their optimal 共continuous-
variable兲 values, and the most critical constraint is active with a
response ratio of unity. The previously described dynamic
Table 2. Performance-Level Site Parameters for Design Examples rounding-off strategy is employed to find the optimal discrete-
section design given in the last row of Table 3, for which, from
Performance Earthquake level Ss S1 T0
the last row of Table 4, the normalized weight objective is slightly
level 共prob. exceed./years兲 共g兲 共g兲 共s兲
increased to f 1 = 0.319 共i.e., the optimal discrete-section weight of
IO 20% / 50 0.33 0.24 0.717 the frame is f 1 ⫻ Wmax = 0.319⫻ 143,136= 45,660 kg兲. A final
LS 10% / 50 0.45 0.31 0.690 pushover analysis for this discrete design is conducted and the
CP 2 % / 50 0.70 0.45 0.637 plastic states of the structure corresponding to the LS and CP
Note: IO= immediate occupancy; LS= life safety; CP= collapse performance levels 共i.e., 2.5 and 5% roof drift兲 are found to be as
prevention; prob. exceed./ years= probability of exceedance/years. shown in Figs. 3 and 4, respectively, where the ovals represent

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006 / 281

J. Struct. Eng. 2006.132:277-286.


Table 3. Example One: Primary Design Variable Results
Cycle A1 A2 A3 A4 A5 A6 A7 A8 A9
index 共mm2兲 共mm2兲 共mm2兲 共mm2兲 共mm2兲 共mm2兲 共mm2兲 共mm2兲 共mm2兲
1 152,900 152,900 152,900 152,900 152,900 152,900 67,100 90,320 92,900
2 93,140 94,400 89,340 94,440 104,800 99,020 45,470 60,512 55,545
3 68,335 65,165 58,595 71,965 82,930 73,660 36,200 46,110 36,910
4 67,440 49,605 42,595 74,970 71,835 59,624 35,095 37,093 26,600
5 61,450 62,340 35,560 71,040 69,845 55,350 34,335 33,755 22,025
6 58,375 40,225 33,780 67,490 66,350 52,585 32,620 32,070 20,920
7 55,460 38,210 32,100 64,115 63,030 49,960 30,985 30,465 19,880
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

8 53,240 36,680 30,810 61,550 61,300 48,080 29,750 29,250 19,080


9 50,820 35,020 29,650 58,760 61,180 49,065 29,350 27,920 18,210
10 50,010 34,460 29,860 57,820 61,250 49,850 29,825 27,680 17,870
Discrete 48,770 33,420 33,420 58,970 65,160 48,770 31,930 28,060 17,900
design W14⫻ 257 W14⫻ 176 W14⫻ 176 W14⫻ 311 W14⫻ 342 W14⫻ 257 W33⫻ 169 W30⫻ 148 W24⫻ 94

