Beruflich Dokumente
Kultur Dokumente
Method to account for the fiber orientation of the initial charge on the fiber
orientation of finished part in compression molding simulation
Yuyang Song, Umesh Gandhi, Camilo Pérez, Tim Osswald, Srikar Vallury,
Anthony Yang
PII: S1359-835X(17)30204-X
DOI: http://dx.doi.org/10.1016/j.compositesa.2017.05.021
Reference: JCOMA 4673
Please cite this article as: Song, Y., Gandhi, U., Pérez, C., Osswald, T., Vallury, S., Yang, A., Method to account
for the fiber orientation of the initial charge on the fiber orientation of finished part in compression molding
simulation, Composites: Part A (2017), doi: http://dx.doi.org/10.1016/j.compositesa.2017.05.021
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
PUBLIC 公開
Method to account for the fiber orientation of the initial charge on the fiber
orientation of finished part in compression molding simulation
Yuyang Song1*, Umesh Gandhi1, Camilo Pérez 2, Tim Osswald2, Srikar Vallury3, Anthony Yang3
Abstract
In the automotive industry, interest in compression molding of discontinuous fiber reinforced
thermoplastic is rapidly increasing. Compression molding offers fast cycle times, allowing us to
maintain longer fibers. Longer fibers translate into improved mechanical properties, and hence,
lighter parts. The mechanical properties of such components also largely depend on the fiber
orientation. It has been observed that the fiber orientation in the finished part depends on the
initial fiber orientation of the bulk charge being compressed. In this paper, we present methods to
measure initial fiber orientation in the bulk charge and demonstrate how it can be included in the
simulation. Computer tomography is used to scan the bulk charge to estimate the orientation of
the fibers in the charge. The estimated orientation through the charge is mapped as the initial
condition for the charge used in the compression molding simulation. We also demonstrate the
improvement in the predicted fiber orientation resulting from the proposed approach with a
physical example.
*Corresponding Author
Toyota Research Institute, NA, 1555 Woodridge, Ann Arbor, MI 48105
Email: Yuyang.song@toyota.com
1
PUBLIC 公開
1. Introduction
Compression molding is a method of plastic molding process in which the molding material,
generally preheated, is placed in an open, heated mold cavity that is suitable for molding
complex, high-strength objects [1-2]. With its short cycle time and high production rate,
compression molding is widely used in the automotive, aerospace, electronics, and sporting
goods industries to produce parts that are large, lightweight, high strength [3-4]. For the fiber
reinforced thermoplastic materials, the initial charge used can broadly be in two forms, bulk or
sheet. Bulk charge is desired to meet the high production volume and lower cost targets that are
typical for the automotive applications [5-6]. Sheet material consists of discontinuous fiber mats
with thermoplastic polymer embedded as copolymers or thin sheets. While such sheets do offer
longer fiber length and improved mechanical properties, the processing is still a challenge. In this
study, we will be focusing on the bulk material only.
Before compression molding, the bulk charge is made using the twin screw or single screw
extruders to compound of fibers and resin. After mixing, the bulk material is placed in the mold
cavity and compressed, as shown in Fig.1. The process offers a higher potential to maintain
longer fibers and better mechanical properties compared to injection molding, as the shear flow
rate through the cavity is quite limited, therefore causing less fiber breakage. The properties of
the finished part are highly dependent on the fiber condition in the part, i.e. fiber length, fiber
orientation and volume fraction distribution through the part. It has also been observed that the
fiber condition varies throughout the part which is a function of process parameters and
geometry of the part.
In order to predict the mechanical properties of the final part, it is important to use the fiber
condition of the final part in considerations. Computation methods based on continuum
mechanics can be used to simulate the compression molding process and estimate the fiber
condition in the final part. Commercial software such as Moldex3D [7] and Moldflow [8] are
available for this purpose. While algorithms to estimate fiber length and fiber concentration are
still evolving [9,10] methods to estimate fiber orientation are well developed. In this study, we
will focus on improving the capability to predict the fiber orientation in the finished part.
