Composite Structures 246 (2020) 112432

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Prediction of void growth and fiber volume fraction based on filament T

winding process mechanics
⁎
Qi Wanga, Tong Lia, Bo Wanga,b, Changzhi Liuc, Qizhong Huangd, Mingfa Rena,b,
a
Department of Engineering Mechanics, Dalian University of Technology, Dalian 116024, China
b
State Key Laboratory of Structural Analysis for Industrial Equipment, Dalian University of Technology, Dalian 116024, China
c
Beijing Institute of Astronautical Systems Engineering, Beijing 100076, China
d
State Key Laboratory of Advanced Fiber Composite, Beijing Composite Materials Co., Ltd, Beijing, 102101, China

A R T I C LE I N FO A B S T R A C T

Keywords: The void growth and fiber volume fraction in filament winding composites are critical to the mechanical per-
Curing formance of composite structures. In this paper, a comprehensive prediction model for void growth and fiber
Finite element analysis (FEA) volume fraction is developed based on various time-dependent manufacturing characteristics in the filament
Void growth winding process. This prediction model is basically decoupled into three sub-models: a diffusion-controlled void
Filament winding
growth sub-model, a thermo-chemical sub-model and a resin flow sub-model for fiber volume fraction modeling.
The relationships between manufacturing parameters and void size, fiber volume fraction are investigated by
employing this new model. The results show that, in conventional models, the predicted size of voids was
underestimated without considering the change of resin pressure and processing temperature. Strict control of
the initial void size has limited benefits on reducing the final size of voids after winding process. However, the
ambient humidity is critical to the control of the final void size in the composite products. Moreover, a higher
winding tension will result in a smaller size of voids and a higher volume fraction of fibers, benefiting the
improvement of product quality.

1. Introduction The growth of voids is one of the most critical defects in the fabrication
of the FRP [6,7]. In recent years, the void defect has been reported as a
Filament winding is one of the most commonly used forming tech- detrimental threat to the strength and fatigue life of FRPs. Furthermore,
niques for the fabrication of continuous fiber reinforced resin compo- the voids will cause further damage to the products, such as cracks and
sites, such as pressure vessels, pipes and so on [1–3]. Comparing to delamination [8–12]. According to the literature [10], an approxi-
other forming techniques for fiber reinforced plastics (FRP), filament mately 20% decrease in shear-, flexural- and tensile strength was
winding has the advantages of high cost-effectiveness and easy auto- caused with an increase of void content by only 2.6%. According to the
mation, making it being widely employed in engineering fields. Among American aeronautics standard, final FRP products with more than 2%
various winding methods, the wet winding is more frequently used for void content should be abandoned [13]. Therefore, the porosity of FRPs
manufacturing thermoset FRP because of the advantages of low mate- is one of the key parameters of FRP products. Winding manufacturing
rial cost and easy modulation of the resin formulation with respect to techniques should be improved to reduce the content of voids for high
specific requirements [4]. A typical configuration of the wet winding quality production of FRPs as critical force-bearing components in ad-
process is shown in Fig. 1, in which the fibers are passed through a resin vanced equipment.
bath and wound on a rotating mandrel in a prescribed geometric path In addition, the fiber volume fraction of filament wound parts also
under controlled tension, followed by the curing and demolding steps to have a significant impact on the performance of FRPs [14]. When fiber
get the final winding FRP products [5]. fraction in the FRP rises from 50% to 65%, the strength of composites
In the wet winding process, the fiber bands are wound on the was improved by over 10% [4]. The content of voids and volume
mandrel after impregnated immediately. However, there is still a lot of fraction of fibers have significant dependency on the process para-
impregnant residing in the resin, and the voids will be easily generated meters during the fabrication, an insightful understanding of the pro-
during the curing step. Fig. 2 shows a typical void in winding FRP part. cess of void growth and evolution of fiber volume fraction related to the

⁎
Corresponding author at: Department of Engineering Mechanics, Dalian University of Technology, Dalian 116024, China.
E-mail address: renmf@dlut.edu.cn (M. Ren).

https://doi.org/10.1016/j.compstruct.2020.112432
Received 2 March 2020; Accepted 28 April 2020
Available online 04 May 2020
0263-8223/ © 2020 Elsevier Ltd. All rights reserved.
Q. Wang, et al. Composite Structures 246 (2020) 112432

