You are on page 1of 12

Proceedings of the IASS Annual Symposium 2023

Integration of Design and Fabrication


10–14 July 2023, Melbourne, Australia
Y.M. Xie, J. Burry, T.U. Lee and J. Ma (eds.)

Numerical and experimental modelling of pneumatic structures


Ruy Marcelo O. PAULETTI*, Karina B. ROCHAa
* Department of Structural & Geotechnical Engineering, Polytechnic School, University of São Paulo, Brazil
pauletti@usp.br
a
Department of Mathematics, Mauá Institute of Technology, São Paulo, Brazil

Abstract
As other types of tensile structures, pneumatics rely on geometric stiffness for stability. However,
pneumatic structures are also endowed with a 'pneumatic stiffness', which corresponds to a reluctance
of the pressure envelope to change its volume. The higher the inside pressure, the more relevant this
component can be. Most civil and architectural applications of pneumatic structures are low to moderate-
energy systems, for which simplified material models may suffice, if a sound model to cope with
membrane wrinkling is also considered. This paper presents a simple yet effective model to cope with
the nonlinear analysis of pneumatic membranes under wind loads, incorporating an effective wrinkling
criterion into a simple membrane element, and deriving new expressions for its pneumatic stiffness. The
model was tested through an inexpensive experiment, comprising the compression of an exercise ball,
which corroborated the importance of considering variable inside pressures for moderate-energy
systems. The relevance reduces, however, as the dimensions of the structures increase, especially in the
case of large, insufflate domes, which are also discussed in the paper.
Keywords: pneumatic structures, membranes, finite elements, nonlinear analysis, pneumatic stiffness, membrane
wrinkling

1. Introduction
Pneumatic structures are paradigms of tension structures, since only in them it is possible to have all
structural elements working exclusively in tension. Practical applications of pneumatic structures started
at the beginnings of the 20th century, to become rather customary nowadays, in many distinct areas, as
large stadia roofs, building façades, or everyday furniture, like sofas, boats and toys. Although the use
of large, insufflated pneumatics for permanent applications has been superseded by other light structural
systems in the last decades, the use of smaller inflated or insufflated domes, as well as inflated beams
and cushions, combined with gridshells and other rigid systems, is becoming increasingly popular
worldwide [1].
In this paper we present a simple model, of easy implementation and yet able to cope with a broad range
of civil and architectural applications, which usually employ (almost) impermeable fabrics or films,
whose material behaviors might be quite complex. Typical materials present nonlinear, anisotropic and
hysteretic behaviors, very much dependent on the fabric internal structure and load history. The
orientation of orthotropy directions in double curvatures surfaces is not a trivial matter, not least because
fabric patterning usually aims to minimizing waste, a conflicting objective with optimizing the yarns
lay-out. Consequently, industry strives to provide the market with fabrics the more isotropic and linear
elastic as possible, and so follows design and analysis software. Furthermore, wind loads might induce

Copyright © 2023 by Ruy M.O. PAULETTI and Karina B. ROCHA. Published in the Proceedings of the IASS
Annual Symposium 2023 with permission.
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

wrinkling on the pneumatic envelope, and practical strategies to cope with wrinkling supersede concerns
about other types of nonlinear material behavior [2].

2. Nonlinear equilibrium
Upon discretization, the problem of equilibrium of a membrane structure can be expressed as finding a
displacement vector 𝒖𝒖∗ such that 𝒈𝒈(𝒖𝒖∗ ) = 𝒑𝒑(𝒖𝒖∗ ) − 𝒇𝒇(𝒖𝒖∗ ) = 𝟎𝟎, where 𝒖𝒖 = 𝒙𝒙 − 𝒙𝒙𝑟𝑟 is the global
displacement vector with respect to an initial configuration 𝒙𝒙0 , 𝒈𝒈(𝒖𝒖) is a residual force vector, 𝒑𝒑(𝒖𝒖)
is the internal load vector and 𝒇𝒇(𝒖𝒖) is the external load vector, all of order 𝑛𝑛𝑑𝑑𝑑𝑑𝑑𝑑 , the number of degrees
of freedom of the system. The equilibrium problem can be solved –within a vicinity of a solution 𝒖𝒖∗ –
−1
𝜕𝜕𝒈𝒈
iterating Newton-Raphson’s recurrence formula, 𝒖𝒖𝑘𝑘+1 = 𝒖𝒖𝑘𝑘 − � � � 𝒈𝒈(𝒖𝒖𝑘𝑘 ) , where the tangent
𝜕𝜕𝒖𝒖 𝒖𝒖𝑘𝑘
𝜕𝜕𝒈𝒈
stiffness matrix is defined as 𝑲𝑲 = . It is usual to assemble the global quantities from their elements’
𝜕𝜕𝒖𝒖
𝜕𝜕𝒈𝒈𝑒𝑒
counterparts, with the element tangent stiffness given by 𝒌𝒌𝑒𝑒 = 𝑒𝑒 , where 𝒖𝒖𝑒𝑒 is the element
𝜕𝜕𝒖𝒖
displacement vector and 𝒈𝒈𝑒𝑒 is the element residual vector.
Whenever analytical expressions for the tangent stiffness matrices are not at hand, a convenient option
is given by finite-difference approximations, which achieve the same convergence rate and precision as
the consistent tangent stiffness matrices, with acceptable extra computational cost, provided calculations
are performed at the element level. Using a central-difference scheme, each column of the tangent
stiffness matrix can be evaluated according to 𝒌𝒌�𝑗𝑗𝑒𝑒 = 1 �𝒈𝒈𝑒𝑒 �𝒖𝒖𝑒𝑒 + ℎ𝛿𝛿𝑗𝑗𝑒𝑒 � − 𝒈𝒈�𝒖𝒖𝑒𝑒 − ℎ𝛿𝛿𝑗𝑗𝑒𝑒 �� , where ±ℎ𝛿𝛿𝑗𝑗𝑒𝑒
2ℎ
are backward and forward perturbations of the jth degree of freedom, which can be taken as small as
allowed by the machine precision [3].
The Newton-Raphson Method (NRM) usually yields the best algorithm for the solution of non-linear
static equilibrium of cable and membrane structures, since it then presents quadratic convergence rate,
in a sufficiently narrow vicinity of the solution. However, cables and membranes have no bending
stiffness and in the absence of a proper tension field, the structure’s tangent stiffness matrix may become
non-positive definite, leading to divergence of the method. In many cases, the Dynamic Relaxation
Method (DRM) offers a convenient alternative, replacing the equilibrium problem by a pseudo-dynamic
analysis, where fictitious mass and damping matrices are arbitrarily chosen to control the stability of the
time integration process. The definition of an arbitrary damping matrix may be circumvented by the
process of kinetic damping, whereby the undamped movement of the system, governed by
𝑴𝑴𝒖𝒖̈ + 𝐠𝐠(𝒖𝒖) = 𝟎𝟎 , is followed until a maximum of the total kinetic energy is reached, whence all the
velocity components are cancelled, and the analysis is restarted, until new kinetic energy maxima
(usually smaller than the precedent ones) are found, and all velocities are zeroed again. Although DRM
shows no advantage for small to medium sized problems, whenever NRM shows good convergence,
there might be economy for large problems. Both methods have been implemented into a coherent
computational framework named SATS – A System for the Analysis of Taut Structures, which has been
applied to several problems involving the shape finding and analysis of cable and membrane structure.
Details can be found in [4].