plastic hinges at the member ends and the digits inside the ovals Example Two
represent the percentage of plastification 共see Hasan et al. 2002兲.
Consider the nine-story framework in Fig. 5 共Hasan et al. 2002兲.
The interstory drift ratios at 5% roof drift are found to be 0.048,
The framework consists of 99 members. Through member link-
0.052, and 0.049 for the first, second, and third story, respec-
ing, the number of design variables is reduced to 18, designated
tively; i.e., at the CP level, the interstory drift distribution is near
as A1 – A18 in Fig. 5. The column design variables A1 – A9 are
uniform and, therefore, the ductility demand is near uniform over
chosen from among available W14 sections, while beams
the height of the building 共see the last column of Table 4 where
A10 – A13, A14, A15 – A16, and A17 – A18 are chosen from among W36,
the coefficient of variation of interstory drift is f 2 = 0.022兲. It is
W33, W30, and W24 sections, respectively. The initial design
seen from Fig. 4 that member plastification takes place somewhat
point is taken to be defined by the maximum available discrete-
uniformly over the structure height when the frame undergoes
section sizes for the design variables. The corresponding upper
nearly uniform interstory drift.
bound section areas are given in Table 1. The maximum weight of
the structure used to normalize the weight objective function is
Wmax = 655,540 kg.
The material strength, design spectra, and allowable roof and
Table 4. Example One: Design History Results
interstory drifts are taken to be the same as those for Example 1.
Roof drift ratio ⌬ / H The load pattern is determined by Eq. 共20兲 using k = 2. The ob-
Critical interstory jective function combination factors in Eq. 共14兲 are taken to be
drift ratio ␻1 = 0.95 and ␻2 = 0.05.
Design ␦s / hs 关at story s兴 Structural Ductility
The optimal design for the framework is given in Table 5 in
cycle weight/Wmaxa demand
index IO LS CP 共f 1兲 共f 2兲 terms of both member cross-section areas A and corresponding
commercial standard section designations. The optimal normal-
1 0.0020 0.0027 0.0042 1.000 0.067 ized value of the weight objective is f 1 = 0.26 共i.e., the optimal
0.0024 关2兴 0.0033 关2兴 0.0051 关2兴 discrete-section weight of the frame is f 1 ⫻ Wmax = 0.26
2 0.0035 0.0047 0.0072 0.639 0.047 ⫻ 655,540= 170,440 kg兲. The optimal value of the damage objec-
0.0040 关2兴 0.0055 关2兴 0.0084 关2兴 tive is f 2 = 0.045 共i.e., the coefficient of variation of interstory drift
3 0.0049 0.0067 0.0151 0.471 0.164 over the structure height兲. The most critical constraint is the roof
0.0056 关2兴 0.0076 关2兴 0.0185 关1兴
drift at the IO level, which is almost active with a response ratio
4 0.0058 0.0078 0.0162 0.402 0.045 of 0.98. The plastic states of the optimal discrete-section design
0.0067 关2兴 0.0089 关2兴 0.0188 关2兴
corresponding to the LS and CP performance levels 共i.e., 2.5
5 0.0063 0.0082 0.0192 0.369 0.037 and 5% roof drift兲 are found to be as shown in Figs. 6 and 7,
0.0070 关2兴 0.0092 关2兴 0.0217 关2兴
6 0.0065 0.0085 0.0238 0.351 0.035
0.0073 关2兴 0.0095 关2兴 0.0270 关2兴
7 0.0067 0.0087 0.0305 0.333 0.037
0.0075 关2兴 0.0098 关2兴 0.0349 关2兴
8 0.0068 0.0089 0.0383 0.320 0.039
0.0077 关3兴 0.0101 关3兴 0.0430 关2兴
9 0.0070 0.0091 0.0471 0.312 0.032
0.0078 关3兴 0.0101 关3兴 0.0518 关2兴
10 0.0070 0.0091 0.0476 0.311 0.026
0.0077 关3兴 0.0101 关3兴 0.0510 关2兴
Discrete 0.0069 0.0090 0.0456 0.319 0.022
design 0.0078 关3兴 0.0102 关3兴 0.0481 关2兴
a
Wmax = 143,136 kg. Fig. 3. Example one: structural plastification at life safety level

282 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006

J. Struct. Eng. 2006.132:277-286.


Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Example one: structural plastification at collapse prevention


level

respectively. The interstory drift ratios at 5% roof drift are found


to be 0.051, 0.050, 0.048, 0.059, 0.056, 0.059, 0.051, 0.050, and
0.026 from the first to the ninth story, respectively; i.e., the plastic
drift distribution is nearly uniform, except at the ninth story where
the interstory drift is smaller as the columns there experience no
plastification 共see Fig. 7兲. For member plastification taken as a
measure of damage, it is observed from Figs. 6 and 7 that damage
is somewhat uniformly distributed over the first eight stories of
the structure where the frame undergoes nearly uniform interstory Fig. 6. Example two: structural plastification at life safety level
drift, and that damage of the ninth story is confined to roof beams.

structural members in search of a uniform-ductility demand de-


sign for the building framework. Two design examples illustrated
Discussion and Conclusions the applicability and practicability of the design optimization
method.
The paper has presented a computer-based synthesis methodology While the pushover analysis technique is widely accepted as a
for the performance-based optimal design of steel building frame- nonlinear procedure capable of evaluating seismic demands with
works subject to earthquakes. Pushover analysis is employed to reasonable accuracy, it has some shortcomings. The pros and cons
establish plastic behavior under equivalent-static seismic loading. of pushover analysis have been well discussed by Krawinkler and
The study demonstrates that it is possible to simultaneously ac- Seneviratna 共1998兲. The major drawback of conventional push-
count for multiple performance levels and design such as to re-
duce both structural cost 共weight兲 and damage 共ductility demand兲.
The damage objective defines the coefficient of variation of
interstory drift, and plays the role of allocating material among

