Sei sulla pagina 1di 10

Eighth South African Conference on Computational and Applied Mechanics

SACAM2012
Johannesburg, South Africa, 35 September 2012
c SACAM
Simulation of the Vacuum Infusion Process
G Stephens & DN Wilke
Dynamic Systems Group, Department of Mechanical and Aeronautical Engineering
University of Pretoria
Pretoria, South Africa
stephensg@gmail.com
AbstractVacuum infusion is a composite manufacturing process
where a vacuum is used to draw resin through a composite
preform. The vacuum infusion process is complicated by changes
in factors as the resin ow progresses through the preform
e.g. preform thickness, permeability, bre compaction, bre
relaxation and the vacuum bag stifness. The numerical solution
strategy presented in this study, is based on a strategy proposed
by Gavignon et. al. . The strategy proposed by Gavignon et.
al. used a xed spatial grid with a superimposed oating node
to track the ow front as the resin ows through the preform.
We simplied this approach by generating the spatial mesh as
the ow progresses. This approach could in future be combined
with an adaptive mesh movement strategy of the existing mesh
(behind the ow front) and the ow front to avoid additional
computational nodes to be added when not required. In this
study we investigate xed spatial parameter to generate the
spatial mesh and show that comparable results to Gavignon
et. al. approach are obtained. In addition validiation against
two experimental data sets are carried out, which shows some
agreement for the lling stage. However, a lack of experimental
agreement for the post-lling stage for both Gavignon et. al. and
our solution strategies most probably indicates that the governing
equations of the underlying mathematical model needs to be
rened.
Keywordsvacuum infusion process; nite element method;
adaptive strategy
I. INTRODUCTION
Vacuum infusion (VI) is a basic but widely used compos-
ite manufacturing process. The vacuum infusion process is
broadly classied into two stages. Firstly, a lling stage during
which new resin is introduced into the preform. Secondly, a
post-lling stage which allows the resin to settle after resin
has been introduced into the preform. The lling stage is
started by drawing a vacuum over a bagged preform (mould).
This creates a pressure difference with respect to atmospheric
pressure. This pressure difference is the driving force that
pushes the resin into the preform or alternatively for the
vacuum to pull the resin through the preform. The result being
a continuously moving ow front that separates dry bres
from wetted (infused) bres. At some stage during the vacuum
infusion process the infusion inlets are closed i.e. no new resin
is added. This marks the onset of the post-lling stage during
which the resin settles and nally cures in the preform.
The aim of the vacuum infusion process is to obtain a
fully infused preform with a near uniform bre volume
fraction over the preform. For complex geometries the success
of the infusion process depends largely on the positioning of
infusion/inlet points and vacuum/outlet point(s). Although the
vacuum infusion process is a basic manufacturing processes,
the modeling thereof is fairly complex due to changes of
various factors over the course of the infusion process such as
thickness, bre compaction, bre relaxation and bag stiffness.
The nal quality and strength of the composite depends largely
on the impregnation quality and nal bre volume fraction.
The former being a physical phenomena dominated on the
microscopic scale while the latter is largely dominated on the
macroscopic scale. This study focuses on the latter and aims
to accurately predict the spatial distribution of the nal bre
volume fraction of a composite component.
Investigations into the modeling of the vacuum infusion
process is by no means new to computational mechanics.
Numerous studies and investigations have been conducted [1],
[2], [14]. These include specic studies related only to resin
properties or bre compaction and also more general studies
that proposed the governing equations that need to be solved in
order to model the ow through a preform. This study focuses
on solving these governing equations using the nite element
method to ultimately obtain the spatial distribution of the bre
volume fraction of a composite component. Specically, we
simplify an existing nite element implementation [1] that
required xed time steps to be taken during the solutions
process on a xed spatial grid. This choice for solving the
system introduced the need for a temporary oating-node to be
superimposed onto the xed grid. We investigate applicability
of generating the spatial grid as the ow front progresses.
We consider three strategies. Firstly, using a xed spatial grid
spacing, secondly a xed temporal grid spacing and spatial
grid spacing based on a constant average velocity of the ow
front. As the spatial mesh is generated on demand the need for
a oating node strategy is alleviated. Although xed spacings
are used in this study, posteriori adaptive approaches to alter
spatial or temporal grid spacings will be investigated in a
future study.
Our numerical strategies are veried against Givignon et.
al. prosed strategy and validated experimentally against two
data sets.
A. Background
Many composite manufacturing technologies exist. One
of the most basic technologies being the vacuum infusion
process. Vacuum infusion (VI) makes use of a solid mould
in which the composite bres are laid out. This preform is
1 c SACAM 2012
then covered by a vacuum bag which acts as a exible
mould. A vacuum is then drawn over the preform which
creates a pressure difference from atmosphere. This pressure
difference is the driving force behind pushing the resin into
the preform. A detailed discussion on VI will follow in
Section I-B. VI has many commercial names which include
vacuum assisted resin transfer moulding (VARTM), resin in-
fusion under exible tooling (RIFT), vacuum moulding (VM),
vacuum bag moulding and SCRIMP (Seeman Composites
Resin Infusion Process).
A popular alternative to VI is RTM (Resin Transfer mould-
ing). RTM uses two solid moulds and additional pressure
to drive the resin through the preform. RTM usually allows
for superior control of the infusion process which ultimately
results in superior quality products. However, RTM usually
results in an increased cost per component for low production
quantities since two solid moulds are required.
The advantages of vacuum infusion are numerous and
include; high bre content (up to 70% (by weight) compared to
hand-layup of 45%), low void content, increased mechanical
strength, fume free, cheaper and complicated structures can
be manufactured . A number of disadvantages also exist for
vacuum infusion moulding. [2] states disadvantages as being
thickness variation due to non-uniform compaction pressure
and limited ability to achieve high bre content. The problem
of certain sections of the mould being un-wetted or only
partially wetted is very prominent, especially in new products
where the process is not fully optimized yet.
At this point in time, many vacuum infusion processes are
set up via trial and error. The resin inlets and vacuum points
are often selected by a person with experience in the eld.
A number of software tools exist to model this ow and thus
most of the guesswork has been eliminated. Most components
that are manufactured using VI are purely cosmetic due to the
variations in part thickness and volume fractions. This results
in a variation of strength over a component. A numerical
model that could accurately predict the spatial bre volume
fraction for the VI process over a component could be used
in conjunction with optimization to improve the uniformity of
