Professional Documents
Culture Documents
n
int
is the mean distance of interior Pareto solu-
tions, and n
int
is the number of interior Pareto solutions;
d
i
P
no
j1
f
i1
j
f
i1
j
is the crowding distance of Pareto design i, de-
ned as the objective difference of its two neighboring solutions
(i + 1) and (i 1); n
o
is the number of objectives which is equal to
3 for this study; d
ex
is the distance between extreme Pareto solu-
tions; and d
a
P
int
d
i
d
m
2
.
n
int
r
for interior Pareto solutions.
Interested readers should refer to MATLAB for a more detailed
description of s
p
.
A design iterative process is considered to converge if the
change of s
p
value from one generation to next is less than a spec-
ied tolerance. Alternatively, a design process may be terminated if
a specied maximum number of generation is reached.
4. Numerical example
4.1. Problem description
Optimal design solutions will be found for a hypothetical three-
story ofce building located in Vancouver, British Columbia,
Canada. The plan layout, as shown in Fig. 2, is symmetric. In each
principal direction, seismic lateral loads are resisted by two pairs of
one-bay eccentrically-braced frames (EBF) located at the perimeter
of the building. For illustration, only the design of the EBF in north
south direction is described herein. The schematic side view of a
frame is illustrated in Fig. 3. The bay size is 9.14 m center-to-center
and all three stories are each 3.96 mhigh. The columns of the frame,
whose bases are pinned-supported, are continuous over the height.
At each story, a symmetrically opposed brace pair is used to main-
tain the overall symmetry of the SFRS. Braces are concentrically
connected to the columns. The brace and beam to column connec-
tions are assumed to be pinned. But brace-to-beam and link con-
nections are designed as rigid, which allows a brace to contribute
to the resisting of the moment developed at the end of link. The ax-
ial deformation of the beams and links will be included in the anal-
ysis model since large axial force may develop in these members.
Fictitious lean-on columns are used to account for the seismic
weight and the destabilizing effects of the gravity loads of interior
simple frames. In this example, two lean-on columns, one at each
side of the frame, are employed to closely mimic the direction of
the inertia loads from the interior gravity frames.
The seismic weights, which are used to form the mass matrix in
NRH analysis, are given in Fig. 3b. The accompanying gravity loads
are given in Fig. 3c where Q
1
to Q
3
are the gravity loads directly ap-
plied to the frame members and P
1
to P
3
are the gravity loads of
interior frames which generate destabilizing effects on the EBF.
Note that the load factors for the limit state design philosophy
are already included in the computation of the seismic weights
and the gravity loads.
Since the gravity columns are designed as continuous along the
height, their bending due to building drift has an impact on the
overall lateral stiffness of the structure. To account for this effect,
the sectional properties of the ctitious columns, including mo-
ment of inertia and cross-sectional area, are taken to be the sum
of the corresponding values of all the gravity columns (which are
HSS254 254 13 [20]) divided by the number of EBFs (which
is four) in the considered direction.
The design spectrum for Vancouver specied by the National
Building Code of Canada [1] is adopted as the datum for scaling
ground motions. The spectral acceleration values are 0.97g, 0.65g,
0.32g, 0.17g, and 0.085g at period T = 0.2, 0.5, 1.0, 2.0, and 4.0 s,
respectively (g is the gravity acceleration). Three ground motions
(i.e., n
g
= 3) are adopted from PEER [21] for this example though
more ground motions are desirable (e.g., FEMA-450 [6] requires se-
ven) to establish average values of the structural response with the
view to mitigate calculation burden. The selected ground motion
time-histories need to be scaled such that their response spectrum
is compatible with the design spectrum throughout the period
range of interest. The names of the adopted ground motions and
their scale factors are shown in Table 1. Illustrated in Fig. 4 are
the response spectra of the scaled ground motions. Note that these
response spectra are equal to or greater than the design spectrum
value throughout the period ranging from 0.3 s to 1.25 s which is
estimated to cover the rst period of all the possible design
solutions.