2.1. Argyris’ membrane element


Figure 1(a) shows the Argyris’ natural triangular membrane finite element defined in an initial
configuration 𝛺𝛺0 , in which it is already under an initial stress field. A reference configuration 𝛺𝛺𝑟𝑟 usually
considers stress-free conditions but, for small strains, we assume 𝛺𝛺𝑟𝑟 ≡ 𝛺𝛺0 . The element’s current
configuration is denoted by 𝛺𝛺. Element nodes and edges are numbered anticlockwise, with edges facing
nodes of same number.

2
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

Figure 1. (a) definitions of Argyris membrane element quantities: An element is defined at a


configuration 𝛺𝛺0 , which may also be used as reference configuration, i.e. 𝛺𝛺𝑟𝑟 ≡ 𝛺𝛺0 ; (b) initial and
current configuration of a pneumatic envelope.

Nodal coordinates are referred to a global Cartesian system and at the current configuration are given
by 𝒙𝒙𝑖𝑖 = 𝒙𝒙0𝑖𝑖 + 𝒖𝒖𝑖𝑖 where 𝒖𝒖𝑖𝑖 are the element’s nodal displacements. The lengths of element edges are
given by ℓ𝑖𝑖 = ‖𝓵𝓵𝑖𝑖 ‖ = �𝒙𝒙𝑘𝑘 − 𝒙𝒙𝑗𝑗 �, with indexes 𝑖𝑖, 𝑗𝑗, 𝑘𝑘 = 1,2,3 in cyclic permutation. Unit vectors
parallel to the element edges are denoted by 𝒗𝒗𝑖𝑖 = 𝓵𝓵𝑖𝑖 /‖𝓵𝓵𝑖𝑖 ‖. At the current configuration 𝛺𝛺 , side length
ℓ𝑖𝑖 and the unit vector 𝒗𝒗𝑖𝑖 , parallel to the nodes are shown in Figure 1, as well as the unit vector normal
to the sides 𝒏𝒏𝑖𝑖 , and the resultant of the stress acting on each element side, 𝝆𝝆𝑖𝑖 . These results are added
up to the nodal loads pi , which may be decomposed into natural load, i.e., components of loads parallel
to the element sides see [2] and [4] for further details).
With these definitions, the vector of internal nodal forces can be decomposed into forces parallel to the
element edges, according to
𝒑𝒑1 𝑁𝑁2 𝒗𝒗2 − 𝑁𝑁3 𝒗𝒗3 𝟎𝟎 𝒗𝒗2 −𝒗𝒗3 𝑁𝑁1
𝑒𝑒 𝒑𝒑
𝒑𝒑 = � 2 � = � 𝑁𝑁3 𝒗𝒗3 − 𝑁𝑁1 𝒗𝒗1 � = �−𝒗𝒗1 𝟎𝟎 𝒗𝒗3 � �𝑁𝑁2 � = 𝑪𝑪𝑪𝑪, (1)
𝒑𝒑3 𝑁𝑁1 𝒗𝒗1 − 𝑁𝑁2 𝒗𝒗2 𝒗𝒗1 −𝒗𝒗2 𝟎𝟎 𝑁𝑁3
where 𝑪𝑪 is a geometric operator, which collects the unit vectors parallel to the element edges and
𝑵𝑵 = [𝑁𝑁1 𝑁𝑁2 𝑁𝑁3 ]𝑇𝑇 is the vector of natural forces.
We assume that the behavior of the element in taut conditions is linear elastic, thus a relationship exists,
such that 𝑵𝑵 = 𝒌𝒌0𝑛𝑛 𝒂𝒂 + 𝑵𝑵0 , where 𝒂𝒂 = [𝛥𝛥ℓ1 𝛥𝛥ℓ2 𝛥𝛥ℓ3 ]𝑇𝑇 is the vector of natural displacements (with
𝛥𝛥ℓ𝑖𝑖 = ℓ𝑖𝑖 − ℓ0𝑖𝑖 and the element natural stiffness is a constant matrix given by
−𝑇𝑇 � −1 −1
𝒌𝒌0𝑛𝑛 = 𝑡𝑡0 𝐴𝐴0 𝑳𝑳−1
0 𝑻𝑻0 𝑫𝑫𝑻𝑻0 𝑳𝑳0 , (2)
where 𝑡𝑡0 𝐴𝐴0 is the element volume, 𝑳𝑳0 = diag�ℓ0𝑖𝑖 � , 𝑫𝑫 � collects the coefficients of Hooke’s law for plane
stresses, such that 𝝈𝝈 � 𝜺𝜺� , and 𝑻𝑻
� = 𝑫𝑫 �0 is a classical transformation matrix, relating the linear Green strains
�0 𝜺𝜺�, highlighting the fact that Argyris’ natural membrane element is
𝜺𝜺� to the natural strains , i.e., 𝜺𝜺𝑛𝑛 = 𝑻𝑻
akin to a strain gauge rosette.
Since 𝒌𝒌0𝑛𝑛 has only six independent components, its storage is economic, reducing the number of
operations required to calculate the internal loads and tangent stiffness, and thus the overall computing
time. The vector of internal forces at each configuration is given by

3
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

𝒑𝒑𝑒𝑒 = 𝑪𝑪(𝒌𝒌0𝑛𝑛 𝒂𝒂 + 𝑵𝑵0 ). (3)