the nal bre volume fraction of a component and ultimately
better control over the strength of a component.
The accuracy of numerical model depends on a number of
factors. Each of these is discussed in more detail in section
II A simple model of the ow through the bre layup could
easily be created; however, accurately predicting the process
requires careful consideration of these factors.
[3] states that VI has been around since the 1950s, however
a recent interest due to new environmental, health and safety
restraints has prompted a renewed interest in the process. The
patented Marco method is considered the rst clearly dened
method for manufacturing composites using a vacuum. The
Marco method was initially developed for use in the boat
manufacturing process; however, the Lotus Car Group Ltd
patented a similar method in 1972. Over time, the process has
been renamed and rened which resulted in various patents.
However, the general principle behind the VI process has
remained unchanged.
In 1994, a slightly different method was developed by DSM
Brittles. This method used two bags on top of each other. The
outer bag used a vacuum to apply a compaction pressure to
the bres and the inner bag was used to pull the resin through
the layup. This resulted in the ability to compact the layup by
different amounts. A number of companies have used female
moulds; however, this is dependent on the geometry of the part.
In addition to improvements on the process various patents for
applications to consumable parts used in the VI process has
been granted. These include piping used to deliver resin into
and out of the preform, the vacuum bag, materials to enhance
ow through the layup and parts ensuring infusion at the inlet
of the preform.
B. Detailed description of the Vacuum Infusion process
The Vacuum infusion processes starts with laying the bre
on the mould that will form the part. Unlike process such as
RTM, VI uses a single solid mould instead of two solid moulds
namely a positive and negative mould. This has a signicant
cost saving for low production volumes. Once the bres had
been laid on the mould, a exible vacuum bag is placed over
it. The inlet/s and outlet/s are selected in such a way that
the entire mould will be fully permeated by the resin. They
are often made into a line inlet or outlet by using a highly
permeable substance such as resin distribution tape. The entire
layup is surrounded by tacky tape to ensure a good seal with
the mould. The outlet usually contains a resin trap before the
vacuum pump to avoid any excess resin to enter and damage
the vacuum pump.
The inlet is then closed and the vacuum switched on. This
part of the process will be called pre-lling from hereon.
This is to check for leaks in the system before the mould is
affected by the resin. An effect of this vacuum being applied
to the layup is that it is compressed. This compression is an
important factor that affects the permeability of the system.
The compression during the pre-lling will be called dry
compression from hereon. The compression of the bres is
based on the approach of Gavignon et. al. [1].
Once it has been ensured that no leaks remain the resin can
be introduced into the system to begin the lling stage. The
resin ows through the laminate and is allowed to permeate all
the bres. A number of other processes occur during lling.
There is a change in the compaction pressure as well as a
phenomenon called relaxation whereby some of the compacted
bres relax during this stage. The ow is modelled with
varying degrees of success and complexity by a number of
studies ( [1] and [4] for example) using a combination Darcys
law and the continuity equation. A number of other variations
on this method are used, however the fundamentals behind the
model remain identical.
The second last step of the process, which will be called
post-lling, is when the resin inlet is closed while the vacuum
is still applied. The point at which the resin inlet is closed
depends on many factors. The resin can be closed before
the infusion is complete, to allow the resin in the mould
2 c SACAM 2012
to permeate through the entire mould. Alternatively, to allow
resin to ll the entire mould before closing the inlet to ensure
that sufcient resin is available. Both approaches can be
solved using the same model as shown by [1] which result
in differences in nal thickness as expected. During the post-
lling process there is again a change in the compaction stress
as the excess resin is sucked out of the layup and the nal part
thickness is established.
The nal part of the process is waiting for the resin to
cure (solidify). The vacuum is usually left on the preform to
ensure that the part thickness does not change again, which is
essentially an extension of post lling stage. It must however
be noted that it is essential that the resin curing time is much
longer than the post lling process as the viscosity of the resin
is assumed to be constant throughout the vacuum infusion
process. This can usually be achieved by introducing chemical
additives into the resin.
II. THEORY
A. Permeability
Before the continuity equation can be solved a number of
relations needs to be investigated and decided upon. Firstly,
we will consider permeability. Naturally this is affected by
the type of bres being used as well as their orientation
and the number of layers. A study on factors which affect
the permeability is done in [5] and a number of factors are
characterised that affect the permeability and these are taken
into account in this study. In order to simplify the model as
much as possible line infusion is used in this study.
Numerous different models [5][9] have been developed
including an approach using micro-models [10], however
generally the empirical Carman-Kozeny approach [1], [11],
[12] is used to relate the permeability to the bre volume
fraction. This relationship is given by [1], [11], [12] and others
as
K = C
(1 V
f
)
n+1
V
n
f
. (1)
Here K is the permeability, V
f
is the bre volume fraction
and C and n are experimentally determined constants. This
approach is not perfect as two layers will never be aligned
in exactly the same way however the best result have been
obtained using this approach.
The constants in this equation were determined using a
simple experiment described later in section III.
B. Viscosity
Another important factor to consider was that of the vis-
cosity of the resin. Most infusions are set up such to avoid
the resin starting to cure while the part is still being infused.
Some studies [9], [13] account for changes in the viscosity,
however this study and others [1] use a constant viscosity
model. This choice is further supported by the development
of resins to have a longer pot life and thus more time available
before the resin starts to cure. Conicting results were found
when replacing the resin with a substance of similar viscosity
[5]. However, as stated in [15] oil is used as the viscosity is
constant with respect to time and is readily available thereby
allowing for simple experiments to be conducted.
C. Compaction
The bre compaction is another area that requires additional
research. Various models ( [3], [11], [14] ) exist to allow for
varying bre compaction. [14] related the compaction stress to
the bre volume fraction. It was also found by [1] and others
that this relation is dependant on whether the bres are wet
or dry. The following relation
V
f
= a
B
f
(2)
has been used by [1], [4] and is also used in this study.
Here,
f
is the bre compaction, a and B are constants
that are determined experimentally. These constants depend
on whether the bres are dry or wetted [1]. The compaction
stress that the bre experiences is governed by the pressure
difference between atmospheric and the vacuum pressure

