Documente Academic
Documente Profesional
Documente Cultură
SUMMARY
This paper presents a novel analytical formulation of an element removal algorithm based on dynamic
equilibrium and the resulting transient change in system kinematics, by applying imposed accelerations
instead of external forces at a node where an element was once connected. The algorithm is implemented
into an open-source finite element code, numerically tested using a benchmark structural system with
simplified element removal criteria, and able to capture the effect of uncertainty in member capacity.
Realistic element removal criteria are introduced for mode-dependent gravity load collapse of seismically
deficient and retrofitted reinforced concrete (RC) columns and unreinforced masonry (URM) infill walls.
Two applications are conducted using structural systems of RC frames with URM infill walls. The first is
a probabilistic study of a one-story model subjected to an ensemble of 14 ground motion recordings from
similar neighboring sites during an earthquake event. The study produces empirical probability curves for
partial and complete collapse conditioned on different hazard levels, and concludes that the intra-event
variability is a major source of uncertainty affecting the outcome of progressive collapse simulations.
The second application is a deterministic sensitivity study of progressive collapse response in a five-story
structural model to uncertainty in live load, stiffness, damping, and seismic hazard level, subjected to
one ground motion record. The analysis identifies the time at incipient collapse as an adequate sensitivity
measure, and the uncertainty in ground motion intensity as the most important, followed by the stiffness
of the URM infill wall. Copyright q 2009 John Wiley & Sons, Ltd.
KEY WORDS: dynamic analysis; element removal; finite element; progressive collapse; reinforced
concrete frame; unreinforced masonry infill
∗ Correspondence to: Khalid M. Mosalam, University of California, 721 Davis Hall, Berkeley, CA 94720-1710, U.S.A.
†
E-mail: mosalam@ce.berkeley.edu
‡
Senior Engineer.
§ Professor and Vice Chair.
Contract/grant sponsor: Earthquake Engineering Research Centers Program; contract/grant number: EEC-9701568
INTRODUCTION
Progressive collapse assessment using nonlinear time-history finite element (FE) simulation is
gaining popularity over traditional methods such as alternate path analysis. Applications of tradi-
tional methods include decommissioned [1] and existing [2, 3] reinforced concrete (RC) buildings,
retrofitted steel frames [4], retrofitted buildings [5], multi-span bridges [6], and probabilistic eval-
uation methods [7]. There is limited literature on this field of research as few experimental inves-
tigations have been conducted on RC frames that are redundant enough to experience progressive
collapse, e.g. [8–10]. Experimental data from these studies are intended to establish the behavior
of RC systems after the loss of one or more deficient columns designed to lose axial load capacity.
However, observed collapse modes are difficult to generalize to cases involving disconnection and
subsequent collision by collapsed elements.
A recent study reported in [11] describes an analytical approach to use post-yield strength and
stiffness degradation to conduct quasi-static progressive failure analysis. Another study reported
in [12] defines a macro-level damage index to predict collapse of yielding beam–column elements
based on maximum element deformations and accumulated plastic energy. Collapsed elements are
removed from the structural system, and external nodal forces are applied at the removed element’s
end-nodes using a step function for the duration of the subsequent load step. This approach is
valid for quasi-static behavior but is sensitive to the time step size during dynamic simulations and
may not be representative of the energy imparted into the damaged structure due to the release of
internal forces. In the approach presented in [13], the downward motion of the collapsed element
is tracked using a condensation method that avoids redefining degrees of freedom (DOFs) and
splitting one node into two. This approach is extended in [12] to include a simplified account for
impact on the structure by a collapsed element.
The effect of locally released internal energy upon element collapse on exciting transient vibra-
tions in the damaged structure was presented in [14] using an energy balance approach. The study
concludes that a quasi-static equilibrium-based progressive failure analysis is not conservative,
and confirms this with both experiments and computations on a simple spoked wheel structure.
Another study on a multiple degree-of-freedom (MDOF) truss system [15] demonstrated similar
conclusions. An analytical formulation was developed in [16] which uses an energy balance method
to amplify the displacement demands on a structural system following the removal of a gravity
load-carrying column. It conservatively ignores the energy dissipated by viscous damping in the
system due to the short duration involved, and characterizes the kinetics of the damaged system
at the onset of column collapse using nodal displacements only.
Two studies in [17] and [18] include the effect of uncertainty on simulated progressive collapse
assessment of MDOF building systems. They assume ductile details and adopt for the definition
of collapse a side sway mechanism of columns within an entire floor, not explicitly conducting
any removal of elements during the simulation.
