You are on page 1of 11

Engineering Structures 126 (2016) 446456

Contents lists available at ScienceDirect

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

Finite element modeling and capacity analysis of post-tensioned steel


frames against progressive collapse
Akbar Pirmoz, Min (Max) Liu
Department of Civil Engineering, The Catholic University of America, 620 Michigan Ave., N.E., Washington, DC 20064, USA

a r t i c l e i n f o a b s t r a c t

Article history: Post-tensioned (PT) steel frames, in which beams are connected to columns through high-strength PT
Received 19 April 2016 strands, have been successfully developed in the past decade as a novel earthquake-resistant structural
Revised 13 June 2016 system. Compared to conventional steel moment frames, a PT steel frame demonstrates superior seismic
Accepted 3 August 2016
performance, notably the minimum damages in main structural components and the self-centering capa-
bility under a design basis earthquake. Despite the abundant research on the seismic behavior of PT steel
frames, there is an apparent lack of study on their load-redistribution behavior upon the notional removal
Keywords:
of critical load-bearing columns, a commonly used threat-independent local structural damage scenario
Progressive collapse
Steel frame
that potentially triggers the progressive collapse of the column-removed frame. This paper presents a
Post-tensioned connection first-of-its-kind numerical investigation on the unique structural behavior of PT steel connections and
Finite element frames in redistributing the unbalanced gravity loads due to column removal. High-fidelity finite element
Column removal structural models are constructed and validated using the available experimental data in the literature.
Load redistribution The capacity of PT steel frames subjected to a gradually increasing vertical displacement along the
Structural capacity removed column line is systematically studied. It is found that, besides the resistance of energy-
Arching action dissipating elements, beam arching action and strand catenary action are the major sources of structural
Catenary action
capacity of a PT steel frame against progressive collapse. The corresponding failure modes are identified
Failure mode
and the design implications are suggested.
2016 Elsevier Ltd. All rights reserved.

1. Introduction possesses a similar level of rotational strength, stiffness, and duc-


tility (realized via energy-dissipating elements), while the costly
The recently emerging post-tensioned (PT) steel frames are repair of cracks in welded structural elements can be favorably
potentially a novel alternative to conventional steel moment- minimized or even eliminated. In addition to experimental investi-
resisting frames [13]. As schematically shown in Fig. 1, at the gation, both high-fidelity finite element (FE) models [7] and
beam-to-column connection of a PT steel frame, PT strands are reduced-order macroelement models [8] have been developed to
installed along the beam web and anchored on the column flanges, study the seismic response of PT steel frames. In particular, using
and the beam is vertically supported by beam-column friction and/ the ANSYS software [9], a parametric FE model can be conveniently
or passive energy-dissipating elements (e.g., angles). Seismic built to generate through computer simulation a vast amount of
design of PT steel frames typically requires that under a design information on the deformation and internal forces of all compo-
basis earthquake, beams, columns, and PT strands provide the pri- nents within a PT steel structure, thereby avoiding the physical
mary sources of strength and stiffness and essentially remain elas- installation of numerous sensors or any restriction owing to vari-
tic, while the inelasticity and seismic damages be concentrated ous experimental conditions, as often encountered in actual struc-
within the replaceable energy-dissipating elements [4]. Thus, after tural tests.
the design basis earthquake, a PT steel frame is expected to regain Parallel to the fruitful research on seismic performance of PT
its original plumb position (i.e., self-center) without incurring steel frames [1012], investigation of the progressive collapse (also
residual deformation in its major structural components. Experi- known as disproportionate collapse) behavior and capacity of such
mental studies [1,2,5,6] have demonstrated that, compared to con- a novel structural system is also very much needed. Progressive
ventional welded moment connections, the PT connection collapse refers to a catastrophic chain reaction of structural fail-
ures, propagating from the initial damages within a local portion
Corresponding author. of a building to the widespread global damages or even total
E-mail address: lium@cua.edu (M.M. Liu). collapse [1315]. Because the final damage severity can be highly

http://dx.doi.org/10.1016/j.engstruct.2016.08.005
0141-0296/ 2016 Elsevier Ltd. All rights reserved.
A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456 447

could lead to gravity-induced progressive collapse. Toward this


goal, we develop high-fidelity FE models for both PT steel beam-
to-column connections and PT steel frames that contain such con-
nections, and validate these FE models using relevant benchmark
experimental data available in the literature. We then use these
models to systematically investigate the capacity of a column-
removed PT steel frame subject to statically increasing displace-
ment along the removed column line, elucidate the role of different
structural components in facilitating the load redistribution upon
column removal, and identify the relevant failure modes and high-
light the implications for progressive collapse design.

2. Benchmark PT structures

Fig. 1. Schematic of a PT steel beam-to-column connection.