Assembling these loads into the global residual vector is all it takes to pose the equilibrium equation of
a taut membrane structure via DRM or NRM, using finite differences. It is also a matter of simple
derivation to obtain the internal stiffness matrix of the taut membrane element, given by:
𝑁𝑁2 𝑁𝑁3 𝑁𝑁3 𝑁𝑁2
⎡� ℓ2 𝑴𝑴2 + ℓ3 𝑴𝑴3 � −
ℓ3
𝑴𝑴3 −
ℓ2
𝑴𝑴2 ⎤
⎢ 𝑁𝑁 𝑁𝑁 𝑁𝑁 𝑁𝑁1 ⎥
𝒌𝒌𝑖𝑖𝑖𝑖𝑖𝑖 = 𝑪𝑪𝒌𝒌0𝑛𝑛 𝑪𝑪𝑇𝑇 + ⎢ − 3 𝑴𝑴3 � 1 𝑴𝑴1 + 3 𝑴𝑴3 � − 𝑴𝑴1 ⎥, (4)
ℓ3 ℓ1 ℓ3 ℓ1
⎢ 𝑁𝑁2 𝑁𝑁1 𝑁𝑁 𝑁𝑁 ⎥
⎣ − 𝑴𝑴2 − 𝑴𝑴1 � 1 𝑴𝑴1 + 2 𝑴𝑴2 �⎦
ℓ2 ℓ1 ℓ1 ℓ2

where, for brevity, we have defined some projection matrices, 𝑴𝑴𝑖𝑖 = 𝑰𝑰3 − 𝒗𝒗𝑖𝑖 𝒗𝒗𝑇𝑇𝑖𝑖 . See details of the above
formulation in [2] and [4].

2.2. External nodal loads and stiffness for pneumatic envelopes


We consider that the pressure envelope displayed in Figure 1(b) is under pressure on both sides. We
denote the total outside pressure as 𝑞𝑞, and we strongly simplify the problem by assuming that its
magnitude is not dependent on displacements. For the inside pressure 𝑝𝑝, we assume Boyle-Mariotte’s
law, for an ideal gas under adiabatic conditions, such that 𝑝𝑝 = 𝑝𝑝0 𝑉𝑉0 /𝑉𝑉𝛤𝛤 , where 𝑝𝑝0 and 𝑝𝑝 are the absolute
values of the pressure acting inside the pneumatic envelope, at the initial and current configurations, of
total volumes 𝑉𝑉0 and 𝑉𝑉𝛤𝛤 , respectively.
The total volume of the pneumatic envelope at some current configuration 𝛤𝛤 can be calculated as
1 1 𝑛𝑛𝑒𝑒𝑒𝑒
𝑉𝑉𝛤𝛤 = ∫𝛤𝛤 (𝒙𝒙 ⋅ 𝒏𝒏 )𝑑𝑑𝑑𝑑 ≃ ∑𝑘𝑘=1 𝒙𝒙1𝑘𝑘 ⋅ 𝐴𝐴𝑘𝑘 𝒏𝒏𝑘𝑘 , where 𝑛𝑛𝑒𝑒𝑒𝑒 is the number of elements of the mesh, 𝐴𝐴𝑘𝑘 is the
3 3
𝑘𝑘 𝑡𝑡ℎ element area, 𝒏𝒏𝑘𝑘 its unit normal vector, and where we arbitrary picked 𝒙𝒙1𝑘𝑘 (the coordinates of node
1 of the element), but could have chosen 𝒙𝒙𝑘𝑘2 or 𝒙𝒙𝑘𝑘3 as well. If positive 𝑝𝑝 acts parallel to 𝒏𝒏, outwards the
envelope, and outside pressure 𝑞𝑞 is positive inwards, the external load vector acting on the element is
given by
𝒎𝒎 𝑇𝑇 (𝑝𝑝−𝑞𝑞)𝐴𝐴 𝑇𝑇
𝒇𝒇 = 𝒇𝒇𝑔𝑔 + 𝒇𝒇𝑝𝑝 = �𝒈𝒈𝑇𝑇 𝒈𝒈𝑇𝑇 𝒈𝒈𝑇𝑇 � + �𝒏𝒏𝑇𝑇 𝒏𝒏𝑇𝑇 𝒏𝒏𝑇𝑇 � , (5)
3 3

where 𝒇𝒇𝑔𝑔 are forces due to self-weight, 𝒇𝒇𝑝𝑝 are forces due to the transversal inside and pressures, 𝒎𝒎 is
the mass of the membrane element, 𝒈𝒈 is the gravity acceleration vector. The awkward double
transpositions were used to allow a single line notation.
Eq. (5) assembled for each element is all what is needed to include the external pressure loads on the
equilibrium problem, solved through DRM or NRM, with finite-difference approximations 𝒌𝒌 �𝑒𝑒𝑒𝑒𝑒𝑒 . A
somewhat lengthy derivation is required to arrive at the analytical expression for the external tangent
stiffness under wind and pneumatic loads, given by:
𝜦𝜦1 𝜦𝜦𝟐𝟐 𝜦𝜦𝟑𝟑 𝜱𝜱1 𝜱𝜱2 𝜱𝜱3
𝑞𝑞−𝑝𝑝 𝑝𝑝𝑝𝑝
𝒌𝒌𝑒𝑒𝑒𝑒𝑒𝑒 = �𝜦𝜦𝟏𝟏 𝜦𝜦𝟐𝟐 𝜦𝜦𝟑𝟑 � + �𝜱𝜱1 𝜱𝜱2 𝜱𝜱3 � = 𝒌𝒌𝛬𝛬 + 𝒌𝒌𝛷𝛷 , (6)
6 18𝑉𝑉𝛤𝛤
𝜦𝜦1 𝜦𝜦𝟐𝟐 𝜦𝜦𝟑𝟑 𝜱𝜱1 𝜱𝜱2 𝜱𝜱3
where 𝜦𝜦𝑖𝑖 = 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠( 𝓵𝓵𝑖𝑖 ) are skew-symmetric matrices, whose axial vectors are given by 𝓵𝓵𝑖𝑖 = ℓ𝑖𝑖 𝒗𝒗𝑖𝑖 and
𝑇𝑇
where 𝜱𝜱𝑖𝑖 = 𝒏𝒏�𝒙𝒙𝑗𝑗 × 𝒙𝒙𝑘𝑘 � , with indices 𝑖𝑖, 𝑗𝑗, 𝑘𝑘 = 1,2,3 in cyclic permutation. The Appendix provided in
[2] presents the deduction of Eq. (6).
Several authors present alternative ways to handle pneumatic loads and stiffness [2], but there is limited
exploration on the relative importance of the resulting terms. Especially in the case of insufflated
structures, due to the envelope permeability and the limited power of the mechanical blowers, a constant
inside pressure might be a more realistic assumption, in which case 𝒌𝒌𝑒𝑒𝑒𝑒𝑒𝑒 = 𝒌𝒌𝛬𝛬 . Moreover, the examples
explored in Section 3 suggest that the external stiffness can be approximated by 𝒌𝒌 �𝑒𝑒𝑒𝑒𝑒𝑒 ≅ 𝒌𝒌𝛬𝛬 , alleviating
computations without degrading convergence. We also remark that although 𝒌𝒌ext is an asymmetric
4
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

matrix at the element level, if the boundary nodes are constrained, all the non-symmetric stiffness terms
cancel out on the global external stiffness 𝑲𝑲ext , therefore the system is conservative. If, however, free
borders exist, or pressure varies along the surface in a generic way, there is no guaranty on the symmetry
of 𝑲𝑲ext .

