Journal of Constructional Steel Research 67 (2011) 322–332

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

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

Progressive collapse design of seismic steel frames using structural optimization

Min Liu ∗
Country Creek, 9677 Scotch Haven Drive, Vienna, VA 22181, USA

article info abstract

Article history: This paper uses structural optimization techniques to cost-effectively design seismic steel moment frames
Received 16 March 2010 with enhanced resistance to progressive collapse, which is triggered by the sudden removal of critical
Accepted 27 October 2010 columns. The potential for progressive collapse is assessed using the alternate path method with each
of the three analysis procedures (i.e., linear static, nonlinear static, and nonlinear dynamic), as provided
Keywords: in the United States Department of Defense United Facilities Criteria (UFC) Design of Buildings to Resist
Steel frame
Progressive Collapse. As a numerical example, member sizes of a two-dimensional, nine-story, three-
Progressive collapse
Alternate path method
bay regular steel immediate moment frame are optimally determined such that the total steel weight
Nonlinear analysis is minimized while the design satisfies both AISC seismic provisions and UFC progressive collapse
Optimization requirements. Optimization results for the example frame reveal that the traditional minimum weight
seismic design, which does not explicitly consider progressive collapse, fails to meet the UFC alternate
path criteria associated with any analysis procedure. Progressive collapse design optimization using
the linear static procedure produces the most conservative and consequently heaviest design against
progressive collapse. In contrast, the more accurate nonlinear static and dynamic procedures lead to
more economical designs with UFC-acceptable resistance to progressive collapse, at the expenses of
considerable modeling and computing efforts.
© 2010 Elsevier Ltd. All rights reserved.

