A hierarchy of models for the design of composite

pressure vessels
Federica Daghia, Emmanuel Baranger, Duy Tien Tran, Pierre Pichon

To cite this version:

Federica Daghia, Emmanuel Baranger, Duy Tien Tran, Pierre Pichon. A hierarchy of
models for the design of composite pressure vessels. Composite Structures, 2020, 235,
�10.1016/j.compstruct.2019.111809�. �hal-02395907�

HAL Id: hal-02395907

https://hal.science/hal-02395907
Submitted on 10 Dec 2019

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.
 A hierarchy of models for the design
of composite pressure vessels
F. Daghiaa,∗, E. Barangera , D.-T. Trana , P. Pichonb
a
LMT (ENS Paris-Saclay, CNRS, Univ. Paris-Saclay)
61 av. du Président Wilson, F-94235 Cachan CEDEX, France
b
CETIM, Technocampus Composites
Chemin du Chaffault, 44340 Bouguenais, France

Abstract
The mechanical response of pressure vessels to an applied internal pressure is essentially
controlled by a few key parameters, related both to the overall geometry of the structure and
to the orientations and thicknesses of the composite layers. The role of each parameter, and
the way they interact to determine the structural response, can be apprehended at a very
early design stage by using simple material and structural models, which enable to explore
a wide range of designs with minimal computational cost. More complex models can then
be called into play to predict the detailed structural response, including crucial information
such as the burst pressure and failure mode. This paper discusses a hierarchy of models with
increasing levels of details and complexity, which are useful to gain increasing insight on the
pressure vessel response all along the design process.
Keywords: pressure vessels, filament winding, netting analysis, shell models

1. Introduction
In recent years, the use of continuous fiber, polymer matrix composites for the devel-
opment of stiff and strong, but lightweight, structures has increased in multiple industrial
domains. Commercially available pressure vessels, in particular, can be classified according
to their composite content into four types:

• type I vessels are purely metallic;

• type II vessels have a metallic structure, reinforced with circumferential composite

layers in their central portion [1];

• type III vessels have a metallic liner, reinforced with composite layers over their entire
body [2];

∗
Corresponding author. Tel. +33 (0)1 47 40 28 31, Fax +33 (0)1 47 40 27 85.
Email address: [email protected] (F. Daghia)

Preprint submitted to Elsevier December 9, 2019

 • type IV vessels have a plastic liner, and their load bearing structure is entirely made
of composite layers [3].
Type V vessels, which do not include a liner, are currently at the research stage [4]. Type
IV pressure vessels are particularly interesting for the transportation industry, for example
for the storage of high-pressure hydrogen in fuel cell vehicles [5, 6].
The mechanical response of type IV composite pressure vessels to an applied internal
pressure is determined by two levels of geometrical features:
• the external geometry, that is the overall shape of the structure, usually composed of
a cylindrical portion and two ends, or domes;
• the internal geometry, that is the orientation, thickness and sequence of the different
composite layers (supposing that the liner bears no load).
All of these features, plus the choice of the composite material itself, constitute the design
space within which the engineers can operate to define the final pressure vessel design. Mak-
ing appropriate choices during the whole design process, and in particular in the preliminary
design phase, is essential to obtain optimal structures, as well as to anticipate the presence of
critical zones, which might require local reinforcements or other specific technical solutions.
In this work, we consider a hierarchy of material and structural models with increasing
levels of details and complexity, which can be used sequentially from the preliminary to
the final stages of structural design. While they are not necessarily very accurate, the sim-
pler models are essential to enable the designer to understand and define the key geometric
parameters, and in particular to analyze the interplay between the external and internal
geometry in determining the mechanical response. The simplicity and extremely fast reso-
lution of these models enables to rapidly explore many design possibilities at a very limited
cost. Once the preliminary design has been defined, the more refined models enable the
designer to predict in detail the mechanical response of the pressure vessel, accessing crucial
information such as the burst pressure and failure mode.
Most of the models discussed in this work already exist in the literature, but they are
revisited and/or extended here to be used within the overall design strategy of composite
pressure vessels. In particular, the two steps strategy for the solution of a shell problem, based
on the successive construction and solution of a membrane problem and a bending correction,
is extended here to stacking sequences with a generally coupled membrane/bending behavior.
The bibliographic references related to the models discussed in the paper are given all
along, where each of the models is presented. A glossary summarizing all of the used nota-
tions is reported in Appendix B.

2. Composite pressure vessel design space: external and internal geometries

2.1. External geometry
The overall shape of the pressure vessel is defined by the shape of the mandrel around
which the composite layers are wound. An internal polymer layer, called liner, manufac-
tured by rotational molding is used to ensure water tightness of the final structure, and the

2
 Figure 1: External and internal geometry of the pressure vessel and associated bases

composite layers are wound directly around the liner. The two ends of the pressure vessel
are sealed by metal bases, which also serve as connections enabling to fill and to empty the
pressure vessel. The number of deposed composite layers determine the thickness of the
pressure vessel, thus it is defined here as part of the internal geometry (see Section 2.2).
In geometric terms, the shape of the pressure vessel can be described as a surface of
revolution. Considering the cylindrical coordinates (er , eθ , ez ), the meridian curve m, whose
rotation around the ez axis generates the surface, is defined as a parametric curve in the
(er , ez ) plane:

m (ξ m ) = fr (ξ m ) er + fz (ξ m ) ez (1)

The definition the functions fr (ξ m ) and fz (ξ m ) for the chosen shape of the meridian
curve is enough to completely define the geometrical properties of the resulting surface. In
particular, at each position ξ m along the meridian, one can define a local orthonormal basis
for the structure (em , ep , en ) where em and ep are the unit tangents to the meridian and the
parallel, respectively, and en is the unit normal vector (see Figure 1). Furthermore, other
useful quantities like the local metric tensor g and the local principal radii of curvature Rm
and Rp can be computed. The specific expressions of these quantities as functions of fr and
fz are reported in Appendix A.
In practice, the pressure vessel is often made up of two portions: a central cylindrical
part of radius R and two heads, or domes, whose shape is dictated by mechanical as well

3
as practical considerations (such as the pressure vessel capacity with respect to its overall
length). The geometry of the domes plays a major role in determining the internal stresses
distribution in the domes themselves, as well as the magnitude of the local effects occurring
at the junction between the domes and the central cylindrical portion of the vessel.

2.2. Internal geometry

The internal geometry is defined by the orientations, thicknesses and sequence of the
composite layers.
In the following, the local orientation of each layer will be defined by the angle θ between
the local vector em associated to the external geometry and the vector e1 defining the fibers’
direction within the layer. The local basis for the material (e1 , e2 , e3 ) is completed by defining
the transverse in-plane direction e2 and the normal direction to the composite layer e3 ≡ en
(see again Figure 1).
Axisymmetric structures such as pressure vessels are generally manufactured by filament
winding. In this process, each composite layer is deposed by winding a fiber/matrix tow of
a given width and thickness around the rotating mandrel. As a result, each layer generally
includes tows oriented at both +θ and −θ, with crossover patterns which depend on the
winding parameters. To simplify matters, we neglect the effect of the crossover patterns in
the following and we consider each layer as made up of tows at angles ±θ in equal proportions.
Filament winding of thermoset matrix composites relies on some initial fibers’ tension and
on geodesic trajectories to keep the layers in place before curing of the polymer matrix, which
occurs after winding. Small angles θ (lower than about 20◦ ) are generally not accessible with
this manufacturing process, and the evolution of the fibers’ orientation along the surface
of revolution is imposed by the geometry of the surface itself. Laser-assisted placement of
thermoplastic matrix composite tape, on the other hand, is a more flexible process in which
complex shapes and trajectories can be obtained, since the deposed tow is immediately
welded to the previous layers by fast laser heating and subsequent cooling. Nevertheless,
geodesic trajectories are still a popular choice for the manufacturing of thermoplastic-based
pressure vessels, and they will be the only kind of trajectories considered here.
Geodesics on a surface of revolution obey Clairault’s relation
fr (ξ m ) sin (θ(ξ m )) = const (2)
which links the angle θ(ξ m ) between em and e1 to the distance fr (ξ m ) from the axis of
revolution. As a consequence, the orientation of each layer is generally set to a certain
constant value θ0 in the cylindrical portion of the pressure vessel, while it evolves in the
domes depending on their geometry. Furthermore, each layer can only be wound up to a
minimum distance from the axis of revolution fr,min , at which the fibers’ direction becomes
aligned with ep (that is, θ(ξ m ) = 90◦ ):
fr,min = R sin (θ0 ) (3)
where R is the radius of the cylindrical part of the pressure vessel. For a given thickness of
the deposed tow, the local thickness of the composite layer also depends on fr (ξ m ), but this
aspect will not be considered here.

4
 This rapid description enables us to underline the complexity of the internal geometry of
the pressure vessels, particularly in the domes. As it was discussed, the layers orientations
and sequence are generally defined for the cylindrical part of the vessel, and their evolution
as a function of the dome’s geometry is endured but not really controlled by designers.

