Finite element formulation for inflatable beams.

Anh Le Van, Christian Wielgosz

To cite this version:

Anh Le Van, Christian Wielgosz. Finite element formulation for inflatable beams.. Thin-Walled
Structures, Elsevier, 2007, 45 (2), pp.221-236. �10.1016/j.tws.2007.01.015�. �hal-01006899�

HAL Id: hal-01006899

https://hal.archives-ouvertes.fr/hal-01006899
Submitted on 6 Oct 2016

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.

Distributed under a Creative Commons Attribution| 4.0 International License

 1

Finite element formulation for inflatable beams

Anh Le van, Christian Wielgosz
GeM—Institute of Research in Civil and Mechanical Engineering, Faculty of Sciences in Nantes, 2, rue de la Houssiniere,
BP 92208, Nantes 44322 Cedex 3, France

The discretized nonlinear equations for bending and buckling of inflatable beams are written by use of the virtual work principle with
Timoshenko’s kinematics, finite rotations and small strains. The linearized equations around a pre-stressed reference configuration are
then deduced, giving rise to a new inflatable beam finite element. The stiffness matrix contains the shear coefficient and the internal
pressure. Use is made of the particular 3-node beam element to investigate the bending and the buckling of a cantilever beam, the
deflection of a pinched torus and the buckling of a torus submitted to a radial compressive force. The numerical results obtained with the
beam element are shown to be close to analytical and three-dimensional (3D) membrane finite element results. The validity of the
numerical results is discussed, in connection with the concepts of the crushing force or the wrinkling pressure of the inflated beam.

Keywords: Inflatable beam; Thin-walled structure; Membrane structure; Follower force; Finite element

1. Introduction

Inflatable beam structures have several interesting properties: they are lightweight and easily folded. The internal
pressure yields tensile pre-stressing in the fabrics and implies a significant bearing capacity, especially when the pressure is
high. First analytical studies on inflatable beams are not new. Comer and Levy [1] studied an inflatable cantilever beam
using the usual Euler–Bernoulli kinematics. Fichter [2] gave the first solution in which the internal pressure appears in the
deflection expressions. Main et al. [3,4] made experiments on a cantilever and improved Comer’s theory. More recently,
Wielgosz and Thomas [5,6] have derived analytical solutions for inflated panels and tubes by using the Timoshenko
kinematics and by writing the equilibrium equations in the deformed state of the beam in order to take into account the
follower force effect of the internal pressure. They have obtained new expressions for the deflection and shown that the
limit load is proportional to the applied pressure and that the deflections are inversely proportional to the material
properties of the fabrics and to the applied pressure, thus improving Fichter’s theory. Lately, Le van and Wielgosz [7] have
used the virtual work principle with finite displacements and rotations in order to derive the nonlinear equations for
inflatable beams; the linearized equations once again improve Fichter’s theory.
Since analytical solutions are only viable for isostatic beams, finite elements are required for computing beam networks.
Their development seems to be a recent topic of research; Wielgosz and Thomas [8] have constructed a finite element
devoted to inflatable panels. This element including the inflation pressure effect was obtained by the equilibrium finite
element method, according to which the global compliance is the sum of the yarn and beam compliances. This element has
been converted in a displacement one in order to be easily used in usual finite element softwares. A finite element for tubes
can also be found in [6], where the equilibrium method has once again been used, giving rise to a non-symmetrical stiffness
matrix depending on the pressure, which is suitable for lightly ventilated structures.
 2

The aim of this paper is to derive a new displacement finite element for inflatable beams directly from the virtual work
principle. The result will be a symmetrical stiffness matrix which holds for airtight tubes.
In the first section of the paper, the finite element is derived for the static in-plane stretching and bending problem of an
inflated cylindrical beam made of a membrane. First, the discretized nonlinear equations will be written by use of the
virtual work principle with Timoshenko’s kinematics, finite rotations and small strains. The procedure will enable one to
correctly account for the shear effect and the pressure in the governing equations. Then, the tangent stiffness matrix will be
derived as the sum of a tangent matrix due to the internal strain work and another tangent matrix due to the internal
pressure. Whereas it is essential to assume finite rotations in order to correctly exhibit all the terms due the internal
pressure, it is sensible to use the inflatable beam in small deformations only. Thus, the nonlinear problem will be linearized
around a well-defined reference configuration. This will lead to the inflatable beam finite element involving the Timoshenko
shear coefficient and the inflation pressure. The particular case of a 3-node beam finite element with quadratic shape
functions will be considered.
The 3-node beam finite element will be used in the next sections to deal with the bending and the buckling of a cantilever
inflatable beam, the bending of a pinched toroidal beam and the buckling of a toroidal beam submitted to a radial
compressive force. In every case, the finite element results will be compared with the analytical or 3D membrane finite
element results, showing the good performance of the proposed finite element. Also, the validity of the numerical results
will be discussed, in connection with the concepts of the crushing force or the wrinkling pressure of the inflated beam.
The formulation and results presented in this paper are the finite element counterpart of Ref. [7].

2. The finite element for an inflatable beam

Consider an inflatable beam made of a cylindrical membrane undergoing axial stretch and bending in the xy-plane
(Fig. 1) under the combined action of an internal pressure and other external dead loads. In the reference (or initial)
configuration, the length of the beam is Lo, the cross-section area S0, the second moment of area I0, and all the centroids Go
of the cross-sections lie on the x-axis. The reference thickness of the membrane is ho.
Remark. For inflatable structures, it should be emphasized that the loading is applied in two successive stages: first, the
beam is inflated to pressure p, and then the other external forces are applied. At the beginning of the first stage, the internal
pressure is zero and the beam is in a natural (or stress free) state. On the other hand, the reference configuration, which
corresponds to the end of the first stage (before external forces other than the internal pressure are applied), is in a pre-
stressed state. To clearly distinguish between the two states, we shall use index + to denote the quantities in the natural
state, as opposed to the usual index o for quantities related to the reference configuration. Thus, L+ and h+ designate the
natural length and thickness; while Lo and ho the reference ones. Computing Lo and ho corresponding to a given pressure
may be difficult depending on the problem at hand, but this is an independent subject which actually is common to every
problem with a pre-stressed reference configuration.
In order to derive the discretized equations for the inflatable beam, use will be made of the principle of virtual work in
the three-dimensional (3D) Lagrangian form
8 virtual displacement field V ,
Z Z Z
 PT : grad V dO0 þ f o  V dO0 þ V  P  N dS0 ¼ 0, ð1Þ
O0 O0 @O0