2.3. A simple wrinkling model


Matrix 𝒌𝒌0𝑛𝑛 defined in (2) provides an efficient scheme to compute the internal loads and stiffness of a
taut membrane, avoiding explicit calculation of strains or stresses, which may be evaluated at the
postprocessing stage. However, before undergoing compressive stresses, a membrane becomes wrinkled
or slack. When wrinkles ought to be accurately represented, a shell-like formulation is required,
demanding relatively heavy computation. Nonetheless, in most civil and architectural applications, there
is no exacting requirement for tautness, nor it is required the knowledge of the precise wrinkle patterns.
What does is required is the knowledge of the basic load transfer mechanisms, which may be incorrectly
assessed, if spurious compression is not averted. Moreover, if wrinkling or slackening are allowed, for
some operational conditions, the resulting anchorage loads might be reduced, because smaller initial
stresses are then required, and because larger displacements of the membrane usually contribute to a
more favorable load distribution. Criteria for identification of the status of a membrane element (taut,
wrinkled, or slack) require consideration of either stresses or strains, or both. When wrinkling or
slackening are likely, equations (3) and (4) ought to be replaced for lengthier calculations. For isotropic
materials, principal stress and strain directions are parallel, and stress, strain or mixed wrinkling criteria
are equivalent. Herein, we choose a pure stress criterion and calculate the element stresses directly from
the natural displacements.
In order to speed up calculations, we first decompose the element natural stiffness according to
𝒌𝒌0𝑛𝑛 = �𝑡𝑡0 𝐴𝐴0 𝑳𝑳−1 �−𝑇𝑇 � �−1 −1 0 0 0 0
0 𝑻𝑻0 ��𝑫𝑫𝑻𝑻0 𝑳𝑳0 � = 𝒌𝒌𝜎𝜎 𝒌𝒌𝑎𝑎 , where we observe that 𝒌𝒌𝜎𝜎 and 𝒌𝒌𝑎𝑎 are two constant matrices
that can be conveniently stored during the pre-processing phase. Then stresses at any configuration can
be evaluated according to 𝝈𝝈 � = 𝒌𝒌0𝑎𝑎 𝒂𝒂 + 𝝈𝝈0 and, after calculating the principal stresses (𝜎𝜎1 , 𝜎𝜎2 ) and the
“principal angle” 𝜃𝜃1 (between the element local axis 𝑥𝑥� and the 𝜎𝜎1 direction), we determine the element
status, and whether it is necessary to modify stresses, according to the following wrinkling criterion:
𝜎𝜎2 > 0 ⇒ TAUT ⇒ � = 𝝈𝝈
𝝈𝝈 �
𝜎𝜎
�𝜎𝜎1 > 0 ∧ 𝜎𝜎2 ≤ 0 ⇒ WRINKLED ⇒ �𝝈𝝈 = 1 [(1 + 𝑐𝑐𝑐𝑐𝑐𝑐 𝜃𝜃1 )
2
(1 − 𝑐𝑐𝑐𝑐𝑐𝑐 𝜃𝜃1 ) 𝑠𝑠𝑠𝑠𝑠𝑠(2𝜃𝜃1 )]𝑇𝑇 (7)
𝜎𝜎1 ≤ 0 ⇒ SLACK ⇒ � = 𝟎𝟎
𝝈𝝈

Thereafter, we consider a modified element internal force vector 𝒑𝒑 � 𝑒𝑒 = 𝑪𝑪𝒌𝒌0𝜎𝜎 𝝈𝝈


�. A mixed stress-strain
criterion could avoid the intermediate calculation of negative stresses, which have no physical meaning
for membranes, but in our framework, we determine stresses and internal loads without the explicit
calculation of strains. Once again, assembling the element internal load vectors into the global residual
vector is all it takes to pose the equilibrium of a wrinkled membrane structure via DRM or NRM using
finite differences. An analytical expression for the internal stiffness, according to the above criterion is
still possible, as given in [5], but it is not as immediate as for the taut element, so we avoid bringing
forth its consideration. Reference [6] details the above formulation.

3. Applications
We present in this section some selected problems to assess the performance of our simplified kinematic
assumptions and the relevance of the pneumatic stiffness for modelling low to moderate-pressure
systems. Initially experimental and numerical results of two small-scale pneumatics are compared, to
assess the performance of the proposed membrane element, and the Boyle-Mariotte’s Law. Then two
pneumatic domes under inside and outside pressures are numerically modeled, one of moderate and
another of large dimensions, under different hypothesis for the inside pressure. Results allow relevant
conclusions about the importance of modeling the variation of inside pressure and considering the
pneumatic stiffness term.
5
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

3.1. Henky’s problem


We first consider Hencky’s problem (1915), on the deformation of an initially flat, circular membrane
fixed at the border and pressurized at one side. Bouzidi et al. [7] compared analytical solutions given by
different authors with a numerical solution using an axisymmetric membrane element and considering
large deformations and hyperelastic material. Figure 2(a) depicts the profiles assumed by a flat
membrane with radius R = 0.1425 m, a product 𝐸𝐸𝐸𝐸 = 311488 𝑁𝑁𝑁𝑁, Poison ratio 𝜈𝜈 = 0, under inflation
pressures 100 𝑘𝑘𝑘𝑘𝑘𝑘 , 250 𝑘𝑘𝑘𝑘𝑘𝑘 and 400 𝑘𝑘𝑘𝑘𝑘𝑘. Up to moderate deformations, isotropic strains are
approximately 𝜀𝜀 ≅ 2(𝛿𝛿 ⁄𝑟𝑟 )2⁄3 , thus, strains are about 3.5%, 7% and 10%, for these three pressures
intensities. We modeled the same problem in SATS, considering the mesh shown in Figure 2(b). For
each pressure intensity, our model took five NRM iteration to achieve equilibrium, for
max (‖𝒈𝒈𝑖𝑖 ‖)/max (�𝒈𝒈0𝑖𝑖 �) ≤ 10−3 , starting from a flat disc configuration. Our results, superimpose well
with those provided in [7], with increasing, but small differences up to the maximum tested load,
showing that our model is applicable to problems up to moderate deformations, which comprises a broad
range of structural applications ((Figure 2(c)).