In order to validate and/or calibrate the present nonlinear FE
structural models, we consider two sets of benchmark experimen-
disproportionate to its triggering event, the associated economic tal data available in the literature. The first set of data was gener-
and societal consequences are often enormous [16]. Therefore, a ated from the testing of a series of PT beam-to-column connections
structural system must be adequately designed/retrofitted to mit- under quasi-static cyclic lateral loads [5]. In the present study, we
igate the risk of such a devastating event. Although the existing use the test data associated with a particular connection specimen
research has offered a comprehensive understanding of the seismic designated as PC2A (Fig. 2). This connection included two beams
behavior of PT steel frames, their structural characteristics against (with a total length of 6096 mm or 240 in.) with an identical PT
progressive collapse can be quite different and are hence not read- connection on each side of an H-shaped column (with a total
ily inferred from the available seismic findings. This is because, as length of 3658 mm or 144 in.) and was post-tensioned by four
clearly revealed by previous studies on other types of structural strands along each side of the beam web. Each seven-wire, 0.6-
systems [1721], although structures that are properly designed in. (15 mm)-diameter strand was initially pre-tensioned to one-
against lateral seismic loads favorably possess some reserve capac- third of its ultimate stress of 1864 MPa (270 ksi). Although not
ity beyond what is needed to resist the normal gravity loads, such a directly produced from a progressive collapse test, such experi-
reserve capacity can be far from adequate to safely redistribute the mental data are used here to confirm the general effectiveness of
unbalanced gravity loads upon the notional removal of critical the FE model for simulating the nonlinear behavior of a PT connec-
load-bearing columns from the original intact building, a com- tion. Table 1 lists the actual (measured) yield stresses of individual
monly used scenario that simulates the sudden column loss caused steel components in the tested connection.
by extreme loading and is often implemented with the widely The second set of data was generated directly from a progres-
adopted threat-independent alternate path method [22,23]. sive collapse experiment on a reduced-scale three-story-two-bay
Indeed, the desirable behavior and corresponding design philos- PT steel frame [24] (Fig. 3), which was fixed at the base and seismi-
ophy of a PT steel frame under unbalanced gravity loads are very cally designed as a lateral load-resisting frame within a prototype
different from those under seismic loads. For seismic design, the building. The tested frame had a uniform bay width of 1219 mm
PT frame is expected to return to its initial position after a design (49.75 in.) and a height of 1264 mm (49.75 in.) for the first story
basis earthquake so that the building can resume its normal func- and 1289 mm (50.75 in.) for both second and third stories. At the
tionality. To this end, the main structural components (i.e., strands, mid-height of each column, restraining steel plates were mounted
beams, and columns) of a PT steel frame must remain elastic while as a lateral support to prevent the out-of-plane instability. The
the energy-dissipating elements (e.g., top-seat angles) function as a frame was tested by applying a monotonically increasing displace-
fuse to bear all structural damages. For gravity-induced progres- ment along the middle column line, where the first-story middle
sive collapse design, however, it is neither necessary nor econom- column was intentionally unsupported to simulate its removal.
ical to make the major structural components of a PT steel frame The beams and columns were W-shape steel sections made of
stay elastic under postulated column removal scenarios, consider- ASTM A572 Grade 50 steel. At each beam-to-column connection,
ing the extremely rare, highly unpredictable nature of a triggering both beam and column flanges were reinforced with steel plates
local damage. Instead, a controlled level of plastic deformation and
local failure of main structural components is acceptable, as long as
the path for damage propagation is blocked and thus the overall
system integrity is maintained to avoid the total frame collapse.
So far, research on the performance of PT steel frames in redis-
tributing the unbalanced gravity loads upon column removal is
very scarce [2426]. In particular, numerical modeling of PT steel
frames specifically for progressive collapse analysis and corre-
sponding investigation of their structural capacity against progres-
sive collapse do not exist. As a result, the capability of a seismically
sound PT steel frame to arrest the progressive collapse risk is
essentially unknown. Lack of such fundamental knowledge can
severely hinder the adoption of PT steel frames in a building sys-
tem that could experience the loss of critical load-bearing columns
during its service life. In response to this concern, the study
reported in this paper aims to understand the behavior and quan-
tify the capacity of PT steel frames under column removal that Fig. 2. Experimental setup of a PT steel connection under a lateral load (adapted
from Ricles et al. [5]).
448 A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456

Table 1 Language capability [9] to create parametric FE models for PT steel


Material properties of a tested PT steel connection. connections and frames, and then validate/calibrate these models
Structural Beam Beam Reinforcing Shim Strand against the afore-mentioned experimental data. The specific mod-
component flange web plate plate eling strategies are described as follows.
Actual yield stress 230 266 843 843 1305
[MPa] 3.1. Modeling of the PT connection

