Case Studies in Construction Materials 10 (2019) e00222

Contents lists available at ScienceDirect

Case Studies in Construction Materials

journal homepage: www.elsevier.com/locate/cscm

Case study

Explanation of the collapse of Terminal 2E at Roissy–CDG

Airport by nonlinear deterministic and reliability analyses
Hikmat Daoua , Walaa Abou Salhaa , Wassim Raphaela,*, Alaa Chateauneufb
a
École Supérieure d’Ingénieurs de Beirut (ESIB), Saint-Joseph University, Beirut, Lebanon
b
Université Blaise Pascal, France

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

Article history: The collapse of Terminal 2E Roissy at Charles de Gaulle Airport that failed eleven months
Received 12 November 2018 after its inauguration in 2004 and left four casualties is investigated in this paper by
Received in revised form 24 January 2019 performing deterministic and reliability analyses using the finite element software Ansys.
Accepted 24 January 2019
The aim of this study is to assess the accuracy of the terminal design by comparing the
calculated reinforcement ratio required with the existing ratio and to evaluate the
Keywords: reliability of Terminal 2E to know if the collapse could be predicted during the design phase.
Case study
Reliability analysis is performed using a combination of Response Surface Method (RSM)
Terminal 2E
Failure
and Monte Carlo Simulation (MCS) and taking into account the nonlinear properties of
Reinforced concrete material and the complexity of geometry to investigate the nonlinear behavior of the
Nonlinear analysis structure in order to reduce computational costs and obtain reliable results. Moreover,
Response surface method sensitivity analysis is performed to determine the effects of the uncertainties on the long
term deflection in order to investigate the causes of the failure.
© 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

The collapse incident of Terminal 2E Roissy bridge at Charles de Gaulle airport occurred eleven months after its
inauguration in 2004 and left four casualties. The collapse of Terminal occurred at 7 a.m., fortunately not at rush hour or it
would have caused a bigger disaster. This failure affected the image of the French civil engineering and led to significant
financial losses, noting that the 650 m length airport costs more than 750 million Euros [1]. Although structural reliability is
considered in several standards such as EN 1990 [2] and ISO 2394 [3], it has recently gained researchers’ attention after
several structural failure incidents such as the collapse of Terminal 2E Roissy bridge. The importance of reliability analysis is
mainly reflected by considering structural performance uncertainties that affect the construction work which is revealed in
structural safety.
Reliability is a concern because structural failure can result in injury or loss of life in addition to significant financial costs.
In order to quantify the reliability of a structure, the probability of failure Pf is calculated under a defined limit state condition
g(x). The probability of failure can be expressed as:
Z
Pf ¼ f x ðxÞdx ð1Þ
gðxÞ<0

* Corresponding author.
E-mail addresses: hikmat.daou@net.usj.edu.lb (H. Daou), walaa.abousalha@net.usj.edu.lb (W. Abou Salha), wassim.raphael@usj.edu.lb (W. Raphael),
alaa.chateauneuf@gmail.com (A. Chateauneuf).

https://doi.org/10.1016/j.cscm.2019.e00222
2214-5095/© 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).
2 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222

