Engineering Structures 33 (2011) 2653–2661

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Pushdown resistance as a measure of robustness in progressive collapse analysis

Kapil Khandelwal a,∗ , Sherif El-Tawil b
a
Department of Civil Engrg. and Geo. Sci., University of Notre Dame, Notre Dame, IN 46556, United States
b
Department of Civil & Env. Engrg., University of Michigan, Ann Arbor, MI 48109-2125, United States

article info abstract

Article history: This paper presents a technique termed ‘pushdown analysis’ that can be used to investigate the robustness
Received 21 September 2010 of building systems by computing residual capacity and establishing collapse modes of a damaged
Received in revised form structure. The proposed method is inspired by the pushover method commonly used in earthquake
4 May 2011
engineering. Three variants of the technique, termed uniform pushdown, bay pushdown and incremental
Accepted 5 May 2011
Available online 25 June 2011
dynamic pushdown, are suggested and exercised using nonlinear analysis on 10-story steel moment
frames designed for moderate and high levels of seismic risk. Simulation results show that the frame
Keywords:
designed for high seismic risk is more robust than the corresponding one designed for moderate seismic
Pushdown analysis method risk. The improved performance is attributed to the influence of seismic detailing, specifically, the
Progressive collapse presence of reduced beam sections and stronger columns. It is shown that the dynamic impact factors
Seismic detailing associated with column removal are significantly lower than the commonly used value of 2.0 and are in
Steel moment frames line with lower values in the guidelines recently proposed by the US Department of Defense. The study
Finite element suggests that seismic ‘fuses’ can play a role in the design for robustness and a discussion of the implications
Macromodel of this observation is provided.
© 2011 Elsevier Ltd. All rights reserved.

