See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329525911

Progressive collapse performance analysis of precast reinforced concrete

structures

Article  in  The Structural Design of Tall and Special Buildings · December 2018

DOI: 10.1002/tal.1588

CITATIONS READS

0 335

3 authors, including:

De-Cheng Feng Zhun Wang

Southeast University (China) Southeast University (China)
30 PUBLICATIONS   113 CITATIONS    3 PUBLICATIONS   0 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Advanced finite beam elements for efficient analysis of large-scale structures View project

Robust numerical framwork for high-fidelity modeling of damage and failure in complex structures View project

All content following this page was uploaded by De-Cheng Feng on 10 December 2018.

The user has requested enhancement of the downloaded file.

Received xx xx 2018; Revised xx xx xx; Accepted xx xx xx
DOI: xxx/xxxx

RESEARCH ARTICLE

Progressive collapse performance analysis of precast

reinforced concrete structures

De-Cheng Feng1,2 | Zhun Wang1 | Gang Wu1,2

1
Key Laboratory of Concrete and Prestressed
Concrete Structures of the Ministry of Summary
Education, Southeast University, Nanjing
210096, China In this paper, the progressive collapse performance analysis of precast reinforced concrete (RC)
2
Laboratory of Industrialized Structural and
structures is performed. A numerical simulation framework for precast RC structures is develo-
Bridge Engineering of Jiangsu Province,
Southeast University, Nanjing 210096, China ped based on the OpenSEES software, where the fiber frame element is used for beam and column
type members and Joint2D element is used for the beam-to-column connections. The conjugated
Correspondence
Gang Wu, Southeast University, 2 Sipailou, material models are then introduced, and a Min-Max failure criterion is imposed on the origi-
Nanjing 210096, China. Email: nal models to reflect the steel fracture and concrete crushing when the structure undergoing
[email protected]
progressive collapse. In addition, to overcome the computational difficulties arisen from pro-
gressive collapse behavior, two enhanced nonlinear solutions ,i.e., the consistent quasi-Newton
algorithm and the explicit KR-α algorithm, are employed respectively for static and dynamic ana-
lysis. A 10-storey prototype precast RC structures is designed to verify the developed numerical
framework, and the progressive collapse resisting mechanism of the structures is investigated
through both static pushdown analysis and dynamic column-removal analysis. Finally, influences
of some typical parameters in precast RC structures on their progressive collapse performance
are studied.

KEYWORDS:
Progressive collapse, precast, reinforce concrete structures, numerical simulation, nonlinear
analysis solution, pushdown analysis, column removal

1 INTRODUCTION

Precast reinforced concrete (RC) structures are widely used in practical structural engineering due to its various advantages, e.g., high efficiency,
product quality, low environmental pollution, etc. Especially, in some developing countries like China, the government has attached great impor-
tance to promote precast RC structural systems in recent years, since there exists a large demand of industrial and civilian buildings in their rapid
process of urbanization. Therefore, it is of significant value to study the actual performance of the precast RC structures subjected to external
loadings for better understanding the failure mechanism and developing appropriate design methods.
In the past three to four decades, most studies on precast RC structures were focused on their seismic performance. The conventional way is
to conduct cyclic loading tests 1,2 of the precast beam-to-column connections and/or develop high fidelity finite element models 3,4 to assess the
seismic performance (e.g., failure mode, cyclic behavior, energy dissipation, etc.) of the corresponding precast structures. Nevertheless, as a new
concern, progressive collapse performance of structures has attracted more and more attentions 5,6,7 , since the unexpected accidental events, e.g.,
malicious attacks, gas explosion, vehicle impact, human error, etc., were frequently happened around the world in recent years, and caused great
loss of human lives and public properties. Although the probability of these extreme events is relatively small, the consequences are unbearable.
The most common and effective way to assess the progressive collapse potential of structures is the alternate load path (ALP) method, which is
defined by several national design codes and/or guidelines, such like DoD 2013 8 and GSA 2013 9 . The fundamental idea is to remove one column
2 D.C. FENG ET AL

of the structure to see whether or not the left structure can bridge over the missing column. Although removing one column can not represent the
real initial damage scenario, and even the location of the removed column is prescribed, it is still the most direct way to assess the progressive col-
lapse risk of a structure 10 . Therefore, experimental studies were first performed based on the idea of ALP method to investigate the progressive
collapse behavior of structures. Sasani and Kropelnicki 11 conducted 3/8 scale model tests of beam-column sub-assemblages under column remo-
val, and studied the corresponding load re-distribution mechanism. Yi et al. 12 tested a three-story four-bay RC frame subjected to middle column
removal, and developed a simplified model to predict its catenary action resistance. Su et al. 13 tested twelve RC frame beams against progressive
collapse and studied the axial restraining effects on beam loading capacity. The progressive collapse behavior of RC structures under removal of
exterior column and corner column was also investigated by Yap and Li 14 and Qian and Li 15 , respectively. Yu and Tan 16 performed two one-half
scale sub-assemblages under middle column removal to study the influence of seismic detailing. Ren et al. 17 , Lu et al. 18 and Qian et al. 19 studied the
progressive collapse behavior of RC floor system under middle column removal scenario and edge column removal scenario, respectively. Although
these experimental studies give realistic results of RC structures against progressive collapse, they may be costly and time consuming. Furthermore,
due to the limitation of the experimental facilities, space, manpower, etc., the influence of some typical parameters on the progressive collapse
behavior of structures cannot be fully investigated.
On the other hand, numerical modeling is an alternative approach to study the progressive collapse performance of structures. Generally, two
families of models have been developed. The first one is the three-dimensional (3D) solid element based finite element models 20,21,22,23 , which
can obtain the detailed responses of structures, e.g., concrete cracking and crushing, reinforcement yielding and fracture, etc. Nevertheless, the
computational cost is really large and several numerical difficulties (e.g., convergence problems for implicit methods and stability problems for
explicit methods) will arise. The other one is the macro-level element based models 24,25,26 , which usually uses fiber beam elements to simulate
the beam and column type structural members and uses macro component-based elements 16 to simulated the beam-to-column connections. This
approach is much more efficient compared with the first one, and some specific local responses of the structures can also be captured, thus it is
treated as a trade-off between numerical accuracy and computational efficiency. Consequently, the macro-level element based models are more
preferred in progressive collapse analysis of RC structures.
However, no matter experimental studies or numerical studies, few of them aimed at precast RC structures, although they were widely used
around the world. It seems that only Kand and Tan’s group has systematically tested progressive collapse behaviors of precast RC beam-column
sub-assemblages 10,27,28 , and several typical factors were studied, e.g., beam-to-column connection types, discontinuous reinforcement, application
of new material in cast-in-situ region, etc. Nevertheless, no numerical models for progressive collapse of precast RC structures has been developed
yet. In fact, the main difference between the monolithic RC structures and precast RC structures lies on the connection region. Based on previous
experimental tests, the shear behavior and bond-slip effect at the beam-to-column connections will particularly significant in precast structures
since the post-cast concrete quality can hardly be guaranteed 10 , thus it should be carefully considered in numerical models. Although the current
macro component-based elements are capable of reflecting these effects, the calibration of the model parameters is extremely complex. Meanwhile,
as an extreme behavior, progressive collapse of structures usually involves material failure, geometric nonlinearities, etc. Thus the numerical per-
formance of the progressive collapse analysis is a big challenge. Both implicit and explicit methods can be adopted to conduct progressive collapse
analysis, however, convergence issue for implicit method and stability issue for explicit method remain a big concern. To overcome these problems,
several enhanced nonlinear analysis solutions are developed, e.g., Krylov Newton algorithm 29 and quasi-Newton algorithm for implicit analysis,
unconditionally stable KR-α algorithm 26,30 for explicit analysis. They can be directly used for analysis of precast RC structures since the modeling
of this kind of structures is usually more complicated than that of monolithic RC structures.
Based on the above-mentioned background, this paper aims at developing a numerical framework for progressive collapse analysis of precast RC
structures, and then investigating the progressive collapse resisting mechanism of precast RC structures. The modeling strategy including element
models, material models and failure criteria is first presented. Then two enhanced nonlinear analysis solutions are introduced to overcome the
numerical difficulties arisen in static and dynamic progressive collapse analysis, respectively. A 10-storey prototype precast RC frames is designed,
and used to investigate the progressive collapse mechanism through both static pushdown analysis and dynamic column removal analysis. Finally,
systematic parametric studies are conducted to see the influences of some critical parameters in precast RC structures on its progressive collapse
performance.