Figure 2. (a) experimental set up of an axisymmetric pressurized membrane; (b) SATS finite element
model (flat initial configuration); (c) comparison between FE and theoretical results;

3.2. A Swiss ball experiment


The problem of the equilibrium of a spherical balloon under axisymmetric loads has become a
benchmark for theoretical and numerical methods on the modelling of pneumatic membranes, and Swiss
balls provide for ready and inexpensive experimental setups. In this section we compare the numerical
modelling of a typical Swiss ball compressed between two parallel planes with experimental results.
The test was freely inspired by Liu et al. [8], where a semi-analytical formulation was solved
numerically, for different contact conditions, comparing numerical results with experimental ones.
We experimentally obtained an average elasticity modulus 𝐸𝐸 = 7.23 MPa and a density 𝜌𝜌 =
0.88 g/mm3 for the ball’s material, as well as an average membrane thickness 𝑡𝑡 = 1.3mm. The tests
were made according to ASTM D638-14. We started the experiment inflating the ball with an initial
inside pressure 𝑝𝑝0 = 2.413 kPa (0.35 psi). The precision of this measurement was limited by the low
resolution of the available pressure gauge (∆𝑝𝑝 = 0.01psi). The initial inflated geometry was then
measured with a digital ruler. Due to slight orthotropy of the material and the presence of some
reinforcement rings, the inflated shape was a slightly prolate spheroid, of equatorial radius 𝑅𝑅𝐸𝐸,0 =
0.2775 m and polar radius 𝑅𝑅𝑃𝑃,0 = 0.2800 m, with an initial volume 𝑉𝑉0 = 𝜋𝜋𝑅𝑅𝐸𝐸2 𝑅𝑅𝑃𝑃 ⁄3 = 9.0318 ×
10−2 m3 . The reinforcement rings were located at symmetrical distances, parallel to the equator and the
air inlet was located at the lower pole. To connect the pressure gauge, the original PVC inlet was
replaced by a metal valve stem.
We tested the ball for a vertical compression load, preserving the rotational symmetry of the problem.
A gap was provided below a stiff base plate, to accommodate the pressure gauge, and the ball was
pushed downward by the top plate. An ink layer was sprayed over the ball’s top hemisphere, and a sheet
of paper was glued to the lower face of the top plate, to obtain an imprint of the contact area. An initial
vertical load of 50 N was required to stabilize the set, before a final vertical load 𝐹𝐹 = 535 N could be
applied, (encompassing both the applied external load and the weight of the top plate). The final pressure

6
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

measured by the pressure gauge was 𝑝𝑝 = 4.481 kPa (0.65 psi). Thus, considering the BML, a final
volume 𝑉𝑉𝐵𝐵𝐵𝐵 = 4.8633 × 10−2 m3 was estimated.
Using a digital ruler, we measured the ball’s final equatorial radius 𝑅𝑅𝐸𝐸,𝑓𝑓 = 0.313 m, equivalent to a
radial displacement Δ𝑅𝑅𝐸𝐸 = 0.0355 m. It was then possible to estimate the meridional stress 𝜎𝜎𝜃𝜃 =
2
�𝑝𝑝π𝑅𝑅𝐸𝐸,𝑓𝑓 − 𝐹𝐹�/�2π𝑅𝑅𝐸𝐸,𝑓𝑓 𝑡𝑡� ≅ 0.32 MPa. The downward displacement of the top plate was
Δ𝑍𝑍,𝑁𝑁𝑁𝑁 = 0.189m. The vertical coordinate of the circular contact disc, with respect to ball equator, was
𝑧𝑧𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 = �𝑅𝑅𝑃𝑃,0 −Δ𝑍𝑍,𝑁𝑁𝑁𝑁 �⁄2 = 0.1855 m. The diameter of the imprint of the contact disc was also
measured with a digital rule, resulting 𝐷𝐷𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 = 0.39350 m. Disregarding any bending stiffness of the
ball, the coherence between these measurements can be verified by considering that the ratio of the
measured vertical force by the measured inside pressure should equate the area of the contact imprint.

That yields and estimated diameter 𝐷𝐷𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 = 0.38987 𝑚𝑚, with a 0.92% error compared to the measured
imprint diameter.

Figure 3. Several stages of the testing of the Swiss ball before, at the onset and at the end of the
compression test and imprint of the contact área.

Figure 4(a) displays our numerical model, which considered the symmetry with respect to the equatorial
plane. Boundary conditions avoided any rigid body movements and preserved the rotational symmetry.
The volume of faceted approximation was 𝑉𝑉�0 = 9.03690 × 10−2 m3 , with an error of 0.39%,
compared to the analytical volume of the spheroid. We assumed a membrane material with 𝐸𝐸 =
7.23 MPa and a zero Poison ration, as well as a membrane thickness 𝑡𝑡 = 1.3 mm.

Figure 4. Numerical model: (a) discretization of the ball’s top hemisphere; (b) Lateral view of
the deformed numerical, reflect with respect to the equatorial plane.

We opted for controlling the vertical displacements of the nodes belonging to the spheroidal cap, which
would become in contact with the top plate. To estimate the size of this cap, we assumed that both the
sag of the cap and the material deformation were small, thus the cap perimeter could be approximated
by the intersection of the spheroid with the plane 𝑧𝑧 = 𝑧𝑧𝑐𝑐𝑐𝑐𝑐𝑐 = 𝑅𝑅𝑃𝑃,0 × cos�𝑅𝑅𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 /𝑅𝑅𝑃𝑃,0 � = 0.2137m. We
enforced a set of nodes of our finite element mesh to conform to this perimeter, highlighted with a
thicker parallel shown in Figure 4(a) . We then deformed the top cap into a flat disco, imposing vertical
displacements such that the final vertical coordinates of the cap’s nodes were all equal to 𝑧𝑧𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 =
0.1855𝑚𝑚, as experimentally measured. Thus, the node located at the upper pole was displaced of 𝑢𝑢𝑧𝑧,𝑃𝑃 =