1. Introduction whereby important buildings could be characterized by some

Robustness is a broad term used in a variety of contexts.
This is usually used in engineering applications to measure ‘‘the
resistance against breaking’’ of an underlying quantity of interest.
In structural control, for instance, a controller is said to be robust if
it can perform satisfactorily (does not break down) for admissible
perturbations in system properties and/or loading conditions.
Similarly, a robust computer code is one that does not crash
when unexpected computational errors are encountered, such as
division by zero. In the context of progressive collapse, robustness
is broadly defined as a measure of the ability of a building system to
carry most of its usual functions in the presence of local component
failures. Specifically, structural robustness is a measure of the
capacity of a building system to withstand loss of local load
carrying capacity.
There is a growing consensus in the structural engineering
community that there is a need to quantify robustness for buildings
that are susceptible to element loss, e.g. due to blast or impact. Such
a measure could be used to provide a means for quantifying desired
system performance, which could then be tied to the economy
of the system. It could be used for the purposes of classification,
∗
Corresponding author.
E-mail addresses: kapil.khandelwal@nd.edu (K. Khandelwal),
eltawil@umich.edu (S. El-Tawil).
0141-0296/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2011.05.013
minimum level of robustness. Information about robustness is also
necessary to decide if a structure is safe for continued occupancy
after local distress or whether extensive repairs are needed before
the structure can be deemed safe.
Research efforts to quantify robustness, as applied to buildings
vulnerable to collapse, are quite limited. In a pilot project for
the study discussed herein, Khandelwal and El-Tawil [1] proposed
pushdown analysis to quantify the robustness of a structure with
lost critical members. Izzuddin et al. [2] proposed three factors
for measuring robustness including energy absorption capacity,
ductility supply and redundancy. They concluded that each of the
three factors cannot be used as a standalone measure of robustness,
but that system pseudostatic capacity, which aggregates all
the three factors, could be a suitable measure. Kim et al. [3]
investigated robustness by gradually pushing down at the location
of a removed column. They showed that collapse capacity was a
function of the number of stories, number of spans, and length
of spans. They compared their results to those from incremental
nonlinear dynamic analyses and concluded that pushing down at
the damaged column location could overestimate the progressive
collapse capacity of a structure.
In other recent studies, the authors have investigated the pro-
gressive collapse behavior of seismically designed steel building
frames [4,5]. In those studies, progressive collapse of steel frames
was investigated using the Alternate Path Method (APM) within a
nonlinear dynamic analysis framework. The APM, which is a threat
2654 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661
independent methodology advocated by the GSA [6] and UFC [7],
is generally applied in the context of a ‘missing column’ scenario
to assess the potential for progressive collapse. The authors con-
cluded that though APM can be used to investigate progressive col-
lapse behavior, it could not be used to measure ‘‘robustness’’ of the
structural system in cases where the structure under consideration
is deemed to be able to survive loss of critical members. In partic-
ular, it is unable to explicitly determine if a structure is near an
incipient collapse state.
In this paper, pushdown analysis is presented as a means for
evaluating the robustness of building systems that have lost critical
members. The study presented herein differs from previous studies
that addressed pushdown response in: (1) its focus on the residual
capacity and associated collapse modes, particularly for seismically
designed buildings, and (2) the means by which the method
is applied. The proposed method is inspired by the pushover
method (both static and incremental dynamic) commonly used
for assessing the seismic resistance of building structures (see, for
example, FEMA-350 [8]).
2. Pushdown analysis method
The proposed pushdown analysis method consists of analyzing
the structure, which has suffered loss of one or more critical
members, under increasing gravity loads. The gravity loads are
incremented until collapse of the structure occurs, defined as
an inability to support the applied loading. Usually this state is
reached after substantial changes in the geometric configuration
have occurred accompanied by member separation from the main
structural system. The load corresponding to this condition is
defined as the failure load. The capacity of the structure at this
point is expressed in terms of the overload factor (Eq. (1)), defined
as the ratio of failure load to the nominal gravity loads.
Failure load
Overload Factor (OF) = . (1)
Nominal gravity loads
Pushdown analysis of a damaged structure is accomplished in
three different ways: Uniform Pushdown (UP); Bay Pushdown
(BP); and Incremental Dynamic Pushdown (IDP). The overload fac-
tors computed from these methods, together with the correspond-
ing collapse modes, are proposed as measures of the robustness of
the structural system in question. In cases where APM shows that