regional temperatures even exceed the curing condition [23–26]. At the

same time, the uneven temperature distribution will affect the resin
flow process, which further changes the growth of voids and the ar-
rangement of fibers in the FRP. Therefore, the filament winding process
is a comprehensive multi-physical process, which includes many factors
such as temperature, viscosity and pressure. Lee and Springer [27]
developed a sequential compaction model in which the fiber motion
was described as sequential compaction of the fibers. The permeability
of the winding parts is assumed to be constant and the applied pressure
is assumed to be entirely borne by the resin. Dave et al. [28] and Gu-
towski et al. [29] proposed that compaction is not sequential and the
applied pressure is shared by both the fiber bed and the resin, which is
known as squeezed sponge model that is widely accepted [30–34].
Smith and Poursartip [35] compared these models and concluded that
the sequential compaction model is a special case of the squeezed
sponge model. The squeezed sponge model can be reduced to the se-
quential compaction model under certain conditions. Cai et al. [36]
analyzed the change of fiber distribution and fiber bed deformation in
Fig. 1. Schematic of the wet filament winding process. the filament winding process using quasi-elastic model, in which the
fiber stiffness was defined as a function of the fiber volume fraction.
However, their model development and validation for wet winding
were limited to the cases with a constant viscosity. In recent years, Zu
et al. [5,37–39] have carried out a lot of work on the winding path
design and winding tension control technology of the filament winding
process.
The first target of this paper is to investigate the resin flow behavior
of the winding FRP parts, calculate the resin pressure during the
winding molding process while considering the influence of perme-
ability change caused by fiber bed compression. At the same time, the
effect of variation of viscosity on the resin flow is evaluated by calcu-
lating the temperature and degree of cure. Afterwards, the diffusion
model developed by Ledru et al. [22] is further improved by con-
Fig. 2. Micrograph of voids in the filament winding part.
sidering the processing parameters which gives a more accurate pre-
diction of void size. Finally, the influence of processing parameters on
processing conditions. the void size and the fiber volume fraction is investigated based on the
For the mechanisms of void growth, various works have been con- above-mentioned comprehensive model.
ducted by Loos and Springer [15,16] and Kardos et al. [17,18]. Kardos
assumed that the growth of voids happens in a pseudo-homogeneous 2. Model
medium because experimental evidence indicates that voids do not
grow preferentially perpendicular to the plies. Based on this assump- Filament winding is a complex multi-physical process involving
tion, a diffusion-controlled void growth model was proposed, which is a chemical and physical changes of the materials. In this study, the fila-
widely accepted model for the prediction of void growth [19–21]. This ment winding process is decoupled into three sub-models, including
model describes that when resin pressure or temperature varies, the thermo-chemical sub-model, resin flow sub-model and void growth sub-
water diffuses from the surrounding liquid resin into the voids due to model. Details of the developments for these sub-models and the in-
pure water vapor or air-water vapor mixture, which caused the further ternal connections between these sub-models are provided in this sec-
growth of voids. The diameter of voids is especially sensitive to the tion.
temperature and resin pressure according to this model. Based on the
work of Kardos, Ledru et al. [22] developed a prediction model by 2.1. Thermo-chemical sub-model
coupling a gas-diffusion-controlled void growth model and a visco-
mechanical/diffusion model. In this coupled model, the viscosity and The axial length of the winding FRP parts is much larger than the
polymer crosslinking effects are added to the diffusion model in order radial thickness. Therefore, the direction of heat transfer and resin flow
to obtain a more accurate prediction void size. Using this model, the is assumed to be along the radial direction. The one-dimensional energy
pressure is found to be one of the most critical parameters to control the equation in the cylindrical winding FRP part can be given by the dif-
final size of void. Although the pressure and temperature have been ferential equation of one-dimensional transient heat transfer with an
considered as important parameters affecting the growth of voids in interior heat source, which is:
previous studies, the resin pressure and parts temperature are simply
∂T 1 ∂ ⎛ ∂T ⎞
assumed to be the processing pressure and curing cycle temperature in ρC = kr + Q̇
these researches. However, experiments indicated that the resin pres- ∂t r ∂r ⎝ ∂r ⎠ (1)
sure is not a constant during the manufacturing process. where ρ is the density of the composite, C is the specific heat, T is the
During the filament winding process, the impregnated fiber is temperature, t is the time, r is the radial coordinate, kr is the thermal
wound on the mandrel under certain winding tension. While heated, conductivity, and Q̇ is the heat generating a rate of resin curing. Q̇ is a
the viscosity of resin decreases, making the resin flow under the function of the rate of cure dβ / dt :
winding tension. During the extrusion of resin, the fiber bed is com-
dβ
pacted and becomes denser. As the temperature further increases, the Q̇ = ρr Vr ⎛ ⎞ Hr
exothermic crosslinking reaction will happen in the resin. The reaction ⎝ dt ⎠ (2)
will cause uneven distribution of temperature in the material and some where ρr is the resin density, Vr is the matrix volume fraction, Hr is the

2
Q. Wang, et al. Composite Structures 246 (2020) 112432

total heat of resin reaction and β is the degree of cure. The curing rate is
an empirically derived function of temperature, time and degree of cure
dβ
= K (T )·f (β )
dt (3)
where f (β ) is conversion function that reflects the type of reaction, and
K (T ) is an Arrhenius-type relation for the temperature dependency
given as:
−E
K (T ) = A·exp ⎛ a ⎞
⎝ RT ⎠ (4)
where A is the pre-exponential factor, Ea is the apparent activation
energy of curing reaction and R is the universal gas constant.
In equation (4), the conversion function f (β ) depends on the type of
the thermoset polymer. There are a few different types of curing reac-
tion for thermoset resins, and the nth order expressions can describe the
curing behavior of most types of resin matrix:
f (β ) = (1 − β )n (5)
where n is the reaction order. Fig. 3. Compaction during the filament winding process.