where O0 is the 3D region occupied by the beam in the reference configuration, qO0 its boundary, P ¼ FR and R the first
(non-symmetric) and second (symmetric) Piola–Kirchhoff stress tensors (F is the deformation gradient), fo the body force
per unit reference volume and N the unit outward normal in the reference contribution. Following the Timoshenko beam
model, the displacement of any material point Po(X, Y, Z) in the beam is given by:
UðPo Þ ¼ U þ RðyÞ  Go Po , (2)

y
θ
G
V x
O
Go
X Lo, So, Io

Fig. 1. The inflated beam model.

 3

where U(X) ¼ (U(X), V(X), 0) is the displacement of the centroid of a current cross-section at abscissa X (all the
components are related to the base (x, y, z), Fig. 1), y(X) the finite rotation of the cross-section and R(y) the rotation
tensor. The virtual displacement of a current material point Po is chosen as

V ðPo Þ ¼ V þ X  GP, (3)

where V*(X) ¼ (U*(X), V*(X), 0) is the virtual displacement of the centroid of a current cross-section, X*(X) ¼ (0, 0, y*(X))
is the virtual rotation. Note that relation (3) involves the current position GP, not the initial position GoPo, like the real
velocity field in rigid body mechanics. The material is assumed to be hyperelastic and isotropic, obeying the Saint-Venant
Kirchhoff constitutive law. The axial and shear components of the second Piola–Kirchhoff stress are then related to the
components of the Green strain tensor by
 
E
SXX ¼ S0XX þ E  E XX SXX ¼ S0XY þ 2G  E XY G¼ , (4)
2ð1 þ nÞ

where E is the Young modulus, n the Poisson ratio and S0XX ; S0XY are the initial stresses induced by the preliminary inflation
of the beam.

2.1. Discretization of the internal virtual work

As shown in [7], the internal virtual work in relation (1) can be recast as
Z Z Lo  
    
PT : grad V  dO0 ¼ N 1 þ U ;X þ M cos yy;X  T sin y U ;X þ NV ;X þ M sin yy;X þ T cos y V ;X
O0 0
       
þ M  1 þ U ;X sin yy;X þ V ;X cos yy;X  T 1 þ U ;X cos y þ V ;X sin y y
Z
   
þ ½M 1 þ U ;X cos y þ V ;X sin y þ Y 2 S;XX dS0 y;X y;X dX ð5Þ
S0

where the generalized stresses—the (material) axial force, the shear force and the bending moment acting on the reference
cross-section S0—are given by
Z  
o 1 2 1 2 Io 2
N¼ SXX dS0 ¼ N þ ES 0 U ;X þ U ;X þ V ;X þ y
S 2 2 2S 0 ;X
Z 0
T¼ SXY dS0 ¼ T o þ GS 0 ðV ;X cos y  ð1 þ U ;X Þ sin yÞ,
S0
Z
M¼  Y SXX dS 0 ¼ M o þ EI 0 ðð1 þ U ;X Þ cos y þ V ;X sin yÞy;X ð6Þ
S0

In relations (6), No, To and Mo are the resultants of the initial stresses on the cross-section. In practice, the coefficient GS0
in (6b) is replaced by kGS0, where the so-called correction shear coefficient k is determined from the shape of the cross-
section. The value usually found in the literature (see, e.g., [9]) for circular thin tubes is k ¼ 0.5. By assuming that the initial
stress S0XX takes the following form general enoughRin practical purposes: S0XX ¼ a0 þ b0 Y þ g0 Y 2 and that S0YY does not
depend on Y, it was shown in [7] that the integral S0 Y 2 S;XX dS 0 in (5) is
Z
NI 0 1
Y 2 S;XX dS 0 ¼ þ KEy2;X þ Kgo , (7)
S0 S0 2
R
where K ¼ S0 Y 4 dS0  I 20 =S 0 is a quantity depending on the initial geometry of the cross-section, like S0 and I0.
This quantity is involved in the nonlinear theory only, not in the linearized theory, as will be seen later. In order to
discretize expression (5), one considers the beam as a finite element of length L0, and interpolates the displacements and the
rotation by

U ¼ hN U ifUge V ¼ hN V ifV ge y ¼ hN y ifyge , (8)

where the brackets /S designate row-vectors and the braces { } column-vectors. The row-vectors /NUS, /NVS, /NyS
are the shape functions for the displacements (U, V) and the rotation y. The column-vectors {U}e, {V}e and {y}e contain the
nodal displacements and rotations (the superscript e stands for ‘element’). The virtual quantities U*, V*, y* are
 4

interpolated in the same way. The virtual internal work in the element writes
8 9
> A>
Z Z > >
> >
Lo D E< B=
PT : grad V dO0 ¼ U ;X V ;X y y;X dX , (9)
Oo 0 >C>
> >
>
: ; >
D

where
8 9
8 9 > Nð1 þ U ;X Þ þ M cos yy;X  T sin y >
> A > >
> >
>
>
> >
> >
> NV þ M sin yy þ T cos y >
>
<B= < ;X ;X =
¼ Mð1 þ U ;X Þ sin yy;X þ MV ;X cos yy;X  T½ð1 þ U ;X Þ cos y þ V ;X sin y (10)
>
> C>> >> >
>
: >
> ; >> NI 0 1 >
>
D >
: Mð1 þ U ;X Þ cos y þ MV ;X sin y þ 3 o
y;X þ EKy;X þ Kg y;X >
;
S0 2

and
8  9 2 3
>
> U ;x >> 8  9e hN U ;x i
> >
< V ;x >
> = <U >
> = 6
6 hN V ;x i
7
7
¼ ½G V  ¼ ½GfU ge ; ½G ¼ 6
6
7. (11)
hN y i 7

>
> y >
> >
: ; > 4 5
>
> >
: y >; y
hN y;x i
;x

The column-vector {U*}e gathers all the degrees of freedom (U*, V*, y*) of the element. Eventually, the virtual internal
work is discretized as
Z
PT : grad V dO0 ¼ hU ie fWge (12)
O0

with
8 9
> A>
>
> >
Z Lo <B> =
f W ge ¼ ½GT dX (13)
0 >
> C>>
: >
> ;
D

denoting the vector of element internal force.