the structure is not capable of successfully absorbing the loss of lo-
cal resistance, the structure is deemed to have no robustness. In
such a situation, the proposed pushdown methods are not appli-
cable.
In the UP case, gravity loads on the entire damaged structure
are increased proportionally within a nonlinear static analysis
framework until the system collapses. A UP analysis will lead to
a collapse state corresponding to failure of the weakest part of
the damaged structure and failure may occur outside the damaged
bays. For example, a gravity bay may dominate the collapse
response by failing prematurely. This method may, therefore, not
adequately consider the damaged bays nor capture the propensity
for collapse to propagate from damaged bays to adjacent ones.
The BP method is proposed to focus attention only on the
damaged bays. In this method, the gravity load is increased
proportionally only in the bays that suffered damage until the
system collapses. The remaining part of the structure is only
subjected to nominal gravity loads. Therefore, this analysis will
lead to a collapse state corresponding to failure in the damaged
bays. The residual capacity of the system is measured in terms of
overload factor calculated as the ratio between the load leading to
failure within the damaged bays and the nominal gravity load.
The IDP method is inspired by the incremental dynamic analysis
method used in earthquake engineering [9]. In IDP, successive
dynamic analyses with increasing gravity loads in the bays of
interest are conducted until an overload factor corresponding to
failure in the damaged bays is established. In each dynamic analysis
case, the system is first assumed to be undamaged while the
loading is being applied. As soon as the dynamic effects associated
with the applied loading die away, members designated as ‘lost’
are instantaneously deleted and the system is allowed to respond
in an inelastic manner. Unlike the UP and BP methods, this analysis
method explicitly accounts for dynamic effects. The disadvantage
is that it is costly in terms of required computational effort because
multiple nonlinear, dynamic analyses must be conducted.
The collapse modes for BP and IDP cases are classified in
terms of the extent of collapse associated with a particular failure
condition. For instance, in some cases collapse may be limited
to just the damaged bay, while in other cases the collapse may
propagate to adjacent bays. The first collapse model is designated
as Contained Collapse Mode (CCM) since the damage is contained
in a defined portion of the structure. The second collapse mode,
where collapse propagates to adjacent bays, is identified as
Propagating Collapse Mode (PCM).
To demonstrate advantages and disadvantages, the proposed
analysis methods are exercised on 10-story steel moment frames
designed for moderate and high levels of seismic risk. The
progressive collapse behavior of these frames was investigated
earlier by the authors [4].
3. Prototype structures
The National Institute of Standards and Technology (NIST)
designed prototype steel framed buildings for the purpose of
studying their response to an event which may cause progressive
collapse [10]. The buildings are 10-story office buildings with plan
dimensions of 45.7 × 30.5 m and utilize moment-resisting frames
as the lateral load-resisting system. The buildings are designed
for: (1) Seismic Design Category C (Atlanta, Georgia), which results
in Intermediate Moment Frames (IMFs) as defined is the AISC
Seismic Provisions [11], and (2) Seismic Design Category D (Seattle,
Washington), which results in Special Moment Frames (SMFs). The
two seismic design categories address moderate and high seismic
risk and are considered to study the effect of seismic design and
detailing on robustness of the steel building systems.
The design loads on the buildings are determined based
on the International Building Code [12]. The material design
standards used in the design of members and their connections
are those referenced in the AISC Load and Resistance Factor Design
Specifications for Structural Steel Buildings [13] and the AISC
Seismic Provisions for Structural Steel Buildings [11]. For typical
floors, the dead load consists of the self-weight of the slab of
2.2 kN/m2 and a super-imposed dead load of 1.44 kN/m2 ; while
the design live load is assumed to be 4.79 kN/m2 . For the roof,
the self-weight of the slab is 2.2 kN/m2 , the super-imposed dead
load is 0.48 kN/m2 ; and the design live load is 0.96 kN/m2 . The
reduction in live loads is based on IBC 1607.9.1 [12].
The lateral load-resisting system is comprised of moment
frames and a gravity system. The design of the gravity system
is the same for the IMF and SMF buildings. Beams and columns
in the gravity system are connected through shear connections,
which are comprised of single plate, shear tab connections that
are fillet welded to the column and bolted using 22 mm, A325
high strength bolts to 9.5 mm A36 shear tabs. The IMF building
employs welded unreinforced flange—welded web moment-
resisting connections. The SMF building employs reduced beam
section (RBS) connections. A992 structural steel (Fy = 345 MPa)
is used for all beams and columns. Furthermore, a 50% flange
reduction is assumed in all RBS connections. Plan views of the
buildings are shown in Fig. 1, while the elevation of the East–West
frames considered in this paper are shown in Fig. 2.
 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661 2655

