Thin-Walled Structures 108 (2016) 20–29

Contents lists available at ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

Full length article

Prediction of the buckling behaviour of thin cement composite shells:

Parameter study
Evy Verwimp a,b, Tine Tysmans a,n, Marijke Mollaert b, Maciej Wozniak a
a
Vrije Universiteit Brussel, Faculty of Engineering, Department of Mechanics of Materials and Constructions (MeMC), Brussels, Belgium
b
Vrije Universiteit Brussel, Faculty of Engineering, Department of Architectural Engineering (ARCH), Brussels, Belgium

art ic l e i nf o a b s t r a c t

Article history: This paper demonstrates the importance of including non-linearities in the design of textile reinforced
Received 9 February 2016 concrete (TRC) shells. By two cases, the influence on the buckling load of considering geometrical non-
Received in revised form linearity and material non-linearity, but also of considering geometrical imperfections is examined. For
17 May 2016
stiff shells mainly in compression, the buckling load is significantly reduced by including imperfections,
Accepted 13 July 2016
analogue to observations with steel reinforced concrete shells. For shells with a low stiffness, experi-
encing large deformations, a non-linear analysis is essential and the buckling load is mainly reduced by
Keywords: considering material non-linearity. Conclusively, the shape and boundary conditions of thin TRC shells
Buckling indicate which parameters should be taken into account in their design.
Geometrical imperfections
& 2016 Elsevier Ltd. All rights reserved.
Geometrically non-linear analysis
Finite elements
Shell structures
Textile reinforced concrete (TRC)

1. Introduction design of the reinforcement in bending, however buckling and

definitely its non-linear analysis is often overlooked (e.g. case
Textile reinforced concrete, cement or mortars (TRC) consist of studies presented in [7–9]). Yet, because of their high slenderness,
continuous fabrics combined with a fine grained matrix. TRC's are buckling may become too significant to not include it early in the
strain hardening materials with high compressive and tensile design stage. Buckling of steel reinforced concrete shells is rela-
strength, ductility and energy dissipation. Thanks to these good tively well examined and general design rules are developed
mechanical properties and the non-corroding reinforcement, throughout the years [10–12], however this is not the case for TRC
several (slender) building applications in TRC were already de- shells. Moreover, the envisioned section thickness is thinner than
veloped over the past few years, like façade elements [1], thin that of steel reinforced concrete shells and the TRC shell's slen-
faces for sandwich panels [2], and lightweight firewalls [3]. These derness approaches that of steel shells. The buckling predictions of
straight-lined applications however do not exploit another ad- very thin steel shells are mostly affected by considering geome-
vantage of TRC, namely the flexibility of the reinforcement. Curved trical non-linearity in the analysis and including imperfections in
elements, like shells, can easily be obtained using TRC, as de- the geometry [13,14]. For this reason, this paper numerically in-
monstrated in [4,5]. Very recently, the RWTH Aachen University vestigates the influence of considering this geometrical non-line-
arity, thus taking into account second order effects in the analysis,
successfully constructed a pavilion composed of four hypar TRC
and including geometrical imperfections on the buckling/bearing
shells [6,7]. Each hypar (7
7 m) was produced as a precast part
behaviour of two very thin TRC shells – with varying boundary
and supported at its centre by a steel reinforced concrete column.
conditions and structural stiffness – under uniform vertical load-
The shells are 60 mm thick, consisting of 12 layers of non-cor-
ing. Moreover, as the material behaviour of steel and TRC is very
roding carbon textile reinforcement and arranged in a 2 by 2 grid,
different, the impact of considering the material non-linearity of
resulting in a pavilion of 14
14 m. These examples clearly show
TRC in tension is also investigated. The aim of this paper is to
the potential of TRC shells, which can be built with a slenderness
evaluate if these non-linearities and imperfections should be taken
which is far more superior to that of traditional steel reinforced
into account (separately or in combination) in the early design
concrete shells. stage of thin TRC shells. As such the results of this paper develop a
In literature, the design of TRC shells often focuses on the certain awareness towards important issues in the design and
assessment of thin shell structures made of cementitious
n
Corresponding author. composites.
E-mail address: [email protected] (T. Tysmans). The first part of the paper describes the materials used and the

http://dx.doi.org/10.1016/j.tws.2016.07.011
0263-8231/& 2016 Elsevier Ltd. All rights reserved.
 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29 21

Fig. 1. Two case studies are examined; (a) circular synclastic shell and (b) a compression-only shell of true scale.

