You are on page 1of 17

Composites: Part B 32 (2001) 413±429

www.elsevier.com/locate/compositesb

The effect of ®ber volume fraction on ®lament wound composite


pressure vessel strength
David Cohen a,b, Susan C. Mantell b,*, Liyang Zhao b
a
Toray Composites (America) Inc., 19002 50th Ave. E., Tacoma, WA 98446, USA
b
Department of Mechanical Engineering , Institute of Technology, University of Minnesota, 111 Church Street SE, Minneapolis, MN 55455-0111, USA
Received 15 November 2000; accepted 21 February 2001

Abstract
This paper is a continuation of previous research reported in Ref. [1]. The previous paper discussed the relationship between ®ber volume
fraction in ®lament wound composite vessels and failure pressure. This research included a design of experiment investigation of manu-
facturing and design variables that affect composite vessel quality and strength. Statistical analysis of the data shows that composite vessel
strength was affected by the manufacturing and design variables. In general, it was found that the laminate stacking sequence, winding
tension, winding-tension gradient, winding time, and the interaction between winding-tension gradient and winding time signi®cantly
affected composite strength. The mechanism responsible for increases in composite strength was related to the strong correlation between
®ber volume fraction in the composite and vessel strength. Cylinders with high-®ber volume in the hoop layers tended to deliver high-®ber
strength. This paper further examines the relationship between ®ber volume fraction and ®ber strain to failure. Data from unidirectional
strand tests and additional vessel tests are presented. A computer program that is based on the thermokinetics of the resin and processing
conditions is used to calculate the ®ber volume fraction distribution in the ®lament wound vessel. The strand's strength-versus-®ber volume
data together with the computer program are used to predict composite vessel burst pressure. In general, good agreement with experimental
data is observed. q 2001 Elsevier Science Ltd. All rights reserved.
Keywords: A. Fibers; B. Residual/internal stress; E. Filament winding; Strength

1. Introduction where sÅ f is the average ®ber strength, sÅm the average matrix
strength and sc is the composite strength.
The idea that composite strength in the ®ber direction is The above statement is not intended to trivialize previous
a function of the ®ber volume fraction is not new, e.g. research. In these works [2±4], the relationship between
Refs. [2±4]. Previous researchers have treated composite volume fraction and ®ber strength was not a simple rule-
strength as a function of the ®ber volume fraction in terms of-mixture as stated by Eq. (1). Rather, these derivations
of the stress. However, in both this paper and the previous involved complex micromechanical models based on the
paper [1], composite strength in the ®ber direction is treated probabilistic failure characteristic of the ®ber. However,
as a function of the ®ber strain-to-failure. Fiber strain-to- when strength data are plotted as a function of stress and
failure may be a better material variable to characterize the volume fraction, e.g. Ref. [3], the true effect of volume
true relationship between composite strength in the ®ber fraction on strength is dif®cult to determine (because the
direction and ®ber volume fraction. From a simplistic cross-sectional area of the ®ber changes). In this regard, it
point of view, a rule of mixtures approach indicates that may be argued that the ability of the ®ber to stretch should
as the ®ber volume fraction decreases, the composite be an independent material property regardless of the ®ber
strength will also decrease proportionally. volume fraction in the composite. Further, unlike stress,
strain may be considered a dimensionless variable (i.e.
s c ˆ Vf s f 1 …1 2 Vf †s m …1† mm/mm), and as such, it does not require normalization
by the ®ber cross-sectional area. In this paper, volume frac-
* Corresponding author. Address: Department of Mechanical Engineer- tion-versus-strain data from both carbon ®ber strands and
ing, Institute of Technology, University of Minnesota, 111 Church Street
carbon/epoxy ®lament-wound pressure vessels are discussed.
SE, Minneapolis, MN 55455-0111, USA. Tel.: 11-612-625-1324; fax: 11-
612-625-9395. The data show a strong relationship between ®ber strain to
E-mail address: smantell@me.umn.edu (S. Mantell). failure and the ®ber volume fraction in the composite.
1359-8368/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved.
PII: S 1359-836 8(01)00009-9
414 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Despite the knowledge that ®ber strength in the compo- load-transfer


p p length decreases with ®ber volume (i.e. {1 2
site is a function of the ®ber volume fraction very little Vf }= Vf is a decreasing function of Vf) Eq. (4) predicts
experimental work has been presented in the literature. that composite strength will increase with ®ber volume.
This aspect of composite quality can be very signi®cant The experimental data from strand and ®lament wound
particularly for strength-critical space structure systems composite vessels that supports the theoretical derivations
where weight is an important variable. If the structure light- of increases in ®ber strain-to-failure with ®ber volume frac-
ens (e.g. by using composite strap-on boosters [5]), the size tion are presented in this paper. No attempt is made to tie the
of the payload can increase for a given propellant volume. empirical observations to the equations presented above.
This paper demonstrates how 10% (or greater) improvement Rather, we focus on the effect of ®lament winding proces-
in composite ®ber strength can be obtained by changing the sing parameters on the ®nal composite quality and the inter-
composite ®ber volume fraction from 50 to 65% by volume. action between quality and vessel strength. In this regard,
It should be stressed that manufacturing high-quality ®la- the approach is empirical in nature and directed toward the
ment wound vessels with high-®ber volume fractions is not engineering and processing of ®lament wound composite
a trivial matter. This is particularly true for very large vessels.
diameter (e.g. the ®lament wound boosters for the Space
Shuttle [5]) wet-wound structures that cannot be cured in
an autoclave. Filament winding with towpreg can improve 2. Test program
the ®nal composite quality and control of the ®ber volume.
However, wet winding is still a more ef®cient and cost In the previous paper [1], it was hypothesized that the
effective method to produce large ®lament wound struc- ®ber volume in the hoop layers of a ®lament wound pressure
tures. Therefore, a method for producing high-quality vessel controls the strength of the vessel. To validate this
composites with high-®ber volume, together with a better hypothesis, it was necessary to develop a `pure' ®ber
understanding of the relationship between volume fraction strength-versus-®ber volume interaction curve that can be
and ®ber strength in the composite, must be developed. used to calculate vessel burst pressure. Because ®ber
The classical approach [6±9] to deriving the relationship volume fraction in wet wound vessels can vary from hoop
between ®ber strength and ®ber volume fraction is based on ply to hoop ply (Fig. 1), that is, each layer can act as the
the characteristic load-transfer length, d , between ®laments. weakest link, and, because strain distribution through the
This relationship depends on the unit cell cross-sectional vessel thickness is not uniform (Fig. 1), failure can initiate
size that is governed by the ®ber arrangement and volume in any of the hoop plies depending on the strength (as deter-
fraction. The dependence of the load-transfer length, d , on mined by the ®ber volume). As the vessel pressure increases
the ®ber modulus, Ef, resin modulus, Em, ®ber diameter df, (for example from P1 to failure pressure), so does the strain
and ®ber volume fraction Vf is given by [7] in the vessel. The ®rst ply to fail is the ply at which the
 p  applied strain equals the failure strain. In the example
3Ef …1 2 Vf † 1=2 presented in Fig. 1, this is hoop layer No. 3. Note that,
d ˆ df p …2†
2Em Vf hoop layer No. 7 has less strength. Nevertheless, the applied
strain in this layer is below the failure strain. It should also
The average ®ber bundle failure strain, e bf , is derived in
be noted that the ®rst hoop ply (closest to the inside
terms of a ®ber bundle length, L, (i.e. ®ber strength is length
diameter), has the highest applied strain, yet this layer is
dependent) and the ®ber probability of failure as described
not the weakest layer (in this case) and, therefore, not the
by the Weibull distribution function [7]
  location of failure. Given the aforementioned hypothesis, it
1 was necessary to develop a method for fabricating ®ber
e bf ˆ e0 L21=a G 1 1 …3†
a strand specimens with controlled ®ber volume fraction
such that a pure ®ber strain-to-failure versus volume frac-
where G(´) is the gamma function, a the Shape parameter tion interaction curve can be obtained. This curve is then
for Weibull distribution of ®ber failure strain, and e0 is the used to correlate the calculated burst pressures of ®lament
Scale parameter for Weibull distribution of ®ber failure wound vessels with test data.
strain (a and e 0 are ®ber properties). In addition, a computer model was developed [10±12] to
The average strength of a ®ber bundle of length L is equal calculate the ®ber volume distribution in full-scale vessels
to s bf ˆ Ef e bf and the composite strength is given by a prior to burst. The ®ber volume distribution from the
modi®ed rule-of-mixture as [7] computer model together with the pure ®ber strain-to-failure
  1=a strength interaction curve are used to predict failure pressure
1 L
sc ˆ 1=a
Vf s f 1 …1 2 Vf †s m …4† of full-scale vessels prior to the actual burst. The material
…ae† G…1 1 1=a† d
from the vessels is then analyzed using the image analysis
in which s f and s m are the average ®ber and matrix (IA) method, and the measured ®ber volume is compared to
strength. Hence, composite strength is inversely propor- predicted distributions. Given the model predicted ®ber
tional to the load-transfer length,d . Therefore, because the volume fractions and the empirically determined relationship
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 415

ASTM D 4018 standard for Tensile Properties of Con-


tinuous Filament Carbon and Graphite Yarns, Strands,
Rovings, and Tows.