The three-dimensional eight-node solid element SOLID185 pro-


vided in the ANSYS element library is used to mesh most structural
components of the tested PT steel connection (Fig. 2). The full inte-
gration option is selected for calculating the element stiffness
matrix. Note that the compressive force at the beam-to-column
interface gives rise to a frictional force that resists the shear in
the PT connection. To model such surface interaction, we use the
contact elements CONTA174 and TARGE170, which are paired to
prevent the penetration of the nodes from one surface into the
other [9], to couple the end contact surfaces of both the beam
and its flange reinforcing plates with that of the flange reinforcing
plates of the column. A friction coefficient of 0.35 is assigned for
such contact surfaces, a normal penalty stiffness factor of 1.0 with
a penetration tolerance factor of 0.05 is chosen when defining the
pair of contact elements, and the close gap option is used to enable
the automatic contact adjustment. In order to prevent the column
stress concentration under a point load and ensure a simple con-
nection condition, we place fictitious rigid elements (i.e., cap plates
with a significantly high, say ten times as high, modulus of elastic-
ity compared to that of steel) at both ends of the column and beam.
We model the PT strands (assuming all wires within a strand
behave identically) by using the one-dimensional two-node ele-
ment LINK180, each node having three degrees of translational
freedom. This element is able to consider the large deformation
and large strain effects, and thus it can be conveniently used to
simulate the necking (i.e., shrinking in the cross section) of a PT
strand. To pre-tension the strands in the FE model, we apply a fic-
Fig. 3. Experimental setup of a PT steel frame under a vertical load (adapted from titious negative thermal gradient to the whole structure but assign
Tsitos et al. [24]). a fictitious thermal expansion coefficient only to the strands. As the
strands tend to shrink under the internal thermal load, they are
restrained from both ends and thus pre-tensioned. Values of the
made of ASTM A514 steel; four dog-bone-shaped energy- two fictitious parameters are adjusted by trial and error to obtain
dissipating rods, which had a cross-sectional area of 50 mm2 at the target initial PT forces.
the second floor and 40 mm2 at the third floor and roof, were In order to trigger the potential beam flange local buckling, the
installed and allowed to yield in tension and compression without initial imperfections in beams are modeled by applying a nominal
buckling. Two PT strands were installed for each beam, and the ini- set of transverse perturbation forces, equivalent to a line load of
tial PT forces per strand were approximately 88.9 kN, 66.7 kN, and 0.5 kN/m, at the flange edge nodes after the strands have been
55.6 kN at the second-floor, third-floor, and roof levels, respec- pre-tensioned and thus the load-carrying capacity of the connec-
tively. The PT strands were of DYWIDAG type and each had a tion has been established. We account for the potential inelastic
cross-sectional area of 140 mm2 at the second and third floors deformation of different structural components by defining their
and 99 mm2 at the roof. The nominal yield stresses of different individual material nonlinearities with appropriate bilinear
components in the tested PT frame are listed in Table 2 based on stress-strain relationships the actual steel yield stresses for dif-
data in the literature [27], where more detailed information about ferent components [5] are used (Table 1), and the strain-hardening
the tested frame can be found. ratios are assumed to be 0.02 and 0.05 for strands and all other
steel components, respectively [7], along with the isotropic hard-
3. FE modeling ening rule and von Mises yield criterion. Note that the elastic mod-
ulus of steel is taken as 200 GPa.
High-fidelity FE modeling enables us to understand the detailed The simple supports at the column ends of the PT connection
structural responses and thus helps identify the potential failure are modeled by restraining the middle nodes of the rigid elements
modes, a salient advantage over the simplified structural models at the column ends in all translational directions. The roller sup-
[28]. In the present study, we exploit the ANSYS Parametric Design ports at the beam ends of the PT connection are modeled by
restraining the bottom row of the beam end nodes in both vertical
and out-of-plane directions (i.e., along y and x axes). The
Table 2
displacement-controlled loading is applied at the upper column
Material properties of a tested PT steel frame.
tip after pre-tensioning the strands and imposing the initial imper-
Structural component Beam & Reinforcing Rod Strand fections. Fig. 4 shows the FE model of the PT steel connection built
column plate
in ANSYS using a total of 7813 nodes and 4720 elements.
Nominal yield stress 345 690 490 1675 The following options and parameters are further considered
[MPa]
when carrying out the numerical procedure in ANSYS. The sparse
A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456 449

Fig. 4. The FE model of the PT connection.

solution method is selected to solve the system of equations for the


FE model, the structural force and displacement convergence crite-
ria are adopted with a tolerance level of 0.001, the minimum and
maximum displacement increments are 0.01 mm and 5 mm,
respectively, and the automatic time stepping and line search
options are activated.

3.2. Modeling of the PT frame


Fig. 5. The FE model of the three-story frame (symmetry considered).

Most modeling strategies for the PT connection are applicable


for the FE modeling of the tested three-story PT steel frame
4. Results and discussion
(Fig. 3). So only the unique aspects of the frame modeling are pre-
sented here. We only consider half of the frame by taking advan-
4.1. Validation of the FE model for the PT connection
tage of the symmetry about a plane passing through the
centerline of the middle column and being perpendicular to the
We compare in Fig. 6 the test data, which were associated with
frame plane. All nodes on this symmetry plane are restrained along
the loading portion of the first hysteretic cycle [5], and the simula-
the beam longitudinal direction. Note that another symmetry
tion results using the FE model (Fig. 4). Note that the vertical axis
about a plane passing through the middle of beam webs is not con-
represents the lateral force (denoted as H in the literature [5]) that
sidered in the FE model otherwise, the nodes within this plane
is applied at the upper column tip in the z direction (Fig. 4), and the
would be prevented from moving transversely and thus the poten-
horizontal axis represents the corresponding lateral displacement
tial beam web local buckling be eliminated. Besides, the reinforcing
in the same direction. It is clearly seen that the FE model
plates are treated as being fully integrated into the corresponding
steel members through the adjacent nodes in the FE model. As a
result, the effect of fillet welds that connect the plates to beams
and columns is ignored, a reasonable approximation considering 250
the minor impact of such welded joints on frame behavior. Each
energy-dissipating rod is modeled as a single two-node LINK180 FE results
element with a constant cross-sectional area associated with the 200
middle portion of the actual rod. Selected nodes at the middle por-
tion of each column are restrained against the out-of-plane dis-
Lateral Force [kN]

placement to simulate the existence of restraining steel plates


mounted on the tested frame. Consistent with the displacement- 150
controlled test, the FE model is loaded by statically increasing Test data [5]
the displacement along the middle column line. Fig. 5 presents
the FE model of the tested three-story frame using a total of 100
16,771 nodes and 10,606 elements in ANSYS.
In order to accurately study the capacity of a frame structure,
the realistic stress-strain relationships of material based on the
measured (actual) strength of coupon should best be used in the 50
FE modeling. Because such data were not reported for the tested
three-story PT frame [24], the expected steel yield stresses are cal-
culated based on their nominal values (Table 2) per AISC provisions 0
[29] and are used in the FE model. For example, the yield stress of 0 20 40 60 80 100 120
steel for beams and columns used in the FE model is taken equal to Lateral Displacement [mm]
an expected value of 375 MPa, as opposed to a nominal value of
345 MPa. Fig. 6. Comparison of test and FE results for the PT connection under a lateral load.
450 A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456

satisfactorily predicts the response of the tested connection in


terms of the initial elastic stiffness, decompression load, and
post-gap-opening stiffness of the PT connection. It appears that
the FE model slightly overestimates the response in the nonlinear
range, likely due to ignoring the inevitable uncertainties associated
with, for example, material properties, residual stresses, and/or
experimental conditions of the tested PT connection.

