International Journal of Hydrogen Energy 31 (2006) 627 – 638

www.elsevier.com/locate/ijhydene

Optimal design of a Type 3 hydrogen vessel: Part I—Analytic

modelling of the cylindrical section
D. Chapelle∗ , D. Perreux
LMARC, Institut FEMTO ST, UMR 6174, 24 rue de l’Epitaphe, 25000 Besançon, France

Available online 22 July 2005

Abstract
The present paper aims to study the cylindrical section of a Type 3 high-pressure hydrogen storage vessel, combining an
aluminium liner which prevents gas diffusion and an overwrapped composite devoted to reinforce the structure. Today, this
technique is widely used but still requires consistent time investments whenever a competitive solution, involving to definitely
increase weight efficiency, is needed. The laminate composite is assumed to be an elasto-damage material whereas the liner
behaves as an elasto-plastic material. Based on the classical laminate theory and on Hill’s criterion to take into account the
anisotropic plastic flow of the liner, the model provides an exact solution for stresses and strains on the cylindrical section
of the vessel under thermomechanical static loading. Part I focuses on the theoretical background. The effect of the stacking
sequence on the gap occurrence, on the residual stress magnitude and on the structure stiffness may then be investigated. This
will be done and be compared with results of experiments which are carried out on prototypes in the second part of this paper
before an optimization is performed.
䉷 2005 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.

Keywords: Anisotropic analysis; Elasto-plastic solution; Filament winding; Metal barrier; Hydrogen

1. Introduction drawback is the smaller gravimetric density compared with

other techniques for an identical efficiency, (c) carbon nan-
Scientists agree on the urgency to reduce pollutant emis- otubes, probably a technology at the early developmental
sions and to seek substitution energies to prevent the sched- stage, and finally (d) compressed hydrogen gas which is re-
uled collapse of fossil fuel. In this context, hydrogen is a viewed below.
promising vehicle for the future. Hydrogen field, and for in- The classification for a hydrogen pressure storage tank
stance hydrogen storage, still requires consequent improve- can be described as follows: Type 1 refers to an all metal
ments to be achieved, all the more this vehicle should be cylinder, Type 2 is a load-bearing metal liner hoop wrapped
available in next 10 years. with resin-impregnated continuous filament, Type 3 is a
Four hydrogen storage media are currently quoted [1]: non-load-bearing metal liner axial and hoop wrapped with
(a) liquefied hydrogen which allows higher volumetric and resin-impregnated continuous filament and, finally, Type 4
gravimetric density but requires the liquefaction of hydro- refers to non-load-bearing non-metal liner axial and hoop
gen gas and an efficient insulated vessel to reduce evapora- wrapped with resin-impregnated continuous filament. The
tion, (b) hydrogen storage materials [2] for which the main fibre is generally a carbon fibre although, for some peculiar
applications, manufacturers may use several layers of glass
fibre to reduce costs.
∗ Corresponding author. Tel.: +33 381666029; When thinking of the low cost for storage and transport,
fax: +33 381666700. and of the relative process simplicity of the latter technique,
E-mail address: [email protected] (D. Chapelle). its wide use is undoubtedly expected. Unfortunately, this

0360-3199/$30.00 䉷 2005 International Association for Hydrogen Energy. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijhydene.2005.06.012
628 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638

Nomenclature

Layer reference axes Specimen reference axes

1 fibre direction z axial direction
2 transverse direction  circumferential direction
3 thickness direction r (=3) radial direction

i (i = 1, 6) (k)
components of the stress tensor of a
ij (i, j = z, , r) components of the stress tensor
layer of the kth layer
i (i = 1, 6) components of the strain tensor of a (k)
ij (i, j = z, , r) components of the strain tensor
layer of the kth layer
E1 , E2 = E3 Young moduli of a layer in the fibre (k) winding angle of the kth layer
and transverse directions ep(k) thickness of the kth layer
23 , 13 , 12 (k)
i (i = 1, 6) thermal expansion coefficients of
layer Poisson ratios
c the kth layer
Sij components of the compliance ten-
Te (), (Ts ()) transforming strain (stress) rela-
sor of the layer
tions from the layer coordinate
i (i = 1, 6) thermal expansion coefficients of the
to the laminate coordinate system
layer
(winding angle is )
H damage contribution tensor