In order to capture the progression of collapse, one must model the behavior of individual
components under extreme loading and identify the criteria for their removal from the structural
model. In this paper, two modes of RC column collapse, namely shear-axial and flexure-axial,
are investigated. Shear-axial collapse was recently investigated in a number of experimental and
analytical studies reviewed in [10]. The shear friction-based model developed in [19] is utilized in
this paper to establish limit-states for shear-axial collapse. The damage indices developed in [20]
for fiber-discretized cross-sections are used for flexure-axial collapse. These indices are material
model dependent and reflect the damage accumulation at individual fibers. For non-ductile and
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 611
max m'a'x X
X
may m'a'y
Figure 1. Dynamic equilibrium of a node connected to a collapsed element: (a) before element
removal and (b) after element removal.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
612 M. TALAAT AND K. M. MOSALAM
the original ones. The dynamic equilibrium of the node under resisting and inertia forces can be
expressed by
⎧ ⎫ ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪ Fx ⎪ ⎪ max ⎪ ⎪
⎪ F ⎪ ⎪ m ax ⎪
⎨ ⎬ ⎨ ⎬ ⎨ x⎪ ⎬ ⎪
⎨ ⎪
⎬
Pex + Fy − ma y = Pex + Fy − m a y (1)
⎪
⎩ ⎪ ⎭ ⎪⎩ ⎪
⎭ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
I ⎩ ⎭ ⎩ ⎭
M M I
where Pex is the vector of externally applied nodal loads, if any, assumed independent of the
element connectivity and not affected by the removal of a collapsed element. The summation is
conducted for the elements connected to the node. Upon abrupt element collapse and removal, this
dynamic equilibrium is disturbed by the sudden release of internal forces from element C1 and must
be restored before the solution continues. The resisting forces in the connected elements, namely
F and M , can only change as a result of updated element deformations, which require changes in
relative displacements (and possibly velocities) between the elements’ end-nodes. These changes
result from updated nodal velocities and accelerations and do not take place instantaneously. The
inertia forces are directly related to nodal accelerations ax , a y , and , which need to change abruptly
to satisfy dynamic equilibrium at time t . Since the connected elements’ end-nodes other than their
shared node are under dynamic equilibrium themselves, this abrupt change is first localized at the
shared node at time t . In the following time steps, element removal superposes a case of transient
loading on the damaged system, whereby the resulting changes in nodal kinetics lead to an updated
set of resisting (and inertia) forces that propagate outward through the connected elements into
neighboring nodes. These updated nodal forces must satisfy dynamic equilibrium of their respective
DOFs during every time step, which results in updated inertia forces and nodal accelerations. This
process continues to propagate in the damaged system through element connectivity until updated
nodal configurations corresponding to a new equilibrium state are reached.
Upon reaching the new equilibrium state, the nodal velocities of the damaged system result in
overshooting the corresponding displacements, amplifying deformation demand in the structural
elements. In the hypothetical absence of external excitation, free vibration of the system ensues.
An initial transient phase is followed by a steady-state phase oscillating about the new equilibrium
state. The new deformation demands may lead to the collapse of additional elements, which would
be ‘safe’ in the new equilibrium state using quasi-static analysis. If so, a second transient phase is
excited, and this process continues until the system either (a) reaches an equilibrium state about
which it can safely oscillate and the progression of collapse is arrested; (b) experiences overall
gravity load collapse; or (c) experiences partial collapse that is nevertheless compartmentalized
while the rest of the structural system survives. In case (c), the structure will probably need to
be ultimately demolished, yet the prospect of maintaining life safety is enhanced compared with
case (b).
In the presence of an earthquake excitation, the input ground motion is treated as equivalent
external loads. The input time step may be larger than that required for capturing transient effects,
particularly if reaching a new equilibrium state requires local axial vibrations in stiff structural
elements with significantly short natural periods. Hence, shorter time steps are needed immediately
after element removal, during which the input ground motion is interpolated. An adaptive time
stepping scheme can be used for computational efficiency.
The advantages of explicitly removing elements instead of assigning them low stiffness are
threefold. First, it avoids numerical problems associated with ill-conditioned stiffness matrices.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 613
Second, enforcing dynamic equilibrium enables the computation of the resulting increase in nodal
accelerations and the inclusion of the system’s complete kinematic state at the time of element
collapse in determining whether it can successfully survive to a new equilibrium state. Third, the
motion of the collapsed element can be tracked relative to the damaged system to estimate the
time and kinetics at subsequent collision, e.g. using rigid body analysis of the collapsed element
motion or a condensation approach [13] where DOFs redefinition in the damaged system is not
required. If collision is detected, a ‘soft-impact’ analysis can characterize the duration, forces,
and redistribution of masses [23]. This approach assumes negligible relative displacement during
impact between the collapsed element and the damaged system at the impact point and low-velocity
impact that precludes punching and penetration [24].
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
614 M. TALAAT AND K. M. MOSALAM
Intact structure
Nodal loads
Dangling
node
Floating
element
Dashes represent
collapsed elements
Figure 3. Elements, nodes, and loads requiring removal due to element removal.
listed element properties (elastic modulus, E, cross-sectional area, A, and moment of inertia, I )
in Table I. The (inclined) truss members exhibit a linear elastic behavior up to an axial load Fy ,
then exhibit either brittle or perfectly plastic behavior up to a ductility limit, Table II, before
removal. Four case studies are used to explore the sensitivity of the algorithm to uncertainty in
element capacities. The (horizontal) beam elements are linear elastic. Nonlinear geometry and large
deformation effects are included in the FE model by using co-rotational geometric transformation
from element quantities, e.g. orientation, to global coordinates [26]. The mass is lumped at the
nodes, Table III. The gravity loads are applied followed by support excitation at nodes 2–5 and 12
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 615
Table II. Ultimate truss element ductility limits for benchmark problem.