7
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕
𝑧𝑧𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 − 𝑅𝑅𝑃𝑃,0 = −0.0945𝑚𝑚, whilst the nodes at cap’s perimeter were displaced of 𝑢𝑢𝑧𝑧 =
𝑧𝑧𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 − 𝑧𝑧𝑐𝑐𝑐𝑐𝑐𝑐 = −0.02817𝑚𝑚, with intermediate 𝑢𝑢𝑧𝑧 values for the intermediate nodes. We remark that
this procedure implies frictionless contact between the membrane and the compression plates. We
considered that the reference experiment reported in [8] actually indicated that the effect should be
relatively small, moreover because in our case the membrane was partially lubricated by the ink layer
and because the contact occurred gradually, so that the contact disc would become stretched even for
high friction conditions. Anyway, the validity of the frictionless assumption could be verified afterward,
by inspecting the results of our model.
Figure 4(b) displays the resulting equilibrium shape. Although only the top hemisphere was modelled.
For easier interpretation, the deformed geometry mirrored with respect to the equatorial plane. Colors
indicate the magnitude of the vertical displacement. Only minor differences were observed using ether
NRM and DRM. Moreover, equal values were obtained, through NRM, either with the analytical tangent
stiffness or its finite-difference counterpart, requiring the same number of 6 iterations, with 𝜖𝜖𝒈𝒈 ≤ 10−3 .
The third column of Table 1 compares the NRM solution with respect to the experimental values. A
discrepant error on maximum 𝜎𝜎1 compared to the meridional stress estimated from the experimental
results, may be associated to the coarseness of the mesh. Nonetheless, a maximum deformation 𝜀𝜀𝑚𝑚𝑚𝑚𝑚𝑚 ≈
𝑚𝑚𝑚𝑚𝑚𝑚(𝜎𝜎1 )⁄𝐸𝐸 = 4.7% is still within the applicable range of our simplified kinematic model.

Table 1. Results for the Swiss ball, 𝑽𝑽𝟎𝟎 = 𝟗𝟗. 𝟎𝟎𝟎𝟎𝟎𝟎𝟎𝟎 × 𝟏𝟏𝟏𝟏−𝟐𝟐 𝒎𝒎𝟑𝟑 , 𝒑𝒑𝟎𝟎 = 𝟐𝟐. 𝟒𝟒𝟒𝟒𝟒𝟒𝟒𝟒 𝐤𝐤𝐤𝐤𝐤𝐤
Hypothesis on inside pressure: BML: 𝑝𝑝 = 𝑝𝑝0 𝑉𝑉0 ⁄𝑉𝑉𝛤𝛤 𝑝𝑝 = 𝑝𝑝0
Error
Solution method: NRM DRM NRM & FDM
NRM vs. expr.
𝑝𝑝 [𝑘𝑘𝑘𝑘𝑘𝑘]: 4.4870 0.1% 4.4913 2.4132
𝑉𝑉Γ [× 10−2 𝑚𝑚3 ]: 9.6980 0.3% 9.7494 13.1740
Equator displacement Δ𝑅𝑅𝐸𝐸 [cm]: 3.532 0.5% 3.553 5.5259
Total vertical reaction ∑ 𝐹𝐹𝑧𝑧 [kN]: 0.5368 0.3% 0.5359 0.5364
max (𝜎𝜎1 ) [MPa]: 0.3392 5.1% 0.3457 0.4748
min (𝜎𝜎1 ) [MPa]: 1.9393 -- 2.027 0.2644
max (𝜎𝜎2 ) [MPa]: 2.4504 -- 2.5739 3.2119
min (𝜎𝜎2 ) [MPa]: 1.5894 -- 1.8917 0.2143

To highlight the importance of considering the BML for this system, we repeated the analyses assuming
a constant inside pressure 𝑝𝑝0 = 2.413kPa. Results for this case are summarized at the fourth column of
Table 1. Both NRM and DRM provides the same numerical results, requiring 4 iterations to achieve
converge, for the same criterion as before. However, large errors are observed with respect to the
experimental values, of about 32% for the final volume, 46% for the final pressure, 56% for the equator’s
displacements and 33% for the final maximum stress. The comparison between the numerical and
experimental results indicates that the assumption of frictionless contact was acceptable, and that at the
scale of a Swiss ball, compliance to BML is required to provide satisfactory results.

3.2. A hemispherical pneumatic dome under wind loads


As a further example, Figure 5(a) shows a model of hemispherical pneumatic dome with radius 𝑅𝑅 =
10.0 m, All the nodes at the base of model are constrained, and we assumed an elastic modulus 𝐸𝐸 =
0.5 GPa, zero Poisson ratio, and membrane thickness 𝑡𝑡 = 0.5 mm. An initial isotropic stress field 𝜎𝜎0 =
0.2 MPa was imposed, coherent to an initial inside pressure 𝑝𝑝0 = 0.2 kPa. We considered two load
cases, superimposed with the inside pressure: (a) a low wind velocity 𝑣𝑣𝑎𝑎 = 47km/h , equivalent to a
low dynamic wind pressure 𝑞𝑞𝑎𝑎 = 0.1 kPa < 𝑝𝑝0 and (b) a high wind velocity 𝑣𝑣𝑏𝑏 = 147 km/h,
equivalent to a high dynamic wind pressure 𝑞𝑞𝑏𝑏 = 1.0 kPa > 𝑝𝑝0 , Figure 5(b) shows the wind pressure

8
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

coefficients, determined using Ansys/Fluent software, considering a uniform inlet velocity profile.
Positive pressures coefficients were obtained at both the weather and leeward sides, which is somewhat
against common knowledge, but coherent with recent research [9]-[11], which indicate that for
Reynold’s numbers 𝑅𝑅𝑅𝑅 > 106 a separation region appears, where the flow detaches from the surface of
the dome, leading to a recirculation area at leeward. Thus, for the current geometry and wind velocities,
with 𝑅𝑅𝑅𝑅 ≈ 107 , compression at this region can indeed appear. The precise location of flow detachment
depends not only on 𝑅𝑅𝑅𝑅, but also on the turbulence intensity at boundary layer and the surface roughness.

Figure 5. (a) FE model; (b) pressure coefficients determined using Ansys/fluent; (c/d) field of principal
stresses 𝝈𝝈𝟏𝟏 and 𝝈𝝈𝟐𝟐 , plotted over the deformed shape. Maximum values are given in Table 2.