 ,  , C  ,  tensors after transforming
D damage variable

F11 , F22 , F1 , . . . classical Tsai–Wu parameters
S damage compliance tensor


C damage stiffness tensor

storage medium, in the classical range of storage pressure in each layer [9] or to focus on the design optimization of
for gas, allows small volumetric density. Considering the car dome contour [10].
industry and assuming a 0.8–1 kg of hydrogen for 100 km This paper is devoted to extend the analytic modelling
consumption, the hydrogen storage vessel of the fuel cell- in order to predict the thermomechanical behaviour of the
powered vehicle (FCV) should contain at least 5 kg of hy- cylindrical section of the hydrogen storage vessel under
drogen to be a competitive solution. As an instantaneous static loading. According to present rules on vessel design,
consequence, the major disadvantage of the compressed hy- failure must occur on the cylindrical part for the vessel solu-
drogen gas technique, regardless of the metal hydrogen em- tion to be validated. A Type 3 vessel is consequently studied
brittlement, is the high pressure required in order to reduce coupling a thin metal liner whose purpose is to prevent gas
the volume of the storage. As a demonstration, the volume diffusion and a composite laminate made by filament wind-
of 5 kg of hydrogen at a hydrogen pressure of 20 MPa is ing to ensure the vessel strength. This work was initiated by
390 l, still 250 l at 35 MPa, and not so reduced at 70 MPa, be- Lifshitz et al. [11], whose analysis, which considers a plane
cause it is 200 l. So, it is clear that no relevant breakthrough stress state, is based on a model of reduced properties, and
should be expected from an increase of internal hydrogen by Xia et al. [12,13] who wrote the theoretical background.
pressure, but attention must be paid to the gravimetric hy- Whereas Xia et al. studied the combined effect of thermo-
drogen density to be reduced. mechanical loading and internal pressure, Parnas et al. [14]
At present, composite pipes have been extensively stud- developed an analytical procedure to predict the behaviour
ied. Many experiments, considering various loadings such of fibre-reinforced composite pressure vessels under com-
as traction or compression [3,4], internal pressure loading bined mechanical and hygrothermal loading. More recently,
with different ratios of applied hoop-to-axial stress [5–7], Kim et al. [15,16] described an optimal design algorithm
have been carried out, all the more on glass/epoxy compos- including the semi-geodesic path algorithm, progressive fail-
ite, in order to identify mechanical properties and failure ure analysis and genetic algorithms, but based on finite el-
mechanisms. Regarding theoretical analysis, they are cur- ement analysis. Main advances of the present study lie in
rently based on the classical laminate theory. Rosenow [8] the following assumptions: the metal liner is supposed to
in his paper predicts the stress and strain responses of tubes be an elastic–plastic material and the laminate is an elastic-
whose winding angle varies from 15◦ to 85◦ . Thus, in thin- damage material. The anisotropic plasticity is introduced
walled filament-wound pipes, the 55◦ angle is well known by way of Hill’s criterion, and the laminate failure happens
to be the optimal angle for internal pressure with close- when the Tsai–Wu criterion [17] is satisfied. The first part
ends effect. Considering a thermomechanical loading such of this paper aims to review in detail the theoretical back-
analysis allows one to assess the residual stress magnitude ground but also to discuss the resolution algorithm. Based on
 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638 629

experimental results on glass/epoxy laminates, the ability Assuming the section remains cylindrical, the axial strain
of the present analysis to model the laminate behaviour is rate of all layers and of the liner are equal to a constant ˙0
studied. The second part will be devoted to compare ex- and the shear strain rate ˙z = 2˙z does not depend on the
perimental results obtained on prototypes of Type 3 vessels z coordinate, so ˙ z = r ˙ 0 , where ˙ 0 is twist rate per unit
and simulations coming from the analytic modelling. Fur- length. Consequently, the incremental strain–displacement
ther works will then lead us to optimize the vessel structure, relations can be written as
so as to reduce the gravimetric hydrogen density.
du̇ u̇
˙r = , ˙ = , ˙z = ˙0 ,
dr r
2. Mechanical analysis jv̇ v̇
˙rz = 0, ˙r  = − , ˙z = r ˙0 (2)
jr r
2.1. Displacements, strains and balance

In the following analysis, attention is focused on the cylin- and the incremental equilibrium equations are reduced to
drical section (see Fig. 1) of the hydrogen storage medium
subjected to internal pressure with close-end effect loading. j
˙ r
˙ r −
˙ 
+ = 0. (3)
The inner radius R0 of the liner and its thickness e are con- jr r
stant all along the cylinder. The vessel strength is ensured
by ns layers of filament-wound composite. The kth layer is All previous relations have to be written for both composite
characterised by its thickness ep(k) and winding angle (k). layers and liner.
The radial, the hoop and the axial coordinate are respectively
denoted by r,  and z. Regarding the usual assumption, the
displacement field is expressed as 2.2. Stress–strain relations

u = u(r), v = v(r, z), w = w(z), (1) The liner is an elasto-plastic material for which plastic
flow is anisotropic. Besides, the laminate behaviour is dif-
where u, v and w are radial, hoop and axial displacements. ferent from one layer to another and each layer behaves ac-
cording to the fibre direction. The fibre is assumed to have
a transverse isotropy and equivalent properties in the (2–3)
plane whose normal axis (1) refers to the fibre longitudinal
direction, as shown in Fig. 1.
General incremental stress–strain relations can be de-
scribed as

˙ = S
˙ + Ṫ ,

˙ = C(˙ − Ṫ ), (4)

where

˙ = (
˙1
˙2
˙3
˙4
˙5
˙6 ), ˙ = (˙1 ˙2 ˙3 ˙4 ˙5 ˙6 ) (5)

and

˙1 = ˙11 , ˙2 = ˙22 , ˙3 = ˙33 ,

˙4 = 2˙23 , ˙5 = 2˙13 , ˙6 = 2˙12 ,

˙ 1 =
˙ 11 ,
˙ 2 =
˙ 22 ,
˙ 3 =
˙ 33 ,

˙ 4 =
˙ 23 ,
˙ 5 =
˙ 13 ,
˙ 6 =
˙ 12 . (6)

C and S are respectively the stiffness and the compliance

instantaneous tensors.
Liner behaviour. Considering the liner as an anisotropic
elasto-plastic material, the incremental stress–strain rela-
tions can be rewritten as

