Professional Documents
Culture Documents
1
1
The art of resin transfer molding (RTM) process optimization requires a clear
understanding of how the process performance is affected by variations in some
important process parameters. In this paper, maximum pressure and mold IXhg
time of the I?TM process are considered as characteristics of the process perform-
ance to evaluate the process design. The five process parameters taken into con-
sideration are flow rate, fiber volume fraction, number of gates, gate location, and
number of vents. An integrated methodology was proposed to investigate the effects
of process parameters on maximum pressure and mold filling tirne and to find the
optimum processing conditions. The method combines numerical simulation and
design of experiments (DOE) approach and is applied to process design for a cylin-
drical composite part. Using IiTM simulation, a series of numerical experiments
were conducted to predict maximum pressure and mold filling time of the RTM pro-
cess. A half-fractional factorial design was conducted to identlfy the significant
factors in the RTM process. Furthermore, the empirical inodels and sensitivity co-
efficients for maximum pressure and mold filling time were developed. Compara-
tively close agreements were found among the empirical approximations, numerical
simulations, and actual experiments. These results were further utilized to find the
optimal processing conditions for the example part.
et aL (5)carried out the transient heat-transfer analy- process parameters. The injection pressure and the
sis using LUSAS finite element package and developed mold temperature were the design variables. A contour
two user programs to simulate resin-cure kinetics. map of the design constraints and process parameters
Recently extensive advances have been made both in was created to obtain the optimal design graphically.
the computational infrastructure and in the efficiency Lin et aL (13) used the quasi-Newtonianmethod imple-
and accuracy of numerical algorithms. Kanapady et aL mented in the optimization software package, Global
(6, 7 ) developed a new, implicit, pure finite-element Local Optimizer (GLO), to optimize the gate location in
approach in symmetric multiprocessor machines for order to minimize mold filling time. However, the ana-
large-scale FXM process simulation. lytical methods are only applicable when the objective
Another important research area in ErrM applica- function is smooth and differentiable with respect to
tions is process optimization. Young (8)proposed the the design variables.
use of genetic algorithm (GA) to search for the param- I n summary, considerable research work has been
eters that could reduce the cycle time and enhance conducted in the analysis, modeling, and optimization
property uniformity of the part. Mathur et aL (9)used of the RTM process. They have provided fundamental
a genetic algorithm to optimize gate and vent locations and preliminary understanding of the RTM process op-
for RTM process in order to minimize mold filling time timization problem. However, because of the extensive
and dry spot formation. The searching results of GA computational effort involved, a complete Simulation
were encouraging because of its potential to find a run of a complex part usually takes several hours. This
near global optimum design in a large search space eventually makes it impractical for a lengthy simula-
with many local maxims. However, the large amount tion to be applied directly to the optimization proce-
of computation needed by a single mold filling simula- dure. This paper presents a new design scheme which
tion limited the number of searching generations in combines the RTM simulation and DOE approach to
the evolution. To improve the computational efforts, optimize the RTM process as shown in Fig. 1. In this
Spoerre et aL (10) developed an RTM process optimiza- new scheme, a set of numerical experiments are con-
tion model using cascade correlation neural network ducted to predict maximum pressure and mold filling
(NN) and GA.Through careful design and proper train- time using RTM simulation before time-consuming and
ing, the cascade correlation NN algorithm was devel- costly actual experiments. Based on the simulation
oped into a n accurate predicative model for optimum results, the DOE approach is used to derive the em-
FCIM process design. In a recent study, Luo et aL (1 1) pirical models for the RTM process optimization. The
introduced an NN-based process optimization scheme. empirical models can reduce computational efforts in
With the simulation results, a neural network is trained the RTM process optimization since they are far less
to create a rapid RTM process model. Compared to complex than the original CVFEM models.
the stochastic global search methods, classical ana- The objectives of this study are: 1) to develop a KI'M
lytic optimization methods have better rate of conver- process design scheme that combines the RTM simu-
gence locally. Mychajluk et al. (12) minimized the lation and DOE approach: and 2) to quantitatively as-
cycle-time with constraints on void content and other sess the effects of process parameters on maximum
Part design
Empirical
I
a Geometry and finite
element models
RTM Simulation 1
Responses
I$ I
Factorial Design
Designed numerical
experiments
I
a
RTM Process
pressure and mold filling time of the RTM process and At the flow front region:
to find the optimum processing conditions.