2.2. Resin flow sub-model k + 1th layer is winding in the winding angle of ϕ , then the stress along
the fiber direction is:
The winding FRP part is assumed to be axially symmetric for sim-
T k+1
plification purpose. Under winding tension, the resin will be squeezed σ fk + 1 =
Af (10)
from the inner layer of the winding cylinder to the outer layer. The
resin flow throughout a fiber sheet under tension is a flow with a low where T is the winding tension, Af is the cross-sectional area of the
Reynolds number. Therefore, the resin flow in the filament winding layer. Then, the circumferential component of the stress in the fiber
process can be considered as a unidirectional flow in porous media layer is:
which can be expressed in Darcy’s law:
k+1
σθθ = σ fk + 1sin2 ϕ (11)
K dp
v=−
μ dr (6) The compaction pressure on the kth layer caused by the k + 1th
layer is:
where v is the fluid velocity, K is the permeability of the medium, μ
k+1 k+1
is the fluid viscosity and dp / dr is the pressure gradient in the direction σθθ t
σrrk =
of the flow. r k+1 (12)
The viscosity of resin depends on the degree of cure and the pro-
where t k+1
and r k+1
are the thickness and inner radius of the k + 1th
cessing temperature [40]. Generally, in the curing process of composite
layer, respectively. Assuming the total number of winding layers is n,
materials, the viscosity of the resin will first decrease with the in-
the stress of the kth layer is the sum of the stress of the stresses caused
creasing temperature while the crosslinking reaction does not happen,
by all the layers above the kth layer
obtaining a better fluidity. As the reaction of crosslinking moves on, the
n
viscosity of the resin increases. This process can be characterized by the
σrrk = ∑ σrri
Roller model: i=k+1 (13)
U The external stress is assumed to be borne by both the fibers and the
μ = μ∞ exp ⎛ + aμ β ⎞
⎝ RT ⎠ (7) matrix [28,29]:
where μ∞ and aμ are constants that are independent on the temperature, σ = pf + pr (14)
U is the activation energy of the viscous flow of resin.
Permeability is an important indicator to evaluate the difficulty of where σ is the total applied stress, pf is the fiber pressure and pr is the
resin flow in the fiber bed, which has a significant impact on the resin resin pressure which can be solved by Darcy’s law. The fiber stress is the
flow [41]. As the resin squeezes out, the fiber bed is gradually com- function of fiber volume fraction Vf , gives as:
pacted, the fiber volume fraction is increasing and the permeability Vf
gradually reduces. Gutowski et al. [29] proposed a permeability model ⎛ − 1⎞
V0
for a variety of composite parts: pf = Es ⎝ ⎠
4
⎛ ⎜
Va
− 1⎞
⎟

r f2 (1 − Vf )3 ⎝ Vf
⎠ (15)
k11 =
4kl Vf2 (8) where Es is the constant related to the state of the fiber bundle, V0 and Va
3 are the initial fiber volume fraction and final fiber volume fraction,
r f2 ⎛
⎜
Va
− 1⎞ ⎟ respectively.
Vf
k22 = ⎝ ⎠
4kl ( Va
Vf
+1 ) (9)
2.3. Void growth sub-model

where rf is the radius of the fiber, Vf is the fiber volume fraction, kl and The voids are assumed to be spherical gas bubbles that contain an
kt are constants, and Va is the theoretical limit of the fiber fraction. air-mixture vapor with an initial radius R 0 in resin. The growth of the
The curing pressure in filament winding is mainly provided by the void is controlled by temperature, pressure, surface tension, and dif-
winding tension on the fiber bands. As shown in Fig. 3, when the fusion, as shown in Fig. 4.

3
Q. Wang, et al. Composite Structures 246 (2020) 112432

∂C (C∞ − Csat ) ⎛ Rv ⎞
= 1+⎜ ⎟

∂r r = Rv Rv ⎝ πDt ⎠ (18)

where C∞ is the uniform dissolved gas concentration in the bulk resin,

Csat is the gas concentration near the bubble wall, D is the diffusion
coefficient defined by:

−Ev
D = D0 exp ⎛ ⎞
⎝ RT ⎠ (19)

where D0 and Ev are constants determined by experiment. C∞ and Csat

are given by [22]:

aφbρr
C∞ =
100Wr (20)

b
⎛ ⎞
aρr ⎜ 100χ H2 O p v ⎟
Csat = p H2 O < pHsat2 O
100Wr ⎜ sat Tref Rδ ⎡ Ec ⎛ 1 1 ⎞ ⎤⎟
⎜ p H2 O [ T ] exp ⎢ R ⎜T − ⎟ ⎥⎟
T
⎝ ⎣ ⎝ ref ⎠ ⎦⎠
aρr
Csat = (100)b p H2 O ⩾ pHsat2 O (21)
Fig. 4. Schematic of void growth state. 100Wr

where φ is the humidity ratio, a and b are experimental values, Wr is the

The pressure in the void pv can be calculated by the equilibrium weight fraction of resin, ρr is the resin density, χ H2 O is the water vapor
stability of resin pressure pr and the surface tension γLV [22]: partial pressure, pHsat2 O is the saturated water vapor pressure at the re-
ference temperature Tref , δ and Ec are adjust parameters of the eva-
2γLV
pv = pr + poration enthalpy.
Rv (16)
Based on the three abovementioned sub-models, a comprehensive
model can be developed based on finite element analysis (FEA) on the
The mass evolution rate of bubble water can be obtained from the
platform of ANSYS (ANSYS, Inc. USA) and MATLAB (MathWorks, Inc.
Fick’s first law and the perfect gas law. Therefore, the bubble radius Rv
USA). The whole process of filament winding of FRP is evaluated by a
can be solved by combining them:
sequential coupling method, whose flow chart is shown in Fig. 5.
Firstly, the material parameters and initial conditions are inputted into
∂C 4 π d ⎛ Rv3
4πDRv2 = ⎜ Mgas pv ⎞⎟ the thermo-chemical sub-model, and the evolution of temperature and
∂r 3 R dt ⎝ T (17)
r = Rv ⎠ degree of cure (T(t), β(t)) are obtained by solving the thermo-chemical
∂C
sub-model which are transferred into the resin flow sub-model. By
where Mgas is the molecular weight of gas in the void, ∂r r = Rv
is the gas solving the resin flow sub-model, the evolution of fiber volume fraction
concentration gradient at the bubble interface which is given by Fick’s (Vf (t)) and resin pressure (Pr (t)) are obtained. Subsequently, the evo-
second law [42]: lution of temperature solved by the thermo-chemical sub-model and the
evolution of resin pressure solved by the resin flow sub-model are

Fig. 5. Comprehensive flow chart.

4
Q. Wang, et al. Composite Structures 246 (2020) 112432

passed to the void growth sub-model. Finally, the void size versus time
(Rv (t)) is obtained.

3. Experiments and model validation

Forming experiments of FRP cylinders were carried out to validate

the comprehensive prediction model in this section. The material of the
cylinders is T700 graphite fibers reinforced TDE-86 epoxy resin. The
diameter of the cylinders is 150 mm. The fiber bandwidth is 6 mm. The
cylinders were wound under the winding tension of 180 N. After
winding, the cylinders were put into an autoclave for curing. The curing
cycle includes a 60 min dwell at 120 °C and a 120 min dwell at 200 °C,
before cooling down to room temperature.

3.1. Differential scanning calorimeter measurements

TDE-86 is a high-performance epoxy resin with a wide range of

applications. The curing kinetics of TDE-86 was studied by non-iso-
Fig. 7. − ln(τ / Tp2) -1/ Tp linear fitting.
thermal differential scanning calorimetry (DSC) technique from 30 to
250 °C at different heating rates (5, 10, 15 K/min). The heat flow data
dβ
were obtained using the area under the peak of exotherm as a function = 1.25 × 105 exp( −82984/ RT )(1 − β )0.951
dt (24)
of temperature and time. The data were processed further to establish
the cure kinetic model of TDE-86. Fig. 6 shows the DSC thermogram.
The curing kinetic parameters can be solved by the Kissinger 3.2. Micro morphology of the FRP
method using the peak temperatures of the exothermic peak at different
heating rates. Kissinger equation is: To observe the void morphology, the FRP cylinder was cut into
several rings along the axial direction, and their sections were observed
τ AR ⎞ E 1
ln ⎛⎜ 2 ⎞⎟ = ln ⎛ ⎜ ⎟− a· by digital optical microscope (AOSVI AO-3M100). The void micro-
T
⎝ ⎠p ⎝ Ea ⎠ R Tp (22) graphs is shown in Fig. 8 and voids are colored in red. From the figures,
the void diameters were up to about 500 μm . In addition, it was ob-
where τ is the heating rate (K/min), Tp is the peak temperature (K), A is
served that there are almost no voids in the inner layers, but more void
the pre-exponential factor (min), Ea is the apparent activation energy
in the outer layers. This distribution of voids is caused by the pressure
(kJ/mol).
gradient of the inner and outer layers.
Peak temperatures at different heating rates were obtained from
The fiber volume fractions of the FRP cylinders were obtained by
DSC data. Ea can be obtained by linear fitting of − ln(τ / Tp2) with 1/ Tp .
image processing. The cylinders were sectioned into several slides along
The result is shown in Fig. 7. The slope of the line is 9.977 and the
the radial direction. The polished sections were observed by scanning
intercept is −13.527. Therefore, Ea can be calculated as 82.984 kJ/
electron microscope (SEM, FEI Quanta 200). Micrographs at the center
mol, ln(A) can be calculated as 15.827. The reaction order n can be
of the cross-section in the specimens (as shown in Fig. 9a) were pre-
obtained by the Crane equation [43]:
processed using the image processing program ImageJ [44]. The com-
d(ln τ ) E ponent material region is segmented by the trainable weka segmenta-
≈− a
d(1/ Tp ) nR (23) tion algorithm to obtain binarized images. The typical binarized image
is shown in Fig. 9b, where the black area represents the fiber and the
The reaction order n is obtained as 0.951, which is similar to the white represents the matrix. The fiber volume fraction of 69.1% can be
first-order reaction and is consistent with the single peak of the DSC calculated by the averaging area ratio of the binarized images.
curve (Fig. 6). Therefore, the curing kinetics model of TDE-86 can be
described as:
3.3. Model validation