3. Material models
The two constituents, fibers and matrix, fulfill different roles in the overall mechanical
response of the composite material. The fibers constitute the main load-bearing element
and they control the stiffness and strength of the composite in the direction e1 , while the
matrix and the fiber/matrix interface enable the load transfer between fibers and control the
transverse and interlaminar properties. Due to the symmetries associated to the fiber/matrix
microstructure, the overall constitutive behavior of a composite tow is orthotropic in the
material basis (e1 , e2 , e3 ).
The different choices of the behavior of the composite tow are discussed in this Section.
Based on the retained structural model, discussed in Section 4, the local behavior at the scale
of the tow can be integrated to yield an overall shell-type behavior for the whole stacking
sequence, or it can be used directly within a three-dimensional structural model.
Different descriptions of the constitutive behavior of the composite tow can be considered
for different phases of the design process, with increasing level of complexity:
• if the stiffness and strength contributions of the fibers alone to the constitutive behavior
is considered, the classical netting analysis is recovered [7, 8];
• if the contribution of both fibers and matrix is considered, but not their evolution
related to progressive degradations, a full linear elastic orthotropic material model is
defined: in the design phase, this elastic behavior is often completed with ply-level
failure criteria (see for example the World Wide Failure Exercises [9, 10, 11]);
• if the progressive degradation of fibers, matrix and the fiber/matrix interface is con-
sidered, more complex models, such as continuum damage models, are introduced:
different irreversible phenomena (microcracking, irreversible strain, ...) are accounted
for in these models, thus enabling load redistribution during progressive degradation
[12, 13].
The key aspects of each of the material models is discussed in the following.
3.1. Netting analysis
Although it is not always presented in classical references about the mechanics of com-
posite materials, the netting analysis is a useful tool for preliminary design of composite
pressure vessels.
The key hypothesis of this approach is that the fibers are the only load carrying element,
thus completely neglecting the matrix contribution to the stiffness and strength of the com-
posite. While this may seem an extreme simplification, it is indeed a sensible way to evaluate
the pertinence and robustness of different choices of orientations and thicknesses to respond
to specific sets of membrane loadings.

5
3.1.1. Classical presentation in terms of static quantities
The typical way to present netting analysis is in terms of static quantities only [7, 5].
As such, it may seem inappropriate to present this approach in the section about material
models, but this apparent contradiction will be resolved later.
The plane stress state associated to each fiber orientation θi , written in the material
coordinates (e1 , e2 , e3 ) (subscript m), is assumed to have the following form:
 T  T
σi = σ11 σ22 σ12 = σ11 0 0

Rotation of σi to the structural basis (em , ep , en ) and summation of the contributions from
the different orientations yields the local equilibrium equations:
X  T  T
σ11,i ti c2i s2i ci si = Nmm Npp Nmp (4)
i

where, for each orientation θi , we have defined ci = cos θi , si = sin θi and the thickness ti ,
 T
while N = Nmm Npp Nmp are the membrane stresses associated to external loads.
This set of equations contains information only about the static part of the problem (local
equilibrium), and so it may be over- or under-determined to compute the local stresses in the
fibers for a given set of external loading, orientations and thicknesses. Its most common use
is as a simple optimal design tool, to determine stacking sequences or choice of thicknesses
in which the fibers in every direction are all equally loaded.
A classical example of the use of netting analysis is to determine the “optimal” winding
angle θ = 54.74◦ for a cylinder with end closures under internal pressure, which is widely
quoted but not always properly used in the literature. This result is obtained from Eq. (4)
by assuming two winding angles θ1,2 = ±θ and a membrane biaxial stress state of the form
(to be justified in Section 5)
 T pR  T
Nmm Npp Nmp = 1 2 0 (5)
2
with p the internal pressure and R the radius of the cylinder. It should be noted that this
is not the only optimal set of orientations and thicknesses, and that, as it will be seen in
the following, it is not robust with respect to deviations from the ideal loading conditions
considered.

3.1.2. Presentation in terms of simplified constitutive behavior

An alternative and very useful way to present the netting analysis in terms of simplified
elastic behavior of the layer was introduced in [8].
There, the plane stress constitutive behavior of each fiber orientation θi in the material
(layer) coordinates is written as
    
σ11 E1 0 0 ε11
σi = Cεi ⇒  σ22  =  0 0 0   ε22 
σ12 i 0 0 0 2ε12 i

6
The layer stiffness matrix C has rank 1, thus the compliance is infinite in all directions except
e1 . In other words, a single orientation has no stiffness to support loads which are not aligned
 T
with the fibers’ direction. Supposing that the membrane strains µ = µmm µpp 2µmp
are constant in each layer, the membrane constitutive behavior for the whole stack in the
structural basis can be defined as follows

N = Aµ

where
X
A= t i TT
i CTi
i

and the rotation operator Ti is defined as

 
c2i s2i ci s i
Ti =  s2i c2i −ci si 
−2ci si 2ci si c2i − s2i

It should be noted that, from a physical point of view, it is the load transfer ensured by the
matrix which justifies the hypothesis of constant membrane strains used to determine the
behavior of the whole stack. Thus, while the matrix contribution to the stiffness of a tow is
neglected, its load transfer role still needs to be assured if the composite is to behave as a
structural assembly.
The membrane stiffness matrix A is rich of information. When only one or two fiber
orientations are included in the stack, A is not a full rank matrix. The image of A, which
defines the set of membrane stresses which can be sustained by the structure, has dimension
equal to the rank and it does not include the whole space of possible membrane stresses.
For this reason, stacks including only one or two orientations are not robust to deviations
from the ideal loading conditions the structure was designed for, as accidental loadings
might end up being supported by the matrix only. Starting from three fibers orientations,
A becomes a full rank matrix and any loading condition can be supported with at least
a partial contribution from the fibers. This is the basis for the design recommendation to
include at least four distinct fiber directions, with at least 10% fibers in each direction [7].
Using the whole stack behavior, the layer behavior and the rotation operator, the stress
σi of layer i in the material (layer) basis can be linked to the membrane stresses N as

σi = Cεi = CTi µ = CTi inv (A) N (6)

where inv (A) is the Moore-Penrose pseudoinverse (since A can be singular). Differently
from the classical formulation based solely on static quantities, this expression enables one
to quantify the fiber stresses for any number of different layer orientations.
Optimal design conditions based on the same ingredients discussed here (namely, a sim-
plified constitutive behavior and the local equilibrium) were derived in [14]. In particular,
it is shown that an infinity of optimal solutions exist, having the same total thickness t and

7
different combinations of orientations and relative thicknesses (θi , ti ). Furthermore, for cases
with three or more different orientations, the optimal solution yields average strains which
 T
are invariant with respect to rotation (that is, µ = µ 1 1 0 ).
Let us consider for example the biaxial membrane stresses discussed previously (Eq.
(5)). In addition to the famous solution θ = 54.74◦ , a family of solutions of the form
(±θ1 , ±θ2 , ..., ±θn ) with n ≥ 2 can be recovered, whose thicknesses can be determined by the
equilibrium condition Eq. (4). A simple example is the set of orientations (90◦ , ±45◦ ), where
each orientation has the same thickness. For n > 2, different combinations of thicknesses are
possible to yield different optimal solutions for the same set of orientations. As discussed
previously, the solutions having three or more fibers orientations are more robust with respect
to accidental loadings when compared to the famous optimal solution with only two winding
angles ±θ.

3.2. Full linear elastic behavior (and failure criteria)

The most classical way to describe the behavior of a composite tow is in terms of a
linear elastic orthotropic solid, whose three-dimensional compliance matrix in the material
coordinates reads
 1 ν21 ν31

E
− E2
− E3
0 0 0
 − ν112 1
− νE323 0 0 0 
 E1 E2 
 − ν13 − ν23 1
0 0 0 
inv (C) =  E1 E2 E3
(7)
 
1
 0 0 0 0 0

G 23

 0 0 0 0 G113 0 
 
0 0 0 0 0 G112

The nine independent parameters (three Young’s moduli Ei , three shear moduli Gij and three
of the six Poisson’s coefficients νij ) result from the homogenized behavior of the fiber/matrix
material. For an undamaged unidirectional tow, the behavior can be further simplified to
transversely isotropic in the (e2 , e3 ) plane, thus resulting in five independent parameters.
Considering the matrix contribution to the stiffness, the compliance and stiffness matri-
ces associated to a single orientation are full rank matrices, thus a single orientation can
withstand any type of loading. Both stiffness and strength of a single tow, however, decline
quickly when moving away from direction e1 , thus a combination of at least three orienta-
tions is still good practice in order to respond more effectively to any loading conditions. In
this sense, the extreme simplification associated to the netting analysis is useful to rapidly
detect potential issues in the choice of the fiber orientations.
The full elastic behavior of a tow can be used directly within a three-dimensional com-
putation, or it can be simplified to its plane stress version and combined with laminated
theories in order to obtain the constitutive behavior of the shell. This enables one to in-
vestigate the response of the structure to any kind of loading conditions, and not only to
membrane loadings as it was the case with the netting analysis.
The linear elastic analysis is usually complemented with one or more failure criteria
associated to each fiber orientation, which are computed during post-processing to evaluate

8
the structural conditions with respect to some admissible values. A variety of criteria exist
(see for example [9, 10, 11]), which can be based on specific failure mechanisms (fiber failure,
matrix cracking, ...) or defined as an overall failure envelope which does not enable a clear
distinction of the phenomena at play.
Composite materials are redundant by construction, and the possibility to share and,
eventually, redistribute load is a key aspect of their mechanical behavior. As such, the
fulfillment of one particular failure criteria may (or may not) be very conservative when
evaluating the limit load. For instance, matrix cracking occurring along a single orientation
hardly affects the overall behavior of the composite if the load can be redistributed to appro-
priately oriented fibers, while it may be catastrophic if (due to a design mistake) the matrix
properties govern the composite behavior in some particular directions. Furthermore, failure
criteria generally do not enable to clearly define the associated structural final failure mech-
anism, which can take a variety of forms: for pressure vessels, one may want to detect the
creation of a communicating crack system which leads to leakage (as in [4]), or the bursting
of the vessel with or without ejection of the metal base, ...
For all these reasons, elastic computations and failure criteria can be useful for prelimi-
nary design, but they are not enough to fully assess the response of the composite structure
with respect to specific criteria in terms of burst pressure or failure mode.

