Professional Documents
Culture Documents
Design of Cylindrical Composite Pressure Vessels Integral Optimisation - Koussios
Design of Cylindrical Composite Pressure Vessels Integral Optimisation - Koussios
S. Koussios
Design & Production of Composite Structures
Faculty of Aerospace Engineering
Delft University of Technology, Kluyverweg 1, 2629 HS, Delft, The Netherlands
s.koussios@tudelft.nl
SUMMARY
In this paper we outline an analytical method for the integral optimisation of
cylindrical composite pressure vessels incorporating additional constraints like
pattern consistency and strain compatibility at the cylinder dome intersection. The
results reveal that designs exclusively based on shape and material parameters are in
some cases far from optimal.
1. INTRODUCTION
z Fa
σθ Fibre bundle
σφ
α
c ρ σφ θ
φ
P
φ R y
σθ
ν xy ν yx
= (1)
Ex Ey
where the first subscript denotes the applied stress direction and the second stands
for the direction of the generated (transverse) strain.
For the stress / strain transformations we employ here the engineering stress &
strain based matrices. The stress transformation matrix is:
1 −ν 12
0
E1 E1
−ν 1
C = 12 0 (4)
E1 E2
1
0 0
G12
S=
1
(S(α ) + S(−α ) ) (5)
2
where:
Due to rotational symmetry for the shape and the applied loads (internal
pressure and axial loading, figure 1), the shell stresses do not have a shear
component hence the applied stress vector is [5]:
σ θ
σ = σ φ (7)
0
The strains in the main shell direction (θ and φ reference system, figure 1) are given
by:
ε = C.σ (8)
Where C = S-1. The strains in a particular layer (with respect to its material axes)
are given by:
The components of the layer stresses vector form the input parameters for the
selected stress-based strength criterion (e.g. Tsai-Hill). Alternatively, one can plug
the result of equation (9) into a strain-based criterion. For example, in [16] the
authors have employed a conservative version of the Hill criterion:
2 2 2
σ 1 σ 2 τ 12
+ + ≤1 (11)
S1 S 2 T12
where S1, S2, are the tensile / compressive strength values of a single layer in
respectively the “1” and “2” direction (1 is the fibre direction), and T12 the in-plane
shear strength. Depending on the sign of σ# one should select the appropriate tensile
or compressive value for S#. Based on this criterion, Vasiliev, Krikanov and Razin
[16] derived the optimality conditions (minimum weight by maximum stored
energy) for a pressure vessel by maximizing the stress invariant σ1+σ2 (=σθ+σφ)
under constraint (11). These conditions imply zero shear stress in the participating
layers, τ12 = 0. The same result, based on the assumption of equal layer strains
(ε1 = ε2) has been derived by de Jong [8]. In this work however, the full form of the
Tsai-Hill criterion is employed [3]:
2 2 2
σ σ σ σ τ
Ω = 1 + 2 − 1 2 2 + 12 ≤ 1 (12)
S1 S 2 S1 T12
σ φ sin 2 α + ke cos 2 α
= (13)
σ θ cos 2 α + ke sin 2 α
where [10]:
E2 (1 + ν 12 )
ke = (14)
E1 (1 + ν 21 )
2
S
k σ = 2 (15)
S1
For the netting case (where the mechanical properties of the matrix are neglected)
relations (14) and (15) simplify to ke = kσ = 0. On the other hand, for k = 1 the
utilised material becomes isotropic. Note that the condition of maximum strength
does not automatically imply maximised vessel performance since the employed k
value will affect the resulting meridian profile, hence the contained volume of the
calculated vessel. From this point we will use the notation k (0 ≤ k ≤ 1) for the
anisotropy parameter.