where x is the vector of the variables, fx(x) is the probability density function of the vector x, and g(x) is the limit state
function which divides the design space from a safe (g(x)  0) to a failure region (g(x) < 0). The variables describe
uncertainties in material properties, geometrical data, loads and modeling calculation. In the reliability analysis, the safety of
a structure can be assessed using 3 level methods:1) Level I methods where the safety of the structure is checked
deterministically, 2) Level II methods such as First-Order Reliability Method (FORM) and Second-Order Reliability Method
(SORM) where the approximation of the limit state function at design points is used to compute the failure probability, and 3)
Level III methods such as Monte Carlo Simulation (MCS) where the probability density function of variables is used to
compute the failure probability of the structure [4]. In a nonlinear finite element analysis, MCS method is not recommended
because it is time consuming and the probability calculated is noisy due to the randomness of the sampling [5]. For that,
Response Surface Method (RSM) is an efficient method for the reliability analysis. The aim of RSM is to calculate an
approximation input-output relationship; in this case, the computation cost will be reduced and very low probability levels
can be evaluated by performing MCS exploiting the approximated response surface instead of re-analysis of the nonlinear
finite element model. Several studies have been performed concerning the reliability analysis using RSM for different
structural elements and components; Zhao et al. [6] investigated the safety level of reinforced concrete beams bridges under
earthquake using RSM, Kartal et al. [7] performed probabilistic analysis of concrete-faced rockfill dams using RSM, Soares R.
C. et al. [8] analyzed nonlinearly reinforced concrete frames using RSM and many others.
The failure of the terminal was investigated by a national committee and several studies (e.g. [1,9], and [10]). The
administrative investigation has concluded that the bridge failure was due to a lack in the structure design and construction.
According to the Administrative Investigation Committee, the stresses and efforts calculated in the design phase were
inaccurate; the load applied at the ultimate limit state was 4.5 times greater than the maximum permissible load. Moreover,
the variation in temperature from 20  C to 4  C a few days before the incident was considered among the causes of failure
because it has led to additional efforts on the bridge shell. The Committee also commented on the hasty construction way, in
particular, the fact that the project construction phase was delayed one month due to technical problems failing to respect
the normal construction rules without genuine monitoring of the technical provided options [11]. Raphael et al. [9] studied
the terminal using the finite element software ST1 to examine the real reasons of its failure. They proved that an inadequate
modeling was carried out by the design office and that the long term concrete creep caused excessive unpredicted
deformations that resulted in excessive forces in the struts that led to shear punching in the shell. Furthermore, the reliability
assessment performed in terms of the short-term creep deflection by coupling ST1 and Phimeca (reliability analysis
software) showed that the terminal presented deficiencies and highlighted the importance of reliability assessment taking
into account the long-term strains of materials to prevent such an incident. In addition, El Kamari Y. et al. [1] studied the
deflection of the terminal using Ansys and simulated the progressive collapse of the Terminal by reducing the rigidity of the
yield elements. They proved that the real deflection exceeded the deflection predicted by the design office and that the
perforation of the shell from one side and its fracture on the other side happened because of the redistribution of efforts and
moments. The sensitivity analysis performed by [1] showed that the compressive strength had the most effect on the
structure strength. It is worth noting that these two previous studies of the terminal [1,10] did not take into account the
nonlinearity of its material properties; therefore, the relationships between stress-strain of concrete and steel were
considered as linear. Moreover, the model performed by [9] was primitive and inadequate enough to represent the real
responses of the terminal since it was drawn using nodes and bars and the reliability method used in the previous studies
[1,9] was classic and rudimentary. Kaljas T [10]. focused on the shape of the terminal and its reinforcement to explain the

Fig. 1. Typical section of boarding area of Terminal [11].

 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222 3

reasons of the failure. The terminal consists of elliptical reinforced concrete shell and external tensile reinforcement (ties and
struts) (Fig. 1). Kaljas T. [10] found that the tensile reinforcement was ineffective due to its geometry, the inadequate tensile
side reinforcement placement as well as the lack of strength and shear stiffness between the tensile and the compressive
sides. Moreover, the reinforcement of the shell was not suitable because the calculated tensile stresses exceeded the mean
tensile strength of concrete used in the terminal construction.
The importance of conducting a reliability assessment of the structure is to avoid structural incidents. As reliability
assessment requires a realistic modeling of the structure which shall represent its real mechanical behavior, a 3D finite
element model of Terminal 2E Roissy was simulated using Ansys by considering the nonlinear properties of material and the
complexity of geometry. Moreover, this study was performed using 3-level reliability analysis method using a combination
of RSM and MCS. Therefore, the purpose of this paper is to determine the reinforcement ratio required in the shell and to
undertake a nonlinear finite element analysis of the terminal carried out by Ansys in order to assess the reliability of the
structure in terms of long-term deflection by using a combination of RSM and MCS.

2. Finite element model

2.1. Overview of terminal 2E at Roissy Airport

Terminal 2E Roissy at Charles de Gaulle Airport opened on June 25, 2003 after several delays caused by security issues and
was designed to handle 10 million passengers per year. Terminal 2E Roissy consists of three main parts: a main building, a
boarding area and an isthmus connecting these two buildings. The boarding area has a length of 650 m and consists of 10
shells that provide access to aircraft through nine gateways and form a tube based on parallel longitudinal beams (Fig. 2). The
30 cm thickness precast concrete shell is 30 m wide with 26.2 m span and divided into 4 m width interlocking arches. The
arches based on horizontal beams are supported by columns by means of elastomeric bearings. The simulation of
elastomeric bearings was performed by using a complete stiffness matrix. In order to stiffen the shallow shell, curved steel
girders (ties) are braced to the two sides and are held away from the shell by regular steel struts. The boarding area is
surrounded by glass and honeycombed with hundreds of square windows to provide natural light while the three
footbridges link the boarding area to the planes [1]. On May 23, 2004, a part of the ceiling in the boarding area collapsed
suddenly after standing for eleven months without any reported problems leaving four casualties as shown in Fig. 3. The
collapse occurred in a section of the terminal containing access points for three footbridges.

