Engineering Structures 237 (2021) 112057

Contents lists available at ScienceDirect

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

Quantification of model uncertainties of the energy-based method for

dynamic column removal scenarios
Luchuan Ding *, Ruben Van Coile, Wouter Botte, Robby Caspeele *
Department of Structural Engineering and Building Materials, Ghent University, Technologiepark Zwijnaarde 60, 9052 Zwijnaarde, Ghent, Belgium

A R T I C L E I N F O A B S T R A C T

Keywords: The Alternative Load Path (ALP) method is widely used to assess progressive collapse resistance of reinforced
Model uncertainty concrete (RC) structures by notional removal of one or more load-bearing elements. In general, a nonlinear time
Energy-based method (EBM) history analysis (NTHA) is needed to perform such an analysis if dynamic effects are explicitly taken into ac­
Dynamic analysis
count. To avoid cumbersome nonlinear dynamic analyses, the energy-based method (EBM) is a promising
Reinforced concrete slab
Column removal scenario
technique to predict the maximum dynamic responses of a structural system. In this article, the accuracy and
Tensile membrane action precision of the EBM is evaluated based on a validated finite element model of a tested RC slab subjected to a
sudden column removal scenario, in particular in relation to the investigation of tensile membrane action (TMA).
Influences of dynamic effects are evaluated, i.e. in relation to strain rate effects, damping, and the time duration
of support removals. Strain rate effects are observed to have only slight influences on the dynamic responses. The
strain rate dependency of reinforcement is found to have a more significant influence on the responses in TMA
stage, although also to a limited extent. The magnitude of the load has a significant influence on the dynamic
response, as do increasing damping ratios due to the corresponding significant energy dissipation. Finally, the
dynamic response reduces with increasing time duration of the column removal. Based on the results of the
stochastic analyses, the EBM is observed to perform well based on a comparison with the results of NTHA in both
flexural and TMA stages. Furthermore, in relation to the analyzed case studies on reinforced concrete slabs, the
model uncertainty of the responses obtained through the EBM compared with the NTHA is found to be repre­
sented well by a lognormal distribution with mean of 0.95 and a standard deviation of 0.20, for evaluating the
loads of first rupture of reinforcement. Furthermore, a lognormal distribution with mean 0.96 and standard
deviation 0.13 is found appropriate to represent the model uncertainty on ultimate load-bearing capacity pre­
dictions. Model uncertainties are also obtained with respect to the model predictions for displacements at the
moment of the first rupture of reinforcement, displacements at the ultimate load-bearing capacities, and both
loads and displacements at second load peaks.

1. Introduction
Progressive collapses caused by extreme events are low-probability/
high-consequence (LPHC) phenomena [1–4]. The occurrence of extreme
events may cause local damage to structures and subsequently lead to
progressive collapse of the entire structure or large parts of them, such as
observed among others in the 1968 Ronan Point collapse in London
(UK), the 1995 Alfred P. Murrah Federal Building collapse in Oklahoma
(USA) [5–8], the 2008 collapse of a road bridge over an important
railway in Studenka (Czech Republic) [9] and the 2018 collapse of the
Morandi Bridge in Genoa (Italy) [10]. In general, the phenomenon of
progressive collapse or disproportionate collapse can be divided into
three stages, i.e. a local failure, the spread of the damage, and a final
collapse. Local failure occurs due to failure of one or more load-bearing
elements, most often triggered by abnormal loadings such as a gas ex­
plosion, a fire, a vehicle impact, a terrorist attack or seismic actions. The
initial damage may then progress throughout the structure or a large
part of it. In most cases, dynamic effects in those responses should be
taken into account to achieve realistic results for the assessment of the
final collapse scenario [11–14]. Note that in this article focus is put on
the evaluation of the performance of the energy-based method (EBM) in
relation to assessment of progressive or disproportionate collapse,
without distinguishing further between those.
At present much experimental and numerical research has been

* Corresponding authors.
E-mail addresses: Luchuan.Ding@UGent.be (L. Ding), Robby.Caspeele@UGent.be (R. Caspeele).

https://doi.org/10.1016/j.engstruct.2021.112057
Received 26 May 2020; Received in revised form 25 January 2021; Accepted 9 February 2021
0141-0296/© 2021 Elsevier Ltd. All rights reserved.
L. Ding et al.
carried out to investigate the dynamic behavior of building structures in
the context of progressive collapse. The studies have been done on sub-
assemblies of frames, flat slabs, and RC frames. Six RC substructures of
varying detailing and dimensions were tested by Qin and Li [15] to study
the dynamic performance of three-dimensional (3D) beam-column
substructures under the removal of a corner support. Two series of RC
beam-column substructures, with and without slabs, were tested in a
dynamic way by the sudden removal of a corner support to evaluate the
influence of the slab on the dynamic performance [16]. Moreover, the
aforementioned dynamic results [16] were compared with the push­
down performance [17] to further evaluate the dynamic effects. Russell
et al. [12] experimentally investigated the dynamic responses of RC flat
slabs after a sudden column loss, where seven RC slabs were tested in a
static way or in a dynamic way for different column removal scenarios.
After that, numerical dynamic column loss analyses of RC flat slabs were
performed [14] and the results were compared with equivalent static
cases in order to determine the dynamic amplification factor. As a more
severe initial damage scenario may occur whereby multiple columns are
removed, two RC flat slabs were tested by Qian et al. [18] and further
studied through numerical simulations to study the impact of two col­
umns missing on the dynamic response. Parisi and Scalvenzi [4]
numerically studied the dynamic performance of progressive collapse
for a five-story RC frame, considering both simultaneous and sequential
removal of ground-floor columns. A half-scale three-story RC frame
subjected to a series of sudden column loss was tested by Xiao et al [19].
Dynamic performance of a real-scale RC building was tested by Adam
et al. [20] under a corner-column removal scenario. In terms of different
analysis methods, Tsai and Lin [21] discussed the advantages and dis­
advantages of different analysis methods (linear static, linear dynamic,
nonlinear static and nonlinear dynamic approaches). Additionally,
experimental and numerical investigations were performed in relation
to the dynamic responses in e.g. [5,13,22–26].
Design methods against progressive collapse can be divided into two
groups in current standards, i.e. direct design methods and indirect
design methods. In direct design methods such as the Alternative Load
Path (ALP) method, the Enhanced Local Resistance (ELR) or the Specific
Local Resistance (SLR) method, the resistance against progressive
collapse is considered explicitly [27–29]. In case of indirect design
methods, the resistance is considered implicitly, as a structure is
designed to have minimum levels of strength, continuity and ductility
for example by means of the Tying Force (TF) method [27,28]. However,
it should be noted that the latter may be found to be insufficient to
reduce collapse risks [30]. The ALP method is a widely adopted method
to evaluate the potential of progressive collapse due to the fact that it
can represent a realistic structural response resulting from a local
damage [4,31]. Most often this method is implemented as a threat-
independent method meaning that it does not explicitly consider the
event that causes the initial damage, only its (conditional) conse­
quences. This improves the practical applicability due to the fact that it
is often difficult to identify and model the extreme events explicitly.
Dynamic effects (inertial effect, damping and strain rate effect etc.)
should be taken into account in the ALP method since the phenomenon
of progressive collapse is a dynamic event [4,20]. However, a dynamic
nonlinear time history analysis (NTHA) is computationally cumbersome
as both nonlinearities and dynamic effects need to be included. Although
the NTHA can provide a more accurate response prediction, it requires
analysts to have significant numerical expertise to perform the analysis
and has heavy computational demands. Such expertise is not commonly
acquired by practicing engineers [31,32]. Therefore, as an alternative, a
static analysis considering a dynamic increase factor (DIF) is often
employed [28,29], which can reduce the computational efforts without
performing the NTHA. However, the DIF depends both on the load level
and the nature of the nonlinear behavior [11,32] and the use of DIF is
associated with considerable model uncertainties as the empirical DIF
equations are typically obtained from a curve-fitting process [33,34].
The result may be highly conservative [35], leading to a considerable
2
Engineering Structures 237 (2021) 112057
bias and large stochastic variability. Furthermore, some effects, e.g. the
compressive or tensile membrane action, are not considered when
applying the DIF, although they have a significant effect on the
nonlinear dynamic response.
In order to tackle some of the concerns with the static-based
assessment method, the energy-based method has been developed as
an alternative approach [32]. The EBM is based on the principle of en­
ergy conservation, where the resistance against progressive collapse is
assessed based on the capability of absorbing and dissipating the energy
introduced by the unbalanced loads [11,32,34]. No dynamic analyses
are needed and the use of DIF is avoided [11,34,36]. In the early 1950′ s,
Newmark [37] formally described this concept for design in relation to
blast loading. This kind of approach has since then been further devel­
oped and used for column removal scenarios [21,23,32,36,38–44],
where good performance has been found for structural response pre­
dictions as obtained from the EBM in comparison to direct dynamic
response calculations.
Although the EBM is based on several simplifications that lead to an
approximate result, it is a compromise between accuracy and
complexity. The NTHA is not carried out in the EBM, since it only
consists of performing a static nonlinear pushdown analysis and subse­
quently predicts the maximum dynamic response through an energy
balance. Therefore, dynamic effects are not taken into account in the
EBM. However, the performance of the EBM at large deformation situ­
ations, e.g. tensile membrane action stage, yet needs to be evaluated. In
addition, the many uncertainties involved in structural analysis may
have a significant influence on the overall behavior, but most of the
existing studies neglect this [30,45]. Therefore, a probabilistic evalua­
tion of the performance of the EBM is necessary. Moreover, a quanti­
tative assessment of the model uncertainty in relation to the EBM is
lacking until now, although such information is crucial for reliability-
based assessments of progressive or disproportionate collapse sce­
narios in relation to structural robustness.
Considering the above mentioned overview of the state of the art in
this research field, the EBM can be a promising alternative approach to
perform a dynamic analysis (through predicting maximum dynamic
response) with significantly less calculation demand in the context of a
sudden column removal scenario. The influences of dynamic effects,
which are not taken into account in the EBM, need to be investigated to
evaluate the effectiveness of the EBM both in bending mechanism stage
(small deformation) and in tensile membrane action stage (large
deformation). Additionally, the model uncertainty in relation to the
EBM is lacking now but it is crucial for reliability-based assessment. This
article hence focuses on quantifying the model uncertainty of the EBM
approach for a RC slab subjected to a notional sudden support removal
scenario.
The sections in this article are organized as follows. Section 2 pre­
sents the methodology of the EBM. Following the validation of a nu­
merical model based on the experimental result in Section 3, the EBM is
then validated by comparing with the results of NTHA calculations in
Section 4. Section 4 concludes with a parametric study in order to
investigate the influence of dynamic effects in light of the considered
applications, which are taken into account in NTHA but not the EBM.
Further, the model uncertainty to be considered when using the EBM is
quantified in Section 5 and a probabilistic model is developed to be
taken into account in reliability-based analyses. Finally, concluding re­
marks are addressed in Section 6.
2. Methodology
2.1. Conservation of energy during collapse
Progressive collapse of structures is a dynamic process, for example
caused by the sudden removal of a column. The released potential en­
ergy resulting from the unbalanced gravity loads for the remaining
structural system leads to dynamic motions and an increase of kinetic
L. Ding et al. Engineering Structures 237 (2021) 112057