Depending on the k parameter, several profiles for the meridian of the optimal dome
can be obtained. Before proceeding to the analysis of such profiles we introduce
some design parameters and dimensionless variables, figure 1:
• Y = ρ/c Y≥1
• Yeq = R/c Yeq ≥ 1
• Z = z/c Z>0
• H = h/c H≥0
Fa
• r =
π PR 2
1
α = arcsin (16)
Y
As derived in [10], for a given every {Yeq, r, k} combination, the meridian profile
Z(Y) is given by integration of:
Y (Y 2 + rYeq2 )
Z '(Y ) = ± (17)
k +1
k + Y −1
(1 + r )2 Yeq6 − Y 2 (Y 2 + rYeq2 )
2
2
k + Yeq − 1
2
The integration interval is [Ymin, Yeq] where Ymin is in general not equal to 1. The
denominator of the above given equation might nullify for Y values slightly bigger
(in some cases also slightly smaller) than 1 [8, 10], so in practice the aimed aspect
ratio R/c might not be exactly achievable. However, the differences are negligibly
small. In the calculations presented here, we assume that:
k +1
k + Y 2 −1
(1 + r ) 2 Yeq6 − Y 2 (Y 2 + rYeq2 ) = 0
2
Y :
= max
k + Yeq − 1
2
Ymin (18)
1
5
1
↑
Z 0.8
[-] 4
0.6
0.4
0.2
3 0
k
2
Y [-] →
2 3 4 5
k +1
Pc (1 + r )Yeq k + Y 2 − 1
3 2
σ θ(d) =
k + Yeq − 1
2 2
2td Y (19)
1 + k (Y 2 − 1) (d)
σ φ(d) = σθ
k + Y 2 −1
where td stands for laminate thickness at the dome. With a given {Yeq, r, k} vector
and known engineering constants for the used material, the shell stresses can be
plugged into the strength analysis procedure as outlined in sub-section 2.2. The
result is the minimally required laminate thickness for the dome as a function of the
Y coordinate:
td (Y ) =
1
2
( {
Pc max Ω(α ), Ω(−α ) }) Y
(20)
where Ω stands for the non-conservative form of the Hill criterion with active
equality, equation (12). The subscript ‘d’ refers to the dome.
Hoop
windings
(incomplete)
1 f
S c = S d (Yeq ) + S(π / 2) (21)
1+ f 1+ f
where:
th = ftd (Yeq )
f ∈ ℕ+ (22)
tc = th + td (Yeq )
The subscript ‘c’ stands for cylinder, ‘h’ for hoop and ‘p’ for polar. The latter refers
to the windings originating at the dome and continue over the cylindrical surface
with an angle α(Yeq), fig. 3. The dome strains at the cylinder dome intersection (Yeq)
are:
(1 + r ) Yeq
ε θ(d) (Yeq ) σ θ(d) (Yeq )
1 + k (Yeq − 1)
2
(d) (d) Pc
ε d (Yeq ) = ε φ (Yeq ) = Cd [α (Yeq )]. σ φ (Yeq ) = Cd [α (Yeq )].
k + Yeq − 1
2
0 0 2t d (Yeq )
0
(23)
The cylindrical part is loaded by [8, 10]:
(1 + r )Yeq
Pc
σc = 2Yeq (24)
2 ( td (Yeq ) + tc )
0
which gives:
k0 ( f ) = k : ε φ( d ) (k ) = ε φ( c ) (k , f ), f ∈ ℕ+ (26)
In practice, consideration of 1 to 4 hoop layers does usually cover the feasible
domain for minimum weight design. With the additional hoop layers however, the
cylindrical part is in most cases too strong so there is some excess weight. Creating
hoop layers with weaker (and probably also lighter) fibres would in this case
improve the performance. In addition, k0 does usually overshadow the profitable
effect of the optimal k (ke for maximum performance or kσ for maximum strength).
In some cases, designing with ke (without taking care for strain compatibility) might
even knock down the performance by a factor 5 [10].
3. VESSEL PROPERTIES
3.1 Definitions
PV
η= [103 m] (27)
W
where P is the applied pressure [106Nm-2], V the contained volume [-] and W the
weight of the vessel per unit volume [103Nm-3].
The applied pressure is given as a design parameter. In addition, we assume
that here that the externally applied axial load is equal to zero. To quantify the
performance, one has to determine the internal volume and the weight of the
laminate. The first unknown parameter is straight forward, but for the calculation of
the resulting laminate thickness along the meridian profile (and thereof the volume
and weight) we can use several approximation theories [12, 16]. Regardless the
method of schematising the thickness distribution, the nature of the winding process
dictates that a certain roving length Ltot (dimensionless) is to be placed on the
mandrel. As this roving passes through the resin bath, a well determined amount of
resin is picked up. Therefore we assume here that the laminate weight is simply
given by:
W = gδ Ltot BT [103 Nm −3 ] (28)
where B and T account for the width, respectively thickness of the roving-resin
combination (both dimensionless) and δ for the associated combined density
(mixture law). Not surprisingly, the result is practically identical to those given by
integrating the various in the literature available thickness approximations over the
vessel surface [12].