4.2. Study of a hypothetical one-story PT frame

Because the outer ends of beams in the tested connection


(Fig. 2) were not restrained in their beam longitudinal direction,
the beneficial beam arching action [21,30] (to be defined next)
and column flexural resistance, which typically exist in the
response of an actual PT steel frame, are unable to be achieved.
As a result, the FE model in Fig. 4 should not be directly used to
study the progressive collapse behavior of a PT steel connection.
To resolve this issue, we hypothetically create a one-story, two-
bay PT frame (Fig. 7) based on the PT connection (Fig. 2) and then
consider a middle column removal scenario. Taking advantage of
the structural symmetry, we only model half of the frame in ANSYS
leading to 9646 nodes and 5621 elements (Fig. 8). An incremen-
tally increasing vertical displacement is applied along the middle
column to investigate the vertical force capacity of the frame upon
the middle column loss.
Before discussing the progressive collapse behavior of the one-
Fig. 8. The FE model of the hypothetical one-story frame (symmetry considered).
story PT frame, we hereby introduce the concept of beam arching
action caused by the rotation of an axially restrained beam with
respect to its original horizontal position. Illustrated in Fig. 9, as
the right-hand-side column is pushed downward by D, the beam
rotates by h. Accordingly, the horizontally projected length of the
beam tends to increase. However, the restraining columns at both
ends of the beam, along with the PT strands, prevent such a length
increase, leading to equal and opposite horizontal forces F that act
at the beam ends and are vertically offset by e. This force couple is
equilibrated by the vertical forces V that also act at the beam ends.
Mathematically,
V Fe=L 1

e d cosh  D  d  D 2
where L and d are the original clear length and depth of the beam,
respectively. Accordingly, the vertical force Varch acting on the right-
hand-side column by the beam is upward and equal in value to the
afore-mentioned V exerted by the column. This vertical beam for-
ce Varch contributes to the frame capacity against the applied ver-
tical force G along the removed column line, as long as the line of
application of the horizontal force F at the right end of the beam Fig. 9. Illustration of the beam arching action in a PT steel frame.
is above that of the horizontal force at the left end of the beam

Fig. 7. Schematic of a hypothetical one-story PT steel frame under a vertical load.


A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456 451

(Fig. 9). Such a beneficial beam arching action continues to exist


Frame
until the beam rotates significantly so that the horizontal force F 300 (a) capacity
at the right end is below the force at the left end. Under this situa- G
tion, the resulting vertical beam force along the removed column 250 Vertical
line starts to point downward and thus reduce the vertical force strand force
capacity of the frame. That is, the beneficial beam arching action Tv
200
is no longer available.
Varch + Tv

Force [kN]
Note that in the above introduction to the beam arching action,
an infinitely rigid beam is assumed to simplify the equation devel- 150
opment and improve the concept understanding. Under this
assumption, the gap between the beam and column opens imme- 100
diately when a vertical displacement is applied, and all forces
apply only at the top/bottom edges of the beam ends. In reality, Vertical
50 beam force
however, the gap starts to open (i.e., the beam end separates from
the column face on the tension flange side) only when a decom- Varch
pression moment capacity (provided by the initial PT forces) is 0
exhausted by the vertical force-induced flexural demand. There-
fore, before gap opens (i.e., when the vertical displacement is very -50
small), the vertical beam force is in the form of beam end shear 0 200 400 600 800 1000
associated with the beam flexural demand (up to the decompres- Vertical Displacement [mm]
sion moment) and thus should not be calculated by Eqs. (1) and
(2). After the gap opens, the vertical beam force is contributed by 2000
both the shear resistance of energy-dissipating elements (if exist-
(b )
ing) and the beam arching action. At this stage, the vertical beam
force Varch due to the arching action can be approximately calcu- Horizontal
lated by Eqs. (1) and (2) while keeping in mind that the horizontal beam force
reaction force is distributed over a finite contact surface at each 1500
F
beam end.
Fig. 10 presents the changes in the overall vertical force capacity
Force [kN]

of the one-story PT frame as well as the vertical and horizontal


forces of the beam (acting on the right-hand-side column) and 1000
the total axial force of strands, as the vertical displacement along
Strand
the removed column increases. Note that both the overall frame axial force
capacity and the strand forces are directly obtained from the T
ANSYS outputs, while the vertical beam force is calculated by 500
Eqs. (1) and (2). It is clearly observed in Fig. 10(a) that the algebraic
sum of the vertical beam force Varch and the vertical strand force Tv
(i.e., the vertical component of axial strand force in Fig. 9) essen-
tially matches the overall vertical force capacity of the frame. This
observation reveals that two major sources beam arching action 0
0 200 400 600 800 1000
Varch and strand catenary action Tv contribute to the vertical force
Vertical Displacement [mm]
capacity G of the frame. Such a relationship is expressed as
G = Varch + Tv in Fig. 9. Fig. 10. Responses of (a) the vertical forces and (b) horizontal beam force and axial
Note that the gap between the beam and column opens at a ver- strand force of the one-story frame.
tical displacement of about 10 mm, when the moment capacity
due to the initial PT forces is exhausted. As discussed above, the
vertical beam force prior to gap opening is offered by the beam vertical offset e (Fig. 9) and almost constant horizontal beam force
end shear corresponding to moment demand and should not be F (Fig. 10(b)). As the vertical displacement further increases to
calculated per Eqs. (1) and (2), and it is equal in value to the differ- 257 mm, the horizontal beam force F becomes 1325 kN, a value
ence between the frame vertical force capacity and the vertical very close to half of 2700 kN, which is the axial yield strength of
strand force. As the vertical displacement goes beyond 10 mm, it beam W24X62. After that, the horizontal beam force vs. displace-
appears in Fig. 10(a) that the accuracy of using Eqs. (1) and (2) to ment curve is almost flattened (Fig. 10(b)), indicating that roughly
calculate the vertical beam force is quite satisfactory, as evidenced half of the beam cross-section yields. This is because only a portion
by the proximity between the curve for the overall frame capacity of the beam cross section (i.e., that along the diagonal line connect-
and the curve for the sum of vertical beam and strands forces. Note ing the localized contact areas at the two ends of the inclined
that for the particular PT connection (Fig. 2), no energy-dissipating beam) is fully engaged to resist the compression by PT strands.
elements were installed. As a result, the vertical beam force is Note that unlike the vertical beam force Varch, the vertical strand
entirely from the beam arching action after the gap opens. force Tv ( = Tsinh) increases with an increasing vertical displace-
It is observed in Fig. 10(a) that the beam arching action domi- ment (Fig. 10(a)). This is quite obvious because as the vertical dis-
nates the vertical force capacity of the frame as long as the vertical placement increases, not only the strands are kept stretched (so
displacement is moderate (less than 346 mm). When the vertical the strand axial force T increases, as seen in Fig. 10(b)) but also
displacement reaches 69 mm, the beam starts to yield under axial the beam rotation angle h (Fig. 9) increases. After a vertical dis-
compression, as verified by the von Mises stress distribution in the placement reaches 346 mm, the vertical strand force Tv starts to
beam (Fig. 11). At a vertical displacement of 117 mm, the vertical exceed the vertical beam force Varch, indicating that the strand
beam force Varch attains its peak value and then decreases after- catenary action begins to dominate the vertical force capacity of
wards, due to, based on Eq. (1), the combined effects of reduced the frame.
452 A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456