3.3. Modeling the progressive degradation

A variety of approaches exist to account for progressive degradation and load redistri-
bution in composite materials, from simple “last ply failure” approaches, which discard the
stiffness associated to a given orientation when some criterion is fulfilled and recompute iter-
atively the solution until failure of the whole stack, to detailed micromechanical descriptions
of matrix cracking, delamination and other degradations through fine-scale models such as
[15]. For computations of a full structure, such as a pressure vessel, ply-scale continuum
damage models appear as a reasonable compromise between accuracy of the description of
the physical mechanisms and computational cost of the simulations. A variety of contin-
uum damage models have been developed since the 1980s; in the following we discuss in
particular the damage mesomodel developed at the LMT, in which the damage variables
are constructed based on an underlying micromechanical description (see [12, 13] for a full
description of the model and its validation in a number of test cases).
Starting from the orthotropic behavior of a healthy composite in the material basis Eq.
(7), a set of internal variables and the associated evolution laws are introduced in the LMT
damage mesomodel to describe the modification of the constitutive behavior as a function of
the different physical degradation mechanisms within the ply. The degradation mechanisms
within the ply which are accounted for include:

• fiber/matrix decohesion;

• transverse matrix cracking;

• fiber failure;

9
which modify the ply’s stiffness in different directions, and eventually generate permanent
deformation at unloading. Furthermore, an interface damage model (analogous to a cohesive
zone model) is generally included in order to account for inter-ply delamination. A detailed
identification procedure for the damage mesomodel parameters starting from standard tests
on composite specimens was developed along with the model itself (see for example [16, 13]).
As it is the case for the full elastic behavior, the damage mesomodel can be used directly
within three-dimensional simulations, or it can be reduced to a plane stress description and
combined with laminate theories.
Due to the nonlinear material behavior, the overall response of the structure becomes
nonlinear, thus generating more complex simulations than its linear elastic counterpart,
including the need for multiple load steps, eventual convergence issues, localization and other
potential numerical difficulties. In the general case, therefore, continuum damage models are
best used in the last phases of the design process, when few potentially interesting designs
need to be validated against some precise specifications (such as the requirement of a “safe”
kind of burst failure mode, which does not eject the metal base). An example of damage
based simulation of a type IV composite pressure vessel can be found in [6].

4. Structural models
The structural response of a pressure vessel to an applied internal pressure, and its even-
tual failure by leakage or burst, is one of the main aspects to be investigated during the
design of these structures. As for the material description, different levels of complexity can
be introduced for the structural description, enabling to gain different insights on the influ-
ence of the different design parameters (external and internal geometries) on the structural
response.
The overall shape of a pressure vessel corresponds to a surface of revolution, with (rela-
tively) small thickness with respect to the internal radius, thus shell models appear pertinent
to describe the response of this type of structures in a (relatively) simple way.
Shell models were particularly popular in the 1960s for the design of homogeneous
isotropic thin shells (see for example [17]). The complete shell model equations being too
complex to solve analytically in the general case, a two-step solution strategy was developed,
based on the successive construction of a main membrane solution, followed by a localized
bending correction which re-establishes kinematic compatibility in the vicinity of critical
points. This approach was the basis of shells design before the democratization of numerical
simulation tools such as finite element analysis.
While membrane [18] and bending [19, 20] models were punctually applied to composite
shell-like structures, to the authors’ knowledge the two-step approach discussed in [17] for ho-
mogeneous shells has not been used for the preliminary design of composite shells. Although
the many approximations involved lead to solutions which are potentially much less accurate
than two-dimensional axisymmetric [21, 22, 23] or full three-dimensional computations, these
simplified models can help the designer to acquire an understanding of the role played by the
different design parameters on the overall structural response, and thus to make pertinent

10
design choices, ultimately decreasing the number of computationally expensive nonlinear
simulations to be carried out to validate the final design.

4.1. Membrane model

Due to their small thickness, shell structures have relatively low bending stiffness, while
their geometrical curvatures enable them to effectively respond to out-of-plane loadings with
membrane stresses. For this reason, the simplest shell model can be obtained by neglecting
the bending terms, yielding the so-called membrane model. Optimal shell design should aim
for pure membrane loading, but, as we will see, this condition is quite hard to achieve in
practice, due to punctual kinematic incompatibilities leading to localized bending effects.

4.1.1. Static analysis

In the membrane model of a general shell, the three unknown membrane stresses (two
normal and one shear component) are related to each other by exactly three equilibrium
equations (two associated to the in-plane directions and one to the out-of-plane direction).
Depending on the boundary conditions, therefore, the problem can be solved purely by
static analysis, without involving kinematic compatibility and material behavior. The in-
dependence of the internal stress state on the material behavior is particularly convenient,
as any material description (from netting analysis to continuum damage models) can in
principle be used simply as a post-processing of the membrane static analysis.
For the pressure vessel subjected to an internal pressure, the membrane stresses are
described in the local shell basis (em , ep , en ) as:
T
N(ξ m ) =

Nmm (ξ m ) Npp (ξ m ) 0

where Nmm (ξ m ) and Npp (ξ m ) are the membrane stresses in the meridian and parallel direc-
tions, respectively, and the membrane shear stress Nmp is zero due to axial symmetry of the
(internal and external) geometry and the load.
The membrane stress in the meridian direction Nmm (ξ m ) at each point of the shell can
be determined by global equilibrium of the shell portion included between 0 and ξ m . The
resultant Q of the applied pressure p = p en between 0 and ξ m is computed as
Z 2π Z ξm " Z #
fr (ξ m )
√
Q(ξ m ) = p(τ )en gdτ dθ = 2π p (fr ) fr dfr ez = Qz (ξ m )ez (8)
0 0 fr (0)

where g is the determinant of the metric tensor. Eq. (8) shows that the resultant Q is
directed along ez , and that Qz (ξ m ) does not depend on the shape of the meridian surface,
but only on the applied pressure p(fr ) and on the radial coordinate fr . In particular, for
a shell which is closed at the top (fr (0) = 0) with constant applied pressure (p(fr ) = p̄),
Qz (ξ m ) becomes

Qz (ξ m ) = p̄πfr2 (ξ m )

11
Global equilibrium of the shell portion then reads

Qz + Nmm (2πfr ) em · ez = 0 (9)

which enables one to determine Nmm .

Local equilibrium along en , on the other hand, leads to the following equation
Nmm Npp
+ +p=0 (10)
Rm Rp

which highlights the role of the radii of curvature Rm and Rp in enabling the shell membrane
response to out-of-plane loads. Since Nmm is already known from global equilibrium, this
equation enables one to compute Npp .

4.1.2. Kinematic description

As shown above, the membrane stresses within the shell can be determined simply by
static analysis if the membrane model is considered. As such, the kinematic description can
be invoked only a posteriori, to recover the strains and the displacements resulting in the
membrane for a given loading condition. We will see here that the strains resulting from
a purely membrane model cannot in the general case be integrated to yield a continuous
displacement: that is where the need for a bending correction arises.
For the membrane model, and due to the symmetries of the problem, the shell displace-
ment can be described by two scalar functions of ξ m , the meridian displacement um (ξ m ) and
the normal displacement un (ξ m ). We have:

u(ξ m ) = um (ξ m )em + un (ξ m )en (11)

It should be noticed that the rotation of the section has not been considered, as the bending
terms are supposed to be negligible.
 T
Starting from Eq. (11), the membrane strains µ(ξ m ) = µmm (ξ m ) µpp (ξ m ) 0 can
be expressed as follows:
√
gmm 0 un
µmm = (um − Γmmm um ) −
g Rm
√ mm
gmm
un
µpp = −Γmpp um − (12)
gpp Rp

where gij are the components of the metric tensor, Γijk are the Christoffel symbols, Ri are the
radii of curvature and the superscript 0 denotes the derivative with respect to the parameter
ξ m . The expressions of all of these terms as functions of the shape of the meridian curve
are given in Appendix A. As it can be noticed, the presence of curvatures leads to coupling
between the membrane strains and the out-of-plane displacement un , which is the kinematic
dual of the local equilibrium equation Eq. (10) with respect to strain energy.