Fig. 1. Cylindrical section of the storage hydrogen vessel, coordi-

nate systems definition on the laminate plane. ˙e + ˙p = (SLi Li ˙ + Ṫ ,
e + Sp )
(7)
630 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638

where e and p denote the elastic and plastic contribution, The stiffness tensor is given by
and Li refers to the liner. Li −1
⎛ S Li CLi = (SLi
e + Sp ) . (15)
Li
e 11 Se 12 Se 13
Li 0 0 0 ⎞
⎜ SeLi 21 SeLi 22 SeLi 23 0 0 0 ⎟ Laminate behaviour. Assuming (1) axis is the longitudinal
⎜ Li ⎟ direction of the fibre, the compliance tensor Sc , traducing
⎜ S S Li S Li 0 0 0 ⎟
SLi ⎜ e 31 e 32 e 33 ⎟
e =⎜ 0 0 0 SeLi 44 0 0 ⎟ the elastic behaviour of the layer, takes the same form as
⎜ ⎟ SLi
⎝ 0 0 0 0 SeLi 55 0 ⎠ e , whereas, taking into account the transversal isotropy,
the compliance constants have the following expressions:
0 0 0 0 0 SeLi 66
(8) c = 1 c = Sc = 1
S11 , S22 33 ,
E1 E2
with
c = S c = −12 , S c = −23 ,
S12
1 13 E1 23 E2
SeLi 11 = SeLi 22 = SeLi 33 = ,
E c c c 2(1 + 23 )
− S44 = 2(S22 − S23 ) = ,
SeLi 23 = SeLi 13 = SeLi 12 = , E2
E
1 c = S c = 2(1 + 12 ) ,
S55
SeLi 44 = SeLi 55 = SeLi 66 = . (9) 66 E1
G
E2 = E3 , 12 = 13 , (16)
E is the Young modulus, G the shear modulus and  the
where Eij refer to Young moduli and ij to Poisson ratios.
Poisson ratio of the metal. Due to the manufacturing pro-
The damage of laminate is assumed to change the com-
cess used to obtain the cylindrical section of the vessel, an
pliance tensor components and is introduced by adding the
anisotropy for which principal axes are r,  and z can occur.
damage contribution tensor H to the compliance tensor Sc
Consequently, Hill’s quadratic yield function is introduced
[6,7,9]. The only non-zero components of H are H22 and
and written in the physical frame:
H66 —H44 has no influence in the present analysis—and a
F (
) = f (
 −
rr )2 + g(
 −
zz )2 + h(
 −
rr )2 description of the effect of the damage on the elastic com-
pliance can be obtained by using the only damage variable
+ 2l
2r + 2m
2rz + 2n
2z
D =
E2 /E2 , it means the relative reduction of the trans-
=
2H , (10) verse Young modulus.
where f, g, h, l, m and n are the anisotropic material proper- DS c22 
2  c
S22
H22 = , H66 = D (17)
ties, and
H , H and ˙H are respectively the Hill equivalent
p p
1−D (1 − D)1/2
stress, plastic strain and plastic strain rate.
if
2 < 0 then 
2  = 0 else 
2  = 1.
 p p p
f ˙11 + g˙22 + h˙33
p
2˙23
p
2˙13
p
2˙12
1/2 Then the damaged compliance tensor
S of a layer takes
˙p = + + + . the following form [18]:
H f g + gh + hf l m n
(11)

S = S + H. (18)
−1
p
Assuming the hardening law is
H = L(H ), after some And the damage stiffness tensor is
C =
S . The damage
calculations, it can be shown that kinetics is obtained using the thermodynamics of the irre-
versible process framework. This method allows one to de-

˙
SLi
p =P p , (12) fine two potentials, one for the free energy, the other for
L (H )
2H dissipation. Thus, the driving force Y of the damage can be
where
⎛ 2 ⎞
1 1 2 1 3 l 1
23 m 1
13 n 1
12
⎜ 1 2 2 l 2
23 m 2
13 n 2
12 ⎟
⎜ 2 2 3 ⎟
⎜ 2 l 3
23 m 3
13 n 3
12 ⎟
⎜ 1 3 2 3 ⎟
P=⎜ 3 ⎟ (13)
⎜ 2l 1
23 2l 2
23 2l 3
23 2l 2
223 2lm
23
13 2ln
23
12 ⎟
⎜ ⎟
⎝ 2m 1
13 2m 2
13 2m 1
12 2lm
23
13 2m2
213 2mn
13
12 ⎠
2n 1
12 2n 2
12 2n 3
12 2ln
23
12 2mn
13
12 2n2
212
with
assessed and is given in Appendix A. The next damage cri-
1 = (h + g)
11 − h
22 − g
33 ,
terion is introduced:
2 = (h + f )
22 − h
11 − f
33 ,
3 = (f + g)
33 − g
11 − f
22 . (14) fD = −Y − Yc − RD
0, (19)
 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638 631

where Yc is a damage threshold and RD is a hardening previously introduced—in order to determine stress, strain
variable for which expression is chosen as and displacement rate. Some additional assumptions are then
specified to simplify the resolution. Assuming each layer is
ln(1 − D) 1/b thin, both stiffness tensor components and temperature rate
RD = − so
a are supposed to not vary through the layer and liner sub-
D = 1 − exp(−a(−(Y + Yc ))b ). (20) layer thicknesses. So, for each layer—and sub-layer—the
following conditions are verified:
The latter expression, in which a and b are parameters de-
pending on the material, is useful to reduce computational (k)
dṪ dCij
error by imposing the condition (D < 1), all the more the = 0, = 0, ∀i, j, k. (24)
only non-linearity taken into account herein is the damage, dr dr
excluding any plasticity. Indeed, taking into account a progressive plastification of the
Considering a rotation around the (3) axis— is the rota- liner involves arbitrarily considering nl sub-layer through
tion angle—the transformation matrix Te which allows one the metal thickness. Consequently, the structure is assumed
to write the strain vector in the cylindrical coordinate sys- to be made of (nl +ns ) layers. The stiffness tensor C(k) —for
tem is given by
⎡ cos2  sin2  0 0 0 sin  cos  ⎤
⎢ sin2  cos2  0 0 0 − sin  cos  ⎥
⎢ ⎥
⎢ 0 0 1 0 0 0 ⎥
Te = ⎢ ⎥. (21)
⎢ 0 0 0 cos  − sin  0 ⎥
⎣ ⎦
0 0 0 sin  cos  0
−2 sin  cos  2 sin  cos  0 0 0 2 2
cos  − sin 
Ts is the transformation matrix for the stress vector: Ts =
(Te−1 )t . As a consequence, we get
k
nl , C(k) = CLi(k) —of the kth layer has the following
˙ = Te ˙, expression in the cylindrical coordinate system:

˙  = Ts
,
˙ ⎛ (k) (k) (k) (k) (k) (k) ⎞
C = Ts
CTe−1 , C11 C12 C13 C14 C15 C16
⎜ (k) (k) (k) (k) (k) ⎟
 = Te  = (1 2 3 4 5 6 ). (22) ⎜
⎜
C22 C23 C24 C25 C26 ⎟
⎟
⎜ C33
(k) (k)
C34
(k)
C35 C36 ⎟
(k)
(k) ⎜ ⎟
The prime denotes the vector or tensor in the cylindrical C =⎜ (k) (k) (k) ⎟ .
⎜ C C C ⎟
coordinate system. ⎜ 44 45 46 ⎟
⎜ (k) (k) ⎟
In order to assess the strength of the structure and con- ⎝ C55 C56 ⎠
sequently to gain an estimation of the burst pressure, it is (k)
C66
necessary to introduce a failure criterion. Among the various
(25)
available criteria, it has been decided to use the Tsai–Wu
criterion, which stipulates that the non-failure is ensured as (k)
Then, the coefficients Ki are introduced:
far as the constraint
F11 (
1 )2 + F22 (
2 )2 + F66 (
6 )2 (k)

6
(k) (k)
Ki = (Cij j ). (26)
+ 2F12
1
2 + F1
1 + F2
2 − 1
0 (23)
j =1
is satisfied. F11 , F22 , F66 , F12 , F1 and F2 are the classical
Tsai–Wu parameters which depend on the composite ma- According to the stress–strain relations, the axial, hoop, ra-
terial tensile, compressive strengths
LU ,
LU ,
TU ,
TU dial and shear stresses can be expressed as
in the longitudinal and transverse direction, and in-plane (k) (k) (k) (k) (k)
shear strength,
TLU . In the light of the recent exercise [19],
˙ z = C11 ˙z + C12 ˙
known as the World-Wide Failure Exercise, and regarding (k) (k) (k) (k) (k)
+ C13 ˙r + C16 ˙ z − K1 Ṫ ,
the comparison of current failure theories for composite lam-
inates [20,21], the choice of the Tsai–Wu criterion is un- (k) (k) (k) (k) (k)

˙  = C21 ˙z + C22 ˙
doubtedly justified.
(k) (k) (k) (k) (k)
+ C23 ˙r + C26 ˙ z − K2 Ṫ ,
2.3. Problem to solve
(k) (k) (k) (k) (k)

˙ r = C31 ˙z + C32 ˙
As far as non-linearities are considered in the present anal-
(k) (k) (k) (k) (k)
ysis, it is unavoidable to regard at incremental relations—as + C33 ˙r + C36 ˙ z − K3 Ṫ ,
632 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638

(k) (k) (k) (k) (k) Displacement continuity:

˙ z = C61 ˙z + C62 ˙
(k) (k) (k) (k) (k) (k) (k)
+ C63 ˙r + C66 ˙ z − K6 Ṫ . (27) ∀k ∈ [1, nl + ns − 1], u̇(k) (Rext ) = u̇(k+1) (Rext ). (34)

Eq. (3) provides the differential equation to be solved along Radial stress continuity—Ṗ0 is the internal pressure in-
the (r) direction in order to guarantee the balance on each crement:
layer:
(k) (k) (k+1) (k)
∀k ∈ [1, nl + ns − 1],
˙ r (Rext ) =
˙ r (Rext ),
(k)
d2 u̇(k) 1 du̇(k) N (1) (nl +ns ) (n +ns )
+ − 12 u̇(k)
˙ r (R0 ) = −Ṗ0 ,
˙ r (Rextl ) = 0. (35)
dr 2 r dr r
(k) (k) 1 (k) Axial equilibrium and zero torsion condition:
= [N2 ˙0 + N3 Ṫ ] + N6 ˙ 0 (28)
r
l +ns  R (k)
n
with ext (k) R0 Ṗ0
(k)
(r
˙ z (r)) dr = ,
(k) (k) (k) Rint 2
C22 C12 − C13 k=1
(k) (k)
N1 = , N2 = , l +ns  R (k)
n
(k) (k) ext (k)
C33 C33 (r 2
˙ z (r)) dr = 0. (36)
(k)
(k) (k) (k) (k) k=1 Rint
(k) K −K (k) C − 2C
N3 = 3 (k) 2 , N6 = 26 (k) 36 . (29)
C33 C33 Finally, the problem can be reduced to a linear system under
the form
(k)
Depending on the N1 value, we get
A.X = B (37)
(k) (k) (k) r ln(r) (k)
N1 = 1: u̇r = D (k) r M1
+ + E (k) r −M1 with
2
(k) (k) (k)
× (N2 ˙0 + N3 Ṫ ) + M6 ˙ 0 r 2 ,
X = (D (1) E (1) D (2) E (2) · · · D (nl +ns ) E (nl +ns ) ˙0 ˙ 0 ). (38)
(30)