Assuming that the winding FRP parts are axisymmetric, a two-di-

mensional axisymmetric finite element model was developed by taking
a microelement on the radial section of the winding cylinder. The

Fig. 6. DSC curves of TDE-86 at different heating rates. Fig. 8. Micrographs of void morphology.

5
Q. Wang, et al. Composite Structures 246 (2020) 112432

Table 1
Input parameters for resin flow sub-model.
Parameters μ∞ aμ U rf kl kt Va As

Unit Pa∙s – J∙mol−1 m – – – Pa

Value 18.8 1.06 × 10−6 3.76 × 10 4 3.5 × 10−6 0.7 0.2 0.8 2072.9

ended. The degree of cure approaches 1 in the second isothermal

platform. The evolution of resin pressure and the fiber volume fraction
during the winding process is shown in Fig. 11b. As illustrated, the resin
pressure is highest at the beginning, after the resin is squeezed out, the
fiber volume fraction increases and the resin pressure gradually de-
creases. At the end, the fiber volume fraction is saturated on 66.9%
which is close to the experimental characterization, which validates the
accuracy of the sub-models of thermo-chemical and resin flow.
After solving the first two sub-models, the results of temperature
and resin pressure are transferred to the sub-model of void growth
(MATLAB, ODE45 function). The results comparison of void growth
prediction between Ledru’s model [22] and the current model is shown
in Fig. 12. It can be found that the results of the models are almost
identical in the first 150 min. Due to the high resin pressure and low
temperature at the initial stage, the void size decreased on the basis of
the initial void size. After 150 min, in the prediction of void size from
the current model, with the decrease of resin pressure and the increase
of temperature, the void growth conditions are satisfied and the void
keeps growing. InLedru’s model, however, the void did not reach the
critical conditions until about 190 min when the temperature is high
enough. It should be noted that, the curing process is almost completed
at about 300 min and the state of void growth should be fixed by
300 min. Therefore, the final void radius predicted by the current
model and the Ledru’s model is 240.8 μm and 65.7 μm, respectively. It
can be seen that the predicted result of the current model is much larger
than Ledru’s model.
Based on the morphology micrographs of the voids, TrackMate
plugin of ImageJ [44] was used to quantify the number of voids with
Fig. 9. (a) Scanning electron micrographs of specimens section center, (b) ty- different equivalent radius according to the size of void pixels
pical binarized image.
(Fig. 13a). The void number ratio versus the void sizes is shown in
Fig. 13b. Although the number of voids decreases with the increase of
void size, the volume of larger void accounts for a considerable pro-
portion of the total void volume, as shown in Fig. 13c. Among the voids,
the equivalent radius of 220–250 μm have the largest contribution to
the overall porosity, accounting for 39%. As a result, larger void sizes
pose a greater threat to the properties of composite structures. These
voids are defined as hazardous voids. Accurately estimation of the ha-
zardous void size is of great significance to the quality evaluation of the
filament winding process. According to the theoretical model, the resin
flow behavior has a great impact on void growth, and the size and
volume fraction of voids will be greatly underestimated without con-
sidering the decrease of resin pressure that is caused by the resin flow.
In summary, the results of the fiber volume fraction and void size of
Fig. 10. The finite element model of the winding cylinder.
the winding FRP cylinders obtained by the experiments and numerical
predictions are compared in Fig. 14, providing validation of our com-
thermo-chemical and resin flow analysis was carried out by FEA prehensive prediction model.
(ANSYS). The finite element model is shown in Fig. 10, there are twenty
elements per layer using 4 nodes plane element (PLANE55). 4. Results and discussion
The input parameters for the thermo-chemical sub-model are ob-
tained from experiments (Section 3.1), while the parameters of the resin Based on the established comprehensive model, the influence of
flow sub-model and void growth sub-model are obtained from the lit- processing parameters on the growth of voids and the evolution of fiber
erature [21,35,36]. The values of these parameters are listed in Tables 1 volume fraction was studied under different working conditions, in-
and 2, respectively. cluding initial void size, relative humidity and compacting pressure.
Numerical prediction results of the sample center temperature and
degree of cure are shown in Fig. 11a. Because the thickness of the part is 4.1. Initial void size
small, the temperature change is consistent with the curing tempera-
ture. The curing reaction began when the temperature reaches the first First of all, the evolutions of void growth with different initial void
isothermal platform and accelerated when the first isothermal platform sizes were compared when other processing parameters are the same.