Pfrunt = 0 (6)
2. RTM PROCESS MODELING AND SIMULATION Solutions to Eq 3 along with the boundary conditions
2.1 Governing Equations [Eq 4-61 yield the resin flow front location as well as
In order for process optimization and control of the pressure distribution as functions of time.
processing conditions, process model has been devel- 2.2 Numerical Method
oped based on the fundamental mechanics. Darcy’s
law can describe resin flow in the fiber preform with The IiTM mold filling process is a moving boundary
acceptable accuracy (14): problem, in which the computational domain changes
continuously at each time step. The control volume
frnite element method is used to solve this problem
I* without mesh regeneration. In the CVFEM, the entire
+
computational domain is discretized into a number
where V is the velocity of resin flow, p. is the viscosity of elements. By connecting the centroids of the ele-
of resin, and [K] is the permeability tensor of the fiber ments and the midpoints of the element borders, the
preform and P is the pressure. computational domain can be divided again into a
As the resin is incompressible, the continuity equa- number of polygonal control volumes. Details of the
tion is given by: CVFEM can be found in (2, 3, 15).
Since the thickness of the part is ofken much
-au+ - +a-u= o aw smaller than the length or width, the resin flow in the
ax ay az mold is considered to be a two-dimensional problem.
Equation 1 and Eq 2 are combined and integrated over Equation 3 can be rewritten as:
a control volume to yield the following expression:
r apl
rzi
where h, is the thickness of the part.
Using linear shape function for a 3-node triangular
Ldz J element, a n approximation of Eq 7 can be given as:
where n,, n,, and n, are the components of the vector P = a,, + a,x+ a,y (8)
normal to the surface of the integrated volume, k , (i,j where x and y are local coordinates, and the local co-
= x, y. and z) are the components of the permeability
ordinate system is shown in Rg. 2. a,,. al, a2 and
tensor [ K ] defined in a three-dimensional Cartesian depend on the corresponding coordinates and node
coordinate system. pressure.
The initial and boundary conditions are as follows. The pressure gradient in the element can be com-
At the injection gate: puted as follows:
US& = uo (41
At the mold wall:
ap
-- -0 (5)
dnwll
)/Jy f ; 1)
A
3
. . -
YA
-nA
1 (om A 2 &I
, 0)
Z Z
Global coordinate system Local coordinate system
Fig. 2. Transformation of coordinates.
Substituting Eq 9 to Eq 7 and integrating, we obtain the our composite laboratory was chosen. Figure 3 shows
equation within a small control volume, which is com- the major dimensions of the composite tube. The outer
posed of 3-node triangular elements. diameters at the top and the bottom of the composite
tube are 13.5 inch and 12.00 inch, respectively. The
thickness of the composite tube is 0.25 inch. In our
numerical experiments, the composite tube was dis-
cretized into 1348 nodes and 1326 quadrilateral ele-
ments for the simulation of resin transfer molding pro-
where rn is the total number of elements in a control
cess. Figure 4 shows the finite element mesh for the
volume, wxl, wx2. and wx3 are coefficients for the
composite tube.
pressure of node.