method to examine buckling numerically. Both the linear and non- structures.
linear buckling analysis, which will be performed in the finite
element software Abaqus, are explained. Furthermore, the method 2.1. Textile reinforced cement: material assumptions
to introduce geometrical imperfections – always present in real
built shells – is also described. For the study presented in this paper, the matrix inorganic
The second part of the paper presents the analysis of the first phosphate cement (IPC) [18], developed at Vrije Universiteit
case study, a synclastic shell of 2 m span, 1.2 m height and 6.5 mm Brussel, is used in combination with chopped E-glass fibre mats
thickness (Fig. 1a), thus it has a slenderness (¼ radius of curvature (300 g/m2). IPC is pH-neutral after hardening and thus does not
to thickness ratio) of λsl ¼400, which is in the same range as the chemically degrade the glass fibres. Moreover, the very small grain
concrete dome examined in [15]. This geometry is chosen because size of the cement allows good impregnation of dense textiles such
it was already experimentally tested and the numerical model was that high fibre volume fractions up to 20% [19] can be achieved.
validated by the authors in [16]. As such the performed experi- The mechanical behaviour of glass fibre textile reinforced IPC
ment in [16] complements this paper. (GFTR-IPC) is very different in tension and in compression due to
The next part describes the analysis of the second case study. the brittle matrix (Fig. 2a [20]). It shows already a non-linear be-
The aim of this case study is to examine if the same reasoning as haviour at low tensile stresses due to the low tensile failure strain
for the 2 m shell can be applied to a true scale shell. This case of the matrix relative to that of the fibres. However, by adding high
study is a synclastic compression-only shell (Fig. 1b) of 10
10 m fibre volume fractions the composite can obtain a significant post-
span, 4.8 m height and 12.5 mm thickness (thus the same slen- cracking stiffness and tensile strength (Fig. 2b [19]). In compres-
derness of λsl ¼400), generated using the form finding software sion GFTR-IPC is linear elastic until failure [20]. The properties of
Easy [9]. This software, based on the force density method [17], the GFTR-IPC used can be found in Table 1.
searches for the equilibrium shape of a flexible fabric under a The chosen materials are not the ones which are traditionally
certain load case, such that this equilibrium shape is in membrane combined to obtain textile reinforced concrete [21], i.e. IPC is ra-
stress state. For this study, a compression-only shape under its ther a cement than a mortar or concrete, and the chopped mats
self-weight was generated. are not bidirectional textiles but consist of randomly oriented fibre
Finally, the last part summarizes the results and presents the bundles of 50 mm length. Using this randomly oriented fibre mat,
conclusions of the paper. the cement composite has an isotropic behaviour and the or-
ientation of the textiles must not be taken into account like in [6].
The isotropic strength and stiffness are an advantage when
2. Materials and methods working with complex shell geometries in which the principal
strains vary in directions throughout the surface under different
This section presents the materials, the methods and geome- loading conditions. Moreover, the fibre mats are generally better
tries used in this paper for analysing the cement composite shell shape adaptive than bidirectional textiles, making more complex

Fig. 2. (a) The mechanical behaviour of GFTR-IPC differs significantly in compression and in tension [20], (b) increasing the fibre volume fraction increases its post-cracking
stiffness and tensile strength [19].
22 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29

Table 1 Ptotal = P0 + λ(Pref − P0) (3)

Properties of glass fibre textile reinforced inorganic
phosphate cement. where P0 is the dead load, Pref is the reference load vector, and λ is
the load proportionality factor. The load proportionality factor is
Density ρ¼ 1900 kg/m3
Compressive characteristic strength sck ¼80 MPa found as part of the solution, and printed at each increment. The
Tensile characteristic strength stk ¼ 40 MPa solution is a single equilibrium path (arc length) in a space defined
Initial Young modulus Ec1 ¼18 GPa by the nodal variables and the loading parameter, and the actual
Fibre density Vf ¼ 20%
load value may increase or decrease as the solution progresses.
Poisson ratio υ ¼0.3