6
Q. Wang, et al. Composite Structures 246 (2020) 112432

Table 2
Input parameters values for void growth sub-model.
Parameters γLV D0 Ev φ a b PHsat2 O Wr ρr δ Ec

Unit Pa∙m m2∙s−1 J∙mol−1 % – – Pa – kg∙m−3 J∙K∙mol−1 J∙mol−1

Value 0.05 1.7 × 10−6 5.33 × 10 4 60 2 × 10−4 1.7 3169 0.4 1280 43.893 57073

As shown in Fig. 15, when the initial void radiuses of parts are 10 μm,
30 μm, and 50 μm the final void sizes are almost identical. This is be-
cause the initial void size has little effect on the critical condition of
void growth within a certain range. Under the same process conditions,
the growth conditions of voids with different initial sizes are close to
each other. Therefore, before the critical condition of void growth is
reached, the voids with larger initial diameter will shrink to a certain
extent. In the void growth stage, the void growth rate of the larger
initial size is slightly lower, resulting in almost the same final void size.
That means, in the final production of winding FRP parts, it is not
significant to improve the product quality by strictly controlling the
initial void size.

4.2. Humidity

The void growth under different conditions of relative humidity are

studied while keeping the other processing parameters the same, and
the results are shown in Fig. 16. When the relative humidity is 50%,
70%, and 90%, the final void radiuses are about 200 μm, 283 μm and
370 μm, respectively. This means the growth of voids is sensitive to the
humidity of the environment. This is because when the ambient hu-
midity is high, the concentration of the water vapor gas dissolved in the
bulk resin is high, resulting in a higher void growth rate. Therefore,
reducing the relative humidity of the manufacturing environment can
effectively reduce the void size of the products, which is positive to
improve the product quality in the FRP fabrication.

4.3. Winding tension

Then, the influence of winding tension is also investigated here.

Three winding tension of 180 N, 270 N and 360 N are applied to the
winding parts to investigate the variation of void size and fiber volume
fraction under different levels of winding tension. The prediction of
void growth is shown in Fig. 17a. The final void radius versus winding
Fig. 11. Prediction results of thermo-chemical (a) and resin flow sub-models tensions is shown in Fig. 17b. The results indicate that the final void
(b). size in the winding FRP parts will decrease under a high winding ten-
sion, but this effect will be saturated as the pressure increases. Similar
conclusions are drawn on the prediction of fiber volume fraction, as
shown in Fig. 18. A higher winding tension results in a higher fiber
volume fraction. However, the fiber volume fraction will not infinitely
increase because the fiber bed cannot be further compressed after a
certain degree, and the fiber volume fraction basically converges when
the tension reaches 600 N.
In conclusion, the initial void size of the prepregs was found to have
a little influence on the final void size, while relative humidity and
winding tension have a great effect on the evolution of void size.
Therefore, in the production of winding FRP parts, it is recommended to
reduce the ambient humidity in the workshop. Meantime, increase the
winding tension in a certain range and moderately relax the initial void
size to balance the quality of products and the cost of production.

5. Conclusions

In this paper, a comprehensive model is developed to predict the

void growth and fiber volume fraction in filament winding products.
The change of resin pressure and the curing temperature was con-
Fig. 12. Void growth prediction results of the original model and the current sidered by carried out a filament winding process mechanics analysis
model. with time-dependent parameters. Filament winding cylinders were

7
Q. Wang, et al. Composite Structures 246 (2020) 112432

Fig. 14. The comparison of the experiment and prediction results.

Fig. 15. The void growth evolution with different initial void sizes.

Fig. 16. The void growth evolution under different relative humidity.
Fig. 13. Statistics of void size in specimens, (a) quantifying of the voids number
with different equivalent radius. (b) Void number ratio versus void equivalent
dangerous estimation of process quality. Based on the current model,
radius, (c) the proportion of different sizes of voids in the overall porosity.
strict control of the initial void size has little impact on the final void
size control. At the same time, the relative humidity in the filament
formed to validate this prediction model. winding process has a great influence on the final void size, reducing
The prediction model is validated by the microstructure of the ambient humidity can effectively reduce the void size in the products. A
manufactured filament winding parts. The original diffusion-based higher winding tension will also result in a smaller void size and a
model without considering the change of process parameters will higher fiber volume fraction.
greatly underestimate the final void size in the product, resulting in a The prediction model developed in this work can be used to guide