Truss element Case 1 Case 2 Case 3 Case 4
1, 11 5 5 5 5
3 5 5 1 1
10 5 1 5 1
◦
in the vertical direction using the earthquake record from Northridge 1994, Tarzana station, 90
component, with scaled peak ground acceleration (PGA) of 1.04 g. For the analysis objective, a
time step of 0.005 s is chosen with no viscous damping.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
616 M. TALAAT AND K. M. MOSALAM
It is assumed that collapsed truss elements separate from the structural system at the location
of their connection with the beams. Hence, the collapse of element 3 causes it to fall and to
collide with the lower canopy, slightly to the right of node 9. For simplicity of the analysis,
collision is assumed to take place exactly at node 9. After collision, element 3 is assumed to
deform in flexure, simply supported at nodes 4 and 9. The energy of the impact is assumed
to dissipate during one half-cycle of such flexural deformation before rapidly decaying, e.g.
due to friction, and element 3 remains passively supported by nodes 4 and 9 until collapse of
the lower canopy. In addition to impact forces, it is assumed that collision with (or loss of) a
truss element at a node results in updating its assigned mass and gravity loads by adding (or
subtracting) 2 ton.
All simulations were successfully conducted without exhibiting numerical convergence prob-
lems. Cases 1 and 2 result in similar responses where no element collapse takes place. In both
cases, the upper canopy elements 1 and 3 exhibit brief yielding, inducing a ductility demand of
1.17 and 1.11, respectively. Cases 3 and 4 result in collapse of element 3 onto the lower canopy
at 8.875 s. As a result, the ductility demand on element 1 increases to only 2.1; and the upper
canopy exhibits no further collapse. Figure 5 illustrates the time history of the normalized axial
forces using their respective values at yield in truss elements 1 and 3 for all four cases. The
response of element 3 is identical for all cases prior to reaching its ultimate ductility limit and
collapsing in Cases 3 and 4. The response of element 1 is identical for all cases prior to 8.875 s.
Subsequent response exhibits higher average force values in Cases 3 and 4 and a longer period of
yielding before the damaged upper canopy starts to oscillate about a new equilibrium state. The
collapse of element 3 results in larger permanent displacement in the vertical direction at node 1,
Figure 6(a), where the nodal displacements are normalized using their respective values caused by
the application of the gravity load.
In the lower canopy, element 3 collides with node 9 at 11.26 s, and the resulting demand causes
yielding in elements 10 and 11. In Case 3, the ductility capacity in both elements is sufficient
to prevent the progression of collapse in the system, which reaches a new equilibrium state and
starts oscillating about it, Figure 6(b). In Case 4, brittle element 10 collapses at 12.37 s. Shortly
afterwards, element 11 reaches its ultimate ductility limit and collapses at 14.57 s, leading to
complete collapse of the lower canopy, Figure 6(b). The deformed shapes at 14.5 s and collapse
1.5
Element 1 Time = 8.875
1
Normalized Force
0.5
Cases 1&2
0 Element 3 Collapse time for brittle
(4 cases overlap element 3 (Cases 3&4) Cases 3&4
-0.5 up to 8.875 sec)
-1
-1.5
0 5 10 15 20 25
Time [sec]
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 617
-0.5 -0.5
Time = 8.875 Time = 11.260
Normalized Displacement
Normalized Displacement
-1 -1
-1.5 -1.5
Cases1&2
Cases 1&2
Cases3&4
Cases 3&4 Cases1&2
Cases 1&2
-2 -2
Case33
Case
Case44
Case
-2.5 -2.5
0 5 10 15 20 25 0 5 10 15 20 25
(a) Time [sec] (b) Time [sec]
Figure 7. Snapshots of benchmark structural model deformation at Time= 14.5 s: (a) cases 1
and 2; (b) case 3 and (c) case 4.
modes, if any, are illustrated in Figure 7 (no displacement magnification) with deformed beam
elements simplified by straight lines connecting the end-nodes. Figure 8 compares the maximum
ductility demands imposed on each truss element during simulations and lends insight to the
sensitivity of the outcome and extent of system collapse to its element properties. There is a
significant contrast between the small difference in ductility demands on members 3 and 10 and
the large difference in simulated progressive collapse outcome for each case. This highlights the
sensitivity of the simulation procedure, because of its inherent path dependency, to uncertainty
in the structural model parameters. Hence, there is an obvious need to conduct simulation-based
progressive collapse assessment within a probabilistic framework.
Three element removal criteria are defined for two different modes of failure in seismically
deficient and retrofitted RC beam–columns and for truss members. These criteria are implemented
in OpenSees for force-based and displacement-based distributed plasticity fiber elements, lumped
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
618 M. TALAAT AND K. M. MOSALAM
5
Cases Case Case
4.5 1&2 3 4
4
3.5
3
2.5
2
Collapsed Collapsed
1.5
1
0.5
0
1 3 10 11
Element
plasticity beam–column elements with fiber-discretized plastic hinges, shear-axial coupled spring
elements [19] and truss elements.
where A and I refer to transformed area and moment of inertia, and h to the distance between the
fiber’s center and the uncracked section centroid. Areas and inertias are transformed using the ratio
between the steel and concrete fibers’ initial stiffness moduli. Iconf is an indicator for the loss of
confinement whose value is 0.0 if the confining medium (e.g. transverse steel ties or FRP jacket)
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 619
fracture is detected and 1.0 otherwise. The proposed damage indices are hence suitable for use
with deficient and FRP-retrofitted RC columns. When D A reaches a threshold value, the associated
column cross-section is considered to have lost its axial load capacity and the element removal
algorithm is invoked on the associated element. The threshold value is defined as D A = 1 in this
paper. A future extension may explicitly reflect the axial load level in the column. An analogous
approach can be defined for D M , but is not of interest in this paper.