(a) IMF building system. (b) SMF building system.

Fig. 1. Plan layout for IMF and SMF building systems.

(a) IMF building frame (SDC-C) E–W elevation (line 6). (b) SMF building frame (SDC-C) E–W elevation (line 6).

Fig. 2. Elevation of frames under consideration.

4. Model description
Appropriate structural models must be used for carrying out
the proposed pushdown analyses and the structural models should
be able to represent all important physical processes associated
with the collapse response of the building system. In particular,
the models must be able to adequately account for the formation of
catenary action and the resulting interaction that occurs between
axial tensile loads and moments in beam–column members.
Moreover, the models must also be capable of representing
the responses that influence beam–column connection response,
such as local buckling at the beam–column interface, beam
lateral torsional buckling, inelastic panel zone behavior, local
flange yielding in reduced beam connections, and connection
fracture [14]. The models shown in Fig. 3 are suitable for modeling
these effects and represent the most important components
contributing to inelastic connection behavior in both shear and
moment-resisting steel connections, respectively [4].
The connection model shown in Fig. 3(a) represents the
commonly used single plate shear tab connection for joining
gravity floor beams and girders. In this model, connection
resistance is modeled by a set of springs, which represent the
binding effect associated with the bottom beam flange bearing
on the column flange, bolt/shear tab interaction, and concrete
slab behavior, respectively. The element formulation recognizes
the interaction between stress tensor components through a J2
plasticity model that was implemented in LS-DYNA as a user
defined model by Khandelwal and El-Tawil [1]. The panel zone
in both shear and moment connections is modeled using a
representation that enforces pure shear deformation. As shown in
Fig. 3, the panel zone model is comprised of 4 rigid bars pinned
together at their ends to permit the desired deformation to occur.
The stiffness and strength of the panel zone is provided by a
diagonal spring joining opposite corners of the panel zone.
Beams and column members outside the transfer area are
represented using a Hughes–Liu beam–column element formula-
tion [15]. This fiber-section formulation samples inelastic behavior
at one point along the axis of the element and at multiple points
across the cross section. The location of integration points in a typ-
ical cross section is shown in Fig. 3(b). The radius cut reduced beam
2656 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661
(a) Shear connection model.
Fig. 3. Details of connection and members models.
sections in SMF connections are modeled with a beam element
of length equal to that of the RBS region but with cross section
properties corresponding to that of the minimum cross section in
the reduced section. As described in [4], the models in Fig. 3 are
capable of representing local behavior such as local buckling and
fracture by carefully tailoring the stress–strain response at each
integration point. The models are also capable of capturing the in-
teraction between moments and axial catenary loads that com-
monly occur during progressive collapse analysis. The models in
Fig. 3 were validated by the authors. Further specifics pertaining to
calibration and validation studies can be found in [4].
The building frames in Figs. 1 and 2 are represented using
the above-described member and connection models. For the
purposes of analysis, only in-plane response of the frame is
considered. However, ground story columns are allowed to deform
out of plane to account for weak axis buckling of those particular
columns. Ground story columns are provided with out of plane
imperfections of L/250 (L = length of column) at midheight to
promote the initiation of buckling, if needed, and ground story
column end conditions are modeled as fixed in plane but hinged
out of plane. The deformed shape of each column is modeled
as sinusoidal and the imperfections for all columns are assumed
to occur in the same direction. The level of imperfection and
boundary conditions are selected so that the load at which the
columns buckle are close to those computed from the AISC-
LRFD [13] Specifications for columns with fixed–pinned ends and
without application of the strength reduction factor.
5. Analysis setup and limitations
Gravity loads are computed from the design specifications
in [10] and account for the dead loads plus 25% of the live loads. The
simulations are conducted with 5% mass proportional damping and
are carried out using the explicit finite element code LS-DYNA [15].
For UP and BP, gravity load is increased to its full value in 5 s and
then the analysis is continued with proportional increase in gravity
loads as discussed earlier. The loading rate employed here ensures
that dynamic effects are minimized in the subsequent portion of
the analysis. For IDP, gravity loads are first slowly applied to avoid
exciting dynamic effects (increased to its full value in 5 s as in the
UP and BP cases). Once the gravity loads have been fully applied,
a 1st floor column is instantaneously deleted and the subsequent
inelastic response of frame is then investigated.
The developed models have a number of assumptions and
limitations that should be considered when evaluating the results
from these models. These limitations are outlined in [4] and
pertain to the assumed nature of the foundations, the effect of
(b) Moment connection model.
Table 1
Alternate path method (APM) analysis cases.
APM case Member removed Building frame
1 Column C-1 IMF
2 Column D-1 IMF
3 Column E-1 IMF
4 Column F-1 IMF
5 Column D-1 SMF
6 Column E-1 SMF
7 Column F-1 SMF
debris impact loads, the location of fracture that leads to member
separation, strain rate effects, 2D versus 3D model representation,
and slab effects. Readers are referred to Khandelwal et al. [4] for
additional details.
6. Analysis results
Pushdown analyses of the above-described IMF and SMF are
carried out using the proposed analysis methods. Key results of
interest, including overload factors and collapse modes for the
three proposed methods, were obtained for analysis cases where
APM showed that the system under consideration is able to survive
member loss as outlined in [4]. Table 1 gives a summary of these
cases. Note that the IMF did not survive loss of columns A and B [4]
and so these cases are excluded from analysis. Also, the SMF is
symmetric, so only half of the column loss cases are considered.
The corresponding pushdown analysis results for the two frames
are shown in Tables 2 and 3 and the results are discussed in detail
below. Both tables summarize the overload factor (computed from
Eq. (1)), collapse mode (CCM or PCM), and the mode by which
failure initiates.
To facilitate the following discussion, the columns and beams
are designated using the notation in Fig. 2. For example, column
C-1 represents a first story column in column line C (Fig. 2).
Similarly, beam CD-2 represents a second story beam in bay CD
(Fig. 2). Analysis cases as designated by the type of analysis and
an appended number that refers to the APM case in Table 1. For
example, UP-1 implies a Uniform Pushdown analysis for APM Case
1 described in Table 1, while IDP-7 is an Incremental Dynamic
Pushdown for APM Case 7.
6.1. Results for IMF building
Several observations can be made from the analysis results in
Table 2. First, it is clear that by any of the measures employed,
the IMF has significant robustness, i.e. resistance to collapse. The
lowest factor in the table is 1.4 and belongs to IDP-1. Second, UP
cases have equal or lower overload factor when compared to the
 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661 2657

Table 2
Pushdown analysis results—IMF building.
Pushdown analysis type Loading type Overload factor Collapse mode Failure initiation

UP-1 Uniform 1.6 – Buckling of column D-1

UP-2 Uniform 2.5 – Buckling of column E-1
UP-3 Uniform 2.5 – Buckling of column D-1
UP-4 Uniform 2.1 – Buckling of column E-1
BP-1 Bay BC & CD 1.6 PCM Buckling of column D-1
BP-2 Bay CD & DE 2.9 PCM Buckling of column E-1
BP-3 Bay DE & EF 2.9 PCM Buckling of column D-1
BP-4 Bay EF 2.3 PCM Buckling of column E-1
IDP-1 Bay BC & CD 1.4 PCM Buckling of column D-1
IDP-2 Bay CD & DE 2.3 PCM Buckling of column E-1
IDP-3 Bay DE & EF 2.3 PCM Buckling of column D-1
IDP-4 Bay EF 1.8 PCM Buckling of column E-1

Table 3
Pushdown analysis results—SMF building.
Pushdown analysis case Loading Overload factor Collapse mode Failure initiation