In a limit state design, member strength equation is generally
written as
P
a
i
S
i
6 /R
n
where / and a
i
are resistance factor and
load factors, respectively; S
i
are load effects; and R
n
is nominal
resistance. To be consistent with the limit state design philosophy,
a resistance factor needs to be included in this study. For an inelas-
tic-analysis-based design method, it is more convenient to move
the resistance factor to the left side of the strength equation, i.e.,
to amplify the load effects by 1//. Such obtained load effects are
called required nominal strength for members. For this study, the
ground motion time-histories are amplied by 1//. Assume /
= 0.9, the ground motions in Table 1 are further multiplied by
1.11 for the design of the EBF.
The design variables are the cross sections of links, beams out-
side the link, braces, and columns. The length of a link is not treated
as a design variable in this example although the determination of
link length is related to the link section. In general, link length e
should be less than 1:6M
p
=V
p
(where M
p
and V
p
are the plastic mo-
ment and shear resistance of the link, respectively) for the link to
act as a shear hinge. Herein, link length e is pre-determined to sat-
isfy this shear hinge requirement. If a link and the beamoutside the
link on the same oor level use one steel component, they are trea-
ted as one design variable. However, one may nd it advantageous
to design a link and the beam outside the link with different sec-
tions due to the following reasons: (1) it is often necessary to have
a stronger beam outside the link to resist the combined effect of
large axial force and bending moment and (2) it allows the practice
of using replaceable links [22].
When a link and the beam outside the link are made using a sin-
gle steel component, a common practice for obtaining a stronger
beam outside of the link is to use ange cover plates. In this case,
the strategy to relate the sectional properties of the link and the
beam outside the link is as follows. Assume the cross sectional
EBF
EBF
EBF EBF
EBF
EBF
EBF
EBF
Fig. 2. Plane view of the three-story building.
Y. Gong et al. / Engineering Structures 48 (2013) 2836 31
area, plastic modulus, and moment of inertia of a link are A
n
, Z
n
,
and I
n
, respectively. The beam outside the link has a cross sectional
area of (1 + k
n
)A
n
where k
n
P0. The total sectional area of the two
ange cover plates is k
n
A
n
. The area of one cover plate is (k
n
A
n
)/2.
The plastic modulus and moment of inertia of the beam outside
the link are Z
n
k
n
A
n
=2d
n
t
c
and I
n
k
n
A
n
=4d
n
t
c
2
,
respectively, where d
n
is link section depth and t
c
is cover plate
thickness.
There are three ways to implement the above strategy. First, a
link and the beam outside the link use the same design variable
with a pre-determined k
n
value. The cover plate size for each link
is thus dependent on the section size of the link itself. Second, a
link and the beam outside the link use the same design variable
with a pre-determined cover plate size. The value of k
n
is thus
dependent on the size of the link. Third, a link and the beam out-
side the link use separate design variables. For the third case, it
is necessary to establish a cover plate size set. The link section is
a design variable while the beam outside the link is replaced by
the cover plate size variable which is chosen from among the cover
plate size set.
For the illustration of this example, six design variables are cho-
sen. All the columns are designed to have the same section and
thus grouped as the rst variable. For each oor, the link and the
beams outside the link are designed as one component and thus
are treated as one variable (if necessary, the beams will be rein-
forced with ange cover plates using the second strategy), which
gives design variables 2, 3 and 4. The two braces at story 1 are
grouped as the fth variable while the braces at stories 2 and 3
are grouped as the sixth variable. This member grouping technique
Table 2
Section sets for design variables.