energy. In contrast, elastic and inelastic strain energy accumulated by ∫ us

1
the deformed structure dissipate the kinetic energy, counteracting the Qd (ud ) = Qs (u)du (1)
ud
motions. Neglecting energy dissipated by other sources such as heat, 0

throughout the structural movement three components of energy are

where Qd is the load in the dynamic load-deformation curve, Qs is the
taken into account when balancing the energy: release of potential en­
load in the static load-deformation curve, ud is the peak dynamic
ergy due to the downward movement of the structure, strain energy due
deflection and us is the static deflection corresponding to the load Qs.
to the structural deformation, and kinetic energy from the moving part.
Hence, the application of Eq. (1) leads to the following analytical pro­
Regarding the motion as a single-degree-of-freedom (SDOF) dynamic
cedure for the EBM in case it is applied in the context of notional
process, the change of the potential energy can be calculated by the
removal scenarios:
product of the weight of the moving part and the corresponding
displacement. The strain energy can be calculated by integrating the
a) Preload the damaged structure with self-weight or other external
product of change in stress and the corresponding change in strain over
loads;
the volume of the structural elements. Subsequently, the kinetic energy
b) Perform a static nonlinear pushdown analysis on the damaged
can be determined as the difference between the potential energy and
structure and record the output of the pushdown curve (Qs, us);
the strain energy, see Fig. 1a.
c) Convert the static pushdown curve into a dynamic load-bearing ca­
If the kinetic energy attains a value of zero, the potential energy can
pacity curve (Qd, ud) through Eq. (1).
be absorbed by the strain energy and the structure can redistribute the
unbalanced loads. Hence, collapse can be averted in this situation, see
3. Validation of numerical model
the curve 1 in Fig. 1a. Otherwise, a collapse will occur, i.e. the curve 2 in
Fig. 1a.
3.1. Finite element model
2.2. Principle of the energy-based method
A finite element model (FEM) is used to assess the performance of the
EBM in case of membrane action following column removal scenarios.
The EBM is based on the principle of conservation of energy.
The finite element model is validated based on an experimentally tested
Regarding the aforementioned energy balance during a typical collapse
slab subjected to large deformations. The real-scale slab was tested by
phenomenon, the EBM can be easily interpreted [11,23]. Consider a
Gouverneur et al. [46] to investigate the tensile membrane action after
structure which is subjected to a sudden column removal at a certain
the removal of a support. The slab specimen had two inner spans of 4 m
moment in time. Before the column is removed, the structure is in
and two outer spans of 3.15 m, i.e. having a total length of 14.3 m. The
equilibrium. Once the column is removed, potential energy (PE) is
width of the specimen was 1.8 m. The test set-up is illustrated in Fig. 2a.
released and transferred into kinetic energy (KE) and strain energy (SE).
The concrete was of class C30/37 as defined in EN 1992–1-1 [47], while
At the origin point in Fig. 1a, the potential energy starts to be released
the flexural reinforcement consisted of 16 bars of type S500 with a
along with the increase of downward displacement. In the meantime,
nominal diameter of 10 mm for both top and bottom reinforcing layers.
the structural system starts to move (kinetic energy) and deform (strain
The concrete cover was 20 mm. The material properties (mean values)
energy) in order to absorb the potential energy. As the kinetic energy is
are summarized in Table 1. Note that the material properties in Table 1
the difference between the released potential energy and the strain en­
are used for deterministic analyses, and uncertainties or probabilistic
ergy due to the principle of conservation of energy, the released po­
models with regard to the material properties are presented in Section
tential energy is equal to the strain energy at the moment when the
5.1. Only the inward movements were restrained by heavily reinforced
kinetic energy is zero. This enables to calculate the dynamic capacity
edge beams at two ends of the slab since the experiment aimed at
curve. The above relates to situations in which no collapse occurs, i.e.
investigating tensile membrane action only. Additional details can be
the curve 1 in Fig. 1a. The kinetic energy becomes zero at the point of
found in the related paper [46].
peak displacement, i.e. the intersection (KE = 0 in Fig. 1a) in the curve 1.
A 2D plane stress finite element model (FEM) of the one-way RC slab
To apply the EBM, the strain energy stored in the deformed system
is built using Abaqus 6.14. Considering the symmetry in both geometry
can be calculated through a static nonlinear pushdown analysis. In that
and loading, only one half of the slab is modelled. For the concrete, 4-
case, the strain energy is the hatched area under the static
node bilinear plane stress elements (CPS4R) are used. As cracks are
load–deflection curve up to a displacement us (Fig. 1b). Moreover, the
expected to occur all over the slab, 8 elements through the slab depth is
released potential energy generated by the unbalance loads is equal to
applied, i.e. a mesh size of 20 mm × 20 mm (Fig. 2b,c). The concrete
the hatched rectangular area for a constant load level Qd. Considering
damaged plasticity (CDP) model is employed, where the parabolic
that the internal stored strain energy and the released potential energy
stress–strain relationship and the Hordijk tensile softening model are
are equal at the same displacement level ud (us = ud), the dynamic load
used for the compressive and tensile behaviors respectively (see Fig. 3a,
Qd can be calculated as
b) [48]. Recommended parameters for the constitutive models are used
as specified in [48]. The softening branch in compression is based on

(a) Energy flow (b) Energy balance

Fig. 1. Concept of the energy-based method: (a) energy flow; and (b) energy balance. Figures redrafted and expanded after [11,23,36].