UP-5 Uniform 1.8 – Failure of shear connections in bay CD

UP-6 Uniform 3.2 – Failure of shear connections in bay CD
UP-7 Uniform 2.9 – Buckling of Column E-1
BP-5 Bay CD & DE 1.8 CCM Failure of shear connections in bay CD
BP-6 Bay DE & EF 3.6 CCM Fracture in RBS in bay DE and EF
BP-7 Bay EF 3.5 CCM Fracture in RBS in bay EF
IDP-5 Bay CD & DE 1.7 CCM Failure of shear connections in bay CD
IDP-6 Bay DE & EF 3.1 CCM Fracture in RBS in bay DE and EF
IDP-7 Bay EF 2.4 CCM Fracture in RBS in bay EF

corresponding BP cases. This is expected because the structure is
under higher overall loads in the former compared to the latter,
and will overload weaker parts of the structure causing premature
failure. Third, loading cases involving only moment bays (e.g. BP-2)
have higher overload factors compared to cases involving gravity
bays (e.g. BP-1), primarily because they are comprised of stronger
members.
Table 2 shows that the collapse initiation mode is similar for
UP, BP and IDP cases. Collapse is typically initiated by out of
plane buckling of the ground story columns. The corresponding
failure modes for these analysis cases are shown in Fig. 4. After an
overloaded column buckles, the loads are transferred to adjacent
bays leading to additional column buckling, i.e. failure propagates
to adjacent bays. The corresponding collapse modes are therefore
designated as propagating collapse modes (PCMs) since failure
extends to adjacent bays compromising the rest of the system.
When failure modes are similar, which they are for the cases
considered herein, the overload factor for IDP cases is lower than
that in the corresponding BP cases because of dynamic effects. The
following discussion focuses only on analysis results for BP cases
since the collapse modes are similar for all three load types.
In case BP-1, the peak axial load supported by column D-1 at
failure is 5516 kN, which closely matches the column’s design
capacity of 5000 kN (W18 × 119 column, Fy = 345 MPa, Ky = 1.0,
φ = 1.0, L = 4.6 m). The vertical displacement at the removed
column is 271 mm at the point of failure and the peak axial forces
developed in beams BC-1 and CD-1 are 249 kN (tension) and
−383 kN (compression), respectively. The tensile load in Beam
BC-1 eventually leads to tension-related failure of the shear tab
connection. Beam CD-1 is in compression because of the global
frame action in the system. As a result of frame-wide cantilever
action in the moment bay CD above the location of the lost column,
beams at lower levels see compression whereas beams at higher
levels are subjected to tensile forces.
In BP-2, failure initiates by buckling of column E-1 (at 5605 kN)
followed by buckling of column C-1. In this case, axial forces
of 236 kN (tension) and 214 kN (tension) developed in beams
CD-1 and DE-1 at a vertical displacement of 143 mm (at the
removed column) when buckling initiates. Case BP-3, is similar
to BP-2. Failure initiates by buckling of column D-1, at a peak
load of 5605 kN in the column. At this stage, the displacement
at the removed column is 142 mm. Buckling of column D-1
is immediately followed by failure of beams in the connection
regions in bays DE and EF. The axial force developed in beams DE-1
and EF-1 are 214 kN (tension) and 235 kN (tension), respectively.
Case BP-4 is similar in many respects to BP-1. Failure initiates
by buckling of column E-1 (at 5542 kN) when the vertical
displacement at the removed column is 180 mm. Beam EF-
1 is actually under compression when failure initiates, again
emphasizing the global cantilever action that takes place above the
removed column.
6.2. Results for SMF building
Analysis results for the SMF building are reported in Table 3,
and some of the corresponding collapse modes are shown in Figs. 5
and 6. The general observations made for the IMF are also seen
here, i.e. the SMF has substantial overload capacity, UP cases have
equal or lower overload factors than corresponding BP cases, and
loading involving moment bays only leads to higher overload
factors compared to cases involving gravity bays.
For analysis cases UP-5 and BP-5 (Fig. 5(a)), collapse starts as
a result of failure of shear connections in gravity bay CD. The
shear connection in beam CD-1 (left end) fails first and is followed
by failure of shear connections in the upper stories. A peak axial
force of 378 kN (tension) is developed in beam CD-1 prior to
separation from the rest of the frame. Failure of shear connections
is followed by failure of some moment connections in bay DE.
An overload factor of 1.8 is achieved in both analyses cases. The
associated collapse mode for BP-5 is considered to be CCM because
the collapse is limited to just the damaged bays.
Unlike the IMF, some differences are noted in the SMF building
in the collapse modes associated with the UP and BP cases. In the
UP-6 case, for example, collapse occurred due to shear connection
failures in gravity bay CD, i.e. collapse occurs outside the damaged
bay due to overloading in the weaker gravity bay. On the other
hand, BP-6 suffers collapse due to fracture in the RBS connection
regions in bays DE and EF (Fig. 5(b)). A similar situation can be seen
2658 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661

(a) Failure modes: Pushdown analysis cases UP-1, (b) Failure modes: Pushdown analysis cases UP-2, BP-2 (c) Failure modes: Pushdown analysis cases UP-3, BP-3
BP-1 and IDP-1. and IDP-2. and IDP-3.