2
PUBLIC 公開
Usually, there are thousands of fibers in the reinforced part, e.g. in 1 cubic mm, 30% glass fiber
(15-µm diameter) could mean 3000-4000 fibers. Therefore, using the orientation distribution
function ψ(p) is more efficient than taking into account every single fiber. The orientation
function represents the probability of a fiber to be in a certain orientation p. Advani and Tucker
[12] proposed a more general and concise expression, a second order tensor , for the
representation of fiber orientation distribution using equation 1. Here, the unitary vector is
averaged over the selected volume. This method is widely used in the industry.
(1)
As this method explained, a high probability of fiber alignment in the specified principal
direction will be indicated by a value close to 1, whereas a low probability is indicated by a value
close to 0. A value of 1/3 in all principal direction means that fibers are oriented randomly. This
method of expressing fiber orientation is concise and easy to use with the numerical calculations
used to estimate mechanical properties of fiber reinforced polymers [13-14], it is also widely
adapted in the industry [7,8,15].
The most widely used computational approaches to simulate the compression molding is
Jeffery’s equation [16] with Folgar-Tucker model [17] to estimate the fiber orientations. This
model considers the material’s velocity gradients, strain rates, and fiber-fiber interaction. Using a
fiber orientation algorithm based on the Folgar-Tucker model, Chih-Chung Hsu [18] predicted
thoroughly the fiber flow and fiber orientation after compression molding simulation.
The traditional approach has been to assume that orientation of fibers in the initial compression
charge is random. However, due to the plasticization process in the extruder, a distinctive non-
random fiber orientation emerges within the charge [19]. For an accurate prediction of fiber
orientation in the finished part, it is important to start with an accurate description of initial
orientations. Until now, there has been no method to include the effect of initial fiber orientation
in the bulk charge for the compression molding process. The main purpose of this research is to
develop a method to account for the initial orientation in the compression charge and study the
effect of initial fiber orientation on the final fiber orientation of the compression molded
thermoplastic parts.
3
PUBLIC 公開
In Step 2, as Fig.4 shows, the measured orientations are mapped on the finite element model of
the charge. Depending on the shape, some scaling might be required. To help this process, an
algorithm, which maps the estimated fiber orientation measurement to each element of the finite
element model for the charge, has been developed in-house using Matlab [20]. After the mapping
process, the fiber orientation in the initial bulk charge model is defined as initial fiber
orientation. Finally, the compression molding simulation with the defined fiber orientation is
carried out using Moldex3D.
4
PUBLIC 公開
Based on the mesh of the charge defined in Moldex3D, the normalized µCT data set is rescaled
to the dimensions of the mesh. The rescaling continues until the µCT data set comprises the
entire mesh of the charge. This approach leads to points lying outside of the mesh as shown in
Fig.6. In the end, all points outside the mesh are removed from the data set. The dimensions
illustrated here are just used for the explanations of the coding algorithm. For the simplification
of the validation purpose of the code, the part designed for this study is a rectangle bar.
In order to have a local orientation for every point of the Moldex3D mesh of the charge, an
interpolation algorithm is applied to the orientation tensor at each node of the mesh. The radial
basis function (RBF)[21-24] is a mesh-less interpolation technique and has been shown to be a
powerful method for representing sparse, scattered data. The RBF interpolation represents
interpolated data as a linear combination of the non-linear basis function that is defined at each
control point and has only a radial component
5
PUBLIC 公開
=
| − | =
(2)
where represents the location of the point where interpolation is performed, represents the
position of the control point in a specified region of interest, is the basis function, is the
weight factor to be determined and N is the number of data points. The distance is defined
between the point where the function will be interpolated to and the points where the function is
known from the µCT data set. For the basis function, , the thin plate spline (TPS)[25] was used
because it gives accurate results for scattered data approximations
The first step in the interpolation is the solving for the weight factors . A system of equations is
calculated, denoted as
⋯
⋮ ⋱ ⋮ … = … (4)
⋯
where describes the distances between all possible combinations of points within the µCT data
set. The vector on the right hand side contains the target values and, in the case of the fiber
orientation distribution, is one component of the orientation tensor with
From the µCT data set, all and the corresponding are known. Thus, the system can be
solved for the weight factors. Solving
r =
(6)
for an arbitrary point of the Moldex3D mesh, the orientation tensor of that point can be
interpolated. This is done for all nodes of the Moldex3D mesh and components of the orientation
tensor .After the interpolation, the mapped fiber orientation and its corresponding coordinates
are imported into Moldex3D. The mapping tool was programmed with Matlab[26].