The structural system is selected to benefit from existing experimental data from a similar physical
model subjected to shake-table ground motion records of increasing intensity at the University of
California, Berkeley, Figure 9(a). The analysis of results is transparent by considering a one-story
system and avoiding dynamic effects of higher vibration modes. The physical model represents
a substructure of the first story of a five-story prototype building. It consists of three seismically
detailed moment-resisting RC frames identical in geometry and reinforcement, placed parallel to
each other and connected by an RC slab and transverse RC beams. The middle RC frame is
infilled with a URM wall. Details of the experiment can be found in [28]. A corresponding FE
model, Figure 9(b), is calibrated and then modified to introduce seismically deficient reinforcement
details in the RC columns. The objective of application ‘A’ is to investigate the effect of intra-event
variability, i.e. uncertainty in the characteristics of ground motion due to site location and building
orientation, on the results of progressive collapse simulations using 14 ground motion records from
a specific seismic event. An empirical probability measure is defined to construct fragility-like
curves for different collapse limit-states.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
620 M. TALAAT AND K. M. MOSALAM
16 B7 14 B5 12
B3 C6
B2 B1
C2
S1
C4
S2
B6 B4
15 11
13
N C5 C3 C1
(a) (b)
Figure 9. Shake-table test structure: (a) physical model in [28] and (b) computational model.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 621
f mo 133 143
Cover concrete fiber
213 214
Parabola Core concrete fiber
Straight line S2
Steel tie
f mu Bending axis
Figure 10. OpenSees FE mesh of the one-story test-bed structural system (infilled frame).
The element mesh of the URM-infilled frame C3-B2-C4 is illustrated in Figure 10. Nodes 3
and 4 are restrained from all translations and rotations. Nodes 30 and 40 coincide with nodes 3
and 4, respectively. The beam–column elements representing the column plastic hinge zones have a
length of 366 mm, based on L p = 0.077 L c +8.16 db , where L c is the length between the column’s
contra-flexure points and db is the longitudinal steel bar diameter [20], and two Gauss numerical
integration points. The cross-section is discretized using the confined concrete section and material
models from [21] with a grid of 10×10 fibers for the core concrete and 2 layers of fibers in each
direction for the concrete cover. The longitudinal steel is modeled using buckling-enabled steel bar
fibers [20] with bar mechanical stress–strain relationship according to [29]. The transverse rein-
forcement is modeled using a bilinear material model. A confinement efficiency factor of 0.75 is
estimated according to [30] for the square shape of the cross-section with four tie branches relative
to the circular shape. The beam–column elements in the middle parts of the columns have nine
Gauss numerical integration points and cross-section discretization identical to that of the plastic
hinge zones. The beam–column elements used to represent the longitudinal beam plastic hinge
zones have a length of 303 mm (using the formula for L p above), and two Gauss numerical integra-
tion points. Their cross-sections are discretized using a grid of 5×10 fibers for the core concrete (10
layers parallel to the bending axis) and one layer in each direction for the concrete cover. According
to recommendations in [31], a 1030-mm effective width of the RC slab is considered in resisting
compressive stresses and meshed using one layer of fibers. Experimental observations showed no
damage taking place in the beams or the middle parts of the RC columns, and hence the steel
bars in these cross-sections are modeled with bar buckling disabled for computational efficiency,
and a constant confining stress computed assuming yielding of the transverse ties is preset and
applied to the core concrete fibers. The URM wall is modeled using the two truss elements S1 and
S2, whose axial force-deformation behavior resembles a compression-only concrete-like material,
Figure 10. The parameters for this stress–strain curve were calibrated in [28] from experimen-
tally measured material properties and are reported as εmo = 0.0028, εmu = 0.0042, f mo = 17 MPa,
and f mu = 2 MPa, corresponding to a strut area of 223 mm2 . Elements 130 and 140 representing
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
622 M. TALAAT AND K. M. MOSALAM
shear-axial interaction of the infilled frame columns are constructed according to [19]. The meshing
of the bare frame members is similar, with the exclusion of the wall struts and the shear-axial
springs.
The transverse beams B4–B7 and the one-way RC slabs were observed to exhibit only minor
flexural cracking during the shake-table tests. Thus, their flexural stiffness is modeled using elastic
beam–column elements with a 1030-mm effective width of the slab contributing to the beam
cross-section and 0.7 times the uncracked concrete Young’s modulus of elasticity [31]. The in-
plane shear stiffness of the RC slab is observed to remain approximately constant throughout the
shake-table tests, and estimated as 406.0 kN/mm [28]. It is modeled using equivalent elastic truss
elements (Figure 9(b)) of stiffness 243.15 kN/mm each, assuming linear geometric transformation
in the plane of the RC slab.