Variable Section C
i
A
(mm
2
)
Z
x
(10
3
mm
3
)
I
(10
6
mm
4
)
Columns W310 107 13600 1770 248
W310 79 10100 1280 177
W310 52 6670 841 119
W310 33 4180 480 65
Link and beams
LK1
W410 100 12700 2130 398
W410 74 9550 1510 275
W410 60 7580 1190 216
W410 46 5890 885 156
Link and beams
LK2
W410 74 9550 1510 275
W410 60 7580 1190 216
W410 46 5890 885 156
W360 39 4980 662 102
Link and beams
LK3
W410 60 7580 1190 216
W410 46 5890 885 156
W360 39 4980 662 102
W360 33 4170 542 82.7
Braces of story 1 HSS203 203 13 9260 651 54.7
HSS203 203 9.5 7150 513 43.9
HSS178 178 8.0 5240 330 24.8
HSS178 178 6.4 4250 271 20.6
Braces of stories 2
and 3
HSS203 203 9.5 7150 513 43.9
HSS178 178 8.0 5240 330 24.8
HSS178 178 6.4 4250 271 20.6
HSS152 152 6.4 3610 196 12.6
Q3
Fictitious
column
Loading on EBF
Fictitious
column
Q2
Q1
P3
P2
P1
P3
P2
P1
P3 = 1100 kN
BE2
P2 = 1321 kN
P1 = 1321 kN
Q3 = 24.0kN/m
Q2 = 29.0 kN/m
Q1 = 29.0 kN/m
BE3
BE1
LK1
LK2
LK3
BR1
BR2
BR3
CL CL
CL
CL
CL
CL
w3
Seismic weight of EBF
w2
w1
D3
D2
D1
D3
D2
D1
D3 = 990 kN
D2 = 1046 kN
D1 = 1046 kN
w3 = 19.7kN/m
w2 = 20.8 kN/m
w1 = 20.8 kN/m
(a) (b) (c)
Fig. 3. Side view of the EW direction MRFs: (a) member numbering, (b) seismic weight distribution, and (c) accompanying gravity loads.
Fig. 4. Design and response spectra.
Table 1
Selected ground motion records.
Record name Earthquake magnitude PGA (g) PGV (cm/s) Scale factor
1979 Imperial Valley: El Centro Array #12 6.5 0.143 17.6 3.2
1989 Loma Prieta: Belmont Envirotech 6.9 0.108 11.8 4.2
1994 Northridge: Old Ridge RT 090 6.7 0.568 52.1 0.8
32 Y. Gong et al. / Engineering Structures 48 (2013) 2836
not only reduces the number of design variables (comparing with
the number of frame members) but also reects the actual con-
struction practice.
Each design variable is to be selected fromamong four candidate
section sizes (though more candidate sections can be added, this
example uses four sections only to mitigate computational burden).
The properties of the sections are provided in Table 2 where the
shape for columns and beams is wide-ange I-section and the shape
for braces is hollow structural section. These candidate sections are
chosen because they are commonly used in construction and they
are close to an estimation of the sizes. If the governing design code
has requirements on section class or slenderness, these require-
ments can be easily accounted for in the selection of the candidate
sections. For example, only compact sections are allowed for the
columns and braces according to Canadian standard CSA/S16-09
[4]. Note that developing an automated selection of section sets is
beyond the scope of this paper, and it is a topic for future work.
The length of all the links are taken to be 780 mm to meet the
shear hinge requirement. The inter-story drift constraint limit is ta-
ken as 2.5% of story height. The link rotation capacity or link rota-
tion limit is taken as 0.08 radians as per CSA/S16-09 [4]. The plastic
rotation limits for columns, beams, and braces are taken as 0.005,
0.005, and 0.010 radians, respectively, as these non-fuse members
(modeled as beamcolumns) are allowed to undergo some minor
plastication corresponding to the Immediate Occupancy level as
per ASCE/SEI 41-06 [14] (the Immediate Occupancy represents a
performance level in which the structure experiences little or no
damage). Note that the plastic deformation limits of non-fuse
members are not required for elastic-analysis-based design meth-
ods since the explicit strength checking is required.
4.2. Analysis model
This study adopts OpenSees software [23] to conduct NRH anal-
ysis. The accuracy of the analytical model of fuse links is essential
for the reliability of the analytical results. To this end, the shear-
link element developed by Ramadan and Ghobarah [24] and later
modied by Richards and Uang [25], as shown in Fig. 5, was cho-
sen. The link model is a hybrid element consisting of a linear elastic
beamcolumn with a nonlinear hinge at each end on which both
the plastic shear and bending deformation concentrate. The mul-
ti-linear functions (Fig. 6), describing the yielding and strain hard-
ening behavior of the nonlinear hinges, are dened by the
following parameters [25].