Fig. 11. Distribution of von Mises stresses in the beam of the one-story frame at a vertical displacement of 69 mm.

Fig. 10(a) shows that at a vertical displacement of 600 mm, the far less than the typical fracture strain of 0.05 [31], indicating that
positive (i.e., upward) vertical beam force Varch reduces to zero, the strands still possess considerable reserve capacity although
suggesting the ending of beneficial beam arching action. In fact, they have already yielded at a vertical displacement of 760 mm
Eq. (2) suggests that, when the vertical displacement D is equal (Fig. 10(b)).
to the beam depth d, the vertical offset e is close to zero and hence
V (which is equal in value to Varch) calculated by Eq. (1) is zero. Note 4.3. Study of the tested three-story PT frame
that the depth of the beam W24X62 used in the one-story PT frame
is about 610 mm, a value practically equal to 600 mm considering We compare in Fig. 13 the vertical force vs. displacement curves
the approximation made in Eq. (2). As the vertical displacement generated from the tested three-story frame (Fig. 3) and its FE
goes beyond 600 mm, the vertical beam force switches its direction model (Fig. 5), respectively. In general, the two sets of data agree
by acting downward along the removed column line (i.e., its value well. For example, the gap opening forces are 176 kN and 167 kN
becomes negative in Fig. 10(a)), thereby reducing the vertical force from the test and FE simulation, respectively. The peak vertical
capacity of the frame. At the same time, the horizontal beam force force capacities of the frame are 497 kN and 502 kN from the test
(acting on the column) decreases due to the significantly inclined and FE simulation, respectively. Note that the FE model exhibits
position of the beam. At a vertical displacement of 950 mm, both a slightly higher initial stiffness than that of the tested frame. As
horizontal and vertical beam forces drop to zero (Fig. 10), indicat- will be discussed in detail later, the actual lower-than-reported
ing that the beam totally separates from the column face (Fig. 12). axial stiffness of strands in the tested frame, along with other
At this stage, the beam completely loses its gravity load-carrying
capacity. The corresponding strain of the strands is calculated to
be 0.018, a value slightly more than the yield strain of 0.012 but
500

Loading FE results

400

Test data [24]


Force [kN]

300 nd
2 - floor strand

200 rd
3 - floor strand

100
Roof strand

0
0 50 100 150 200
Vertical Displacement [mm]
Fig. 12. Deformation of the PT connection of the one-story frame at a vertical Fig. 13. Comparison of tested and simulated vertical force vs. displacement curves
displacement of 950 mm. of the three-story frame. Simulated axial forces of individual strands also shown.
A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456 453