2.2. Problem definition

The finite element software Ansys was used to perform the reliability evaluation of Terminal 2E Roissy at Charles de
Gaulle Airport. The precast concrete shell in the collapse area consisted of six full and perforated arches. In this study, we
chose to model the first nine rings from the isthmus including the six collapsed rings (Fig. 4). The finite element model was
developed considering structure complexity, holes in the shell, connection between arches (fragile corners iron), shortened
arches, asymmetry of the structure and loads applied. Moreover, MCS were performed exploiting the calculated response
surface in order to obtain statistical results such as sensitivity analysis, histograms and cumulative function distributions.

Fig. 2. General view of Terminal before the collapse [1].

4 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222

Fig. 3. Collapsed zone of Terminal 2E [1].

Fig. 4. Nonlinear finite element Ansys model of the Terminal 2E.

2.3. Model characteristics

In this study, nine arches, including the collapsed ones, were modeled. Arches consisted of two types: full and perforated
arches. Some arches were shortened due to existing footbridges. The 3D model was drawn in the same way as described by El
Kamari [1] (Fig. 4). Thus, the model was formed of 10,488 shell elements, 12,978 nodes and 814 bars. The arches were
modeled by shell elements while the struts and ties by bar elements. The nine arches were linked together by transversal
angle irons. The meshes used in the finite element Ansys are quadratic elements (12,740 nodes and 11,082 elements). Dead
and live loads were applied to the structural model: gravity, glazed roof and gateways loads were considered as dead loads
while thermal gradient and wind loads represented the live loads. The boarding area of the terminal was surrounded by a
glass of 36 mm thickness, and the gateways loads were considered equal to 120 kN. Table 1 presents the mechanical
characteristics of the concrete and steel materials used in this study. The modulus of elasticity of concrete in this study has
been modified to be more accurate. For that, it decreased from 37.95 GPa [1] to 33.34 GPa as we took into account the effect of

Table 1
Characteristics of materials used in the model [1].

Concrete Compressive strength fc 40 MPa

Density r 2.5 t/m3
Poisson’s ratio n 0.2
Thermal expansion coefficient (K1) 105
Steel Yield stress fy 460 MPa
Density r 7.85 t/m3
Poisson’s ratio n 0.3
Young’s modulus Ec 210 GPa
Thermal expansion coefficient (K1) 105
 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222 5

temperature on the value of the first loading date t0. We have considered the same characteristics of the reinforcement for
the external ties and struts. The compressive behavior of concrete is considered as elastic perfectly plastic where the
nonlinear relation between the compressive stress sc and strain ec is shown in Fig. 5 while the mean flexural tensile strength,
which depends on the mean axial tensile strength and the depth of the cross-section, is calculated as follows:

f ctm;f l ¼ maxfð1:6  h=1000Þf ctm ; f ctm Þ ð2Þ

where h is the total member depth in mm and fctm is the mean axial tensile strength [12].
As the structure safety is highly affected by the creep of concrete [13], several models were proposed to evaluate it such as
the ACI 209R-92 [14], the Bažant-Baweja B3 [15,16], the CEB Model Code 1990-99 (CEB MC90-99) [17–20], the GL2000
[21,22], and the Eurocode 2 (EC2) [12]. The concrete creep is affected by relative humidity, dimension of the element,
composition of the concrete, type of cement, maturity of the concrete when the load was first applied (t0) and duration and
magnitude of the sustained load. In this study, Ansys calculates the total deflection taking into account the creep of concrete
which was estimated using the creep coefficient w(t,t0) according to EC2 [12].

2.4. Reliability analysis of the terminal

Fig. 6 shows the probabilistic reliability assessment procedure using a combination of RSM and MCS.
For reinforced concrete (RC) structures, excessive deformation is a major cause of structural damage that can lead to
failure. As a result, deflection control is a must to ensure structure serviceability. It is important to notice that the aging of
concrete results in a reduction in the performance level [23]; thus, this paper studied the long-term deflection. The long-
term RC structure deflection is affected by the uncertainties of concrete mechanical properties: modulus of elasticity,
compressive strength, tensile strength, and creep parameters [24]. In addition, the mechanical behavior of RC structures is
also affected by the environmental conditions such as temperature and moisture.
In this study, Ansys calculates the total deflection of the terminal at long-term, which implicitly consists of the short-term
deflection and shrinkage deformation as well as the time-dependent deflection.