(d) Failure modes: Pushdown analysis cases UP-4,

BP-4 and IDP-4.

Fig. 4. Propagating collapse modes (PCMs) associated with IMF building.

in UP-7 and BP-7. The former collapses due to buckling of column
E-1 (column force at buckling is 7473 kN), while the latter fails by
fracture in the connection regions (RBS zones) in bay EF (Fig. 5(c)).
There is computational [14] and experimental [16] evidence that
RBS connections will fracture when subjected to collapse-type
loading.
The collapse modes for IDP cases are similar to BP cases except
for a minor variation between BP-5 and IDP-5. In the latter, collapse
is triggered by shear connection failure in beam CD-1 (right end) as
opposed to shear connection failure in beam CD-1 (left end) for the
former. Another important feature observed for case IDP-5 is that
shear connections failed incrementally as opposed to complete bay
collapse. For instance, only a partial story collapse occurs with
failure of shear connections in gravity beams CD-1 and CD-2 at an
overload factor of 1.5 (Fig. 6(a)). The number of shear connection
failures in gravity bay CD increases with increasing load factor
(Fig. 6(b)) and complete collapse of bays CD and DE occur at a load
factor of 1.7 (Fig. 6(c)). It is notable that all the associated collapse
modes for the SMF building are CCM.
7. Discussion of results
7.1. Dynamic increase factor and analysis methods
The simulation results offer a means for computing the dynamic
increase factor (DIF), which embodies the effect of dynamic loading
on system response. The DIF can be computed as the ratio of
overload factors for BP (quasi-static loading) and IDP (dynamic
loading) cases. The DIF for the IMF building ranges from 1.14 to
1.27, whereas it ranges from 1.06 to 1.45 for the SMF building.
Lower DIFs are observed when one of the bays involved in the
computation is a gravity bay. This suggests that dynamic effects
vary depending on the type of structural system, and that in
both systems, dynamic effects are not as high as 2.0, which is
specified in GSA [6]. This observation is in accord with the growing
consensus that using DIF = 2.0 is too conservative. The most recent
version of the UFC [7] has specified lower values that are tied
to the type of structure being analyzed. Following the UFC [7]
guidelines, the DIF for the moment bays of the IMF is 1.40 and the
corresponding number for the SMF is 1.22, numbers that are more
in line with the values computed herein. Another key observation
is that while significant differences in collapse modes occurred
between UP and BP cases, especially for SMF building, there is good
correlation between IDP and BP cases, suggesting that BP analysis
is a reasonable way to measure pushdown resistance for the types
of frames discussed herein.
7.2. Seismic detailing
The simulation results suggest that the SMF building designed
for high seismic risk is generally more resistant to progressive
collapse and hence more robust than the IMF building designed for
moderate seismic risk. This is evident from the overload factors,
which for the SMF building range from 1.7 to 3.6, while the IMF
building has overload factors in the range of 1.4–2.9. The better
performance of the SMF building as compared to the IMF one is
directly attributed to the use of seismic detailing, specifically, the
use of RBS connections and stronger columns. In some of the BP and
IDP cases, collapse in the SMF frame is associated with fracture in
the reduced part of the RBS connections. Under overloading, the
reduced sections acted as structural fuses, failing relatively early,
thus limiting the loads transferred to the columns and shielding the
system from a propagating collapse mode. This issue is discussed
 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661
(a) Failure mode: Pushdown analysis case BP-5 and IDP-5.
(c) Failure mode: Pushdown analysis case BP-7 and
IDP-7.
Fig. 5. Contained collapse modes (CCMs) associated with SMF building.
further on in the paper. In contrast, collapse modes for all the
IMF frame cases are associated with buckling of ground story
columns. Such collapse modes are undesirable as they jeopardize
the stability of the entire structural system, i.e. they entail a PCM.
7.3. Development and role of catenary action
The term catenary action refers to the ability of beams to resist
vertical loads through formation of a catenary-like, or cable-like,
mechanism. In general, catenary action is associated with the
development of large enough deformations such that the applied
loads are mainly resisted by the vertical components of axial
forces that develop in the beams, i.e. catenary forces, rather than
the original flexural action. For catenary action to successfully
develop, it is necessary that: (a) beam-to-column connections have
enough ductility to facilitate beam deformation into the catenary
configuration and at the same time have sufficient structural
integrity to support the loads; and (b) sufficient anchoring capacity
exists at the member ends to transfer the resulting catenary forces
to the remainder of the system. Violation of either of these two
conditions implies that catenary action cannot play a significant
role in collapse resistance.
In the IMF frame, the failure modes for UP-2, UP-3, BP-2, BP-3,