2.2.3. Geometrical imperfections

shapes possible. However, on a macro scale GFTR-IPC and TRC Real shells are built and executed imperfect, this is particularly
show similar behaviour. First, the chopped glass fibre mats provide the case for thin shells. These shape imperfections can reduce the
a continuous reinforcement just like textiles and second, the high buckling resistance significantly, as was already demonstrated for
fibre volume fraction grants a significant strain hardening beha- concrete shells [25] and steel shells [14]. One of the investigated
viour in tension just like traditional TRC. For this reason, we parameters is therefore the effect of the size and shape of those
choose to refer to GFTR-IPC as a kind of TRC. imperfections on the buckling load of thin cement composite
shells. In literature, there are two general methods to introduce
2.2. Prediction of the buckling behaviour of thin GFTR-IPC shells geometrical imperfections in a FE model: (1) In most references,
the shape of the first buckling mode – obtained by an eigenvalue
2.2.1. Linear analysis: eigenvalue buckling prediction buckling prediction of the perfect shell – is used as the initial
Generally, when the linear buckling theory is applied, the aim imperfection shape [11,25,26]. (2) A more elaborated way is
is to obtain a first approximation of the value of the critical probably to use a geometrical imperfection shape which is a linear
buckling load. In this linear theory, the shell's behaviour prior to combination of the first few buckling modes [12]. If the eigenva-
the critical load – also called bifurcation point or limit load – is lues are closely spaced, there is no physical reason why the next
assumed to be in a primary stable linear state [22]. At the bi- mode would be less likely to appear.
furcation point, the initial primary behaviour mode crosses a Concerning the magnitude of geometrical imperfections to be
secondary mode of the shell, which is usually an unstable mode. considered in the analysis, Medwadowski [10] recommends for
The basis of the underlying linear buckling analysis performed in concrete structures to use the shell's thickness as a rather con-
Abaqus is explained below [23]. Consider the equilibrium between servative limit. Eurocode 3 [27] presents the design rules con-
internal and external forces; cerning the strength and stability of steel shell structures. Steel
shell structures have a far higher slenderness than TRC shells can
⎡⎣ K⎤⎦u = F (1) achieve, and Eurocode 3 recommends to include an imperfection
size which is not smaller than 0.625  thickness for a normal fab-
in which [K] is the global stiffness matrix of the structure, F is the
rication tolerance. For this reason, the effect of the imperfection
vector of the nodal forces and u is the vector of the nodal dis-
size is also studied, and referred to as a fraction of the shell's
placements. At the point where the structure becomes unstable,
thickness. The geometrical imperfections are introduced into the
external disturbances are no longer necessary to further displace
numerical model as described in the Abaqus User's Manual [28].
the structure. Hence;
The geometrically non-linear numerical approach (as discussed
⎡⎣ K⎤⎦u = 0 (2) in Section 2.2.2) including geometrical imperfections was already
validated in [16] by the experimental assessment of the buckling
At buckling u cannot be zero, in other words the stiffness behaviour of a thin spherical GFTR-IPC dome.
matrix [K] becomes singular or |K| ¼0. Within the total applied
loads a preload P – e.g. self-weight – and an incremental loading 2.2.4. Material non-linearities
pattern Q are distinguished. Consequently, the tangent stiffness Whereas the consideration of imperfections and geometrical
matrix [K] equals the sum of a stiffness matrix corresponding to non-linearity affect the buckling load predictions of very thin steel
the preload [K0] and a so called initial stress stiffness matrix [KΔ]. shells (r/t 4500), the plasticity of steel plays no role in the buck-
The critical buckling loads are then equal to Pþ λiQ, with λi the ling analysis [13,14]. However, the material behaviour of steel is
eigenvalues. The corresponding mode shapes (eigenvectors) are typically linear up to high tensile stresses (e.g. 235 MPa for widely
given by ui. used steel S235) while the TRC used is non-linear at already very
low tensile stresses (around 7 MPa). This non-linear tensile be-
2.2.2. Geometrically non-linear analysis haviour might be important for thin TRC shells which encounter
As thin shell structures can suffer large deformations, the effect large deformations. Therefore the buckling prediction of both case
of including these pre-buckling deformations can become too studies is investigated considering this material nonlinearity and
significant to not consider them early in the design process. compared to analyses with linear material assumptions.
Therefore the influence of considering geometrical non-linearity in When the material behaviour of GFTR-IPC is assumed linear
the buckling analysis, using the modified Riks method in Abaqus elastic, only the initial Young's modulus and Poisson's ratio (Ta-
[24], is studied and compared with the linear analysis (see Section ble 1) are needed in Abaqus. To model the non-linear material
2.2.1). behaviour of the GFTR-IPC material, the Concrete Damaged Plas-
The modified Riks [24] method is a load-deflection method, ticity (CDP) model is used, which is inbuilt in Abaqus. Validation
which uses the load magnitude as an additional unknown, and studies [29] have documented the capability of the CDP model to
calculates simultaneously loads and displacements. Therefore, closely predict the nonlinear tensile response of GFTR-IPC for both
another quantity must be used to measure the progress of the uniaxial and biaxial stress states.
solution; Abaqus [24] uses the arc length l along the static equi-
librium path in load-displacement space. The loading during a Riks 2.3. Finite element modelling of the geometries
step is always proportional. The current load magnitude, Ptotal, is
defined by Both geometries are modelled in Abaqus as a continuous shell
 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29 23

section. Two types of linear elements are used, namely the S3 and
the S4R elements, i.e. 3-node and 4-node shell elements with re-
duced integration respectively. These elements are three-dimen-
sional general-purpose elements which can be used for both thin
and thick shells in all kinds of boundary conditions and loading
conditions [30]. Mostly the S4R elements are used, but if locally
the geometry of the shells does not allow them, S3 elements are
used instead. Mesh convergence was checked and a mesh with
16,014 elements for the 2 m shell and 16,357 elements for the
compression-only shell are chosen as a compromise between ac-
curacy and calculation time. Both case studies are subjected to a
vertical uniformly distributed load. Due to the low thickness of the
studied shells compared to their other dimensions, the self-weight
is neglected. The 2 m synclastic shell is examined both for pinned
and vertically supported boundary conditions. The compression- Fig. 4. Pinned shell: considering geometrical non-linearity and especially including
only shell has pinned boundary conditions in its four corners. geometrical imperfections (based on mode 1) affect the buckling load prediction
significantly.