2 PROGRESSIVE COLLAPSE SIMULATION STRATEGY FOR PRECAST RC STRUCTURES

As mentioned before, the modeling strategy for precast RC structures is first presented. The strategy is based on OpenSEES software and the
adopted element models, material models and the failure criteria are introduced in detail herein.
D.C. FENG ET AL 3

2.1 Finite element models

2.1.1 Beams and columns
For beam and column type structural members, the force-based (FB) fiber beam-column element 31 is used, as shown in Fig. 1 . The element adopts
force interpolation function to describe the internal force distribution along the element length, thus the equilibrium can be satisfied even for strong
material nonlinearity. This feature enables us using only one element for each beam or column member in simulation. Meanwhile, the geometric non-
linearity is also considered through co-rotational formulation. The section is discretized into several fibers according to plane section assumption,
and each fiber is subjected to uniaxial stress state. Different stress-strain relations can be assigned to different fibers, thus the stirrup-confinement
effect can be reflected. The incorporated concrete and steel constitutive models will be introduced in next sub-section. Integrating over the entire
section, the mechanical behavior of the section can be obtained, and the axial-flexure interaction can be also represented through the fiber model.
To overcome the localization issues 32,33,34 arisen from the softening behaviors, plastic hinge length integration method 35,36 is adopted for FB
element. The integration weights for the end integration points are assumed as the plastic hinge length Lp , which can be obtained from experimental
results or empirical formulation. In this paper, the formulation proposed by Paulay and Priestley 37 is used, i.e.,

Lp = 0.08L + 0.022fy db (1)

where L is the member length, fy and db are the yielding strength and diameter of the longitudinal bars, respectively. With the plastic hinge length
integration method, objective responses with respect to number of integration points can be achieved.

2.1.2 Beam-to-column connections

For beam-to-column connections, the Joint2D element developed by Altoontash 38 is used, as shown in Fig. 1 . Joint2D element is a simplified
version of the BeamColumnJoint element Lowes and Altoontash 39 and the macro-component element by Bao et al. 40 and Yu and Tan 16 , which all
require a complex calibration of model parameters and/or component properties. As a simpler model, the Joint2D element only consists five spring
components, which respectively represent the shear behavior of the joint panel and the moment-rotation behavior including bond-slip effect of the
four sections at beam-column interface. The behaviors of the five springs are all defined by uniaxial force-deformation relations. For the shear panel,
the shear force-shear distortion behavior can be determined by modified compression field theory (MCFT) 41 or the softened truss model (STM) 42 ,
while for the beam-column interface springs, the moment-rotation behavior can be calculated through a unit length quasi-fiber section model which
replaces the stress-strain relation for the steel fibers by the stress-slip relation to include the bond-slip effect at the member ends. Once the force-
deformation relationships of the five springs are determined, they are then simplified into a tri-linear relationship by getting the critical points and
then assigned to the springs, which can be done by the Hysteretic material model in OpenSEES. It should be noted that for common precast RC
structures the column end springs can be further simplified as rigid since no bond-slip failure would occur at the column-joint interface when the
sub-assemblage is under a column removal scenario.

Column

Shear spring
Rigid

Force-based fiber element Joint2D element

Beam Beam

Joint panel
Rigid Interface spring

Column

Integration point
Fiber section

FIGURE 1 Finite element models for precast RC structures

4 D.C. FENG ET AL

2.2 Material models

2.2.1 Concrete
The uniaxial plastic-damage model is adopted for concrete fibers. The model is based on damage mechanics 43,44,45 and has been implemented into
OpenSEES software as ConcreteD material. The general form of the stress-strain relation can be written as

σ ± = 1 − d± Ec ± − p± = 1 − d± Ec e± (2)

where σ ± is the tensile or compressive stress; ± is the tensile or compressive strain; e± and p± are the elastic and plastic components of the
strain, respectively; Ec is the initial elastic modulus; d± is the tensile or compressive damage variable; subscript ” ± ” represents tensile part or
compressive part, respectively.
The damage variables can be obtained through an experimental damage function 46,47 , i.e.,

ρ± n±
1 − ±


n±
for x± ≤ 1
± n −1+(x± )
d = (3)
ρ±
1 − ± ± for x± > 1

 2
α (x −1) +x±

in which
e± fc± 1
x± = , ρ± =
, n± = (4)
±
c Ec ±
c 1 − ρ±
where fc± and ±
c are the tensile or compressive peak strength and corresponding strain, respectively; α is the tensile or compressive descending
±

parameter that controls the descending branch of the stress-strain curve.

The plastic strain is defined by an empirical plastic evolution model, which is expressed as

ηp
p+ = 0, p− = ξp d− (5)

where ξp and ηp are the plastic coefficients, and usually assumed to be 0.6 and 0.1 in analysis, respectively. Note that the tensile plastic strain is
neglected since it is relatively small and has little influence on the overall behavior of concrete structures.
The typical cyclic behaviors are also depicted in Fig. 2 . The curve starts in compression from O-A while the compressive damage increases, and
then linearly unloads to the plastic strain point B. In the unloading path A-B, the compressive damage remains unchanged. Then the curve enters
into the tension part, and monotonically develops along path B-C, in which the tensile damage develops. At point C, it unloads to point B firstly, then
reverses into compression and reloads along B-A. Afterwards, another similar reversal regime is imposed. Obviously, in the above loading scheme,
both damage and residual deformation are observed, and the resulting strength softening and stiffness degradation are also obtained.

Compression

Ec B
O
C Tension

FIGURE 2 Typical cyclic behavior of concrete model

In addition, two more modifications are made on the original model to consider the concrete-steel interactions, i.e., tension stiffening effect for
tensile property 48 and stirrup-confinement effect for compressive property 49 , as shown in Fig. 3 . For the first one, the decay rate of the tensile
descending branch will slow down due to contributions of longitudinal bars, while for the second one, the compressive strength and ductility of will
D.C. FENG ET AL 5

Confined concrete
fc fcc

fc
With tension stiffening
fc c
Peak strength (unconfined)
corresponding strain (unconfined)

Without tension stiffening Unconfined fcc cc

concrete
Peak strength (confined)
corresponding strain (confined)

c c cc

(a) Tension stiffening effect (b) Stirrup-confinement effect

FIGURE 3 Concrete model considering concrete-steel interaction

get improved. In this paper, the model by Stevens et al. 48 is used to identify the tensile parameters for concrete fibers and the Mander model 49,50 is
used to calculate the corresponding parameters for confined concrete fibers.

2.2.2 Steel
The Menegotto-Pinto (MP) model 51 is used for reinforcement steel fibers, which is available in OpenSEES as the Steel02 material. The monotonic
skeleton is given by 
Es s for s ≤ y
σs = (6)
f + E ( −  ) for s > y
y h s y

where σs and s are the stress and strain of the steel, respectively; fy and y are the yielding stress and strain of the steel, respectively; Es is the
elastic modulus; Eh = bEs is the hardening modulus and b is the hardening ratio.

Bare bar Embedded bar

1 1
r
, r
2 2
0
, 0
Eh

Es

1 1
0
, 0

2 2
r
, r

FIGURE 4 Typical cyclic behavior of steel model

To account for the tension stiffening effect, the skeleton of the stress-strain curve of steel should also be modified. The properties of reinfor-
cement embedded in concrete is different from that of bare reinforcement, since the former actually corresponds to the average stress and strain
over a certain length which contains several cracks, rather than a local point. Based on the experimental studies conducted by Belarbi and Hsu 52 ,
6 D.C. FENG ET AL

the yield strength fyavg and the harden ratio bavg of the embedded reinforcement can be given by

fyavg = (0.91 − 2B) fy , bavg = (0.02 + 0.25B) b, B = fc+ / (fy ρ) (7)

where ρ is the reinforcement ratio of the longitudinal bars.

The hysteretic behavior is determined by two sets of asymptote lines, as shown in Fig. 4 . The stress firstly unloads as the initial stiffness Es , and
then transits to the opposite direction through the curve defined by the asymptote lines with slope Es and Eh , respectively. The expression for the
process is
(1 − b) ∗
σ ∗ = b∗ + (8)
(1 + ∗R )1/R
with
 − r σ − σr
∗ = , σ∗ = (9)
0 − r σ0 − σr
where (0 , σ0 ) correspond to the strain and stress at the intersection point of the two asymptote straight lines; (r , σr ) correspond to the strain and
stress at the last reversal point; R is the coefficient that controls the shape of the transition curve in order to better represent the Bauschinger0 s
effect. Meanwhile, the point sets (0 , σ0 ) and (r , σr ) will be updated.

2.2.3 Shear panel

The shear panel spring in the Joint2D element is defined by a uniaxial Pinching4 material in OpenSEES 39 , which is shown in Fig. 5 . The monotonic
curve is defined by 8 points (symmetrically 4 points in each direction), while the unloading and reloading paths are defined by 6 additional points
(symmetrically 3 points in each direction). Stiffness and strength are assumed to deteriorate due to the imposed loading history. The MCFT or
STM can be used to calculate the envelope of shear stress-strain relation τ − γ of the panel, and then converted to the moment-rotation relation
M − θ with the following expressions: M = τ VJ , θ = arctanγ , where VJ is the volume of the panel. The parameters for cyclic behavior are defined
according to the recommendations by Lowes and Altoontash 39 .