3
L. Ding et al. Engineering Structures 237 (2021) 112057

Fig. 2. Test set-up and FEM: (a) test set-up [46]; (b) Edge beam and lateral boundary condition; and (c) FEM in Abaqus.

relationship based on laboratory testing is employed, i.e. the strain

Table 1
hardening of steel is taken into account (Fig. 3c). A sudden decrease in
Material properties for concrete and reinforcement.
strength after rupture of the reinforcing bars is used in order to enable
Material Parameter Units Mean Formula the modelling of the observed failure phenomenon of the slab [6].
Value
Furthermore, the Poisson ratio for the reinforcement steel is taken as
Concrete Compressive strength MPa 36.2 – 0.3.
fc* Since in the real test set-up the inward movement of the edge beams
Tensile strength fct ** MPa 2.78 fct = 0.3⋅
is restricted, two connector elements (Fig. 2b) are employed in Abaqus
(fc − 8)2/3
to simulate springs simulating the relationship between the occurring
Elastic Young’s GPa 31.97 –
horizontal force and the corresponding inward displacement. Based on
modulus Eci*
Tensile fracture energy N/ 0.074 the measured membrane forces versus the corresponding displacements
Gf1 *** mm Gf1 = 0.03⋅ of the edge beams and anchorage blocks, the spring stiffness is
( )
fc 0.7 approximated as a constant value of 151.5kN/mm for each spring.
10
Compressive fracture N/ 18.46 Gfc = 250⋅Gf1
energy Gfc *** mm
3.2. Validation of the developed finite element model
Reinforcement Yield strength fy* MPa 555 –
Tensile strength ft* MPa 605 –
Ultimate strain εu* % 8.31 – To validate the numerical model, a loading scheme is applied ac­
Elastic Young’s GPa 207.9 – cording to the loading procedure of the actual test performed. The
modulus Es* loading scheme consists of three loading phases. Initially, the self-weight
Source: * Gouverneur et al. [46]; ** EN 1992 [47]; *** fib [49]. and a service load of 60kN are applied followed by the removal of the
service load. Subsequently, the middle support is removed, where the
compressive fracture energy according to the fib Model Code 2010 [49] two inner spans of 4 m then become one span of 8 m. Finally, a
and the fracture energy is divided by the equivalent length in order to displacement-controlled vertical load is imposed on the slab until the
reduce the sensitivity of mesh size. The tensile behavior is governed by failure of the slab. Abaqus/Standard is employed to perform the static
tensile strength and fracture energy, where the tensile strength is nonlinear pushdown analysis. Note that geometrical nonlinearity is
calculated according to EN 1992-1-1 [47], see Table 1. The Poisson ratio taken into account.
is take as 0.15. Finally, other default input values for the CDP model are Fig. 4 presents the vertical load–displacement curves obtained from
based on the Abaqus user manual [50]. both the test and the numerical pushdown analysis. Comparing with the
For the reinforcing steel, 2-node linear truss elements (T2D2) are experimental curve, good agreement is observed for the pushdown curve
applied. Perfect bond between the reinforcement elements and the until the first load peak. Similar as observed in the experiment [46], the
neighboring concrete elements is assumed. A multi-linear stress–strain slab experiences an elastic stage, an elastic–plastic stage, and a tensile
membrane action (TMA) stage. Little difference is found between the