1. Introduction capacity of floor/roof systems to improve the building’s load

Progressive collapse refers to a chain reaction of structural
element failures following the initial damage to a localized portion
of a building, eventually causing widespread structural damages
that are disproportionate to the triggering event. The risk of
progressive collapse shall be adequately mitigated during the
building design process in order to safeguard the structure against
this catastrophic event. In traditional structural codes, building
design against progressive collapse has often been indirectly
addressed by prescribing a certain level of structural integrity [1].
Recently, design guidelines have been developed to explicitly
take progressive collapse into account for building design. For
example, the UFC 4-023-03 (hereafter referred to as UFC) is
one of the frequently referenced documents for structural design
against progressive collapse [2]. Instead of preventing or limiting
initial localized damages, these design guidelines try to provide
satisfactory continuity, ductility, and redundancy in buildings such
that the spread of local damage can be confined. Specifically, UFC
allows both indirect and direct methods to be used to achieve
this goal. The tie force method is an indirect design approach,
with which the structure is tied together by employing the tensile
∗ Tel.: +1 703 541 7416; fax: +1 703 541 7418.
E-mail address: maxminliu@gmail.com.
0143-974X/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2010.10.009
redistribution capability. There are two direct design methods in
UFC: the enhanced local resistance method and the alternate path
method. The enhanced local resistance method aims to harden
critical structural members (e.g., perimeter columns, wall sections)
to offer satisfactory strength and ductility to resist progressive
collapse. In comparison, using the alternate path method, a
designer must ensure that the building is able to bridge over
selected load-bearing elements that are notionally removed, one
at a time, from the original intact building to simulate their sudden
loss when subjected to damaging extreme loads.
Of the three different UFC design methods, the alternate path
method provides a systematic way of evaluating the potential
of buildings for progressive collapse. For each predetermined
scenario of removing key structural elements, structural analysis
is carried out for the damaged structure (i.e., the structure with
an element missing). Instead of simulating the possible chain
reaction of structural failures following the notional element
removal, structural deformations and internal forces resulting
from structural analysis are examined against the UFC acceptance
criteria associated with the specific type of structural analysis
being carried out. The building design is considered acceptable
if these acceptance criteria are met. The UFC alternate path
method can be applied by using one of the three analysis
options: linear static, nonlinear static, and nonlinear dynamic. As
the simplest analysis option, the linear static procedure offers
preliminary, often conservative assessment of the progressive
 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
collapse potential. The nonlinear dynamic procedure is the
most accurate and also most computationally expensive option,
as it necessitates sophisticated finite element modeling and
considerable computing efforts to carry out the time history
analysis. As an intermediate analysis option, the nonlinear static
procedure also involves modeling of both material and geometrical
nonlinearities while it does not perform time history analysis
to simulate the load redistribution behavior. To approximately
compensate for the dynamic effects corresponding to the actual
load redistribution, a dynamic increase factor is used in the
nonlinear static procedure to increase the gravity loads acting on
the areas that are immediately affected by the removed structural
elements.
Seismic design provisions have undergone significant revisions
after the recent major earthquakes that exposed deficiencies
in previous generations of engineering practice. As a result,
seismically safer buildings can generally be designed per the
current seismic provisions. However, unless it is intentionally
proportioned for enhanced resistance to progressive collapse, a
seismically designed building does not necessarily have sufficient
capacity to redistribute loads to other regions of the building
upon the sudden loss of certain critical load-bearing elements. This
is because buildings behave very differently when withstanding
an earthquake and when resisting progressive collapse. While
seismic design primarily focuses on lateral loads, progressive
collapse design is more concerned with gravity loads acting on the
structure [2]. The inherent resistance of seismic building structures
to progressive collapse has been investigated using the alternate
path method [3]. In order for seismic-resistant buildings to have
adequate load redistribution capability, the progressive collapse
requirements need to be explicitly considered during the structural
design process.
One important issue in progressive collapse design is how to
achieve cost-effectiveness in proportioning structural components
to facilitate load redistribution. Use of the alternate path method
requires structural analysis and acceptance evaluation of trial
designs under different scenarios of removing critical elements,
making the design process very repetitive in nature. The efforts
to obtain an economical design can be formidable if a manual
trial-and-error design approach is employed. However, this
process can be efficiently handled by computerized structural
optimization, which uses appropriate numerical search algorithms
to maximize or minimize predefined objective function(s) by
selecting appropriate values for a set of design variables while
conforming to relevant design constraints. For design of practical
building structures, design constraints are usually derived from
structural criteria set forth in relevant code standards. The
commonly used design variables are often discrete-valued, such as
cross-sectional dimensions in concrete design and commercially
available standard sections in steel design. The material cost is
usually used as an objective function subjected to minimization.
For steel design, the material cost is often simply expressed
as the total steel weight, recognizing that this weight metric
alone may not completely quantify the actual expenses associated
with construction of a steel building [4]. Because objective
functions and design constraints are typically non-differentiable
with respect to discrete-valued design variables, most gradient-
based optimization algorithms are not readily applicable. In
contrast, heuristic search algorithms are particularly effective
for solving practical steel design optimization problems. These
algorithms include simulated annealing [5], genetic algorithm
(GA) [6], tabu search [7], particle swarm optimization [8], and ant
colony optimization [9].
The ever-increasing concerns over the susceptibility of building
structures to progressive collapse have spurred extensive research
to accurately simulate such a chain reaction and understand
323
the underlying mechanisms (e.g., [10–13]). Design optimization
against progressive collapse provides a promising approach to
achieving economy in building design while mitigating such rare
yet devastating events. Grierson and Khajehpour carried out
earlier, relevant research [14]. They considered the progressive
collapse risk by defining a single load-path redundancy factor,
which is a function of bay numbers and degree of connectivity
between the floor system and columns/shear walls in each story.
They emphasized the importance of enhancing structural safety
against progressive collapse in achieving an overall cost-effective
design. However, integration of structural optimization techniques
with the progressive collapse design by direct approaches (e.g.,
alternate path method) has not been available in the literature.
In this paper, a GA-based structural optimization is presented
to cost-effectively design code-compliant seismic-resistant steel
frame structures that simultaneously satisfy the UFC progressive
collapse criteria associated with the alternate path method. A
number of column removal scenarios are considered per the UFC
guidelines. The load redistribution capability is assessed by using
each of the three analysis procedures (i.e., linear static, nonlinear
static, and nonlinear dynamic) provided in UFC. To illustrate the
usefulness of the present progressive collapse design optimization,
a numerical example is provided, in which member sizing of a
planar nine-story, three-bay seismic steel moment frame is carried
out in order to minimize the total steel weight while possessing
UFC-acceptable resistance to progressive collapse. Designs thus
obtained using different UFC analysis procedures are critically
compared. In order to demonstrate the importance of intentionally
considering progressive collapse in the structural design process,
the traditional minimum weight design of the same example
frame without explicitly accounting for progressive collapse is also
obtained as a baseline design.
2. Problem statement
2.1. Overview
The present progressive collapse-resistant design optimization
of seismic steel frame structures can be conceptually stated as
Objective: To reduce the total steel weight, which acts as an
approximate indicator of frame construction cost.
Subject to: AISC-LRFD seismic provisions [15];
UFC progressive collapse design requirements associ-
ated with the alternate path method.
The design variables considered in the optimization are steel
member sizes that are selected from commercially available hot-
rolled, wide-flange standard steel sections. A standard GA is used
as a numerical solver to find the optimized combination of section
types for beams and columns of the frame. Fig. 1 gives a flowchart
for this structural optimization framework. GA strategies and
AISC-LRFD seismic requirements are briefly described in the next
two sub-sections. The UFC alternate path method is discussed in
Section 3.
2.2. Genetic algorithm
A GA scheme that was previously developed for optimal design
of steel frames [4] is used for the present member-sizing problem.
Specifically, a steel frame design is encoded into a two-portion
string of pointers, which are associated with two subsets of
commercially available standard sections used for column and
beam members, respectively. GA starts with a set of initial trial
frame designs that are generated from an exhaustive combination
of different section types, one from the column section subset and
the other from the beam section subset. Each of the following
324 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
Fig. 1. A flowchart for the progressive collapse design optimization of seismic steel
frames.
offspring generations contains 200 design candidates that are
reproduced from their parent generation using genetic operations
(i.e., crossover and mutation). A two-point crossover operation,
one in each portion of the string representation, is adopted with
a probability of 50%. The mutation operation is performed with a
probability of 10% through perturbing a randomly selected pointer
anywhere in the string representation. The present member-sizing
problem with multiple design constraints is converted into an
unconstrained optimization problem by penalizing the fitness
values in GA, depending on the degree of constraint violation [4].
During the optimization process, the steel weight of the best frame
design resulting from each generation is continuously monitored.
The GA computation is performed until the pre-determined
stopping criterion (e.g., the steel weight of the current optimized
frame is reduced sufficiently compared with that of the initial
design) is considered met.
2.3. AISC-LRFD seismic design requirements
The following load combinations are used for code-compliance
seismic checking [15]:
1.4D

1.2D + 1.6L + 0.5 (Lr or S )