f
= P
atm
P. (3)
D. Continuity
Consider the continuity equation [15],
=
V
V oids
V
Total
, 0 1 (4)
V
f
=
V
Solids
V
Total
, 0 V
f
1, (5)
which gives
V
f
= 1 . (6)
Where V
V oids
is the volume of the voids, V
Total
is the total
volume of the preform, is the porosity and V
f
is the
bre volume fraction. Assuming bres are incompressible and
compaction causes strain only in normal direction, i.e. in the
height direction:

f
h
0
V
f0
= h V
f
(7)
Where V
f0
,
0
and h
0
are the bre volume fraction, dentsity
and height respectively, all under zero compaction stress
(Denoted by the subscript zero). Assuming a constant bre
density,
V
f0
V
f
=
h
h
0
(8)
Equation (8) represents the conservation of bre mass for
the system. The uid mass increases as resin is introduced into
the system and is assumed to be incompressible. The volume
ow rate is dened by
Q
x
(x, t) = u
x
(x, t) h(c, t). (9)
Here u
x
is the volume averaged velocity or Darcy velocity.
Due to the model being developed in 1D the volumetric ow
rate is actually an area ow rate and can be obtained by
3 c SACAM 2012
dividing by the width of the part and thus has the units of
m
2
.s
1
. This means that u
x
has the correct units of m.s
1
.
Based on the incompressibility and mass conservation of the
uid we obtain
Q
x
(x, t) t Q
x
(x + x, t) t = V
Pores
(t), (10)
with V
Pores
(t) the change in porous volume in the preform.
By considering the limits as x 0 and t 0 we get