3. Case study 1: synclastic GFTR-IPC shell 51% (105.1 kN/m2 to 51.8 kN/m2). Considering geometrical non-
linearity in the analysis reduces the buckling load prediction with
3.1. Results of linear analysis only 12% (105.1 kN/m2 to 92.2 kN/m2), while considering both
geometrical nonlinearity and geometrical imperfections reduces
Generally, the intention of a linear buckling analysis is to de- the buckling load prediction with 57% with reference to the linear
termine the most probable failure pattern of a structure. In this buckling prediction of the perfect shell (105.4 kN/m2 to 45.4 kN/
context, the first eigenmode λ1 (Fig. 3a) is often considered as the m2). This illustrates not only the importance of performing a non-
buckling mode requiring the smallest amount of energy to occur. linear analysis on the thin GFTR-IPC shell, but proves especially the
The first four eigenvalues and accompanying buckling modes of predominance of geometrical imperfections. Finally, including the
case study 1 under a vertical uniform distributed load are shown non-linear material behaviour does not affect the buckling re-
in Fig. 3. In this case study, the eigenvalues are very closely spaced, sponse of this shell in a significant way. Considering the im-
meaning there is no physical reason the next modes would be less portance of these geometrical imperfections, a more elaborated
likely to appear than the first mode. A series of closely spaced study is performed to examine the influence of their size and their
eigenvalues often indicates the imperfection sensitivity of the shape, and this assuming non-linear material behaviour.
structure. An eigenvalue buckling analysis does not predict the As previously mentioned (Section 2.2.3), the most elaborated
buckling load correctly for imperfection sensitive structures and a way to introduce geometrical imperfections in the numerical
non-linear procedure should be used instead [23]. This indicates model, is by means of a linear combination of different buckling
the importance of performing a geometrically non-linear analysis modes obtained from the linear analysis. To distinguish the shapes,
to predict the buckling behaviour of the GFTR-IPC shell. four different linear combinations are studied. The first three
groups consist of linear combinations of the buckling modes
3.2. Results of non-linear analysis where the eigenvalues are very closely spaced within one group.
For example, group 1 consists of four buckling modes with ei-
3.2.1. Pinned shell genvalues between 106.0 kN/m2 and 106.3 kN/m2. The last group
To investigate separately the impact of considering geometrical consists of a linear combination where the shapes of the buckling
nonlinearity, geometrical imperfections and material nonlinearity, modes are similar and the combination of these imperfection
following cases are analysed: (i) Linear analysis of perfect shell and shapes fortify each other (on certain locations). The different
(ii) of shell with geometrical imperfections; (iii) geometrically groups are:
nonlinear analysis of perfect shell assuming linear material beha-
viour and (iv) considering nonlinear behaviour; (v) nonlinear  group 1 ¼[λ2, λ3, λ4, λ5] ¼[106.0 kN/m2, 106.0 kN/m2, 106.3 kN/
analysis of geometrically imperfect shell assuming linear material m2, 106.3 kN/m2] and Δx1,j ¼ ω1,j  ∑i∈(2,3,4,5)ui
behaviour and (vi) considering nonlinear material behaviour.  group 2 ¼[λ6, λ7] ¼[108.2 kN/m2, 108.2 kN/m2] and Δx2,j ¼ ω2,j
Fig. 4 shows the applied load as a function of the displacement  ∑i∈(6,7)ui
for the different cases. The applied imperfection is based on the  group 3 ¼[λ8, λ9, λ10, λ11] ¼[109.2 kN/m2, 109.2 kN/m2, 109.5 kN/
first buckling mode of the linear analysis and has a size equal to m2, 109.5 kN/m2] and Δx3,j ¼ ω3,j  ∑i∈(8,9,10,11)ui
the thickness of the shell (6.5 mm). Including the geometrical  group 4¼[λ2, λ4] ¼ [106.0 kN/m2, 106.3 kN/m2] and Δx4,j ¼ ω4,j
imperfection in the linear model decreases the buckling load with  ∑i∈(2,4)ui

Fig. 3. Pinned shell: the eigenvalues of the first buckling modes are closely spaced, indicating the spherical shell is sensitive to imperfections. (a) λ1 ¼105.1 kN/m2. (b)
λ2 ¼ 106.0 kN/m2. (c) λ3 ¼106.0 kN/m2. (d) λ4 ¼106.3 kN/m2.
24 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29

Fig. 5. Pinned shell: the greater the imperfection size, the greater the impact and the lower the buckling load. (a) Group 1. (b) Group 2. (c) Group 3. (d) Group 4.

Fig. 7. Vertically supported shell: the considered geometrical imperfection (mode

1) does not affect the buckling behaviour, but including the non-linear material
behaviour significantly decreases the buckling load.
Fig. 6. Pinned shell: both size and shape of the geometrical imperfection affect the
buckling load.
thickness⋅imperfection share j
ωk, j =
# modes in group k (4)

with Δx the imperfection size, ui the shape of the normalised Fig. 5 shows the applied load as a function of the displacement
buckling mode (eigenvector), ωk,j the parameter which scales the for the synclastic shell with the four different imperfection shapes
buckling modes to obtain the desired imperfection size and is (group 1–4) and with imperfection sizes Δx varying from 0%, 3%, 15
defined by: to 100% of the shell's thickness. In all these results, non-linear
 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29 25

Fig. 8. Vertically supported shell: the deformations are large and a non-linear analysis including material non-linearities is more important than including geometrical
imperfections. (a) Group 1. (b) Group 2. (c) Group 3. (d) Group 4.