4
L. Ding et al.
Fig. 3. Stress–strain relationships for (a) concrete in compression; (b) concrete in tension [48]; and (c) reinforcement.
Fig. 4. Vertical load – displacement relationship.
values of the first load peak, which are 158.1kN and 156.6kN, obtained
in the numerical analysis and the experiment respectively. This first load
peak corresponds to rupture of the reinforcement over one of the inner
support. The subsequent structural response is highly complex which is
reflected by a significant discrepancy between experiment and numer­
ical analysis. In the experiment the slab failed as the rupture of the
bottom layer of reinforcement, which is also predicted as such by the
numerical model, although with an overestimation of the corresponding
load. In case structural failure is defined as first rupture of reinforce­
ment, it can be concluded that the FEM has a good performance.
4. Assessment of the performance of the energy-based method
(EBM)
4.1. EBM versus dynamic nonlinear analysis
Based on the numerical pushdown curve presented in Section 3.2, a
dynamic load-bearing curve is calculated according to the EBM by using
Eq.(1). The result is presented in Fig. 4. It can be seen that the dynamic
load-bearing curve is more smooth than the static pushdown curve. The
ultimate dynamic load-bearing capacity is 100.4kN, compared to a
maximum load of 158.1kN in case of the static pushdown analysis. The
sudden decrease in strength occurring on the pushdown curve is not
observed on the dynamic capacity curve since the latter is calculated
from the former based on the energy balance. However, a slight soft­
ening stage on the dynamic capacity curve is observed between 80kN
and 100kN. The latter stage should be treated with care, as the real post-
peak behavior after the rupture of the top reinforcement is highly
complex and seems to be overestimated by the FEM analysis (see section
3.2), likely leading to also an overestimation in the range between 80kN
and 100kN.
As the curve following from the EBM is directly derived from the
pushdown curve without performing any dynamic analyses, the effec­
tiveness of the EBM should be validated by comparing it with results of
direct dynamic analyses. The incremental dynamic analysis (IDA)
5
Engineering Structures 237 (2021) 112057
technique is used here to execute the dynamic nonlinear time history
analysis (NTHA) on the same FE model. The IDA consists of performing a
series of dynamic analyses from a lower load value to a higher load value
until the collapse of the slab occurs.
A loading scheme with four phases is applied to every dynamic
analysis (Fig. 5). First, self-weight and vertical loads are applied in phase
1 from 0 to T1 (1.5 s). Next, the loads are kept the same from T1 to T2
(1.5 s − 2.0 s). Subsequently, the middle support is removed by linearly
reducing the force in the support in phase 3 in a time duration of 10-5s
from T2 to T3, where this short time duration is applied to simulate an
instantaneous removal scenario. Following T3, the structure oscillates
together with the external loads.
Abaqus/Explicit is employed to carry out the dynamic analyses to
avoid convergence problems. The stable time increment in the explicit
analysis in Abaqus is kept smaller than 10-6, allowing to model this short
removal time duration.
The results of the IDA are presented in Fig. 6a, where every star
represents a peak displacement of the dynamic response under the
corresponding vertical imposed load as shown in Fig. 6b. Good agree­
ment is observed between EBM and IDA, which indicates that the EBM
predicts the maximum dynamic displacement well. Moreover, the ulti­
mate dynamic load-bearing capacity is approximately 100.0kN, which
agrees well with the prediction through the EBM with a value of
100.4kN. For higher applied loads in the IDA, the structure fails in the
dynamic analysis after removal of the middle support (Fig. 6b). Tensile
membrane effect is observed after 50kN as shown in Fig. 6c, where no
horizontal force is found under 40kN but it occurs at 50kN and higher
(Fig. 6c). This can also be verified by the horizontal displacement in the
upper spring as presented in Fig. 6d, i.e. an inward displacement is
observed only when the loading level reaches 50kN. Note that an out­
ward residual displacement can be observed when the load level is
smaller than 50kN as no outward restraint is applied to the edge beam.
It is important to emphasize that the failure mode observed in the
dynamic case is different from the static case. For the latter, the top layer
of reinforcement over the inner support in the slab firstly ruptured when
the first load peak on the pushdown curve was reached. The slab how­
ever had a remaining load-bearing capability resulting from the bottom
Fig. 5. Loading scheme for dynamic analysis.
L. Ding et al.
(a) Comparison for EBM to IDA
(c) Horizontal force in the upper spring
Fig. 6. Response of dynamic analysis (IDA): (a) load–displacement relationship; (b) time history response of displacement; (c) horizontal force in the upper spring;
and (d) horizontal displacement in the upper spring.
layer of reinforcement until also the rupture of that reinforcement layer
occured [46]. For the dynamic case, the slab directly fails when the
imposed load reaches the ultimate load-bearing capacity of 101kN, i.e.
the two layers of reinforcement fail almost at the same time (i.e. going
from 100kN to 101kN, see Fig. 6).
It should be emphasized that different dynamic effects, i.e. the strain
rate effects, damping and removal duration of the mid-support, cannot
be taken into account in the EBM. Therefore, their influence is assessed
in the following sections.
4.2. Influence of strain rate
The EBM assumes that the strain energy absorbed by the system for a
giving displacement is the same for both dynamic and static responses.
However, in reality material properties of a system vary depending on
the deformation rate, i.e. the so-called strain rate dependency. Different
materials exhibit different strain rate properties, which are moreover
vary both in time and in space considering the dynamic response. There
properties are influenced by a large number of parameters, which makes
it difficult to evaluate their impact on a structural system. For RC
structures, the most critical parameters are the increase in yield stress of
the reinforcement and the increase in cracking stress for concrete. Here,
the recommendations in the fib Model Code [49] are used to account for
the increase of concrete cracking stress and the increase of the rein­
forcement yield stress and ultimate strength. The material properties are
adjusted by dynamic increase factors (DIFs), i.e. the ratios of dynamic
stresses to static stresses with regard to strain rates. For concrete, a two-
step model, distinguishing between strain rates lower and higher than
Engineering Structures 237 (2021) 112057
(b) Time history response of displacement
(d) Horizontal displacement in the upper spring
10 s− 1, is used as follows:
/ {
(ε̇ct /ε̇ct0 )0.018 for ε̇ct ⩽10s− 1
DIFt = ft fts = (2)
0.0062(ε̇ct /ε̇ct0 )1/3 for ε̇ct > 10s− 1
ε̇ct0 = 10 - 6 s− 1
(3)
where DIFt is the concrete tensile DIF at tensile strain rate ε̇ct . ft and fts
are the tensile strengths at an elevated strain rate ε̇ and at the quasi-
static condition, respectively. ε̇ct0 is the quasi-static strain rate with a
value of 10-6s− 1.
Considering the range of applications in this contribution, however,
applying a fixed value is prefered over the complex formulations (e.g.
Eq. (2)) and numerically much more efficient [14]. Although a
maximum stain rate can reach up to 2.06 s− 1, strain rates larger than 0.6
s− 1 are only observed in some elements in the top layer bar over the
inner support. Therefore, peak strain rates in the range of 0.01 s− 1 –0.6
s− 1 are expected for most of the elements. Consequently, two cases are
considered, i.e. with fixed DIFs of 1.2 and 1.3 applied to the cracking
stress for all concrete elements during the entire dynamic simulation.
Taking into account that the strain rate is smaller than 0.6 s− 1 for most
elements and this DIF will be applied to all elements and during the
entire analysis, the DIF value of 1.2 has been selected to avoid over­
estimating the strain rate effect by applying a fixed DIF according to
Russell et al. [14]. The latter value of 1.3 is based on the maximum strain
rate to determine the upper boundary of the response, i.e. 2.06 s− 1.
Actually, the results presented below show that the influence of strain
rate effects of concrete on the displacement response is minor and can
hence be adequately captured by this less cumbersome procedure.
6
L. Ding et al.
DIFs for hot rolled reinforcing steel [49,51], valid for strain rates up
to 10 s− 1, are calculated as follows:
( )
( )
DIFy = fy,d /fy,s = 1 + 6/fy,s ln ε̇/ε̇0 (4)
( )
( )
DIFu = fu,d /fu,s = 1 + 7/fu,s ln ε̇/ε̇0
ε̇0 = 5 × 10 - 5 s− 1
where DIFy and DIFu are the DIFs for the yield stress and the ultimate
strength of reinforcing steel, respectively. fy,d and fu,d are the yield stress
and the ultimate strength at an elevated strain rate ε̇, respectively. fy,s
and fu,s are the yield stress and the ultimate strength for a static condi­
tion, respectively. ε̇0 is the quasi-static strain rate with a value of 5 × 10-
5 − 1
s .
The influence of the strain rate effects is investigated based on the
FEM with the same loading scheme as in section 4.1 for IDA (with a
support removal duration of 10-5s). Six cases are investigated: (1)
without accounting for any strain rate effect (designated by ’Case 1’); (2)
accounting only for the strain rate effect of reinforcing steel (designated
by ’Case 2’); (3) accounting only for the strain rate effect of concrete
with a DIF of 1.2 (designated by ’Case 3’); (4) accounting for the strain
rate effects of both reinforcement and concrete with a DIF of 1.2
(designated by ’Case 4’); (5) accounting only for the strain rate effect of