V
y1
1:1V
y
; V
y2
1:3V
y
; V
y3
1:5V
y
; V
y
V
p
7
M
y1
M
y
; M
y2
1:03M
y
; M
y3
1:06M
y
; M
y
M
p
8
K
1m
2GA
w
=e; K
2m
0:03K
1m
; K
3m
0:015K
1m
; K
4m
0:002K
1m
9
K
1M
6EI=e; K
2M
0:03K
1m
; K
3m
0:015K
1M
; K
4M
0:002K
1M
10
where the plastic moment capacity M
p
= F
ye
Z
b
; F
ye
and Z
b
are the ex-
pected yield strength and plastic modulus of the section. The plastic
shear capacity V
p
F
ye
=
3
p
d 2t
f
t
w
; d, t
f
and t
w
are the section
depth, ange thickness and web thickness, respectively; E and G
are Youngs modulus and shear modulus, respectively; I and A
w
are the moment of inertia and area of the web of the link section,
respectively.
Fig. 7 illustrates the traditional method to obtain link rotation
demand in which the EBF is assumed to experience the target plas-
tic mechanism. By ignoring elastic deformation, the plastic rotation
of the shear link is obtained as c L=eh
p
where h
p
is the overall
drift angle [26]. For this study, the traditional method needs to
be modied since the target plastic mechanism is not warranted
and the inter-story drift angle may not be uniform along building
height. The new equation for link rotation demand is
c
y
D
0 y
C
0
e
h
p
11
where y
D
0 and y
C
0 are the vertical displacements of the master nodes
(Figs. 5 and 7) of the link element. Using c L=eh
p
, Eq. (11) can be
re-written as
c
y
D
0 y
C
0
e
L
L e
12
Eq. (12) expresses link rotation demand as a function of the link end
displacements only. Thus, the equation is applicable whether Fig. 7
mechanism exists or not.
Since inelastic deformations may occur in non-fuse members
for a candidate design solution, the non-fuse member elements
must also be able to model material yielding in the analysis proce-
dure. To this end, the exibility-based nonlinear beamcolumn
element in OpenSees is employed. This element uses ber discret-
ization of the cross section [23] in which each ange and web is
discretized into eight layers as recommended by Filippou and
Fenves [27]. Each ber mimics the material behavior of steel by a Fig. 5. Shear link element.
Fig. 6. Multi-linear forcedeformation relationship of link model.
L
p
D
C
Fig. 7. Link rotation.
Y. Gong et al. / Engineering Structures 48 (2013) 2836 33
bilinear elastoplastic stressstrain relationship with 5% strain
hardening as suggested by OpenSees material model STEEL02.
The interaction of axial force and bending moment on a cross
section is explicitly considered through the integration of bers
material response over the section. Each beamcolumn element
further uses ve numerical integration points along the element
length to form its exibility matrix.
For this example, each non-fuse member is represented by two
beamcolumn elements which allows the modeling of initial
imperfection through directly adjusting the coordinates of element
nodes. Herein, 1/500 of the member length imperfection is consid-
ered. The post-buckling behaviors of the non-fuse members are not
considered in the model since buckling failure is not permitted for
non-fuse members in EBFs. However, if post-buckling behaviors
were to be included in models, then eight elements would be
needed for each member [28].
Since capacity design principle requires that non-fuse members
are proportioned with respect to the actual resistance of the links,
the material strength for links should use probable or expected
yield strength in the NRH analysis. Assuming that all the frame
members use CSA G40.21 [29] 350 MPa grade steel, the expected
yielding strength of links is then taken as 385 MPa. Since frame
beams use the same component of links, the strength of the beams
is also taken as 385 MPa. The material strength of columns and
braces is taken as the specied value, 350 MPa, as required by limit
state design philosophy.
Seismic mass is concentrated on column lines as well as the
master end nodes of link elements. Different damping models are
employed to construct the damping matrices for fuse and non-fuse
members as recommended by Ricles and Popov [30] and Koboevic
and Redwood [31]. For non-fuse members, the Rayleigh damping
proportionality constants for the mass matrix and tangent stiffness
matrix are computed based on the frequencies of modes 1 and 3
with damping ratio of 0.05. For fuse links, only mass-related damp-
ing is considered with the proportionality constant computed in
the same way.
Hysteretic energy and relative seismic input energy are com-
puted using the structural responses obtained from OpenSees.