8
Q. Wang, et al. Composite Structures 246 (2020) 112432

Fig.17. (a) The void growth evolution under different winding tension, (b) the
final void radius versus winding tension. Fig. 18. (a) The fiber volume fraction evolution under different winding ten-
sion, (b) the final fiber volume fraction versus winding tension.

the manufacture of filament winding products. At the same time, this

study can also provide inputs for the subsequent research on the pre- Acknowledgments
diction of the mechanical properties of composite structures with
manufacturing defects. This work was supported by the National Natural Science
Foundation of China (Nos. U1837204, 11802053, 11602030 and
11825202).
6. Data availability
Appendix A. Supplementary data
The raw and processed data required to reproduce these findings are
available to download from: https://doi.org/10.17632/b7y24n6dxz.2. Supplementary data to this article can be found online at https://
doi.org/10.1016/j.compstruct.2020.112432.

Conflict of interest References

We wish to confirm that there are no known no known competing [1] Wang R, Jiao W, Liu W, Yang F, He X. Slippage coefficient measurement for non-
financial interests or personal relationships that could have appeared to geodesic filament-winding process. Compos Part A Appl Sci Manuf 2011;42:303–9.
https://doi.org/10.1016/j.compositesa.2010.12.002.
influence the work reported in this paper.
[2] Vasiliev VV, Krikanov AA, Razin AF. New generation of filament-wound composite
pressure vessels for commercial applications. Compos Struct 2003;62:449–59.
https://doi.org/10.1016/j.compstruct.2003.09.019.
CRediT authorship contribution statement [3] Munro M. Review of manufacturing of fiber composite components by filament
winding. Polym Compos 1988;9:352–9. https://doi.org/10.1002/pc.750090508.
[4] Zhao L, Mantell SC, Cohen D, McPeak R. Finite element modeling of the filament
Qi Wang: Conceptualization, Methodology, Software, Validation, winding process. Compos Struct 2001;52:499–510. https://doi.org/10.1016/
Writing - original draft, Formal analysis. Tong Li: Validation, Writing - S0263-8223(01)00039-3.
review & editing, Visualization. Bo Wang: Validation, Supervision, [5] Lei Z, Hui X, Bing Z, Li D, Zi B. Design of filament-wound composite structures with
arch-shaped cross sections considering fiber tension simulation. Compos Struct
Funding acquisition. Changzhi Liu: Formal analysis. Qizhong Huang: 2018;194:119–25.
Resources. Mingfa Ren: Writing - review & editing, Investigation, [6] Chang X, Ren MF, Li T, Guo X. Evaluation of mechanical behaviour of unidirectional
Funding acquisition. fibre-reinforced composites considering the void morphology. J Reinf Plast Compos

9
Q. Wang, et al. Composite Structures 246 (2020) 112432

2017;36:1817–28. https://doi.org/10.1177/0731684417727358. 2016;80:171–81.