concrete with a DIF of 1.3 (designated by ’Case 5’); and (6) accounting
for the strain rate effects of both reinforcement and concrete with a DIF
of 1.3 (designated by ’Case 6’). The results are shown in Fig. 7a. The
results agree well with the result of EBM, except for responses corre­
sponding to an imposed load higher than 80kN, where the rupture of
reinforcement results in softening of the EBM curve (as discussed in
section 4.1). Little difference is observed under small imposed loads. A
slight influence is observed in case the strain rate effect of steel is
considered (Case 2, Case 4 and Case 6). Although the differences are
limited, the influence of the strain rate effect of reinforcement rather
than concrete is higher under higher imposed loads (after 40kN), i.e.
when the reinforcement governs the structural response in the tensile
membrane stage.
Little influence of the strain rate effect of concrete is observed. A
similar conclusion is drawn by Russell et al. [14]. Note that four dis­
placements (Fig. 7a), i.e. corresponding to 90 kN in case of curve ’Case
3’, 60kN and 80kN in case of curve ’Case 5’, and 110kN in case of curve
’Case 6’ are considerably larger than those assumed for other strain rate
effect assumptions. Analysis of the simulations indicates this is due to
the rupture of the top layer reinforcement.
Differences for ultimate load-bearing capacities prove to be rather
insignificant except in case of the curve ’Case 6’ leading to a value of
130kN. However, for the latter case, it should be emphasized that the
(a) Vertical load – displacement
Fig. 7. Influence of strain rate effect: (a) load–displacement relationship; and (b) time-history response of strain rate and displacement.
Engineering Structures 237 (2021) 112057
fixed DIF of 1.3 may significantly overestimate the load-bearing
response as the strain rates for most elements are much smaller than
2.06 s− 1. The ultimate load-bearing capacities are 100kN, 104kN, 98kN,
116kN, 104kN and 130kN for the six cases (Table 2), respectively.
Comparing with the value of EBM, i.e. 100.4kN, the corresponding de­
viations are 0.40%, − 3.46%, 2.45%, − 13.45%, − 3.46%, and –22.77%
(5) (Table 2), respectively. As expected on the basis of previous observa­
tions, the deviations are slightly higher in case the strain rate effects of
both concrete and reinforcing steel are taken into account (curves ’Case
(6)
4’ and ’Case 6’).
Influences of the strain rate effects are hence considered not to be
significant, which corresponds to the observations by Russell et al. [14]
and Pham et al. [24]. This is due to the fact that large values of strain
rate only occur in limited areas on the structure and for only a short
duration. This can be observed in Fig. 7b, where a typical time-history
response of the strain rate at the element with maximum strain rate is
illustrated (for the case relating to the curve ’Case 1’ in Fig. 7a). By
comparing with the corresponding displacement under the vertical load,
it is shown that the maximum strain rate only occurs for a quite short
duration and reaches its peak value before the slab arrives to its peak
displacement, limiting the effect of the strain rate effect.
4.3. Influence of damping
In order to assess the influence of energy dissipated by damping,
viscous damping (Rayleigh damping) [52] is employed to model the
energy dissipation mechanism in the IDA. Considering to the first and
second natural periods (0.141 s and 0.025 s) of the structural system, the
mass proportional Rayleigh damping coefficient α and the stiffness
proportional Rayleigh damping coefficient β can be calculated from
ξi =
α
+
βωi
(7)
2ωi 2
where ξi is the damping ratio for the ith mode and ωi is the ith natural
frequency.
Based on common damping ratios in structural engineering, the
measured data from the experimental results of Russell et al. [12] and
the investigation by Xu and Ellingwood [32], damping ratios of 1%, 2%,
and 5% are investigated here. The results with different damping ratios
are presented in Fig. 8a. Although differences are not significant under
small loads, damping is observed to have a significant influence on the
dynamic response under high loads, which leads to significant larger
load-bearing capacities. Moreover, larger damping ratios result in larger
capacities, which becomes more apparent at higher load levels. Ultimate
dynamic capacities with damping are significantly higher than the ul­
timate capacity obtained based on the EBM or without damping. Note
that the situations both without and with strain rate effects of rein­
forcement have been investigated (designated by ‘WR’ and ‘R’,
(b) Time-history response (80kN)
7
L. Ding et al.
Table 2
Ultimate load-bearing capacity (Units: kN) and the corresponding deviation.
Case EBM Case 1 Case 2
Resistance (R) 100.4 100 104
Deviation* – +0.40% − 3.46%
*
The deviations are calculated as (REBM-RCase)/REBM.
Fig. 8. Influence of damping: (a) load–displacement relationship; and (b) time history response of displacement. (DR – damping ratio; WR – without strain rate
effect; R – with strain rate effect).
respectively).
The results confirm again that the EBM works well in the elastic
stage. However, the dynamic capacities can be significantly higher than
predicted by the EBM in the tensile membrane action stage. For a
commonly adopted damping ratio of 5% for RC structures, the obtained
dynamic load capacities are almost as high as the capacities obtained in
the quasi-static case and the ultimate dynamic load-bearing capacity is
almost two times larger as compared to the case of zero damping.
Furthermore, the peak displacement is observed to decay fast under a
lower load level for the damping ratio of 5%. For instance, Fig. 8b shows
the time-history response of displacement considering a load of 40kN
(without TMA as shown in section 4.1), where a poor oscillation is
oserved for damping ratio 5%. The peak displacements are 98 mm, 90.4
mm and 74.4 mm for the considered damping ratios of 1%, 2% and 5% .
Comparing with the peak displacement of 103.9 mm without damping,
deviations are 5.7%, 12.9% and 28.4%, respectively.
However, it should be emphasized that assessing the appropriate
damping ratio is not at all easy and highly controversial. Almost no test
results are available to assess the result for these extreme loading situ­
ations. Herraiz et al. [23] reported about a series of RC flat slabs sub­
jected to loss of corner columns and penultimate columns, where a good
agreement between the EBM and maximum dynamic response was
found for the latter case, i.e. the loss of penultimate columns. However,
this observation relates only to two datasets. On the other hand, a poor
agreement was observed in case of corner column removal scenarios
where the maximum dynamic responses were very similar to the static
ones for the same loading levels [23]. However, other reasons such as
incomplete similarity of the specimens, excessive support release times,
etc. were believed to lie at the basis of the large deviations, not larger
damping ratios. Moreover, it was reported for these tests by Russell et al.
[12] that the damping ratios for the residual vibration for the elastic
range were lower than 1%, while for the inelastic range damping ratios
were up to 24%. These data were extracted from the residual vibrations,
while maximum dynamic responses under sudden column removal
scenarios were reached almost at one half of the first oscillation, i.e.
8
Engineering Structures 237 (2021) 112057
Case 3 Case 4 Case 5 Case 6
98 116 104 130
+2.45% − 13.45% − 3.46% –22.77%
before the residual vibration. It is not clear if this damping ratio was
similar to the mean value before the peak respose was reached. If they
are similar, the use of Rayleigh damping is an incorrect assumption
when the damping ratio is larger than 10% [50], i.e. the responses under
large loading levels may not be reliable. This can be verified by the study
of Hall [53], where it is explicitly indicated that the damping forces can
become unrealistically large compared to the restoring forces, resulting
in an analysis being unconservative under a certain condition, e.g. a
nonlinear analysis with softening non-linearity.
Comparing with the total absorbed strain energy, energy dissipated
through damping and cracking during such a short period should not be
significant [23]. The investigation by Pham et al. [24] shows that the
EBM is suitable for the analysis of structures under a threat-independent
scenario of sudden column loss without considering damping. A
maximum deviation of almost 22% was observed for the ultimate dy­
namic load-bearing capacity if the critical damping was 5% in case of
extrem large deformation situation (780 mm) after catenary action was
fully developed. The deviation (22%) is however much smaller than the
value (almost 100% in case DR 5%) obtained here.
In addition, according to the investigation by Xu and Ellingwood
[32] where the stiffness-proportional damping was employed, a
maximum difference of 10% was observed in case of damping ratios of
5%.
In the previous analyses Rayleigh damping was applied, combining
mass-proportional damping (in case of lower frequency ranges) and
stiffness-proportional damping (in case of higher frequency ranges)
[50]. However, the natural frequency decreases when a higher load is
imposed due to the stiffness degradation [12], which may lead to an
overestimation of the dissipated energy by the mass-proportional
component. As such, also simulations considering only stiffness-
proportional damping (SPD) are executed and compared with the sim­
ulations using Rayleigh damping (RD). The results are presented in
Fig. 9a,b. It is observed that considering stiffness-proportional damping
leads to a considerable reduction of the load–displacement curve
compared to the simulations employing Rayleigh damping.
L. Ding et al. Engineering Structures 237 (2021) 112057