Table 2 - Selected results for a hemispherical dome, 𝑽𝑽𝟎𝟎 = 𝟐𝟐. 𝟎𝟎𝟎𝟎𝟎𝟎𝟎𝟎 × 𝟏𝟏𝟏𝟏𝟑𝟑 𝒎𝒎𝟑𝟑 , 𝒑𝒑𝟎𝟎 = 𝟎𝟎. 𝟐𝟐 𝐤𝐤𝐤𝐤𝐤𝐤

Basic wind pressure Hypothesis on inside pressure BML: 𝑝𝑝 = 𝑝𝑝0 𝑉𝑉0 ⁄𝑉𝑉𝛤𝛤 Constant: 𝑝𝑝 = 𝑝𝑝0
𝑝𝑝 (kPa): 0.1994 0.2000
𝑉𝑉Γ (× 103 m3): 2.0898 2.0898
𝑣𝑣𝑎𝑎 = 47 𝑘𝑘m/h max (𝜎𝜎1 ) (MPa): 0.4174 0.4175
min(𝜎𝜎1 ) (MPa): 1.6841 1.6847
𝑞𝑞𝑎𝑎 = 0.1 kPa max (𝜎𝜎2 ) (MPa): 0.2886 2.8887
min(𝜎𝜎2 ) (MPa): 0.0262 0.0262
max (‖𝒖𝒖‖) (𝑚𝑚): 0.0048 0.0048
𝑝𝑝 (kPa): 0.2161 0.2000
𝑉𝑉Γ (× 103 m3): 1.7987 1.9336
𝑣𝑣𝑏𝑏 = 147 km/h max (𝜎𝜎1 ) (MPa): 2.6224 2.6846
min(𝜎𝜎1 ) (MPa): 0 0
𝑞𝑞𝑏𝑏 = 1.0 kPa max (𝜎𝜎2 ) (MPa): 1.0073 1.0115
min(𝜎𝜎2 ) (MPa): 0 0
max (‖𝒖𝒖‖) (𝑚𝑚): 2.2094 2.2660

For the low wind velocity case, the analysis converged in only 3 NRM iterations, with 𝜖𝜖𝑢𝑢 ≤ 10−3 . Very
small displacements were observed, ‖𝒖𝒖‖ < 5 mm , with all elements in taut or marginally taut
conditions. For the high wind velocity case, however, convergence by NRM becomes problematic, since
large wrinkling areas developed over the membrane, with a ring of elements effectively becoming slack,
at the transitions between the convex and concave regions of the membrane at the deformed
configuration, as can be seen on the plotting of the 𝜎𝜎1 field (Figure 5(c)). The onset of slackness renders
the system’s tangent stiffness 𝑲𝑲 non-positive definite. To circumvent this nuisance using NRM, we
have implemented a naïve, yet enough penalty method, simply adding artificial nodal stiffnesses 𝑘𝑘� for
every degree of freedom and gradually reducing it until artificial reactions becomes either null or small,
if compared with the external forces 𝒇𝒇 .Assuming as correct the results obtained using BML, shown at
the 2nd column of Table 2 , we compare them with the results obtaining with constant inside pressure,
shown at the 3rd column of Table 2. We find an error about 7.5% for the inside pressure and the final
volume. The maximum displacement becomes 6 cm smaller, with an error about 2.6%. Maximum 𝜎𝜎1
and 𝜎𝜎2 principal stresses present errors of 2.32% e 0.41%. It is seen that in comparison to the Swiss ball
experiment, the consideration of BML in this example is not as relevant, although not insignificant.
9
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

3.3. A large, insufflated dome reinforced by meridian cables


As a final example, we will consider the ‘Megadome’, a large, inflated dome built for the 2022 Christmas
and New Year's celebrations in the city of Canela, in the South of Brazil. Figure 6(a/b) depict the installed
structure. Only the main dome, with a diameter of 𝐷𝐷 = 60m and a height 𝐻𝐻 = 30m, was modeled. The
pneumatic envelope is constituted by single layer of PVC-polyester fabric, reinforced with 40 meridian
steel cables anchored to a perimeter concrete ring. The inside pressure was guaranteed by a set of fans
with enough redundancy and capacity to vary the internal pressure for different operational scenarios.

Figure 6. The Canela’s Megadome, Brazil (2022).

The meridian cables converge to a parallel upper cable ring, with a diameter of 5m. The spacing between
the cables varies from 4.7m at the base of the dome to 0.40m at the top. Although a larger number of
smaller diameter cables might be desirable to increase the rigidity of the dome and minimize the
wrinkles of the membrane, they would also significantly increase the practical difficulties of detailing
and assembling the structure. The internal surface of the Megadome was designed for cinematographic
projections, and to avoid significant creases in the pneumatic envelope under normal operational
conditions, the reinforcing cables should be installed with small initial loads, which should however
suffice to prevent cable slippage, under normal operating conditions. Morover, the inside pressure under
normal operational conditions should not be as high as to excessive, to avoid an excessiveley scaloped
surface.

Figure 7. Wind pressure coefficients for a semispherical dome; (a) NBR-6123 standard; (b) reference [12].

Reference [12] lists some of the most notable spherical domes in the world. The dimensions of the
Megadome place it among the largest pneumatic domes in the world, and while there are domes with
larger diameters, those that reach its height and aspect ratio are rare. It is common for large spherical
domes to be reinforced by cables, with the most frequent option being a meridian arrangement, although
in some cases, cross-cable arrangements are also used. The option of pneumatic envelopes without
reinforcement cables is less frequent, usually limited to domes with smaller diameters. Although the
response of pneumatic domes to wind actions is a dynamic phenomenon, the consideration of equivalent

10
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

static analyses is reasonable for typical dimensions [12]. Moreover, in this case we considered constant
inside pressure, since in [2] we concluded that this hypothesis suffices for large, insufflated domes.
Wind loads were defined considering Brazilian NBR6123 Standard. A characteristic wind speed
𝑣𝑣𝑘𝑘 = 37m/s = 133km/h was defined. The dome was classified as a temporary building, since public
access is restricted for operational wind speeds 𝑣𝑣𝑜𝑜𝑜𝑜 > 30m/s . A dynamic wind pressure 𝑞𝑞 =
0.613𝑣𝑣𝑘𝑘2 = 0.85kPa was defined. The external pressure coefficients were assumed by considering an
envelope of the coefficients provided by the NBR-6123 and [12], see Figure 7.

MX

MX

MN MN

Z Z Z
Y Y MN
MX X X Y X

Figure 8. field distributions for Concern conditions: (a) norm of the displacements; (b/c) principal 𝜎𝜎1 and 𝜎𝜎2
stress fields; (d) normal loads on cables and achorages. Maximum valures are given in Table 3 for different
operational conditions.