12
 Since the membrane stresses have already been determined through static analysis, the
membrane strains can be computed from the constitutive behavior as

µ(ξ m ) = inv (A(ξ m )) N(ξ m ) (13)

where the membrane stiffness matrix A, obtained by integration of the composite’s stiffness
over the thickness, also depends on ξ m since the orientation and number of plies is not
generally the same at each point along the meridian curve (particularly in the dome). As it
was pointed out before, any of the constitutive models discussed in Section 3 can be used to
determine the matrix A, yielding generally different results and, in the case of the netting
analysis or of very significant damage evolution, a possibly singular A. The strains computed
from Eq. (13) can then be injected in Eq. (12).
Integration of Eq. (12) to obtain displacement functions um (ξ m ) and un (ξ m ) which are
continuous over the whole domain is not always possible. Indeed, integration of the second
line of Eq. (12) over the whole domain would require the membrane strain µpp to be contin-
uous in ξ m , but this is not guaranteed by the purely static solution. Discontinuities in the
membrane strain µpp are generated by the external and/or the internal geometries:

• for the external geometry: a discontinuity of the meridian radius of curvature Rm (for
example, at the junction between the cylinder and the dome) generates a discontinuous
membrane stress Npp via Eq. (10);

• for the internal geometry: a ply drop generates a discontinuous stiffness matrix A.

A discontinuity of Npp or A translates into a discontinuity of µpp via Eq. (13).

The lack of a continuous displacement solution for the general case highlights a limit
of the purely membrane model. A two-step solution can nevertheless be devised, where the
incompatibilities eventually identified by integrating the purely membrane model are resolved
by introducing (localized) bending corrections to restore the continuity of displacements.

4.2. Bending correction

Restoring the compatibility of the displacements resulting from the membrane model
requires the introduction of bending terms in the shell theory.
As it was stated earlier, the complete analytical solution of the whole shell model in the
general case is complex even for homogeneous isotropic shells, let alone for the composite
case. Luckily, the bending solution is localized in the vicinity of the disturbance, with a
penetration length that decreases as the ratio of the membrane to bending stiffness increases,
as it will be shown later in this Section. For this reason, a first estimation of the bending
effects for any shell geometry can be obtained by considering the much simpler solution for
a cylindrical shell, whose radius and (constant) stiffness are equal to those of the general
shell in correspondence of the disturbance. This approximation was regularly used for the
design of thin shells in the past [17] and its bounds were rigorously defined for homogeneous
isotropic shells. Here, the aim is just to obtain a first estimate of the bending effects,
and in particular to understand their sources and the key parameters at play, while the final

13
validation of the retained geometry will be performed using two-dimensional axisymmetric or
full three-dimensional finite element simulation. As such, the accuracy of this approximation
is less crucial here, and it will not be the focus of the discussion.
The equations for the cylindrical shell under bending loadings are derived in the following.
With respect to the classical derivation, we consider here the more general case of unbalanced
laminates, which introduces coupling terms between the membrane and bending constitutive
behavior.

4.2.1. Model equations for a cylinder under bending loading

In the bending model, the equilibrium equations alone do not enable one to determine
the static solution, thus the full model equations (kinematic, static and constitutive) need
to be considered together.
The geometry of the cylinder is described here by the following parametrized meridian
curve

m(ξ m ) = Rer − Rξ m ez , ξm ≥ 0

The rest of the geometrical quantities are derived according to the equations given in Ap-
pendix A.
The kinematic description needs to be enriched with respect to the membrane model, by
including the section rotation and the associated bending strains. We have:

u(ξ m , ξ n ) = [um (ξ m ) + ξ n φm (ξ m )] em + un (ξ m )en (14)

where ξ n denotes the normal coordinate with respect to the shell midplane, and the rotation
φm is related to the normal displacement according to the Kirchhoff-Love hypotheses, as
φm = −u0n /R. The strains are expressed as
0

µmm = um u00n


R
χ
mm = −
membrane u bending R2
µpp = n
χpp = 0
 
R
where the membrane strains are identical to Eq. (12) specialized for the cylindrical geometry,
while the bending strain χmm (change in curvature of the meridian) appears as an additional
term.
The static description involves the following two equilibrium equations:
0
Nmm =0
00
Mmm Npp
2
− +p=0 (15)
R R
The first of Eqs. (15) immediately yields a constant meridian membrane stress Nmm in the
cylinder, while the second one corresponds to Eq. (10), specialized for the cylinder and
including also the bending contribution. Solving it requires to express Mmm and Npp in
terms of displacements, calling upon the constitutive and compatibility equations.

14
 Considering a general stack of ±θ layers, not necessarily balanced with respect to the
shell midplane, we obtain the following constitutive behavior
    
N A B µ
= (16)
M BT D χ
where A was previously defined and we have

2
n
X ξi+1 − (ξin )2 T
B= Ti C i Ti
i
2

3
n
X ξi+1 − (ξin )3 T
D= Ti C i Ti
i
3

where ξin and ξi+1

n
are the positions of the bottom and the top of layer i. Due to the
symmetries of the problem, the three constitutive matrices A, B and D have the following
structure
 
•11 •12 0
• =  •12 •22 0 
0 0 •33
As for the matrix A already considered in the membrane model, the constitutive matrices
B and D could, in principle, be determined using any of the material behaviors described
in Section 3. The use of the simplified “netting analysis” behavior, however, is not rec-
ommended, as the contribution of each orientation also depends on its position relative to
the midplane, thus making it not obvious to presuppose that certain stiffness terms can be
neglected. Continuum damage models, on the other hand, lead to stiffness matrices which
vary in both space and time, since the degradation evolves according to the local loading
conditions: using this kind of material models significantly complexifies the problem, and
would not lead to the simple closed-form solutions which are discussed in the following, and
which enable one to understand the main parameters controlling the amplitude and pen-
etration length associated to the bending problem. For this reason, the full linear elastic
behavior for each ply appears as the wisest choice of material model in conjunction with the
shell bending problem.
Since Nmm is constant and known from equilibrium, Eq. (16) can be used to express
µmm as a function of Nmm and of the other strain components, which depend on the normal
displacement un only. Substitution of the constitutive and compatibility equations into the
section of Eq. (15) yields
D̃ 0000 2B̃ 00 Ã A12 Nmm
4
un − 3 un + 2 un = p − (17)
R R R A11 R
where the modified stiffness terms D̃, B̃ and à are defined as
2
B11 B11 A12 A212
D̃ = D11 − , B̃ = B12 − , Ã = A22 −
A11 A11 A11

15
 The solution of this equilibrium equation is composed of two portions: the particular
solution associated to the second member yields the constant un already determined in
Section 4.1 through the membrane model, while the solution to the homogeneous equation
recovers the bending terms.
The homogeneous equation can be put in the form

u0000 00 4
n − βun + 4α un = 0 (18)

where α and β are expressed as

! 41
R2 Ã 2B̃R
α= , β= (19)
4 D̃ D̃

In particular, α is related to the ratio between the membrane and the bending stiffness, while
β appears due to the coupling term B̃. The solution reads
m m
un (ξ m ) = e−α1 ξ [C1 sin (α2 ξ m ) + C2 cos (α2 ξ m )] + eα1 ξ [C3 sin (α2 ξ m ) + C4 cos (α2 ξ m )]
(20)

where
√
 
φ
α1 = α 2 cos
2
√
 
φ
α2 = α 2 sin
2
 √ 
16α 4 −β 2
 arctan if β > 0


β
φ= √ 
16α 4 −β 2
 arctan +π if β < 0


β

The four integration constants C1 − C4 can be determined for specific values of concentrated
transverse shear forces Q0m and moments Mmm 0
applied at ξ m = 0. In particular, C1 = C2 = 0
in the shell portion with decreasing ξ m , and C3 = C4 = 0 in the shell portion with increasing
ξ m , as the exponential term should decrease far from the disturbance.
The solution of the bending problem, Eq. (20), is well known for balanced laminates, and
it was derived here for the more general case. Due to the real exponential term (±α1 ξ m ), it
is localized in the vicinity of the disturbance, and its penetration length along the cylinder
in the dimensionless coordinate ξ m is proportional to 1/α1 . Setting B̃ = 0, we recover the
well-known result for the uncoupled case: φ = π/2 and α1 = α2 = α, thus the penetration
length depends exclusively on the ratio of the membrane to bending stiffness.

4.2.2. Bending correction strategy

The solution of the complete elastic bending problem can be built by a two step procedure:

16
 1. first, a membrane solution is established according to Section 4.1, for which the nor-
mal displacement un may be incompatible in some specific points of the shell due to
discontinuities associated to the external or internal geometry;
2. secondly, a local bending correction is introduced and computed at each discontinuity
in order to restore compatibility everywhere in the shell.
In the following, we illustrate the approach for a single discontinuity. For multiple disconti-
nuities, the same principle can be adopted, but the specific formulas given in the following
require that the bending corrections associated to two different discontinuities do not inter-
act with each other; in other words, the distance between the discontinuities must be greater
than the penetration length.