(u
x
h)
x
=
( h)
t
(11)
Which can be simplied using the above equations to:

u
x
h
x
=
h
t
, (12)
with
h
t
usually ignored. Many steps have been omitted from
the derivation however the full derivation is available from
[15].
Darcys law is then used to relate the volume averaged
velocity to the bre volume fraction in conjunction with the
permeability given in (1) to obtain
u
x
=
K

dP
dx
. (13)
As in [1] we also compute the volume averaged velocity at
the ow front
u
x
=
K
xx

P
x
+
_
x x
c
h
__
K
xx

h
x
P
x

h
t
_
, (14)
with x is the ow front position, x
c
is the centre point about
which a Taylor series expansion is computed. This is used to
estimate the distance the ow front will traverse in a given
time step. Alternatively, to compute the require time step to
ow a given distance. The volume averaged velocity is then
related to the velocity of ow front
v
x
=
u
x

, (15)
which is what is observed experimentally.
The combination of all of these equations results in a system
of 6 equations with 6 unknowns where height is the only
independant eld variable. The model is described as a 1.5
dimensional model as one dimension related to the direction
of ow along the preform is explicitly discretized, whilst the
second half dimension is related to the height which is
calculated.
E. FEM Implementation
The strong form of the governing equation is obtained by
substituting (13) into equation (12)

x
_
K

h
dP
dx
_
=
h
t
. (16)
The use of Darcy intrinsically implies that the uid is New-
tonian and the ow laminar which is used in this study. A
modied Darcys law was considered by [1] with an empirical
type correction factor to aim to improve the experimental
correlation. Applying the weighted residual method on (16)
at elemental level we obtain
I =
_
xi+1
xi
_

x
_
K

h
dP
dx
_

h
t
_
(x)dx = 0. (17)
Applying integration by parts to (17) we nally obtain the
weak form
_
xi+1
xi
__
K

h
dP
dx
_
d
dx
+
h
t
(x)
_
dh = (18)
_

x
_
K

h
dP
dx
_
(x)
_
xi+1
xi
The element interpolation for the 1 dimensional isoparamet-
ric elements are given by
h h
i
N
i
+h
i+1
N
i+1
(19)
with the shape functions given by
N
1
() =
1
2
(20)
and
N
2
() =
1 +
2
. (21)
The inverse of the Jacobian is given by
d
dx
=
2
x
i+1
x
i
. (22)
As this study considers varying element lengths we cannot
simplify the Jacobian as in the case of Gavignon et. al. [1].
Finally from (18) we obtain the Algebraic system of nonlinear
equations that need to be solved
K(h)
el
+C
el
h
t
= F(h)
el
. (23)
Here K(h)
el
is the element stiffness vector, C
el
the element
capacitance matrix and F(h)
el
the element force vector which
is zero as the boundary conditions account for inlets and
outlets. h is the vector containing the element thicknesses
for the preform and unless otherwise shown is the current
iteration, i.e. h
J
where J is the iteration number.
The capacitance matrix and stiffness vectors are given in
[15] as:
C
el
=
_
xixi1
3
xixi1
6
xixi1
6
xixi1
3
_
(24)
K(h)
el
=
_
PiPi+1
6(xixi1)
Pi+1Pi
6(xixi1)
_
K
s
(25)
K
s
(h, K) = (2K
i
h
i
+K
i
h
i+1
+K
i+1
h
i
+2K
i+1
h
i+1
) (26)
The
h
t
is dealt with using a backwards difference approxi-
mation:
h
t

h(t)
i
h(t)
j1
t
(27)
4 c SACAM 2012
Substituting and simplifying:
K(h)
el
+C
el
h(t) h(t t)
t
= 0 (28)
An iterative method was used to solve the problem and thus
an error residual R(h)
el
is dened as:
R(h)
el
= C
el
_
h(t) h(t t)
_
+ t
_
K(h(t))
el
_
(29)
Applying the Newton-Raphson algorithm:
_
K
T
(h(t))
el
_
h
j
= R(h
J1
) (30)
_
K
T
(h(t))
el
_
=
R(h
K1
)
h