3.2 Dome
1 + Z '2 (Y )
φ 'd (Y ) = tan α (Y ) (29)
Y
The φ-propagation over a single trajectory (e.g. departing at the equator and arriving
at the pole) can be obtained by integration of (29) over the interval
[Yeq, Ymin]. As the dome is covered by two of these trajectories
(equator→pole→equator) and the vessel has obviously two domes, the total, dome
related φ-contribution is obtained by multiplying the integration result by 4.
The roving length differential is given by:
1 + Z '2 (Y )
L 'd (Y ) = (30)
cos α (Y )
With a similar reasoning, the total, dome related roving length is 4 times the integral
of (30) over [Yeq, Ymin]. The internal volume differential is:
Hence the total dome related volume is 2 times the integral of (31) over [Yeq, Ymin].
Note that all quantities presented here are dimensionless.
H
Φ (p)
c =2 tan α (Yeq ) (32)
Yeq
where ‘p’ stands for polar winding. The associated roving length is:
H
c =2
L(p) (33)
cos α (Yeq )
The hoop layers result in a laminate thickness as given in equation (22a). The
total laminate volume of the hoop layers is 2π f YeqHTeq, which is equal to Lc(h)BT.
Hence:
H Teq
c = 2π f
L(h) (34)
B T
Vc = π Yeq2 H (35)
Yeq Teq
2π YeqTeq = 2 nd
BeqT → ζ = π (36)
ζ Beq T
where:
B
Beq = (37)
cos α (Yeq )
With the number of windings ζ (also referred to as coverage parameter) given, the
total roving length (including the hoop layers) becomes:
Yeq (h)
Ltot = ζ 4 ∫ L 'd (Y )dY + L(p)
c + Lc (38)
Y
min
In combination with equation (28) one can calculate the weight of the laminate per
unit of volume.
8
7 C
9
6
10
5 11
A, D
4 12
13
1
3 B
E 14
2 ∆Φ
lagging
15
leading
∆ϕ
4. WINDING PATTERNS
4.1 Principles
( p + 1)kd − nd = +1 Leading
pkd − nd = −1 Lagging (39)
p, k , nd ∈ ℕ +
The realisation of a pattern is entirely dependent on the angle ∆Φ, measured on the
Yeq-periphery, between two adjacent windings. In figure 4 this would be the
. The unbounded value of this angle is given by:
angle BE
Yeq
Φ tot = Φ + 4 ∫ φ'
(p)
c d dY (40)
Ymin
{
∆Φ = min mod (2π ) ( Φ tot ) , mod ( −2π ) ( Φ tot ) } (41)
In addition, the effective roving width Beq (equation (37)) is converted in:
Beq
∆ϕ = (42)
Yeq
2π
p = IP
∆Φ
(43)
∆Φ
k = CE
∆ϕ
where IP stands for ‘integer part’ and CE for ‘ceiling’. The last parameter, nd, is the
biggest integer containing ζ (equation (36)):
nd = CE (ζ ) (44)
Apparently, the coverage parameter is the only link between patterns and
compliance with the minimally required thickness Teq.
With the pattern constants given, the angle between two neighbouring windings
should satisfy (see also equation (39)) [14]:
2π 1
∆Φ leading = 1 +
p +1 ζ
(45)
2π 1
∆Φ laging = 1 −
p ζ
4.3 Solution procedure
For a given set of vessel design parameters {Yeq, r} and a vector of realistic
numbers for the additional hoop layers, e.g. f = (1, 2, 3, 4) a series of k0(f) values
will emerge (equation (26)). Among them, a particular f value in combination with
H (given dimensionless cylindrical length) will provide the best performance η
(equation (27)). The latter is independent of the selected {B, T} so the performance
is not affected. After all, we can achieve the same laminate thickness with a big
number of narrow/thin rovings or a smaller number of broad/thick fibre bundles
without affecting the weight. Having a predetermined range for {B, T} (usually
given in the form of discrete values due to availability), the first condition that must
be satisfied is:
?
∆Φ leading = ∆Φ tot
?