Table 3. Some selected maximum results for different operational conditions (without partial safety factors)
Max. Membrane Max.
𝑣𝑣𝑜𝑜𝑜𝑜 Max cable loads 𝑁𝑁𝑚𝑚𝑚𝑚𝑚𝑚
Beauford 𝑝𝑝0 Displacements Membrane
Conditions [𝑘𝑘𝑘𝑘/ [kN]
wind Scale [mmH2O] [m] stress [kN/m]
ℎ] ‖𝒖𝒖‖𝑚𝑚𝑚𝑚𝑚𝑚 𝑆𝑆1 = 𝜎𝜎1 × 𝑡𝑡 Meridian Ring
Assembling -- 0 30 0,03 4,7 0 0
Attention Fresh Gales 70 30 0,36 11,4 23,4 37,6
Concern Storms 100 30 2,02 18,0 47,4 87,9
Violent
Preservation 134 40 2,70 25,0 87,8 196,6
Storms

The operational conditions of the Magadome are associated to daily wind forecasts. Several conditions
have been considered, including Testing, Instalation, Maintenance, Normal Operation, Attention,
Concern and Preservation. The Concern condition corresponds to ‘Storms’ on the Beaufort Scale and,
in antecipation of that forescat, the structure must be closed to the public, with access allowed only to
the maintenance team. In the Preservation conditions, corresponding to ‘Violent Storms’, the inside
pressure must be increased to 𝑝𝑝0 =0.4kPa, the structure must be sealed, and access allowed only to tackle
the envelope’s jeopardy. We remark that even moderate winds can generate projectiles such as fallen
tree branches [13], which might puncture the pnaumetic envelope, but with low probability of tearing
the membrane, under normal operational conditions. At Conservation conditions, the membrane will be
under high stresses, and tear progagation cannot be precluded.
Figure 8 shows the overal spatial distribution of main quantities for Concern condition. Maximum
values, relevant to the design, are given in Table 3 for differet operational conditions. Since the direction
of the wind can be any, all the cables and anchorages must be designed to withstand maximum loads.

4. Conclusions
We have presented a simple yet effective model to cope with the nonlinear static analysis of pneumatic
membranes under wind loads, incorporating an effective wrinkling criterion into Argyris’ triangular
membrane element and deriving a new expression for its pneumatic stiffness, 𝒌𝒌𝑒𝑒𝑒𝑒𝑒𝑒 = 𝒌𝒌𝛬𝛬 + 𝒌𝒌𝛷𝛷 . The
model was tested through an inexpensive experiment, comprising the compression of a typical exercise
11
Proceedings of the IASS Annual Symposium 2023
Integration of Design and Fabrication

ball, which corroborated the relevance of the variation of the inside pressure, for moderate-energy
systems. However, further numerical explorations, on domes of realistic dimensions and pressure
intensities, indicated that this relevance decreases, as the dimensions of the pressure envelopes increase.
In the case of a large, insufflated dome, have found that although 𝒌𝒌𝛷𝛷 did complete 𝒌𝒌𝑒𝑒𝑒𝑒𝑒𝑒 , its contribution
did not improve convergence of NRM. On the other hand, when wrinkling occurs, 𝒌𝒌𝛬𝛬 was required to
guaranty converge, but it was not necessary while the pressure envelope was fully taut. The relative
influence of these stiffness terms is an open subject, but our results indicate that, at least for low to
moderate-energy pneumatic systems, the external stiffness is fairly approximated by 𝒌𝒌 �𝑒𝑒𝑒𝑒𝑒𝑒 ≅ 𝒌𝒌𝛬𝛬 ,
without degrading convergence, while alleviating computations.

References
[1] Pauletti RMO. “Some issues on the design and analysis of pneumatic structures”. Int J Struct Eng
2010; 1(3/4): 217-240.
[2] R.M.O. Pauletti, K.B. Rocha, “A simple finite element framework for modelling pneumatic
structures”. Engineering Structures vol. 235 (2021) 111812
https://doi.org/10.1016/j.engstruct.2020.111812
[3] Pauletti RMO, Almeida Neto ES. “A finite-difference approximation to Newton’s Method
Jacobian Matrices”. IABSE-IASS Symposium, London, 2011.
[4] Pauletti, RMO. “Static Analysis of Taut Structures. in Comput Meth Appl Sci 2008; 8: p.117-139.
Textile Composites and Inflatable Struct II, E. Oñate, B. Kröplin (eds). Springer, Dordrecht.
[5] Akita , Nakashino K, Natori MC, Park KC. “A simple computer implementation of membrane
wrinkle behaviour via a projection technique”. Int J Numer Meth Eng 2007; 71(10): 1231–1259.
[6] Pauletti RMO, Guirardi DM. “Implementation of a simple wrinkling model into Argyris’
membrane finite element”. VI International Conference on Textile Composites and Inflatable
Structures 2013; Munich.
[7] Bouzidi R, Ravaut Y, Wielgosz C. “Finite elements for 2D problems of pressurized membrane”.
Comput Struct 2003; 81(26–27): 2479–2490.
[8] Liu MX, Wang CG, Li XD. “Rigid-flexible contact analysis of an inflated membrane balloon with
various contact conditions”. Comput Meth Appl Mech Eng 2018; 144–145: 218-229. Elsevier BV.
[9] Sun Y, Su N, Wu Y, Jin Q. “Experimental Investigation of the Aerodynamic Forces and Pressures
on Dome Roofs: Reynolds Number Effects”. World Congress on Advances in Civil, Environmental
and Materials Research (ACEM16). Jeju Island, Korea; 2016.
[10] Wood JN, De Nayer G, Schmidt S, Breuer M. “Experimental Investigation and Large-Eddy
Simulation of the Turbulent Flow past a Smooth and Rigid Hemisphere”. Flow, Turbulence and
Combustion 2016; 97(1): 79–119.
[11] Wood JN, Breuer M, De Nayer G. “Experimental studies on the instantaneous fluid–structure
interaction of an air-inflated flexible membrane in turbulent flow”. J of Fluids Struct 2018; 80:
405–440. Elsevier BV.
[12] Chen, Z.; Wei, C.; Li, Z.; Zeng, C.; Zhao, J.; Hong, N.; Su, N., Wind‐Induced Response
Characteristics and Equivalent Static Wind‐Resistant Design Method of Spherical Inflatable
Membrane Structures. Buildings 2022, 12, 1611.
[13] J. D. Holmes, Wind Loading of Structures, Taylor & Francis, London, 2001.

12

You might also like