Load

Skeleton curve

Deformation

Cyclic paths

FIGURE 5 Pinching4 material for shear panel spring in Joint2D

2.2.4 Bond-slip
The force-deformation behavior of the beam-column interface springs is determined through a quasi-fiber section model approach, as shown in
Fig. 6 . The stress-strain model of the reinforcement fibers is replaced with the stress-slip model, thus the bond-slip effect can be represented. The
generated force-deformation curve is then simplified to a tri-linear model (i.e., Hysteretic material in OpenSEES) and assigned to the springs. The
stress-slip model can be obtained through either experimental results or analytical derivations. In this paper, the analytical model developed by the
author 53 considering different anchorage types and anchorage lengths for precast RC structures is adopted.
The slip s is derived based a stepped bond stress filed assumption, as shown in Fig. 7 , where the hooked bar an be simplified as a straight bar
with an equivalent straight length. The bond stresses in the elastic and plastic regions of the reinforcement are assumed as constant, thus the total
D.C. FENG ET AL 7

y
M
Fiber analysis result
Concrete fiber
(using stress-strain)

z Simplified tri-linear relation

Steel fiber
(using stress-slip)
q
Unit length fiber section analysis

FIGURE 6 Quasi-fiber section model approach for interface springs in Joint2D

hooked straight bond stress bar stress bar strain

uby fs s
Lydb

ube fy y
Ld
Ledb

embedded length

FIGURE 7 Stepped bond stress distribution

slip can be expressed by the following equation according to equilibrium condition 53,54,55
Z Ld
s=  (x) dx (10)
0
where  (x) is the strain field of the reinforcement; Ld is the developed bond length.
The detailed expression depends on the relation of the bar embedded length Lembd and bond developed length Ld . The embedded length Lembd
for continuous and lap-spliced bars are the straight length, while the embedded length for hooked bars is approximated as Lembd = Lbembd + 5db , in
which Lbembd is the length of the straight part and db is the bar diameter. Consequently, the slip is

• If Lembd > Ld 
 s Ledb fs db
s ≤ y Ledb =

2 4ube
s=
(fs −fy )db
(11)
 y Ledb
 +
y +s
Lydb s >  y Ledb =
fy db
, Lydb =
2 2 4ube 4uby
p p
where s is the bar strain at the loaded end; fy and y are the yield stress and strain of the bar; ube = 1.0 fc− and uby = 0.5 fc− are the
bond stresses for the elastic and plastic parts, respectively; Ledb and Lydb are the elastic part and plastic part of the developed bond lengths,
respectively.

• If Ld > Lembd > Ledb

∗ For continuous bar:


 fs db
 2s Ledb s ≤ y Ledb =


 4ube
(fs −fy )db

y  + fy d b fy db
s= L + y 2 s Lydb
2 edb
s > y Ledb = 4ube
, Lydb = 4uby
≤ Lembd − 4ube
(12)

(fs −fy )db

 end +y  +
Ledb + y 2 s Lydb s > y Lydb = , Ledb = Lembd − Lydb


2 4uby

where end is the strain at the bond stress end, and can be determined through similar triangle method.
8 D.C. FENG ET AL

∗ For lap-spliced and hooked bar:


s fs db

 Ledb s ≤ y Ledb =
2 4ube


y  + fy d b (fs −fy )db fy db
s= L + y 2 s Lydb
2 edb
s > y Ledb = 4ube
, Lydb = 4uby
≤ Lembd − 4ube
(13)

(fs −fy )db

   +
s0 + 2y Ledb + y 2 s Lydb s > y Lydb = , Ledb = Lembd − Lydb


4uby

where s0 is the free-end slip which can be calculated according to the model by Alsiwat and Saatcioglu 56 .

• If Lembd < Ledb

∗ For continuous bar: 

s fs db
 Ledb s ≤ y Ledb = ≤ Lembd
2 4ube


s= end +s
2
Lembd s ≤ y Ledb = Lembd (14)

(fs −fy )d

 end +y L + y +s L

s > y Lydb = 4u b , Ledb = Lembd − Lydb
2 edb 2 ydb
by

∗ For lap-spliced and hooked bar:


 fs db
 2s Ledb

 s ≤ y Ledb = 4ube
≤ Lembd

s = s0 + 2s Lembd s ≤ y Ledb = Lembd (15)

(fs −fy )d

s + y L +
 y +s
Lydb s > y Lydb = 4u b , Ledb = Lembd − Lydb
0 2 edb 2 by

2.3 Failure criteria

Material failure may occur when the structure is subjected to progressive collapse, e.g., concrete crushing and reinforcement fracture. Thus additi-
onal criteria should be applied to the original material models. OpenSEES provides a Min-Max material to reflect the material failure, as shown in
Fig. 8 . If the strain of the model exceeds the pre-defined minimum or maximum values, the material is assumed to be failed and zero stress and stif-
fness are returned. Specifically, the thresholds for reinforcement are min = −0.01 (buckling) and max = 0.15 (fracture); for unconfined concrete
are min = −0.0033 (crushing) and max = 0.001 (split); for confined concrete are min = −0.03 (crushing) and max = 0.001 (split) 26 .

The original material model

Min value

min max

Max value

The original material model

FIGURE 8 Min-Max model for material failure

2.4 Model validation

To validate the above proposed modeling approaches, the tests of precast beam-to-column sub-assemblages under middle column removal con-
ducted by Kang and Tan 10 are modeled. Four specimens are selected, which are named by MJ-B-0.52/0.35S, MJ-B-0.88/0.59R, MJ-L-0.52/0.35S
D.C. FENG ET AL 9

and MJ-L-0.88/0.59R, respectively, where "MJ" denotes middle joint; "B" and "L" denote bent or lap-splice connecting bar types; the first and second
numbers denote the top and bottom longitudinal reinforcement ratios; "R" and "S" denote rough or smooth surfaces of the precast specimens. More
details about the geometrical dimensions, material properties and the boundary conditions can be found in Kang and Tan 10 .
The simulated responses of the specimens by the proposed numerical modeling approach, i.e., the vertical displacement of the middle column
versus the vertical applied load and the horizonal reaction forces of the beams versus the vertical applied load, are demonstrated in Fig. 9 . Obvi-
ously, the numerical and experimental results match each other very well for nearly all the specimens. It can be found from the applied load-vertical
displacement curves that the initial stiffness, flexural beam action, and the effects of compressive arch action (CAA) and catenary action all can be
well reflected by the numerical model. Furthermore, the bar fracture failure at the middle column joint and end column stubs can also be reprodu-
ced. In general, the comparison of numerical and experimental results indicates that the proposed numerical modeling approach can reproduce the
realistic responses of precast beam-to-column sub-assemblages, and therefore can be used as an effective tool in a progressive collapse analysis of
precast RC structures.

60 100
Test data
50 Simulation

Horizontal reaction force (kN)

0
40
Applied load (kN)

30 CAA Catenary action

-100

20 Beam action

-200
10 Test data
Simulation
0 -300
0 100 200 300 400 500 0 100 200 300 400 500
Middle joint displacement (mm) Middle joint displacement (mm)

(a) MJ-B-0.52/0.35S-vertical (b) MJ-B-0.52/0.35S-horizonal

100 400

300
Horizontal reaction force (kN)

80
200
Applied load (kN)

60
100

40 0
Catenary action

CAA -100
20 Beam action Test data Test data
-200
Simulation Simulation
0 -300
0 200 400 600 800 0 200 400 600 800
Middle joint displacement (mm) Middle joint displacement (mm)

(c) MJ-B-0.88/0.59R-vertical (d) MJ-B-0.88/0.59R-horizonal

FIGURE 9 To be continued

3 ENHANCED NONLINEAR ANALYSIS SOLUTIONS FOR PROGRESSIVE COLLAPSE ANALYSIS

The progressive collapse performance of precast RC structure can be analyzed through either static method or dynamic method. However, both
the method will meet the numerical difficulties since the progressive collapse behavior involves several extreme behaviors, e.g., material failure and
strong geometric nonlinearity. Therefore, to improve the numerical performance, two enhanced algorithms are respectively introduced for static
and dynamic progressive collapse analysis.
10 D.C. FENG ET AL

60 200
Test data
50 Simulation

Horizontal reaction force (kN)

100
40
Applied load (kN)

30 0

20
CAA Catenary action -100
Beam action
10 Test data
Simulation
0 -200
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
Middle joint displacement (mm) Middle joint displacement (mm)

(e) MJ-L-0.52/0.35S-vertical (f) MJ-L-0.52/0.35S-horizonal

80 300

200

Horizontal reaction force (kN)

60
Applied load (kN)

100

40 0
CAA Catenary action
-100
20
Beam action
Test data -200 Test data
Simulation Simulation
0 -300
0 200 400 600 800 0 100 200 300 400 500 600 700
Middle joint displacement (mm) Middle joint displacement (mm)