(1)
(1.2 + 0.2SDS ) D + ρ QE + 0.5L + 0.2S

(0.9 − 0.2SDS ) D + ρ QE
where D, L, Lr , and S are dead load, floor live load, roof live
load, and snow load, respectively; SDS is the five-percent damped
design spectral response acceleration parameter at short periods;
ρ is a redundancy factor; and QE is the effect of horizontal
seismic forces. If use of these load combinations reveals that the
axial strength demand-to-capacity ratio of any column is greater
than 0.4, additional two load combinations that include amplified
seismic loads are used to check column strength [16]:
(1.2 + 0.2SDS ) D + Ω 0 QE + 0.5L + 0.2S

(2)
(0.9 − 0.2SDS ) D + Ω 0 QE
where Ω 0 is an overstrength factor.
A steel frame design is considered code-compliant if the
required strengths of all structural components are no greater
than their respective capacities. Also, the deformation and overall
stability of the frame are controlled by satisfying code-specified
inter-story drift requirements. The relevant details are not
included herein and can be found elsewhere [15,16].
3. UFC alternate path method
The UFC alternate path method examines the ability of a
structure to bridge over the removed critical structural elements.
According to UFC, structural elements in the first story above
grade, in the story directly below the roof, in the story at the
mid-height of the building, and in the story above the location
of column splices are all possibly subjected to removal. As a
minimum, corner and middle elements in a selected story shall
be considered for removal. For each removal scenario, structural
analysis is carried out for the damaged structure to determine the
acceptability of all structural components that contribute to load
redistribution capacity. Instead of dealing with the entire failure
progression process, UFC defines acceptance criteria for different
types of structural components. If the resulting deformations and
internal actions satisfy the acceptance criteria for all structural
components, the overall structure is considered capable of
resisting progressive collapse initiated by the particular element
removal scenario. UFC distinguishes deformation-controlled and
force-controlled actions and have different acceptance criteria
for each type of action. For moment frames, flexure and axial
tension are considered deformation-controlled actions, while axial
compression and shear are considered force-controlled actions.
A structural component can be either deformation-controlled
or force-controlled. In particular, a beam–column member is
considered force-controlled if the calculated axial compression
force is greater than half of its nominal lower-bound axial
strength. In the following sub-sections, the three different analysis
procedures used for the UFC alternate path method are briefly
discussed. The differences among these procedures in load
combinations, modeling and analysis, and acceptance criteria are
emphasized.
3.1. The linear static procedure
The linear static procedure is the simplest option to carry out
the UFC alternate path analysis. Two types of load combinations
are defined in this procedure. In general, a deformation-controlled
load combination is used to calculate the deformation-controlled
actions, while a force-controlled load combination is required to
obtain the force-controlled demands. For floor and/or roof areas
that are immediately affected by a removed element, the following
increased gravity loads are applied:
GLS_above = ΩLS [(0.9 or 1.2)D + (0.5L or 0.2S )]
(3)
 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
where ΩLS is the load increase factor that approximately accounts
for dynamic and nonlinear effects following the sudden element
removal. For steel frames,
2,

for force-controlled
 angle of a structural component, respectively. The yield rotation
load combination

ΩLS = (4)
0.9mmin + 1.1,
 for deformation-controlled
load combination.
Here, mmin is the smallest value of m-factors of all primary struc-
tural components, excluding columns, that contribute to progres-
sive collapse resistance and that are within the immediately af-
fected portion of the structure. The m-factors are adopted from
ASCE 41 [17] and are indirect measures of the nonlinear deforma-
tion capacity of structural elements. For steel members, m-factors
are determined as a function of member compactness for beams
and a function of both member compactness and axial compression
force level for columns. In addition, there are m-factors for different
types of beam-to-column connections. For example, for the fully
restrained ‘‘WUF (welded unreinforced flange)’’ moment connec-
tions used in steel structures, the m-factor is equal to 4.3 − 0.083d,
where d is the depth (in inches, with 1 in = 0.0254 m) of the con-
nected beam member. The m-factor of a steel connection shall be
modified to account for the effects of continuity plate detailing,
panel zone strength, beam span-to-depth ratio, and beam web and
flange slenderness. For the floor areas outside the region that is
immediately affected by the removal of structural elements, the
following gravity loads are applied:
GLS_outside = (0.9 or 1.2)D + (0.5L or 0.2S ). (5)
In addition to gravity loads, the following notional lateral load is
also applied at each floor or roof level:
Llateral = 0.002Σ P (6)
where Σ P is the sum of gravity loads acting at each level.
Linear static models are used in structural analysis to calculate
both deformation-controlled and force-controlled demands of
structural components. For a deformation-controlled action, the
capacity of any structural component is calculated by multiplying
the factored expected strength by the corresponding m-factor.
However, for a force-controlled action, the factored lower-bound
strength is used as the capacity. It is considered that a structure has
adequate resistance to progressive collapse if, for each contributing
structural component, the following condition is satisfied:
Demand
≤ 1. (7)
Capacity
3.2. The nonlinear static procedure
Compared with the simple elastic analysis, an inelastic analysis
is able to more accurately predict structural performance and
thereby leads to more reliable structural design. However,
the increase in accuracy is achieved at the expense of more
sophisticated modeling and extra computational efforts. In the UFC
nonlinear static procedure, the gravity loads applied to floor areas
not immediately affected by removed structural elements and the
lateral loads applied at each floor/roof level are the same as those
defined for the UFC linear static procedure. For the immediately
affected bays, the following increased gravity loads are applied:
GNS_above = ΩNS [(0.9 or 1.2)D + (0.5L or 0.2S )] (8)
where ΩNS is a dynamic increase factor. For steel frame structures,
this factor is calculated as
ΩNS = 1.08 + 0.76/ min θacc /θy + 0.83
   