For a 4-node quadrilateral element, the pressure dis- 3.2 Selection of Factors and Responses
tribution of the quadratic element is represented by
Since the RTM process involves a large number of
P = Nipi (111 variables, the major variables with respect to the pro-
cess performance and product quality have to be iden-
1
where N i = - ( l + < i < ) ( l + q i q ) ,C i = - 1 , + 1 . + 1 , - 1 , tified. Based on our laboratory experience, our nu-
4 merical experiments were limited to five important
qi = -1, -1, +1, + l . variables: flow rate, fiber volume fraction, number of
Similarly, we get the equation within a small control gates, gate location, and number of vents. The other
volume composed of 4-node quadrilateral element. variables such as resin viscosity, resin temperature,
and mold temperature were kept as constant as pos-
sible in this study. The levels of the five variables and
their coded values used in the virtual experiments are
shown in Table 1 . These values were selected in order
to cover the normal range of our laboratory operations
Then a set of linear algebraic equations can be ob- for the composite tube. The flow rate of our RTM ma-
chine was limited to 1.0-2.0 cc/s. The high level, ? =
tained by writing Eq 12 for all control volumes in the
computational domain. Solving the linear algebraic 50%. was the highest fiber volume fraction by which
equations together with the appropriate boundary the RTM tube could be conveniently fabricated. At the
conditions, we can obtain the pressure of each node low level. vf = 45%. two plies less than the high level
in the flow field. were used. The fiber volume fraction of vf = 45% was
high enough to give proper contact with both mold
After the pressure field is determined, the velocity
field can be calculated according to Darcy’s law. If a surfaces. Some preliminary actual experiments were
control volume is completely filled (f= l),it is classi- conducted to find two practically applicable injection
fied as the main flow region. If a control volume is
-r
empty (f= 0).it belongs to the empty region. The con-
trol volumes partially filled with the resin is presented
-
9.50 e
-P
as the flow front region (0< f < 1).For a selected time J.50
increment, the volume of resin into control volumes
at the flow front region is calculated and added to the
original resin volume in the control volume at flow front
region. If the total resin volume in a control volume is
equal to the volume of the control volume, that control
volume is considered “full.”The mold f;ning is regarded
as a quasi-steady-state process by assuming a steady
state condition at each time step. Thus, the time in-
crement is determined in such a way that minimum
control volumes will be filled in each time step to en-
sure the stability of the quasi-steady-state approxima- 50
tion (15). After the flow front region at each time step
is updated, another pressure computation is performed
for the main flow region. The procedure is repeated 2.80 4
until the whole mold cavity is filled.
The factorial design of numerical experiments and hypothesis that there is no relation between factor
the values of the responses corresponding to each and response. We see that both models have P-values
simulation trial are reported in Table 2. less than 0.0001 and indicate that they are signifcant
at the confidence level limit of 99.9%.In ANOVA, R2 is
4. DATA ANALYSIS a measure of the amount of variability in the response
explained by the model. The value of R2 ranges 0-1
In this study. the data analysis and optimization and the larger value is more desirable. Adding a vari-
was carried out using Design Expert, a commercial
able to the model will always increase R2, regardless
experimental design software package developed by of whether the additional variable is statistically sig-
Stat-Ease, Inc. The main results of the analysis are
nificant or not. Thus it is possible for models that have
presented here (16).
large values of R2 to yield poor predictions of new ob-
servations or estimates of the mean response. Adjusted
4.1 Analysis of Variance
R-square (Ra2)is better suited for assessing predica-
The results of the design of numerical experiments tive accuracy. The value of Ra2 will often decrease when
can be analyzed through a technique called analysis of unnecessary terms are added to the model. When R2
variance (ANOVA). The analysis of variance (ANOVA) and R 2 differ dramatically there is a good chance that
of the model for maximum pressure and mold filing non-signifcant terms have been included in the sta-
time is summarized in Table 3 and Table 4, respec- tistical model. Since Ra2 is very close to the ordinary
tively. The column "Prob > F" is the P-value of the null R2, the models for maximum pressure and mold filing
Source Sum of squares Degrees of freedom Mean square F Value Prob > F
Model 2.149E+11 9 2.388E+ 10 276.66 < 0.0001
Residual 5.1 79E+08 6 8.361E+07
Cor Total 2.1 54E+ 11 15
Root MSE 9290.58 R-Squared 0.9976
Dep Mean 2.1 54E+05 Adj R-Squared 0.9940
C.V. 4.31
Source Sum of squares Degrees of freedom Mean square F Value Prob > F
Model 5.486Ei05 6 91,427.81 288,808.76 < 0.0001
Residual 28.56 9 3.17
Cor Total 5.486E+05 15
Root MSE 1.78 R-Squared 0.9999
Dep Mean 548.69 Adj R-Squared 0.9999
C.V. 0.32
time are good. The coefficient of variation (C.V.) meas- Similarly, examination of Table 6 indicates that the
ures the unexplained or residual variability in the data following design variables can be declared highly sig-
as a percentage of the mean of the responses variable nificant to mold filling time: (a)flow rate, (b) fiber vol-
(1 7). The C.V. values for maximum pressure and mold ume fraction, (c) gate location, and (d) the interaction
filling time indicate that both models have low vari- of flow rate and fiber volume fraction. Although they
ability. are statistically signifcant, the following design vari-
This approach ensures that only variables that have ables are not included in the practical model for mold
a si@icant effect are included in the statistical model. filling time: (a)number of gates, and (b) the interaction
The statistical effect of each variable on the responses of flow rate and gate location.