The hysteretic energy of element k at the end of an earthquake
E
h,k
is expressed as the sum of the works of internal forces, i.e.,
E
h;k
W
Moment;k
W
Axial;k
W
Shear;k
k 1; 2; . . . n
e
13
where n
e
is the number of elements; W
Moment;k
; W
Axial;k
; W
Shear;k
are
the works done by the bending moment, axial force and shear force
of element k, respectively; and
W
Moment;k
X
T
t0
1
2
Da
t
Da
tDt
M
t
14
W
Axial;k
X
T
t0
1
2
Dx
t
Dx
tDt
N
t
15
W
Shear;k
X
T
t0
1
2
Dy
t
Dy
tDt
V
t
16
where M
t
, N
t
and V
t
are internal moment, axial force and shear force
at time t, respectively; Da
t
, Dx
t
, and Dy
t
are the increments of end
rotation, axial deformation, and shear deformation of the element
from time (t Dt) to time t, respectively; Dt is the time-increment
for the dynamic analysis; T is the duration of the earthquake. The
total hysteretic energy of the entire seismic force resisting system
is E
h
P
ne
k1
E
h;k
. The total hysteretic energy of fuse members is
E
hf
P
n
f
k1
E
h;k
.
The relative seismic input energy E
i
(t) imparted to the SFRS is
expressed as
E
i
t
X
nm
m1
E
i;m
t 17
where n
m
is the number of the degree-of-freedom. The input energy
of mth degree of freedom at time t is
E
i;m
t E
i;m
t Dt
1
2
Du
t;m
Du
tDt;m
F
i
t;m
18
in which inertial force F
i
t;m
M
m
x
g;t
, and M
m
and x
g;t
are the mass
of the mth degree of freedom and ground acceleration, respectively;
Du
t.m
is the incremental displacement of the mth degree of freedom
from time (t Dt) to time t. The seismic input energy at the end of
an earthquake is E
i
(T). More details about the analysis technique
and energy calculation can be found in [32].
Table 3
Search results of multi-objective genetic algorithm.
Solution # CL (W) LK1 (W) LK2 (W) LK3 (W) BR1 (HSS) BR2 and BR3 (HSS)
1 310 52 410 100 410 46 360 33 203 203 13 178 178 8.0
2 310 52 410 74 410 46 360 33 203 203 13 178 178 8.0
3 310 52 410 74 410 46 360 33 203 203 9.5 178 178 8.0
4 310 52 410 60 410 46 360 33 203 203 13 178 178 8.0
5 310 52 410 46 360 39 360 39 203 203 13 178 178 8.0
6 310 52 410 46 360 39 360 33 203 203 13 178 178 8.0
7 310 52 410 46 360 39 360 33 178 178 8.0 178 178 8.0
8 310 52 410 46 360 39 360 33 178 178 8.0 178 178 6.4
Table 4
Structural responses of Pareto-optimal designs.
Solution
#
O
1
(kN)
O
2
(kN m)
b
(%)
Max. link rotation
a
(%
radians)
Max. story drift
a
(%
radians)
Max. beam plastic rotation (%
radians)
Max. brace plastic rotation (%
radians)
Period
(s)
1 51 743 99 5.8 (0.73) 0.96 (0.38) 0.029 0.19 0.597
2 48 757 99 5.8 (0.72) 0.95 (0.38) 0.029 0.16 0.606
3 47 790 98 5.7 (0.71) 0.95 (0.38) 0.031 0.18 0.623
4 47 805 99 6.5 (0.82) 0.82 (0.33) 0.018 0.12 0.616
5 46 820 99 6.9 (0.86) 0.93 (0.37) 0.041 0.05 0.638
6 45 822 99 6.6 (0.82) 0.90 (0.36) 0.038 0.05 0.641
7 42 834 98 7.0 (0.88) 0.98 (0.39) 0.046 0.19 0.684
8 40 810 97 7.0 (0.87) 1.04 (0.42) 0.056 0.37 0.703
a
The number inside parenthesis is the ratio of demand to capacity.
34 Y. Gong et al. / Engineering Structures 48 (2013) 2836
4.3. Design solutions
For six design variables that each has four candidate sections,
there are 4096 possible design solutions. For GA, each design is rep-
resented by a 12-bit binary string in which each variable corre-
sponds to a 2-bit string with 2
2
= 4 values being available to the
design. The population size is taken as 30 while the mutation rate
is taken as 0.05. During an iterative design process, any new design
solution produced by GAis evaluated using the OpenSees NRHanal-
ysis procedure, and the analysis results are stored for a later recov-
ery incase the same designsolutionis repeatedina later generation.