h
j1
(31)
In these equations the subscripts i denote the element number
whereas the superscripts j denote the iteration number. Up to
this point the model is similar to that of [1], however this study
uses a different approach to the mesh generation and tracking
of the ow front as well as calculating the sensitivities.
F. Govignons Solution Method
Tracking the ow front as it progresses through the preform
can be accomplished a number of ways. The previous study
[1] that is being adapted here used a method involving a
oating node. The meant that the time step for each iteration
was xed and because the velocity changes as the ow
progresses the ow front would not be at a node for every
time step and hence a oating node was introduced to indicate
the postion of the ow front. If this oating node was close
to an origional mesh node the timestep was adjusted slightly
to have these points corrospond. This method thus also ment
that more timesteps were used than mesh points to solve
the problem and introduced unnecesary complications to the
solution method. It is assumed that this method was used
as the sensitivities required for the Newton-Raphson scheme
were easier to calculate this way.
G. Proposed Solution Method
In order to simplify the problem this study proposes that
standard nite elements are used in the solution. In doing so
the ow front was made to corrospond with a mesh point. This
means that the time step had to be adjusted for every iteration.
In other words for a mesh that is 100 elements long, 100 time
steps have to be calculated for the ow to reach the end of
the preform. This does however mean that the sensitivities are
more difcult to derive and hence the complex step method
is used to calculate them. The complex step method [17] is a
simple way of calculating the derivative of a function which
may be difcult to calculate analytically. The derivative of a
function is thus given as:
f
x

Im[f(x +iw)]
w
(32)
Here the f is the function being evaluated, x is the point at
which it is being evaluated, i indicates the complex part and
10
0
10
1
10
2
10
3
10
1
10
2
10
3
Mesh Size
E
r
r
o
r
Fig. 1. Error compared to mesh size
w is size of the complex step. The smaller this step size is the
more accurate the derivative becomes to the limit of machine
precision. Due to this method not using any subtraction as a
nite difference method does it is a lot more accurate. Using
this the consistent tangent can be calculated as follows:
_
K
T
(h(t))
el
_
=
R(h
K1
)
h

h
j1

Im[R(h +iw)]
w
(33)
Using this method of calculating the sensitivities preserved
the quadratic convergence of the Newton-Raphson scheme. For
most of the time steps only 3 iterations were needed to achieve
a h < 10
12
. This method of calculating the sensitivities
enables adaptive meshing techniques to be implemented in
the model. This is currently being investigated and different
adaptive meshing techniques are being implemented. These
include using a constant time step in or constant velocity and
adapting the mesh accordingly.
In order to ensure that results obtained were not dependent on
the size of the mesh being used a mesh dependency analysis
was performed where the error int he infusion time was
compared for increasing mesh sizes. Fig. 1 shows that the
error decreases as the mesh size increases, that is, has more
nodes, thus proving that the problem is mesh independent.
The owchart in Fig. 2 provides a general overview of
the process that is followed in the coding. The code was
developed using GNU Octave. The post-lling process is not
shown on the ow chart, however it is a relatively simple
process of changing the boundary conditions.
H. Comparison of Solution Methods
The biggest difference in the methods is the calculation
of the sensitivities. Because this study does not do this
analytically the solution is slightly slower than if they had
been calculated analytically as is usually done. The advantage
is that many other changes can be made, such as changes to
how the mesh is generated and so forth and the sensitivities
will still be calculated accurately and the system will still
converge quadratically. The overall solution scheme is also a
lot simpler by having the ow front corresponding to a mesh
point for every time step. This may also prove useful when
5 c SACAM 2012
Fig. 2. Flowchart of Code
solving the problem for the 2D case where ow front tracking
is more important.
I. Boundary Conditions
The boundary conditions of the problem are relatively
simple due to the model only being 1.5D. The boundary
conditions are thus the inlet pressure, and the vacuum pressure.
The test section is chosen to be wide enough to avoid any edge
6 c SACAM 2012
effects. It had been shown [1] that there is very little pressure
change over the preform with regards to the dry bre and hence
the vacuum pressure is used at the ow front and the pressure
drop over the dry preform neglected. The inlet pressure is
affected by the pressure drop of the resin as it ows through
the inlet pipe as well as any head that is may have to name