(9)
325
where the minimum operator is applied to all structural compo-
nents, excluding columns, that contribute to progressive collapse
resistance and that are within the immediately affected areas; θy
and θacc are the yield rotation angle and acceptable plastic rotation
angle of a beam is calculated in radians as
Zb Fye lb /6EIb ,

for a framed beam
θy = (10)
Zb Fye lb /4EIb , for a cantilevered beam
where Zb , Ib , and lb are the plastic section modulus, moment of
inertia, and length, respectively, of the beam; E and Fye are the
elasticity modulus and expected yield strength, respectively. For
a framed column, the yield rotation angle is calculated in radians
as
θy = 1 − PU /Pye Zc Fye lc /6EIc
 
(11)
where Zc , Ic , and lc are the plastic section modulus, moment of
inertia, and length, respectively, of the column; PU and Pye =
Ac Fye are the demand and expected-yield axial strength of the
column with Ac the cross-sectional area. θacc is defined as a
multiple of θy [17]. Similar to the m-factors used in the linear
static procedure, θacc is also a function of section compactness
for beams and a function of both section compactness and axial
force level for columns. For deformation-controlled actions, the
acceptance criterion is met if the calculated deformation demand
is no greater than the expected deformation capacity. For example,
the acceptance criterion for the flexure of steel members is
θcalc ≤ θacc (12)
where θcalc is the calculated plastic rotation demand. In contrast,
the acceptance criteria for force-controlled actions and force-
controlled columns are the same as those in the UFC linear static
procedure, except that force-controlled demands are calculated
by using a nonlinear analytical model and the nonlinear static
analysis-related load combination. In addition, UFC defines θacc for
different types of beam-to-column connections. For example, for
the fully restrained WUF moment connections in steel structures,
θacc is equal to 0.0284 − 0.0004d, where d is the depth (in inches,
with 1 in = 0.0254 m) of the steel member being connected.
3.3. The nonlinear dynamic procedure
Because the nonlinear dynamic procedure explicitly accounts
for both dynamic and nonlinear effects, amplification factors are
no longer needed in the load combination. Instead, whether or not
they are immediately affected by the element removal scenarios,
all floor and roof areas are subjected to the following gravity loads:
GND = (0.9 or 1.2)D + (0.5L or 0.2S ). (13)
Meanwhile, the same notional lateral load as used in the linear
and nonlinear static procedures is applied at each floor or roof
level. For each removal scenario, after the intact structure is
statically loaded with the gravity loads and notional lateral loads,
the selected structural element is instantly removed. The time
history analysis is then carried out for a time duration during which
the maximum displacement can be reached or one cycle of vertical
motion can occur at the removal location. The acceptance criteria,
yield rotation angles, and acceptable plastic rotation angles for
the nonlinear dynamic procedure are all the same as those for
the nonlinear static procedure, except that both deformation-
controlled and force-controlled demands are calculated from
nonlinear dynamic analyses using a single load combination.
326 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
Fig. 2. An example frame: (a) configuration and member grouping patterns; (b) deformed shapes under different column removal scenarios.
4. A numerical example
4.1. Example seismic frame
As shown in Fig. 2(a), a two-dimensional, nine-story, three-bay
regular steel intermediate moment frame (IMF) with a fixed base
is used to demonstrate the usefulness of the present optimization-
based design of seismic frames against progressive collapse. This
frame is considered one of the two lateral force-resisting moment
frames located along the building perimeter and resists half of the
total seismic loads exerted on the building along the transverse
direction. This building has five bays in the orthogonal direction
with a total length of 150 ft (45.7 m). The height of the first
story is 18 ft (5.5 m) and the height of each of other stories is
13 ft (4.0 m). The member grouping patterns are also shown in
Fig. 2(a). Specifically, between any two vertically adjacent column-
splice locations, columns within all stories take a single member
size and beams at all floor/roof levels take another single member
size. Therefore, a total of eight member size design variables
(i.e., four for columns and four for beams) are considered in the
optimization. The nominal depth of all columns is limited to 18 in
(0.46 m). However, there is no restriction on beam depths. All fully
restrained moment connections in this frame belong to the same
WUF type. The steel material used for all beams and columns is
ASTM A992, with a lower-bound yield and tensile strength values
equal to 50 ksi (345 MPa) and 65 ksi (450 MPa), respectively. An
overstrength factor of 1.10 is used to convert the steel strength
from the lower-bound to expected value.
This example IMF is generically designed for the Seismic Design
Category C with SDS = 0.232, representing a moderate seismic risk
level. The redundancy factor ρ is equal to 1, and values of other
relevant seismic parameters are adopted from ASCE 7 [16]. The
total design base shear is 146 kips (650 kN), which is distributed
over the frame height to each floor/roof level. It is assumed that
seismic loads control over wind loads. The design dead loads
consist of slab self-weight and super-imposed miscellaneous dead
load, resulting a total of 80 psf (3.83 kPa) for each floor and 60
psf (2.87 kPa) for the roof, plus a 20 psf (0.96 kPa) cladding load
and frame self-weight. Subjected to appropriate load reduction,
the design live loads are 50 psf (2.39 kPa) plus 20 psf (0.96 kPa)
partition load for each floor and 20 psf (0.96 kPa) for the roof.
The snow load is 30 psf (1.44 kPa) after considering all applicable
factors. Fig. 2(b) shows a total of eight different column removal
scenarios that are used in the alternate path analysis of the
example frame for the progressive collapse design optimization.
These removed columns are corner and interior members in
the first, fifth, seventh, and ninth stories, respectively. Note that
symmetry in the frame layout has been considered in selecting
these column removal scenarios.
4.2. Structural modeling
For the alternate path analysis using the linear static pro-
cedure, an in-house linear elastic program is used. For the
alternate path analysis using nonlinear static and nonlinear
dynamic procedures, the DRAIN-2DX program [18] is used
to create a planar analytical structural model that accounts
for both material and geometrical nonlinearities. The type-2
element is used to model columns and beams. It is assumed that
plastic hinges yield only in bending and can occur only at element
ends. One type-2 element per column is used, while each beam is