(k) (k) (k) (k) (k) The components of A and B are given in Appendix B.
N1 = 4: u̇r = D (k) r M1 + E (k) r −M1 + r(M2 ˙0
(k)
(k) N6 2.5. Algorithm
+ M3 Ṫ ) + ˙ r 2 ln(r), (31)
2 0
(k) (k)
This section focuses on the computation algorithm which
(k)
else: u̇r = D (k) r M1 + E (k) r −M1 has been developed to finally solve the linear system 37, see
(k) (k) (k) Fig. 2. As far as non-linearities are considered, whether it
+ r(M2 ˙0 + M3 Ṫ ) + M6 ˙ 0 r 2 (32)
be plastic flow of the metal liner or damage of the laminate,
with it is unavoidable to use a step by step procedure—p is the
number of steps—in order to be able to update characteristic
 (k)
(k) (k) (k) N2 values such as the damage tensor or the plastic contribution.
M1 = N1 , M2 = (k)
, First, it is worth noting that not only the liner thickness is
1 − N1
arbitrarily subdivided into sub-layers to consider a progres-
(k) (k)
(k) N3 (k) N6 sive plastic flow but also a specific control allows one to
M3 = (k)
, M6 = (k)
(33)
1 − N1 4 − N1 better circle the plastic threshold. Its purpose is to decrease
the loading step (
Ch = Ch /p) when the plastic transition is
and where E (k) and D (k) , k ∈ [1, nl +ns ], are the integration supposed to occur on the kth layer of the liner, until the next
constants. inequality for the Hill equivalent plastic stress is checked:
|
H −
0 |
. Only then and for next steps, the stiffness ten-
sor for the layer is updated, so plastic contribution is added.
2.4. Boundary conditions
Secondly, structure failure happens when the Tsai–Wu cri-
terion is no more satisfied on the ns th layer of the laminate
The number of unknown constants is 2(nl + ns + 1).
composite. Finally, the initial state of the structure, it means
These unknown constants are determined by regarding the
for instance the residual stresses resulting from the manu-
boundary conditions. In the following, inner and outer radii
(k) (k) facturing process and due to the shrinkage occurring during
of each layer are introduced, respectively Rint and Rext the cooling phase that follows the polymerisation, makes
(1) (n +n ) 
with Rint = R0 and Rextl s = R0 + e + ni=1 s
ep(i). The part of the input data as well as the material properties that
boundary conditions are then summarised as: are used all along the procedure.
D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638 633

Fig. 2. Resolution algorithm.

634 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638

3. Input data, model reliability The parameters a and b are identified using experimental
data [6]. Fig. 3 shows the analytic response and the quali-
Tables 1 and 2 summarise, respectively, the properties of tative good agreement obtained between experiments and
both composites glass/epoxy (G/E) and carbon/epoxy (C/E) modelling—for the G/E system, a = 4, b = 0.15 and for
and the aluminium alloy—Al 6060—properties used in the the C/E system, a = 5, b = 0.1. On the sub-mentioned
present analysis. graph are plotted, for G/E composite tubes submitted to
Units for dilatation coefficients, for Young moduli and internal pressure with close-ends effect, the hoop and axial
for other parameters (except Poisson ratio and h ) are simulated strains in the physical frame and on the surface
10−5 ◦ C−1 , GPa and MPa, respectively. of the specimen, as a function of the hoop stress according
Moreover, parameters K and h are introduced in order to to the stacking sequence. Experimental data [6] are not
define the plastic behaviour, i.e., the hardening law, of the plotted in Fig. 3. The number of layers is 6—no liner is
liner with a so-called Hollomon law: considered for identification and model control—and the
stacking sequences are [+45◦ , −45◦ ]3 , [+50◦ , −50◦ ]3 ,
L(H ) = K(H )h .
p p
(39) [+55◦ , −55◦ ]3 , [+60◦ , −60◦ ]3 . The model reliability is

Table 1
Composite properties

L T EL ET G LT
LU
LU
TU
TU
TLU

G/E 0.46 2.64 55 21 9 0.27 1300 900 35 110 70
C/E −0.065 2.7 150 11 4 0.3 1500 1500 50 250 70

Table 2
Aluminium alloy properties

 E 
0
r K h
Al alloy 2 72 0.25 200 250 310 0.09

Fig. 3. Simulated hoop and axial strains versus hoop stress for G/E composite with 6 layers according to the modelling. Considered winding
angles are 45◦ , 50◦ , 55◦ , 60◦ .
 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638 635

then fully established all the more the large difference be- Y I1
tween strains for 55◦ and 60◦ winding angles is clearly in- YI = − ,
Y I2
troduced. Based on macroscopic considerations, it is well
known that the optimal winding angle for internal pressure 26 S22
c (1 − D)(2 − D)
Y II = − ,
with close-ends effect is 55◦ . Moreover, as experimentally c (1 − D)1/2 + S c D)2
4(1 − D)3/2 (S66 22
observed, this optimum is a highly local optimum.
Let us emphasise that it seems possible for any compos- Y = Y I + Y II.
ite material, characterised by the fibre and the resin, to find
a couple (a, b) of values, inducing a specific damage be-
haviour, in order to simulate the mechanical experimental
Appendix B.
response of the laminate. This ability to report experimen-
tal investigations is essential when the will is to design a
Strain continuity. Condition (34) on displacements allows
structure reinforced by the same composite and to assess a
one to write the ( − 1) first lines of A and components of
theoretical burst pressure.
B as follows: ∀i ∈ [1, − 1], = nl + ns :