2.2. Discretization of the external virtual work due to the internal pressure

Before external dead loads are applied, the beam is inflated by an internal pressure p  n (where n is the outward unit
normal in the current configuration), which gives the beam its bearing capacity. The contribution of the internal pressure is
determined under the assumption that the reference volume O0 is a circular cylinder of radius Ro and thickness ho 5Ro .
This assumption, combined with the Timoshenko kinematics (2), implies that any cross-section remains a circular disk of
radius Ro during the deformation. This means that the change in shape (ovalization or warping) of the cross-section is not
taken into account, as is the case of all usual beam models. The following expression for the virtual work of the pressure is
obtained in [7]
Z Lo
  
W pressure ¼ P U sin yy;X  V  cos yy;X þ y ðcos yV ;X  sin yð1 þ U ;X ÞÞ dX
0
þ P½U  cos y þ V  sin yL0 o , ð14Þ
where P ¼ ppR2o .
It should be mentioned that expression (14) is obtained since the current vector GP is involved in the
virtual displacement (3) instead of the initial one GoPo. In order to get a symmetrical expression for the tangent stiffness
 5

matrix, let us recast the bracketed terms ½ L0 o in relation (14) as follows:
Z Lo Z Lo
½U  cos yL0 o ¼ ðU  cos yÞ;X dX ¼ ðU ;X cos y  U  sin yy;X Þ dX , (15a)
0 0
Z Lo Z Lo
½V  sin yL0 o ¼ ðV  sin yÞ;X dX ¼ ðV ;X sin y þ V  cos yy;X Þ dX . (15b)
0 0

Hence,
Z Lo
W pressure ¼ P ½U ;X cos y þ V ;X sin y þ y ðV ;X cos y  sin y  U ;X sin yÞ dX : (16)
0

Eventually, relation (14) is discretized as

 e
W pressure ¼ hU ie Upressure ðUe Þ , (17)
where
8 9
> cos y >
Z >
> >
>
 e Lo < sin y =
Upressure ðUe Þ ¼P ½GT dX . (18)
0 >
> V ;X cos y  sin y  U ;X sin y >
>
>
: >
;
0

2.3. Tangent stiffness matrix

The element tangent stiffness matrix is defined by

 e
qfWge q Upressure
½Ke ¼  , (19)
qfUge qfU ge
where vectors {W}e and {Upressure}e are given by (13) and (18). The first matrix in the right-hand side of (19) is standard:
2 3
e Z Lo ES 0
qfWg 6 7 T
¼ ½GT ð½S þ ½T4 EI 0 5½T Þ½G dX , (20)
qfUge 0
kGS0
where the symmetrical square matrix [S] and the rectangular matrix [T] are
2 3
N 0 M sin yy;X  T cos y M cos y
6 7
6 N M cos yy;X  T sin y M sin y 7
6 7
6 7
6 M½ð1 þ U ;X Þ cos y þ V ;X sin yy;X 7
6 M½ð1 þ U ;X Þ sin y þ V ;X cos y 7
½S ¼ 6 7.
6 þT½ð1 þ U ;X Þ sin y  V ;X cos y 7
6 7
6 7
4 NI 0 3 5
symmetrical þ EKy2;X þ Kg0
S0 2
2 3
1 þ U ;X cos yy;X  sin y
6 7
6 V ;X sin yy;X cos y 7
6 7
6 7
½T ¼ 6 0 ½ð1 þ U ;X Þ sin y þ V ;X cos yy ;X ð1 þ U ;X Þ cos y  V ;X sin y 7. ð21Þ
6 7
6 7
4 I0 5
y;X ð1 þ U ;X Þ cos y þ V ;X sin y 0
S0
The second matrix in the right-hand side of (19) is the element stiffness matrix due to the internal pressure and is essential
for inflatable structures. After algebraic manipulations, one gets
2 3
0 0 P sin y 0
 e Z L 60
q Fpressure o
T6
0 P cos y 077
¼ ½G 6 7½G dX , (22)
qfU ge oe 4 P sin y P cos y PðV ;X sin y þ cos y þ U ;X cos yÞ 0 5
0 0 0 0
 6

which is a symmetrical matrix. Eventually, the global stiffness matrix is symmetrical and contains the material and the
geometric matrices, including the presence of the internal pressure.

2.4. General stiffness matrix for the linearized problem

In engineering practice, the inflatable beam should be used in small deformations only. In this subsection, the nonlinear
equations obtained above are linearized in order to get the stiffness matrix at the equilibrium state. It is noteworthy that it
is crucial to derive the nonlinear equations first, assuming finite displacements and rotations, as done in the previous
section, and then to carry out the linearization. Otherwise, if the linearization were carried out too early, some of the terms
due the internal pressure would be missing. The linearization of the discretized equations will be performed about the
reference configuration, which is in a pre-stressed state. For instance, in the case of bending, the pre-stress is due to the
preliminary inflation of the beam; in the case of buckling, the pre-stress also includes the compressive load. The following
assumptions are made on the initial stresses:

(i) The initial axial stress S0XX is constant over the cross-section. Thus, Mo in (6c) and go in (7) are zero.
o
(ii) The initial shear stress S0XY is zero. Thus, T in (6b) is zero, too.