The post-tensioning loads on the columns are applied as external nodal forces at nodes 11–16,
which would result in P- effects more similar to realistic structures. Masses and distributed
gravity loads are lumped at the nodes. The infilled frame is assumed to carry twice as much gravity
load as the bare frames. Numerical time integration uses the Newmark -method with constant
acceleration. Based on snap back test results in [28], a stiffness-proportional viscous damping
ratio = 0.045 is estimated at the fundamental mode of vibration. The stiffness used to determine
the damping matrix during each analysis step is computed as the mean of the initial stiffness
and the tangent stiffness of the last converged analysis step, in order to avoid rapid numerical
reduction in damping caused by decreased and possibly momentary non-positive tangent stiffness
matrix due to structural damage. The FE model’s estimation of the fundamental vibration period is
0.121 s, compared with the experimentally recorded value of 0.135 s, and it reasonably reproduces
the force–displacement time histories of the test structure at levels of ground motion up to and
including collapse of the URM wall [20]. As an example of such reasonable match, Figure 11
shows the comparison of experimental and simulated lateral roof displacement at the middle frame
of the specimen for an input ground motion obtained from the Northridge (1994) earthquake
(Tarzana station in the 90◦ direction) and scaled to PGA = 0.61 g.
After calibration, the FE model is artificially weakened by introducing seismically deficient
details in individual RC columns where plastic hinges are expected to form. The transverse tie
spacing in all columns is increased to 150 mm, the maximum value allowed by the 1963 ACI
building code [32]. This increases the potential for bar buckling and shear failure, especially in
columns C3 and C4 due to the formation of short columns following URM wall cracking and
Roof Horizontal Displacement [mm]
8
Shake-table experiment
OpenSees simulation
4
-4
-8
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 11. Simulated and experimental time-history results of shake-table test structure.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 623
partial collapse. Lap splices are inserted at the column-footing level of columns C5 and C6 in order
to observe the resulting difference in response and avoid simultaneous, non-progressive collapse
of the otherwise symmetric bare frames. The length of the lap splice is 20 times the bar diameter
[32] which is inadequate by current standards [31]. The diagonal strut elements are assumed to
collapse and are removed upon softening to 0.75 f mo , representing brittle URM wall collapse.
Preliminary analysis of the modified FE model identified columns C3 and C4 as likely to collapse
first, imposing large deformation demands on beams B4-B7, which redistribute their loads to the
corner columns. Excessive vertical displacements lead to crushing of concrete in compression and
loss of bending moment capacity, but the continuity of the longitudinal steel bars over the beam
length forms catenaries supporting the vertical load through the action of inclined axial tension
until the longitudinal steel fails due to fracture or pull out. An independent FE simulation was
conducted for the setup in the inset of Figure 12, showing a side view of the modified FE model
with C3 removed and a displacement-controlled downward motion imposed at node 13. Beams B4
and B6 are modeled using a total of 12 displacement-based beam–column elements with five Gauss
integration points each and cross-section discretization similar to that in Figure 10. The reinforcing
steel fracture strain (pull out failure is excluded in the present study) is assumed to be 0.13. The
simulated force-displacement relationship, Figure 12, is idealized by a tri-linear envelope defined
by points (0, 0), (20, 245), (45, 150), and (490, 450); and is used to construct four nonlinear
springs connecting nodes 13 and 14 to the interior ends (at the infilled frame) of beams B4–B7.
500
Ultimate displacement leading to
450 Idealized fracture of longitudinal steel bars
and spring element removal
OpenSees simulation
Vertical Reaction Force at Node 1 [kN]
400
350
300
250
200
B4 B6
11 15
150 13
100 C1 C5
50
1 5
0
0 50 100 150 200 250 300 350 400 450 500
Downward Displacement at Node 13 [mm]
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
624 M. TALAAT AND K. M. MOSALAM
34.5
Selected site
location
Latitude
Recording station
Epicenter
Surface projection of
fault rupture plane
34
-119 -118.5 -118
Longitude
Figure 13. Northridge (1994) earthquake locality [10] and building site.
Table VI. Summary of ground motion recording stations used in application ‘A’.
The time-history profile and intensity of the individual records show considerable differences,
representing a measure of the irreducible aleatory uncertainty due to the site location and orientation
of the building. The modified FE model is subjected to the 14 ground motion records, each scaled
to represent seven seismic hazard levels. The simulation results are used to construct fragility-like
curves for four collapse limit-states defined below. Each curve represents the conditional proba-
bility of reaching a limit-state given the specified hazard level. An empirical probability is defined
as the number of simulations in which a limit-state is reached divided by the total number of
simulations per hazard level, i.e. 14.
The directions of the horizontal ground motions recorded at each recording station are not
coincidental, and hence referred to as ‘Component 1’ and ‘Component 2’ in Table VII. Significant
differences in characteristics can be observed, with no discernible independent effects of site
location and measurement orientation. Component 1 is the direction per record in which the elastic
spectral acceleration response is higher for the considered structural system. An average value of
the 5%-damped elastic spectral acceleration Sa (T̄1 , 5%) is computed for the vibration period range
0.121–0.192 s (T̄1 ) corresponding to the FE model’s fundamental periods with the diagonal struts
included and removed, respectively. This spectral acceleration is adopted as a seismic intensity
measure. The scaling of individual components is based on the USGS uniform hazard curves [34]
via matching the system-specific Sa (T̄1 , 5%) to the hazard-specific values in Table VIII, linearly
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 625
Table VII. Unscaled seismic intensity measures of records used in application ‘A’.