Fig. 9. For the compression-only shell, the eigenvalues of the first four buckling modes are closely spaced. (a) λ1 ¼ 1589 N/m2. (b) λ2 ¼ 1590 N/m2. (c) λ3 ¼ 1617 N/m2. (d)
λ4 ¼ 1625 N/m2.

material behaviour is assumed. Fig. 6 shows the buckling load of 56–72%.

the different groups as a function of the imperfection share and The maximum deformation of the GFTR-IPC shell at the point of
gives a more concise overview of the results. buckling lies between 4.2 and 4.5 mm ( ¼L/476  L/444) for 3%
Comparing the different groups, applying a different im- imperfection, between 4.1 and 5.0 mm (¼L/488 L/400) for 15%
perfection shape affects the buckling load prediction significantly. imperfection and between 4.1 and 8.4 mm (¼L/488 L/238) for an
For example considering a constant imperfection size equal to imperfection equal to the shell's thickness. These deformations are
100% of the shell's thickness, the buckling load of group 1 is 37% relatively small and remain in the same order of magnitude as the
lower than the buckling load of group 4 (25.5 kN/m2 vs. 40.6 kN/ shell's thickness ( ¼maximum imperfection size) due to the
m2) and even 41% lower than when only mode 1 is included as structural stiffness of the shell.
imperfection shape (25.5 kN/m2 vs. 43.1 kN/m2; see Fig. 4). Within These results clearly prove the importance of considering
one group, the buckling load decreases nonlinearly with increasing geometrical imperfections when analysing the thin GFTR-IPC shell.
size of geometrical imperfection. For example, an imperfection When designing a shell, it is evidently not possible to know in
equal to 15% of the shell's thickness decreases the initial buckling advance which imperfection shape or size will occur in the man-
load with 16–34%, while increasing the imperfection size to 100% ufactured shell. For this reason, the most critical shape must be
of the shell's thickness, 6.5 mm, decreases the buckling load with determined in the design process. Therefore, the method of using
26 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29

while assuming non-linear material behaviour results in a model

where stresses over the shell's section are continuously redis-
tributed (when the non-linear stage in tension is reached) and
thus compressive stresses increase. The redistribution changes the
stiffness of the shell's section and eventually the shell fails because
of buckling and not because of excessive stresses. This explains
why the model with a linear material assumption fails at a higher
load than with non-linear material assumption.
To make sure the shape of the geometrical imperfection does
not affect the buckling load, the analysis is also performed for the
four different imperfection shapes (groups 1–4) and with a size
equal to the shell's thickness. The results are shown in Fig. 8.
Changing the shape of the geometrical imperfection influences the
buckling behaviour slightly and decreases the initial buckling load
with 9–19% when linear material behaviour is assumed. This has
however no physical meaning as the limit stress in tension is again
reached at 10 kN/m2 and the shells will not fail due to buckling
Fig. 10. Both geometrical imperfections and non-linear material behaviour affect instabilities. When the non-linear material behaviour is taken into
the buckling behaviour. account the buckling load decreases maximally with 5% for group
2 (8.4 kN/m2 to 8.0 kN/m2). For all four groups the imperfections
have little influence and the buckling load remains at the same
only mode 1 as geometrical imperfection, which is often done in level. For the vertically supported shell the most important factor
literature [11,25,26], is in this case not a safe approach. The size of remains thus performing a non-linear analysis which includes the
the imperfection must be assessed by the designer for every case non-linear material behaviour rather than the integration of
individually as the size depends on different parameters, like e.g. imperfections.
the manufacturing method (formwork) and the materials used. The maximum deformation of the shell at buckling – assuming
non-linear material behaviour – lies between 31.0 mm and
3.2.2. Vertically supported shell 37.0 mm ( ¼L/65  L/54), which is much larger than the applied
The same analysis is now performed on the case study 1 shell imperfection size (6.5 mm). Even when assuming linear material