Figure 2: Concentrated loads applied for the bending correction

The bending correction is introduced by applying equal and opposite loads, in the form of
a transverse shear force Q̂ and a moment M̂ , on either side of the discontinuity (the positivity
convention adopted in the following calculations is given in Figure 2). Any values of Q̂ and M̂
preserve equilibrium, while an appropriate choice of their values, computed in the following,
enables to restore the compatibility of displacements associated to the membrane solution.
Let us set the origin of ξ m at the discontinuity for convenience. The compatibility con-
dition to restore displacement continuity at ξ m = 0 reads
   
membrane bending membrane bending
z }| { z }| { z }| { z }| {
 un un   un un 
   
+ = +
 0 φm   0 φm 
  

ξ m =0− ξ m =0+

The membrane displacements are already known from step 1. The bending displacements
at the discontinuity can be written as functions of the concentrated loads, as
bending
z  }|  {
un Q̂
= Si
φm ξ m =0i M̂

17
where S− and S+ represent the concentrated bending compliance of the two portions of the
shell with ξ m < 0 and ξ m > 0, respectively, as seen from the discontinuity. The values of Q̂
and M̂ can thus be determined by solving the simple system:
bending membrane
z }|  z }| !{
{    
Q̂ un un
(S− − S+ ) + − =0 (21)
M̂ 0 ξ m =0−
0 ξ m =0+

The amplitude of the bending correction terms is thus directly proportional to the mismatch
in the membrane displacements. The determination of the matrices S− and S+ is discussed
in the following.
As it was stated earlier, the analytical determination of S− and S+ is quite straight-
forward for a cylindrical shell, but difficult in the general case. Since the bending solution
is localized, only the portion of shell in the vicinity of the applied Q̂ and M̂ plays a role
in the determination of the concentrated bending compliance. For this reason, a classical
approximation is to replace the matrices S− and S+ for a general shell with the one associ-
ated to a cylinder with the same local radius and constitutive behavior [17]: the accuracy
of this approximation increases as the bending solution becomes more localized, that is for
increasing values of α1 .
The general solution on the bending problem for a cylinder, Eq. (20), can be easily used
to determine the analytical expressions of S− and S+ . The integration constants C1 − C4 on
either side of the discontinuity can be determined for an applied concentrated force Q̂ and
moment M̂ at ξ m = 0. The displacement and rotation at ξ m = 0 can then be expressed as
a function of Q̂ and M̂ , yielding S− and S+ . In particular, we have
   
1 −S11 S12 1 S11 S12
S− = , S+ =
K S21 −S22 K S21 S22

where

S11 = 2R3 D̃α1

S12 = −R3 B̃ + R2 D̃ 3α12 − α22

S21 = R3 B̃ + R2 D̃ α12 + α22

S22 = 2RD̃α1 α12 + α22

2
K = R2 B̃ 2 + 2RB̃ D̃ −α12 + α22 + D̃2 α12 + α22

Obviously, the terms of the compliance matrices depend on the radius R of the cylinder, as
well as on the stiffness terms Ã, B̃ and D̃. These terms are supposed to be constant on each
side of the discontinuity, thus this model cannot account for ply reorientation in the dome, or
for variations in stiffness which could be associated to progressive degradation. On the other
hand, a localized jump in stiffness at the discontinuity, due for example to a ply drop, can
easily be taken into account by computing S− and S+ using different stiffness coefficients.

18
If the coupling is set to zero (B̃ = 0, α1 = α2 = α), the simpler and well-known expressions
for balanced laminates and homogeneous shells are recovered.
Once the values of Q̂ and M̂ which restore the displacement compatibility have been
determined from Eq. (21), the numerical values of the coefficients C1 − C4 on each side of
the discontinuity can be determined as:

• on the left of the discontinuity:

1   1  
C1 = C2 = 0, C3 = K11 Q̂ − K12 M̂ , C4 = −K21 Q̂ + K22 M̂ ,
K K

• on the right of the discontinuity:

1   1  
C3 = C4 = 0, C1 = K11 Q̂ + K12 M̂ , C2 = K21 Q̂ + K22 M̂ ,
K K
ou
1 

K11 = −R4 B̃ + R3 D̃ α12 − α22
α2
α1  3 2 2 2


K12 = −R B̃ + R D̃ α1 − 3α2
α2
K21 = 2R3 D̃α1
= −R3 B̃ + R2 D̃ 3α12 − α22


K22

Using Eq. (20) and the definition of all of the shell quantities as functions of un , given in
Section 4.2.1, the bending correction can be calculated everywhere in the shell. In particular,
knowledge of the membrane and bending strains everywhere in the shell enables one to
compute the local stresses in each ply in the material coordinates, in order to detect the
most critical points in terms of potential failure locations.
As it was underlined multiple times, the solution resulting from this two steps procedure
is still an approximation for multiple reasons:

• the shell theory is only an approximation of the full three-dimensional description;

• in the bending correction, the bending compliances S− and S+ are determined for a
cylindrical shell, and not for a general shell;

• in the bending correction, the constitutive behavior was considered constant along the
shell, thus not accounting for the evolution of the plies orientation and their eventual
degradation.

For this reason, this approach is only suitable for preliminary design, whereas the final
design should be validated via two-dimensional axisymmetric or full three-dimensional finite
element computation, with a constitutive model which is able to account for the progressive
degradation at the scale of the tow.

19
 Table 1: Material properties of the composite tow
E1 (GPa) 150 E2 (GPa) 10
G12 (GPa) 5 ν12 (-) 0.3

5. Some illustrations
The range of material and structural models presented in Sections 3 and 4 can be used to
quickly gain insight on the mechanical response of the pressure vessel, and in particular on
the role of the different design parameters in determining this response. Some illustrations
of these preliminary design steps are given in the following.
All of the examples presented here are run using a pressure vessel constituted of a cylindri-
cal portion of radius R = 100 mm, closed by ellipsoidal domes with different ratios between
the principal axes. Different stacking sequences are considered, all of which have the total
thickness of t = 12 mm. This thickness to radius ratio is quite far from the ‘thin shells’ which
are the classical domain of validity of the simplified models discussed here, but it is repre-
sentative of real ratios for pressure vessels. The simple membrane and bending shell models
discussed here are therefore intended only as preliminary design tools, and their results are
not expected to be extremely accurate.
The internal loading pressure is taken as p = 80 MPa. The material properties of the
composite are given in Table 1.

5.1. External geometry

The meridian curve of the ellipsoidal dome is parametrized as follows
m(ξ m ) = R sin(ξ m )er + bR cos(ξ m )ez , 0 ≤ ξ m ≤ π/2
where the dimensionless constant b enables one to choose the ratio between the principal
axes. The signed radii of curvature (obtained from the expressions given in the Appendix)
are
R
3 R
1
Rm = − cos2 (ξ m ) + b2 sin2 (ξ m ) 2 , Rp = − cos2 (ξ m ) + b2 sin2 (ξ m ) 2
b b
Obviously, the sphere can be recovered as a special case by setting b = 1, which gives
Rm = Rp = −R.
The meridian curve of the cylinder is parametrized as follows
m(ξ m ) = Rer − Rξ m ez , ξm ≥ 0
The signed radii of curvature are:
Rm = −∞, Rp = −R
The two portions of the meridian curve are continuous at z = 0, but their curvature Rm
jumps from a finite value (in the ellipsoid) to infinity in the cylinder. As we will see, this is
a source of discontinuity in the membrane stresses, and thus it may engender local bending
effects.

20
 Figure 3: Membrane stresses along z for different dome geometries (b = 1, b = 3/4 and b = 4/3)

5.2. Membrane static analysis

As it was underlined in Section 4.1, the membrane model is isostatic, therefore the mem-
brane stresses N can be determined independently of the choice of the material model.
Using Eqs. (9) and (10), N is determined as

• for the ellipsoid

p̄R
1 h 1
iT
N= cos2 (ξ m ) + b2 sin2 (ξ m ) 2 1 2 − cos2 (ξ m )+b2 sin2 (ξ m )
0
2b

• for the cylinder

p̄R  T
N= 1 2 0
2

The values for the cylinder were already introduced in Section 3.1 to discuss optimal
orientations and thicknesses determined by netting analysis.
The membrane stresses in the meridian and parallel directions, Nmm and Npp , are plotted
in Figure 3 for different values of the axes ratio b in the ellipsoid. While in the cylinder the

21
membrane stresses are constant and Npp = 2Nmm , their values in the dome are highly
dependent on the chosen axes ratio b. For b < 1, the radii of curvature, and thus the
membrane stresses, are maximum at the dome top (ξ m = 0) and decrease towards the
junction with the cylinder (ξ m = π/2), while the opposite is true for b > 1. For b = 1
(spherical dome) the constant curvatures engender equal and constant membrane stress
Nmm = Npp .
While the membrane stress in the meridian direction evolves continuously towards its
value Nmm = p̄R2
in the cylindrical part, the membrane stress in the parallel direction is always
discontinuous at the junction between the dome and the cylinder, due to the discontinuity
of the meridian radius of curvature Rm . The discontinuity is greater for √ smaller values of
b, with Npp in the dome eventually becoming compressive (for b < 1/ 2). Depending on
the choice of the stacking sequence, this discontinuity could generate a discontinuity in the
membrane strains, and thus the need for a local bending correction, as we will discuss in the
next Section.