The value of W
0
, which is used to normalize O
1
, is taken as the
weight of the heaviest design of the current generation. For the nor-
malization of O
2
, E
imax
is taken as the largest seismic input energy of
the current generation. The GA search is considered to converge if
the average change of s
p
is less than 10
6
over 50 generations, or
the algorithm is terminated if the generation number reaches 150.
A test run is executed under the assumption that at each oor,
the link and the beams outside the link are of a single component
and have the same cross section. The test run nds no feasible
solutions. Examination of the analysis results reveals that the plas-
tic rotation limits of the beams outside the links are violated in all
the candidate solutions. Therefore, it was decided to add ange
cover plates to the beams. The second strategy was adopted, i.e.,
two cover plates of 12 mm 100 mm are welded to every beam.
The number of the design variables remains 6.
Because GA is not guaranteed to converge to all the Pareto-
optimal solutions in a single run, it is common practice to conduct
multiple runs. The solutions of the multiple runs are then
combined together to determine a set of Pareto-optimal designs.
Table 3 lists the solutions from two runs in which 1142 designs
were evaluated through using 105 h of computer time on a desktop
with Intel Core i5 750 CPU and 4 GB random access memory.
The structural responses of the eight Pareto-optimal designs are
listed in Table 4. Note that the plastic rotation demands of the
beams outside the links are not zero even they are reinforced by
ange cover plates. The #8 is the design solution with the least
weight (O
1
= 40 kN), and the #1 is the design having the minimum
seismic input energy (O
2
= 743 kN m). All the Pareto-optimal solu-
tions have a near 100% b value indicating that the plastic deforma-
tions almost exclusively occur within links (note that the plastic
rotations allowed for non-fuse members are very small).
The maximum inter-story drift and the maximum link rotation
are also provided in Table 4. It shows that link rotation constraint is
more likely to govern the design solution as the demand-to-capac-
ity ratio of link rotation is much closer to unity.
In general, it can be said that each Pareto-optimal solution is an
equally worthy candidate design for further consideration. Though
design #1 is 27 percent heavier than design #8, it absorbs 9% less
of seismic energy. Design #2 is 6% lighter than design #1, but its
seismic input energy is only 2% greater than that of design #1. If
one considers the difference of b values among all the designs as
negligible, then designs #5, #6 and #7 may be regarded as inferior
to design #8. In summary, although the Pareto solutions cannot
dominate each other, with the help of above discussion, one can
easily choose the nal solution based on his/her preference.
Though it might appear that the design objective O
3
(or b) plays
little role in selecting the nal design among the Pareto-optimal set,
O
3
still plays the vital role in driving GAto obtain the Pareto-optimal
set since a feasible solution with no yielding of the links is possible.
The rst period of the Pareto solutions has a value ranging
approximately from 0.6 to 0.7 s (Table 4) which is considerably
greater than the 0.3 s estimated using the empirical method in
NBCC [1]. The same discrepancy existed for the moment frame
in the previous study [8]. This discrepancy may incur some
restriction on an inelastic-analysis-based design method. For
example, NBCC specify that the theoretical period must not ex-
ceed 1.5 times the empirical period for moment frames for fear
that the uncertainties associated with mathematical modeling of
a structure (e.g., participation of nonstructural components in
seismic response) might lead to the underestimation of earth-
quake demands. Though a simple remedy of scaling up ground
motions in the aforedescribed design process is available, further
research is needed to calibrate between various design methods
to determine this scale factor. Alternatively, methods to model
the contribution of nonstructural components could be included
in future research.
5. Conclusions
This paper presented an optimal capacity design method for
steel building frameworks subjected to earthquake loading. The
design example was dedicated to eccentrically-braced frames,
though the method is also applicable to other steel frameworks
such as moment-resisting frames [8]. Such a design approach is
made possible by using an inelastic analysis tool which, for this
study, is nonlinear response history analysis procedure.