2.4.1. Limit state function

In order to evaluate structural integrity, we assess the risk by calculating the probability that the deflection of the
Terminal exceeds the allowable limit. The permissible limit of deflection ycritic is equal to l/250 [9] where l is the span length;
thus the limit state condition g(x) is expressed as
l
gðxÞ ¼ y ð3Þ
250
where y is the long term deflection of the Terminal and
  
l
Pf ¼ prob fgðxÞ  0g ¼ prob y 0 ð4Þ
250

Fig. 5. Parabola-rectangle diagram for concrete under compression [12].

6 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222

Fig. 6. Flowchart showing the probabilistic reliability assessment procedure.

2.4.2. Random variables, distributions and parameters

In this study, the random variables are: the density, the compressive strength and the modulus of elasticity of concrete [9],
the temperature and the wind velocity. The mean and standard deviation values of the temperature and the wind velocity are
calculated from the recorded values in the weather archive of Paris at Charles de Gaulle Airport [25] during the projected
lifetime of the terminal. These random variables were assumed to be statistically independent.
Table 2 presents a summary of the mean, standard deviation, coefficient of variation and probability distribution of the
design variables used in this study.

2.4.3. Response surface method using Ansys

The finite element simulations were performed using the finite element software Ansys, which includes the Probabilistic
Design System (PDS) tool that allows the user to perform a reliability analysis. The finite element model used in this study
was parametrically built, and the statistical distribution functions were defined for each variable as shown in Table 2. Ansys
performed parametric analyses in which simulation software automatically solves entire ranges of specified variables. The
aim of RSM is to approximate the input-output relationship by the mathematical function y ^ which was considered in this
study as a quadratic polynomial in the form:
X
n n X
X n
^ ¼ c0 þ
y ci :xi þ cij :xi :xj ð5Þ
i¼1 i¼1 j¼1

where c0, ci and cij with i,j = 1, . . . ,n are the regression coefficients, and xi and xj are the n input variables.
The regression coefficients are determined by a regression analysis and are usually evaluated, so that the sum of squared
differences between the true simulation results and the values of the approximation function are minimized [26].
The response surface analysis consisted of two steps. The first step was to perform the simulation loops to calculate the
deflection value that corresponds to the sample points of the random variables listed in Table 2. The second step was to
perform a regression analysis to compute the coefficients of the approximation function. The response surface is obtained
based on Design of Experiments which identifies the points in the space of input variables in an efficient and accurate way
[27]. In this study, Design of Experiments were generated using 27 training samples with respect to density, compressive
strength and modulus of elasticity of concrete, temperature and wind velocity and taking into account their specific

Table 2
Mean, standard deviation and coefficient of variation of the variables [10,26].

variable mx sx Vx Probability density function (PDF)

Compressive strength fc 40 MPa 6 MPa 0.15 lognormal
Modulus of elasticity of concrete Ec 33.34 GPa 5 GPa 0.15 lognormal
Concrete density 2.5 T/m3 250 T/m3 0.1 Gauss
Temperature 11  C 6.74  C 0.612 Weibull
Velocity of wind 10.71 m/s 3.77 m/s 0.352 Weibull
 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222 7

probabilistic distributions. The studied response was the deflection of the terminal shell which is solved at each training
sample using Ansys.

2.4.4. Probability of failure

Once the response surface was built, MCS were performed using this response surface. The probability of failure is
calculated as the fraction of Monte Carlo samples lying in the failure region.
In our study, the target reliability index is equal to 4.3 corresponding to Pf =8.539  106 according to (ISO 2394 1998) [3]
because we are in the case of great consequences and low relative costs of safety measures (Table 3).

2.4.5. Validity and accuracy of RSM

The response surface is an adequate representation of the true input-output relationship to give reliable and accurate
results. However, if the random output parameter is a non-smooth function of the random input variables, the
approximation function is insufficient. For that, the goodness-of-fit parameters implemented in Ansys warn the user if the
approximation function is insufficient. Therefore, it is necessary to check some parameters in order to evaluate the quality of
fitting [7,27–29]:

 Error of sum of squares SSE: the target value is zero.

X
n  2
SSE ¼ yi  yi ð6Þ
i¼1

 Coefficient of determination R2: the target value is one but a value superior to 0.95 is acceptable [30].

P
n
^ i Þ2
ðyi  y
2 i¼1
R ¼1 ð7Þ
Pn
ðyi  yÞ2
i¼1

 Adjusted coefficient of determination (adjusted R2): the target value is one.