discretized into multiple type-2 elements in order to capture possi-
ble formation of plastic hinges along the beam. A strain-hardening
ratio of 3% is used to simulate the post-yield behavior. Note that
this type-2 element can only represent a bilinear moment–rotation
relationship, while the UFC backbone moment–rotation curve is
multi-linear and includes sudden strength degradation. However,
 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
Fig. 3. Moment–rotation curves of a steel member.
the present DRAIN-2DX model is sufficient for evaluating the valid-
ity of a trial frame design in satisfying the UFC progressive collapse
requirements. This is because, as illustrated in Fig. 3, the acceptable
plastic rotation angle, θacc , of a steel member, as defined in ASCE
41, is always within the first post-yield linear branch (i.e., prior
to strength degradation) of the moment–rotation curve. Beam-to-
column connections and panel zones are not modeled. For WUF
connections, the m-factors used in the linear static procedure and
the acceptable plastic rotation angles used in the nonlinear static
and nonlinear dynamic procedures are modified to include the
effects of beam span-to-depth ratios and beam web/flange slender-
ness. It should be noted that use of the present analytical model is
based on several assumptions, including: (1) the foundation can ac-
commodate the redistributed loads following any column removal;
and (2) use of a planar analytical model can acceptably predict the
actual load redistribution behavior of the original spatial building
under each column removal scenario.
For the nonlinear static procedure, under each column removal
scenario, gravity and lateral loads are applied to the damaged
frame using a load history that starts at zero and then incremen-
tally increases to the final values with a total of ten steps. For the
nonlinear dynamic procedure, under each column removal sce-
nario, the first step is to statically apply gravity and lateral loads to
the intact frame, from which end forces of the to-be-removed tar-
get column are recorded. Then, gravity loads, lateral loads, and the
recorded end forces (in reversed directions) are statically applied
to the damaged frame. Once the damaged frame reaches a static
equilibrium, the recorded end forces (in original directions) are
applied instantaneously to the damaged frame to simulate the sud-
den removal of the target column. The time history analysis is car-
ried out using a critical damping ratio of 5%.
4.3. Results and discussion
Optimal frame designs resulting from the present progressive
collapse design optimization are presented and discussed in this
section. To clearly identify the changes made in the member
sizes of optimal designs due to consideration of progressive
collapse requirements, the traditional optimal seismic frame
design without accounting for progressive collapse is also obtained
as a baseline structure. The designations of structural members of
the example frame in Fig. 2 are defined as follows. Columns are
designated by both column line and story number. For example,
the column in the first story along column line A is denoted as A-1.
Similarly, beams are designated by both bay location and floor/roof
level. For example, the middle beam at the 4th floor level is denoted
as BC-4, while the leftmost beam at the roof level is denoted
as AB-R.
The four optimal designs are: MWD, which is the traditional
optimal seismic design that minimizes the steel weight but does
not consider the UFC progressive collapse requirements; and
PCND, PCNS, and PCLS, which are the minimum weight seismic
designs considering progressive collapse by the nonlinear dy-
namic, nonlinear static, and linear static procedures, respectively.
327
Fig. 4. Generation-wise evolution of different design optimization processes.
Fig. 4 illustrates the evolutions of the GA optimization for obtain-
ing these optimal designs over the first 30 generations. Clearly
all these curves flatten out within early generations. It is pos-
sible that lighter designs could be obtained by running the op-
timization process for additional generations. However, it turns
out that such a weight reduction is not significant for the present
study: the weights of MWD, PCND, PCNS, and PCLS at the 60th
generation are only 2.9%, 5.4%, 1.1%, and 0% less than those
weights at the 30th generation, respectively. More importantly,
the general trends observed in the following comparison among
different 30th-generation optimized designs with or without con-
sidering progressive collapse are still valid for the 60th-generation
optimized designs. Note that it is not the goal of this study to obtain
globally optimal designs. Instead, this study aims to show the feasi-
bility of obtaining cost-effective designs using optimization, espe-
cially within only a few GA generations. Therefore, the optimized
designs obtained at the 30th generation are used in the following
discussion.
Table 1 gives the steel weight and member size information for
these optimized designs obtained at the 30th generation. Clearly
MWD is the lightest frame since it only needs to satisfy the seis-
mic provisions but not progressive collapse requirements when
its structural members are proportioned. Note that, for code-
compliant seismic designs that do not result from structural opti-
mization, more inherent resistance to progressive collapse may be
available due to the presence of extra capacity beyond what meets
the minimum seismic strength/stiffness requirements. Naturally,
seismic designs with progressive collapse considerations require
stronger/stiffer structural members, generally leading to heavier
designs. Percentage-wise, PCND, PCNS, and PCLS are 8.2%, 13.2%,
and 38.8%, respectively, heavier than MWD. The linear static pro-
cedure is the most conservative option, as it requires the overall
heaviest structural members to resist progressive collapse. In con-
trast, use of the nonlinear dynamic procedure produces the lightest
progressive collapse-resistant designs.
The height-wise seismic interstory drift ratios of these four
optimized designs are compared in the left plot of Fig. 5. Note
that the maximum allowable drift ratio is 2% for the present
seismic design. The AISC-LRFD strength demand-to-capacity ratios
for column and beam groups are shown in the middle and right
plots of Fig. 5, respectively. It is well known that lateral seismic
drifts often control the member proportioning in seismic design,
thereby leading to reserve strength that favorably contributes
to progressive collapse resistance. This is because only notional
lateral loads are used in the UFC alternate path analysis. Fig. 5
indicates that, beyond the lateral drift-induced reserve strength,
more strength is needed by MWD to meet the UFC progressive
collapse requirements. This is particularly true for columns at
bottom portion of the MWD frame where axial compression
328 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332