IDP-2 and IDP-3 entail relatively small vertical displacements be-
cause the primary failure mechanism is column buckling. In other
words, these systems do not change their geometric configuration
sufficiently to mobilize catenary resistance mechanisms. On the
2659
(b) Failure mode: Pushdown analysis case BP-6 and
IDP-6.
other hand, the strong columns in the SMF preclude premature
column buckling and force true catenary mechanisms to develop
in response to lost columns. For instance, in BP-5, a catenary force
of 378 kN (tension) is developed in beams CD-1 at a vertical dis-
placement of 492 mm before the shear connections fail in direct
tension.
A different situation occurs in BP-6 (Fig. 7) where tensile forces
of 814 kN and 845 kN are developed in beams DE-1 and EF-1,
respectively, at a vertical displacement of 1143 mm. These forces
are well below the catenary capacity of the beams and are
essentially limited by the capacity of the frame. Fig. 7 shows that
widespread axial force demands develop in the affected bays of
BP-6 to help resist the applied loads. Some of the beams are
subjected to compressive axial forces that were developed to
counteract the tensile forces generated at the lowest floor.
However, the axial forces in the frame, even the tensile ones at the
first floor, are still considered to be relatively low, suggesting that
flexural action in the beams still plays a significant role in resisting
the applied loads. Clearly, the resistance mechanism in this case is
a global one that engages the entire frame.
8. Practical implications
Insight into the robustness of seismically designed building
systems can be obtained through the limited simulation studies
conducted with the proposed pushdown analysis methods. In
particular, the study shows that an improved layout (more
perimeter moment bays) and seismic detailing both play a
2660 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661
(a) Overload factor: 1.5.
(c) Overload factor: 1.7.
Fig. 6. Pushdown analysis case IDP-5: D-1 column removed.
Fig. 7. Development of axial forces in the SMF building: BP-6.
significant role in determining robustness. Khandelwal et al. [4]
had also previously noted that improved system layout is a key
reason for the greater collapse resistance associated with SMF
versus IMF.
The IMF and SMF frames fail in fundamentally different ways
under pushdown loading. The IMF, with less stringent seismic
detailing, failed by out of plane column buckling, while the SMF
failed by a more ductile collapse mechanism that entailed the
formation of catenary action. The ductility of the SMF was clearly
(b) Overload factor: 1.6.
necessary for permitting the system to mobilize and benefit from
catenary action. In addition, the seismic RBS detail played the
beneficial role of a ‘fuse’, limiting the load transferred to the
columns and shielding the system from a propagating collapse
mode.
The observation that ‘fuses’ are beneficial after a structure
has exhausted its bridging capacity suggests that structural
systems could potentially be designed in a manner that explicitly
prohibits widespread propagation of collapse. However, practical
implementation of such a design philosophy may be difficult and is
complicated by a number of issues. In particular, system response
is three dimensional, not two dimensional as idealized in this
paper. It is entirely feasible that the 3D response of the system
will mobilize other resistance mechanisms that should be carefully
considered when designing the location and resistance of the
fuses. For example, slabs will contribute significantly to collapse
resistance as discussed in [17,18]. If not taken into consideration,
parts of the system that fail at a fuse may engage the slab,
shifting loading to other parts of the structure and potentially
leading to premature or possibly more widespread failure than
intended. Also, implementation of a fuse design strategy will also
face the hurdle of long-entrenched ideas within the structural
engineering community regarding the link between well-tied
systems and robustness. Regardless of the hurdles, technical or
otherwise, implementation and eventual acceptance of a fuse
or compartmentalized design strategy will require extensive
experimental and computational research and the deployment
of models that are able to explicitly track the 3D effects of fuse
failures.
 K. Khandelwal, S. El-Tawil / Engineering Structures 33 (2011) 2653–2661
9. Summary and conclusions
This paper proposed new analysis techniques that could be
used for investigating the robustness of building systems. Three
pushdown methods were proposed—uniform pushdown (UP),
bay pushdown (BP) and incremental dynamic pushdown (IDP).
The proposed methods were then exercised to investigate the
robustness of two dimensional, 10-story seismically designed
frames, one of which was an intermediate moment frame and the
other a special moment frame. Based on the limited simulation
studies conducted, and within the assumptions and limitations
described in the paper, the following conclusions can be drawn.
1. The proposed pushdown analysis methods can be used to
investigate the robustness of a damaged building system in
terms of residual capacity and associated collapse modes.
2. Incremental dynamic pushdown gives the most realistic