is identified by the qualitative significance level, P-val-
ue. That is, P 5 0.05 indicates that the effect of the 4.2 Effect of Factars
variable is significant, and P > 0.05 indicates that the
effect of the variable is not significant. The results of The effect graph is very useful in interpreting the
factorial analysis for maximum pressure and mold f i - main effects and interactions. The effects of factors on
ing time are summarized in Table 5 and Table 6, re- maximum pressure and mold filling time are shown in
spectively. Factors with the value of P > 0.05 were Figs. 6 and 7, respectively. It is observed that the B-
removed from Table 5 and Table 6. Examination of and B+ lines are not parallel. This indicates an in-
Table 5 shows that the following design variables can teraction between factors A and B. The main effects
be declared highly sigmficant to maximum pressure: or two-factor interactions which have large effects on
(a) flow rate, (b) fiber volume fraction, (c) number of maximum pressure are B, A, AB,followed by C and E.
gates, (d) number of vents, and (e) the interaction be- The main effects or two-factor interactions which have
tween flow rate and fiber volume fraction. The following large effects on moldl filling time are A, B, AB, followed
design variables can be declared sigruficant to maxi- by D.
mum pressure: (a) the interaction between flow rate
and number of gates, (b)the interaction between flow 5. MODEL BUILDING AND OPTIMIZATION
rate and number of vents, (c) the interaction between 5.1 Empirical Models
fiber volume fraction and number of gates, and (d) the
interaction between fiber volume fraction and number Once the significant factors are identified through
of vents. But practically maximum pressure will not analysis of variance, regression analysis techniques are
be significantly affected by the variations in these four used to build the quantitative model relating the sig-
interactions. They become statistically sgnitkant be- nificant factors to t h e response. To build the empirical
cause the numerically generated data do not have the model, only significant variables are included. Based
system noise. Therefore, the four interactions are not on the estimated co'efficients in Table 5 and Table 6,
included in the practical model for maximum pressure the empirical models for estimating maximum pres-
although they are statistically significant. sure and mold f m time are
462ml
396950
7'7mi I
717mi
A. A+ B- Bt
(a) A main effect (A-Flow rate) (b) B main effect (B-Fiber volume fraction)
1
462000 462000
396950 396950
9 331900
K
266850
5
1
2 201800
-l
136750 136750
71700 71700
C- C+ E. E+
(c) C main effect (C- Number of gates) (d) E main effect (E-Number of vents)
462000.
396350
331900
f
i
I
266850
8 Zo'QW
136750
71700
A. A+
779 I 719 _I
I
1
3431
A A+
(a) A main effect (A-Flow rate) (b) B main effect (€3-Fiber volume fraction)
779 779
:
a 488333
2 488393
343 sp3
D D+ A A+
(c) D main effect (D-Gate location) (d) AB interaction (A-How rate, B-Fiber volume
fraction)
Fig. 7. Effectgraphsfor moldfling time.