By taking these assumptions into account and making U(X) ¼ V(X) ¼ y(X) ¼ 0 in the tangent stiffness matrices
(20)–(22), taking the definition (11) for matrix [G] into account, one gets the element stiffness matrix for the linearized
problem:
2  3
ðES 0 þ N o Þ N U;X hN U;X i
6
 7
  Ny
Z Lo 6
6 ðkGS 0 þ N o Þ N V ;X hN V ;X i ðkGS 0 þ PÞ N V ;X
7
7
e
½K ¼ 6 Lo 7 dX .
6  
 7
o 6 Ny o
ðE þ N =S 0 ÞI 0  Ny Ny 7
4 ðkGS 0 þ PÞ hN V ;X i N hN i þ ðkGS þ PÞ 5
y;X y;X 0
Lo L2o Lo Lo

(23)

2.5. Particular case of the 3-node beam finite element

All the subsequent numerical computations will be carried out using a 3-node element, with 3 degrees of freedom at each
node (axial displacement U, deflection V and rotation y). The three degrees of freedom are interpolated by the same
quadratic shape function:

hNi ¼ h12xðx  1Þ 1  x2 12xðx þ 1Þi x 2 ½1; þ1: (24)

Evaluating (23) with two integration points, at x ¼ p1ffiffi3, leads to the stiffness matrix for a 3-node beam element of length
Lo, as shown in expression (25). It should be noted that the two point integration corresponds to a reduced integration of
the term {Ny}/NyS and the full integration of the other terms. In the classical beam theory, P is zero and the reduced
integration provides exact solution when kGSo tends to infinity:.
2 3
7
3ðES 0 þ N oÞ 83ðES 0 þ N o Þ 1
3ðES 0 þ N oÞ
6 7 o 1 o o 7
6 3ðkGS 0 þN Þ 2ðkGS 0þ PÞ 83 ðkGS 0 þ N Þ 2
3ðkGS 0 þ PÞ 1
3ðkGS 0 þN Þ 16ðkGS 0þ PÞ 7
6 7
6 7 ðE þ N o =S 0 ÞI 0 8 ðE þ N o =S0 ÞI 0 1 ðE þ N o =S 0 ÞI 0 7
6 23ðkGS 0 þ PÞ  1
þ PÞ 7
6 6ðkGS 0 7
6 3 L2o 3 L2o 3 L2o 7
6 7
6 þ19ðkGS 0 þ N Þ o
þ19ðkGS 0 þ PÞ 181
ðkGS 0 þ PÞ 7
6 7
6 7
6 16
3 ðES 0 þ NoÞ 83ðES 0 þ N o Þ 7
6 7
6 16
þ N oÞ 0 83ðkGS 0 þ N o Þ 2
PÞ 7
6 3 ðkGS 0 3ðkGS 0 þ 7
16 7
½K ¼ 6
e 16 ðE þ N o =S 0 ÞI 0 o
8 ðE þ N =S 0 ÞI 0 7
7:.
Lo 6
6 23ðkGS 0 þ PÞ  7
6 3 L2o 3 L2o 7
6 7
6 þ49ðkGS 0 þ PÞ þ19ðkGS 0 þ PÞ 7
6 7
6 7
þ N oÞ 7
6 3ðES 0 7
6 7
6 7
ðkGS 0 þ N o Þ 1
2ðkGS 0 þ PÞ 7
6 3 7
6
6 7 ðE þ N o =S 0 ÞI 0 7
7
6 symmetric 2 7
6 3 Lo 7
4 5
þ19ðkGS 0 þ PÞ

(25)
 7

3. Example 1. Bending of an inflatable cantilever beam

Consider the linearized problem of an inflated cantilever beam under bending. The beam is made of a cylindrical
membrane, its reference length is ‘o, its reference radius Ro and its reference thickness ho. The beam is built-in at end X ¼ 0,
subjected to an internal pressure p and a external force Fy at end X ¼ ‘0 (see Fig. 4 and replace there the compressive force
by the transverse load Fy).

3.1. Comparison with analytical results

Let us solve the problem using one single 3-node finite element (25). Here, the initial axial force is No ¼ P. Since the axial
displacement U is zero, we will confine ourselves to the degrees of freedom (V, y). Taking into account the boundary
conditions V1 ¼ y1 ¼ 0 yields the nodal deflections and rotations
5F ‘3o F ‘o F ‘3o F ‘o
V2 ¼ þ ; V3 ¼ þ ,
48ðE þ P=S 0 ÞI 0 2ðkGS 0 þ PÞ 3ðE þ P=S0 ÞI 0 kGS0 þ P
3F ‘2o F ‘2o
y2 ¼ ; y3 ¼ . ð26Þ
8ðE þ P=S 0 ÞI 0 2ðE þ P=S0 ÞI 0
Relations (26) coincide with the analytical solution proposed in [7]. The deflections and rotations are nonlinear functions of
the pressure in two ways: First, the pressure is involved in the denominators through P; second, the reference dimensions
‘o, S0 and I0 themselves depend on the pressure, as mentioned in Section 2.
Relations (26) are similar to those obtained by Fichter [2], apart from the additional term P=S 0 ¼ ppR2o =ð2pRo ho Þ ¼
pRo =ð2ho Þ. In some cases, this term may cause noticeable differences between Fichter’s theory and the inflatable beam
model proposed here. For instance, assume that p ¼ 2  105 Pa, Ro ¼ 0.1 m, ho ¼ 50  106 m and E ¼ 2.5  109 Pa, then
the term P/So in (26) is equal to 8% of the Young’s modulus E and is thus not negligible compared to E.
If the internal pressure is zero, then P ¼ 0 and (26) coincides with the nodal values derived from the well-known
analytical solution for a Timoshenko beam. However, contrary to a classical beam, here the inflatable beam is made
of a membrane, so the pressure cannot be equal to zero for the beam not to collapse. This fact will be discussed below in
connection with the validity of the solution. The influence of the internal pressure on the beam response is clearly
shown in the previous relations: the inflation amounts to strengthen the Young’s modulus and the shear modulus. In
particular, when pressure p tends to infinity, so do the equivalent material properties and the deflection and the rotation
tend to zero.