a few and this value must be obtained experimentally. This
value could be calculated experimentally, however resistance
factors etc for resin and the pipe would be needed and this
data is not readily available.
The post-lling stage is modelled in the same way as the lling
stage and the boundary conditions are simply changed. The
inlet is closed and thus the inlet pressure is not set to a value
after every time step and the ow front is not advanced. The
vacuum is still applied to last node and thus any ow of resin
that occurs is resin that is being removed from the system.
This is an important process as it determines the nal part
thickness.
The vacuum bag is modelled as a free surface and can thus
displace in the height direction, whereas the bottom surface is
solid. With thin preforms (<10 layers and no core) the ow
front is assumed to be in the same place over the preform,
however with thicker preforms and preforms with a core of
resin distribution medium on the top this assumption would
not be valid.
J. Verication of Model
Verication is done by comparing the results of the com-
puter model accurately represent the mathematics behind the
model. In this case it has been done by comparing to the results
from [15]. Due to the models being similar in many ways the
constants and experimentally determined values could be used
and a very similar system solved for using the new model.
Verifaction was made difcult due to results presented by
[15] not being given numerically but only graphically. The
graphical results were converted to numerical results using
a graph digitizer. Fig. 3 shows the comparison between the
two models. The graph shows the pressure prole over time
for different points along the preform for the two models. The
pressure prole was used as it was the best represented in [15].
There is a direct link between the pressure and height through
the volume fraction and compaction stress. The positive gradi-
ent (lling stage) shows a very close comparison, however the
negative gradients (Post-lling) show some variations. This is
attributed to this model not taking the relaxation of the bres
into account. This may seem like an oversight at this point, but
it will be shown later that the relaxation does not accurately
model post-lling when compared to the experimental results.
Validation will be discussed in section III.
III. MODEL CONSTANT EXPERIMENTAL RESULTS
Before the model could be solved constants for the various
variables such as permeability had to be determined. Each of
these needed separate experiments and are discussed in the
following sections. There constants were obtained from [15]
for the verication process.
100000
[15]
Time(s)
P
r
e
s
s
u
r
e
(
P
a
)
0
0
1000
New
Fig. 3. Comparison for verication purposes
TABLE I
PERMEABILITY CONSTANTS
C n
This studies values 1.73e-10 0.8374
[15]s Values 9.5e-11 2.6
TABLE II
MATERIAL PROPERTIES
Material and Property Value
E-Glass Bulk Density 2550kg/m
3
SAERTEX Bi-Diagonal Area Density 0.409kg/m
3
SAE 30 Viscosity (30
0
) 0.2Pa s
SAE 30 Oil Density (30
0
) 876kg/m
3
A. Permeability Experiments
The constants in (1) needed to be determined. This was
done by xing the V
f
of a layup and then measuring the mass
ow through the preform for a specic driving pressure. Doing
this for a number of different volume fractions the constants
C and n were found using the method of least squares and
given in Table I which also contains the values found by
[15]. This study found that the number of layers made little
difference at high bre volume fractions, which is where VI
takes place. The bre that was used in all of this studies
experiments is SAERTEX Bi-Diagonal Glass (45
0
-45
0
). This
bre was chosen for its linearity and cost. The layers were
always orientated the same way so that the direction of ow
was into the 45
0
angle. Table II shows details about the glass
and oil used in these experiments. SAE 30 oil was used as
it has a similar viscosity to some infusion resin systems and
was readily available. Unfortunately the use of a temperature
controlled table was not possible however the lab where the
experiments were performed does not have windows and is
not inuenced much by weather conditions.
B. Compaction Tests
Many other studies [4], [13], [15] used expensive force
control apparatus to perform compaction tests. This study
7 c SACAM 2012
TABLE III
COMPACTION CONSTANTS
Wet Dry
A 0.3432 0.3326
b 0.0218 0.0352
R
2
0.9424 0.9371
set out to perform these tests as cheaply as possible with
the normal equipment that would be available in a vacuum
infusion lab. Once again two constants had to be determined,
this time from (2) with a and B being the unknowns.
Equation (3) shows how the compaction stress can be related
to the vacuum pressure and atmospheric pressure. Thus it
was decided that the by changing the vacuum pressure and
measuring the height of the preform the constants could be
determined. The height of the preform is directly related to
the bre volume fraction by the following relation:
V
f
=