2.2. Subscale 51-cm vessel test program

In many ®lament wound vessels, a critical parameter in


performance is the burst pressure. Because of the large cost
associated with hydroburst testing large-diameter full-scale
vessels, accurate prediction of performance has increased in
importance. A subscale test analog to measure the vessel's
hoop strength and stiffness can be an effective tool. Small-
scale test articles can also provide a cost effective method to
assess the effect of materials, fabrication, design, and
Fig. 1. Schematic presentation of composite case failure scenario as a
processing variables on vessel strength and stiffness perfor-
function of the hoop layer strain to failure, which is dependent on the
®ber volume fraction for that particular layer. mance. A recent paper [13] describes the development of a
test method that is capable of pressure testing 51 cm
diameter rings with representative composite pressure
between vf and ®ber strain to failure, vessel burst strength is vessels' thickness and laminate lay-ups. This paper demon-
predicted. Good correlation was obtained between the strated that tests of rings with various laminate thicknesses
predicted and actual burst pressures. The computer and lay-ups have excellent agreement with measured hoop
program, together with the test methods discussed, are ®ber strain to failure and stiffness measured in full-scale
useful tools in assessing the performance penalty for vessels composite pressure vessels. The test data showed near
in which manufacturing deviations (e.g. longer winding one-to-one scaling in strength and stiffness. The issue of
delays) are encountered during production. ®ber volumetric scaling in a ®lament wound structure was
also discussed in Ref. [14]. Wrestling with this issue
2.1. Fiber strand test program provided motivation for developing the 51 cm ring test
method.
A method for fabricating carbon ®ber strands with As discussed in Refs. [1,13], the 51 cm rings are fabri-
controlled ®ber volume fractions was developed. The strand cated (machined) from a 51 cm ®lament-wound vessel. The
(also called tow) test was used to determine ®ber strain-to- composite vessel is wound over a 51 cm nominal diameter
failure variation as a function of volume fractions. Strands steel mandrel. The mandrel has a 76 cm long cylindrical
with different ®ber volume fractions were fabricated by section that ¯ares into semi-elliptical domes on each end.
passing high-strength intermediate modulus IM7 carbon The vessel can be wound using a multitude of winding
®ber impregnated with 55A epoxy resin through an ori®ce parameters such as: tension, helical layer angle, helical
with a controlled opening diameter. 1 The hole radius, r, was and hoop layers thickness, bandwidths, and winding
varied to achieve the desired ®ber volume fraction as speed. As discussed in Section 3, these parameters are
certain to in¯uence the ®nal part quality. This is particularly
Vf …pr 2 † ˆ nAe …5† true for wet-wound structures in which ®ber motion occurs
where Vf, n, and Ae are the ®ber volume fraction, number of continuously with time. After curing, the end domes are cut
tows (strands) used to fabricate the specimen, and the tow off and the 76 cm long composite cylinder is extracted from
area (mm 2)/end, respectively. Tow specimens were fabri- the mandrel. The composite cylinder is then cut into 20
cated with ®ber volume fractions between 45 and 65%. nominally 0.54 cm wide rings using a diamond-cutting
The specimens fabrication procedure includes: (1) impreg- wheel. The rings are numbered in consecutive order during
nating the tows with a high resin content, (2) extruding the machining, establishing their location along the cylinder
tows through the ori®ce to a set ®ber volume, and (3) wind- length. The rings are then tested in a pressurized ring test
ing under tension on a paddle wheel mandrel. The paddle ®xture as discussed in Ref. [13].
wheel mandrel is designed so that there is no radial contact To assess laminate quality (i.e. ®ber and void volume
associated with a solid mandrel. During the 24 h gellation fractions), three rings are cut and analyzed for ®ber, resin,
and 1208C cure, constant tension was maintained in the tow void volume fractions, and ply thickness using the IA method.
specimens. Once the specimens were cured, they were cut to The IA method consists of examination of the composite
length, end tabbed, and tested in tension to failure using the cross-sectional area to determine the ®ber, void, resin volume
fractions, and ply thicknesses on a ply-by-ply basis. The IA
1
All ®lament wound vessels data reported in this paper were fabricated data are compared to the computer model's predictions.
from IM7 carbon ®ber and 55A epoxy resin. The computer model utilizes cylinder-manufacturing data in
416 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

particular. A more detailed discussion of the IA method is 2. C2-type (biased hoop plies): (^22.52/904)2/(^22.52/
provided in Ref. [13] and Appendix A. 903)//(^22.52/902)/(^22.52/90)3/^22.52
Several 51 cm cylinders were fabricated following the
design of experiment (DOE) test protocols [15,16]. The The dispersed hoop plies' lay-up follows typical pressure
ring test method was used in these experiments to investi- vessel design in which the hoop layers of an approximately
gate the in¯uence of wet-winding parameters on composite equal thickness are dispersed between helical layers. The
vessel quality and strength/stiffness response [1]. DOE is an dispersed hoop plies help to compact the helical layers
experimental method in which the main effects of a variable that have low radial compaction force because of the
(factor) and its interaction can be measured. In the ®rst DOE acute winding angle. The biased lay-up has a large percent
study, ®ve variables were selected for study using a 1/4 of the hoop plies closer to the inside diameter (ID) of the
fractional factorial design. The ®ve variables selected for vessel. The motivation for this lay-up type is to increase the
study were: (1) winding tension, (2) laminate stacking damage tolerance of the case by removing a large percen-
sequence, (3) winding-tension gradient, (4) winding time tage of the hoop plies away from the outside diameter (OD)
between layers, and (5) cut versus uncut helicals. The surface that can be subjected to impact events. Fiber volume
®ve-variable 1/4 fractional factorial test matrix and the fraction was selected as one of the variables because of its
settings for each variable are given in Table 1. A total of potential to improve vessel strength performance as noted in
nine cylinders were built and tested. The ninth cylinder the ®rst DOE. Data from the ®rst DOE indicate that ®ber
(designated as 2B) was a duplication of the second cylinder strength is maximized for vessels with near 62% ®ber
and was needed to establish cylinder-to-cylinder variability. volume in the hoop layers. The ®ber sizing/precursor vari-
Detailed discussion of the experimental setup and variables able was selected to address the issue of new ®ber sizing (GP)
selection procedure is discussed in Ref. [1]. and a new IM7 carbon ®ber's precursor from a domestic
Based on the results from the ®rst DOE study, a second supplier. The ®ber sizing/precursor was treated as a system
DOE study was conducted. This DOE focused on evaluating (i.e. no attempt was made to separate the effect of sizing or
processing, material, and design variables with potential precursor). The second DOE's 51 cm cylinders (referred to as
weight reduction payoff for strap-on composite boosters. C-type) were wound with a full-scale vessel membrane thick-
The study also utilized the subscale, 51 cm pressurized ness of approximately 1.97 cm The winding tension used to
ring test article. Three variables were investigated in a 1/2 wind the ®rst two cylinders was selected to be proportional to
fractional factorial DOE. These were: (1) the hoop plies' the full-scale booster diameter [5]. This resulted in a 1±8
lay-up sequence, (2) ®ber volume in the hoop plies, and (3) winding tension scale factor between the full-scale and
®ber sizing/®ber precursor. Cylinders from this DOE group subscale 51 cm vessel. Material from each cylinder was used
are referred to as C-type vessels. The two types of lay-up in: pressurized ring tests to determine strength and stiffness,
sequences used for the C-type cylinder are: split ring tests to measure residual stresses, ®nal dimensions
examination to determine case shrinkage, and IA to deter-
1. C1-type (dispersed hoop plies): (^22.52/902)8/^22.52 mine ®ber and void volumes and ply thicknesses.