Minimum structural weight, minimum seismic input energy,
and maximum hysteretic energy of fuse members were adopted
as the design objectives. No pre-determined weights were assigned
to the design objectives, which was to allow the designer to choose
the nal design solution based on his/her preference.
The technical merits of the design method include: (1) the de-
sign formulation is generally applicable to various steel frameworks
such that the current a-set-of-rules-for-each-type-of-system ap-
proach is not necessary (e.g., ductility-related force reduction fac-
tors [4] are not needed); (2) the analytical tool is most accurate
which helps to obtain a structural design solution with a better per-
formance under earthquake hazards; (3) the limit state design phi-
losophy is achieved at structural system level; (4) the proposed
computer-aided design tool will help engineering ofces carry out
capacity design principle without arduous hand calculations.
The selection of the design objectives and the design constraints
fully reects the strength of the adopted evaluating tool. The earth-
quake input energy, hysteresis energy of structure and plastic
deformation of members are the structural response information
which can only be obtained through a nonlinear dynamic analysis.
Comparing with displacement or force responses, an energy re-
sponse has the advantage of not only reecting the duration of
ground motions but also accounting for the accumulative effect
of inelastic deformations. While the inter-story drift constraints
govern the overall structural performance, the plastic deformation
constraints ensure the satisfactory local behavior of individual
members. Note that the member strength checking is implicitly
carried out as an integral part of the structural analysis.
For the three-story EBF example, load and resistance factors
were included in the analysis procedures. Eq. (12) was proposed
to compute link rotation by using nodal displacement information.
Eqs. (13)(18) described the method to evaluate seismic energies
using the structural responses obtained from an existing computer
program. In the example, it was necessary for the beams outside
the links to be reinforced by ange cover plates in order to prevent
the beams from undergoing detrimental plastic rotation. It is pos-
sible, though not conducted herein, a more economic design with-
out beam reinforcement might be obtained by trying variant link
lengths and different link sizes.
The example demonstrated that through the help of the third
objective, the design formulation was able to obtain a set of design
solutions that all had the same plastic mechanismunder the design
earthquake loading. Therefore, among these well-designed Pareto
optimal structures, it is plausible to directly relate the seismic in-
put energy with the future risk of structural failure.
Y. Gong et al. / Engineering Structures 48 (2013) 2836 35
The numerical example also illustrated that designcriteria O
1
and
O
2
were generally competing with each other (Table 4) among the
Pareto optimal set. But O
3
or b was not competing with either O
1
or
O
2
, because the yielding of fuse members in a structure is not depen-
dent on its overall structural weight or seismic input energy but
rather the relative strength between fuse and non-fuse members.
The fundamental period of the design solutions appeared to be
much greater than an empirical value. This entails further research
to compare the design results of the proposed design method with
traditional elastic-analysis-based design methods (especially when
an empirical period is used) in order to develop a guideline for the
inelastic-analysis-based design method. Other possible design
objectives concerning maximum story drifts, oor accelerations
and base shear will be investigated in future study. The long com-
puter hours for the design example also demonstrated the necessity
in future endeavor to develop strategies to mitigate computational
burden.
Acknowledgement
The authors wish to thank NSERC for the nancial support of
this work.
References
[1] NRCC. National building code of Canada 2010. Ottawa (Ontario, Canada):
National Research Council of Canada; 2010.
[2] Engelhardt MD, Popov EP. On design of eccentrically braced frames. Earthq
Spectra 1989;5(3):495511.
[3] Tremblay R. Achieving a stable inelastic seismic response for multi-story
concentrically braced steel frames. Eng J 2003;40(2):11129.
[4] CSA. CAN/CSA-S16-09 design of steel structures. Mississauga (Ontario,
Canada): Canadian Standards Association; 2009.
[5] FEMA-273. NEHRP guideline for the seismic rehabilitation of buildings.
Washington (DC, USA): Building Seismic Safety Council, Federal Emergency
Management Agency; 1997.
[6] FEMA-450. NEHRP recommended provisions for seismic regulations for new
buildings and other structures, part 1 provisions. Washington (DC, USA):
Building Seismic Safety Council, Federal Emergency Management Agency;
2004.
[7] Foley CM, Pezeshk S, Alimoradi A. Probabilistic performance-based optimal
design of steel moment-resisting frames, I: formulation. J Struct Eng
2007;133(6):75766.