(a) Vertical load – displacement (b) Time-history response (40kN)

Fig. 9. Comparison of Rayleigh damping (RD) and stiffness proportional damping (SPD): (a) load–displacement relationship; and (b) time history response of
displacement.

However, considering results reported in literature, the results ob­
tained with stiffness-proporational damping are believed to lead to more
reasonable results. A further study is however neccessay to further
investigate the influence of damping, preferably supported by more
extensive experimental data, which is currently lacking.
Considering all these observations, it is clear that the type of
damping to apply as well as the values to consider are not univocal and
still subject to discussion when relating to the extreme deformations
associated with column removal scenarios in concrete structures. As
such, applying for the time being calculations without damping proves
to be a conservative approach and for those situations the EBM proves to
provide a reasonable estimation.
4.4. Influence of column removal duration
As the speed of support removal can affect the dynamic response, the
U.S. Department of Defense [28] stipulates that the duration of the
column removal needs to be less than one-tenth of the first natural
period of vertical oscillation. The natural period of the vertical vibration
approximates 0.141 s for the slab without the middle support (i.e. after
column removal). Based on previous investigations [14,15] on the in­
fluence of the support removal duration, a series of removal durations
are selected, i.e. 10-2ms (which is used in the previous analyses), 20 ms,
50 ms, and 100 ms, where the IDA is executed by linearly reducing the
force in middle support with these assigned durations T3 as shown in
Fig. 5. In order to investigate the influence of the removal time only,
both damping and strain rate effects are not considered in the dynamic
analyses here.

(a) Influence of removal duration (b) Time-history response under 30kN

(c) Time-history response under 90kN (d) Time-history response under 100kN
Fig. 10. Influence of support removal duration: (a) load–displacement relationship; (b) time history response of displacement under 30kN; (c) time history response
of displacement under 90kN; and (d) time history response of displacement under 100kN.