Table 1
Design and manufacturing variables test matrix for the ®rst doe

Cylinder Stacking Cut versus Winding band Winding tension Winding time
number sequence uncut helicals tensions gradient between layers

1 Biased Cut Low No Long


2 Biased Uncut High Yes Short
3 Dispersed Uncut Low No Short
4 Dispersed Cut High Yes Long
5 Biased Cut High No Short
6 Dispersed Cut Low Yes Short
7 Biased Uncut Low Yes Long
8 Dispersed Uncut High No Long
X Biased hoop layers lay-up ˆ (^16.4/904)/(^16.4)3/(90/^102)/(90/^10)
X Dispersed hoop layers lay-up ˆ (^16.4/90)4/(^10/90)2/(^10)
X Cut versus uncut helicals Ð All helicals are cut on the cut option; none on the uncut option
X Low-band tension ˆ 142 N for 16.4 helicals and hoops
X Low-band tension ˆ 94.7 N for 10 helicals
X High-band tension ˆ 320 N for 16.4 helicals and hoops
X High-band tension ˆ 214 N for 10 helicals
X High-band tension gradient ˆ 320 N, 256 N, and 192 N Ð applied to the ®rst three hoops
X Low-band tension gradient ˆ 142 N, 114 N, and 85.4 N Ð applied to the ®rst three hoops
X Short winding time ˆ 2 h for hoops and 3 h for helicals
X Long winding time ˆ 9 h for both hoops and helicals
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 417

Data from another set of three subscale 51 cm vessels

Design and manufacturing parameters for thick-wall cylinders (Lay-up types. CD-type: (^22.52/902)8/^22.52; CB-type: (^22.52/904)2/(^22.52/903)//(^22.52/902)/(^22.52/90)3/^22.52; A-type: [(^152/904)5/
(referred to as A-type) are used to supplement the C-type
cylinder data. A total of three A-type cylinders, designated

tension (N)
A-1, A-2, and A-3, were built. The A-type cylinder has a

64.5
64.5

64.5
Band width Band

93.4
laminate thickness similar to the C-type cylinders with

160

145

145
nominal laminate thickness of 2.03 cm. In this respect, the
C-type and A-type cylinders are very similar and are
grouped together for this study.
13.97
13.97
13.97
13.97
18.80
18.80
18.80
(mm)
The nominal design and manufacturing parameters used
to fabricate the seven, thick-wall, A-type and C-type 51 cm
volume (%) thickness (mm)

cylinders are summarized in Table 2. This group of cylin-


ders is referred to as thick-wall vessels because of their
relatively small R/t ratio of approximately 10. In contrast,
0.417
0.417
0.417
0.417
0.292
0.307
0.292
Ply

the ®rst DOE's cylinders have a R/t ratio of approximately

Band for all vessels, except for the hoop layers of the A-type vessel, was formed with 12 tows. The band for the A-type vessel was formed with 7 tows.
41. As the table demonstrates, vessels A-1 and A-3 were
Design nominal helical plies

fabricated using the identical design/manufacturing para-


Fiber

meters (except for the ®ber sizing). The A-3 cylinder was
52
52
52
52
55
52
55

fabricated with R type ®ber sizing. The A-2 vessel was


fabricated with almost identical parameters to the A-3
content (% wt)

vessel (except for the winding tension). This vessel was


wound with near full-scale to subscale hoop layers band
Resin

tension scaling. The actual full-scale to subscale radii


38.5
38.5
38.5
38.5
36.0
38.5
36.0

scale is 1±6. Because of winding machine limitations, the


actual hoop band tension scaling was approximately 1±4 (or
tension (N)

0.26). The water-emulsi®ed R-sizing has a lower friction


coef®cient (0.26) as compared to the solvent-emulsi®ed
64.5
64.5

64.5

54.7
Band width a Band

320

236

236

W-sizing, which has a 0.34 coef®cient of friction. In


contrast, the water-emulsi®ed GP-size was reported to
have 0.36±0.39 coef®cient of friction. All of the above
sizings are epoxy based and were formulated to maximize
19.56
19.56
19.56
19.56
12.45
12.45
12.45
(mm)

®ber handling during usage.


Ply thickness

2.3. Full-scale vessel test program


0.279
0.248
0.279
0.279
0.257
0.259
0.257
volume (%) (mm)

Full-scale test data reported in this paper come from


hydroburst tests that took place during the development
Design nominal hoop plies

program of various strap-on composite booster programs


for space applications [5]. Three types of full-scale vessels
Fiber

55
62
55
55
55
55
55

are discussed in this paper. The full-scale vessels types are:


A-type (same laminate layup as the sub-scale A-type 51 cm
Resin content

vessel), B2-type, and B3-type. The B2-type vessel has the


(% by wt)

same layup as the ®rst DOE with dispersed hoop layers


36 ^ 1
29 ^ 1
36 ^ 1
36 ^ 1
36 ^ 2
36 ^ 2
36 ^ 2

laminate lay-up (Table 1). The B3-type vessel is similar to


the B2-vessel except that each hoop layer has a double-ply
instead of a single-ply (as in the B2-type vessel). These
Lay-up

vessels represent a range of manufacturing conditions.


type

CD

CD
CD
CB

A
A
A

All full-scale vessel data reported here are for vessels that
were fabricated by the wet-winding method using Hexcel's
Domestic
Domestic

Domestic
Domestic
precursor

Foreign
Foreign

Foreign

IM7 R-sized domestic precursor carbon ®ber and 55A


Fiber

epoxy resin. The vessels are pressurized to failure with


water. During the test, the average hoop strain and pressure
Fiber
(^152/905)3 ^ 152])

are recorder digitally. On all vessels, multiple hoop strain


size

GP
GP
W
W

W
R
R

gages are used. Following the hydroburst test, material from


designation

the cylinder section of each vessel is removed and analyzed


Table 2

by the IA method for ®ber volume, void volume, and


Vessel

ply-thickness on a ply-by-ply basis through the thickness.


A-1
A-2
A-3
C-1
C-2
C-3
C-4

a
418 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Multiple IA samples are removed from each vessel. The as a speci®c process in which ®ber bed compaction domi-
data reported are for average values (for at least six speci- nates. For processing conditions in which the ¯ow time tf is
mens) from this analysis. Note, full-scale vessel data are less than the winding time tw for a layer, the resin will have
included to demonstrate that the proposed strain-to-failure ample time to ¯ow through the compacting ®ber bed. For
criterion holds even for full-scale cylinders wound under the wet winding conditions and materials currently con-
varying conditions. sidered, the ratio tf/tw is typically 0.01. Thus, before the
resin cures, the ®ber motion within the bed is modeled
by compaction of a bed of ®bers subjected to an exter-
3. Computer model for prediction of ®ber volume nal pressure. Once the resin matrix gels in a layer, the
fraction layer stiffness is modi®ed to include the matrix stiffness.
In this formulation, the cylinder/mandrel assembly is
In conjunction with these experiments, an elasticity- modeled as an elastic solid. As each new layer is wound,
based model was developed to describe the physical an external pressure load is applied to the outside of the
phenomena that occur during ®lament winding. Much previously wound layer. This pressure load is treated as a
research has been done regarding ®lament winding (e.g. boundary condition. The pressure boundary condition is
Refs. [17±20]). In general, wet ®lament winding consists applied incrementally, Dp, as each layer is wound to account
of three major processes: (1) tow impregnation, (2) winding for the nonlinear stiffness. Once the layer stiffness and
the tows, and (3) curing the cylinder. Most ®lament winding boundary conditions have been de®ned, the resulting dis-
process models consider: the thermochemical and rheologi- placement of each layer is calculated. A brief summary of
cal behavior of the resin, the motion of the ®ber sheet as it the relevant equations follow; a more detailed derivation
moves through the uncured resin, and the stress/strain devel- can be found in Ref. [12].
oped during cure. The model developed incorporates several The layer displacement in the jth layer uj ( j ˆ 1¼k)
key features from Lee and Springer [19] and Cai and when an external load Dp is applied to a k layered cylinder
Gutowski [20] particularly in the thermochemical, rheology, is [19]:
and stress±strain modeling. Detailed discussion of the For l ˆ 1
model is provided in Ref. [12]. The following sections high-  j
bj C
light some aspects of the model. Du j ˆ a j1 r 1 1 1 c j1 C26 2 36 2 C22 r 2
Fiber motion can be broadly divided into the winding r 2C33
phase and the curing phase of the ®lament winding process.  j
C
In the winding phase, the tension in the layer being wound 1 d j1 C12 2 13 r In r …6†
2C33
applies compaction pressure to the layers beneath, con-
solidating these previously wound layers. In the oven-curing For l ± 1
phase, temperature changes in the mandrel and composite,  j
as well as chemical changes in the resin may cause the
jb j1 C36
Du j ˆ a j1 r l 1 1 c j
1 C26 2 2 C22 r
2