fibrearea

fibrebulk
h
b (34)
Where b is the number of layers in the layup,
fibrearea
the
area density of the bre given in table II and
fibrebulk
the
bulk density of E-Glass, also given in table II.
These experiments were done for both wet and dry bres.
The drawback of this method was that the experiments were
difcult to perform at low pressures due to the vacuum pump
being used. Values were however obtained that were of the
same order as [15] and thus with a some more work it is
expected that this method could be used for compaction tests.
[15] used a multi-power law to match the data as a single
power law is not very accurate. This study simplied the
approach to a single power law, however with more work a
better relation could be found. Values close to atmospheric
pressure proved troublesome and this is also where the power-
law method seems to fail. This is because at zero compaction
stress the power law gives a zero bre volume fraction.
Table III shows the values for a and B obtained from these
experiments. this table also contains the R
2
values which is the
square of the residuals and gives an indication of the closeness
of the t to the experimental data.
IV. INFUSION EXPERIMENTS RESULTS
For these experiments and the compaction experiments
a stereophotogrammetry measuring device was used. This
enabled the heights of the preform to be measured to within
a few micron but also to be measured over the whole eld
of interest. What the device does is take two photos from
different angles and triangulate points on the preform and then
calculate the heights and subsequently the displacements. With
this device a very accurate full eld set of result were obtained.
Fig. 4 shows the raw data that was obtained from the device
as well as showing the smoothed data for a point along the
preform over time. The data was smoothed using the Tikhonov
regularization scheme. As can be seen from Fig. 4 there is an
error of around 15m in the values obtained from the device,
Time(Non-Dimentional)
P
r
e
f
o
r
m
H
e
i
g
h
t
(
m
)
0 600
0.0034
0.0031
Raw
Smoothed
Fig. 4. Raw and Smoothed Experimental Data
thus data obtained from the device is accurate enough and can
capture the required variations in the preform thickness. Fig.
4 also shows a dip in the preform height which may seem
unusual, however this is due to the bres compacting more
when wet, thus as the resin reaches the bres and the become
wet they compact slightly more and thus the drop in the gure.
The turning point at the top of the gure is where the inlet
is closed and the post lling begins. It is worth noting that
both of these processes are smooth in the experimental case,
however the model does not have a way of determining how
wet the bres are and the response time is slightly slower.
For these reasons the same points on the plot of the models
results are discontinuous points as something in the model
itself changes and this propagates immediately to show in the
results.
Due to the system being solved for in terms of height a
comparison can easily be drawn between the results of the
model and experiments. The experiments were done for a
number of different layers and it was found that this has
little effect on the ow, which to some extent proves that
there is little ow between the layers. The one area that was
affected by many layers was the inlet and it is expected that
this is because the compaction pressure is very low at the
inlet which allows resin or oil to ow between the layers.
Future experiments will include measuring the pressure along
the preform.
V. RESULTS AND COMPARISON
This section is divided into two pieces. Unfortunately the
materials used in [15] were not easily available and hence
this study made use of other materials. There are thus results
of the experiments performed by this study compared to the
model using the constants calculated for this study as well as
the new model compared to the experimental and numerical
results obtained in [15].
A. This Study- Experiments and Model Comparison
Fig. 5 shows the comparison of the experiments performed
in this study compared to results obtained using the model and
8 c SACAM 2012
Time(s)
P
r
e
f
o
r
m
H
e
i
g
h
t
(
m
)
0 7000
0.003
0.0055
Fig. 5. This Studys Model and Experimental Data Compared
constants calculated in this study. Each line represents a point
in the preform. The black lines are representative of the model
and the green lines are the experimental results. It can clearly
be seen that there are a number of problems and inaccuracies.
The rst is that there are vast differences in the height values.
This can be attributed to inaccuracies in the compaction model
and the experiments performed to calculate the constants.
A positive factor of the model is that the time that the
infusion has taken to reach the point where the inlet is close
(Top peaks) is similar. The post lling process also shows
a variation in the results as the model has a steep gradient,
whereas experimentally the excess resin takes a lot longer to
be removed. This phenomenon is discusses in more detail in
the next section.
B. Model Compared to Experiments in [15]
Fig. 6 shows the comparison between this model and the
experiments done in [15]. It is once again clear that more work
needs to be done. Here the time taken to ll the preform do
not correspond. Once again the shape of the lling stage of the
lines is similar, however post lling has a similar problem to
that of the previous comparison. There is also a problem with
the compaction model as the initial heights do not correspond
for the preforms, thus proving to a certain extent that the power
law is not well suited to relating the bre volume fraction and
compaction stress.
The discrepancy in the post ling stages for both of the