(g) MJ-L-0.88/0.59R-vertical (h) MJ-L-0.88/0.59R-horizonal

FIGURE 9 Comparison of numerical and experimental results

3.1 Consistent quasi-Newton algorithm for nonlinear static analysis

For static analysis, a suitable algorithm is needed to solve the nonlinear equations. Usually the full Newton-Raphson method is adopted since it
has the quadratic rate of convergence. However, the computational cost is relatively large since the tangent stiffness must be re-computed at each
iteration, and convergence issues may arise due the negative stiffness of the system. To improve the performance, several method has been propo-
sed, e.g., modified Newton-Raphson, quasi Newton-Raphson, Krylov Newton-Raphson, etc. In this paper, a recently developed two-level consistent
quasi Newton-Raphson algorithm 57 is adopted.
In general, the nonlinear finite element equilibrium equation can be expressed as follows

f = P (ũ) (16)

where P and f denote the internal and external force vectors of the system, respectively; ũ denotes the nodal displacement vector. To solve the
above equation, the following iterative scheme is adopted

Km ∆ũm+1 = f − P (ũm ), ũm+1 = ũm + ∆ũm+1 (17)

where Km denotes the iterative stiffness matrix; ∆ is the increment symbol; m is the iterative number index.
sec . Numbers of celebrated
In the quasi Newton-Raphson method, the secant stiffness matrix Ksec is chosen as the iterative matrix, i.e., Km = Km
works have been done to develop the secant matrix, the most famous one is the Broyden-Fletcher-Goldfarb-Shannon (BFGS) 58 update, which is
introduced by Matthies and Strang 59 (M-S) to solve the nonlinear finite element equation, the secant matrix for each iteration can be written as
 h i −1 h i
(Km )−1 = Λ − ρm ∆f m (∆ũm )T Ksec m−1
Λ − ρm ∆f m (∆ũm )T + ρm ∆ũm (∆ũm )T
sec
(18)
ρ = (∆ũm )T ∆f m
m
D.C. FENG ET AL 11

where Λ is the identity matrix.

Although the quasi-Newton-Raphson method needs more iterations compared with the full Newton-Raphson method, it is still computationally
efficient because the inverse secant matrix is computed directly while Newton-Raphson method performs inversion of matrix in each iteration.
Besides, the M-S secant stiffness is also numerically robust due to its well-posed resulting secant matrix. The main difficulty of the M-S method is
the selection of the initial secant matrix K0sec . In this paper, a secant operator derived from the material level is developed. Generally, the explicit
expression of the iterative secant matrix can be summarized as
Z
Ksec = N T SN dΩ (19)
Ω
where S = σ/ is the secant stiffness at the material level; N is the shape function in a displacement-based finite element method; Ω is the entire
problem domain. For this study, since we use the force-based beam element, Eq. (19) can be re-expressed as
Z −1
Ksec = B T S−1 BdΩ (20)
Ω
where B is the force shape function in the force-based beam element.
The material level based secant stiffness matrix Ksec defined in Eq. (19) caries more information from the material level, with which the nonlinear
behavior of the structure can be better described, but it is not consistent with the BFGS update scheme. The structural level based secant stiffness
matrix Ksec defined in Eq. (18) is consistent with the BFGS update and has a faster convergence rate, but the bandwidth may be large if the update
of Eq. (18) has been performed for too many times. To address the problems encountered at different levels, a two-level update strategy of secant
matrix is proposed in the paper. Just as shown in Fig. 10 , the main idea is to start the secant iteration with the material level based matrix Ksec and
then followed by the structural level based matrix Ksec . Furthermore, to reduce the bandwidth and the cumulative errors of the structural level
based secant matrix updated by Eq. (18), the procedure is restarted by the material level based secant matrix every certain numbers of iterations,
e.g., 8 iterations according to the previous computational experiences.

f
K 1sec K sec
2 3
sec
4
K sec
f n +1
0
sec restart

start
fn
o un un1 +1 un2+1 un3+1 un4+1 un5+1 ...... u

FIGURE 10 Two-level update strategy for quasi Newton-Raphson algorithm

3.2 Explicit KR-α algorithm for nonlinear dynamic analysis

For the dynamic analysis, either implicit method or explicit method can be adopted to solve the equations of motion. Implicit method is unconditi-
onally stable but requires iteration at each time step, while the explicit method does not need iteration but is generally conditionally stable, and an
extreme small time step is needed. Recently, Kolay and Ricles 30,60 developed a family of unconditionally stable explicit algorithm named as KR-α
algorithm, and has been successfully applied in collapse analysis of RC structures 26 , which is also adopted here.
12 D.C. FENG ET AL

In general, the equations of motion of a dynamic system can be written as

 
M Ü (t) + C U̇ (t) + R U (t) , U̇ (t) = F (t) (21)

where M and C are the mass and damping matrices, respectively; U (t), U̇ (t) and Ü (t) are the displacement, velocity, acceleration vectors,
respectively; R (t) and F (t) are the resisting force and applied force vectors, respectively; t is the time.
According to the explicit KR-α algorithm, the displacement and velocity updating, and the discretization of the equations of motion are expressed
respectively by
Ẋi+1 = Ẋi + α1 ∆tẌi , Xi+1 = Xi + ∆tẊi + α2 ∆t2 Ẍi (22)

M Ẍ i+1 + C Ẋi+1−αf + Ri+1−αf = Fi+1−αf (23)

c

where Xi , Ẋi and Ẍi are the displacement, velocity, acceleration vectors of the i − th step, respectively; α1 and α2 are the integration parameter
matrices; ∆t is the size of the time step; Xi+1 , Ẋi+1−αf , Ri+1−αf and Fi+1−αf are the equivalent acceleration, velocity, restoring force and applied
b̈

force vectors, and can be defined as

Ẍ i+1 = (I − α3 ) Ẍi+1 + α3 Ẍi (24)
c

(25)




Ẋi+1−αf = 1 − αf Ẋi+1 + αf Ẋi , Ri+1−αf = 1 − αf Ri+1 + αf Ri , Fi+1−αf = 1 − αf Fi+1 + αf Fi

in which I is the unit matrix; α3 is the integration parameter matrix; αf is the scalar integration parameter; Ri (Ri+1 ) and Fi (Fi+1 ) are the i − th
(i + 1 − th) restoring force and applied force vectors, respectively. The integration parameter matrices above can be determined by setting the
eigenvalues of the amplification matrix of the KR-α method equaling that of the generalized-α method, i.e.,
−1
α1 = M + γ∆tC + β∆t2 K (26)

M
 
1
α2 = + γ α1 (27)
2
−1 
α3 = M + γ∆tC + β∆t2 K αm M + αf γ∆tC + αf β∆t2 K (28)
 

where K is the initial stiffness matrix; parameters γ , β , αm and αf are the functions of the user defined high-frequency spectral radius ρ∞ , and the
expressions are given by
1 1
γ= − αm + αf , β = (1 − αm + αf ) (29)
2 4
2ρ∞ − 1 ρ∞
αm = , αf = (30)
ρ∞ + 1 ρ∞ + 1
As can be seen from Eqs. (26)-(29), KR-α algorithm only has one user-defined parameter, i.e., high-frequency spectral radius ρ∞ , whose value
varies from 0-1. ρ∞ = 1 means no numerical energy dissipation is introduced, and the numerical energy dissipation increases with the grow of
ρ∞ , until ρ∞ = 0 where the maximum numerical energy dissipation occurs. The accurate value for ρ∞ depends on the problem and the size
of the time step. Higher value of ρ∞ (smaller numerical energy dissipation) will cause spurious high-frequency oscillations of the internal force
responses, which can be removed by either reducing the time step or introducing appropriate numerical energy dissipation, i.e., decreasing the
value of ρ∞ 61 .. However, reducing the size of the time step will lose the computational efficiency of the KR-α method. Consequently, a convergence
study is commonly used, the user selects some response parameter of interest and studies how the response converges with reduced values of ρ∞
starting with ρ∞ = 1, thus the appropriate value of ρ∞ can be determined when a converged response is achieved. More details can be seen in
Kolay and Ricles 30,61 , and see Appendix ?? for the commands used in OpenSEES.