mandrel and the composite to expand or contract. For rl


j
2C33
thermoset resins that go through a low viscosity phase  j
j C13
during cure (e.g. prepreg resins), these expansions or 1 d 1 C12 2 2 C22 r …7†
contractions can cause radial displacement of the ®bers C33
[19]. Unlike many other wet winding resins, 55A resin where
reaches a gelation within 48 h prior to oven cure [21], v
u j
and, therefore, ®ber position will be unchanged during uC
high-temperature cure. lj ˆ t 22j …8†
C33
There is a trade-off between resin ¯owing through the
®ber bed and the compaction of the bed itself [22] when The constants a 1j , b 1j , c 1j , and d 1j are evaluated by applying
modeling ®ber motion. As a new layer is wound, an external continuity at the interfaces and the external pressure bound-
pressure is applied, creating a pressure gradient in the ary conditions. Cij are the material off axis stiffness coef®-
previously wound layers. The external pressure load is cients, and are functions of the material engineering
shared between the ®ber and the resin. The pressure gradient constant [25]: the tensile modulus E, shear modulus G,
in the resin can cause resin to ¯ow through the porous ®ber and Poisson ratio n . In evaluating these constants, we
bed. This mechanism is modeled by Darcy's ¯ow. In the consider the modulus as a function of both the ®ber volume
second mechanism, the load on the ®ber bed causes the bed fraction and the matrix viscosity as described in Ref. [12].
to compact similarly to a nonlinear spring. Researchers Perhaps the most critical aspect in accurately predicting
[19,20,22±24] have studied both of these mechanisms the ®ber volume fraction lies in tracking the resin bleed
extensively and identi®ed winding conditions in which through from previously wound layers [12]. In the winding
one or both ®ber motion mechanisms dominate. process, resin from one layer can ¯ow to an adjacent layer.
For example, Cai and Gutowski [20] identi®ed wet winding For long winding times, the viscosity of the resin in the
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 419

layers already wound is quite high as compared to the new


layer. Uncured resin from previously wound layers can ¯ow
into the newly wound layer and the effective viscosity can
be calculated by a rule of mixtures [26].
In the mth layer, given the viscosities m j of its con-
stituents and the relative concentration of each w w(m), the
viscosity m can be calculated as follows:
X
m
mˆ w…m†
j mj …9†
jˆ1

w w(m) is de®ned as the volume of resin vrj from the jth ply into
the mth layer ( j , m) divided by the total volume of resin
Vr(m) in that layer
vr j
w…m†
j ˆ …10†
Vr…m†
These volumes, vrj and Vr(m), are readily calculated once
the resin velocity at the layer boundaries are known. In this
mixing formulation, an averaging rule of mixtures was
selected because the mixing behavior was viewed as being Fig. 2. Strand specimen ®ber volume versus ori®ce opening.
sequential (as opposed to parallel). Reid et al. [27] cite both
sequential and parallel mixing rules as acceptable. strength increases with an increase in the ®ber volume up
Once the change in displacements Du associated with the to approximately 52% ®ber volume. Beyond 55% ®ber
pressure increment Dp have been determined (Eqs. (6) and volume fraction, the data show a decrease in strain-to-
(7)), the radial positions of the layers in place are updated failure. Based on the subscale vessel data, discussed in
Section 4.2, this decrease in strength beyond 55% ®ber
r jo ˆ r jo 1 Du j ; r ji ˆ r ji 1 Du j21 …11† volume is more a function of the tow specimen fabrication
rather than a `true' effect caused by ®ber volume. This state-
where roj and rij are the outer and inner radii of the jth layer
ment is made based on manufacturing records that show
respectively and j is evaluated from j ˆ 1¼k. At each load
that, beyond the 55% ®ber volume, signi®cant force was
or time increment, the corresponding ®ber volume fraction
required to pass the tow through the ori®ce. The increase
v fj is calculated
in force along with an increase in ®ber volume, may have
v jf ˆ h jf =…r jo 2 r ji † …12† damaged the tow, causing a decrease in strength. There is no
way to verify this hypothesis beyond conducting additional
where h fj is the ®ber sheet thickness in the jth layer. This tests in which different specimen geometries and ®ber
thickness remains constant throughout the manufacture of volume control methods are tried.
the cylinder. To support the argument that the observed decrease in
strength beyond the 55% ®ber is not `truly' caused by
4. Test results

4.1. Fiber strand test results

Strand specimens with varying ®ber volume were


produced using the method discussed previously. The speci-
mens ®ber volume was determined using the acid digestion
method. Fig. 2 depicts the ®ber volume variation in the
strand specimens versus ori®ce opening. Using this method,
strand specimens with ®ber volume fraction varying
between 35% and 65% were produced. A minimum of 50
specimens were fabricated and tested in tension to failure
for each of the ®ber volume groups. The specimens were
instrumented with an extensometer, and the failure load and
strain were recorded for each specimen.
The average ®ber strain-to-failure versus ®ber volume Fig. 3. Tow specimens average strain to failure versus ®ber volume in the
test data is plotted in Fig. 3. The plot shows that ®ber tow.
420 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Fig. 4. Fiber strain translation (FST) versus ®ber volume; regression models are for tow specimens and thick-wall vessel data.

®ber volume, the tow data are plotted (Fig. 4) together with method by which vessel FST factor is calculated is also
case strength versus ®ber volume data from the A-type and discussed.
C-type cylinders. These data are discussed in Section 4.2. The vessel failure pressure analysis is conducted as
The data in Fig. 4 are plotted in terms of the ®ber strain follows: (a) using an elasticity program with ®ber and
translation (FST) factor, rather than the ®ber strain to failure. resin material nonlinearities each ply strains are calculated
The ®ber strain translation factor is de®ned as: as outlined in Ref. [13], (b) using the regression model for
FST versus ®ber volume in Fig. 4, the translation factor for
Fiber strain translation factor each individual ply is calculated, (c) using the average ®ber
lot acceptance strain to failure (as recorded by the ®ber
Maximum fiber strain measured …tow=case†
ˆ manufacturer) together with the FST factor, the strain to
Average fiber lot acceptance strain to failure failure for each individual hoop ply is calculated, (d) the
pressure is continuously increased (in the elasticity analysis)
The ®ber strain translation factor normalizes the data to a
until ®rst hoop ply failure occurs, and, ®nally, (e) the
common basis that allows comparison of strength data for
program terminates and prints the failure pressure, ®ber
vessels that are wound with different ®ber lots and strengths.
strain at failure, hoop ply at which failure initiated, and
The regression line for the tow strength data (D) is calcu-
the FST factor. The analysis requires a ply-by-ply ®ber
lated for tow specimens with up to 52% ®ber volume. Tow
volume and ply thickness. These type of data can come
specimens (7) with higher than 52% ®ber volume are not
from two sources: (1) IA of the failed material and/or cut
included in the regression line calculation. The regression
specimens from untested rings as in the case of the ring test
line calculated for the thick-wall C and A-type vessels is
method, and (2) analytical programs, such as those
based on the FST and ®ber volume fraction in the ®rst hoop
discussed previously, that can determine each individual
ply to fail in each individual vessel. The method for identi-
ply thickness and ®ber volume based on the manufacturing
fying the ®rst hoop ply to fail and the vessel's FST factor are
data. Because the wet ®lament winding manufacturing is a
discussed in Section 4.2. It is important to note for now that
very dynamic process, the ®nal plies' ®ber volumes and
the regression lines for the strand and vessel data show
thicknesses are strongly in¯uenced by the manufacturing
similar trends (i.e. slope). A similar trend of increasing
parameters (e.g. winding time, winding tension, winding
®ber strength with ®ber volume for pultruded carbon ®ber
speed) as was demonstrated in Ref. [1] and further discussed
rods (that in essence are like tow specimens) was shown in
in the Section 4.4 of this paper. Because specimens cannot
Ref. [3]. The regression equations for the tow and thick-wall
be obtained from vessels that are fabricated as end items, a
vessels test data are listed in Fig. 4.
method that combines the accurate prediction of the ®ber
volume and ply thickness with determining the vessel's failure
4.2. Vessel failure analysis results pressure can be very useful for engineering applications.
To test the accuracy of the aforementioned method, the
In this section, the method for predicting vessel failure failure pressures of the ®rst DOE cylinders were calculated.
pressure based on the ®ber volume distribution in the hoop A detailed discussion of the ®rst DOE cylinder test results
layers and the ®ber lot ®ber strain to failure is discussed. The takes place in Section 4.3. In this analysis, the individual
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 421