with an identical geometry but with an edge which is only verti- behaviour, the deformation at the point of material failure
cally supported instead of pinned. The linear eigenvalue prediction (9.3 mm) is already larger than this imperfection size.
is not applicable for this shell, as the unrestricted shell will show Comparing the results of the pinned and vertically supported
large pre-buckling deformations and hence it might also experi- shell, which show exactly the opposite phenomenon, clearly in-
ence high stresses in the nonlinear tensile behaviour stage. dicates the importance of the shell's structural stiffness and
Fig. 7 shows the effect of introducing a geometrical imperfec- loadbearing behaviour. For a shell with a high structural stiffness
tion of 6.5 mm ( ¼shell thickness) based on buckling mode 1 – experiencing mainly compression (like the pinned shell), the de-
obtained from the linear eigenvalue prediction in Section 3.1 – and formations and tensile stresses remain small when loading the
of including non-linear material behaviour. At first sight, in- structure. This implicates a relatively small influence of consider-
troducing the imperfection slightly increases the buckling load ing geometrical non-linearity or including material non-linearity
with 2% when linear material behaviour is assumed (19.7 kN/m2 to on the buckling load prediction, however these structures are
20.0 kN/m2). However, when analysing this structure better, the highly sensitive to geometrical imperfections, corresponding to
limit design stress in tension of GFTR-IPC (40 MPa) is already [25]. If a shell with a low structural stiffness and experiencing
reached at a load of 10 kN/m2, long before the structure buckles significant bending and/or tension is considered (like the vertically
(49% lower load). At this point the shell deformed already 9.3 mm supported shell), geometrical imperfections hardly affect the
(¼ L/215), meaning the impact of material non-linearities becomes buckling behaviour because deformations and high tensile stresses
more important. When the non-linear material behaviour is as- in the nonlinear stress-strain stage are predominant, making a
sumed, including geometrical imperfections does not affect the geometrically non-linear calculation – which takes second order
buckling behaviour in a significant way. Comparing the results effects into account – and consideration of material non-linearity
between linear and non-linear material behaviour, the buckling essential.
load prediction decreases by 57% (19.7 kN/m2 to 8.4 kN/m2) if
material failure is not taken into account and is 16% lower than the
load at material failure (10.0 kN/m2 vs. 8.4 kN/m2). These results 4. Case study 2: compression-only GFTR-IPC shell
clearly prove that the non-linear material behaviour cannot be
neglected for the analysis of this vertically supported shell. The reasoning of Section 3 is now evaluated on a true scale
As already mentioned, for the linear material assumption the shell. The results of both linear and non-linear analyses are pre-
limit tensile stress (40 MPa) is already reached at a load of 10 kN/ sented in the next paragraphs.
m2 (near the edges), long before the shell buckles. The compres-
sive stresses in the shell remain limited at this point (  35.4 MPa). 4.1. Results of linear analysis
When the non-linear material behaviour is assumed, both tensile
and compressive stresses remain below the limit values at the The first four buckling modes and the accompanying eigenva-
point of buckling (8.4 kN/m2). However, the maximum tensile lues are shown in Fig. 9. Just as for the 2 m shell, these eigenvalues
stress at buckling equals 38.0 MPa (corresponds to 9090 mm/m), are closely spaced (1589 N/m2–1625 N/m2) which indicates the
which means that the material is far in the non-linear stage of the imperfection sensitivity of the shell. However, this linear analysis
stress-strain curve (Fig. 2a). The deflection of the shell at this point gives a first indication of the buckling behaviour, as all buckling
equals 36.3 mm. Assuming linear material behaviour, the model modes show the compression-only shell buckles always at the free
thus predicts the shell to fail because of excessive tensile stresses edges, where the structure is the least stiff.
 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29 27