4 PROGRESSIVE COLLAPSE PERFORMANCE ANALYSIS OF PRECAST RC STRUCTURE

4.1 Design of prototype structure

To assess the performance of the proposed numerical framework, a prototype precast RC frame structure of 10-storey tall is designed and its
progressive collapse resisting mechanism is investigated via both static pushdown analysis and dynamic column-removal analysis. The structure is
designed in accordance with the Chinese design code of concrete structures 62 , and the precast beam-to-column connection type recommended by
the Chinese technical specification for precast concrete structures 63 is adopted, as shown in Fig. 11 , in which the anchorage/embedment length
of the bottom reinforcement passing through the middle joint is Lembd = 44db . The location of the prototype structure is assumed to be at Nanjing,
China, where the site class is defined as group II (the soil shear wave velocity vs is between 250 m/s to 500 m/s), the seismic intensity is categoried
as 7.0 (the peak ground acceleration (PGA) with a 10% probability of exceedance in 50 years is 0.1 g, where g is the acceleration of gravity), and the
D.C. FENG ET AL 13

characteristic site period Tg is 0.40 s. Dead load (DL) for roof is set as 7.0 kN/m2 while it is 5.0 kN/m2 for floor. Live loads (LL) for roof and floor are
both set as 2.0 kN/m2 . The configuration of the structures, geometric size of the sections, and the reinforcing details are shown in Fig. 11 , where
the diameters of the reinforcement bars outside the brackets are for 1-4 storeys while those in the brackets are for 5-10 storeys. The material
properties are given in Table 1
Two widely adopted progressive analysis procedures are used in this paper to assess the performance of the prototype structures, namely, the
static pushdown analysis and the dynamic column removal analysis, which are also recommended in several national codes 8,9 . Actually, the two
methods have no relationship between each other, but have its own characteristics. The static pushdown analysis is easy to perform, but it assesses
the progressive collapse potential of the structures at the macro level. On the other hand, the dynamic column removal analysis is relatively compli-
cate, and can obtain the detailed internal force re-distribution during the progressive collapse process. Consequently, the first method can be used
to evaluate the general progressive collapse risk, while the second method can be used to study the progressive collapse resistance mechanism.

precast column

8@100/200
grout sleeve
cast-in-situ region

500
8@100/200
precast beam
3600×9

8@100/200 600 8@100/200

Beam-to-column connection

4D25(20)
4D20

380 120
2D25(20)
600

500
2D12

4D25(20) 4D20(18)
4500

600 250

A B C D E F Columns Beams

6000×5

FIGURE 11 Design of the prototype precast RC structure

TABLE 1 Material properties of the prototype precast RC structure

Concrete Steel
Storey
Ec (MPa) fc+ (MPa) +
c α+ fc− (MPa) −
c α− Es (MPa) fs (MPa) b
1-4 32500 2.6 0.0001 1.0 26.7 0.002 2.5 200000 400 0.01
5-10 30000 2.0 0.0001 1.0 20.1 0.002 2.5 200000 400 0.01
14 D.C. FENG ET AL

4.2 Static pushdown analysis

First the static pushdown analysis is performed to assess the progressive collapse behavior of the frame. In the pushdown analysis procedure, the
column at the base storey is firstly removed, and then the designed gravity load is applied proportionally in the damaged bays, namely, the vertical
load for the damaged bays is α(1.2×DL+0.5×LL), while that for the remaining bays is 1.2×DL+0.5×LL, where α is the proportional load factor 8 .
The load factor α versus the corresponding vertical displacement ∆ is the pushdown curve, and α is actually the dynamic increase factor which can
be regarded as the upper boundary of the progressive collapse capacity. In the static analysis, the displacement increment is set as 0.1 mm and the
tolerance for convergence is set as 10−6 on the energy norm.
Single column of the frame in Fig. 11 is removed firstly, and the corresponding pushdown curves are shown in Fig. 12 (a). As we can see, no
matter removing the exterior column A or removing the interior column C, both the curves exhibit obvious three stages. At first the gravity load is
resisted by the beam-end moments, and the load factor increases with the vertical displacement. In this stage, some parts of the sections will crack
due to tension and the remaining part is subjected to compression, leading to the compressive arch action (CAA), which is corresponding to the first
peak in the figure. After the plastic hinges are fully developed at the beam ends, the state of sections of the beam transfers from compression to
tension, causing the tensile catenary action (TCA), which can further improve the vertical load capacity. Finally longitudinal bars of the beams at the
damaged bay will progressively fractured due to the large vertical displacement, thus the curves start to decay, indicating the structure is gradually
losing the capacity to resist the gravity load, i.e., collapsed. It should be noted that the pushdown curves for removal of column A and removal of
column C are nearly the same, the maximum load factors are both around 2.0, which means the capacities of the frame to resist progressive collapse
under these two cases are very close.

2.4 1.2
Bar fracture occurs
2.0 1.0
First peak Bar fracture occurs
1.6 0.8
Load factor α

Load factor α

First peak
1.2 0.6

0.8 Stage 1 Stage 2 Stage 3 0.4 Stage 1 Stage 2 Stage 3

CAA TCA Collapse CAA TCA Collapse

0.4 Removal of column A 0.2 Removal of column A & B

Removal of column C Removal of column C & D
0.0 0.0
0 200 400 600 800 1000 1200 0 400 800 1200 1600
Vertical displacement (mm) Vertical displacement (mm)
(a) Removal of single column (b) Removal of double columns

FIGURE 12 Pushdown curves of the prototype structure

Removal of double columns of the frame are also analyzed. Fig. 12 (b) gives the pushdown curves for cases of removal of columns A & B and
removal of columns C & D. Similarly, the curves also have three stages, however, the progressive collapse resisting capacities of removing double
columns is much lower than that of removing single column, the maximum load factor is only half of that for removing single column. This is because
the span of the damaged bay for removing double columns is much larger than that for removing single column, thus the structure will become much
more unstable under gravity load. In addition, the corresponding progressive collapse behaviors for these two double columns removal scenarios
are very different. The load factors corresponding to the first peaks for removing A & B and C & D are 0.47 and 0.74, respectively, while the maximum
load factors are 0.71 and 1.08, respectively. The damaged bays are actually cantilevers when removing columns A & B, while the damaged bays are
restrained by the left structure at both directions when removing columns C & D. Consequently, the curve for removing C & D has higher initial and
hardening stiffnesses since the damaged bays have a stronger boundary constraint. Meanwhile, the maximum collapse capacity of the frame under
removal of columns C & D is also larger than that under removal of columns A & B, since the remaining structure can provided more alternate paths
to resist the gravity load.
D.C. FENG ET AL 15

4.3 Dynamic column-removal analysis

Dynamic column-removal analysis is the most direct way to study the progressive collapse behavior of the structure. Unlike the static pushdown
analysis, the vertical gravity load (i.e., 1.2×DL+0.5×LL) is first applied to the structure and the column is removed subsequently in a very short time
in this procedure. In this paper we also conduct dynamic column removal analysis, and in the analysis the time step is set as 0.001s and the value of
ρ∞ is 0.5.
Both removal of single and double columns are performed, but only removing the exterior columns are analyzed in this part, since we have found
from the static pushdown analysis that the progressive collapse potential of removing exterior columns is much higher than that of removing interior
columns. Fig. 13 gives the responses for removing single column A, i.e., the displacement time history of the upper node of column A, and the axial
forces time history of the remaining columns and beams. It can be seen that the vertical displacement of the upper node of column A will suddenly
drops to 71 mm after the column is removed, and then vibrates in the vertical direction until finally reaching a steady state around displacement 65
mm. Evidently, the left structure can bear the gravity load in this case. On the other hand, after the column A is removed, the gravity load will cause
an overturning effect on the remaining structure, thus the axial force of column B will increase dramatically from -1160 kN to -2100 kN, while the
axial force of column F will decrease from -600 kN to -340 kN. The axial forces of the other columns are nearly unchanged. Large variation of axial
force in beam AB is also observed, it develops from 10 kN to -90 kN . In general, the frame can offer enough alternate paths to bear the gravity load
although the exterior column A is removed. Removal of single column will not lead to instability of the frame.
The results for removing columns A & B are shown in Fig. 14 . Unlike the case of removing single column, the vertical displacement of upper node
of column B develops quickly (only 1.5s) to 3000 mm, which means the second floor nearly reaches the ground, and structure can be regraded as
collapsed. The overturning effect is much more serious than that of removing single column. The axial force of column C may develop to 3000 kN
in compression while that of column F may develop to 150 kN in tension. The axial forces of the beams AB and BC firstly increase in compression,
then gradually transit to tension, during which procedure bar fracture will occur and thus oscillation appears in the curves. From the results we can
conclude that the stability of the frame is lost and finally progressive collapse happened when double columns are removed.

0 0 50

−500
Displacement (mm)

−20
Axial force (kN)

Axial force (kN)

0
−1000
−40
−1500
−50
−60
−2000
B C D E F
−80 −2500 −100
0 2 4 6 8 10 0 5 10 0 2 4 6 8 10
Time (s) Time (s) Time (s)
(a) Node vertical displacement (b) Column axial forces (c) Beam AB axial force

FIGURE 13 Dynamic responses of the frame: removal of column A

0 1000 200

0
Displacement (mm)

−1000 100
Axial force (kN)

Axial force (kN)

−1000
−2000 0
−2000
−3000 −100
−3000
C D E F AB BC
−4000 −4000 −200
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5
Time (s) Time (s) Time (s)
(a) Node vertical displacement (b) Column axial forces (c) Beam axial forces

FIGURE 14 Dynamic responses of the frame: removal of columns A & B

16 D.C. FENG ET AL

5 INVESTIGATION OF TYPICAL PARAMETERS IN PRECAST RC STRUCTURES

To further study the performance of the numerical modeling approach in progressive collapse analysis of precast RC structures, parametric studies
are performed to investigate the influences of a variety of typical factors. It is worth noting at this juncture that many basic design parameters, inclu-
ding reinforcement ratio, beam depth, concrete strength, slab effect, boundary condition, etc., have been widely studied before. The present study
therefore mainly focuses on the numerical framework developed for precast structures. It should be also noted that we only perform removing of
exterior columns since the progressive collapse risk is relatively higher.