Table 3
Calculated ®rst doe cylinders failure pressure based on regression model from strand data (Fig. 4)

Cylinder Measured average Calculated


number
Failure OD hoop Failure Percent Hoop layer Fiber strain in Fiber volume
pressure (MPa) strain (%) pressure (Mpa) difference to fail ®rst hoop layer (%) in hoop layer

1 20.80 1.461 21.99 5.7 4 1.537 44.6


2 21.21 1.483 22.17 4.6 5 1.584 53.4
3 20.53 1.441 20.97 2.2 6 1.536 44.4
4 20.71 1.457 22.53 8.8 2 1.557 48.3
5 20.94 1.494 22.26 6.3 6 1.560 49.0
6 20.00 1.415 20.39 2.0 4 1.522 41.9
7 20.11 1.425 21.87 8.8 4 1.527 42.8
8 20.71 1.459 23.13 11.7 4 1.556 48.1
2 rep. 21.16 1.484 22.08 4.3 6 1.579 52.4

ply's ®ber volume and thickness was determined by IA link failure). Figs. 5 and 6 depict typical plots of the strain
method on untested rings. Two regression models were distribution versus the ply strength distribution through the
used: one based on the strand strength data, the second laminate of the ®rst DOE (cylinders 3 and 4, respectively).
based on the thick-wall C and A-type cylinders data. The In all cylinders, failure initiated in a hoop layer other than
regression equations for the two data sets are provided in the inner most hoop layer. Both cylinders (No. 3 and 4)
Fig. 4. The results of this analysis are summarized in Tables could have potentially achieved higher burst pressure if ®ber
3 and 4. On average, the tables show that the ®rst model volume through the thickness was more uniform.
(based on the strand data) consistently over-predicted the Because the two regression lines are essentially parallel to
case performance by 6%. The second model predicted the each other, the predicted ®rst hoop ply to fail will be the
burst pressure of the nine cylinders within 2.5% on average. same. As was demonstrated, the predicted failure pressure
The error in the predicted pressure for seven cylinders was will be different. For this reason, the columns (in Table 3)
less than 3.7%. For only two cylinders, 6 and 8, failure listing the ®rst hoop ply to fail and ®ber volume correspond-
pressure was predicted with a larger error (5 and 6%, respec- ing to this ply are not repeated. Rather, the calculated
tively). Although in some pressure vessel applications, outside diameter hoop strain and ®ber strains at the failed
errors larger than 5% may be signi®cant, overall these hoop ply are listed for each cylinder. The OD hoop strain is
results are very good. The tables show that, for most calculated at the actual measured failure pressure. The ratio
cylinders, the failure is predicted to take place in either of these two strains are used to multiply (correct) the calcu-
the 4, 5, and/or 6 hoop plies (only one cylinder, No. 4, is lated hoop ®ber strain in the failed hoop ply. That is, the
predicted to fail in the second hoop ply). As discussed calculated OD hoop strain is matched with the measured OD
previously, the data show that the hoop layer with low-®ber hoop strain at failure, and it is assumed that the internal
volume will dominate the strength of the laminate (i.e. weak hoop strain distribution will change proportionally. Based

Table 4
Calculated ®rst doe cylinders failure pressure based on regression model from thick-wall cylinders data (see Fig. 4)

Cylinder Measured average Calculated


number
Failure OD hoop Failure Percent OD hoop Fiber strain FST c
pressure (MPa) strain (%) pressure (MPa) difference strain (%) a at failed ply (%) b

1 20.80 1.461 20.75 20.2 1.424 1.463 0.831


2 21.21 1.483 21.28 10.3 1.499 1.501 0.853
3 20.53 1.441 19.77 23.7 1.495 1.453 0.825
4 20.71 1.457 21.42 13.4 1.393 1.500 0.852
5 20.94 1.494 21.18 11.2 1.465 1.500 0.852
6 20.00 1.415 18.93 25.3 1.469 1.456 0.828
7 20.11 1.425 20.57 12.3 1.375 1.460 0.829
8 20.71 1.459 21.98 16.2 1.378 1.477 0.839
2 rep. 21.16 1.484 21.14 20.1 1.509 1.491 0.847
a
Calculated at the measured failure pressure.
b
Calculated ®ber strain at failure pressure at the ®rst hoop ply to fail, then multiplied by the ratio measured to calculated OD hoop strain.
c
Calculated by dividing the ®ber strain at failed hoop ply that is divided by the ®ber lot strain-to-failure (note: all cylinders were fabricated from the same
®ber lot with 1.76% average strain to failure as reported by the ®ber manufacturer).
422 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Fig. 5. Hoop ®ber strain, ®ber strain to failure, and ®ber volume fraction distribution through the thickness of DOE cylinder No. 3.

on Fig. 1, where the strain distribution at failure pressure is valid for only 55A resin type. However, the phenomenon
identical in shape at lower pressure P1 (but incrementally of increased ®ber strain to failure with ®ber volume should
displaced by D), this correction method is reasonable. Once hold regardless of the resin type, except that the regression
the ®ber strain-to-failure is determined, the FST factor is equation may be different than that reported in Fig 4.
calculated by dividing this strain by the average ®ber lot
acceptance strain to failure. 4.3. Subscale 51-cm vessel test results
At this point, it is appropriate to note that the ®ber lot
strain-to-failure used in this type of analysis must be based Tests of the ®rst DOE's nine cylinders, reported in Ref. [1],
on conformance. This is not a trivial point because each showed that there was a strong correlation between average
®ber manufacturer uses its own test method that consists burst pressure and ®ber volume in the composite. The
of a different specimen fabrication, resin type, gripping results from this experiment (also reported in Tables 3
type, and test method. Some manufacturers follow the and 4) are plotted in Fig. 7, Chart A (referred to as thin-
ASTM standard, however, the standard is applied to what- wall cylinders). The reported, average measured pressure
ever resin they have found provides high and consistent ®ber and hoop strain (in Tables 3 and 4) for each cylinder are
strain to failure. Others may use the single ®lament test based on at least 10 rings tested to failure. In this ®gure, the
method that has very little bearing on how the ®ber will ®ber volume fraction in the failed hoop layer (last column
behave in the composite. The results reported here are in Table 3) for each cylinder is plotted against the FST

Fig. 6. Hoop ®ber strain, ®ber strain to failure, and ®ber volume fraction distribution through the thickness of the ®rst DOE cylinder No. 4.
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 423

Fig. 7. Composite vessel strength versus minimum hoop layers ®ber volume fraction.