Component 1 Component 2
Record PGA (g) Sa (T̄1 , 5%) (g) Record PGA (g) Sa (T̄1 , 5%) (g)
JEN292 1.023870 1.8011 JEN022 0.570622 0.9417
NWH090 0.583030 1.3155 NWH360 0.589780 1.0988
WPI046 0.454930 0.5943 WPI316 0.325431 0.4173
RRS228 0.825195 1.2973 RSS318 0.486504 1.0854
SCS142 0.897237 1.2291 SCS052 0.612467 0.8901
SCE018 0.828279 1.3825 SCE288 0.493035 1.2162
SYL360 0.843306 1.2507 SYL090 0.604490 0.8726
Table VIII. Hazard-specific seismic intensity measures at selected building site location.
Hazard Level 1/50 2/50 3/50 5/50 10/50 20/50 50/50
RP (years) 4975 2475 1642 975 475 224 72
PGA (g) 1.8407 1.5887 1.4344 1.2339 0.9676 0.6790 0.2706
Sa (0.1, 5%) (g) 3.6857 3.1366 2.7767 2.3784 1.8124 1.2484 0.5058
Sa (0.2, 5%) (g) 4.6824 3.8789 3.4726 2.9261 2.2447 1.5323 0.6389
Sa (T̄1 , 5%) (g) 4.1841 3.5078 3.1247 2.6523 2.0286 1.3904 0.5724
averaged between periods of 0.1 and 0.2 s (RP refers to return period and A/50 refers to A%
probability of being exceeded in 50 years). This procedure could have been more complex if the
mechanism of collapse were sensitive to the values of the scaling factors being computed. In the
present application, however, collapse always initiated in the URM walls due to its significantly
higher lateral stiffness. Four collapse limit-states are investigated:
(1) Partial URM wall collapse: removal of at least one diagonal strut.
(2) Complete URM wall collapse: removal of both diagonal struts.
(3) Partial structural system collapse: removal of at least two RC columns.
(4) Complete structural system collapse: removal of all six RC columns.
The conditional probability curves are shown in Figure 14 where the probabilities of the URM wall
collapse and the partial RC frame collapse are non-zero even at the lowest considered hazard level
of 50/50. The interpretation of these results deserves attention. The structural system is subjected
to ground motions recorded in similar, neighboring sites during the same earthquake event, which
nevertheless exhibited large variation in time-history profiles and response spectra. This variability
is reduced by scaling to the same elastic spectral acceleration at each hazard level. However, for
each hazard level between 20/50 and 2/50, the probability of reaching all limit-states is larger than
zero and less than unity, i.e. a simulation using at least one-scaled component reaches all limit-states
while using another similarly scaled component from the selected 14 records reaches none. This
emphasizes the paramount importance of fundamentally addressing intra-event variability within
the framework of performance-based earthquake engineering to develop consistent and unified
selection and scaling of ground motion records for simulating structural progressive collapse.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
626 M. TALAAT AND K. M. MOSALAM
1
Limit-state 1
0.9 Limit-state 2
Limit-state 3
Limit-state 4
0.8
Probability of Collapse
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
50/50 20/50 10/50 5/50 3/50 2/50 1/50
Hazard Level
The structural system is selected to represent complex multi-story behavior, where the collapse of
stiff URM walls may lead to the formation of a soft-story mechanism. The objectives of application
‘B’ are to investigate the collapse pattern and its sensitivity to uncertainties in selected system
parameters and in seismic hazard.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 627
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
628 M. TALAAT AND K. M. MOSALAM
acceleration of the FE model is computed as Sa (T̄1 , 5%) = 1.038 g, Figure 16. The target spectral
acceleration value, corresponding to the reference hazard level (2/50) at the selected building site,
is computed for the average first-mode period as 2.493 g [34].
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 629
where fˆ¯mo , fˆmo(0.5) , and ˆ are the estimated mean, median, and COV. For the viscous damping
ratio, mean, maximum and minimum values of 0.05, 0.10 and 0.02 are commonly assumed in
practice. For a 0.997 confidence interval and lognormal distribution, this corresponds to a median
value of 0.5 = 0.045 [28] and COV of 0.4. The reference seismic hazard level is set to 2/50 [36].