5.1 Influence of modeling strategies

First the Influence of modeling strategies is analyzed. Specifically, the considerations in modeling of the post-cast joint part, i.e., shear behavior of
the joint panel and bond-slip effect at the beam-column interface, are discussed, since this is the major part that differs from modeling of monolithic
RC structures. The results of pushdown analysis by removing column A, and columns A & B are shown in Fig. 15 . It is obvious that considering the
shear behavior of the joint panel or not in the modeling actually has little influences on the progressive collapse behavior of the frame. Actually
similar conclusion can be seen in research by Yu and Tan 16 . The joint panel is subjected to vertical translation and it is restrained by the connected
beams and columns, such that shear deformation is very little. However, bond-slip effect has a very significant effect on the progressive collapse
performance of the structure. The maximum load factor will increase from 1.9 to 2.25 for loss of column A, and 0.7 to 0.85 for loss of columns A & B
if bond-slip is neglected in the analysis. Bond-slip effect will cause fixed-end rotation at the beam ends, therefore, the responses will become softer
and the onset of bar fracture will be later than that of neglecting bond-slip.

2.5 1.0
Bar fracture occurs Bar fracture occurs

2.0 First peak 0.8

First peak
Load factor α

Load factor α

1.5 0.6

1.0 0.4

Bond-slip and shear Bond-slip and shear

0.5 Without bond-slip 0.2 Without bond-slip
Without shear Without shear
0.0 0.0
0 200 400 600 800 1000 1200 0 400 800 1200 1600
Vertical displacement (mm) Vertical displacement (mm)
(a) Removal of column A (b) Removal of columns A & B

FIGURE 15 Influence of modeling strategies on pushdown analysis

The dynamic responses of removing column A and columns A & B are also given in Figs. 16 -17 . Close to the conclusions drawn in pushdown
analysis, for either removing single column or double columns, the results of modeling with and without joint shear are nearly coincided with each
other, while the results of modeling with and without interface bond-slip vary very much. For removing single column A, the frame will remain stable
finally, but the nodal displacement time history and internal forces are much more stiffer. The responses will eventually damp out, but the rate of
damping is more fast if bond-slip is neglected. For removing double columns A & B, the frame finally collapsed. If neglecting bond-slip, the nodal
vertical displacement of the removed column B develops slower at first but then develops faster than the other models. This is because the bar
fracture will occur earlier if the fixed-end rotation is not considered, thus in this stage the displacement develops quickly. It is also depicted in the
internal forces time history. For model without bond-slip, the axial forces of beams and columns are at first larger but then drops smaller than the
other models, e.g., axial force of column C drops due to bar fracture at 0.8s for model with bond-slip effect while that drops at 0.6s for model without
bond-slip effect.
D.C. FENG ET AL 17

0 −1000 50
Bond−slip and shear Bond−slip and shear Bond−slip and shear
Without bond−slip Without bond−slip Without bond−slip
Displacement (mm)

Without shear Without shear Without shear

−20

Axial force (kN)

Axial force (kN)

−1500 0

−40

−2000 −50
−60

−80 −2500 −100

0 2 4 6 8 10 0 5 10 0 2 4 6 8 10
Time (s) Time (s) Time (s)
(a) Node vertical displacement (b) Column B axial force (c) Beam AB axial force

FIGURE 16 Influence of modeling strategies on dynamic responses: removal of column A

0 −1000 400
Bond−slip and shear Bond−slip and shear
Without bond−slip Without bond−slip
−1500
Displacement (mm)

Without shear Without shear

−1000 200
Axial force (kN)

Axial force (kN)

−2000
−2000 0
−2500
−3000 Bond−slip and shear −200
−3000
Without bond−slip
Without shear
−4000 −3500 −400
0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5
Time (s) Time (s) Time (s)
(a) Node vertical displacement (b) Column C axial force (c) Beam BC axial force

FIGURE 17 Influence of modeling strategies on dynamic responses: removal of columns A & B

5.2 Influence of solution algorithms

Secondly the influence of solution algorithms adopted in the modeling is studied, since they are greatly related to the accuracy and efficiency of the
analysis. Specifically, some pushdown analysis defines that the structure is collapsed if the analysis is non-converged. However, it is really hard to
identify whether the non-convergence is caused by spurious numerical errors or by stiffness loss and/or inability of the system to sustain the gravity
loading. In addition, the progressive collapse behavior is an extreme event, thus the dynamic responses analysis is also very difficult to perform using
implicit integration method, and the computational cost of using normal explicit integration method will be especially high.
In the static pushdown analysis, the performances of full Newton-Raphson method, Krylov-Newton method and the adopted consistent quasi-
Newton method are compared. The convergence tolerance is set as 10−6 on energy norm and the maximum iterative times for each step is set as
200. The numerical results are shown in Fig. 18 . Obviously, full Newton-Raphson method can hardly get converged in the progressive collapse
analysis, it diverges at 102 mm for single column removal and at 104 mm for double columns removal. The performance of Krylov-Newton method
and consistent quasi-Newton method are very similar in analysis of removing single column, both of them can capture the descending branch of
the pushdown curve, i.e., the collapse stage. However, for analysis of removing double columns, the Krylov-Newton method diverges at 1120 mm,
which is the end point of the ascending branch, while the consistent quasi-Newton method can still catch the collapse stage.
In the dynamic column-removal analysis, the performances of implicit Newmark method and the adopted explicit KR-α method are compared.
Time step is unified as 0.001s. The responses of nodal displacement and member axial force for loss of column A are shown in Fig. 19 . The results
by the two different methods are coincide with each other. However, the computing time is only 184s for explicit KR-α method while that is 304s
for implicit Newmark method. Fig. 20 shows the responses of the frame under removing columns A & B. Due to the instability of the frame, implicit
Newmark method diverges at 0.7s, which is not sufficient to study the progressive collapse mechanism of the structure. Explicit KR-α method can
compute to 1.5s, at which time the vertical displacement of the second floor reaches 3000 mm, and the structure can be regard as collapsed. In
summary, the explicit KR-α method enables us capture more information of the structure collapse process.
18 D.C. FENG ET AL

2.4 1.2

2.0 1.0
Krylov Newton diverges here
1.6 0.8
Load factor α

Load factor α
Full Newton diverges here
1.2 0.6
Full Newton diverges here

0.8 0.4
Consistent quasi-Newton Consistent quasi-Newton
0.4 Krylov Newton 0.2 Krylov Newton
Full Newton Full Newton
Full Newton
0.0 0.0
0 200 400 600 800 1000 1200 0 400 800 1200 1600
Vertical displacement (mm) Vertical displacement (mm)
(a) Removal of column A (b) Removal of columns A & B

FIGURE 18 Influence of static solution algorithms on pushdown analysis

0 50
KR−α explicit KR−α explicit
Newmark implicit Newmark implicit
Displacement (mm)

−20
Axial force (kN)

−40

−50
−60

−80 −100
0 2 4 6 8 10 0 2 4 6 8 10
Time (s) Time (s)
(a) Node vertical displacement (b) Beam AB axial force

FIGURE 19 Influence of dynamic solution algorithms on dynamic responses: removal of column A

5.3 Influence of joint variations

Finally the influence of the variations of the joint property is discussed, since the joint region (or beam-to-column connection region) in precast
RC structures is usually post-cast. Although the precast specimens are produced in factories thus the quality should be good and variations of the
property can be controlled to be very low, nevertheless, the quality of the post-cast concrete in joint region can be hardly guaranteed. Thus, great
variations may occur for the properties of the joint region. Note that we have studied that shear deformation is very limited in progressive collapse,
so the variations of shear behavior of the joint panel actually has very little influence on the overall performance. Consequently, the influence of the
variations of the interface springs reflecting bond-slip of the Joint2D element are studied.
A 15% coefficient of variation is assumed for the moment-rotation relation of the interface springs in analysis, and the generated mean value and
mean value ± 1 standard derivation (SD) curves are shown in Fig. 21 . The pushdown results using the three different curves are given in Fig. 22 . As
we can see, for both column removal scenarios, the pushdown curves have an approximately variation of 10%. Especially, the onset of bar fracture
is highly depended on the property of the interface spring. The higher of the moment-rotation relation, the earlier will the bar fractured, because
for the same curvature applied at the beam end, the section with higher moment will cause larger stress in the longitudinal bars.
The dynamic responses of the frame subjected to removal of column A and columns A & B also change with the interface spring properties, just
as shonw in Fig. 23 . The variation of the nodal displacement in case of loss of column A may develop to nearly 20%. However, at the axial force
level, the variation is much smaller. In a word, it is very important to consider influence of the variation of the post-cast joint property on the overall
progressive collapse behavior of precast RC structures.
D.C. FENG ET AL 19