calculated for the same cylinder (last column in Table 4). ended up with high levels of porosity and ®ber volume in the
This ®gure shows that the ®ber strain translation increases hoop layers well below the nominal (intended) design
with increased ®ber volume fraction in the failed hoop layer. values (Table 2). The high porosity could have been caused
Note that all of the ®rst DOE cylinders were fabricated from by low winding tension. Another factor contributing to the
the same ®ber lot. As was shown in Ref. [1], ®ber volume ineffectiveness in controlling the vessel composite quality
distribution in the ®lament wound composite was strongly was the inability of the wet winding ®ber impregnation
affected by winding tension and winding time, with high system to decouple the ®ber impregnation operation from
winding tension producing higher ®ber volume. Similarly, the material application process. That is, the tension during
shorter winding time produced composite with higher ®ber impregnation was the same tension during application.
volume as compared to longer winding times. The mechan- Following a review of the data (Table 3), the last two cylin-
ism by which these variables produce higher ®ber volume is ders (in the C-type group) were wound using IM7GP/
related to the ®ber motion through the resin. First by tension domestic precursor 2 with the following parameters:
and second by lower resin viscosity. The increase in ®ber
volume due to tension or winding time is well predicted by 1. Vessel C-3 be wound using the same lay-up as vessel C-1
the computer model developed. Both parameters (high with dispersed hoop plies and high winding tension
tension and lower viscosity) produce greater ®ber motion 2. Vessel C-4 be wound using identical setup to vessel C-1
and, therefore, larger ®ber compaction, which leads to
higher ®ber volume.
The second set of thick-wall subscale 51 cm vessels were Using the above manufacturing parameters, the essential
fabricated using various ®ber sizes, winding tension, resin elements of the original material evaluation were preserved.
content (hoop layers), and lay-up. As noted previously, both The third vessel was used to show that a composite with
the A-type and C-type cylinders are considered thick-wall higher ®ber volume and low voids can be fabricated and that
because their R/t ratio is near 10. Initially, the four C-type such a vessel delivers high-®ber strength. The fourth vessel
cylinders were intended to be part of a DOE setup. As part was used as a direct comparison between IM7W/foreign
of this experiment, the ®rst two 51 cm vessels (Table 2) precursor and IM7GP/domestic precursor. It is not believed
were wound with scaled winding tension (64.5 N/band). that the carbon ®ber precursor has any effect on the ®nal
The scaled tension in the subscale 51 cm diameter vessel composite quality. However, ®ber sizing can have signi®-
corresponded to 1:8 ratio of the full-to-sub-scale winding cant effect on the ®nal case quality both in terms of the ®nal
tension (calculated based on the ratio of the two vessels
radii). Following the winding and cure, rings were cut and 2
Note that these changes resulted in the violating the initial design of
analyzed by IA to determine ®ber volume, resin content, and experiment setup. Nevertheless, a very useful data were produced. The
void volume. The IA results for these two vessels are change to a different ®ber size and precursor was required due to program-
summarized in Table 5. The data show that the two vessels matic constrains.
424 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Table 5
Image analysis data summary for C-type vessels 1 and 2

Cylinder number Physical property Hoop layers Helical layers

Design nominal IA measured Percent difference Design nominal IA measured Percent difference

C-1 Resin volume (%) 45 46 2.2 48 44 28


Fiber volume (%) 55 47 215 52 47 210
Void volume (%) N/A 7 N/A N/A 8 N/A
Ply thickness (mm) 0.279 0.335 20 0.833 0.899 8
C-2 Fiber volume (%) 62 44 229 52 47 10
Void volume (%) N/A 10 N/A N/A 8 N/A
Ply thickness (mm) N/A N/A 0.833 0.889 7

®ber and void volume fractions. Additional data for vessels 4.4. Fiber volume model results
that were fabricated with R-size ®ber are also presented for
the A-type cylinders. The ®ber sizing issue is addressed in In order for the aforementioned vessel failure pressure
both Sections 3 and 4 of this paper. (i.e. strength) calculation method to be useful, one needs
The pressurized ring test results for the seven thick-wall to be able to accurately calculate the ®ber volume and ply
cylinders are plotted in Fig. 7, Chart B (labeled as thick-wall thickness distribution through the vessel thickness. In this
cylinders). These data are again plotted in terms of the FST section, the results obtained by the computer model
factor versus ®ber volume fraction in the predicted hoop discussed previously are presented. These results are then
layer to fail. For all cylinders the predicted ®rst ply to fail used to predict the burst pressure of two full-scale cases.
was the ®rst or second hoop ply from the vessel's ID. The An early version of the wet winding computer model,
data in Fig. 7, Chart B, show the close correlation between tswind was written in fortran [12]. In tswind, solutions
the cylinder average FST factor and ®ber volume in the for the ®ber motion were found by following exact analytic
failed hoop ply for all seven cylinders. This trend holds expressions for the displacements of each layer. Also in
for vessel types, three ®ber size types, and two ®ber pre- tswind, the temperature, cure and viscosity for each layer
cursor types. The ®gure includes two additional data points during winding and cure were found by following a ®nite
from 91.4 cm diameter A-type and C-type cylinders that difference formulation. The entire winding model was
were tested in a closed vessel arrangement (as opposed to recently run on a commercial ®nite element code (abaqus).
the uniaxial ring tests for the other cylinders). It appears that Variables unique to wet winding (®ber volume fraction,
the biaxial state of stress does not effect the observed trend. viscosity and cure) as well as the unique nonlinear stiffness,
Unlike the trend for the ®rst DOE cylinders, the increase in cure kinetics, rheometry, and mixing relations have been
FST factor is very signi®cant (<10% between the low and included in user-de®ned subroutines. Both models perform
the high ends). similarly, in terms of prediction of key quality measures.

Fig. 8. Comparison of model predictions and ®ber volume fraction data for Fig. 9. Comparison of model predictions and ®ber volume fraction data for
®rst DOE cylinder 1. ®rst DOE cylinder 7.
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 425

Fig. 12. Fiber volume distribution through the full-scale B-type vessel
thickness; abaqus process model versus measured values by the IA
method.
Fig. 10. Comparison of model predictions and ®ber volume fraction data for
the ®rst DOE cylinder 8.
in predicting these low-®ber volume fraction layers. A
comparison of results for cylinders 1 and 8 shows the effects
However, the commercial ®nite element formulation is of low vs. high band tensions. The low-band tensions in
more easily adapted to new material models and part cylinder 1 result in lower overall ®ber volume fractions.
geometry. The model validation data presented herein In contrast, cylinder 8 shows alternating layers with high
represents results from both solution approaches. and low ®ber volume fraction, particularly at the outermost
Comparisons of model predictions and ®ber volume frac- layers of the cylinder. Again, the importance of including
tion data for several of the ®rst DOE (subscale) cylinders are the resin viscosity-mixing behavior is indicated. When the
shown in Figs. 8±10. In general, there is good agreement band tension is high, a signi®cant volume of higher viscosity
between model predictions and the data. For each cylinder, resin mixes with the adjacent layer. When the winding time
the model predicts that there will be regions in the cylinder is long, this resin will reach gelation, causing a low-®ber
with lower ®ber volume fraction. Cylinders 1 and 7 show volume fraction layer. The model accurately predicts this
low ®ber volume fractions at layer 10. In both cylinders, the trend. Additional examples of the validity of the ®ber
winding time for layer 10 is long. The high viscosity resin motion model for full-scale cylinders with various ®ber
from the previous layers bleeds through the cylinder, creat- sizes may be found in Ref. [12].
ing local layers with low-®ber volume fraction. It was Figs. 11 and 12 show a comparison between model
shown in Ref. [12] that both the viscosity/cure effects on predictions of ®ber volume fraction and experimental data
®ber bed stiffness and the resin mixing algorithm are critical for two full-scale cylinders. For the B2-type vessel, the low-
®ber volume fraction corresponds to a long delay time in
winding at the ninth layer. The predictions for ®ber volume
fraction for the A-type cylinder are reasonably good. The A-
type vessel-manufacturing time is much longer and has
many more layers than the other model validation runs
presented. The model predicts the cyclic variation in the
®ber volume fraction. However, there are several layers of
groups for which the model over predicts the ®ber volume
fraction. The particularly low-®ber volume fraction
predicted for the ®nal two layers is associated with the
accumulation of resin that bled through previously wound
layers. Improvements to the model are underway, including
additional veri®cation of the resin rheology characterization
and mixing algorithm. Changes in these areas will have a
signi®cant effect on ®ber volume fraction prediction, parti-
cularly for vessels with many layers and long winding times
Fig. 11. Fiber volumes distribution through the full-scale A-type vessel such as the A-type vessel. Additional examples of the valid-
thickness; abaqus process model versus measured values by the IA ity of the ®ber motion model for full-scale cylinders with
method. various ®ber sizes may be found in Ref. [12].
426 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Table 6
Predicted failure pressure for full-scale composite vessels