The higher and lower hazard levels of 1/50 and 3/50 are investigated. The values of the corre-
sponding elastic spectral accelerations are listed along with the uncertain system parameter ranges in
Table IX.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
630 M. TALAAT AND K. M. MOSALAM
14 14
Vertical Co-ordinate [m] 12 12
8 8
6 6
4 4
2 2
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(a) Horizontal Co-ordinate [m] (b) Horizontal Co-ordinate [m]
14 14
12 12
Vertical Co-ordinate [m]
Vertical Co-ordinate [m]
10 10
8 8
6 6
4 4
2 2
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
(c) Horizontal Co-ordinate [m] (d) Horizontal Co-ordinate [m]
Figure 17. Collapse progression in reference five-story structural system: (a) time = 7.7 s
(removal of C21); (b) time = 7.9 s (removal of C11); (c) time = 8.2 s (collapse mechanism);
and (d) time = 8.3 s (free-fall).
story, and overall drifts prior to collapse are 0.037, 0.013, and 0.017, respectively, with mode and
sequence of collapse given in Table X.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 631
Figure 18. Simulated drift ratio time-histories of reference five-story structural system.
direct correlation with the induced change in uncertain system parameters. Instead, the maximum
drift is controlled by local variables such as the axial load and the corresponding deformation
ductility of RC columns. Accordingly, Tornado diagram results of story drifts are found less
valuable by comparison.
The plot of the swings resulting from the deterministic sensitivity analysis in ascending order is
the graphical output of a Tornado diagram analysis. In the Tornado diagram shown in Figure 19,
the largest swing is associated with the earthquake intensity. Increasing the hazard level to 1/50
results in a stronger earthquake and earlier collapse at 6.92 s, while decreasing it to 3/50 results in
no collapse other than the first story URM wall. Next in importance is the URM wall stiffness. The
higher stiffness value results in earlier collapse at 7.469 s. The lower stiffness results in lower force
redistribution to the neighboring columns. In addition, the low-stiffness URM walls exhibit damage
in the upper stories, which reduces the subsequent localization of damage and story drift demand
in any one story and enables the structural system to survive the earthquake without gravity load
collapse in any RC column. Next in importance is the live load on the structure. Full occupancy
results in higher gravity-induced axial loads on the RC columns and the URM walls. This leads
to the collapse of at least one diagonal strut in each of the first four stories and to slightly earlier
collapse at 7.681 s. On the other hand, no occupancy results in lower gravity-induced axial loads.
As a result, the structure survives the first interval of high-amplitude shaking but collapses during
the second high-amplitude shaking interval at 16.37 s. Finally, the least important parameter is the
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
632 M. TALAAT AND K. M. MOSALAM
No incipient collapse
Reference value
Sa (T1 5,%)
Uncertain Parameter
Emi
LL
6 8 10 12 14 16 18 20 22 24
Time at Incipient Collapse [sec]
Figure 19. Tornado diagram for simulated sensitivity of time at incipient collapse.
viscous damping ratio of the undamaged structure, unless the increase in frictional damping due
to damage accumulation is explicitly modeled, which is not the case in the present study.
The swings for the top two uncertain parameters in Figure 19 are close in magnitude. However,
for the lower stiffness URM wall, the survival of the structural system comes at the expense of more
extensive damage propagation into the upper stories of the structure and a larger maximum first
story drift. Moreover, the minimum URM wall stiffness corresponds to a 0.997 confidence interval,
while the minimum ground motion intensity corresponds to decreasing the reference hazard level
by only one percentage point of being exceeded in 50 years. Thus, the most important uncertain
parameter in deciding the possibility of progressive collapse is the seismic intensity. However,
the most important system parameter when subjected to a given ground motion is the URM wall
stiffness.
CONCLUSIONS
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
MODELING PROGRESSIVE COLLAPSE 633
outcome, followed by URM wall stiffness for a given ground motion. Finally, uncertainties in live
load (occupancy level) and viscous damping ratio (excluding frictional damping due to damage)
have moderate and minor influences, respectively.
ACKNOWLEDGEMENTS
This study was supported by the Earthquake Engineering Research Centers Program of the NSF under
Award No. EEC-9701568 to PEER at University of California, Berkeley. Financial support from the
research sponsor is gratefully acknowledged. Opinions and findings presented are those of the authors
and do not reflect views of the sponsor.
REFERENCES
1. Virdi KS, Beshara FB. Numerical simulation of building decommission as a progressive collapse process. Third
International Conference on Decommissioning and Demolition. UMIST: U.K., 1992.
2. Gilmour JR, Virdi KS. Numerical modeling of the progressive collapse of framed structures as a result of impact
or explosion. Second International PhD Symposium in Civil Engineering, Budapest, 1998.
3. Krauthammer T, Hall RL, Woodson SC, Baylot JT, Hayes JR, Sohn Y. Development of progressive collapse analysis
procedure and condition assessment for structures. National Workshop—Prevention of Progressive Collapse,
Rosemont, IL, 2002.
4. Houghton DL, Karns JE. Post 9-11 multi-hazard mitigation in steel frame structures as a function of connection
geometry. Seventy-first International Convention of SEAOC, Santa Barbara, CA, 2002.
5. Crawford JE. Retrofit methods to mitigate progressive collapse. Technical Report TR-02-7, Karagozian and Case,
Glendale, CA, 2002.
6. Starossek U. Progressive collapse study of a multi-span bridge. Structural Engineering International, IABSE
1999; 16(2):121–125.
7. Ellingwood BR. Mitigating risk from abnormal loads and progressive collapse. Journal of Performance of
Constructed Facilities 2006; 20(4):315–323.
8. Kim Y, Kabeyasawa T. Dynamic test and analysis of an eccentric reinforced concrete frame to collapse. Thirteenth
World Conference on Earthquake Engineering, Vancouver, BC, Canada, 2004.