0 200
KR−α explicit
Diverge point Newmark implicit

Displacement (mm)
−1000 100

Axial force (kN)

−2000 0

−3000 −100 Diverge point

KR−α explicit
Newmark implicit
−4000 −200
0 0.5 1 1.5 0 0.5 1 1.5
Time (s) Time (s)
(a) Node vertical displacement (b) Beam BC axial force

(c) Deformed shape by Newmark implicit at 0.7s (d) Deformed shape by KR-α explicit at 1.5s

FIGURE 20 Influence of dynamic solution algorithms on dynamic responses: removal of columns A & B

350

300

250
Moment (kN⋅m)

200

150

100
Mean value
Mean value+1SD
50 Mean value−1SD

0
0.000 0.005 0.010 0.015 0.020
-1
Curvature (mm )

FIGURE 21 Variation of the bond-slip spring properties

6 CONCLUSIONS

In this paper, a numerical framework for progressive collapse analysis of precast RC structures is developed, and a 10-storey precast RC frame is
designed and used to verify the framework and conduct parametric studies. Based on the numerical investigation, the following conclusions can be
drawn:
20 D.C. FENG ET AL

2.4 1.0

2.0
0.8

1.6
Load factor α

Load factor α
0.6
1.2
0.4
0.8
with mean value bond-slip spring with mean value bond-slip spring
with mean value+1SD bond-slip spring 0.2
0.4 with mean value+1SD bond-slip spring
with mean value−1SD bond-slip spring with mean value−1SD bond-slip spring
0.0 0.0
0 200 400 600 800 1000 1200 1400 0 400 800 1200 1600
Vertical displacement (mm) Vertical displacement (mm)
(a) Removal of column A (b) Removal of columns A & B

FIGURE 22 Influence of joint variation on pushdown analysis

0 50
with mean bond−slip spring with mean bond−slip spring
with mean+1SD bond−slip spring with mean+1SD bond−slip spring
−20
Displacement (mm)

with mean−1SD bond−slip spring with mean−1SD bond−slip spring

Axial force (kN)

0
−40

−60
−50
−80

−100 −100
0 2 4 6 8 10 0 2 4 6 8 10
Time (s) Time (s)
(a) Node vertical displacement (b) Beam BC axial force

0 400
with mean bond−slip spring
with mean+1SD bond−slip spring
Displacement (mm)

with mean−1SD bond−slip spring

−1000 200
Axial force (kN)

−2000 0

−3000 with mean bond−slip spring −200

with mean+1SD bond−slip spring
with mean−1SD bond−slip spring
−4000 −400
0 0.5 1 1.5 0 0.5 1 1.5
Time (s) Time (s)
(c) Node vertical displacement (d) Beam BC axial force

FIGURE 23 Influence of dynamic solution algorithms on dynamic responses: removal of columns A & B

• The developed numerical framework for progressive collapse analysis of precast RC structures involves three levels, i.e., material models,
element models and nonlinear solution algorithms. It can reflect the typical properties of precsat structures, i.e., the modeling of the post-
cast joint region, and can improve the numerical performance in both static and dynamic analysis procedures. The numerical example verifies
that the framework is very effective in both static pushdown analysis and dynamic column-removal analysis.

• Consideration of shear of the joint panel has little influence on the overall collapse behavior since the joint panel is restrained by the con-
nected beams and columns, thus the shear deformation is not significant. However, bond-slip at the beam-column interface affects the
D.C. FENG ET AL 21

progressive collapse behavior greatly. Neglecting bond-slip effect will overestimate the responses of the structure and cause earlier fracture
of the reinforcement bars.

• For static pushdown analysis, the adopted consistent quasi-Newton method can capture the collapse stage (descending stage), which can not
be captured by full Newton-Raphson method or Krylov-Newton method. For dynamic column-removal analysis, the adopted explicit KR-α
method is superior to implicit Newmark method in both efficiency and convergence issue.

• The variation of the joint region properties has an influence on the collapse performance of the total frame. A 15% variation on the moment-
rotation relation of the interface spring may lead to up to 20% variation on the nodal displacement response. However, the variation of the
axial forces is much smaller.

ACKNOWLEDGEMENTS

Financial supports from the Natural Science Foundation of Jiangsu Province (Grant No. BK20170680), the National Natural Science Foundation
of China (Grant Nos. 51708106, 51778130), the Fundamental Research Funds for the Central University, and the National Key Research and
Development Program of China (No. 2016YFC0701400) are greatly appreciated.

APPENDIX

A THE OPENSEES COMMAND LINE FOR KR-α ALGORITHM

The explicit KR-α algorithm is used in OpenSEE through the following command lines as:

- integrator KRAlphaExplicit $ρ∞

where $ρ∞ is the value of the high-frequency spectral radius, ranging from 1.0 (no numerical energy dissipation) to 0.0 (maximum numerical
energy dissipation).

Meanwhile, the lumped mass MUST be used and the following commands are required in company with the KR-α algorithm:

- algorithm Linear

- system FullGeneral

- analysis Transient

References

1. Chen Shicai, Yan Weiming, Gao Jie. Experimental investigation on the seismic performance of large-scale interior beam-column joints with
composite slab. Advances in Structural Engineering. 2012;15(7):1227-1237.

2. Im HyeongJu, Park HongGun, Eom TaeSung. Cyclic loading test for reinforced-concrete-emulated beam-column connection of precast concrete
moment frame. ACI Structural Journal. 2013;110(1):115-126.

3. Hawileh RA, Rahman A, Tabatabai H. Nonlinear finite element analysis and modeling of a precast hybrid beam–column connection subjected
to cyclic loads. Applied Mathematical Modelling. 2010;34(9):2562-2583.

4. Bahrami Saeed, Madhkhan Morteza, Shirmohammadi Fatemeh, Nazemi Nima. Behavior of two new moment resisting precast beam to column
connections subjected to lateral loading. Engineering Structures. 2017;132:808–821.

5. Kim Jinkoo, An Dawoon. Evaluation of progressive collapse potential of steel moment frames considering catenary action. The Structural Design
of Tall and Special Buildings. 2010;18(4):455-465.
22 D.C. FENG ET AL

6. Pachenari Alireza, Keramati Abolghassem, Pachenari Zahra. Investigation of progressive collapse in intermediate RC frame structures. The
Structural Design of Tall and Special Buildings. 2013;22(2):116-125.

7. Tsai Meng Hao, Huang Tsuei Chiang. Progressive collapse analysis of an RC building with exterior partially infilled walls. The Structural Design of
Tall and Special Buildings. 2013;22(4):327-348.

8. DoD . Unified facilities criteria: design of buildings to resist progressive collapse. UFC 4-023-03: Department of DefenseWashington DC, United
States; 2013.

9. GSA . Alternate path analysis & design guidelines for progressive collapse resistance. : General Service AdministrationWashington DC, United States;
2013.

10. Kang Shao-Bo, Tan Kang-Hai. Behaviour of precast concrete beam–column sub-assemblages subject to column removal. Engineering Structures.
2015;93:85–96.

11. Sasani Mehrdad, Kropelnicki Jesse. Progressive collapse analysis of an RC structure. The Structural Design of Tall and Special Buildings.
2008;17(4):757–771.

12. Yi Wei-Jian, He Qing-Feng, Xiao Yan, Kunnath Sashi K. Experimental study on progressive collapse-resistant behavior of reinforced concrete
frame structures. ACI Structural Journal. 2008;105(4):433-439.

13. Su Youpo, Tian Ying, Song Xiaosheng. Progressive collapse resistance of axially-restrained frame beams. ACI Structural Journal.
2009;106(5):600-607.

14. Yap Sim Lim, Li Bing. Experimental investigation of reinforced concrete exterior beam-column subassemblages for progressive collapse. ACI
Structural Journal. 2011;108(5):542-583.

15. Qian Kai, Li Bing. Performance of three-dimensional reinforced concrete beam-column substructures under loss of a corner column scenario.
Journal of Structural Engineering. 2012;139(4):584–594.

16. Yu Jun, Tan Kang-Hai. Experimental and numerical investigation on progressive collapse resistance of reinforced concrete beam column sub-
assemblages. Engineering Structures. 2013;55:90–106.

17. Ren Peiqi, Li Yi, Lu Xinzheng, Guan Hong, Zhou Yulong. Experimental investigation of progressive collapse resistance of one-way reinforced
concrete beam–slab substructures under a middle-column-removal scenario. Engineering Structures. 2016;118:28–40.

18. Lu Xinzheng, Lin Kaiqi, Li Yi, Guan Hong, Ren Peiqi, Zhou Yulong. Experimental investigation of RC beam-slab substructures against progressive
collapse subject to an edge-column-removal scenario. Engineering Structures. 2017;149:91-103.