various experimental uncertainties, can be a plausible reason for are compared in Fig. 14. It is clearly seen that both the overall
such a relatively stiffer FE model, which is originally built by deflection patterns and the member deformation details (e.g.,
assuming a defect-free strand assembly. beam local buckling locations and gap opening sizes) match satis-
It is observed in Fig. 13 that the peak capacity of the FE model factorily, confirming the high accuracy of the FE model in predict-
occurs shortly after the second-floor strands yield but slightly ing the nonlinear behavior of a column-removed PT frame.
before the vertical displacement reaches 89 mm, when the tested The overall discrepancy between the tested and simulated
frame was reported to gradually lose PT forces due to the succes- responses of the three-story PT frame could be attributed to ignor-
sive fracture (failure) of wires in its second-floor strands [24]. ing in the FE model miscellaneous experimental errors, for exam-
We calculate the strain of the second-floor strands at this displace- ple, the inherent imperfection and residual stresses in structural
ment to be 0.014, a value far below a typical ultimate (i.e., fracture) members, out-of-plumbness of columns, deviation of beam axes
strain of 0.05 [31]. Such a considerable strain difference implies with respect to one another, inherent flexibility at column bases,
that the strand fracture was unlikely to be the true cause of PT and non-uniform beam-to-column contact surfaces, as well as
force loss in the tested frame. Accordingly, it is not possible to potential defects of the steel materials and/or strand anchorage.
determine whether a tested strand failed or not based on its strain Indeed, because the tested three-story frame is more complex than
level. In order to simulate the successive wire failures, we decide to the tested connection, more experimental uncertainty and errors
model each wire of the second-floor strand using a LINK180 ele- can be expected. As a side note, such factors could also be respon-
ment, and sequentially remove each element using the KILL ELE- sible for the irregular fluctuation in the test data (Fig. 13). In the
MENT option, leading to a stepped decrease in the second-floor following subsection, the effect of possible strand defects (as an
strand force (Fig. 13). It was also reported [24] that the gradual fail- illustration of such experimental errors) on the performance of
ure of second-floor strands, along with the beam local buckling and the three-story PT frame is investigated.
energy-dissipating rod failure, caused the third-floor strands of the
tested frame to yield, a phenomenon clearly captured by the FE
model, as observed in Fig. 13. Note that as the second-floor strands
4.4. Effect of strand defects on the capacity of the three-story PT frame
gradually fail, an increase in the vertical displacement is accompa-
nied by a decrease in the overall frame capacity (Fig. 13). This
As already discussed, the strain associated with the reported
decrease is contributed not only by the reduced cross-sectional
failure of the second-floor strand wires is 0.014, which is much less
area of strands due to successive wire failure but also by the weak-
than the typical ultimate strand strain of 0.05. Besides, it was
ened beam arching action, the latter similarly observed in Fig. 10
reported that at a vertical displacement of 239 mm, the third-
(a) for the one-story frame case.
floor strands of the tested frame began to lose their PT forces
Besides, it was reported [24] that at a vertical displacement of
[24]. We calculate the corresponding strain of the third-floor
239 mm, the third-floor strands of the tested frame lost their PT
strands to be 0.024, again far below a typical ultimate strain of
forces, causing another major decrease in the vertical force capac-
0.05. The calculated strains at which the second and third-floor
ity of the frame. The corresponding deformed shapes of the three-
strands reportedly lost their PT forces would have been similar
story frame obtained from the test and predicted by the FE model
and both close to the ultimate strain of typical strand materials,
if the strand fracture were indeed the true cause of the PT force
losses during the test. In other words, the existence of such a con-
siderable difference in the strains of failed strands implies that the
loss of PT forces in the tested three-story frame might not neces-
sarily be contributed by PT strand rupture. Therefore, it is interest-
ing to explore probable reasons for the strands to lose their PT
forces at such a low strain level.

250
Load cell
capacity [24]

200
Axial Force [kN]

150 Test data [24]

FE results
100

50

0
0 20 40 60 80 100 120
Vertical Displacement [mm]
Fig. 14. (a) Tested [24] and (b) simulated deformed shapes of the three-story frame. Fig. 15. Tested and simulated second-floor strand forces of the three-story frame.
454 A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456

Fig. 15 compares the tested and simulated PT forces of a second- 4.5. Capacity analysis of the three-story PT frame with defect-free
floor strand in the three-story frame. It is observed that at the strands
beginning, both strand forces are equal to the initial pre-
tensioning force of 88.9 kN. As the vertical displacement increases, So far we have demonstrated that an impaired strand assembly
the axial stiffness of the tested strand is lower than that of the sim- could noticeably reduce the PT frame capacity against progressive
ulated strand. At a vertical displacement of 52 mm, the simulated collapse due to the premature loss of PT forces. Therefore, it is
strand yields at an axial force of 233 kN, which well exceeds the important to ensure that every component of a strand assembly
load cell capacity of 204.6 kN in the test [24]. For this reason, it works properly in order to prevent the gradual or sudden loss of
is unclear whether the tested strand indeed yielded or not. Since PT forces. In the following discussion, we report the results from
it is unlikely that major mechanical properties (e.g., elastic modu- a re-analysis of the three-story frame while assuming that the
lus and cross-sectional area) of the tested strands were consider- strands would possess their ideal stiffness as the vertical displace-
ably less than the reported values [24], we hypothesize that the ment increases. Under this assumption, a strand would rigorously
following two issues cause the relatively low axial stiffness of a follow the assigned bilinear stress-strain relationship, that is, it
strand in the actual test. First, because the wires of a tested strand would yield at a strain of 0.008 (which is calculated by dividing
were reported to rupture one by one as the vertical displacement the strand yield stress by its elastic modulus) and fracture (and
increased gradually [24], it is reasonable to infer that the wires thus completely loses its PT force) at an ultimate strain of 0.05.
of the tested strand also prematurely yielded one by one at a low We statically push the FE frame model down along the middle
strain level. Thus, such successive yielding of strand wires gradu- column until reaching a vertical displacement of 100 mm. The
ally reduced the overall axial stiffness of the tested strand. Second, resulting forces (including the vertical force capacity of the frame
it is also possible that the strand anchorage gradually slipped, thus and the axial forces of individual strands) vs. displacement curves
relaxing the strand and decreasing the overall axial stiffness of the are plotted in Fig. 17. It is observed that at a vertical displacement
strand assembly that can be viewed as a series system. of 42 mm the second-floor strands yield first. Then, as the middle
To test this hypothesis, we numerically adjust the axial stiffness column continues to be pushed down, the third-floor strands yield
of strands to reflect the effects of one or both of the above strand at a vertical displacement of 66 mm, and almost simultaneously
defect issues (i.e., premature wire yielding and anchorage slip- the overall force capacity of the frame reaches a peak value of
page). A trial-and-error approach is used to reduce the elastic mod- 497 kN. Then, there are slight drops in the forces of the frame
uli of individual strands in the FE model to equivalently albeit and third-floor strands, likely due to the initiation of local buckling
approximately account for such discrepancy in strand stiffness, at the bottom flange of the third-floor beam (Fig. 18). Note that the
until the simulated axial force vs. displacement curves of individ- roof strands remain elastic during this process. Similarly, the
ual strands satisfactorily match those from the test, respectively, decrease in the axial force of the roof strands after a vertical dis-
up to a vertical displacement of 89 mm. The curves obtained from placement of 88 mm is likely due to the initiation of local buckling
the test and the FE model using the properly reduced strand mod- at the bottom flange of the roof beam and the inward flexural
uli are compared in Fig. 16. It is observed that such curves of all deflection of columns, both factors relaxing the PT strands at the
individual strands match very well, respectively. However, the ver- roof level. Because the strands at the second and third-floor levels
tical force vs. displacement curve of the overall frame obtained have already yielded or nearly so, the gradual decrease in the PT
from the FE model exhibits a somehow lower post-gap-opening force of roof strands causes the vertical force capacity of the frame
stiffness than that from the test. Such discrepancy could be attrib- to drop accordingly.
uted, at least in part, to the miscellaneous sources of structural We also perform a parametric analysis to examine the effect of
restraining (e.g., friction between the column end and the sliding strand numbers on the response of the three-story PT frame
support plates) in the actual assembly of the tested frame. against column removal, using the same FE model except that
the number of defect-free strands is doubled (i.e., from the original