3.2. Limit of validity of the results

Of course, one has to check a posteriori that the obtained deflection and rotation satisfy the small deformation
hypothesis required by the linerization process. Yet there is another condition for the solution to hold: since the inflatable
beam is made of a membrane, the principal stresses at any point in the beam must be non-negative. This amounts to
satisfying the following inequality (see, e.g., [1,10–13] for more details):
pR3o p
Fp . (27)
2‘o
Recall that the reference length ‘o and radius Ro are (increasing) functions of the pressure p. Inequality (27) shows that,
given a force F, the internal pressure must be high enough for the bending solution to be meaningful. There exists a lower
bound for the pressure, referred to as the wrinkling pressure of the beam, below which a wrinkle appears first at the point
(X ¼ 0, Y ¼ Ro, Z ¼ 0) and the bending solution is no longer valid. For a given force F, Relation (27) is a rather intricate
nonlinear inequality with the wrinkling pressure p as unknown. A first-order approximation can be obtained by making in
this inequality ‘oE‘+ and RoER+, which leads to the following simple bound where the right-hand side is known:
2F ‘+
pXpwrinkling  . (28)
pR3+
Table 1 gives the wrinkling pressures for F ¼ 1 N and for a set of typical values for natural radius R+ and natural length ‘+.
Eventually, note that the moment at which the wrinkling occurs, which can be obtained from equality (28),
M wrinkling F ‘+ pðpR3+ p=2Þ, is about 1.5 times less than the collapse moment given in [5], namely M collapse ¼
ðp=2ÞðpR3+ p=2Þ (at the collapse state, an increase in deflection does not result in an increase in moment) For a survey of
various expressions for the collapse bending moment, see [14].
 8

Table 1
Wrinkling pressure pwrinkling (N/m2) given by relation (28) with F ¼ 1 N

Natural radius, R+ (m) Natural length ‘+ (m)

0.65 0.90 1.15

0.04 6466 8952 11439
0.06 1916 2653 3389
0.08 808 1119 1430

Table 2
Data for the inflatable cantilever beam

Natural thickness, h+ 125  106 m

Young modulus, E 2.5  109 N/m2
Poisson ratio, n 0.3
Correction shear coefficient, k 0.5

Natural radius, R+ (m) 0.04 0.06 0.08

Natural length, ‘+ (m) 0.65 0.90 1.15
Pressure, p (N/m2) 0.5  105 105 1.5  105 2  105

4. Example 2. Buckling of an inflatable cantilever beam

Consider now the same cantilever beam as in the previous section, and replace the bending force with an axial
compressive force. As in the case of a classical beam, experiments show that, given an internal pressure, for low values of
the compressive force F there exists a unique solution corresponding to a uniaxial stress state where the beam remains
straight. On the other hand, when force F reaches some critical value, non-trivial solutions are possible, which correspond
to a bent position of the beam. When computing the critical values by means of the finite element (25), one has to carry out
numerical computations since it is impracticable to invert the stiffness matrix and obtain the closed-form expression for the
critical force, even with one single element. The numerical results have been compared with the membrane finite element
ones. Here, the initial axial force in relation (25) is No ¼ PF.

4.1. Data for numerical computations

The numerical computations are conducted with four values of the pressure p, three values of natural radius R+ and
three values of natural length ‘+, while other quantities remain fixed, as shown in Table 2.
As mentioned after relation (6), the correction shear coefficient k is taken equal to the usual value for circular thin tubes,
i.e. k ¼ 0.5. The reference dimensions (‘o, Ro, ho), defined as those when the beam is pressurized, are computed as functions
of the internal pressure by means of the well-known elastic small strain solution for thin tubes:
 
1  2n pR+
‘0 ¼ ‘+ 1 þ ,
2 Eh+
 
2  n pR+
R0 ¼ R+ 1 þ ,
2 Eh+
pR+
ho ¼ h+  3n . ð29Þ
2E
Recall that ‘+, R+ and h+ are the natural length, radius and thickness, respectively. With the data in Table 2, it is
found that difference between the reference thickness and the natural thickness is 2.3% at most.

4.2. Membrane finite element computations

The numerical results obtained with the beam element will be compared with 3D membrane finite element results. The
membrane computations are carried out using a general purpose nonlinear finite element program, based on the total
Lagrangian formulation. The beam is modelled as a 3D membrane structure; the membrane elements have zero bending
stiffness and satisfy the usual plane stress condition. The 3D constitutive law is the Saint-Venant Kirchhoff one,
characterized by the Young’s modulus E and Poisson ratio n. The path-following and branch switching techniques are
 9

Z X

Fig. 2. A typical mesh used in 3D membrane finite element computations.

R∅ = 0.04 m, ∅ = 0.9 m
203
202
201
200
← p = 2 105 N/m2
199
198
Fc (N)

197
← p = 1.5 105 N/m2
196
195
194 ← p = 105 N/m2
193
192
191 ← p = 0.5 105 N/m2

190
0 1 2 3 4 5 6 7 8 9 10 11
Element number

Fig. 3. Buckling of an inflatable cantilever beam. Mesh convergence test.

included in the numerical scheme, in order to deal with possible limit and bifurcation points. Fig. 2 shows a typical mesh
used in 3D membrane finite element computations, containing 2401 nodes and 768 eight-node quadrilaterals and six-node
triangles.
There is one significant difference between the inflatable beam model and the 3D membrane finite element solution.
When solving by membrane finite element, the reference configuration is the natural one where the internal pressure is zero.
On the other hand, when solving with the inflatable beam model the reference configuration is a pre-stressed state (after the
pressure and the force F are applied).