19. Qian Kai, Li Bing, Zhang Zhongwen. Influence of Multicolumn Removal on the Behavior of RC Floors. Journal of Structural Engineering.
2016;142(5):04016006.

20. Sasani Mehrdad, Werner Andre, Kazemi Ali. Bar fracture modeling in progressive collapse analysis of reinforced concrete structures. Engineer-
ing Structures. 2011;33(2):401–409.

21. Qian Kai, Li Bing. Experimental and analytical assessment on RC interior beam-column subassemblages for progressive collapse. Journal of
Performance of Constructed Facilities. 2011;26(5):576–589.

22. Bao Yihai, Lew HS, Kunnath Sashi K. Modeling of reinforced concrete assemblies under column-removal scenario. Journal of Structural
Engineering. 2012;140(1):04013026.

23. Pham Anh-Tuan, Tan Kang-Hai, Yu Jun. Numerical investigations on static and dynamic responses of reinforced concrete sub-assemblages
under progressive collapse. Engineering Structures. 2017;149:2-20.

24. Li Yi, Lu Xinzheng, Guan Hong, Ye Lieping. An improved tie force method for progressive collapse resistance design of reinforced concrete
frame structures. Engineering Structures. 2011;33(10):2931–2942.

25. Brunesi E, Nascimbene R. Extreme response of reinforced concrete buildings through fiber force-based finite element analysis. Engineering
Structures. 2014;69:206–215.
D.C. FENG ET AL 23

26. Feng De-Cheng, Kolay Chinmoy, Ricles James M, Li Jie. Collapse simulation of reinforced concrete frame structures. The Structural Design of Tall
and Special Buildings. 2016;12(25):578–601.

27. Kang Shao Bo, Tan Kang Hai, Yang En Hua. Progressive collapse resistance of precast beamâĂŞcolumn sub-assemblages with engineered
cementitious composites. Engineering Structures. 2015;98:186-200.

28. Kang Shao Bo, Tan Kang Hai. Robustness Assessment of Exterior Precast Concrete Frames under Column Removal Scenarios. Journal of
Structural Engineering. 2016;142(12):04016131.

29. Scott Michael H., Fenves Gregory L.. Krylov Subspace Accelerated Newton Algorithm: Application to Dynamic Progressive Collapse Simulation
of Frames. Journal of Structural Engineering. 2015;136(5):473-480.

30. Kolay Chinmoy, Ricles James M. Development of a family of unconditionally stable explicit direct integration algorithms with controllable
numerical energy dissipation. Earthquake Engineering & Structural Dynamics. 2014;43(9):1361–1380.

31. Spacone Enrico, Filippou Filip C, Taucer Fabio F. Fibre beam-column model for non-linear analysis of RC frames: Part I. Formulation. Earthquake
Engineering & Structural Dynamics. 1996;25(7):711–726.

32. Coleman J, Spacone Enrico. Localization issues in force-based frame elements. Journal of Structural Engineering. 2001;127(11):1257–1265.

33. Wu Jian-Ying. New enriched finite elements with softening plastic hinges for the modeling of localized failure in beams. Computers & Structures.
2013;128:203–218.

34. Feng De-Cheng, Ren Xiaodan, Li Jie. Implicit Gradient Delocalization Method for Force-Based Frame Element. Journal of Structural Engineering.
2015;:04015122.

35. Scott Michael H, Fenves Gregory L. Plastic hinge integration methods for force-based beam-column elements. Journal of Structural Engineering.
2006;132(2):244–252.

36. Feng De-Cheng, Ren Xiao-Dan. Enriched Force-Based Frame Element with Evolutionary Plastic Hinge. Journal of Structural Engineering.
2017;143(10):06017005.

37. Paulay Thomas, Priestley MJ Nigel. Seismic design of reinforced concrete and masonry buildings. New York: Wiley; 1992.

38. Altoontash Arash. Simulation and damage models for performance assessment of reinforced concrete beam-column joints. PhD thesisStanford
University2004.

39. Lowes Laura N., Altoontash Arash. Modeling Reinforced-Concrete Beam-Column Joints Subjected to Cyclic Loading. Journal of Structural
Engineering. 2003;129(12):1686-1697.

40. Bao Yihai, Kunnath Sashi K, El-Tawil Sherif, Lew Hai S. Macromodel-based simulation of progressive collapse: RC frame structures. Journal of
Structural Engineering. 2008;134(7):1079–1091.

41. Vecchio Frank J, Collins Michael P. The modified compression-field theory for reinforced concrete elements subjected to shear. ACI Structural
Journal. 1986;83(2):219–231.

42. Hsu Thomas T C. Softened truss model theory for shear and torsion. ACI Structural Journal. 1988;85(6):624-635.

43. Wu Jian Ying, Li Jie, Faria Rui. An energy release rate-based plastic-damage model for concrete. International Journal of Solids and Structures.
2006;43(3-4):583–612.

44. Wu Jian-Ying, Cervera Miguel. A thermodynamically consistent plastic-damage framework for localized failure in quasi-brittle solids: Material
model and strain localization analysis. International Journal of Solids and Structures. 2016;88:227–247.

45. Wu Jian-Ying, Cervera Miguel. A novel positive/negative projection in energy norm for the damage modeling of quasi-brittle solids. International
Journal of Solids and Structures. 2018;139:250–269.

46. Feng De-Cheng, Li Jie. Stochastic Nonlinear Behavior of Reinforced Concrete Frames. II: Numerical Simulation. Journal of Structural Engineering.
2015;142(3):04015163.
24 D.C. FENG ET AL

47. Feng De-Cheng, Wu Gang, Sun Ze-Yang, Xu Ji-Gang. A flexure-shear Timoshenko fiber beam element based on softened damage-plasticity
model. Engineering Structures. 2017;140:483–497.

48. Stevens NJ, Uzumeri SM, Will GT, others . Constitutive model for reinforced concrete finite element analysis. ACI Structural Journal.
1991;88(1):49–59.

49. Mander John B, Priestley Michael JN, Park R. Theoretical stress-strain model for confined concrete. Journal of Structural Engineering.
1988;114(8):1804–1826.

50. Feng De-Cheng, Ding Zhao-Dong. Confined concrete model with strain gradient effect for RC columns under eccentric loading. Magazine of
Concrete Research. 2018;70(23):1–16.

51. Menegotto M.. Method of Analysis for Cyclically Loaded R. C. Plane Frames Including Changes in Geometry and Non-Elastic Behavior of
Elements under Combined Normal Force and Bending. In: :15-22; 1973.

52. Belarbi Abdeldjelil, Hsu Thomas T. C.. Constitutive Laws of Concrete in Tension and Reinforcing Bars Stiffened by Concrete. ACI Structural
Journal. 1994;91(4):465-474.

53. Feng De-Cheng, Wu Gang, Lu Yong. Numerical Investigation on the Progressive Collapse Behavior of Precast Reinforced Concrete Frame
Subassemblages. Journal of Performance of Constructed Facilities. 2018;32(3):04018027.

54. Feng De-Cheng, Xu Jun. An efficient fiber beam-column element considering flexure–shear interaction and anchorage bond-slip effect for cyclic
analysis of RC structures. Bulletin of Earthquake Engineering. 2018;:1–28.

55. Feng De-Cheng, Wu Gang, Lu Yong. Finite element modelling of precast reinforced concrete beam-to-column connections under cyclic loading.
submitted to Engineering Structures. 2018;.

56. Alsiwat Jaber M, Saatcioglu Murat. Reinforcement anchorage slip under monotonic loading. Journal of Structural Engineering.
1992;118(9):2421–2438.

57. Ren Xiaodan, Li Jie. Two-Level Consistent Secant Operators for Cyclic Loading of Structures. Journal of Engineering Mechanics.
2018;144(8):04018065.

58. Dennis Jr John E, Schnabel Robert B. Numerical methods for unconstrained optimization and nonlinear equations. SIAM; 1996.

59. Matthies Hermann, Strang Gilbert. The solution of nonlinear finite element equations. International Journal for Numerical Methods in Engineering.
1979;14(11):1613–1626.

60. Kolay Chinmoy, Ricles James M. Assessment of explicit and semi-explicit classes of model-based algorithms for direct integration in structural
dynamics. International Journal for Numerical Methods in Engineering. 2016;107(1):49–73.

61. Kolay Chinmoy, Ricles James M, Marullo Thomas M, Mahvashmohammadi Akbar, Sause Richard. Implementation and application of the uncon-
ditionally stable explicit parametrically dissipative KR-α method for real-time hybrid simulation. Earthquake Engineering & Structural Dynamics.
2015;44(5):735–755.

62. Code for design of concrete structures. Beijing: Ministry of Housing and Urban-Rural Development of the PeopleÂąÂŕs Republic of China; 2010.

63. Technical specification for precast concrete structures. Beijing: Ministry of Housing and Urban-Rural Development of the PeopleÂąÂŕs Republic of
China; 2014.

View publication stats