9. Wu C-L, Loh C-H, Yang Y-S. Shake table tests on gravity load collapse of low-ductility RC frames under near-
fault earthquake excitation. Proceedings, 1st International Conference on Advances in Experimental Structural
Engineering, Nagoya, Japan, Itoh Y, Aoki T (eds), 2005; 725–732.
10. Ghannoum W. Experimental and analytical dynamic collapse study of a reinforced concrete frame with light
transverse reinforcements. Ph.D. Dissertation, University of California, Berkeley, 2007.
11. Grierson DE, Xu L, Liu Y. Progressive-failure analysis of buildings subjected to abnormal loading. Computer-aided
Civil and Infrastructure Engineering 2005; 20(3):155–171.
12. Kaewkulchai G, Williamson EB. Modeling the impact of failed members for progressive collapse analysis of
frame structures. Journal of Performance of Constructed Facilities 2006; 20(4):375–383.
13. Kaewkulchai G, Williamson EB. Beam element formulation and solution procedure for dynamic progressive
collapse analysis. Computers and Structures 2004; 82(7–8):639–651.
14. Pretlove AJ, Ramsden M, Atkins AG. Dynamic effects in progressive failure of structures. International Journal
of Impact Engineering 1991; 11(4):539–546.
15. Ramsden M. Dynamic effects in the progressive failure of lattice structures. Ph.D. Dissertation, University of
Reading, U.K., 1987.
16. Powell G. Progressive collapse: case studies using nonlinear analysis. Structures Congress and Forensic
Engineering Symposium, New York, 2005; 2185–2198.
17. Haselton CB, Deierlein GG. Benchmarking seismic performance of reinforced concrete frame buildings. Structures
Congress and Forensic Engineering Symposium, New York, 2005; 1891–1902.
18. Zareian F, Krawinkler H. Assessment of probability of collapse and design for collapse safety. Earthquake
Engineering and Structural Dynamics 2007; 36(16):1901–1914.
19. Elwood KJ, Moehle JP. Axial capacity model for shear-damaged columns. ACI Structural Journal 2005;
102(4):578–587.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe
634 M. TALAAT AND K. M. MOSALAM
20. Talaat M, Mosalam KM. Computational modeling of progressive collapse in reinforced concrete frame structures.
PEER Technical Report 2007/10, 2008.
21. Mosalam KM, Talaat M, Binici B. A computational model for reinforced concrete members confined with fiber
reinforced polymer lamina: implementation and experimental validation. Composites Part B: Engineering 2007;
38(5–6):598–613.
22. Mazzoni S, McKenna F, Scott MH, Fenves G. Opensees user’s manual, 2004. Available from:
www.opensees.berkeley.edu.
23. Zineddin M, Krauthammer T. Dynamic response and behavior of reinforced concrete slabs under impact loading.
International Journal of Impact Engineering 2007; 34(9):1517–1534.
24. Schonberg WP, Keer LM, Woo TK. Low velocity impact of transversely isotropic beams and plates. International
Journal of Solids and Structures 1987; 23(7):871–896.
25. The MathWorks, Inc. Getting Started with Matlab 7 (Release 2007b), Natick, MA, 2007.
26. Crisfield MA, Moita GF. Unified co-rotational framework for solids, shells and beams. International Journal of
Solids and Structures 1996; 33(20–22):2969–2992.
27. Brown J, Kunnath SK. Low-cycle fatigue failure of reinforcing steel bars. ACI Materials Journal 2004; 101(6):
457–466.
28. Hashemi A, Mosalam KM. Seismic evaluation of reinforced concrete buildings including effects of masonry
infill walls. PEER Technical Report 2007/100, 2007.
29. Menegotto M, Pinto P. Method of analysis for cyclically loaded reinforced concrete plane frames including
changes in geometry and non-elastic behavior of elements under combined normal force and bending. Proceedings,
Symposium on Resistance and Ultimate Deformability of Structures Acted on by Well Defined Repeated Loads,
International Association for Bridge and Structural Engineering (IABSE), Lisbon, 1973; 15–22.
30. Mander JB, Priestley MJN, Park R. Theoretical stress–strain model for confined concrete. Journal of Structural
Engineering 1988; 114(8):1804–1826.
31. ACI. Standard building code requirements for reinforced concrete. ACI 318-02, Detroit, MI, 2002.
32. ACI. Standard Building Code Requirements for Reinforced Concrete. ACI 318-63, Detroit, MI, 1963.
33. ICC. International Building Code. International Code Council, Falls Church, VA, 2003.
34. USGS Probabilistic Hazard Curves, 2007. Available from: http://earthquake.usgs.gov/research/hazmaps/design/.
35. Lee T-H, Mosalam KM. Seismic demand sensitivity of reinforced concrete shear-wall building using FOSM
method. Earthquake Engineering and Structural Dynamics 2005; 34(14):1719–1736.
36. NEHRP. Recommended Provisions for Seismic Regulations for New Buildings and Other Structures. Building
Seismic Safety Council, Washington, DC, 2001.
Copyright q 2009 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2009; 38:609–634
DOI: 10.1002/eqe