(i)
4. Conclusion M
A(i, 2i − 1) = −Rext1 (i),
(i)
In the first part of the present paper, attention is paid to −M1
writing the theoretical background devoted to model analyt- A(i, 2i) = −Rext (i),
ically the cylindrical section of a Type 3 vessel for hydro- (i+1)
M
gen storage applications. The analysis gives an exact solu- A(i, 2i + 1) = Rext1 (i),
tion to the linear problem, but has to be solved with a step
(i+1)
by step procedure as far as non-linearities—the anisotropic −M1
A(i, 2i + 2) = Rext (i),
plastic flow of the liner, the damage for the composite—are
taken into account. Actually, this analysis allows one to pre- (i+1) (i)
2 (i).
A(i, 2 + 2) = (M6 − M6 )Rext
dict the structure response for both thermal and mechani-
cal static loadings, and consequently to study the durabil- (i) (i+1)
ity of the cylindrical section of a Type 3 hydrogen stor- If (M1  = 1 and M1  = 1)
age vessel. In part II of the paper, complete results will
be presented: the magnitude of the thermal gap and of the (i+1) (i)
A(i, 2 + 1) = (M2 − M2 )Rext (i),
residual stresses according to the stacking sequences of
the composite structure, the expected influence of the liner (i+1) (i)
B(i) = −Ṫ (M3 − M3 )Rext (i).
anisotropy not only on the mechanical response but also on
the burst pressure, and finally the predictions of burst pres- (i) (i+1)
sure that will be compared with results of experiments that If (M1 = 1 and M1  = 1)
are presently carried out on prototype vessels. The final step
(i+1) (i)
of this study will be devoted to establish an optimisation pro- A(i, 2 + 1) = (M2 − 0.5N2 ln(Rext (i)))Rext (i),
cedure of the Type 3 structure based on previous results and
(i+1) (i)
simulations. B(i) = −Ṫ (M3 − 0.5N3 ln(Rext (i)))Rext (i).

(i) (i+1)
Appendix A. If (M1  = 1 and M1 = 1)

Y I1 = 21 S22
c S c 2 (S c − S c )2
(i+1) (i)
12 23 22 A(i, 2 + 1) = (0.5N2 ln(Rext (i) − M2 ))Rext (i),
+ 22 S22
c (S c 2 − S c S c )2
12 22 11
(i+1) (i)
+ 23 S22
c (S c 2 − S c S c )2 B(i) = −Ṫ (0.5N3 ln(Rext (i)) − M3 )Rext (i).
12 23 11
c 2 − S c S c )(S c 2 − S c S c )S c
− 23 2 (S12 23 11 12 22 11 22 Else
c 2 − S c S c )(S c − S c )S c S c
− 23 1 (S12 23 11 22 23 12 22
(i+1) (i)
c S c − S c 2 )(S c − S c )S c S c ,
− 22 1 (S22 A(i, 2 + 1) = 0.5(N2 − N2 )Rext (i) ln(Rext (i)),
11 12 22 23 12 22
(i+1) (i)
c S c 2 − S c S c 2 (1 − D)
Y I2 = 2(S11 B(i) = −0.5Ṫ (N3 − N3 )Rext (i) ln(Rext (i)).
22 11 23
c S c 2 (1 − D) − S c S c 2
− S22 12 22 12 Stress continuity. Condition (35) on stress allows one
c S c 2 (1 − D))2 ,
+ 2S23 12 to write the ( + 1) next lines of A and components of
636 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638

B—∀i ∈ [1, − 1]: Else

(i)
(i) (i) (i) M −1 A( − 1 + i, 2 + 1)
A( − 1 + i, 2i − 1) = −(C23 + M1 C33 )Rext1 (i),
(i+1) (i+1)
(i) (i) (i) −M1 −1
(i) = (0.5N2 (C23 ln(Rext (i))
A( − 1 + i, 2i) = −(C23 − M1 C33 )Rext (i),
(i+1) (i+1)
A( − 1 + i, 2i + 1) + C33 ln(Rext (i) + 1)) + C13 )
(i+1) (i) (i)
(i+1) (i+1) (i+1) M1 −1 − (0.5N2 (C23 ln(Rext (i))
= (C23 + M1 C33 )Rext (i),
A( − 1 + i, 2i + 2) (i) (i)
+ C33 ln(Rext (i) + 1)) + C13 ),
(i+1)
(i+1) (i+1) (i+1) −M1 −1
= (C23 − M1 C33 )Rext (i), B( − 1 + i)
A( − 1 + i, 2 + 2) (i+1) (i+1)
(i+1) (i+1) (i+1) (i+1) = Ṫ [−(0.5N3 (C23 ln(Rext (i))
= (M6 (C23 + 2C33 ) + C36 )
(i+1) (i+1)
(i) (i) (i) (i) + C33 ln(Rext (i) + 1)) − K3 )
− (M6 (C23 + 2C33 ) + C36 )Rext (i).
(i) (i)
(i) (i+1) + (0.5N3 (C23 ln(Rext (i))
If (M1 = 1 and M1 = 1)
(i) (i)
+ C33 ln(Rext (i) + 1)) − K3 )].
A( − 1 + i, 2 + 1)
(i+1) (i+1) (i+1) (i+1) For internal pressure,
= (M2 (C23 + C33 ) + C13 )
(i) (i) (i) (i) (1)
M1 −1
− (M2 (C23 + C33 ) + C13 ), (1)
A(2 − 1, 1) = (C23 + M1 C33 )R0
(1) (1)
,
B( − 1 + i) (1)
(1) (1) (1) −M1 −1
(i+1) (i+1) (i+1) (i+1) A(2 − 1, 2) = (C23 − M1 C33 )R0 ,
= Ṫ [−(M3 (C23 + C33 ) − K3 )
(i) (i) (i) (i) (1) (1) (1) (1)
+ (M3 (C23 + C33 ) − K3 )]. A(2 − 1, 2 + 2) = (C36 + M6 (C23 + 2C33 ))R0 .
(i) (i+1) (1)
If (M1 = 1 and M1 = 1) If (M1 = 1)

A( − 1 + i, 2 + 1) (1) (1) (1) (1)