5.3. Membrane model and netting analysis: optimal orientations and critical points
Here, the netting analysis material model is used to post-treat the results of the membrane
structural model. This combination of models is used in the literature for preliminary design
of the stacking sequence in the cylinder [5] or to design the optimal dome shape considering
only a couple of orientations ±θ [18]. Here, we show here that it is particularly useful
to anticipate possible critical points in the dome design for any number of orientations ±θi .
Furthermore, the interplay between features associated to the external and internal geometry
can be used to remove some of these critical points very early in the design process.
The membrane stresses N being known from the membrane model, the membrane strains
µ can be easily computed from Eq. (13), and the presence of discontinuities in µpp , which
require the introduction of a local bending correction, can be identified. Furthermore, the
stresses σ11,i for each orientation θi can be determined from Eq. (6), thus identifying the
position and orientation of the most loaded tows, which will determine the safety factor with
respect to bursting in the preliminary design phase.
Three different internal geometries are considered in the following examples. The orien-
tations in the cylindrical portion of the vessel are given as:
1. [±54.74◦ ] in equal proportions;
2. [±20◦ , ±45◦ , ±70◦ ] in the following proportions: [14, 28, 58] · 100
t
;
3. [±20◦ , ±45◦ , ±70◦ , 90◦ ] in the following proportions: [22, 22, 22, 33] · t
99
(where the 90◦
ply is dropped at the junction).
while their evolution along the dome, computed via Eq. (2), is represented in the left column
of Figure 4. At this stage, only the orientations and thicknesses of the different layers are
specified: indeed, the stacking sequence, that is the order in which the different orientations
and laid on the mandrel, does not modify the response of the purely membrane model. Its
role will be investigated later, when the bending correction is considered.
In the cylinder, all three choices are optimal and they have identical total thickness t.
In the dome, each orientation evolves according to Clairault’s relation (Eq. (2)), dropping

22
one by one at their respective minimum distances from the axis of revolution (Eq. (3)): in
particular, the 90◦ ply is dropped right at the junction between the cylinder and the dome.
The interplay between the external geometry, which determines the membrane stresses N,
and the internal geometry, which controls the evolution of the stiffness matrix A, gives rise
to complex and sometimes surprising results.

5.3.1. Effect of the internal geometry (orientations and thicknesses)

A spherical dome geometry is considered first. The evolution of the angles, membrane
strains and fiber stresses along z for each orientation are plotted in Figure 4 for different
internal geometries.
All three of the internal geometries are optimal in the cylinder, thus the stresses in the
fibers are all identical in that portion of the pressure vessel. As for the membrane strains
µmm and µpp , they are identical for cases 2 and 3, while µpp = 2µmm for case 1.
The situation is significantly different in the dome.
The “famous” optimal solution considered in case 1 has only two orientations, therefore it
is not robust with respect to a change in the loading conditions. For this reason, no response
can be computed in the dome, in which the membrane stresses N contain a component
which is not included in the image of A. If a fully elastic, rather than a netting, constitutive
model was used, the response in the dome would be controlled by the matrix, thus making it
much less stiff and prone to failure associated to matrix damage, such as transverse matrix
cracking.
The second and third solutions, including more than one couple of orientations, have a
much more robust response, at least until the next to last orientation is dropped. After
that, they incur in the same problems as described above. The minimum radius at which
the next to last orientation is dropped defines the minimum radius of the metal base, which
should be present at this point in order to withstand the loads which cannot be carried by
the composite.
The membrane strains have complex evolutions along z, in particular in correspondence
of geometric accidents and ply drops. For case 2, discontinuities in µpp occur both at the
dome to cylinder junction (due to the discontinuity of Npp ) and at the position in which the
70◦ ply is dropped (leading to the discontinuity of A). Bending corrections should therefore
be introduced in both positions, with the ply drop appearing more critical as the jump in
strain is higher. For case 3, on the other hand, a single discontinuity at the ply drop is
observed. Indeed, at the dome to cylinder junction, two different effects compensate each
other:
1. the discontinuity of Npp , generated by the discontinuous curvature (external geometry),
which would tend to decrease the strain µpp in the dome (as it occurs for case 2);
2. the discontinuity of A, generated by dropping the 90◦ ply (internal geometry), which
compensates exactly the first effect, thus finally leading to a continuous µpp at the
junction.
This is a nice example of how the external and internal geometries can be controlled together
to remove potentially critical situations.

23
Figure 4: Angles evolution, membrane strains and fiber stresses for different internal geometries, circular
dome (membrane model and netting analysis)

24
Figure 5: Geometry, membrane strains and fiber stresses for case 3, dome shapes b = 3/4 (top figures) and
b = 4/3 (bottom figures) (membrane model and netting analysis)

The fiber stresses in the different orientations are also significantly different for cases
2 and 3. In both cases, the fibers stresses evolve differently in the different tows, whose
orientations change along the dome while the membrane stresses N stays the same (spherical
dome). As it happened for the membrane strains, discontinuities of the fiber stresses occur
at the junction and at the 70◦ ply drop for case 2, and only at the ply drop for case 3. For all
three internal geometries, furthermore, the maximum fiber stress always occurs in the dome,
and it is significantly higher than the “optimized” value in the cylinder. This confirms the
common observation that the dome is indeed the critical zone in composite pressure vessels,
and it highlights the importance of this simple tool to anticipate the position and value of
the maximum stress due to membrane loadings. A solution to reduce the fiber stresses in the
dome without significant oversizing of the cylinder is to add doily layers in specific positions
in the dome [5]. This is rather easy to manufacture by laser assisted tape placement in the
case of thermoplastic matrix composites.

5.3.2. Effect of the external geometry

The effect of different dome geometries is investigated here for case 3. The dome shape,
as well as the evolution of the membrane strains and fiber stresses, are plotted in Figure 5
for two different choices of ellipsoid (b < 1 and b > 1 respectively).

25
 As it can immediately be noticed, the ellipsoid with b < 1 yields lower values of both
membrane strains and fiber stresses than the one with b > 1. The first geometry is therefore
to be preferred. This is coherent with previous findings on the optimal dome shape based
on only two orientations [18].
For b < 1, indeed, the membrane stress in the meridian direction Nmm is higher, but the
membrane stress in the parallel direction Npp is lower (and may even become compressive)
for a significant portion of the dome height. In particular, it is rather low in correspondence
of the 70◦ ply drop, thus the discontinuity in both membrane strain and fiber stresses in this
position nearly disappears. The discontinuity at the junction, on the other hand, re-appears,
as the jump in Npp for the ellipsoidal dome is not exactly compensated by 90◦ ply drop.
This last example (ellipsoidal dome with b = 3/4, internal geometry example 3) is used
in the following to illustrate the introduction of the bending correction.

5.4. Bending correction and full elastic material model: penetration length and role of the
stacking sequence
The first step for the evaluation of the bending terms is the computation of the membrane
solution, which enables one to localize and quantify the discontinuities in the displacement,
requiring correction. The amplitude of the bending correction is directly proportional to the
displacement jump, thus the first approach for limiting the bending response is to design
the external and internal geometries in order to optimize as much as possible the membrane
response, as it was seen in Section 5.3.
Once the external geometry and the set of tows orientations and thicknesses have been
chosen, a last parameter remains in composites design: the stacking sequence. Changing
the order in which the different layers are deposited on the mandrel enables the designer a
(limited) control over the bending stiffness and the coupling term (D̃ and B̃), while leaving
the membrane stiffness (Ã) untouched. As the ratio between membrane and bending stiffness
is the main parameter which controls the penetration length, the distance at which the
bending effects will be perceived can be partially controlled by appropriate layup choices.
In the following we consider again the example configuration with b = 3/4 and the set
of orientations and thicknesses associated to case 3. Four different stacking sequences are
considered (where the angles are given in the order of deposition on the mandrel):
3.1 [±20◦ , ±45◦ , ±70◦ , 90◦ ] in the following proportions: [2, 2, 2, 3] · 9t ;
3.2 [±90◦ , ±70◦ , ±45◦ , 20◦ ] in the following proportions: [3, 2, 2, 2] · 9t ;
3.3 [±20◦ , ±45◦ , ±70◦ , 90◦ , ±70◦ , ±45◦ , ±20◦ ] in the following proportions:
[1, 1, 1, 3, 1, 1, 1] · 9t ;
3.4 [90◦ , ±70◦ , ±45◦ , ±20◦ , ±45◦ , ±70◦ , 90◦ ] in the following proportions:
[1.5, 1, 1, 2, 1, 1, 1.5] · 9t .
In all of the cases, the 90◦ plies are dropped at the junction between the cylinder and the
dome.
The results of the complete procedure, membrane solution followed by bending correction,
for the four stacking sequences are given in Figures 6 and 7. In particular, Figure 6 gives
the minimum and maximum strains in the problem coordinates, while Figure 7 gives the