4.3. Numerical results

As shown in Fig. 3, the numerical results converge rapidly as the element number increases. Considering the geometry
R+ ¼ 0.04 m, ‘+ ¼ 0.9 m for instance, the critical force value Fc obtained with more than 5 elements remains unchanged
up to the fourth or even the fifth digit. Thus, the subsequent results are presented using 5 elements.
Fig. 4 gives the buckling load Fc as a function of the length of the beam and the internal pressure. The values for Fc
obtained by the beam element are in good agreement with that obtained with the membrane finite element. Over the whole
range of the computation, the differences are less than 0.9%. The buckling force Fc decreases as the tube length increases,
and it increases along with the internal pressure. With R+ ¼ 0.08 m, for instance, Fc is divided by about three from
‘+ ¼ 0.65 m to ‘+ ¼ 1.15 m. With R+ ¼ 0.08 m and ‘+ ¼ 1.15 m, Fc rises by about 10% when the pressure varies from
0.5  105 to 2  105. Other computations, which are not presented here, show that the influence of the pressure on the
buckling force is stronger if a material with lower Young’s modulus is chosen. Nevertheless, the strains can then be so large
that a fully nonlinear computation is required.
Note that the pressure plays a crucial role in the inflatable beam formulation: if the pressure were discarded in the
stiffness matrix expression (25) (so that No ¼ F and P ¼ 0), the discrepancy between the beam and the membrane finite
element computations should reach 3.9%. Concerning the role of the correction coefficient k, other numerical
computations show that if k is given very large values—so as to cancel the shear effect and switch from the Timoshenko to
 10

R∅ = 0.04 m
400

∅=0.65 m

300
Fc (N)

∅=0.90 m y
200

F
∅=1.15 m
o
100

0
0 50000 100000 150000 200000
p (N/m2)

R∅ = 0.06 m R∅ = 0.08 m ∅=0.65 m

1400 3000
∅=0.65 m
1200
2500

1000
2000

800 ∅=0.90 m
Fc (N)

Fc (N)

∅=0.90 m 1500
600
∅=1.15 m
∅=1.15 m 1000
400

500
200

0 0
0 50000 100000 150000 200000 0 50000 100000 150000 200000
p (N/m2) p (N/m2)

Fig. 4. Critical load of a cantilever.

, inflatable beam finite element; ————, membrane finite element; —  —  —, crushing force Fcrushing given by
relation (30).

the Euler–Bernoulli model—then the discrepancy reaches 8.7%. In Ref. [7], the comparison was made between theoretical
and membrane finite element results. The difference found therein are not the same, since the analytical and numerical
results for the inflatable beam do not behave alike.

4.4. Limit of validity of the numerical results

Like the bending case, the computed buckling force is meaningful only if the internal pressure is high enough. Before
buckling takes place, the principal stress directions at every point are directed along the cylindrical base vectors, so that the
validity condition of the solution writes

SoXX 402N o ¼ P  F 402F oF crushing  P ¼ ppR2o , (30)

 11

where the reference radius Ro itself is a function increasing with the pressure p. Inequality (30) shows that, given an internal
pressure, the Fc value obtained numerically must not be too high for the buckling solution to be meaningful. If the
compressive force is greater than the upper bound specified by (30), the inflated beam collapses by crushing rather than by
bending buckling. Thus, the bound given by (30) may be referred to as the crushing force of the inflated beam.
The crushing force is plotted versus the pressure in dashed lines in Fig. 4. Since reference radius Ro is found to vary little
with the pressure, these curves are almost straight lines. If the critical force Fc is found lower than these curves, then the
beam buckles at this value indeed. If not, the beam is crushed down before bending buckling occurs and the corresponding
point (p, Fc) is not represented in Fig. 4.
It is worth noting that the 3D membrane finite element computation does correctly detect the crushing forces as
bifurcation points. However, it fails to determine the bifurcation eigenvector, which would mean that the beam is indeed in
ultimate collapse. As expected, the plots of the critical force obtained by the 3D finite element computation versus the
internal pressure stem from the crushing curves (dashed lines in Fig. 4). However, for the sake of clarity, these portions of
curves are not represented.

5. Example 3. Pinched toroidal beam

Now consider a toroidal beam pressurized and compressed by two forces F acting along a diameter (Fig. 5). The
geometry is defined by the small radius r+, the large radius R+ and the membrane thickness h+, which are natural
quantities defined in the stress-free state before the internal pressure is applied.
The numerical computations are conducted on two geometries and four values of internal pressure p, using the data
shown in Table 3.
The analytical solution for a torus subjected to an internal pressure was investigated by several authors. One of the most
recent works is due to Redekop [15] who solved the problem in linear elasticity by means of successive approximations and
obtained the analytical displacement field in the axisymmetrical case. The analytical solution for a non-axisymmetric
pinched torus is not available to the authors’ knowledge. Here, the toroidal beam will be computed by means of beam finite
elements. The quarter of the beam (loaded by half of force F) is modeled with 20 three-node elements, which is more than

F
y
r∅
F/2
A

R∅
x
z

x
F B

Fig. 5. Pinched toroidal beam and the modeled quarter.

Table 3
Data for the toroidal beam

Natural thickness, h+ 50  106 m

Young’s modulus, E 2.5  109 N/m2
The Poisson ratio, n 0.3
Correction shear coefficient, k 0.5

Geometry 1 Geometry 2

Natural small radius, r+ (m) 0.01 0.02

Natural large radius, R+ (m) 1.00 1.00

Internal pressure, p (N/m2) 0.5  105 105 1.5  105 2  105

 12

enough (Fig. 6a). The numerical results will be compared with the 3D membrane finite element results. The membrane
meshes contains 7817 nodes and 8  300 eight-node elements for geometry 1, and 3917 nodes and 8  150 eight-node
elements for geometry 2, Fig. 6(b).
When the beam is pressurized, it is assumed to remain toroidal, with small radius ro and large radius Ro. For lack of
analytical expressions, these reference dimensions are taken equal to the values obtained by the 3D membrane finite
element computations. Concerning the reference thickness ho, it is approximated by relation (29c) for a straight beam. This
is not accurate since, contrary to cylindrical beams studied above, here the pre-stresses in the reference state are not
uniform over the beam section and so is the membrane thickness. Eventually, the equilibrium of half of the torus (without
force F) shows that the initial axial force is No ¼ P.
Fig. 7 shows a typical deformed shape of the torus. Table 4 and Fig. 8 give the displacements at points A and B of the
torus (see Fig. 5) pinched by F ¼ 102 N. The displacements from the membrane computations can be obtained by two
different manners, which result in identical values to within 6  105 or less: for instance, the deflection vB is computed by

1 1

0.75

0.5
y

0.5

0.25

0 0
0 0.25 0.5 0.75 1 0 0.5 1
x x

Fig. 6. Meshes of the pinched torus: (a) mesh with beam elements, (b) mesh for geometry 2, with membrane elements.