A(2 − 1, 2 + 1) = C13 + M2 (C23 + C33 ),
(i+1) (i+1) (i+1) (i+1)
= (M2 (C23 + C33 ) + C13 ) (1) (1) (1) (1)
(i) (i) B(2 − 1) = −Ṗ0 + Ṫ [K3 − M3 (C23 + C33 )].
− (0.5N2 (C23 ln(Rext (i))
(i) (i) Else
+ C33 ln(Rext (i) + 1)) + C13 ),
(1) (1) (1)
B( − 1 + i) A(2 − 1, 2 + 1) = C13 + 0.5N2 (C23 ln(R0 )
(i+1) (i+1) (i+1) (i+1) (1)
= Ṫ [−(M3 (C23 + C33 ) − K3 ) + C33 (ln(R0 ) + 1)),
(i) (i)
+ (0.5N3 (C23 ln(Rext (i)) (1) (1)
B(2 − 1) = − Ṗ0 + Ṫ [K3 − 0.5N3
(i) (i)
+ C33 ln(Rext (i) + 1)) − K3 )]. (1) (1)
(C23 ln(R0 ) + C33 (ln(R0 ) + 1))].
(i) (i+1)
If (M1 = 1 and M1 = 1)
For external pressure,
A( − 1 + i, 2 + 1) ( )
( ) ( ) ( ) M −1
= (0.5N2
(i+1)
(C23
(i+1)
ln(Rext (i)) A(2 , 2 − 1) = (C23 + M1 C33 )Rext1 ( ),
(i+1) (i+1) −M1 −1
( )
+ C33 ln(Rext (i) + 1)) + C13 ) ( )
A(2 , 2 ) = (C23 − M1 C33 )Rext
( ) ( )
( ),
(i) (i) (i) (i)
− (M2 (C23 + C33 ) + C13 ), ( ) ( ) ( ) ( )
A(2 , 2 + 2) = (C36 + M6 (C23 + 2C33 ))Rext ( ).
B( − 1 + i)
( )
(i+1) (i+1) If (M1 = 1)
= Ṫ [−(0.5N3 (C23 ln(Rext (i))
(i+1) (i+1) ( ) ( ) ( ) ( )
+ C33 ln(Rext (i) + 1)) − K3 ) A(2 , 2 + 1) = C13 + M2 (C23 + C33 ),
(i) (i) (i) (i)
+ (M3 (C23 + C33 ) − K3 )]. ( ) ( ) ( ) ( )
B(2 ) = Ṫ [K3 − M3 (C23 + C33 )].
 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638 637

Else Then
( ) ( ) ( ) n

A(2 , 2 + 1) = C13 + 0.5N2 (C23 ln(Rext ( ))
A(2 + 1, 2 + 1) = ( f (i)),
( ) i=1
+ C33 (ln(Rext ( )) + 1)),
( ) ( ) ( ) n

B(2 ) = Ṫ [K3 − 0.5N3 (C23 ln(Rext ( )) R0 Ṗ0
B(2 + 1) = ( f (i)) + . (40)
2
( ) i=1
+ C33 (ln(Rext ( )) + 1))].
Zero torsion condition. The second equation of (36) is
Axial strength balance. The first equation of (36) is here here concerned—i ∈ [1, ]:
concerned—i ∈ [1, ]:
(i) (i)
(i) (i) M +2 M +2
M +1 M +1 Rext1 (i) − Rint1 (i)
R 1 (i) − Rint1 (i) A(2 + 2, 2i − 1) =
A(2 + 1, 2i − 1) = ext (i)
(i)
M1 + 2
M1 + 1
(i) (i) (i) (i) (i) (i)
× (C12 + M1 C13 ), × (C26 + M1 C36 )

A(2 + 1, 2i) A(2 + 2, 2i)

⎧ (i)  
(i) Rext (i) (i)
⎧ (i)  
⎪ (C − C ) ln if M1 = 1, (i) (i)
⎪ 12
⎪
⎪
13 Rint (i) ⎪ (C26 − 2C36 ) ln R ext (i)
if M1 = 2,
⎨ −M (i) +1 (i) ⎪
⎪ Rint (i)
−M +1 ⎪
⎪
= Rext 1 (i)−Rint 1 (i) ⎪
⎨ −M (i) +2 (i)
⎪
⎪ (i) otherwise, −M +2
⎪
⎪ 1−M1 = Rext 1 (i)−Rint 1 (i)
⎩ (i) (i) (i) ⎪
⎪ (i) otherwise,
×(C12 − M1 C13 ) ⎪
⎪ 2−M1
⎪
⎪
n
 ⎩ (i) (i) (i)
 3 (i) − R 3 (i)
Rext ×(C26 − M1 C36 )
A(2 + 1, 2 + 2) = int
3 n

i=1  4 (i) − R 4 (i)
Rext
 A(2 + 2, 2 + 2) = int
(i) (i) (i) (i) 4
× (C16 + M6 (C12 + 2C13 )) . i=1

(i) (i) (i) (i) (i)
If (M1 = 1) × (C66 + M6 (C26 + 2C36 )) .