Fig. 7. Example two: structural plastification at collapse prevention


Fig. 5. Example two: nine-story five-bay moment frame level

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006 / 283

J. Struct. Eng. 2006.132:277-286.


W14⫻ 370 W14⫻ 233 W14⫻ 211 W14⫻ 159 W14⫻ 455 W14⫻ 342 W14⫻ 311 W14⫻ 257 W14⫻ 159 W36⫻ 210 W36⫻ 182 W36⫻ 182 W36⫻ 170 W33⫻ 152 W30⫻ 148 W30⫻ 116 W24⫻ 94 W24⫻ 62
over analysis is that the pattern of lateral inertial loads is pre-

17,900 11,740
共mm2兲
A18
scribed to be invariant, which implies that seismic demands are
dominated by one vibration mode 共typically the fundamental
共mm2兲 mode兲. Chopra and Goel 共2002兲 have proposed a modal pushover
analysis 共MPA兲 technique accounting for several vibration mode
A17

effects that is a significant improvement over single-mode push-


over analysis. Ongoing research by the writers suggests that this
22,060
共mm2兲

MPA technique is readily implemented for the design optimiza-


A16

tion procedure presented herein 共Grierson et al. 2005兲.


The exponent k value in Eq. 共20兲 is a function of the natural
period of the structure 共FEMA 1997兲. For this study, it is taken as
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

28,060
共mm2兲
A15

a fixed value for each design example even though the natural
period changes from one design iteration to the next. This treat-
ment appears to be consistent with current practice employed by
28,840
共mm2兲

the engineering community. For example, using the National


A14

building code of Canada 共NRC 1995兲, the natural period of a


steel moment frame is approximated by an empirical formula
without explicit reference to the actual design of the frame, and
32,260
共mm2兲
A13

the k value is then determined using this estimated period value.


Alternatively, though not done herein, the value of k can be
treated as a variable during the design process since it is a func-
tion of T and the sensitivity of T is already available 共Gong et al.
34,580
共mm2兲
A12

2005兲.
Since stiffness constraints concerning building roof drift and
interstory drift are the primary design requirements considered by
34,580
共mm2兲
A11

this study, the proposed design optimization procedure is most


applicable for steel moment frames in high seismic zones, where
member sectional sizes are almost exclusively determined by drift
requirements 共Foutch and Yun 2002兲. If a few member stress
39,870
共mm2兲
A10

requirements are subsequently found to be violated, often the af-


fected sections can be appropriately resized without the need for a
complete redesign of the structure. Alternatively, though not done
30,130
共mm2兲

herein, it is possible to directly account for strength constraints


A9

concerning stability and stresses explicitly in the design optimi-


zation procedure 共Chan et al. 1995兲.
48,770
共mm2兲

Whether the column base is pinned or fixed does not affect the
A8

optimization algorithm. In fact, when the example structures were


at the collapse prevention performance level, the column bases
were all at almost 100% plastification 共i.e., virtually pinned兲 and
58,960
共mm2兲

the algorithm did not encounter difficulty during any of the design
A7

iterations over the loading history. New research currently under-


way to extend the design automation to steel frameworks with
65,160
共mm2兲

“semirigid” connections also finds this to be true for the case of


A6

initially “partially restrained” column bases.


The strong-column weak-beam 共SC/WB兲 requirement that is
Table 5. Example Two: Discrete-Section Design Results

mandatory for certain steel moment frameworks in seismic zones


86,450
共mm2兲

共FEMA 2000兲 was not included in the design formulation pre-


A5

sented by this study. The SC/WB requirement imposes additional


member-capacity constraints on the design process, and a pilot
30,130
共mm2兲

study has found that such constraints tend to generate an optimal


A4

design having less plastification occurring in column members;


specifically, a heavier-weight structure with reduced likelihood of
a soft or weak story 共Gong 2003兲. Research is currently underway
40,000
共mm2兲

by the writers toward integrating the SC/WB concept into the


A3

design optimization formulation.