26
minimum and maximum fiber stresses for each ply. Indeed, due to the bending component,
strains and stresses are no longer constant within each orientation, but they vary linearly
between the top and the bottom of each layer. The purely membrane solution, identical
for all four sequences, is also reported for completeness: it is slightly different from the one
reported in Figure 5, as here the full elastic material behavior was considered. The values
of strains and stresses above the next to last ply drop are not taken into consideration, as
they are controlled essentially by the matrix response.
Looking at the strain maps (Figure 6) we can notice that the continuity of µpp was
indeed restored at the junction. The strains in the meridian directions, on the other hand,
contain both a membrane (µmm ) and a bending (χmm ) component: here, the values at the
intrados (εmm,b ) and extrados (εmm,t ) of the cylinder are depicted. As it can be observed,
the membrane solution is recovered at some distance from the discontinuity. This distance
is not equal in all four cases: in particular, it is obviously longer for case 3, in which the 20◦
orientations, which contribute most to the bending stiffness D̃ in the meridian direction, are
positioned at the greatest distance from the composite midplane. This stacking sequence,
therefore, is the worst solution if one wants to limit the penetration length of the bending
effects. All of the other cases have similar penetration lengths, of the order of R/2. This
value is quite high: indeed, as it was pointed out earlier, the shell considered in the examples
is far from being a thin shell, with a ratio t/R = 0.12. This large penetration length still
appears quite reasonable in terms of approximating the involved portion of the dome as the
local equivalent cylinder.
The stress maps (Figure 7) are rather complex, as maximum and minimum values for
each layer are reported in the same graph. A simplified color code, associated only to the
orientation of the layer, was chosen to simplify the reading. Overall, the stresses in the 70◦
and 90◦ layers, which are mostly related to the strain µpp , are rather independent of the
stacking sequence, while the layers at 45◦ and 20◦ have rather different responses depending
on the chosen sequence. Although the most critical stress values are still those computed
in the dome for the membrane solution, in an area which is not affected by the junction,
some local increases in the fiber stresses with respect to the purely membrane solution can
be noticed, particularly in the cylindrical portion of the shell. Again, the worst case scenario
seems to be case 3, for which the maximum fiber stress in the cylinder is 20% higher than the
one computed with the membrane solution. The other three solutions seem quite similar,
with cases 1 and 4 yielding better results for the stresses in the dome. If the critical zones
in the dome are resolved with local doily layers or by the presence of the metal base, the
overstresses generated by the bending correction may become the critical values for the
design of the pressure vessel.

5.5. Comparison with a two-dimensional axisymmetric finite element model

Once a wide range of external and internal geometries have been explored with the mem-
brane model and bending correction, one or two configurations can be retained as a prelim-
inary design and further explored with three-dimensional or two-dimensional axisymmetric
finite element models, eventually including the progressive degradation of material proper-
ties. Here, the stresses in the fibers direction predicted by the full shell model (membrane

27
Figure 6: Strains the the problem coordinates for b=3/4, case 3: membrane solution, total solution (mem-
brane and bending correction) for stacking sequences 1 to 4 (full elastic material model)

28
Figure 7: Fiber stresses in the material coordinates for b=3/4, case 3: membrane solution, total solution
(membrane and bending correction) for stacking sequences 1 to 4 (full elastic material model)

29
 100

r(mm) 80

60

40

20

100 120 140 160 180 200 220

z(mm)

Figure 8: Geometry generated by OptiTank (left) and map of the fiber stresses obtained in post-processing
(right) for the example illustrated in this work

analysis and bending correction) are compared to those of a linear elastic two-dimensional
axisymmetric finite element computation of the pressure vessel provided by Cetim. This en-
ables us to illustrate the following steps of the design process, but also to assess the suitability
of the shell model as a preliminary sizing tool.
The model proposed by CETIM is built with an in-house MatLab tool named OptiTank.
The starting point is a two-dimensional axisymmetric master finite element model, which
includes the liner and metallic base geometry and materials, as well as the boundary con-
ditions and loadings. The OptiTank tool generates the geometry and material properties of
the composite layers, based on a parametric version of the Vasiliev model [24], yielding final
geometries in good agreement with the real manufactured parts (see Figure 8, left). The
computation is carried out using commercial finite element software (here, Abaqus), then the
results are imported back into OptiTank for post-processing and visualization in Hyperview
(see Figure 8, right).
OptiTank can be used in two differents modes. In the first mode, the user can impose
a specific stacking sequence in the cylinder, which is then propagated automatically to the
domes. This sequence can be taken as-is, or it can be corrected based on the real tape width
to ensure end to end contact of the deposed tapes during manufacturing. In the second
mode, the software creates a random set of sequences within a design space to be defined,
and the tested sequences are classified according to their mass and mechanical strength.
The major advantage of the OptiTank tool is its good performances resulting from au-
tomation. Due to its flexible design, additional functionalities can be integrated. One of
the outlooks is the evolution of the material laws describing composite. By integrating a
damage model as the ones discussed in this work, the OptiTank user will be able to simulate
the damage and plasticization of the composite for the last iterations of his pressure vessel
designs.
The configuration retained here is an ellipsoidal dome with b = 3/4, its stacking sequence
is number 3.1 from Section 5.4. The whole structure (composite layup, polymer liner, metal
base) is accounted for in the finite element model, whereas only the composite layup is

30
Figure 9: Fiber stresses in the material coordinates for b=3/4, case 3.1: approach proposed here (left) versus
two-dimensional axisymmetric finite element computations with constant (center) and variable (right) layer
thickness

modeled in the shell description. The external geometry (shell surface) considered for the
shell model coincides with the internal surface of the composite (and not with its middle
surface): this enables a more accurate evaluation of the overall load seen by the composite,
which is given by the pressure p applied over its internal surface.
Two different finite element models were provided by OptiTank. In the first, a constant
thickness of the composite layers everywhere in the dome is assumed (as in the shell model).
In the second (depicted in Figure 8), a more accurate description of the geometry is consid-
ered, with the thickness of each composite layer increasing as the local radius decreases.
The results obtained with the different models are displayed in Figure 9. For the shell
model, the values at the top and bottom of each ply are plotted as a function of the z
coordinate of the shell: these correspond to the maximum and minimum stress seen by each
ply. For the axisymmetric finite element models, on the other hand, the stress values are
available at the center of each element (Q4 elements with reduced integration), and all of
the stress values are plotted as a function of the z coordinate of the element center. Due to
the choice of finite element discretization, bending effects might therefore be underestimated
by the finite element model.
As it can be observed in the three plots, the values and the evolution of the stresses along z
for the different plies are quite similar for the three different models. In particular, the stress
values in the central portion of the vessel are well predicted by the simple shell model (the
three-dimensional effects are most visible in the 70◦ and 90◦ plies, whose fiber stress values
are not constant but spread over a interval of a few tens of MPa). The penetration length
of the bending effects at the junction is also well predicted by the shell model; the bending
effects in the 20◦ and in the 45◦ plies seem to be overestimated, but this may be related
to the choice of discretization of the finite element model. In the dome, the three models
predict very similar stress values until the 70◦ ply drop (end of the pink line). Above this
point, the increase in layer thickness starts to be play a significant role: this effect is visible
in the difference between the two finite element models, with the shell model comparing

31
well to the constant thickness finite element model and overestimating the maximum fiber
stresses with respect to the more accurate, variable thickness solution. Finally, the results
in the greyed area should not be considered in the comparison: indeed, in this portion of
the shell, the metallic base is present in the finite element models and it carries a significant
part of the load, which cannot be sustained by the composite layers alone after the 45◦ ply
drop.
Overall, notwithstanding its many approximations, the shell model proposed in this work
appears to give good results which can successfully be used for preliminary sizing. Its
computational cost is, of course, much lower than its finite element counterpart. The main
advantage, however, consists in the clear relationships established between the design choices
(external and internal geometry) and the structural response. Such clear understanding and
indications for the design are hard, if not impossible, to achieve through a finite element
model, even associated to a (cumbersome) parametric study.

6. Conclusions and perspectives

In this work, a hierarchy of material and structural models with increasing levels of
detail and complexity is presented and discussed for the design of type IV composite pressure
vessels. While most of the models exist in the literature, they were revisited and/or extended
here to be used within the overall design strategy. In particular, the membrane solution and
bending correction strategy, which was currently used in the past for the design of metal
shells of revolution, was extended here to composites with stacking sequences displaying
general membrane/bending coupled behavior.
The combination of a simple structural membrane model and a material netting analysis
description was extremely useful to understand the crucial role of the shell curvatures and
of the orientations and thicknesses of the composite layers in determining the mechanical
response of the pressure vessel. In particular, it enabled one to detect and quantify the dis-
continuities in membrane strains, which give rise to localized bending effects. Furthermore,
the interplay between external and internal geometry can be used to remove some of these
discontinuities at a very early design stage.
Once these key parameters are fixed, the bending correction can be computed with a
full elastic material description. This result is useful to quantify the penetration length of
bending effects, which was of the order of the cylinder radius here due to the high thickness
to radius ratio. Furthermore, the bending model gives some indications on the best choice
of stacking sequence, as the bending effects decrease with decreasing bending stiffness.
The last step of the design strategy consists in a two-dimensional axisymmetric or full
three-dimensional computation of the pressure vessel. The use of a material model describing
the progressive degradation of the composite material, such as the ones discussed in Section
3.3, constitutes a perspective of this study.