500 500

400 400 Frame

Frame
Force [kN]

Force [kN]

300 300
nd
2nd-floor strand
2 -floor strand

200 200
3rd-floor strand

100 3rd-floor strand 100


Roof strand
Roof strand
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Vertical Displacement [mm] Vertical Displacement [mm]
Fig. 16. Tested forces (symbols) and simulated forces (lines, using reduced strand Fig. 17. Simulated force vs. displacement curves of the three-story frame (with two
moduli) vs. displacement curves of the three-story frame. defect-free strands per beam) and its individual strands.
A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456 455

strands in place, the strands are partially relaxed and thus unable
to fully stretch to reach their desired strength. Therefore, doubling
the number of strands alone does not cost-effectively increase the
overall force capacity of this particular frame. In order to fully
engage the strands, beams should be strengthened to increase their
yield and local buckling capacities. In the meantime, columns
should not only have adequate flexural capacity to avoid excessive
inward deflection but also possess enough axial strength to resist
the elevated axial compression due to the increased PT forces.

5. Concluding remarks

We have successfully developed, calibrated, and validated high-


fidelity FE models to investigate the load redistribution capability
of post-tensioned (PT) steel frames against progressive collapse
upon notional column removal. For the studied PT connections
and frames, we have found that the vertical force capacity of a
column-removed PT steel frame is primarily contributed by the
Fig. 18. Local buckling at the bottom flange of the third-floor beam of the three- vertical components of both beam reaction force acting on the col-
story frame (assuming defect-free strands) at a vertical displacement of 66 mm.
umn (i.e., arching action) and strand axial forces (i.e., catenary
action). A major failure mode of a PT steel frame subject to column
two to four) per beam. The corresponding vertical force vs. dis- removal is the beam axial yielding and/or local buckling under the
placement curve of the PT frame, along with the axial forces of large axial compressive force imposed by PT strands (along with
individual strands, is shown in Fig. 19. Comparison to Fig. 17 the column restraining effect) as the frame moves downwards
reveals that, for this particular PT frame, doubling the number of under the unbalanced gravity loads. After that, the role of beam
strands increases the gap-opening load (from 119 kN to 231 kN) as a support to hold the strands in place is weakened. Unless col-
of the frame but does not significantly affect the peak force capac- umns have enough flexural capacity to avoid excessive inward
ity (from 493 kN to 513 kN, i.e., only a 4% increase) of the frame deflection, the strands are relaxed and thus the vertical force
within the studied vertical displacement range. This is because, capacity of the frame is negatively affected. Eventually, the beams
as discussed in the previous paragraph, the peak force capacity of completely separate from columns and/or the strands attain an
the original frame (i.e., the one with two defect-free strands per unacceptable inelastic strain (even if they have not yet fractured),
beam) is controlled by the yielding of second and third-floor beams signifying the imminent widespread failure of the PT steel frame.
followed by the local buckling of the third-floor and roof beams. An immediate implication for the design of a PT steel frame
Although the strands are doubled, such a beam failure sequence against progressive collapse is that the beams and columns must
essentially remains, yet the higher PT forces cause the beams to fail be adequately sized so that PT strands can be fully engaged to
at smaller vertical displacements, when all individual strands resist the unbalanced gravity loads resulting from a missing col-
behave elastically and their axial forces are well below those of umn scenario. In other words, the strengths and stiffness of beams,
their counterparts in the original frame (Fig. 17). As beams fail columns, and PT strands within a PT steel frame should be properly
while columns do not have enough flexural capacity to hold the balanced in order to achieve an overall cost-effective design. Once
such a design consideration has been implemented, it is expected
that under the unbalanced gravity loads, beams reach their axial
compression capacity while PT strands yield or experience an
acceptable level of plastic deformation. At this stage, the gravity-
500 carrying capacity of the frame is safeguarded by the available shear
resistance (through friction or energy-dissipating elements), trans-
Frame verse structural members, and/or strand catenary action. In the
400 meantime, columns should be able to resist the large compression
forces imposed by the sloped PT strands. Furthermore, it is empha-
sized that in order to attain the desired capacity of a PT steel frame
Force [kN]

against progressive collapse, early failure (e.g., premature wire


300 fracture and gradual anchorage slippage) of the strand assembly
should be avoided by all means.
Finally, it is worth noting that a more accurate study of the pro-
2nd-floor strand
200 gressive collapse potential of a frame structure requires that the
three-dimensional (3D) effect be considered. Indeed, load redistri-
rd bution upon sudden column removal is by nature a 3D phe-
3 -floor strand
100 nomenon due to the existence of transverse beams and floor
slabs. It is expected that considering such a 3D effect can have an
important impact on the prediction of progressive collapse
Roof strand response and capacity of PT steel connections and frames. The pre-
0
0 20 40 60 80 100 sent FE modeling and capacity investigation has focused on two-
dimensional (2D) structures solely because it is based on the 2D
Vertical Displacement [mm]
benchmark test data available in the literature. However, it is
Fig. 19. Simulated force vs. displacement curves of the three-story frame (with four straightforward to extend the FE models developed in the present
defect-free strands per beam) and its individual strands. study to account for the 3D effect by explicitly including the trans-
456 A. Pirmoz, M. Liu / Engineering Structures 126 (2016) 446456