P
n
^i Þ2
ðyi  y
2 n  1 i¼1
Adjusted R ¼ 1  ð8Þ
np1P n
ðyi  yÞ2
i¼1

 Maximum relative residual (MRR): the target value is zero.

 
yi  y^i
MRR ¼ MaxðAbs ð9Þ
i¼1:n yi

 Root mean square error (RMSE): the target value is zero.

Table 3
Target reliability index in accordance with (ISO 2394 1998).

Reliability costs of safety measures Consequences of failure

small some moderate great

high 0 1.5 2.3 3.1
normal 1.3 2.3 3.1 3.8
low 2.3 3.1 3.8 4.3
8 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u n
u1 X
RMSE ¼ t ðy  y ^i Þ2 ð10Þ
n i¼1 i

^i are the values of the output parameter and the regression model at the ith sampling point, respectively, y is
where: yi and y
the arithmetic mean of the values yi expressed as:

1X n
y¼ y ð11Þ
n i¼1 i

where n is the number of sampling, and p is the number of polynomial terms of quadratic response surface [25].
These parameters were calculated by Ansys and presented in Table 4.

2.4.6. Sensitivity analysis

Reliability sensitivity analysis studies the influence of the changes in distribution input parameters on the rate of change
in the probability of failure [31]. For mean value and standard deviation parameters, the reliability sensitivities can be
expressed as:

@Pf @PðgL ðxÞ  0Þ @FðbÞ @FðbÞ @b

¼ ¼ ¼ ð12Þ
@mxi @mxi @mxi @b @mxi

@Pf @PðgL ðxÞ  0Þ @FðbÞ @FðbÞ @b

¼ ¼ ¼ ð13Þ
@s xi @s xi @s xi @b @s xi
1
where β is the reliability index (b ¼ F ðPf Þ), F is the standard normal cumulative distribution function, gL(x) is the
linearized limit state function, and mxi and sxi are the mean value and standard deviation of the variable xi.

3. Results and analysis

As mentioned previously, the aim of this study is to calculate the reinforcement ratio required in the shell and to conduct a
reliability assessment of Terminal 2E Roissy at Charles de Gaulle Airport by performing a nonlinear finite element analysis to
know if the collapse could be predicted during the design phase by studying the parameters affecting the long-term
deflection and by considering the permanent as well as the thermal and wind loads. The model used was parametrically built
in Ansys.
Fig. 7 shows the flexural moment in the shell at the ultimate limit state without taking into account the nonlinearity of
materials’ properties. The maximum moment in the shell is equal to 0.111 MN.m which requires a reinforcement ratio of 0.22
according to Eurocode 2. The existing reinforcement ratio was 0.15 which is less than the required value. Consequently, the
reinforcement of the shell was insufficient, and the design was inaccurate.
Reliability analysis was performed using a combination of RSM and MCS to calculate the long term terminal deflection by
varying the density, compressive strength, and modulus of elasticity of concrete, temperature and wind velocity parameters’
values. Table 2 presents the model variables with the corresponding probability distributions and their mean, standard
deviation and coefficient of variation. RSM (Fig. 8) approximates the deflection value as a function of these variables;
therefore, the response surface determined by performing regression analysis was used for the determination of the
deflection value. By checking the goodness-of-fit measures (Table 4), it was found that the response surface is accurate and
sufficient, so the results obtained by RSM are true and reliable. The shape of our response surface is similar to the shape of
response surface of deflection obtained by [32,33]. Once the response surface was built, 107 Monte Carlo simulations were
performed exploiting this response surface. The results obtained by deterministic and probabilistic analyses are summarized
in Table 5.
The mean deflection value obtained using the deterministic analysis (9.09 cm) is approximately equal to the mean
deflection value obtained using the probabilistic analysis (9.6 cm). The deflection value varies from 4.91 cm to 15.6 cm with a

Table 4
Values of goodness-of-fit measures calculated by Ansys.

Goodness-of-fit parameters target value Obtained value

Error sum of squares SSE 0 6.428  106
Coefficient of determination R2 1 0.9989
Adjusted coefficient of determination (Adjusted R2) 1 0.9979
Maximum relative residual (MRR) 0 0.0165
Root mean square error (RMSE) 0 5.8165  104
 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222 9

Fig. 7. Results of moment (in N.m) in the shell using Ansys.

Fig. 8. Approximated response surface using Ansys.

Table 5
Values of deflection obtained by Ansys in deterministic and reliability analyses.