P = 215400 + 71856.25A + 81231.25B were performed. The actual experimental work was
-21518.75C - 16643.75E + 27018.75- (13) conducted with the: resin transfer molding machine
manufactured by Liquid Control Corp. The resin was
T = 548.69 - 182.94A - 26.06B + 7.190 + 8.81AB
delivered to an alurninum mold through clear plastic
(14) hose at a constant llow rate. To reduce the actual ex-
where A, B, C, D,and E are the coded values of the perimental cost, the mold was designed only for one
flow rate, fiber volume fraction. number of gates, gate quarter of the tube clue to the symmetry of shape. The
location, and number of vents, respectively. Based on knitted glass fiber mats were cut into preformed shape
the analysis of variance, both of the models are ade- and placed in the mold cavity before resin injection.
quate. Multiple gates and vents were produced so that a wide
variety of experiments could be performed. These could
5.2 Comparison Between Numerical Simulation be plugged with small screws when not in use. To re-
and Actual Experiments cord the inlet pressure during the experiment, the
To check predicative accuracy of the empirical mod- pressure sensor W a s screwed into the plastic hose
els, four numerical simulations and actual experiments near the gate and then connected to the resin transfer
molding machine. In the Experiment 1, the fiber vol- within 3.96% in maximum pressure and 6.73% in
ume fraction = 45% was used. The resin was in- mold filling time. A good agreement exists between
jected at a flow rate of 1.25 cc/s on the top of the RTM approximations and numerical simulations. On the
tube. Air was vented on the bottom of the tube. The other hand, the empirical models for maximum pres-
resin flow at 100 s after initial injection was repre- sure and mold filling time are also proven to be good
sented in Q. 8. due to the large values of R2 and R2. Therefore, com-
The numerical simulation was compared with the paratively close agreements are found among empiri-
results taken from actual experiments as shown in cal approximations, numerical simulations, and actual
Table 7. Their relative errors were within 1.02% in experiments.
maximum pressure and 4.36% in mold filling time.
Therefore, the numerical simulation yields a good pre- 5.3 Sensitivity Coefficients
diction of maximum pressure and mold filling time. I t To quantify the effects of variations in some design
can be seen that the CVFEM models are reasonable variables on maximum pressure and mold filling time,
representations of the actual resin transfer molding the sensitivity coefficients are incorporated into the
process. In this study, the empirical models for esti- analysis. The sensitivity coefficient is defined as the
mating maximum pressure and mold f w g time were partial derivatives of the numerical-type design vari-
derived from numerical simulation results. The approx- ables such as maximum pressure (Sp,p= dP/dp) and
imations were compared with numerical simulation mold filling time (ST,e= dT/dp) with respect to the nu-
results as shown in Table 8. Their relative errors were merical design variable. In this study, the sensitivity
coefficients of maximum pressure and mold filling time
to number of gates, gate location and number of vents
cannot be calculated because they are discrete varia-
bles. But the sensitivity coefficients of maximum pres-
sure for flow rate (SPA)and fiber volume fraction (SP,*)
can be calculated from Eq 13,respectively.
SPA = 71856 + 27018.75B (15)
Sp., = 81231.25 + 27018.75A (16)
Similarly,the sensitivity coefficientsof mold ming time
for flow rate (S,) and fiber volume fraction (ST,B)can
be calculated from Eq 14, respectively.