verse structural elements. Research in this direction is warranted [13] ASCE. Minimum design loads for buildings and other structures (ASCE/SEI 7
10); 2010.
but out of the scope of the present study.
[14] Liu J, Tian Y, Orton S. Resistance of flat-plate buildings against progressive
collapse. II: System response. J Struct Eng 2015;141(12):04015054.
Acknowledgment [15] Liu M, Pirmoz A. Energy-based pulldown analysis for assessing the progressive
collapse potential of steel frame buildings. Eng Struct 2016;123:3728.
[16] Janssens VO, ODwyer DW, Chryssanthopoulos MK. Assessing the
The material is based on work supported in part by the Grant- consequences of building failures. Struct Eng Int 2012;22(1):99104.
in-Aid fund of The Catholic University of America. The financial [17] Corley WG. Applicability of seismic design in mitigating progressive collapse.
support is gratefully acknowledged. The opinions expressed herein, In: Proceedings of national workshop on prevention of progressive collapse. p.
101.
however, are those of the authors and do not necessarily reflect [18] Ellingwood B. Mitigating risk from abnormal loads and progressive collapse. J
those of the sponsor. Perform Constr Facil 2006;20(4):31523.
[19] Khandelwal K, El-Tawil S, Sadek F. Progressive collapse analysis of seismically
designed steel braced frames. J Constr Steel Res 2009;65(3):699708.
References [20] Liu M. Progressive collapse design of seismic steel frames using structural
optimization. J Constr Steel Res 2011;67(3):32232.
[1] Ricles J, Sause R, Garlock M, Zhao C. Posttensioned seismic-resistant [21] Pirmoz A. Performance of bolted angle connections in progressive collapse of
connections for steel frames. J Struct Eng 2001;127(2):11321. steel frames. Struct Des Tall Spec Build 2011;20(3):34970.
[2] Christopoulos C, Filiatrault A, Uang C, Folz B. Posttensioned energy dissipating [22] DoD. United facilities criteria design of buildings to resist progressive collapse
connections for moment-resisting steel frames. J Struct Eng 2002;128 (UFC 4-023-03). Washington, DC: Department of Defense; 2013.
(9):111120. [23] GSA. Alternate path analysis & design guidelines for progressive collapse
[3] Chancellor N, Eatherton M, Roke D, Akbas T. Self-centering seismic lateral force resistance. Washington, DC: US General Services Adminstration; 2013.
resisting systems: high performance structures for the city of tomorrow. [24] Tsitos A, Mosqueda G, Filiatrault A, Reinhorn AM. Experimental investigation
Buildings 2014;4(3):520. of progressive collapse of steel frames under multi-hazard extreme loading. In:
[4] Garlock M, Sause R, Ricles J. Behavior and design of posttensioned steel frame The 14th world conference on earthquake engineering.
systems. J Struct Eng 2007;133(3):38999. [25] Tsitos A, Mosqueda G, Filiatrault A, Reinhorn AM. Experimental investigation
[5] Ricles J, Sause R, Peng S, Lu L. Experimental evaluation of earthquake resistant of the progressive collapse of a steel post-tensioned energy dissipating frame.
posttensioned steel connections. J Struct Eng 2002;128(7):8509. In: 9th US national and 10th Canadian conference on earthquake engineering.
[6] Lin Y, Sause R, Ricles J. Seismic performance of steel self-centering, moment- [26] Tsitos A, Mosqueda G. Experimental investigation of the progressive collapse
resisting frame: hybrid simulations under design basis earthquake. J Struct Eng of a steel special moment-resisting frame and a post-tensioned energy-
2013;139(11):182332. dissipating frame. In: Fardis MN, Rakicevic ZT, editors. Role of seismic testing
[7] Moradi S, Alam M. Finite-element simulation of posttensioned steel facilities in performance-based earthquake engineering. Geotechnical,
connections with bolted angles under cyclic loading. J Struct Eng 2016;142 geological, and earthquake engineering, vol. 22. Springer; 2011. p. 36782.
(1):04015075. [27] Wang D. Numerical and experimental studies of self-centering post-tensioned
[8] Guo T, Song L-L, Zhang G-D. Numerical simulation and seismic fragility steel frames PhD Thesis. State University of New York at Buffalo; 2007.
analysis of a self-centering steel MRF with web friction devices. J Earthquake [28] Sadek F, Main J, Lew H, El-Tawil S. Performance of steel moment connections
Eng 2015;19(5):73151. under a column removal scenario. II: Analysis. J Struct Eng 2012;139
[9] ANSYS. ANSYS mechanical 16.2. Canonsburg, PA; 2015. (1):10819.
[10] Chou CC, Chen JH. Analytical model validation and influence of column bases [29] AISC. Seismic provisions for structural steel buildings (ANSI/AISC 341-
for seismic responses of steel post-tensioned self-centering MRF systems. Eng 05). Chicago, IL: American Institute of Steel Construction, Inc; 2005.
Struct 2011;33:262843. [30] Izzuddin BA, Vlassis AG, Elghazouli AY, Nethercot DA. Progressive collapse of
[11] Tzimas AS, Dimopoulos AI, Karavasilis TL. EC8-based seismic design and multi-storey buildings due to sudden column loss Part I: Simplified
assessment of self-centering post-tensioned steel frames with viscous assessment framework. Eng Struct 2008;30(5):130818.
dampers. J Constr Steel Res 2015;105:6073. [31] Devalapura RK, Tadros MK. Stress-strain modeling of 270 ksi low-relaxation
[12] Shiravand MR, Mahboubi S. Behavior of post-tensioned connections with prestressing strands. PCI J 1992;37(2):1006.
stiffened angles under cyclic loading. J Constr Steel Res 2016;116:18392.

You might also like