Minimum value Mean value Maximum Value Standard variation

Deterministic analysis – 9.09 cm – –
Reliability analysis (values obtained by MCS exploiting the response surface) 4.91 cm 9.60 cm 15.6 cm 0.15 cm

standard deviation of 0.15 cm in the probabilistic analysis. As shown in Fig. 9, the maximum deflection occurred in the part
near to the footbridges where crash happened. The collapsed zone is in a section of the terminal that is structurally different
from the rest of the building where the access points to the footbridges are presented; therefore, an interface problem
between structurally dissimilar portions of the building existed. The introduction of connecting footbridges into the
boarding area worked liked stress concentrations, and as a consequence, additional strain to the structure was created; thus
the excessive unpredicted deflection in this part had increased forces in the struts that led to shear punching in the shell.
However, cracks have been observed in a structurally similar area to the collapsed area
Sensitivity analysis shows the influence of the input variables on the structure; therefore, finds out the factors that led to
the collapse. The sensitivity analysis was made to assess the influence of the model variables on the structure. Based on the
previous studies, the creep of concrete [9] and the compressive strength of concrete [1] had the major effect on the long-term
terminal deflection while there was no effect of the thermal gradient. Noting that these studies used a 2-level reliability
analysis (FORM), and they did not take into account the nonlinearity behavior of concrete and steel. In this study, a nonlinear
3-level reliability analysis finite element analysis using response surface method was performed. Moreover, the shrinkage
10 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222

Fig. 9. Nonlinear deterministic results of deflection of Terminal using Ansys (in m).

and the creep of concrete were taken into consideration according to EC2 [22]. As shown in Fig. 10, the modulus of elasticity
of concrete and the density of concrete have the main effects on the long- term terminal deflection value. As a result, any
error in the mixing and casting of the concrete shell as well as the fabrication process will create unpredictable weakness in
addition to the concrete creep which reduces the modulus of elasticity of concrete. The perforation of several struts into the
shell revealed the weakness of concrete; thus, the poor quality of the concrete was one of the causes of the failure. The
sensitivity analysis also shows that the temperature was among the reasons of collapse (as obtained by the investigators)
while the previous studies have concluded that there is no influence for this parameter. The temperature during the week
before the collapse was around 25  C, but at night it decreased to 4  C. This rapid change in temperature has created
additional efforts in the shell because the expansion of ties and concrete and the contraction of struts, and led to failure.
The interpretation of the results of MCS is based on statistical methods. Histogram and cumulative deflection distribution
function are presented in Figs. 11 and 12. For the cumulative distribution function (CDF), the data is sorted in an ascending
order and the CDF of the ith data point, denoted with Fi, can be derived from:

X
N
N!
F k ð1  F i ÞNk ¼ 50% ð14Þ
k¼1
ðN  kÞ!k! i

From the CDF curve showed in Fig. 12, the probability of failure as well as the inverse probability can be derived.
In a prior study [9], the reliability analysis showed that the terminal presented deficiencies in the long-term deflections.
The permissible level of deflection obtained ycritic is equal to l/250 [9] and Pf is given by Pf = prob{(l/250 – y) <0} where l is the
span length and y is the long-term deflection. The probability of failure obtained is equal to Pf = 1.6478  102. This
probability of failure is very high and unacceptable compared to the target value which equal to 8.53  10-6. Thus, this
reliability study shows that the structure is prone to failure where it presents a high probability of failure while the
deterministic calculation of the deflection did not present any deficiency in a long- term condition.

Fig. 10. Sensitivity analysis of the variables for the terminal (in %).
 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222 11

Fig. 11. Distribution of long-term Terminal deflection using Ansys.

Fig. 12. Cumulative distribution function of long-term Terminal deflection using Ansys.

4. Conclusion

In this study, geometrically and materially nonlinear reliability analysis of Terminal 2E Roissy was investigated using a
combination of Response Surface Method (RSM) and Monte Carlo simulation (MCS). For this purpose, finite element model
consisting of nine arches and taking into account structure complexity, holes in the shell, connection between arches (fragile
corners iron), shortened arches and asymmetry of the structure was carried out using Ansys. Dead load (gravity, glazed roof
and gateways loads) and live loads (thermal gradient and wind loads) were applied to this model. The nonlinear behavior of
structural elements was modeled using elastic-perfectly plastic for concrete and steel, and the shrinkage and creep of
concrete were calculated according to EC2 in the nonlinear analysis.
Based on the deterministic analysis of the bending moment, the design of the terminal made by the design office was
inaccurate because the shell was insufficiently reinforced.
The reliability analysis according to the long-term deflection was done by performing the following:

 A representative finite element model of the terminal was parametrically built using Ansys.
 The variables taken into account were the modulus of elasticity, the compressive strength and the density of concrete, the
temperature and the wind velocity.
 RSM was performed to calculate the approximate response surface.
12 H. Daou et al. / Case Studies in Construction Materials 10 (2019) e00222

 107 MCS were performed exploiting the approximate response surface.

 The deterministic analysis showed that the maximum deflection occurs in the part near to the footbridges where crash
happened. According to the reliability analysis, the modulus of elasticity of concrete and the concrete density had the main
effects on the long-term Terminal deflection value. The sensitivity analysis had also highlighted the effects of the
environmental conditions (temperature and wind) on the structure.
 The reliability analysis showed that the structure represented deficiencies and was prone to failure in terms of the long-
term conditions. Excessive deformation was a major cause of the structural damage, and thus, led to the terminal failure.

References

[1] Y. El Kamari, W. Raphael, A. Chateauneuf, Reliability study and simulation of the progressive collapse of Roissy Charles de Gaulle Airport, J. Case Stud.
Eng. Fail. Anal. 3 (2015) 88–95, doi:http://dx.doi.org/10.1016/j.csefa.2015.03.003.
[2] EN, 1990 Eurocode, Basis of Structural Design, CEN, 2002.
[3] ISO, 2394 General Principles on Reliability for Structures, ISO, 1998.
[4] Hasan Basri Basaga, Alemdar Bayraktar, Irfan Kaymaz, An improved response surface method for reliability analysis of structures, J. Struct. Eng. Mech.
42 (2) (2012) 175–189, doi:http://dx.doi.org/10.12989/sem.2012.42.2.175.
[5] Anirban Basudhar, Samy Missoum, Antonio Harrison Sanchez, Limit state function identification using support vector machines for discontinuous
responses and disjoint failure domains, J. Probabilist. Eng. Mech. 23 (2008) 1–11, doi:http://dx.doi.org/10.1016/j.probengmech.2007.08.004.
[6] Jun Zhao, Junqi Lin, Jinlong Liu, Jia Li, Study on safety level of RC beams bridges under earthquake, AIP Conference Proceedings 1864 (2017), doi:http://
dx.doi.org/10.1063/1.4992879 020062.
[7] Murat Emre Kartal, Hasan Basri Basaga, Alemdar Bayraktar, Probabilistic nonlinear analysis of CFR dams by MCS using response surface method, J. Appl.
Math. Model. 35 (2011) 2752–2770, doi:http://dx.doi.org/10.1016/j.apm.2010.12.003.
[8] R.C. Soares, A. Mohamed, W.S. Venturini, M. Lemaire, Reliability analysis of non-linear reinforced concrete frames using the response surface method, J.
Reliab. Eng. Syst. Safe. 75 (2002) 1–16.
[9] Wassim Raphael, Rafic Faddoul, Roy Feghaly, Alaa Chateauneuf, Analysis of Roissy Airport Terminal 2E collapse using deterministic and reliability
assessments, J. Eng. Fail. Anal. 20 (2012) 1–8, doi:http://dx.doi.org/10.1016/j.engfailanal.2011.10.001.
[10] Toomas Kaljas, Reasosns for Charles de Gaulle Airport collapse, J. Civ. Eng. Archit. 11 (2007) 411–419, doi:http://dx.doi.org/10.17265/1934-7359/
2017.05.001.
[11] L’express, Après 10 Ans d’enquête, Procès En Vue Pour l’effondrement Du Terminal 2E, (2014) . (accessed 1 October 2014) https://www.lexpress.fr/
actualites/1/societe/apres-10-ans-d-enquete-proces-en-vue-pour-l-effondrement-du-terminal-2e_1605785.html.
[12] CEN. EN 1992-1-1, Eurocode 2: Design of Concrete Structures – Part 1-1: General Rules and Rules for Buildings, (2004) Brussels.
[13] Wassim Raphael, Elise Zgheib, Alaa Chateauneuf, Experimental investigations and sensitivity analysis to explain the large creep of concrete
deformations in the bridge of Cheviré, J. Case Stud. Constr. Mater. 9 (2018), doi:http://dx.doi.org/10.1016/j.cscm.2018.e00176b.
[14] ACI Committee 209, Prediction of Creep, Shrinkage, and Temperature Effects in Concrete Structures (ACI 209R-92), American Concrete Institute,