estimate of residual capacity and collapse modes. However,
collapse modes associated with bay pushdown analysis cases
agree well with IDP for the building systems considered in
this study suggesting that static BP analysis is a simpler, more
economical, but still reasonable substitute for IDP.
3. While less conservative and definitely more reasonable than
GSA [6], the UFC guidelines produce DIFs that are more in line
with the values computed in this study.
4. The development of tensile catenary action in some compo-
nents of the damaged system necessitates development of com-
pressive forces in other parts of the system. These force patterns
develop as a result of frame action within the structural system.
5. The simulation results suggest that the SMF building designed
for high seismic risk is generally more resistant to progressive
collapse and hence more robust than the IMF building designed
for moderate seismic risk. This is evident from the overload
factors, which for the SMF building range from 1.7 to 3.6, while
the IMF building has overload factors in the range of 1.4–2.9.
The better performance of the SMF building as compared to the
IMF one is attributed to better layout and the use of seismic
detailing, specifically the use of RBS connections and stronger
columns.
6. A key observation in this study is that seismic ‘fuses’ can
play an important role in collapse response, localizing failure
in the affected bays and preventing a propagating collapse
mode. This observation suggests that structural systems could
be designed in a manner that explicitly prohibits propagation
of a collapse However, practical implementation of such a
design philosophy may be complicated by the need to explicitly
consider and adequately model 3D system response to ensure
that the desired response will occur. Development of such
a design strategy will require extensive experimental and
computational research as well as a shift in current thinking.
2661
Acknowledgments
The presented work was supported in part by the University of
Michigan and the US National Science Foundation through grants
CMMI-0726493 and CMMI-0928547. Any opinions, findings,
conclusions, and recommendations expressed in this paper are
those of the authors and do not necessarily reflect the views of the
sponsors.
References
[1] Khandelwal K, El-Tawil S. Assessment of progressive collapse residual capacity
using pushdown analysis. In: Proceedings of the ASCE structures congress.
2008.
[2] Izzuddin BA, Vlassis AG, Elghazouli AY, et al. Assessment of progressive
collapse in multi-storey buildings. Proc Inst Civ Eng Struct Build 2007;160:
197–205. [publication doi].
[3] Kim T, Kim J, Park J. Investigation of progressive collapse-resisting capability of
steel moment frames using push-down analysis. J Perform Constr Facil, ASCE
2009;23(5):327–35.
[4] Khandelwal K, El-Tawil S, Kunnath SK, Lew HS. Macromodel-based simulation
of progressive collapse: steel frame structures. J Struct Eng 2008;134:1070–8.
[5] Khandelwal K, El-Tawil S, Sadek F. Progressive collapse analysis of seismically
designed steel braced frames. J Constr Steel Res 2009;65:699–708.
[6] GSA. Progressive collapse analysis and design guidelines for new federal office
buildings and major modernization projects. Washington (DC): US General
Service Administration; 2003.
[7] UFC. UFC 4-023-03: design of buildings to resist progressive collapse.
Washington (DC): Department of Defense; 2009.
[8] FEMA-350 . Recommended seismic design criteria for new steel moment-
frame bulidings. Washington (DC): Federal Emergency Management Agency;
2000.
[9] Villaverde R. Methods to assess the seismic collapse capacity of building
structures: state of the art. J Struct Eng 2007;133:57–66.
[10] Liang X, Shen Q, Ghosh SK. Report—assessing ability of seismic structural
systems to withstand progressive collapse: seismic design and progressive
collapse analysis of steel frame buildings. In: SK Ghosh and Associates. 334
E Colfax. Unit E. Palatine, IL 60067. 2007.
[11] AISC-Seismic . ANSI/AISC 341-02: seismic provisions for structural steel
buildings. Chicago (IL): American Institute of Steel Construction; 2002.
[12] IBC. International building code. Washington (DC): International Code Council;
2003.
[13] AISC-LRFD . Manual of steel construction—load and resistant factor design. 3rd
ed. Chicago (IL): American Institute of Steel Construction; 1999.
[14] Khandelwal K, El-Tawil S. Macro models for progressive collapse analysis
of steel moment frame buildings. In: Structures congress 2007. Reston, VA
20191-4400. vol. 2007. 2007.
[15] Hallquist J. LS-DYNA. Livermore (CA): Livermore Software Technology Corp.;
2008.
[16] Sadek F, Main JA, Lew HS, Robert SD, Chiarito VP, El-Tawil S. An experimental
and analytical study of steel moment connections under a column removal
scenario. NIST report. Gaithersburg (MD): National Institute of Standards and
Technology, NIST, US Department of Commerce; 2010.
[17] Sadek F, El-Tawil S, Lew HS. Robustness of composite floor systems with shear
connections: modeling, simulation, and evaluation. J Struct Eng, ASCE 2008;
134(11):1717–25.
[18] Alashker Y, El-Tawil S, Sadek F. Progressive collapse resistance of
steel–concrete composite floors. J Struct Eng 2010;136(10):1187–96.