comparisons with experimental data show that the model is
lacking some form of resistance which causes the ow to be
so slow in the post lling stage. It is thought that it is due to
the bres absorbing some of the resin and thus having a much
higher permeability which in turn would cause the ow rate
to decrease.
VI. CONCLUSION
Fig. 5 and 6 show that there is still work to be done in terms
of validation of the model, that is, getting it to correspond
with reality. The model presented here however provides a
simple framework which is adaptable and would allow many
Time(s)
P
r
e
f
o
r
m
H
e
i
g
h
t
(
m
)
0 1000 2500
0.0034
0.0046
Fig. 6. This Studys Model and [1]s Experimental Data Compared
different factors to be tested to improve the models accuracy
when compared to experimental data. The exibility that the
new model offers has yet to be fully tested, but will be the
subject of future studies.
Besides work that needs to be done on the model an accurate
way of determining material constants needs to be developed.
The methods presented in this study provide a much cheaper
approach to determining these constants, however more work
is needed in this area as well.
The model presented was veried, as shown in Fig. 3 as
well as the quadratic convergence being maintained without
analytically calculated sensitivities. The method used in this
study for tracking the ow front also simplies the problem.
REFERENCES
[1] Q. Govignon, S. Bickerton, and P. A. Kelly, Composites :
Part A Simulation of the reinforcement compaction and resin
ow during the complete resin infusion process, Composites
Part A, vol. 41, no. 1, pp. 4557, 2010. [Online]. Available:
http://dx.doi.org/10.1016/j.compositesa.2009.07.007
[2] B. Yenilmez, M. Senan, and E. M. Sozer, Variation of part thickness and
compaction pressure in vacuum infusion process, Composites Science
and Technology, vol. 69, no. 11-12, pp. 17101719, 2009. [Online].
Available: http://dx.doi.org/10.1016/j.compscitech.2008.05.009
[3] C. Williams, S. Grove, and J. Summerscales, The compression
response of bre-reinforced plastic plates during manufacture
by the resin infusion under exible tooling method,
Composites Part A: Applied Science and Manufacturing,
vol. 29, no. 1-2, pp. 111114, Jan. 1998. [Online]. Available:
http://linkinghub.elsevier.com/retrieve/pii/S1359835X97000389
[4] N. C. Correia, F. Robitaille, A. C. Long, C. D. Rudd, and S. G. Advani,
Use of Resin Transfer Molding Simulation to Predict Flow , Saturation
, and Compaction in the, Journal of Fluids Engineering, vol. 126, no.
March, pp. 16, 2004.
[5] A. Hammami, Key factors affecting permeability measurement
in the vacuum infusion molding process, Polymer composites,
vol. 23, no. 6, pp. 10571067, 2002. [Online]. Available:
http://onlinelibrary.wiley.com/doi/10.1002/pc.10501/abstract
[6] J. B. Alms, L. Garnier, J. L. Glancey, and S. G. Advani, In-
plane permeability characterization of the vacuum infusion processes
with ber relaxation, International Journal of Material Forming,
vol. 3, no. S2, pp. 12671275, Mar. 2010. [Online]. Available:
http://www.springerlink.com/index/10.1007/s12289-010-0690-7
[7] S. Mekic, I. Akhatov, and C. Ulven, A radial infusion
model for transverse permeability measurements of ber
reinforcement in composite materials, Polymer Composites,
9 c SACAM 2012
vol. 30, no. 7, pp. 907917, 2009. [Online]. Available:
http://onlinelibrary.wiley.com/doi/10.1002/pc.20632/abstract
[8] Y. Ma and R. Shishoo, Permeability Characterization of
Different Architectural Fabrics, Journal of Composite Materials,
vol. 33, no. 8, pp. 729750, Apr. 1999. [Online]. Available:
http://jcm.sagepub.com/cgi/doi/10.1177/002199839903300805
[9] Y. Yoo, Numerical simulation of the resin transfer mold
lling process using the boundary element method, Polymer
composites, vol. 17, no. 3, pp. 368374, 1996. [Online]. Available:
http://onlinelibrary.wiley.com/doi/10.1002/pc.10623/abstract
[10] M. Chohra, S. G. Advani, A. Gokce, and S. Yarlagadda, Modeling of
ltration through multiple layers of dual scale brous porous media,
Polymer composites, vol. 27, no. 5, pp. 570581, 2006. [Online].
Available: http://onlinelibrary.wiley.com/doi/10.1002/pc.20228/abstract
[11] J. M. Bayldon and I. M. Daniel, Flow modeling of
the VARTM process including progressive saturation effects,
Composites Part A: Applied Science and Manufacturing,
vol. 40, no. 8, pp. 10441052, Aug. 2009. [Online]. Available:
http://linkinghub.elsevier.com/retrieve/pii/S1359835X09001079
[12] Y. Song and J. Youn, Numerical investigation on ow through
porous media in the post-infusion process, Polymer Composites,
vol. 30, no. 8, pp. 11251131, 2009. [Online]. Available:
http://onlinelibrary.wiley.com/doi/10.1002/pc.20668/abstract
[13] X. Song, Vacuum Assisted Resin Transfer Molding ( VARTM ): Vac-
uum Assisted Resin Transfer Molding ( VARTM ):, Ph.D. dissertation,
2003.
[14] H. Andersson, T. S. Lundstr om, and B. R. Gebart, Numerical
model for vacuum infusion manufacturing of polymer composites,
International Journal of Numerical Methods for Heat & Fluid
Flow, vol. 13, no. 3, pp. 383394, 2003. [Online]. Available:
http://www.emeraldinsight.com/10.1108/09615530310464553
[15] Q. Govignon and Others, Monitoring and simulation of the
lling and post-lling stages of the resin infusion process, Ph.D.
dissertation, University of Aukland, 2010. [Online]. Available:
http://researchspace.auckland.ac.nz/handle/2292/5630
[16] D. Camotim and M. Ritto-Correa, On the arc-length and other quadratic
control methods : Established , less known and new implementation
procedures, Computers and Structures, vol. 86, pp. 13531368, 2008.
[17] J. R. R. a. Martins, The complex-step derivative approximation, ACM
Transactions on Mathematical Software, vol. 29, no. 3, pp. 245262,
Sep. 2003.
10 c SACAM 2012

Potrebbero piacerti anche