9
L. Ding et al.
Fig. 10a shows that a more abrupt removal (shorter removal dura­
tion) can result in a larger displacement. This can also be observed from
the dynamic time-history responses of the vertical displacements for
different removal durations, see Fig. 10b. Fig. 10a also shows that the
influence between different removal durations is insignificant when the
loads are smaller than 90kN, and moreover the results are close to those
predicted by the EBM. According to the guidelines of the DoD [28] the
removal duration for this particular case must be less than 14.1 ms (i.e.
one tenth of the first natural period), which is between 10-2 ms and 20
ms. The response of such a prescribed removal time is almost the same as
when considering a sudden column removal scenario and is also accu­
rately predicted by the EBM.
More significant differences are observed under higher load levels, i.
e. when the rupture of the top layer reinforcement results in larger
displacements, see Fig. 10a,c,d.
5. Quantification of the model uncertainty
5.1. Probabilistic modelling of input variables
Based on previous investigations [2,6,54–56,59], eight parameters
are selected as input variables to quantify the uncertainty propagation
when applying the EBM. The probabilistic models are presented in
Table 3. These parameters are the material properties of concrete, the
material properties of reinforcment, the cross-section of the reinforce­
ment and the stiffness of the horizontal springs (see Fig. 2b). The other
input parameters are considered deterministic.
Latin Hypercube Sampling (LHS) is used in combination with the
developed FEM to perform stochastic analyses. The advantage of LHS is
that it allows to limit the number of calculations to an acceptable
amount. As the standard LHS may bring undesired spurious correlation
into the sample scheme, correlation reducing Latin Hypercube Sampling
(CLHS) is used to avoid this unwanted effect [57]. 60 Latin-Hypercube
samples are generated based on the probabilistic models. Eventually,
each set of sampled realizations is used as an input for the FEM to
determine the responses obtained by the EBM and by IDA.
5.2. Stochastic analyses
All Latin-Hypercube samples are evaluated considering both static
pushdown analyses and direct dynamic analyses. For the former, only
one pushdown analysis is executed for every realization. For the latter,
the IDA technique is used for every slab realization. Firstly, a load in­
terval of 10kN is employed from 40kN to 160kN, i.e. 13 dynamic sim­
ulations. Next, to obtain a 2kN resolution for the maximum dynamic
load, another 4 simulations are executed for every 2kN in the interval
between the last non-failed simulation and the first failed simulation.
Subsequently, the curves from the pushdown analyses are converted into
predicted dynamic capacity curves according to the energy-based
method (EBM) and compared with the results of direct dynamic ana­
lyses (IDA). Note that strain rate effects and damping are not taken into
account (considering the analysis and observations from section 4). A
Table 3
Probabilistic models for the random variables.
Name Units Distribution Mean
Density of concrete ρc kg/m3 N 2400
Concrete compressive strength fc MPa LN 36.2
Reinforcement yield stress fy MPa N 555
Reinforcement tensile strength ft MPa LN 605
Reinforcement strain at maximum % LN 8.3
stress εu
Young’s modulus of reinforcement Es GPa N 207.9
Cross-section of reinforcement As mm2 N 1256
Stiffness of horizontal spring k kN/ LN 151.5
mm
Engineering Structures 237 (2021) 112057
support removal duration of 10-2ms is adopted here for the IDA.
Furthermore, it needs to be mentioned that the numerical results deviate
from the observed experimental data after the first load peak is reached,
but this does not prevent to quantify the model uncertainty of the EBM
compared to direct dynamic analyses as performed here. Nevertheless,
this indicates the importance to quantify also a model uncertainty with
respect to the prediction of the highly nonlinear post-peak behavior in
case of large deformations and membrane actions, for which however at
this stage only very limited data is available.
Fig. 11 shows the results of static pushdown analyses (designated by
the curve ‘pushdown’), the results of EBM (designated by ‘EBM’) and the
results of the direct dynamic analyses (designated by ‘IDA’) as well as its
maximum load value (designated by ‘Max-IDA’), and this for every
realization of the Latin-Hypercube sampling. It is observed that the re­
sponses are highly sensitive to the realisations of the variables. However,
good agreement is observed between EBM and IDA in almost all cases.
The EBM curves are derived from the pushdown curves, i.e. the EBM
curves depend on the pushdown curves and are calculated on the basis of
conservation of energy. The results of the EBM evaluation corresponding
to the stage before the first load peak (the rupture of top reinforcement)
in the pushdown curve are almost identical to the results of IDA for all
the realizations (Fig. 11). Note that the first load-peaks for IDA in cases
No. 22 and No. 35 have been calculated separately and both values are
30kN (the first load-peak for IDA is lower than 40kN). However, notable
discrepancies are found between EBM and IDA in the stage after the
aforementioned first load-peak. This can be explained by the fact that
rupture of the top reinforcement causes a sudden decrease of load-
bearing capacity in the pushdown curve (i.e. the first load-peak). The
EBM curve is affected by this sudden decrease of the pushdown curve.
For instance, the rupture of top reinforcement may occur at different
stages in the pushdown curves, see e.g. No. 6, No. 7, and No. 8. The
rupture of the top reinforcement occurs much earlier in case No. 8
compared to cases No. 6 and No. 7. Moreover, the relative early rupture
of top reinforcement in case No. 8 even causes the EBM curve to
decrease. The deviation between the EBM and the IDA in case No. 8 is
therefore more significant than in the other two cases.
Those cases that show a clear difference between the two approaches
relate to differences in predictions occurring after the first peak of the
static pushdown analysis, e.g. leading to an immediate follow-up failure
of the second reinforcement layer in the IDA (e.g. No. 17 and No. 26),
whereas this is not predicted as such by the EBM. Two situations are
observed for the pushdown curves. The first situation indicates two load
peaks, where the second load peak is the ultimate load-bearing capacity,
and the smaller first load peak is due to the rupture of the top layer
reinforcement over the inner support (e.g. No. 2 and No. 45). The other
situation is where the first load peak has a larger value than the second
load peak (e.g. No. 3 and No. 51). As the dynamic load-bearing curves of
EBM completely depend on the pushdown curves, the two situations
result in different ultimate capacities for EBM if different failure criteria
for the pushdown curve are taken into account.
As a conservative approach, failure can be defined as the rupture of
one reinforcement layer (i.e. the top reinforcement layer in the case
under investigation) although the slab can in some cases still take a
higher load. This failure criterion is hence associated with the appear­
ance of the first load peak in the pushdown curve and the rupture of the
top layer of the reinforcement in the IDA curve. Note that even if the
COV second load peak is lower in some cases, it results in an additional
0.04 branch in the EBM curve.
0.10 In case the failure criterion relates to the first load peak (associated
0.03 with rupture of the top reinforcement layer), the results of the load-
0.03 bearing capacities are illustrated in Fig. 12a for all Latin-Hypercube
0.15
simulations. Again, it is observed that the EBM results correspond well
0.08 to the results obtained through the IDA, both with respect to the loads
0.02 corresponding to the first load peaks as well as for the associated dis­
0.25 placements (see Fig. 12b and Fig. 12c, respectively).
In case a quantification of the ultimate load-bearing capacity is
10
L. Ding et al. Engineering Structures 237 (2021) 112057

Fig. 11. Comparison for the results of stochastic analyses for every realization.

11
L. Ding et al. Engineering Structures 237 (2021) 112057

Fig. 11. (continued).

envisaged (regardless of whether this is occurring as a result of a post- load-bearing capacity. Fig. 13b,c present the comparison between the
peak behavior associated with larger displacements and a complex EBM and IDA for ultimate load-bearing capacities as well as the dis­
stress redistribution), Fig. 13a represents the results of this ultimate placements corresponding to them, respectively. Again, good agreement

12
L. Ding et al. Engineering Structures 237 (2021) 112057

(a) Load vs displacement (b) Comparison of load (b) Comparison of displacement

Fig. 12. Comparison of the first load peak: (a) load–displacement relationship; (b) comparison of load corresponding to the rupture of one reinforcement layer; and
(c) comparison of the associated displacement.

(a) Load vs displacement (b) Comparison of load (b) Comparison of displacement

Fig. 13. Comparison of the ultimate load-bearing capacity: (a) load–displacement relationship; (b) comparison of load; and (c) comparison of displacement.

is found for the prediction of the ultimate load-bearing capacity by the

PEBM DEBM
EBM, although the performance is less when assessing the associated R= or (8)
PIDA DIDA
displacements.
Finally, for completeness, also the predictability of the load-bearing where R is the ratio of EBM/IDA, PEBM is the load-bearing capacity
capacity and associated displacements by the EBM in relation to the following from the EBM calculations, PIDA is the load-bearing capacity
second load peak in the pushdown analysis is analysed (i.e. relating to following from the IDA calculations; DEBM is the displacement calculated
the rupture of both reinforcement layers), see Fig. 14a. Although the according to the EBM, DIDA is the displacement calculated according to
second load peaks may be lower than the highest load peaks, the dy­ the IDA. As mentioned before, these values can be related to load-
namic ultimate capacities of EBM based on the pushdown curves will be bearing capacity associated to the rupture of one reinforcement layer,
larger in these cases (Fig. 14a). The comparison of the loads associated the ultimate load-bearing capacity (as the maximal load-bearing
to this failure criterion and the corresponding ultimate displacements
are presented in Fig. 14b,c, respectively. Again, reasonably good
agreements are observed for the EBM, when comparing to the IDA Table 4
Ratios of EBM/IDA.
results.
Case R [-]
Mean (µ) Standard deviation (σ)
5.3. Quantification of the model uncertainty of EBM versus IDA analyses
PEBM First peak 0.95 0.20
R =
PIDA Ultimate load-bearing capacity 0.96 0.13
To evaluate the performance of the EBM quantitatively, ratios of the
Second peak 0.98 0.13
loads of the EBM to those from the direct dynamic analysis (IDA), see DEBM First peak 1.02 0.38
R =
Fig. 12b, Fig. 13b and Fig. 14b are calculated using Eq.(8). Furthemore, DIDA Ultimate load-bearing capacity 1.00 0.15
ratios of the displacements, see Fig. 12c, Fig. 13c and Fig. 14c are also Second peak 1.03 0.13
calculated.

(a) Load vs displacement (b) Comparison of load (b) Comparison of displacement

Fig. 14. Comparison of the second load peak (ultimate displacement): (a) load–displacement relationship; (b) comparison of load; and (c) comparison of
displacement.