Acknowledgements
This work was funded by Cetim within the framework of the Laboratoire Commun
Comp’Innov. The PhD scholarship of Duy-Tien Tran was funded by the Doctoral School

32
SMEMaG (ED 579).

Appendix A. Geometric properties of the surface of revolution

The shape of the pressure vessel is described in this work as a surface of revolution. The
placement of the meridian curve in the Euclidean space R3 was defined in Eq. (1) as

m (ξ m ) = fr (ξ m ) er + fz (ξ m ) ez

The expressions of the geometric properties of interest of the surface of revolution as a

function of fr (ξ m ) and fz (ξ m ) are reported here.
The metric tensor g on the surface is expressed as
  02
fr + fz02 0
 
gmm 0
g= =
0 gpp 0 fr2

where the superscript 0 indicates derivation with respect to the parameter ξ m .

The local orthonormal basis for the structure, (em , ep , en ) is expressed as
1 ∂m fr0 er + fz0 ez −fz0 er + fr0 ez
em = √ = p , e p = e θ , en = e m × e p = p
gmm ∂ξ m fr02 + fz02 fr02 + fz02

The (signed) normal radii of curvature of the surface in the meridian and parallel direc-
tions are expressed as
3
(fr02 + fz02 ) 2
Rm =
−fr00 fz0 + fr0 fz00
1
fr (fr02 + fz02 ) 2
Rp =
fz0
The Christoffel symbols are expressed as
fr0 fr00 + fz0 fz00
Γm
mm =
fr02 + fz02
−fr fr0
Γm
pp = 02
fr + fz02
fr f 0
Γpmp = Γppm = 2r
fr

and they are zero for all other permutations of the indexes.

Appendix B. Notations

33
A membrane shell stiffness matrix
à modified membrane shell stiffness coefficient
B membrane/bending shell stiffness matrix
B̃ modified membrane/bending shell stiffness coefficient
ci cosine of θi
C stiffness matrix of a layer in the material (layer) basis
C1 , C 2 , C 3 , C 4 integration constants for the bending solution
D bending shell stiffness matrix
D̃ modified bending shell stiffness coefficient
(e1 , e2 , e3 ) unit vectors for the local material (layer) basis
(em , ep , en ) unit vectors for the local structural (shell) basis
(er , eθ , ez ) unit vectors for the cylindrical coordinates
E1 , E2 , E3 Young’s moduli in the material directions
fr , fz scalar functions describing the meridian curve
g local metric tensor
g determinant of the metric tensor
G12 , G13 , G23 shear moduli in the material directions
m meridian curve
M = [Mmm , Mpp , Mmp ]T bending moment per unit width in the local structural (shell) basis
M̂ moment required to restore continuity of the shell displacement
N = [Nmm , Npp , Nmp ]T membrane stresses in the local structural (shell) basis
p = p en applied internal pressure
Q resultant of the applied internal pressure on a portion of the shell
Q̂ = Qz ez transverse shear required to restore continuity of the shell displace-
ment
R radius of the cylindrical portion of the pressure vessel
Rm , Rp principal (signed) radii of curvature of the meridian and of the parallel
si sine of θi
S− , S+ concentrated shell bending compliance of the portions with ξ m < 0
and ξ m > 0, respectively, as seen from the discontinuity
t thickness of the shell
ti thickness of the layer i
Ti rotation operator from the structural (shell) to the material (layer)
basis for the layer i
u shell displacement
um , un components of the shell displacement in the meridian and normal
directions
α parameter related to the penetration length of the bending effects in
the uncoupled case
α1 parameter related to the penetration length of the bending effects in
the coupled case
α2 parameter related to the oscillation of the bending effects in the cou-
pled case
β corrective parameter related to membrane/bending coupling
Γkij Christoffel symbols

34
  T
εi = ε11 ε22 2ε12 strain in the material (layer) basis for the layer i
θ0 ply orientation in the cylindrical portion of the pressure vessel
θi local ply orientation of layer i
µ = [µmm , µpp , µmp ]T membrane shell strains in the structural (shell) basis
ν12 , ν13 , ν23 Poisson’s coefficients in the material (layer) directions
ξm parameter of the meridian curve
ξn thickness coordinate of the shell
ξin , ξi+1
n bottom and top positions in the thickness of layer i
 T
σi = σ11 σ22 σ12 stress in the material (layer) basis for the layer i
φ corrective parameter related to the membrane/bending coupling
φm rotation around the ep axis of the shell
χ = [χmm , χpp , χmp ]T bending shell strains in the local structural basis
•0 derivative of • with respect to the parameter ξ m

Data availability
The data required to reproduce these findings are available to download from link to be
given when submission is accepted.

References
[1] ISO11119-1. Gas cylinders - Refillable composite gas cylinders and tubes - Design,
construction and testing. Part 1: Hoop wrapped fibre reinforced composite gas cylinders
and tubes up to 450L, 2012.

[2] ISO-11119-2. Gas cylinders - Refillable composite gas cylinders and tubes - Design,
construction and testing. Part 2: Fully wrapped fibre reinforced composite gas cylinders
and tubes up to 450L with load-sharing metal liners, 2012.

[3] ISO-11119-3. Gas cylinders - Refillable composite gas cylinders and tubes. Part 3: Fully
wrapped fibre reinforced composite gas cylinders and tubes up to 450L with non-load-
sharing metallic or non-metallic liners, 2013.

[4] H. Laeuffer. Caractérisation et modélisation des réseaux de fissures pour la prédiction

de la perméabilité des réservoirs composites stratifiés sans liner. PhD thesis, ENSAM,
2017.

[5] H. S. Roh, T. Q. Hua, and R. K. Ahluwalia. Optimization of carbon fiber usage in type
4 hydrogen storage tanks for fuel cell automobiles. International Journal of Hydrogen
Energy, 38:12795–12802, 2013.

[6] B. Gentilleau, F. Touchard, and J. C. Grandidier. Numerical study of influence of

temperature and matrix cracking on type IV hydrogen high pressure storage vessel
behavior. Composite Structures, 111:98–110, 2014.

35
 [7] Composites Materials Handbook - Volume 3. Polymer matrix composites materials us-
age, design, and analysis. Department of Defense Handbook, 2002.

[8] G. Verchery. The netting analysis as a limit case of the laminated structure theory.
In Proceedings of the International Conference on Composite Materials (ICCM-19),
Montréal, Canada, 28 July-2 August 2013, page 1724, 2013.

[9] M. J. Hinton and P. D. Soden. Predicting failure in composite laminates: the background
to the exercise. Composites Science and Technology, 58:1001–1010, 1998.

[10] M. J. Hinton and A. S. Kaddour. The background to the Second World-Wide Failure
Exercise. Journal of Composite Materials, 46(19-20):2283–2294, 2012.

[11] A. S. Kaddour, M. J. Hinton, P. A. Smith, and S. Li. The background to the Third
World-Wide Failure Exercise. Journal of Composite Materials, 47(20-21):2417–2426,
2013.

[12] E. Abisset, F. Daghia, and P. Ladevèze. On the validation of a damage mesomodel for
laminated composites by means of open-hole tensile tests on quasi-isotropic laminates.
Composites: Part A, 42:1515–1524, 2011.

[13] F. Daghia and P. Ladevèze. Identification and validation of an enhanced mesomodel for
laminated composites within the WWFE-III. Journal of Composite Materials, 47(20-
21):2675–2693, 2013.

[14] V. V. Vasiliev and E. V. Morozov. Mechanics and analysis of composite materials,

chapter Optimal composite structures. Elsevier, 2001.

[15] D. Violeau, P. Ladevèze, and G. Lubineau. Micromodel-based simulations for laminated

composites. Composites Science and Technology, 69(9):1364–1371, 2009.

[16] P. Ladevèze and E. Le Dantec. Damage modelling of the elementary ply for laminated
composites. Composites Science and Technology, 43(3):257–267, 1992.

[17] E. H. Baker, A. P. Cappelli, L. Kovalevsky, F. L. Rish, and R. M. Verette. Shell analysis

manual. Technical Report NASA-CR-912, NASA, 1968.

[18] M. Hojjati, V. Safavi Ardebili, and S. V. Hoa. Design of domes for polymeric composite
pressure vessels. Composites Engineering, 5(1):51–59, 1995.

[19] N. J. Pagano and J. M. Whitney. Geometric design of composite cylindrical character-

ization specimens. Journal of Composite Materials, 4:360–378, 1970.

[20] A. A. Vicario and R. R. Rizzo. Effect of length on laminated thin tubes under combined
loading. Journal of Composite Materials, 4:273–277, 1970.

36
[21] Olivier Allix, Emmanuel Baranger, and L Blanchard. An efficient strategy for the calcu-
lation of end effects on composite pipes: the thermoelastic case. Composite structures,
76(4):291–302, 2006.

[22] E Baranger, O Allix, and L Blanchard. A dedicated fourier analysis for the simula-
tion of composite pipes with defects. International journal for numerical methods in
engineering, 71(1):81–101, 2007.

[23] Emmanuel Baranger, Olivier Allix, and Laurent Blanchard. A computational strategy
for the analysis of damage in composite pipes. Composites Science and Technology,
69(1):88–92, 2009.

[24] V. V. Vasiliev. Composite Pressure Vessels: Design, Analysis and Manufacturing. Bull
Ridge Publishing, 2009.

37