Vessel Measured failure Predicted based on abaqus model Predicted based on actual IA data
type pressure (MPa)
Pressure Percent Failed Pressure Percent Failed
(MPa) error hoop ply (MPa) error hoop ply

A-type 11.74 11.52 21.9 8th 11.49 22.1 8th


13.09 12.64 23.4 18th N/A N/A N/A
B-type 10.85 10.86 10.1 4th 10.76 20.8 5th
16.20 15.45 24.6 11th 16.23 10.2 11th

4.5. Full-scale failure pressure analysis results These ®ber data are part of the material certi®cation
received from the ®ber manufacturer. In cases were multiple
The ®nal part of this paper uses the aforementioned ®ber doffs and/or lots are used to wind the same ply, the
method for calculating composite pressure vessel strength lowest average strain to failure is used in the analysis.
based on the ®ber volume distribution through the thickness. Finally, a somewhat different regression equation, than the
The method is used to predict full-scale composite case one given in Fig. 4, was used in the analysis. The regression
failure pressure. Two approaches to predict vessel failure equation that was used incorporates additional vessel data
pressure are utilized. One method is based on the ®ber that were not presented in Fig. 4. The following regression
volumes and ply thicknesses determined by the abaqus equation was used (both FST and Vf are in percents):
process model. The second method is based on ply-
by-ply ®ber volumes and ply thicknesses from the FST ˆ 35:7 1 0:8781V f :
actual IA data. The process model analyses uses actual
The analysis results are summarized in Table 6. The table
processing data (e.g. winding time, delay time, resin
lists the vessel type, measured failure pressure, predicted
content, winding tension) for the individual cases that are
failure pressure based on the data from abaqus process
considered.
model, and calculated failure pressure based on data from
Two types of full-scale cases are analyzed: one case has
the IA conducted post failure. For each of the analysis cases,
the A-type laminate lay-up and is referred to as the A-type
we also include the hoop ply through the thickness where
full-scale vessel, the second case type is referred to as B-
failure is predicted to happen ®rst. There are two B-type
type and has laminate lay-up and thickness similar to the
vessels as indicated by the different diameters. The two
®rst DOE cylinder's laminate. A comparison between the
vessels have different failure pressure by design. Table 6
abaqus process model and the IA ®ber volume data through
shows that the calculated failure pressure based on the IA
the thicknesses of A-type and B-type vessels is plotted in
data is very accurate. This accuracy is an indication of
Figs. 12 and 12, respectively. The ®ber volume distribution
the accuracy of the ®ber strength versus ®ber volume
determined by the process model agrees reasonably well
model. The errors between the predicted failure pressure
with the measured values. Note that the thick A-type lami-
by the abaqus model versus the measured failure pressure
nate, with multiple hoop plies in each layer, shows signi®-
are somewhat higher but still in good agreement. It should
cant variation in ®ber volume through the thickness. This is
be noted that because the vessel strength cannot be tested
a typical variation observed in a wet-wound case where
nondestructively, the combined process model and ®ber
signi®cant ®ber motion (and therefore resin bleeding)
volume versus strength models are promising approaches
occur during the winding. The individual ply's IA data are
to determine ®nal case strength based on the processing
represented by circle and/or square symbols. The circles
condition. This is of particular importance in pressure
represent helical plies and the squares identify the hoop
vessel of processing anomalies (higher than expected
plies. The process model (and IA) also calculates the indi-
winding time, higher than expected resin content, etc.)
vidual ply thicknesses needed for the ply-by-ply case failure
where a decision must be made if the case is suitable for
analysis (discussed in Section 4.2).
end-usage.
Case failure pressure predictions are as follows: The ply-
by-ply ®ber volumes and ply thicknesses from the IA and/or
abaqus process models are used in conjunction with the 5. Conclusions
elasticity program as discussed in Section 4.2. The failure
analysis uses the ®ber strength data used in each ply are This paper presents additional data to support the
used. That is, in the winding of full-scale case, multiple observed relationship between ®ber volume fraction and
doff and/or lots of ®ber may be used. This ®ber doffs and/ strength in graphite-epoxy composite pressure vessels.
or lots may have different average ®ber strain to failure. More speci®cally, it demonstrates the relationship between
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 427

the ®ber volume fraction in the hoop layers and the pressure Appendix A. Composite ®ber volume IA method
vessel failure pressure (or ultimate strain-to-failure). The
ultimate strain-to-failure increases as the ®ber volume Composite samples ®ber, resin, void volume fractions,
fraction of the hoop plies increases. Data for multiple and ply thicknesses through the laminate thickness were
vessels that were fabricated by the wet winding process determined using an automated IA method. The IA method
under varying conditions were presented. Some of the consists of examination of the composite cross-sectional
data come from design of experiment runs in which the area to determine the ®ber, resin, and void volume fractions.
winding parameters, such as winding tension, winding Inherent to this method is the assumption that the ®ber,
times, and lay-up were varied systematically to produce resin, and void cross-sectional areas are uniform along the
composite vessels with varying ®ber volume fractions length (perpendicular to the cross section). Because
throughout the thickness. These data, and additional vessel the ®bers have nearly uniform cross-sectional area along
data, demonstrate that there is a strong correlation between the length, the method is more accurate in determining
the ®ber volume fraction in the hoop layers of these vessels ®ber volume fraction. However, the void cross-sectional
and the failure pressure and/or strain-to-failure. area along the length is not necessarily (and probably not)
An attempt to demonstrate this relationship in single- uniform along the length, and therefore, the method is much
epoxy impregnated strands is also reported. Graphite- less accurate in determining void volume faction. The resin
epoxy strands with varying levels of ®ber volume frac- volume fraction is determined by subtracting the ®ber and
tion were fabricated and tested in tension to failure. The void volume fractions from one.
test data show increase in ®ber strain-to-failure of up to In the IA method, composite samples are mounted in
52% ®ber volume fraction beyond which the strain epoxy plugs and the cross-section surface is polished
decreased. The initial increasing strength trend followed (Fig. A1). The polished surface or cross-section area is
the vessel's increasing strength. The vessel's strength then observed under a re¯ected light microscope and the
increases up to 65% ®ber volume fraction. Data for image is magni®ed 1263X. The analysis region consists of
higher ®ber volume fraction in the vessel are not avail- rectangular area of 0.125 cm (0.049 in.) by 0.16 cm
able. Beyond the 52% ®ber volume fraction, the strand test (0.063 in.). The relative size of the frame to the ply thick-
results are not valid. This may be due to the specimen ness is shown in Fig. A2. Typically, one measurement
geometry and/or ®ber damage at the higher ®ber volume (reported number) consists of 100 such frames along the
fractions. ply length with one to two frames across the ply thickness
A computer model was developed to simulate the ®ber (depending on ply thickness). The images are acquired auto-
motion during the wet-winding process. One of the objec- matically using the Leco Auto Image Analysis System
tives for developing this model was to predict the ®ber (Fig. A3). The system consists of an Olympus BX60M
volume fractions in the wet-wound graphite-epoxy vessels. microscope equipped with 1.25X to 100X objectives. The
The model uses the actual winding parameters, such as microscope has a Hitachi KP-M1 black and white CCD
winding tension, winding time, initial resin content, resin camera which is interfaced with the computer system.
thermo-kinetics, and resin viscosity. The model results are There is a Prior auto positioning stage that is also connected
compared on a ply-by-ply basis to actual full-scale vessel to the computer system that moves the sample during
data obtained by the IA method. A reasonably good agree- analysis. The voids and ®bers (which are analyzed at different
ment between the two is shown. times) appear as gray regions in the projected image
Finally, the ®ber strength versus ®ber volume relationship (Fig. A4). For each frame, the computer captures the
obtained from subscale tests together with the process image and converts it to a gray scale image. The system
model predictions are used to calculate full-scale vessel
failure pressure. These predictions are conducted prior to
the actual tests. In addition, posttest comparisons are
conducted using IA data from the failed vessels. The calcu-
lated failure pressures based on the IA data are accurate to
within 2% error of the actual failure pressures. These
results support the accuracy of the ®ber strength ®ber
volume fraction relationship. The predicted failure pres-
sures based on the process model are accurate to within
5% error. Although this error is larger, it is still reason-
able for engineering analysis. This is particularly signif-
icant because the vessel strength cannot be tested
nondestructively. Therefore, combining the process model
and ®ber volume-versus-strength models to determine ®nal
vessel strength based on the processing conditions is a
promising approach. Fig. A1. Image analysis composite sample.
428 D. Cohen et al. / Composites: Part B 32 (2001) 413±429

Fig. A4. Magni®ed projected IA.