Fig. 11. Generally, both the imperfection shape and size have little influence on the buckling load. The deformation is always predominant. (a) Group 1 linear material.
(b) Group 1 non-linear material. (c) Group 2 linear material. (d) Group 2 non-linear material. (e) Group 3 linear material. (f) Group 3 non-linear material.

4.2. Results of non-linear analysis analysis. The linear results are not representative as the eigenvalue
prediction serves to examine stiff structures with a response
To investigate separately the impact of considering geometrical which usually involves very little deformation prior to buckling
nonlinearity, geometrical imperfections and material nonlinearity, [23]. This compression-only shell is clearly not a stiff structure
the same six cases as for case study 1 are analysed (see Fig. 10). The which undergoes finite deformations before instabilities actually
introduced geometrical imperfection equals the shell's thickness occur, meaning the parameters of the shell change along with its
(12.5 mm) and is based on the first buckling mode of the eigen- deformation. For this reason only the results of the non-linear
value prediction. Strikingly, the results of the linear analysis give a analysis will further be discussed.
buckling load which is lower than the geometrically non-linear Taking a first look at the results in Fig. 10, the consideration of
28 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29

Clearly, the deformation at the point of buckling is very large, i.e.

between 205.6 mm ( ¼L/49) and 233.1 mm ( ¼L/43) and pre-
dominant to the imperfections. Strikingly, all curves (except group
1) follow the same trend and at a load of approximately 1650 N/m2
the stiffness of the structure decreases significantly as was seen in
Fig. 10. This stiffness reduction corresponds to a deformation of the
edges (around 60 mm) which are in compression. The load on the
shell can be further increased until the buckling load is reached
and the corners buckle.
For group 1 though, considering an imperfection size of 75%
and 100% of the shell's thickness does reduce the buckling load
with respectively 13% and 17% for the linear material behaviour
assumption and with 12% for the non-linear material behaviour
assumption. The deformations at point of buckling are also large,
i.e. between 160.8 mm ( ¼L/62) and 212.1 mm (¼ L/47). The failure
takes place at the edges, however, there is one edge which de-
forms differently than the others. As Fig. 12 shows, both for the
linear material behaviour (Fig. 12a) and non-linear material
(Fig. 12b) assumption, only the imperfection shape of group
1 seems to influence the predicted buckling load. For all other
groups a similar trend is observed and a geometrical imperfection
hardly affects the buckling load.
Comparing the results of the models with and without con-
sidering the non-linear material behaviour, the buckling load
prediction is maximally 12% lower when considering material
Fig. 12. Except for group 1, introducing a geometrical imperfection hardly affects
the predicted buckling load. (a) Linear material. (b) Non-linear material.
nonlinearity, while the deformations are similar.
Comparing these results, the large deformations imply that
non-linear material behaviour and more importantly, the geome-
both the non-linear material behaviour and the geometrical im- trically non-linear calculation should be used. However, as group
perfections affect the buckling load prediction. Introducing the 1 and Fig. 10 show, the shape and size of the geometrical im-
geometrical imperfection reduces the buckling load with 12% perfections do have an influence on the structure and can reduce
when assuming linear material behaviour and with 10% when the buckling load prediction. Therefore, the geometrical im-
assuming non-linear material behaviour. Including the non-linear perfections cannot be ignored for this case study and the most
material behaviour decreases the buckling load with 9% and con- critical shape must be determined.
sidering both material nonlinearity and geometrical imperfections
decreases the buckling load prediction with 18%.
A more elaborated study is now performed to examine the 5. Conclusions
influence of the material non-linearity as well as the size and
shape of the geometrical imperfection. To vary in shape, three This paper presented the numerical analysis of the buckling
different linear combinations of buckling modes are studied. The behaviour of two thin-walled GFTR-IPC shells subjected to uni-
first two groups consist of linear combinations of the buckling formly distributed vertical loading. During this analysis, the in-
modes where the value of the eigenvalue is very closely spaced fluence of performing a geometrically non-linear analysis, in-
within one group. The third group consists of a linear combination cluding material non-linearities and introducing geometrical im-
where the shapes of the buckling modes are similar and the perfections is examined. As the results of the two cases with dif-
combination of these imperfection shapes fortify each other (on ferent boundary conditions showed, the buckling behaviour is
certain locations). All groups are studied with the geometrically highly depending on the structural stiffness and loadbearing be-
non-linear analysis, considering either linear or nonlinear material haviour of the shell.
behaviour. The groups are: For stiff shells experiencing mainly compression, like the syn-
clastic shell (2 m diameter, 0.2 m height, 6.5 mm thick) with re-
 group 1 ¼[λ1, λ2] ¼ [1589 N/m2, 1590 N/m2] and Δx1,j ¼ ω1,j stricted boundary conditions, the results first proved the im-
 ∑i∈(1,2)ui portance of performing a geometrically non-linear analysis as the
 group 2 ¼ [λ3, λ4] ¼[1617 N/m2, 1625 N/m2] and Δx2,j ¼ ω2,j buckling load was reduced by 12% for this shell. However, espe-
 ∑i∈(3,4)ui cially the influence of including geometrical imperfections in the
 group 3 ¼[λ4, λ5]¼ [1625 N/m2, 1705 N/m2] and Δx3,j ¼ ω3,j model is significant. The larger the imperfection the more the
 ∑i∈(4,5)ui buckling load is reduced, with a maximum decrease of 72% for a
geometrical imperfection equal to the shells thickness. The buck-
with Δx the imperfection size, ui the shape of the normalised ling load is moreover not only depending on the imperfection size,
buckling mode (eigenvector), ωk,j the parameter which scales the but also on the imperfection shape, meaning the most critical
buckling modes to obtain the desired imperfection size and de- shape must be determined. The consideration of material non-
fined by Formula (4). Fig. 11 shows the applied load as a function of linearity had little influence on the buckling load prediction of the
the displacement for the compression-only GFTR-IPC shell for the pinned shell. These results are analogue to the observations on
different groups and with imperfection sizes Δx equal to 0%, 50%, traditional steel reinforced concrete shells.
75% and 100% of the shell's thickness. For very slender GFTR-IPC shells with a lower structural stiff-
Except for group 1, including an imperfection has no influence ness, like the vertically supported shell and the compression-only
on the buckling behaviour (maximum difference of 3%) regardless shell (10
10 m span, 4.8 m height, 12.5 mm thick), the geome-
whether linear or non-linear material behaviour is assumed. trical imperfections proved to have a much smaller impact. As the
 E. Verwimp et al. / Thin-Walled Structures 108 (2016) 20–29 29