Table 1
Optimized seismic frame designs with or without progressive collapse considerations.
Optimized designs through weight minimization Steel Member Member groups
weight types
1–3 stories 4–5 stories 6–7 8–9 stories
stories

Without considering progressive collapse 99.1 kips Columns W18 × 119 W18 × 97 W18 × 97 W18 × 50
(MWD) (440.6 kN) Beams W27 × 84 W24 × 68 W24 × 55 W21 × 44
Nonlinear dynamic 107.2 kips Columns W18 × 158 W18 × 106 W18 × 76 W18 × 50
(PCND) (476.8 kN) Beams W24 × 76 W24 × 76 W24 × 55 W24 × 62
Considering progressive collapse by different Nonlinear static 112.2 kips Columns W18 × 158 W18 × 119 W18 × 76 W18 × 76
UFC analysis procedures (PCNS) (498.9 kN) Beams W24 × 76 W24 × 68 W24 × 62 W24 × 68
137.5 kips Columns W18 × 192 W18 × 130 W18 × 97 W18 × 76
Linear static (PCLS)
(611.6 kN) Beams W24 × 76 W27 × 94 W27 × 94 W30 × 99

Fig. 5. Seismic interstory drift ratios and seismic strength ratios of different optimized designs.

forces in columns are large enough to make these columns force-
controlled as a result of axial compression and flexure interaction.
Moreover, beams in the top portion of the MWD frame are clearly
inadequate to meet the UFC progressive collapse requirements and
therefore also need to be significantly toughened.
The acceptability of the four optimized designs is checked
using the progressive collapse criteria associated with different
UFC analysis procedures and is reported in Table 2. As expected,
MWD, which does not explicitly consider the UFC progressive
collapse requirements in the design process, fails to meet the UFC
criteria associated with any of the three analysis procedures. In
contrast, PCLS, which is the most conservative design, meets the
UFC criteria associated with all analysis procedures. In addition to
naturally meeting the UFC criteria associated with the nonlinear
static procedure, PCNS also satisfies the UFC criteria associated
with the nonlinear dynamic procedure. However, PCNS does not
pass the UFC criteria checking associated with the linear static
procedure. PCND only meets the UFC criteria associated with the
nonlinear dynamic procedure. Note that these observations are
solely based on designs of the particular example frame used in
this paper.
Fig. 6 gives the controlling UFC demand-to-capacity ratios
and the corresponding column removal scenarios (shown in
parentheses) for all structural members of the four designs,
which are checked by the acceptance criteria associated with the
UFC linear static procedure. In this figure, a ratio greater than
unity indicates that the corresponding member does not meet
the progressive collapse criteria. Note again that, due to frame
layout symmetry, critical columns subjected to removal are only
selected within the left portion of the frame. As a result, the
actual controlling UFC demand-to-capacity ratio for any structural
member should be considered equal to the larger value of the ratio
for that member itself and the ratio for its mirrored member with
respect to the centerline of the frame. It can be seen that all beams
and most columns of MWD fail to meet the progressive collapse
criteria associated with the linear static procedure. The largest UFC
demand-to-capacity ratio is equal to 2.380 (beam AB-R). Moreover,
failures of columns in the lower stories of MWD are mostly caused
by axial compression–flexure interaction and thereby are force-
controlled (denoted by a superscript ‘‘F ’’), while columns in the
upper stories fail in flexure only and thereby are deformation-
controlled. Although PCND and PCNS have both taken progressive
collapse into account, their capacities are still not sufficient to
meet the UFC criteria associated with the linear static procedure,
with the largest UFC demand-to-capacity ratios equal to 1.789
(beam AB-7) and 1.650 (beam AB-6), respectively. Again, structural
 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332 329

Fig. 6. Controlling UFC demand-to-capacity ratios calculated by the linear static procedure.

Table 2
Progressive collapse acceptance matrix.
Optimized Acceptability using UFC alternate path criteria associated with different analysis procedures
designs
Nonlinear dynamic Nonlinear static Linear static

MWD Fails in column removal scenarios 1–7 Fails in all column removal scenarios Fails in all column removal scenarios
PCND Passes, with the largest demand-to-capacity Fails in column removal scenarios 3, 5, and 7 Fails in all column removal scenarios
ratio equal to 0.931
PCNS Passes, with the largest demand-to-capacity Passes, with the largest demand-to-capacity Fails in column removal scenarios 1–7
ratio equal to 0.952 ratio equal to 0.973
PCLS Passes, with the largest demand-to-capacity Passes, with the largest demand-to-capacity Passes, with the largest demand-to-capacity
ratio equal to 0.640 ratio equal to 0.137 ratio equal to 0.999