2 2
f (i) = Rext (i) − Rint (i) (C (i) + M (i) (C (i) + C (i) )), (i)
11 2 12 13 If (M1 = 1)
2
2 2
f (i) = Ṫ Rext (i) − Rint (i) (K (i) − (M (i) (C (i) + C (i) ))).
3 3
c (i) = Rext (i) − Rint (i) (C (i) + M (i) (C (i) + C (i) )),
2 1 3 12 13 16 2 26 36
3
Else 3 3
 (i) c (i) = Ṫ Rext (i) − Rint (i) (K (i) − M (i) (C (i) + C (i) )).
f (i) = 1 N2 (i) 2 3 4 3 26 36
(C12 (Rext (i)(ln(Rext (i) − 1))))
2 2 Else
2 (i)(ln(R (i) − 1))
− Rint  (i)
int N2
(i) 2 c (i) = 1 (i)
C26 3 (i) ln R (i) −
Rext
1
+ C13 (Rext (i)(ln(Rext (i) + 1))) 3 2
ext
3
2 (i)(ln(R (i) + 1))
− Rint int
 3 (i) ln R (i) − 1
(i) 2
− Rint int
+ C11 (Rext 2 (i)) ,
(i) − Rint 3

(i) 3 (i) ln R (i) + 2

+ C36 Rext ext
f (i) = Ṫ [K (i) (R 2 (i) − R 2 (i)) 3
ext int
2 1
(i) (i) 2 3 (i) ln R (i) + 2
− 0.5N3 (C12 (Rext (i)(ln(Rext (i) − 1)))) − Rint int
3
2 (i)(ln(R (i) − 1))
− Rint int 
(i) 2 (i)(ln(R (i) + 1))) (i) 2
+ C13 (Rext ext + C16 (Rext 2 (i)) ,
(i) − Rint
2
− Rint (i)(ln(Rint (i) + 1))].
638 D. Chapelle, D. Perreux / International Journal of Hydrogen Energy 31 (2006) 627 – 638


c (i) = Ṫ
[7] Perreux D, Lazuardi D. The effects of residual stress on
(i)
3 (i) − R 3 (i))
K4 (Rext int the non-linear behaviour of composite laminates. Part I.
3 Experimental results and residual stress assessments. Compos
(i) (i) 3 (i) ln R (i) − 1 Sci Technol 2001;61:167–75.
− 0.5N3 C26 Rext ext
3 [8] Rosenow MWK. Wind angle effects in glass fiber-reinforced
1 polyester filament-wound pipe. Composites 1978;9:263–71.
3 (i) ln R (i) −
− Rint int [9] Perreux D, Lazuardi D. The effects of residual stress on the
3
non-linear behaviour of composite laminates. Part II. Layer,
+ C36 3 (i) ln R (i) + 2
(i)
Rext laminate non-linear models and the effect of residual stress
ext
3 on the model parameters. Compos Sci Technol 2001;61:

3 (i) ln R (i) + 2 177–90.
−Rint int . [10] Liang CC, Chen HW, Wang CH. Optimum design of dome
3
contour for filament-wound composite pressure vessels based
Then on a shape factor. Comp Struct 2002;58:469–82.
[11] Lifshitz JM, Dayan H. Filament-wound pressure vessel with
n
 thick metal liner. Comp Struct 1995;32:313–23.
A(2 + 2, 2 + 1) = ( c (i)), [12] Xia M, Takayanagi H, Kemmochi K. Analysis of multi-layered
i=1 filament-wound composite pipes under internal pressure.
n
 Comp Struct 2001;53:483–91.
B(2 + 2) = ( c (i)). (41) [13] Xia M, Kemmochi K, Takayanagi H. Analysis of filament-
i=1 wound fiber-reinforced sandwich pipe under combined
internal pressure and thermomechanical loading. Comp Struct
(i)
The case N1 = 4 is not detailed herein but is taken into 2001;51:273–83.
account in the calculation procedure. [14] Parnas L, Katırcı N. Design of fiber-reinforced composite
pressure vessels under various loading conditions. Comp
Struct 2002;58:83–95.
References [15] Kim C.-U, Hong C.-S, Kim C.-G, Kim J.-Y. Optimal design
of filament wound type 3 tanks under internal pressure using
[1] Züttel A. Materials for hydrogen storage. Materials Today a modified genetic algorithm. Comp Struct 2004, in press.
2003;6(9):24–33. [16] Kim C-U, Kang J-H, Hong C-S, Kim C-G. Optimal design
[2] Carter TJ, Cornish LA. Hydrogen in metals. Eng Failure Anal of filament wound structures under internal pressure based
2001;8:113–21. on the semi-geodesic path algorithm. Comp Struct 2005;67:
[3] Bai L, Seeleuthner P, Bompard P. Mechanical behaviour 443–52.
of ±55◦ filament-wound glass-fiber/epoxy-resin tubes: I. [17] Tsai SW, Wu EM. A general theory of strength for anisotropic
Microstructural analysis, mechanical behaviour and damage materials. J Comp Mater 1971;5:58–80.
mechanisms of composite tubes under pure tensile loading, [18] Perreux D, Oytana C. Continuum damage mechanics of
pure internal pressure and combined loading. Compos Sci microcracked composites. J Comp Eng 1993;2(3):115–22.
Technol 1997;57:141–53. [19] Hinton MJ, Soden PD. Predicting failure in composite
[4] Smerdov AA. A computational study in optimum formulations laminates: the background of the exercise. Compos Sci
of optimization problems on laminated cylindrical shells for Technol 1998;58:1001–10.
buckling: I. Shells under axial compression. Compos Sci [20] Hinton MJ, Kaddour AS, Soden PD. A comparison of the
Technol 2000;60:2057–66. predictive capabilities of current failure theories for composite
[5] Rousseau J, Perreux D, Verdière N. The influence of winding laminates, judged against experimental evidence. Compos Sci
patterns on the damage behaviour of filament-wound pipes. Technol 2002;62:1725–97.
Compos Sci Technol 1999;59:1439–49. [21] Kaddour AS, Hinton MJ, Soden PD. A comparison of the
[6] Perreux D, Thiebaud F. Damaged elasto-plastic behaviour predictive capabilities of current failure theories for composite
of [+, −] fibre-reinforced composite laminates in biaxial laminates: Additional contributions. Compos Sci Technol
loading. Compos Sci Technol 1995;54:275–85. 2004;64:449–76.