Farmington Hills, MI, 1992, pp. 47.
[15] Z.P. Bažant, S. Baweja, Creep and shrinkage prediction model for analysis and design of concrete structures—model B3, Mater. Struct. 28 (1995) 357–
365 415-430, 488-495.
[16] Z.P. Bažant, S. Baweja, Creep and shrinkage prediction model for analysis and design of concrete structures: model B3, in: A. AI-Manaseer (Ed.), The
Adam Neville Symposium: Creep and Shrinkage-Structural Design Effects, SP-194, American Concrete Institute, Farmington Hills, MI, 2000, pp. 1–83.
[17] H.S. Muller, H.K. Hilsdorf, General task group 9, CEB Comire Euro-International du Beton, (1990) , pp. 201 Paris, France.
[18] CEB, "Evaluation of the Time Dependent Properties of Concrete," Bulletin d’lnformation No. 199, Comite European du BetonlFederation Internationale
de la Precontrainte, Lausanne, Switzerland, 1991, pp. 201.
[19] CEB, CEB-FIP Model Code 1990, CEB Bulletin d’lnformation No. 2131214, Comite Euro-International du Beton, Lausanne, Switzerland, 1993, pp. 33–41.
[20] CEB, Structural Concrete-Textbook on Behaviour, Design and Performance. Updated Knowledge of the CEBI FIP Model Code 1990, fib Bulletin 2, V. 2,
Federation Internationale du Beton, Lausanne, Switzerland, 1999, pp. 37–52.
[21] N.J. Gardner, M.J. Lockman, Design provisions for drying shrinkage and creep of normal strength concrete, Aci Mater. J. 98 (Mar.-Apr.2) (2001) 159–167.
[22] N. Gardner, Comparaison of prediction provisions for drying shrinkage and creep of normal-strength concretes, Am. J. Civ. Eng. Archit. 31 (5) (2004)
767–775.
[23] Behnam Kiani, Robert Y. Liang, Jacob Gross, Material selection for repair of structural concrete using VIKOR method, J. Case Stud. Constr. Mater. 8 (2018)
489–497, doi:http://dx.doi.org/10.1016/j.cscm2018.03.008.
[24] Tengfei Xu, Tianyu Xiang, Renda Zhao, Guotao Yang, Cheng Yang, Stochastic analysis on flexural behavior of reinforced concrete beams based on
piecewise response surface scheme, J. Eng. Fail. Anal. 59 (2016) 211–222, doi:http://dx.doi.org/10.1016/j.engfailanal.2015.10.004.
[25] Weather archive Charles De Gaulle Airport. https://www.meteoblue.com/en/weather/forecast/archive/paris-charles-de-gaulle-
airport_france_6269554.
[26] Probabilistic design techniques. https://www.sharcnet.ca/Software/Ansys/16.2.3/enus/help/ans_adv/Hlp_G_ADVPDS5.html (last modified 29-Jun-
2015).
[27] Stefan Reh, Jean-Daniel Beley, Siddhartha Mukherjee, Enh Hui Khor, Probabilistic finite element analysis using ANSYS, J. Struct. Saf. 28 (2006) 17–43,
doi:http://dx.doi.org/10.1016/j.strusafe.2005.03.010.
[28] Flavio Stochino, RC beams under blast load: reliability and sensitivity analysis, J. Eng. Fail. Anal. 66 (2016) 544–565, doi:http://dx.doi.org/10.1016/j.
engfailanal.2016.05.003.
[29] Goodness of fit criteria, https://www.sharcnet.ca/Software/Ansys/17.0/en-us/help/wb_dx/dx_gofcriteria.html.
[30] M. Pendola, A. Mohamed, M. Lemaire, P. Hornet, Combination of finite element and reliability methods in nonlinear fracture mechanics, J. Reliab. Eng.
Syst. Safe. 70 (2000) 15–27.
[31] Ning-Cong Xiao, Hong-Zhong Huang, Zhonglai Wang, Yu Pang, Liping He, Reliability sensitivity analysis for structural systems in interval probability
form, J. Struct. Multidiscip. Optim. 44 (2011) 691–705, doi:http://dx.doi.org/10.1007/s00158-011-0652-9.
[32] Thanasis K. Barlas, Busra Akay, Optimization of morphing flaps based on fluid structure interaction modeling, Conf. 2018 Wind Energy Symposium,
January (2018), doi:http://dx.doi.org/10.2514/6.2018-0998.
[33] Jorge Crempien, Ferdinando Auricchio, Carlo G. Lai, Fabio Nobile, Response statistics of uncertain dynamical systems subjected to stochastic loading
usimg sparse grid collocation techniques, Istituo Universitario di Studi Superiori, Università degli Studi di Pavia, 2008.