Fig. A2. Image analysis frame size relative to ply thickness.


can distinguish more than 100 different shades of gray). The
computer assigns a false color to each shade of gray and the
percent area is calculated based on the number of pixels for
each color for each frame. The resin volume is calculated
based on the remaining area fraction.
Typically six to eight samples are analyzed from each
vessel. The samples are taken at different axial and circum-
ferencial locations. Due to the nature of oven cured ®lament
wound material, there is considerable variation from sample
to sample and from frame to frame.
Typical frame-to-frame and sample-to-sample varia-
bility is summarized in Table A1. It should be noted
that the hoop layers frame-to-frame variation for the
most part is below 5%. This is considered to be an
excellent result. The frame-to-frame variation within
the helical layers is signi®cantly higher. This variation
is in¯uenced by the helical cross over where there are
resin rich areas that tend to have low ®ber volume.
Fig. A3. Automated IA system. However, it should be noted that the between sample

Table A1
Typical frame-to-frame and sample-to-sample variability IA results

Layer 0 deg. location 180 deg. Location Combined

Identi®cation Average S.D. C.V. (%) Average S.D. C.V. (%) Average S.D. C.V.

Helical #1 61.3 11.9 19.4 59.1 11.5 19.5 60.2 1.6 2.6%
Helical #1 62.5 10.6 17.0 60.4 12.7 21.1 61.5 1.5 2.4%
Hoop 70.8 3.7 5.3 69 3.2 4.7 69.9 1.3 1.8%
Hoop 68.7 2.7 4.0 66.6 2.0 3.0 67.7 1.5 2.2%
Helical #2 59.9 10.2 17.0 66.6 7.4 11.1 63.3 4.7 7.5%
Helical #2 61.1 8.1 13.3 61.7 11.3 18.4 61.4 0.4 0.7%
Hoop 67.7 2.2 3.2 63.3 4.2 6.7 65.5 3.1 4.8%
Hoop 66.1 2.7 4.1 65.7 1.6 2.5 65.9 0.3 0.4%
Helical #3 62.7 6.1 9.8 59.4 7.7 13.0 61.1 2.3 3.8%
Helical #3 57.7 12.9 22.4 58.2 7.5 13.0 58.0 0.4 0.6%
Hoop 65.4 3.8 5.8 65.5 3.0 4.5 65.5 0.1 0.1%
Hoop 66.8 2.4 3.6 65.4 3.5 5.4 66.1 1.0 1.5%
Helical #4 68.0 11.5 16.9 63.4 6.6 10.3 65.7 3.3 5.0%
Helical #4 68.2 5.6 8.2 63.8 7.0 11.0 66.0 3.1 4.7%
D. Cohen et al. / Composites: Part B 32 (2001) 413±429 429

variation is relatively low, validating the accuracy of the [11] Sun L, Mantell SC, Banerjee A, Cohen D. Resin Flow and Compac-
analysis method. In addition, because the aforementioned tion in Filament Wound Cylinder. 1995.
[12] Banerjee A, Sun L, Mantell SC, Cohen D. Model and experimental
strength relationship with ®ber volume depends on the hoop
study of ®ber motion in wet ®lament winding. Compos A±Appl Sci
layers, for which the analysis is more accurate, the results Manufac 1997. p. 251±630.
should be less dependent on the variability in the ®ber [13] Cohen D, Toombes YT, Johnson AK, Hansen MF. Pressurized ring
volume measurements. test for composite pressure vessel hoop strength and stiffness evalua-
tion. J Compos Technol Res 1995;17(4):331±40.
[14] Cohen D. Application of reliability and ®ber probabilistic strength
References distribution concepts to composite vessel burst strength design.
J Compos Mater 1992;26(13):1984±2014.
[1] Cohen D. In¯uence of ®lament winding parameters on composite [15] Box GEP, Hunter WG, Hunter JS. Statistics for experimenters: an
vessel quality and strength. Compos Part A±Appl Sci Manufac introduction to design, data analysis, and model building. New
1997. p. 1035±47. York: Wiley, 1978.
[2] Chamis CC, Hanson MP, Sera®ni TT. Criteria for selecting resin [16] Montgomery DC. Design and analysis of experiments. New York:
matrices for improved composite strength. In: 28th Annual technical Wiley, 1984.
Conference, Reinforced Plastics/Composites Institute, The Society of [17] Cirino M. Axisymmetric and Cylindrically Orthotropic Analysis of
the Plastics Industry, Inc., 1973. Filament Winding, Newark, Delaware: University of Delaware, 1989.
[3] Chen-Chi M, Goang DY, Han LJ, Hsieh KH. Processibility and prop- [18] Calius EP, Springer GS. A model of ®lament wound thin cylinders. Int
erties of pultruded ®ber reinforced polyurethane/furfuryl alcohol J Solids Struct 1990;26:271±97.
interpenetrating polymer network composites (II). In: 39th Inter- [19] Lee SY, Springer GS. Filament winding cylinders: I. Process model.
national SAMPE Symposium, SAMPE, 1994. J Compos Mater 1990;24:1270±98.
[4] Yushanov SP, Joshi SP. Stochastic Modeling of Fracture Processes in [20] Cai Z, Gutowski T, Allen S. Winding and consolidation analysis for
Fiber Reinforced Composites, vol. 33, No. 9 (September), 1995. cylindrical composite structures. J Compos Mater 1992;26(9):1374±
p. 1689±97. 99.
[5] Riddle RA, Beckwith SW. Development of test and analysis methods [21] Bhi ST, Hansen RS, Wilson BA, Calius EP, Springer GS. Degree of
for thick-wall, graphite/epoxy ®lament wound composite materials cure and viscosity of hercules HBRF-55 resin. In: Advanced Material
fracture toughness. In: Composite Materials: Testing and Design Technology '87, Society for the Advancement of Materials and
(Seventh Conference). Philadelphia: American Society for Testing Process Engineering, 1987.
and Materials, Philadelphia, 1984. [22] Smith GD, Poursartip A. A comparison of two resin ¯ow models for
[6] Shih GC, Ebert LJ. Theoretical modeling of the effect of the inter- laminate processing. J Compos Mater 1993;27(17):1695±711.
facial shear strength of the longitudinal tensile strength of uni- [23] Gutowski T, Dillon G. The elastic deformation of lubricated carbon
directional composites. J Compos Mater 1987;21:207±24 (March). ®ber bundles: comparison of theory and experiments. J Compos Mater
[7] Tsai SW, Hahn HT. Introduction to composite materials. Westport, 1992;26:2330±47.
CT: Technomic Publishing, 1980. [24] Dave R, Kardos JL, Dudukovic MP. A model for resin ¯ow during
[8] Rosen BW. Tensile failure of ®brous composites. AIAA Journal composite processing: Part 1. General mathematical development.
1964;2:1985±91. Polym Compos 1987;8(1):29±38.
[9] Zweden C, Rosen BW. A statistical theory of material strength with [25] Tsai S. Composites Design. 3rd ed. Dayton: Think Composites, 1987.
application to composite materials. J Mech Phys Solids 1970;18:180± [26] Sun L. Process model for ®lament winding of thermosetting matrix
206. composite cylinders, University of Minnesota, 1995.
[10] Sun L, Mantell SC, et al. Thermoset ®lament winding process model [27] Reid RC, Prausnitz JM, Sherwood TK. The properties of gases and
and parametric study, International mechanical engineering congress liquids. New York: McGraw-Hill, 1968.
and exposition. San Francisco, CA: American Society of Mechanical
Engineers, 1995.

You might also like