[8] Gong Y, Xue Y, Xu L, Grierson DE. Energy-based design optimization of steel
building frameworks using nonlinear response history analysis. J Constr Steel
Res 2012;68(1):4350.
[9] Xu L, Gong Y, Grierson DE. Seismic design optimization of steel building
frameworks. J Struct Eng 2006;132(2):27786.
[10] Grierson DE, Gong Y, Xu L. Optimal performance-based seismic design using
modal pushover analysis. J Earthq Eng 2006;10(1):7396.
[11] Gong Y. Optimal stiffness distribution of steel moment frames under extreme
earthquake loading. Adv Struct Eng 2005;8(6):57384.
[12] Leger M, Dussault S. Seismic-energy dissipation in MDOF structures. J Struct
Eng 1992;118(5):125169.
[13] Uang CM, Bertero VV. Evaluation of seismic energy in structures. Earthq Struct
Dynam 1990;19:7790.
[14] ASCE. Seismic rehabilitation of existing buildings, ASCE/SEI 41/06 standard.
Reston (VA): American Society of Civil Engineers; 2007.
[15] Goldberg DE. Genetic algorithms: search, optimization and machine
learning. Reading (MA, USA): Addison Wesley; 1989.
[16] The MathWorks, Inc. MATLAB the language of technical computing, version
7.8; 2009. <http://www.mathworks.com>.
[17] Gong Y, Xu L, Grierson DE. Sensitivity analysis of steel moment frames
accounting for geometric and material nonlinearity. Comput Struct
2006;84(7):46275.
[18] Gong Y, Xu L, Grierson DE. Performance-based design sensitivity analysis of
steel moment frames under earthquake loading. Int J Numer Meth Eng
2005;63:122949.
[19] Deb K. Multi-objective optimization using evolutionary algorithms. West
Sussex (England): John Wiley & Sons, Ltd; 2001.
[20] CISC. Handbook of steel construction. 10th ed. Toronto (Ontario, Canada):
Canadian Institute of Steel Construction; 2010.
[21] PEER. Pacic Earthquake Engineering Research Center; 2008. <http://
www.peer.berkeley.edu>.
[22] Mansour N, Christopoulos C, Tremblay R. Experimental validation of
replaceable shear links for eccentrically-braced steel frames. J Struct Eng
2011;137(10):114152.
[23] OpenSees. Open system for earthquake engineering. Berkeley (California):
Pacic Earthquake Engineering Research Center; 2008. <http://
opensees.berkeley.edu>.
[24] Ramadan T, Ghobarah A. Analytical model for shear-link behavior. J Struct Eng
1995;121(11):157480.
[25] Richards PW, Uang C-M. Testing protocol for short links in eccentrically-
braced frames. J Struct Eng 2006;132(8):118391.
[26] Kasai K, Popov EP. General behavior of WF steel shear link beams. J Struct Eng
1986;112(2):36282.
[27] Filippou FC, Fenves GL. Methods of analysis for earthquake-resistant
structures. In: Bozorgnia Yousef, Bertero Vitelmo V, editors. Earthquake
engineering: from engineering seismology to performance-based
engineering. CRC Press; 2004.
[28] Lamarche CP, Tremblay R. Seismically induced cyclic buckling of steel columns
including residual-stress and strain-rate effects. J Constr Steel Res
2011;67(9):140110.
[29] CSA. CAN/CSA-G40.21-04, Structural quality steel. Toronto (Ontario) Canadian
Standards Association; 2004.
[30] Ricles JM, Popov EP. Inelastic link element for EBF seismic analysis. J Struct Eng
1994;120(2):44163.
[31] Koboevic S, Redwood R. Design and seismic response of shear critical
eccentrically braced frames. Can J Civil Eng 1997;24:76171.
[32] Xue Y. Capacity design optimization of steel building frameworks using
nonlinear time-history analysis. A thesis presented to the University of
Waterloo in fulllment of the thesis requirement for the degree of Doctor of
Philosophy in Civil Engineering. University of Waterloo, Ontario, Canada;
2012.
36 Y. Gong et al. / Engineering Structures 48 (2013) 2836