sTA - -182.94 + 8.81B (17)
Rg. 8. Resinflow at 100 s after initial injection. S,, = -26.06 + 8.81A (18)
Numerical Actual
Simulations Experiments Error
Case
Number Processing Conditions P (Pa) T (s) P (Ye) T (%)
Flow rate: 1.25 ccls
Fiber volume fraction: 45%
1 Number of gates: 8 107,000 647 106,900 655 0.09 1.22
Gate location: Top
Number of vents: 4
Flow rate: 1.5 cc/s
Fiber volume fraction: 47.5%
2 Number of gates: 4 216,000 523 213,800 530 1.02 1.32
Gate location: 6ottom
Number of vents: 8
Flow rate: 1.75 cc/s
Fiber volume fraction: 50%
3 Number of gates: 8 346,000 401 344,830 410 0.33 2.20
Gate location: Bottom
Number of vents: 4
Flow rate: 2.0 cc/s
Fiber volume fraction: 47.5%
4 Number of gates: 4 281,000 373 282,760 390 0.62 4.36
Gate location: Top
Number of vents: 8
Case
Number Processing Conditions
Approximations
Numerical
Simulations I Error
p (Pa) T (4 P (%) T (Yo)
Flow rate: 1.25 cc/s
Fiber volume fraction: 45%
1 Number of gates: 8 106,875 678 107,000 647 0.12 4.80
Gate location: Top
Number of vents: 4
Flow rate: 1.5 cc/s
Fiber volume fraction: 47.5%
2 Number of gates: 4 220,275 541 216,000 523 1.98 3.44
Gate location: Bottom
Number of vents: 8
Flow rate: 1.75 cc/s
Fiber volume fraction: 50%
3 Number of gates: 8 341,193 428 346,000 401 1.39 6.73
Gate location: Bottom
Number of vents: 4
Flow rate: 2.0 cc/s
Fiber volume fraction: 47.5%
4 Number of gates: 4 292,131 373 281,000 373 3.96 0
Gate location: Top
Number of vents: 8
It can be seen that the sensitivity coefficient of max- The variable, gate location, is kept constant at the
imum pressure to fiber volume fraction is very large coded level -1. Frorn the empirical models, the linear
and also depends on variations in flow rate. The sen- terms of number of gates and number of vents ap-
sitivity coefficient of maximum pressure to flow rate is pears have a negative influence on maximum pres-
large and depends on fiber volume fraction. It was also sure and mold fillirig time which implies that an in-
found that the sensitivity coefficient of mold filling time crease in number of gates and number of vents would
to flow rate is very large and depends on fiber volume result in a less maximum pressure and mold filling
fraction. But the sensitivity coefficient of mold filling time. Therefore, the two variables, number of gates and
time to fiber volume fraction is relatively small and de- number of vents, are kept constant at the coded level 1.
pends on flow rate. The graphical opiimization criterion is specified by
choosing the response limits,lower and upper, for each
5.4 Optimization of RTM Process response to be included in the optimization. Based on
In fiTM process design, one of the objectives is to our laboratory experience, the mold cavity for the com-
minimize mold filling time subject to all necessary pro- posite tube needs to1 be completely filed with the resin
cess constraints such as pressure. To solve this opti- within 360 to 480 seconds subject to the maximum
mization problem, the simultaneous optimization of pressure 200,000 to 300,000 Pa. The optimum region
multiple responses can be performed graphically based was found by overlaying maximum pressure and mold
on the empirical models. Graphical optimization dis- filling time response surfaces shown in Rg. 9. The op-
plays the area of feasible response values in the factor timum conditions for flow rate and fiber volume frac-
space. The area is the overlapping shaded area, which tion range 0.5-1.00 and -0.5-0.5 at the coded level,
satisfies the multiple constraints on the responses. respectively. That is, the optimum flow rate is in the
In the graphical optimization, the contour plot will range of 1.75-2.0 cc/s, and the optimum fiber volume
be displayed in two dimensions. If there are more fac- fraction is in the rarige of 46.25%-48.75%.
tors than axes, the levels of the other variables need
to be set to identify the appropriate slice of the design 6. CONCLUSIONS
space. The sensitivity coefficients lead us to recognize The following conclusions/recommendationscan be
that the flow rate and fiber volume fraction are the drawn from the results and analysis:
most sensitive factors with regard to maximum pres- 1. Numerical simulation and DOE approach can be
sure and mold filling time. Then flow rate and fiber vol- employed to investigate the effects of process pa-
ume fraction are selected as the axe variables, while rameters on maximum pressure and mold filling
the other three factors such as number of gates, gate time and to firid the optimum processing condi-
location and number of vents are held constant. Since tions. This melhod can reduce computational ef-
gate location has a positive influence on mold filling forts in the RIM process optimization since the
time, the injection of resin on the bottom of the com- empirical models are far less complex than the
posite tube is suggested to minimize mold filling time. original CVFEM models.