44,190
共mm2兲
A2

Acknowledgments
70,320
共mm2兲

This work is sponsored by the National Science and Engineering


Research Council of Canada, and forms part of the PhD research
A1

284 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006

J. Struct. Eng. 2006.132:277-286.


study of the second writer under the supervision of the other two References
writers.
American Institute of Steel Construction, Inc. 共AISC兲. 共1994兲. Manual of
steel construction, load & resistance factor design, 2nd Ed., Vol. I,
Chicago.
Notation Arora, J. S. 共2002兲. “Methods for discrete variable structural optimiza-
tion.” Proc., Structures Congress 2000, ASCE, Reston, Va.
The following symbols are used in this paper: Bhatti, M. A., and Pister, K. S. 共1981兲. “A dual criteria approach for
A ⫽ cross-sectional area; optimal design of earthquake resistant structural systems.” Earth-
a j ⫽ set of discrete commercial-section areas quake Eng. Struct. Dyn., 9, 557–572.
possible for member j; Bhatti, M. A., Pister, K. S., and Polak, E. 共1979兲. “Optimization of con-
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

Ck ⫽ constants for sectional properties; trol devices in base isolation systems for aseismic design.” Proc.,
CP ⫽ collapse prevention level; IUTAM Symp. Structural Control, Univ. of Waterloo, Waterloo,
d共 兲 / dx ⫽ derivative of structural response; Canada.
f ⫽ combined objective function; Chan, C.-M., Grierson, D. E., and Sherbourne, A. N. 共1995兲. “Automatic
f s ⫽ shape factor of cross section; optimal design of tall steel building frameworks.” J. Struct. Eng.,
f 1 ⫽ weight objective function; 121共5兲, 838–847.
f 2 ⫽ damage objective function; Chan, C.-M., and Zou, X.-K. 共2004兲. “Elastic and inelastic drift perfor-
G ⫽ seismic weight of story; mance optimization for reinforced concrete buildings under earth-
H ⫽ height of building; quake loads.” Earthquake Eng. Struct. Dyn., 33, 929–950.
Chan, S. L. 共2001兲. “Nonlinear behaviour and design of steel structures.”
Hs ⫽ vertical distance from base of building to
J. Constr. Steel Res., 57, 1217–1231.
story s;
Chopra, A. K. 共2001兲. Dynamics of structures, theory and applications to
I ⫽ moment of inertia of cross section;
earthquake engineering, Prentice–Hall, Upper Saddle River, N.J.
IO ⫽ immediate occupancy level;
Chopra, A. K., and Goel, R. K. 共2002兲. “A modal pushover analysis
k ⫽ constant depending on period of structure; procedure for estimating seismic demands for buildings.” Earthquake
L ⫽ length of member; Eng. Struct. Dyn., 31, 561–582.
LS ⫽ life safety level; Federal Emergency Management Agency 共FEMA兲. 共1997兲. “NEHRP
n ⫽ number of members; guideline for the seismic rehabilitation of buildings.” FEMA-273,
ns ⫽ number of building stories; Building Seismic Safety Council, Washington, D.C.
P ⫽ lateral earthquake load applied to one story; Federal Emergency Management Agency 共FEMA兲. 共2000兲. “Recom-
S ⫽ elastic modulus of cross section; mended seismic design criteria for new steel moment-frame build-
Sa ⫽ spectral acceleration; ings.” FEMA-350, SAC Joint Venture, Washington, D.C.
Ss ⫽ short-period response acceleration; Fleury, C. 共1979兲. “Structural weight optimization by dual methods of
S1 ⫽ 1 s period response acceleration; convex programming.” Int. J. Numer. Methods Eng., 14, 1761–1783.
T ⫽ elastic fundamental period of structure; Foley, C. M., and Schinler, D. 共2003兲. “Automated design of steel frames
T0 ⫽ period at which constant acceleration and using advanced analysis and object-oriented evolutionary computa-
tion.” J. Struct. Eng., 129共5兲, 648–660.
constant velocity regions of response
Foutch, D. A., and Yun, S. Y. 共2002兲. “Modeling of Steel Moment Frames
spectrum intersect; for Seismic Loads.” J. Constr. Steel Res., 58, 529–564.
Vb ⫽ design base shear; Ganzeri, S., Pantelides, C. P., and Reaveley, L. D. 共2000兲. “Performance-
W ⫽ seismic weight of structure; based design using structural optimization.” Earthquake Eng. Struct.
Wmax ⫽ maximum possible weight of framework; Dyn., 29, 1677–1690.
X j ⫽ set of reciprocal discrete commercial-section Gong, Y. 共2003兲. “Performance-based design of building frameworks
areas for variable j; under seismic loading.” PhD thesis, Univ. of Waterloo, Waterloo,
x ⫽ vector of design variables; Canada 具http://flash.lakeheadu.ca/~ygong/典 共May 26, 2005兲.
x ⫽ reciprocal variable; Gong, Y., Xu, L., and Grierson, D. E. 共2005兲. “Performance-Based Sen-
Z ⫽ plastic modulus of cross section; sitivity Analysis of Steel Moment Frameworks Under Seismic Load-
⌬ ⫽ roof drift; ing.” Int. J. Numer. Methods Eng., 63共9兲, 1229–1249.
¯ ⫽ allowable roof drift; Grierson, D. E., Gong, Y., and Xu, L. 共2005兲. “Optimal Performance-
⌬ Based Design Using Modal Pushover Analysis.” J. Earthquake Eng.,
␦ ⫽ interstory drift; in press.
¯␦ ⫽ allowable interstory drift; Gupta, A., and Krawinkler, H. 共1999兲. “Seismic demands for perfor-
␯s ⫽ lateral deflection of story s; mance evaluation of steel moment resisting frame structures.” SAC
␳ ⫽ steel mass density; and 5.4.3, Rep. No. 132, Stanford Univ., Stanford, Calif.
␻1 , ␻2 ⫽ combination factors for objective functions. Haftka, R. T. 共1989兲. “Integrated nonlinear structural analysis and de-
sign.” AIAA J., 27共11兲, 1622–1627.
Subscripts Hasan, R., Xu, L., and Grierson, D. E. 共2002兲. “Pushover analysis
for performance-based seismic design.” Comput. Struct., 80共31兲,
j ⫽ structural member index; and
2483–2493.
m , s ⫽ building story index.
Jao, S. Y., and Arora, J. S. 共1993兲. “Design optimization of nonlinear
Superscripts structures with rate-dependent and rate-independent constitutive mod-
els.” Int. J. Numer. Methods Eng., 36, 2805–2823.
CP ⫽ collapse prevention level; Krawinkler, H., and Seneviratna, G. D. P. K. 共1998兲. “Pros and Cons of a
L ⫽ lower-bound value; Pushover Analysis of Seismic Performance Evaluation.” Eng. Struct.,
U ⫽ upper-bound value; and 20共4–6兲, 452–464.
0 ⫽ structural response of current design. National Research Council 共NRC兲. 共1995兲. National building code of

JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006 / 285

J. Struct. Eng. 2006.132:277-286.


Canada, Ottawa. T132-2, Elsevier Science Ltd., New York.
Schmit, L. A., and Farshi, B. 共1974兲. “Some approximation concepts for Wu, C. C., and Arora, J. S. 共1987兲. “Design Sensitivity Analysis and
structural synthesis.” AIAA J., 12共5兲, 692–699. Optimization of Nonlinear Structural Response Using Incremental
Walker, N. D. 共1977兲. “Automated design of earthquake resistant multi- Procedure.” AIAA J., 25共8兲, 1118–1125.
story steel building frames.” Rep. No. UCB/EERC 77-12, Earthquake Xu, L., and Grierson, D. E. 共1993兲. “Computer-automated design of semi-
Engineering Research Center, Univ. of California, Berkeley, Calif. rigid steel frameworks.” J. Struct. Eng., 119共6兲, 1740–1760.
Wen, Y. K., and Kang, Y. J. 共1998兲. “Design criteria based on minimum Xu, L., Sherbourne, A., and Grierson, D. E. 共1995兲. “Optimal cost design
expected life-cycle cost.” Struct. Eng. World Wide 1998, Paper Ref.: of semi-rigid low-rise industrial frames.” Eng. J., 3, 87–97.
Downloaded from ascelibrary.org by UNIVERSITY OF REGINA LIBRARY on 06/11/13. Copyright ASCE. For personal use only; all rights reserved.

286 / JOURNAL OF STRUCTURAL ENGINEERING © ASCE / FEBRUARY 2006

J. Struct. Eng. 2006.132:277-286.

You might also like