deformations were always very large (a lot larger than the applied Spat. Struct. 45 (144) (2004) 131–148.
imperfection size) a linear calculation is not suitable and a geo- [11] E. Dulacska, L. Kollar, Design procedure for the buckling analysis of reinforced
concrete shells, Thin Walled Struct. 23 (1995) 313–321.
metrically non-linear analysis, which takes second order effects [12] M. Andres, R. Harte, Buckling of concrete shells: a simplified numerical ap-
into account, must be used instead. Moreover, because of the large proach, J. Int. Assoc. Shell Spat. Struct. 47 (3) (2006).
deformations and high occurring tensile stresses, material non- [13] L. Chen, Buckling of Circular Steel Cylindrical Shells under Different Loading
Conditions (Doctoral thesis), The University of Edinburgh, Edinburgh, Scot-
linearities also become more important. This was especially the
land, 2011.
case for the vertically supported shell, where the limit stress in [14] ECCS EDR5, European Recommendations for Steel Construction: Buckling of
tension was already reached at a much lower load (49%) than the Shells, 5th ed., J.M. Rotter, H. Schmidt (Eds.), Brussels, Belgium, 2008.
buckling load. [15] Z.T. Chang, M.A. Bradford, R.I. Gilbert, Short-term behaviour of shallow thin-
walled concrete dome under uniform external pressure, Thin-Walled Struct.
Conclusively, when designing thin GFTR-IPC shells their lower 49 (2011) 112–120.
structural stiffness requires that a non-linear analysis must always [16] E. Verwimp, T. Tysmans, M. Mollaert, S. Berg, Experimental and numerical
be performed, and the results should be taken into account early buckling analysis of a thin TRC dome, Thin-Walled Struct. 94 (2015) 89–97.
[17] H. Schek, The force density method for form finding and computation of
in the design process. As such, the results presented in this paper
general networks, Comput. Methods Appl. Mech. Eng. 3 (1) (1974) 115–134.
highlight some important issues in the design and assessment of [18] X. Wu, J. Gu, Inorganic Resin Composites, Their Preparation and Use Thereof,
thin TRC shells and should sensitise designers of thin TRC shell European Patent EP 0 861 216 B1, 29.05.1997.
structures. [19] O. Remy, Lightweight Stay-In-Place formwork: “A Concept for Future Building
Applications” (Doctoral thesis), Vrije Universiteit Brussel, Faculty of En-
gineering, Department of Mechanics of Materials and Constructions, Brussel,
Belgium, 2012.
References [20] H. Cuypers, Analysis and Design of Sandwich Panels with Brittle Matrix
Composite Faces for Building Applications (Doctoral thesis), Vrije Universiteit
Brussel, Faculty of Engineering, Department of Mechanics of Materials and
[1] E. Engberts, Large-size façade elements of textile reinforced concrete, in: Constructions, Brussel, Belgium, 2002.
Proceedings of International RILEM Symposium Textile Reinforced Concrete, [21] W. Brameshuber, State-of-the-Art Report of RILEM Technical Committee TC
Aachen, Germany, 2006, pp. 309–318. 201-TRC: Textile Reinforced Concrete, RILEM Publications, Bagneux, France,
[2] C. De Roover, J. Vantomme, J. Wastiels, K. Croes, H. Cuypers, L. Taerwe,
2006.
H. Blontrock, Modelling of an IPC-concrete modular pedestrian bridge, Com-
[22] F. Frey, M. Studer, Analyse des Structures et Milieux Continus: Coques, Vol. 5,
put. Struct. 80 (27–30) (2002) 2133–2144.
Presses Polytechniques et Universitaires Romandes, Lausanne, Switzerland,
[3] C. Hiel, Improving the reliability of the electric grid infrastructure: case study
2003.
of firewalls manufactured with fiber reinforced inorganics (FRI), in: Proceed-
[23] Abaqus 6.12–1 Documentation, Abaqus analysis users manual – Chapter 6:
ings of the American Society for Composites, Newark, Delaware, US, 2009.
Analysis procedures, solution, and control, Section 6.2.3: Eigenvalue Buckling
[4] N. Cauberg, T. Tysmans, S. Adriaenssens, J. Wastiels, M. Mollaert, B. Belkassem,
Prediction, 2012.
Shell elements of textile reinforced concrete using fabric formwork: a case
[24] Abaqus 6.12–1 Documentation, Abaqus analysis users manual – Chapter 6:
study, Adv. Struct. Eng. 15 (4) (2012) 677–689.
[5] D. Ehlig, F. Schladitz, M. Frenzel, M. Curbach, Textile concrete: an overview of Analysis procedures, solution, and control, Section 6.2.4: Unstable Collapse
executed projects, Beton- und Stahlbetonbau 107 (2012) 777–785. and Postbuckling Analysis, 2012.
[6] A. Scholzen, R. Chudoba, J. Hegger, Thin-walled shell structure made of textile- [25] R. Reitinger, E. Ramm, Buckling and imperfection sensitivity in the optimi-
reinforced concrete. Part I: structural design and realization, Struct. Concr. 16 zation of shell structures, Thin-Walled Struct. 23 (1–4) (1995) 159–177.
(1) (2015) 106–114. [26] E. Ramm, W.A. Wall, Shell structures – a sensitive interrelation between
[7] A. Scholzen, R. Chudoba, J. Hegger, Thin-walled shell structure made of textile physics and numerics, Int. J. Numer. Methods Eng. 60 (2004) 381–427.
reinforced concrete. Part II: experimental characterization, ultimate limit state [27] European Committee for Standardisation, Eurocode 3: Design of steel struc-
assessment and numerical simulation, Struct. Concr. 16 (1) (2015) 115–124. tures: Part1–6: Strength and Stability of Shell Structures, CEN, 2004.
[8] T. Tysmans, S. Adriaenssens, J. Wastiels, O. Remy, Textile reinforced cement [28] Abaqus 6.12–1 Documentation, Abaqus analysis users manual – Chapter 11:
composites for the design of very thin saddle shells: a case study, in: Pro- Special-purpose techniques, Section 11.3.1: Introducing a Geometric Im-
ceedings of the 18th International Conference on Composite Materials, Jeju perfection into a Model, 2012.
Island, South Korea, 2011. [29] T. Tysmans, M. Wozniak, O. Remy, J. Vantomme, Finite element modelling of
[9] E. Verwimp, T. Tysmans, M. Mollaert, Reinforcing concrete shells with cement the biaxial behaviour of high-performance fibre-reinforced cement compo-
composite stay-in-place formwork: numerical analysis of a case study, in: sites (HPFRCC) using concrete damaged plasticity, Finite Elem. Anal. Des. 100
Proceedings of the International Conference on High Performance and Opti- (2015) 47–53.
mum Design of Structures and Materials, Ostend, Belgium, 2014. [30] Abaqus 6.12–1 Documentation, Abaqus analysis users manual – Chapter 27:
[10] S.J. Medwadowski, Buckling of concrete shells: an overview, J. Int. Assoc. Shell Elements, Section 27.1.1: Element Library: Overview, 2012.