[7] Guo Z-S, Liu L, Zhang B-M, Du S. Critical void content for thermoset composite [25] Park HC, Goo NS, Min KJ, Yoon KJ. Three-dimensional cure simulation of com-
laminates. J Compos Mater 2006;43:1775–90. https://doi.org/10.1177/ posite structures by the finite element method. Compos Struct 2003;62:51–7.
0021998306065289. [26] Ren M, Wang Q, Cong J, Chang X. Study of one-dimensional cure simulation ap-
[8] Hamidi YK, Aktas L, Altan MC. Effect of packing on void morphology in resin plicable conditions for thick laminates and its comparison with three-dimensional
transfer molded E-glass/epoxy composites. Polym Compos 2005;26:614–27. simulation. Sci Eng Compos Mater 2018;25:1197–204.
https://doi.org/10.1002/pc.20132. [27] Lee S-Y, Springer GS. Filament winding cylinders: I. Process model. J Compos Mater
[9] Costa ML, Almeida SFM, d., Rezende MC. The influence of porosity on the inter- 1990;24:1270–98.
laminar shear strength of carbon/epoxy and carbon/bismaleimide fabric laminates. [28] Dave R, Kardos JL, Duduković MP. A model for resin flow during composite pro-
Compos Sci Technol 2001;61:2101–8. https://doi.org/10.1016/S0266-3538(01) cessing: Part 1—General mathematical development. Polym Compos 1987;8:29–38.
00157-9. [29] Gutowski TG, Morigaki T, Cai Z. The consolidation of laminate composites. J
[10] Liu L, Zhang B-M, Wang D-F, Wu Z-J. Effects of cure cycles on void content and Compos Mater 1987;21:172–88.
mechanical properties of composite laminates. Compos Struct 2006;73:303–9. [30] Costa VAF, Sousa ACM. Modeling of flow and thermo-kinetics during the cure of
https://doi.org/10.1016/j.compstruct.2005.02.001. thick laminated composites. Int J Therm Sci 2003;42:15–22.
[11] Jeong H. Effects of voids on the mechanical strength and ultrasonic attenuation of [31] Shin DD, Hahn HT. Compaction of thick composites: simulation and experiment.
laminated composites. J Compos Mater 1997;31:276–92. https://doi.org/10.1177/ Polym Compos 2004;25:49–59.
002199839703100303. [32] Hubert P, Vaziri R, Poursartip A. A two-dimensional flow model for the process
[12] Mueller de Almeida SF, Chaves de Mas Santacreu A. Environmental effects in simulation of complex shape composite laminates. Int J Numer Methods Eng
composite laminates with voids. Polym Polym Compos 1995;3:193–204. 1999;44:1–26.
[13] Lee Il W, Springer GS. A model of the manufacturing process of thermoplastic [33] Li M, Gu Y, Zhang Z, Sun Z. A simple method for the measurement of compaction
matrix composites. J Compos Mater 1987;21(11):1017–55. https://doi.org/10. and corresponding transverse permeability of composite prepregs. Polym Compos
1177/002199838702101103. 2007;28:61–70.
[14] Cohen D, Mantell SC, Zhao L. The effect of fiber volume fraction on filament wound [34] Young W-B. Compacting pressure and cure cycle for processing of thick composite
composite pressure vessel strength. Compos Part B Eng 2001;32:413–29. https:// laminates. Compos Sci Technol 1995;54:299–306.
doi.org/10.1016/S1359-8368(01)00009-9. [35] Smith GD, Poursartip A. A comparison of two resin flow models for laminate pro-
[15] Springer GS. A model of the curing process of epoxy matrix composites. Prog Sci cessing. J Compos Mater 1993;27:1695–711.
Eng Compos 1982:23–35. [36] Cai Z, Gutowski T, Allen S. Winding and consolidation analysis for cylindrical
[16] Loos AC, Springer GS. Curing of epoxy matrix composites. J Compos Mater composite structures. J Compos Mater 1992;26:1374–99.
1983;17:135–69. https://doi.org/10.1177/002199838301700204. [37] Zu L, He QX, Shi JP, Li H. Non-geodesic trajectories for filament wound composite
[17] Kardos JL, Duduković MP, Dave R. Void growth and resin transport during pro- truncated conical domes. Appl Mech Mater 2013;281:304–7.
cessing of thermosetting — Matrix composites BT - Epoxy Resins and Composites [38] Lei Z, Koussios S, Beukers A. Design of filament–wound domes based on continuum
IV. In: Dušek K, editor., Berlin, Heidelberg: Springer Berlin Heidelberg; 1986, p. theory and non-geodesic roving trajectories. Compos Part A Appl Sci Manuf
101–23. 2010;41:1312–20.
[18] Kardos JL, Duduković MP, McKague EL, Lehman MW. Void formation and transport [39] Zu L, Koussios S, Beukers A. Design of filament-wound isotensoid pressure vessels
during composite laminate processing: an initial model Framework. Compos. Mater. with unequal polar openings. Compos Struct 2010;92:2307–13. https://doi.org/10.
Qual. Assur. Process. ASTM International; 1983. 1016/j.compstruct.2009.07.013.
[19] Anderson JP, Altan MC. Formation of voids in composite laminates: coupled effect [40] Banerjee A, Sun L, Mantell SC, Cohen D. Model and experimental study of fiber
of moisture content and processing pressure. Polym Compos 2015;36:376–84. motion in wet filament winding. Compos Part A Appl Sci Manuf 1998;29:251–63.
[20] Grunenfelder LK, Nutt SR. Void formation in composite prepregs–effect of dissolved [41] Parnas RS, Howard JG, Luce TL, Advani SG. Permeability characterization. Part 1: a
moisture. Compos Sci Technol 2010;70:2304–9. proposed standard reference fabric for permeability. Polym Compos
[21] Boey FYC, Lye SW. Void reduction in autoclave processing of thermoset composites: 1995;16:429–45.
Part 1: high pressure effects on void reduction. Composites 1992;23:261–5. [42] Wood JR, Bader MG. Void control for polymer-matrix composites (1): theoretical
[22] Ledru Y, Bernhart G, Piquet R, Schmidt F, Michel L. Coupled visco-mechanical and and experimental methods for determining the growth and collapse of gas bubbles.
diffusion void growth modelling during composite curing. Compos Sci Technol Compos Manuf 1994;5:139–47.
2010;70:2139–45. [43] Crane L, Dynes PJ, Kaelble DH. Analysis of curing kinetics in polymer composites. J
[23] Bogetti TA, Gillespie Jr JW. Two-dimensional cure simulation of thick thermoset- Polym Sci Polym Lett Ed 1973;11:533–40.
ting composites. J Compos Mater 1991;25:239–73. [44] Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, et al. Fiji:
[24] Weiland JS, Hartmann MP, Hinterhölzl RM. Cure simulation with resistively in situ an open-source platform for biological-image analysis. Nat Methods 2012;9:676.
heated CFRP molds: implementation and validation. Compos Part A Appl Sci Manuf

10