inadequacies are widespread within the frame: all beams and
several columns fail to meet the UFC criteria associated with the
linear static procedure. As expected, PCLS meets the UFC criteria
associated with the linear static procedure: the largest demand-to-
capacity ratios are 0.999 (beam AB-2) for beams and 0.983 (column
B-1) for columns.
PCLS is analyzed using both UFC nonlinear static and nonlinear
dynamic procedures; the locations of plastic hinges formed in
PCLS are shown in Fig. 7. It is found that plastic hinges are
only formed under the first and second removal scenarios. PCLS
responds elastically under other column removal scenarios. Except
for a plastic hinge formed in column A-9 under the second
removal scenario, all plastic hinges are formed in beams of the
lower/middle-story bays that are immediately affected by each
removed column. It is also observed that the nonlinear dynamic
analysis identifies more plastic hinges than the nonlinear static
analysis. However, the demand-to-capacity ratios corresponding
to these additional plastic hinges are smaller than the ratios
330 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
Fig. 7. Plastic hinges formed in PCLS under different column removal scenarios.
corresponding to the plastic hinges identified by the nonlinear
static analysis alone. The nonlinear dynamic analysis shows that
the controlling structural member is column B-4, which is force-
controlled by the axial compression and flexure interaction under
the first removal scenario. The axial force in this column predicted
by the nonlinear dynamic analysis is 956 kips (4252 kN), which
very slightly exceeds half of the nominal axial capacity of 1910
kips (8496 kN). With an associated bending moment of 54.9 kip-ft
(74.5 kN-m), the resulting interaction ratio is equal to 0.640.
In comparison, the axial compression force in column B-4 as
predicted by the nonlinear static analysis is 954 kips (4244 kN),
which is just less than half of the nominal axial capacity; with no
plastic hinge taking place, this column is force-controlled by shear
and the corresponding UFC strength-to-capacity ratio is merely
equal to 0.024.
The locations of plastic hinges formed in PCNS under each
column removal scenario are illustrated in Fig. 8. This frame is
separately analyzed by UFC nonlinear static procedure and by
UFC nonlinear dynamic procedure. It is seen that uses of these
two procedures identify the same locations of plastic hinges
under all scenarios except under the first and third scenarios,
where additional plastic hinges are identified by the nonlinear
dynamic analysis. However, rotations of these additional plastic
hinges are all less than the UFC acceptable plastic rotations and
therefore do not turn PCNS into an invalid design. The largest
Fig. 8. Plastic hinges formed in PCNS under different column removal scenarios.
UFC strength demand-to-capacity ratio is equal to 0.973 calculated
by the nonlinear static procedure and 0.952 by the nonlinear
dynamic procedure. These largest ratios are both associated with
the force-controlled column B-1 due to axial compression and
flexure interaction under the first removal scenario.
Fig. 9 shows the locations of plastic hinges formed in PCND
under each column removal scenario. Compared with Fig. 8, it
is seen that, in general, more plastic hinges are present under
different removal scenarios. This observation is expected since
PCND is lighter and thereby weaker than PCNS in resisting
progressive collapse. As mentioned earlier, PCND does not meet
the progressive collapse criteria associated with either linear static
or nonlinear static procedure. It is found from the nonlinear
dynamic analysis that the largest UFC strength demand-to-
capacity ratio is 0.931, which is also associated with the force-
controlled column B-1 as a result of axial compression and flexure
interaction under the first removal scenario.
Finally, PCNS is used to demonstrate the significant force
redistribution that takes place in the frame upon column removal.
Take columns A-1 and B-1 as two illustrations. The initial axial
compression forces in these columns of the intact PCNS frame are
311 kips (1381 kN) and 586 kips (2607 kN), respectively. Using the
nonlinear dynamic analysis, under the second removal scenario
where column B-1 is suddenly removed, the axial compression
force in column A-1 of the damaged PCNS frame reaches a peak
value of 830 kips (3692 kN) and eventually damps out to rest at
600 kips (2669 kN). As a comparison, use of the nonlinear static
analysis predicts that, under the second removal scenario, the axial
compression force in column A-1 is 846 kips (3763 kN), a value
1.9% more than the peak force predicted by the nonlinear dynamic
procedure. For column B-1, under the first removal scenario
where column A-1 is suddenly deleted, its axial compression force
predicted by the nonlinear dynamic analysis attains a peak value
of 1290 kips (5738 kN) and finally settles down at 1058 kips
(4706 kN). The nonlinear static analysis predicts that the axial
compression force in column B-1 is equal to 1377 kips (6125 kN),
which is 6.7% more than the maximum force obtained by the
nonlinear dynamic procedure.
 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
Fig. 9. Plastic hinges formed in PCND under different column removal scenarios.
5. Conclusions
Using the alternate path method provided in the 2009 edition of
the United States Department of Defense United Facilities Criteria
(UFC 4-023-03), weight minimization for the gravity-induced
progressive collapse resistant design of a planar nine-story, three-
bay regular seismic steel intermediate moment frame (IMF) leads
to the following conclusions:
1. The traditional optimized seismic IMF (‘‘MWD’’), which is
obtained in this study as a baseline design by weight
minimization without explicitly accounting for progressive
collapse, does not meet the UFC alternate path criteria
associated with any of the three analysis procedures. In addition
to the inherent reserve strength resulting from satisfying the
seismic strength and lateral drift requirements, more member
strength is needed at different portions of the frame in order
to enable load redistribution following the sudden removal of
critical columns.
2. The present progressive collapse design optimization success-