13
L. Ding et al.
capacity observed, regardless of whether or not this relates to a post-
peak behavior after rupture of the first reinforcement layer) or the
load-bearing capacity associated to the rupture of both reinforcement
layers. The results of the ratios R are shown in Table 4.
Further, histograms of the ratios are presented in Fig. 15. Fig. 15a
shows ratios of the loads of the first load peaks, where a lognormal
distribution LN(0.95, 0.20) is found to represent the histogram, while a
LN(1.02,0.38) is fitted for the displacements of the first peaks as pre­
sented in Fig. 15b. Fig. 15c presents the ratios of the loads of the ultimate
load-bearing capacity, where a lognormal distribution LN(0.96, 0.13) is
used to fit the probability density function (PDF), while a LN(1.00,0.15)
is considered for the displacements of the ultimate load-bearing peaks as
shown in Fig. 15d. Fig. 15e,f show the ratios for the loads and the dis­
placements of the second load peaks respectively, where a lognormal
distribution LN(0.98, 0.13) for the loads and a lognormal distribution LN
(1.03,0.13) for displacements are found to fit the histograms,
respectively.
6. Conclusions
To avoid cumbersome nonlinear dynamic analyses, the energy-based
method (EBM) is a promising technique to predict the maximum dy­
namic responses of a structural system on the basis of a pushdown
analysis. In this article, the accuracy and precision of the EBM is eval­
uated based on a validated finite element model of a tested RC slab
subjected to a sudden column removal scenario, in particular in relation
to the investigation of tensile membrane action (TMA). After validation
against experimental results, the model is employed to verify the accu­
racy and precision of the EBM in comparison to incremental dynamic
analyses (IDA). The analyses indicate that the simple EBM efficiently
predicts the maximum dynamic response for a concrete slab, even in the
tensile membrane stage. Note that the FE model is only validated up to
the first load peak and should be further investigated by experimental
data when this becomes available in the future. The post-peak behavior
after the rupture of the top reinforcement is highly complex and seems to
be overestimated as indicated in sections 3.2 and 4.1. However, the
performance of the EBM as an approximation of the NTHA appears to be
very good also in this range, although for very large displacements with
reinforcement rupture the FE model itself is not validated and the results
need to be treated with care.
(a) Load (first peak) (b) Displacement (first peak)
(d) Displacement (highest peak) (e) Load (second peak) (f) Displacement (second peak)
Fig. 15. Histograms and PDFs of the ratios of EBM/IDA: (a) load corresponding to first peak; (b) displacement of first peak; (c) load corresponding to ultimate peak;
(d) displacement of ultimate peak; (e) load corresponding to second peak; and (f) displacement of second peak.
14
Engineering Structures 237 (2021) 112057
As the EBM neglects the influence of specific dynamic effects, such as
strain rate effects, damping, and support removal duration, their effect
on the dynamic capacity and the EBM’s approximation is evaluated.
Next, stochastic analyses were executed considering eight stochastic
input variables using Latin Hypercube sampling. The results of EBM are
compared with the results of direct dynamic analyses to assess the
performance of EBM in a probabilistic way. On the basis of these sim­
ulations, probabilistic models have been proposed for the model un­
certainty of EBM compared to direct dynamic analyses, in particular in
relation to the prediction of the loads and displacements associated to
the first rupture of reinforcement, the ultimate load-bearing capacity
(regardless whether or not this is the result of a post-peak behavior after
the rupture of the first reinforcement layer) and the rupture of the sec­
ond reinforcement layer. The most important findings are summarized
as follows:
(1) Based on the validated FEM, results of both EBM and IDA are
obtained. A good performance in both flexural stage and tensile
membrane action stage is found for the EBM when comparing
with the IDA if damping and strain rate effects are not taken into
account.
(2) The strain rate effect is found to have limited influence on the
dynamic response in sudden support removal scenarios. The
occurring strain rates of most finite elements are in general small
and only localized elements experience large strain rates. The
influence of the strain rate effect of reinforcement is slightly more
significant than that of concrete since the tensile membrane stage
resistance is heavily influenced by the capacity of the reinforcing
steel.
(3) Damping has a slight influence in the elastic stage. However, a
significant influence on the dynamic response under a high load
and large damping ratio is observed, since the energy dissipation
leads to a larger capacity. It is however difficult to say that
viscous damping (Rayleigh damping) is suitable in case of these
extremely damaged states [53,58]. In addition, it is not clear
whether the damping mechanism for the residual oscillation is
similar to that before the peak response is reached [23]. More­
over, different damping mechanisms lead to different results and
it is difficult to say which one is more realistic as there is no
experimental result to compare with. Hence, damping needs to be
(c) Load (highest peak)
L. Ding et al.
further investigated, preferably in relation to experimental re­
sults which currently are lacking in relation to the applications
under consideration. Neglecting damping however leads to a
conservative estimation of the load-bearing capacity and the
associated displacements, and this is moreover predicted well
also by the simplified EBM calculations.
(4) Good agreement was found between EBM and direct dynamic
analyses. Furthermore, the model uncertainty of the loads at the
first load peaks obtained through the EBM compared to direct
dynamic analysis is found to be represented well by a lognormal
distribution with mean of 0.95 and a standard deviation of 0.20,
while for the ultimate load-bearing capacities, a lognormal dis­
tribution with mean of 0.96 and a standard deviation of 0.13 is
obtained. When comparing the associated displacements, the
model uncertainty for displacements at the first load peaks is
found to be represented well by a lognormal distribution with
mean of 1.02 and a standard deviation of 0.38, while for the
displacements at the ultimate load-bearing capacities, a
lognormal distribution with mean 1.00 and standard deviation of
0.15 is found to be appropriate.
Although the EBM is validated and its model uncertainty is evaluated
as well in this contribution, it is necessary to emphasize that the results
are based on a single structural sub-assemblage, i.e. an RC slab. The
performance of the EBM and the associated model uncertainty in rela­
tion to the assessment of dynamic column removal scenarios still need to
be investigated for other types of structures, e.g. frames or entire 3D
structural systems.
Declaration of Competing Interest
The authors declared that there is no conflict of interest.
Acknowledgements
This research has been made possible through funding from the
China Scholarship Council (No. 201706260259) for the first author,
which is gratefully acknowledged.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.
org/10.1016/j.engstruct.2021.112057.
References
[1] Ellingwood BR. Mitigating risk from abnormal loads and progressive collapse.
J Perform Constr Facil 2006;20:315–23.
[2] Feng DC, Xie SC, Xu J, Qian K. Robustness quantification of reinforced concrete
structures subjected to progressive collapse via the probability density evolution
method. Eng Struct 2020;202. 109877.
[3] Weng YH, Qian K, Fu F, Fang Q. Numerical investigation on load redistribution
capacity of flat slab substructures to resist progressive collapse. J Building Eng
2020;29. 101109.
[4] Parisi F, Scalvenzi M. Progressive collapse assessment of gravity-load designed
European RC buildings under multi-column loss scenarios. Eng Struct 2019;
110001.
[5] Adam JM, Parisi F, Sagaseta J, Lu X. Research and practice on progressive collapse
and robustness of building structures in the 21st century. Eng Struct 2018;173:
122–49.
[6] Droogné D, Botte W, Caspeele R. A multilevel calculation scheme for risk-based
robustness quantification of reinforced concrete frames. Eng Struct 2018;160:
56–70.
[7] Ding L, Botte W, Van Coile R, Caspeele R. Robustness-evaluation of a stochastic
dynamic system and the instant equivalent extreme-value event: The PDEM-based
structural reliability evaluation of a dynamic system. Beton-und Stahlbetonbau
2018;113:33–7.
[8] Feng DC, Wang Z, Wu G. Progressive collapse performance analysis of precast
reinforced concrete structures. Structural Design Tall Special Buildings 2019;28.
e1588.
[9] Agarwal J, Haberland M, Holický M, Sykora M, Thelandersson S. Robustness of
structures: Lessons from failures. Structural Eng Int 2012;22:105–11.
15
Engineering Structures 237 (2021) 112057
[10] Malomo D, Scattarreggia N, Orgnoni A, Pinho R, Moratti M, Calvi GM. Numerical
study on the collapse of the Morandi bridge. J Perform Constr Facil 2020;34:
04020044.
[11] Izzuddin B, Vlassis A, Elghazouli A, Nethercot D. Progressive collapse of multi-
storey buildings due to sudden column loss—Part I: Simplified assessment
framework. Eng Struct 2008;30:1308–18.
[12] Russell J, Owen J, Hajirasouliha I. Experimental investigation on the dynamic
response of RC flat slabs after a sudden column loss. Eng Struct 2015;99:28–41.
[13] Feng DC, Wu G, Lu Y. Numerical investigation on the progressive collapse behavior
of precast reinforced concrete frame subassemblages. J Perform Constr Facil 2018;
32:04018027.
[14] Russell J, Owen J, Hajirasouliha I. Dynamic column loss analysis of reinforced
concrete flat slabs. Eng Struct 2019;198. 109453.
[15] Qian K, Li B. Dynamic performance of RC beam-column substructures under the
scenario of the loss of a corner column—Experimental results. Eng Struct 2012;42:
154–67.
[16] Qian K, Li B. Quantification of slab influences on the dynamic performance of RC
frames against progressive collapse. J Perform Constr Facil 2015;29:04014029.
[17] Qian K, Li B. Slab effects on response of reinforced concrete substructures after loss
of corner column. ACI Struct J 2012;109:845–55.
[18] Qian K, Weng Y-H, Li B. Impact of two columns missing on dynamic response of RC
flat slab structures. Eng Struct 2018;177:598–615.
[19] Xiao Y, Kunnath S, Li FW, Zhao YB, Lew HS, Bao Y. Collapse test of three-story half-
scale reinforced concrete frame bsuilding. ACI Struct J 2015;112.
[20] Adam JM, Buitrago M, Bertolesi E, Sagaseta J, Moragues JJ. Dynamic performance
of a real-scale reinforced concrete building test under a corner-column failure
scenario. Eng Struct 2020;210. 110414.
[21] Tsai MH, Lin BH. Investigation of progressive collapse resistance and inelastic
response for an earthquake-resistant RC building subjected to column failure. Eng
Struct 2008;30:3619–28.
[22] Tsai MH. An analytical methodology for the dynamic amplification factor in
progressive collapse evaluation of building structures. Mech Res Commun 2010;37:
61–6.
[23] Herraiz B, Russell J, Vogel T. Energy-based method for sudden column failure
scenarios: theoretical, numerical and experimental analysis. IABSE Symposium
Report: International Association for Bridge and Structural Engineering; 2015. p.
70-7.
[24] Pham AT, Tan KH, Yu J. Numerical investigations on static and dynamic responses
of reinforced concrete sub-assemblages under progressive collapse. Eng Struct
2017;149:2–20.
[25] Peng Z, Orton SL, Liu J, Tian Y. Experimental study of dynamic progressive
collapse in flat-plate buildings subjected to an interior column removal. J Struct
Eng 2018;144:04018094.
[26] Parisi F, Scalvenzi M, Brunesi E. Performance limit states for progressive collapse
analysis of reinforced concrete framed buildings. Structural Concrete. 2019;20:
68–84.
[27] EN1991-1-7. Eurocode 1: Actions on structures. Part 1.7: General
actions—Accidental actions. Brussels, Belgium: Comité Européen de
Normalisation; 2006.
[28] DoD. Unified facilities criteria: design of buildings to resist progressive collapse.
UFC 4-023-03. United States Department of Defense; 2009.
[29] GSA. Alternate path analysis & design guidelines for progressive collapse
resistance. General Services Adm 2016.
[30] Song X. Reliability analysis of code-conforming steel-frame structures against
column-loss scenarios. J Struct Eng 2020;146:04020025.
[31] Subki NEA, Mansor H, Hamid YS, Parke G. Progressive collapse assessment: a
review of the current energy-based Alternate Load Path (ALP) method. MATEC
Web of Conferences: EDP Sciences 2019:02012.
[32] Xu G, Ellingwood BR. An energy-based partial pushdown analysis procedure for
assessment of disproportionate collapse potential. J Constr Steel Res 2011;67:
547–55.
[33] Liu M. Pulldown analysis for progressive collapse assessment. J Perform Constr
Facil 2013;29:04014027.
[34] Liu MM, Pirmoz A. Energy-based pulldown analysis for assessing the progressive
collapse potential of steel frame buildings. Eng Struct 2016;123:372–8.
[35] Izzuddin BA, Nethercot DA. Design-oriented approaches for progressive collapse
assessment: load-factor vs. ductility-centred methods. Structures Congress 2009:
Don’t Mess with Structural Engineers: Expanding Our Role 2009:1–10.
[36] Dusenberry DO, Hamburger RO. Practical means for energy-based analyses of
disproportionate collapse potential. J Perform Constr Facil 2006;20:336–48.
[37] Newmark NM. An engineering approach to blast resistant design. New York:
American Society of Civil Engineers; 1953.
[38] Powell G. Progressive collapse: case studies using nonlinear analysis. Structures
Congress 2005: Metropolis Beyond 2005:1–14.
[39] Szyniszewski S, Krauthammer T. Energy flow in progressive collapse of steel
framed buildings. Eng Struct 2012;42:142–53.
[40] Main JA. Composite floor systems under column loss: collapse resistance and tie
force requirements. J Struct Eng 2014;140:A4014003.
[41] Kim S, Lee C-H, Lee K. Effects of floor slab on progressive collapse resistance of
steel moment frames. J Constr Steel Res 2015;110:182–90.
[42] Bao Y, Main JA, Noh S-Y. Evaluation of structural robustness against column loss:
Methodology and application to RC frame buildings. J Struct Eng 2017;143:
04017066.
[43] Yu XH, Qian K, Lu DG, Li B. Progressive collapse behavior of aging reinforced
concrete structures considering corrosion effects. J Perform Constr Facil 2017;31:
04017009.
L. Ding et al.
[44] Naji A. Progressive collapse analysis of steel moment frames: an energy-based
method and explicit expressions for capacity curves. J Perform Constr Facil 2019;
33:04019008.
[45] Feng DC, Xie SC, Deng WN, Ding ZD. Probabilistic failure analysis of reinforced
concrete beam-column sub-assemblage under column removal scenario. Eng Fail
Anal 2019;100:381–92.
[46] Gouverneur D, Caspeele R, Taerwe L. Experimental investigation of the
load–displacement behaviour under catenary action in a restrained reinforced
concrete slab strip. Eng Struct 2013;49:1007–16.
[47] EN1992-1-1. Eurocode 2: design of concrete structures–part 1-1: general rules and
rules for buildings. European Committee for Standardization. 2004.
[48] Hendriks MA, de Boer A, Belletti B. Guidelines for nonlinear finite element analysis
of concrete structures. Rijkswaterstaat Technisch Document (RTD), Rijkswaterstaat
Centre for Infrastructure, RTD. 2017;1016-1:2012.
[49] fib. fib model code for concrete structures 2010: Wiley-vch Verlag Gmbh; 2013.
[50] Abaqus. ABAQUS user’s manual. Version 6.14. ABAQUS Providence, RI; 2014.
[51] CEB. Concrete structures under impact and impulsive loading: Synthesis report:
Comite euro-international du beton; 1988.
[52] Yılmaz İ, Arslan E, Kızıltaş EÇ, Çavdar K. Development of a prediction method of
Rayleigh damping coefficients for free layer damping coatings through machine
learning algorithms. Int J Mech Sci 2020;166. 105237.
16
Engineering Structures 237 (2021) 112057
[53] Hall JF. Problems encountered from the use (or misuse) of Rayleigh damping.
Earthquake Eng Struct Dyn 2006;35:525–45.
[54] Yu XH, Lu DG, Qian K, Li B. Uncertainty and sensitivity analysis of reinforced
concrete frame structures subjected to column loss. J Perform Constr Facil 2016;
31:04016069.
[55] Botte W. Quantification of structural reliability and robustness of new and existing
concrete structures considering membrane action: Doctoral Thesis; 2017.
[56] Ding L, Droogné D, Botte W, Van Coile R, Caspeele R. Structural reliability
calculations considering concrete tensile membrane action using the Probability
Density Evolution Method. 13th International Conference on Applications of
Statistics and Probability in Civil Engineering (ICASP13): Seoul National
University; 2019. p. 1-7.
[57] Olsson A, Sandberg G, Dahlblom O. On Latin Hypercube sampling for structural
reliability analysis. Struct Saf 2003;25:47–68.
[58] Charney FA. Unintended consequences of modeling damping in structures. J Struct
Eng 2008;134:581–92.
[59] Ding L, Botte W, Van Coile R, Caspeele R. Model Uncertainty Quantification for
Column Removal Scenario Calculations Using the Energy-based Method. The 13th
fib International PhD-Symposium In Civil Engineering. Paris2020.