1 00
0.50
\
R g . 9. Optimum region of maxi-
0.00
\
IMaximum
mum p r e s s u r e a n d mold filling
time.
-050
-1 00
I I I I
3 -050 0 00 050 1.go
Flow rate
2. It was found that maximum pressure is sensitive reduce the computational time for simulation it-
to fiber volume fraction and flow rate, followed by erations in optimization, as the less important
number of gates and number of vents. Together factors can be excluded from the simulation. A
with multiple gates and vents, the low fiber vol- future work of this study will investigate the ef-
ume fraction and flow rate will yield a low maxi- fectiveness of this approach.
mum pressure at the end of the filing stage.
3. Mold filling time is sensitive to flow rate and fiber REFERENCES
volume fraction, followed by gate locations. To 1. M. V. Bruschke and S. G. Advani, SAMPE (1991).
reduce the mold filling time, a higher flow rate 2. W. B.Young, K. Han, L. H. Fong, J. Lee, and M. J. Liou,
and fiber volume fraction and gates on the bot- PoZyrn Compos., 12,6 (1991).
tom of the RTM tube are suggested. 3. W. B. Young, P o b m Compos., 15.2 (1994).
4. M. K. Kang and W. I. Lee, composites Science and Tech-
4. The region of the optimum was found by over- nology. 59 (1999).
laying maximum pressure and mold filling time 5. S.C.Joshi, X L. Liu, and Y. C . Lam, Composite Science
response surfaces based on the empirical mod- and Technology, 59 (1999).
els. Together with multiple gates on the bottom 6. R. Kanapady, K. K. Tamma, M. Baddourah, a n d A.
of the composite tube and multiple vents on the Mark, Advances in Engineering S o j b a r e , 29 (1998).
7. R. Kanapady, K. K. Tamma, and A. Mark, Numerical
top of the composite tube, the optimum con- Heat nansfer, Part B: Fundamentals. 36.4(1999).
ditions for flow rate and fiber volume fraction 8. W. €5. Young, J. CompositeMaterials, 28. 12 (1994).
are in the range of 1.75-2.00 cc/s and 46.25%- 9. R. Mathur, B. K. Fink, and S. G . Advani, PoZyym Corn-
48.75%. respectively. ps., 20,2 (1999).
10. J. Spoerre, C. Zhang, H. P. Wang, and R. Parnas, J.
5. The proposed models are applicable within the CompsiteMaterials, 32,13 (1998).
selected range in this study. For understanding 11. J. Luo, Z. Liang, C. Zhang, and B. Wang, Composites
the responses beyond this range, suitable modi- PartA:Applied Science and Manufacturing, 32 (2001).
fications of the models have to be made. More- 12. G. Mychajiuk, S. Manoochehri, and R. Parnas, J. Ad-
over, the influence of other parameters such as vanced Materials, 28, 1 (1996).
13. M. Y. L h ,M. J. Murphy, and H. T. Hahn. Composites
material properties, resin and mold temperature, Part A: Applied Science and Manufacturing, 31 (2000).
and the curvature of mold need to be examined. 14. S. G. Advani, Flow and Rheology in Polymeric C o m p s -
6, Another branch of the research is to derive the ites Manufaturing, Elsevier Publishers (1994).
empirical process model from the results of ac- 15. T. G. Gutowski, Advanced Composite Manufacturing,
John Wiley & Sons (1997).
tual experiments using DOE approach. Then, the 16. Design-Expert 6 User's Guide, Stat-Ease, Inc. (2000).
numerical simulation will be performed to opti- 17. D. C. Montgomery, Design and Analysis of E.xpriments,
mize the RT!M process with the empirical relation- John Wiley & Sons (2001).
ship as the starting point. This will signifkantly