fully produces a seismic IMF design that satisfactorily meets
the UFC alternate path criteria by enhancing load redistribution
capability through appropriate member sizing. Meanwhile, the
cost of constructing this frame is considerably reduced through
minimization of the overall steel weight.
3. Use of the UFC linear static procedure leads to a seismic
IMF design (‘‘PCLS’’) that is the most conservative since this
design also satisfies the progressive collapse criteria associated
with both nonlinear static and nonlinear dynamic procedures.
Consequently, this design is also the heaviest. For the example
frame used in this study, the weight of PCLS is 38.8%, 28.3%, and
22.6% heavier than MWD, the optimized designs considering
progressive collapse by the UFC nonlinear dynamic procedure
(‘‘PCND’’) and by the UFC nonlinear static procedure (‘‘PCNS’’),
respectively.
4. PCNS satisfies the UFC progressive collapse criteria associated
with both nonlinear static and nonlinear dynamic procedures.
However, PCNS does not meet UFC criteria associated with the
linear static procedure. In contrast, PCND satisfies only the UFC
criteria associated with the nonlinear dynamic procedure but
331
neither linear static nor nonlinear static procedure. Although
use of nonlinear static and dynamic procedures leads to more
economical designs, the associated modeling and computing
efforts can be enormous.
5. Beam failures identified by the UFC alternate path analysis
in this paper are all due to the inadequate plastic rotation
capacities of beam-to-column connections. Because WUF-type
connections are used in the present frame design, an increase
in plastic rotation capacities of connections inevitably requires
an increase in beam sizes. This observation implies that,
for practical design or retrofit purposes, it is also possible,
sometimes maybe more cost-effective, to enhance structural
resistance to progressive collapse by adopting other types
of connections that have significantly more plastic rotation
capacities.
It should be emphasized that the above conclusions may hold
valid only for the present example frame with the chosen layout
configuration. For demonstration purposes, a 2D analytical frame
model has been used in the alternate path analysis. However,
for design of practical building structures, a 3D analytical model
may be needed to reflect the actual load redistribution behavior
by taking into account important structural effects such as
membrane actions due to the presence of floor/roof slabs. Besides,
frame action is the sole source of progressive collapse resistance
considered in this paper. This is because inelastic deformations
corresponding to the UFC acceptance criteria are not large enough
to activate catenary actions in beams. Finally, the conclusions
drawn in this study are solely based on comparison of different
designs obtained by weight minimization. For frames that are
not designed through optimization, additional inherent reserve
strength could be available to help resist progressive collapse.
Acknowledgement
The author thanks Professor Mark Aschheim of Santa Clara
University for his valuable advice on nonlinear analysis using the
DRAIN-2DX program.
332 M. Liu / Journal of Constructional Steel Research 67 (2011) 322–332
References
[1] ACI. Building code requirements for structural concrete (ACI 318-05).
Farmington Hills (MI): American Concrete Institute; 2005.
[2] UFC. United facilities criteria design of buildings to resist progressive collapse
(UFC 4-023-03). Washington (DC): Department of Defense; 2009.
[3] Foley CM, Schneeman C, Barnes K. Quantifying and enhancing the robustness
in steel structures: part 1—moment-resisting frames & part 2—floor framing
systems. AISC Engineering Journal 2008;45(4):247–86.
[4] Liu M, Burns SA, Wen YK. Genetic algorithm based construction-conscious
minimum weight design of seismic steel moment-resisting frames. ASCE
Journal of Structural Engineering 2006;132(1):50–8.
[5] Balling RJ. Optimal steel frame design by simulated annealing. ASCE Journal of
Structural Engineering 1991;117(6):1780–95.
[6] Grierson DE, Pak WH. Optimal sizing, geometrical and topological design using
a genetic algorithm. Structural and Multidisciplinary Optimization 1993;6(3):
151–9.
[7] Bland JA. Discrete-variable optimal structural design using tabu search.
Structural and Multidisciplinary Optimization 1995;10(2):87–93.
[8] Gholizadeh S, Salajegheh E. Optimal seismic design of steel structures by
an efficient soft computing based algorithm. Journal of Constructional Steel
Research 2010;66(1):85–95.
[9] Kaveh A, Azar BF, Hadidi A, Sorochi FR, Talatahari S. Performance-based
seismic design of steel frames using ant colony optimization. Journal of
Constructional Steel Research 2010;66(4):566–74.
[10] Ettouney M, Smilowitz R, Tang M, Hapij A. Global system considerations for
progressive collapse with extensions to other natural and man-made hazards.
ASCE Journal of Performance of Constructed Facilities 2006;20(4):403–17.
[11] Bažant ZP, Verdure M. Mechanics of progressive collapse: learning from world
trade center and building demolitions. ASCE Journal of Engineering Mechanics
2007;133(3):308–19.
[12] Sasani M, Sagiroglu S. Progressive collapse resistance of hotel San Diego. ASCE
Journal of Structural Engineering 2008;134(3):478–88.
[13] Fu F. Progressive collapse analysis of high-rise building with 3-D finite
element modeling method. Journal of Constructional Steel Research 2009;
65(6):1269–78.
[14] Grierson DE, Khajehpour S. Method for conceptual design applied to office
buildings. ASCE Journal of Computing in Civil Engineering 2002;16(2):83–103.
[15] AISC. Seismic provisions for structural steel buildings (ANSI/AISC 341-05).
Chicago (IL): American Institute of Steel Construction; 2005.
[16] ASCE. Minimum design loads for buildings and other structures (ASCE 7-05).
New York (NY): American Society of Civil Engineers; 2006.
[17] ASCE. Seismic rehabilitation of existing buildings (ASCE 41-06). New York
(NY): American Society of Civil Engineers; 2007.
[18] Prakash V, Powell GH, Campbell S. DRAIN-2DX base program description
and user guide. Report no. UCB/SEMM-93/17, version 1.10. Berkeley (CA):
Department of Civil and Environmental Engineering, University of California,
Berkeley; 1993.