F/2

1
y

0.5

0
0 0.5 1
x

Fig. 7. Deformed shape of the pinched torus, r+ ¼ 0.02 m, R+ ¼ 1 m, p ¼ 105 Pa, F ¼ 102 N. Magnifying factor for the displacement ¼ 500.
 13

Table 4
Displacements at points A and B of the torus pinched by F ¼ 102 N

p (N/m2) Using beam element Using 3D membrane element Difference

vA (m)
0.5  105 0.1880E–02 0.1891E–02 0.6%
1.0  105 0.1865E–02 0.1868E–02 0.1%
1.5  105 0.1851E–02 0.1848E–02 0.2%
2.0  105 0.1837E–02 0.1828E–02 0.5%
uB (m)
0.5  105 0.1725E–02 0.1732E–02 0.4%
1.0  105 0.1711E–02 0.1711E–02 0.0%
1.5  105 0.1698E–02 0.1692E–02 0.4%
2.0 105 0.1686E–02 0.1674E–02 0.7%
(a) Geometry 1: R+ ¼ 1 m, r+ ¼ 0.01 m
vA (m)
0.5  105 0.2343E–03 0.2355E–03 0.5%
1.0  105 0.2308E–03 0.2297E–03 0.5%
1.5  105 0.2274E–03 0.2248E–03 1.1%
2.0  105 0.2241E–03 0.2202E–03 1.8%
(b) Geometry 2: R+ ¼ 1 m, r+ ¼ 0.02 m
uB (m)
0.5  105 0.2145E–03 0.2155E–03 0.5%
1.0  105 0.2112E–03 0.2102E–03 0.5%
1.5  105 0.2081E–03 0.2057E–03 1.1%
2.0  105 0.2051E–03 0.2015E–03 1.8%

uB for geometry 1
0.0015

0.001

0.0005
uB for geometry 2
0
vA for geometry 2
-0.0005

-0.001

-0.0015
vA for geometry 1
-0.002
0 50000 100000 150000 200000
p (N/m2)

Fig. 8. Displacements (unit: m) at points A and B (Fig. 5) of the torus pinched by F ¼ 102 N.
, inflatable beam finite element; ————, membrane
finite element.

taking either the displacement at the point (X ¼ 0, Y ¼ Ro, Z ¼ ro), or the average of the displacements at the points
(X ¼ 0, Y ¼ Ro+ro, Z ¼ 0) and (X ¼ 0, Y ¼ Roro, Z ¼ 0). The results obtained by the inflatable beam element are in
quite good agreement with those by the 3D membrane element, the maximum difference on vA or uB is 1.8%.
Unlike other examples, a formula for the limit of validity for the pinched beam seems unavailable, because the loading is
not axisymmetric and the stress state in the membrane due to the internal pressure and the pinching forces is rather
intricate. The 3D membrane computations are the only direct way to know when a wrinkle appears and when the beam
solution is no longer valid. It has been checked that the value F ¼ 102 N chosen for the pinching force does not yield
wrinkles.
 14

6. Example 4. Buckling of an inflatable toroidal beam

Consider now the same toroidal beam as in the previous section, and replace the pinching force F with a radial
compressive force q per unit length uniformly distributed on the external circumference, Fig. 9. During the deformation
process, the load q is assumed to remain radial, i.e. directed toward the initial center of the ring. Let us compute the critical
value of q, which produces some slight deflection from the circular form of equilibrium. We will deliberately preclude out-
of-plane buckling modes and confine ourselves to in-plane modes, although the former may occur before the latter.

6.1. Numerical results

As in the case of the pinched torus, only a quarter of the beam is modeled using 20 three-node elements, see Fig. 6(a).
Table 5 and Fig. 10 show the buckling force as a function of the internal pressure, for the two considered geometries
(r+ ¼ 0.01 m, R+ ¼ 1 m and r+ ¼ 0.02 m, R+ ¼ 1 m). The critical force qc increases as a quasi-linear function of the
pressure. In addition, when the small radius r+ is doubled, the critical force qc is multiplied by about 8. The maximal
difference with the 3D membrane finite element is 2.4%.
The obtained results can also be compared with Weeks’ analytical solution [16]. Weeks determined the buckling load qc
of a pressurized toroidal ring subjected to a uniformly distributed radial load q, assuming linear elasticity and accounting
for the prebuckling membrane stresses. The critical load is the root of the determinant of the global stiffness:
detðK mat þ qK geom Þ ¼ 0, (31)
mat geom
where K and K denote the material and the geometrical stiffness matrices, respectively. Eq. (31) leads to solving a
cubic equation with unknown q. However, since the matrices in (31) are rather intricate, Weeks gave a first-order
approximation to the critical load by expanding the determinant and keeping only the terms which are linear in q and ro/Ro.
The result of this procedure is found to give answers within 5% of the exact solution for ro/Rop0.1. Using our notations, it

r∅

R∅
θ
x
z
q

Fig. 9. Inflatable toroidal beam under radial compressive force.

Table 5
The critical force of a radially loaded inflatable toroidal beam

Pressure, p (N/m2) Reference small radius, ro (m) Reference large radius, Ro (m) Critical force qc (N/m)

Using beam element Using 3D membrane element Difference

(a) Geometry 1: r+ ¼ 0.01 m, R+ ¼ 1 m

0.5  105 1.003E–02 1.0008E+00 1.566E+00 1.556E+00 0.7%
1.0  105 1.007E–02 1.0016E+00 1.579E+00 1.576E+00 0.2%
1.5  105 1.010E–02 1.0025E+00 1.591E+00 1.605E+00 0.9%
2.0  105 1.013E–02 1.0034E+00 1.603E+00 1.615E+00 0.7%
(b) Geometry 2: r+ ¼ 0.02 m, R+ ¼ 1 m
0.5  105 2.013E–02 1.0016E+00 1.246E+01 1.216E+01 2.4%
1.0  105 2.027E–02 1.0034E+00 1.265E+01 1.253E+01 1.0%
1.5  105 2.040E–02 1.0052E+00 1.284E+01 1.280E+01 0.3%
2.0  105 2.054E–02 1.0071E+00 1.302E+01 1.309E+01 0.5%

Comparison with 3D membrane element.