(46)
∆Φ lagging = ∆Φ tot
These relations (if there is a solution within the provided {B, T} range) result in a
series of:
TB = const . (47)
With these expressions, the coverage parameter ζ can be quantified. The resulting
nd value(s) (equation (44)) can now be plugged in equation (39) to exactly pinpoint
the B-value(s) that satisfies(y), next to the structural requirements, the pattern
conditions. The number of feasible patterns depends on the number of feasible
solutions for (47).
5. EXAMPLE
• P = 10 [MPa]
• r=0 [-]
• c = 25 [mm]
• Yeq = 8 [-]
• H = 15 [-]
• {Bmin, Bmax} = {0.1, 0.5} [-]
• {fmin, fmax} = {1, 5} [-]
Note that c does not influence the performance neither does it have any impact on
the winding patterns. As everything here is dimensionless it serves only the purpose
of “translating” the design into a real volume. Therefore, every desired volume can
be obtained by properly adjusting this parameter. In addition, the roving thickness is
here a result so it does not have to be known in a-priori fashion.
The layer properties for the employed laminate are assumed as:
• Ex = 60000 [MPa]
• Ey = 9800 [MPa]
• vxy = 0.3 [-]
• Gxy = 3400 [MPa]
• S1, tension = 1500 [MPa]
• S1, compression = -1350 [MPa]
• S2, tension = 40 [MPa]
• S2, compression = -210 [MPa]
• T12= 50 [MPa]
With c = 25 [mm] this would mean that the suitable roving thicknesses are
{0.4, 0.84}. As a thickness of 0.4 [mm] is more realistic, it is advised to choose here
for the second option. It should be noted however that neither the structural
performance nor the winding time is affected by this selection.
6. CONSLUSIONS
1. Baker EH, Kovalevsky L, Rish FL. Structural analysis of shells. New York: McGraw Hill Book
Company, 1972.
2. Cho-Chung Liang, Heng-Wen Chen, Cheng-Huan Wang. Optimum design of dome contour for
filament winding composite pressure vessels based on a shape factor. Composite Structures
2002; 58: 469-482.
3. Daniel IM, Ishai O. Engineering Mechanics of Composite Materials. New York, Oxford: Oxford
University Press, 2006.
4. Design of Filament Wound Pressure Vessels. European Space Agency, February 1994.
Structural Materials Handbook: Vol. 1: Polymer Composites: Section VI: Design of Structures,
chapter 29.
5. Flügge W. Stresses in Shells. Berlin / Heidelberg / New York: Springer Verlag, 1966.
6. Fukunaga H, Uemura M. Optimum Design of Helically Wound Composite Pressure Vessels.
Composite Structures 1 (1983) 31-49.
7. Gray A. Modern Differential Geometry of Curves and Surfaces. CRC press, 1993.
8. Jong de Th (in Dutch). “Het wikkelen van drukvaten volgens de netting theorie”. Report VTH-
166. Structures and materials laboratory, Faculty of Aerospace Engineering, Delft University of
Technology. Delft, April, 1971.
9. Jong de Th. A theory of filament wound pressure vessels. Report LR-379. Structures and
materials
10. Koussios S, Bergsma OK, Beukers A. The Role of Material Anisotropy in the Optimal Design
of Cylindrical Composite Pressure Vessels. 22nd Annual Conference of the American Society for
Composites. Seattle, WA, September 2007.
11. Koussios S, Beukers A. Composite Pressure Vessel Design: Integral Determination of Winding
Patterns. 16th International Conference on Composite Materials. Kyoto, Japan, July 2007.
12. Koussios S, Beukers A. Influence of Laminate Thickness Approximation Methods on the
Performance of Optimal Filamentary Pressure Vessels. 23rd Annual Conference of the American
Society for Composites. Memphis, TN, September 9-11, 2008.
13. Koussios S, Beukers A. Manufacturability of Composite Pressure Vessels: Application of Non-
Geodesic Winding. 16th International Conference on Composite Materials. Kyoto, Japan, July
2007.
14. Koussios S. Filament Winding: a Unified Approach. PhD. thesis. Faculty of Aerospace
Engineering, Delft University of Technology. Delft University Press, 2004.
15. Peters ST, Humphrey WD, Foral RF. Filament Winding Composite Structure Fabrication.
Covina CA: SAMPE International Business Office, 1999.
16. Vasiliev VV, Krikanov AA, Razin AF. New generation of filament-wound composite pressure
vessels for commercial applications. Composite Structures 2003: 62: 449-459.