 15

15
14
13
12
11
10
9

qc (N/m)
8
7
6
5
4
3
2
1
0
0 50000 100000 150000 200000
p (N/m2)

Fig. 10. The critical force of a radially loaded inflatable toroidal beam.
, inflatable beam finite element; ————, membrane finite element; - - - - - -,
weeks’ solution, relation (32); —  —  — : crushing force qcrushing given by relation (33).

reads
EI 0
4:5 3
Ro E
qcr ¼   with G ¼ ; S o ¼ 2pro ho ; S̄o ¼ pr2o ; I 0 ¼ pr3o ho . (32)
4 þ 4:5ro =Ro EI 0 =R2o 2ð1 þ nÞ
1 þ 3ro =2Ro þ
GS 0 =2 þ pS̄0
As Weeks did not tell how to compute the reference dimensions ro, Ro and ho, these are obtained here from the 3D
membrane finite element computations or approximated by relation (29c), as said above. The numerical values for qc from
(32) are reported in Fig. 10 in dashed lines. For the ratios ro/Ro retained here, the differences between Weeks’ results and
ours are about 10%, which could be considered as reasonable in view of the various approximations made.

6.2. Limit of validity of the numerical results

By denoting y the circumferential direction of the large circle (see Fig. 9), the membrane solution is meaningful as long as
the hoop stress Soyy —or the initial axial force No—is positive. The expression for No is obtained from the equilibrium of half
the torus under the combined action of the internal pressure p and the external load q:
P ppr2o
Soyy 4022N o ¼ 2P  2qðRo þ ro Þ402qoqcrushing  ¼ , (33)
Ro þ ro Ro þ ro
where the reference radii Ro and ro vary as functions of the pressure p. If the compressive force q is greater than the upper
bound specified by (33), the inflated torus collapses so that the bound may be referred to as the crushing force of the inflated
torus. Using results in Table 5, the crushing force is plotted versus the pressure in dash–dot lines in Fig. 10. As in the case of
the bent cantilever beam, these curves are almost straight lines. For the two considered geometries, the critical forces qc
obtained in Table 5 are acceptable since they are lower than the crushing forces.

7. Conclusions

Inflatable structures can be computed with 3D thin shell or membrane finite elements. However, a lot of structures can
be analyzed with beam finite elements so as to significantly reduce the computing time. The main features of the paper are
summarized below:
(i) A new beam finite element has been obtained, involving the Timoshenko shear coefficient and the inflation pressure.
The explicit expression for the stiffness matrix has been derived in the particular case 3-node beam element.
 16

(ii) It is crucial to distinguish between the natural configuration where the internal pressure is zero and the pre-stressed
reference configuration around which the linearization is performed. The dimensions defining the reference configuration
of the beam depend on the prescribed internal pressure.
(iii) For the bending problem of a cantilever beam, one single element is able to provide the analytical solution. For the
buckling of a cantilever beam, the bending and the buckling of a toroidal beam, the numerical results have compared well
with the membrane finite element results. Over the whole range of the computations, the differences are lower than 2.4%
compared with those given by the membrane analysis, showing the good performance of the proposed finite element.
(iv) The numerical results for the bent cantilever beam only holds if the pressure is greater than the so-called wrinkling
pressure. Similarly, the solutions for the buckled cantilever beam and the toroidal beam are valid only if the compressive
force is less than the so-called crushing force.
The proposed formulation applies to the static in-plane stretching and bending problem of inflated beams. Taking into
account inertia effects is quite straightforward unless fluid interaction is significant (here the fluid is the inflated air). On the
other hand, the equations for 3D problems should be rather lengthy. Further investigations are in progress in order to
obtain finite elements for inflated beam structures in 3D space, in statics and dynamics.

References

[1] Comer RL, Levy S. Deflections of an inflated circular cylindrical cantilever beam. AIAA J 1963;1(7):1652–5.
[2] Fichter WB. A theory for inflated thin-wall cylindrical beams. NASA Technical Notes D-3466, 1966.
[3] Main A, Peterson SW, Strauss AM. Load–deflection behaviour of space-based inflatable fabric beams. J Aerospace Eng 1994;2(7):225–38.
[4] Main JA, Peterson SW, Strauss AM. Beam-type bending of space-based membrane structures. J Aerospace Eng 1995;8(2):120–5.
[5] Wielgosz C, Thomas JC. Deflections of inflatable fabric panels at high pressure. Thin-Walled Struct 2002;40:523–36.
[6] Thomas JC, Wielgosz C. Deflections of highly inflated fabric tubes. Thin-Walled Struct 2004;42:1049–66.
[7] Le van A, Wielgosz C. Bending and buckling of inflatable beams: some new theoretical results. Thin-Walled Struct 2005;43:1166–87. Erratum:
Thin-Walled Struct 2006; 44: 822–23.
[8] Wielgosz C, Thomas JC. An inflatable fabric beam finite element. Commun Num Methods Eng 2003;19:307–12.
[9] Cowper GR. The shear coefficient in Timoshenko’s beam theory. J Appl Mech 1967;33:335–40.
[10] Stein M, Hedgepeth JM. Analysis of partly wrinkled membranes. NASA Technical Note, NASA TN D-813, 1961. p. 1–23.
[11] Bulson PS. Design principles of pneumatic structures. Struct Eng 1973;51(6):209–15.
[12] Webber JPH. Deflections of inflated cylindrical cantilever beams subjected to bending and torsion. Aeronaut J 1982;86(10):306–12.
[13] Haughton DM, McKay BA. Wrinkling of inflated elastic cylindrical membranes under flexure. Int J Eng Sci 1996;34:1531–50.
[14] Veldman SL, Bergsma OK, Beukers A. Bending of anisotropic inflated cylindrical beams. Thin-Walled Struct 2005;43:461–75.
[15] Redekop D. A displacement solution in toroidal elasticity. Int J Vessels Piping 1992;51:189–209.
[16] Weeks GE. Buckling of a pressurized toroidal ring under uniform external loading. NASA Technical Note, NASA TN D-4124, August 1967.