6
PUBLIC 公開
A hat section part is designed and manufactured at a Magna facility in Toronto using the D-LFT
process. The material used for the prototyping is 40% glass fiber reinforced polypropylene
(40GFPP). Fig.8 shows the images for charge located in the tool.
Table.1 shows the processing condition used for the part fabrication. The finished sample is presented in
Fig.9.
Process conditions Actual values
Mold Temperature 70 oC
7
PUBLIC 公開
Fig.9 Sample made after compression molding, A, B and C are fiber orientation measurement locations
a. µ CT scanned image and selected cells through thickness b. 3D structure of the fibers inside selected cell
Fig.10 Fiber orientation tensor measurement from the µ CT scanning images of a sample cut from part
8
PUBLIC 公開
Fig.11 shows the actual scaling process used to map the estimated orientation from the
measurement to the actual finite element model used as initial charge for the compression
molding simulation. Getting CT scan of an actual charge and creating map of orientation is quite
complex and time-consuming process, due to the large size of data (~80GB). To address this
challenge, a smaller charge from similar twin screw was CT scanned and a map of orientation
distribution was created. Next, the orientation from the smaller charge was mapped on the larger
actual charge using radial basis function (RBF) approach. We recognize that this is an
approximation, which is practical and offers improvement over current method assuming the
fibers to be completely random or aligned in only one direction. The details about the
compression molding simulation set-up and boundary conditions are in the next section.
Fig.11 Initial fiber orientation mapping from the real charge to simulation
9
PUBLIC 公開
the average fiber length by weight is about 8mm. The fibers in the compression molded charge
are highly bundled. Previous work by Perez [20] has reported average fiber length the
compression charge at around 15 mm.
Fig.12 Fiber length distribution from compression molded part at location 1 and location 2
∂T
ρC P + u ⋅ ∇ T = ∇ (k ∇ T ) + Φ
∂t (9)
C
where vis velocity vector, p is pressure, T is temperature, τ is the stress tensor, and ρ , η, p and k
represents density, viscosity, heat capacity, and thermal conductivity of the plastic material,
respectively.
10
PUBLIC 公開
A volume fraction function f is introduced to track the evolution of the melt front. Where f = 0
is defined as the air phase, f = 1 as the melt phase and the melt front is located within control
volume between 1 and 0. The advancement of f over time is described by the following
transport equation:
∂
f + ∇⋅ fv = 0
∂t (10)
The collocated cell-centered finite volume method proposed by Chang and Yang [32] was
modified to discrete the governing equations (6) -(8). The numerical method is basically a
SIMPLE-like finite volume method with improved numerical stability and convergence for non-
isothermal and non-Newtonian flow. Pressure, velocity, and temperature fields are segregated in
the solver so that the efficiency can be achieved even for three-dimensional calculation. Special
numerical techniques are incorporated in the flow solver so that the numerical convergence and
accuracy can be achieved for the highly non-linear system.
Moldex3D developed the iARD-RPR model [33,34] to predict the fiber orientation change over
time during the filling and compression phase. This is an improvement over the Anisotropic
Rotary Diffusion (ARD) model [16]. The RPR model is indicative of Retarding Principal Rate to
slow down the fast transient orientation rate. iARD-RPR model uses 3 parameters to define fiber
orientations:
=A
+A
A HD IARD (CI , CM ) + ARPR (α ) (11)
The iARD-RPR model is used in this compression molding simulation to predict fiber
orientation at the final part.
The geometry of the model used for simulation is presented at Fig.13 and the detail of the model
is listed as well. The total cavity volume is 167cc, and the charge volume is 193cc (about 115%
of the cavity size). The detailed material properties used for the simulation is shown in Table.2.
The process condition used for the simulation followed the same as the manufacturing process
such as compression time, melt temperature, etc. Fiber orientation tensors are extracted after
filling process at locations from point A, B and C—which is exact the same location used from
the actual finished part, showing at Fig.14.
11
PUBLIC 公開
12
PUBLIC 公開
in Fig.16, that the fiber orientation in point B has two different results between these two
scenarios. The case when initial fiber orientation is unidirectional shows higher fiber orientation
results than the case when the initial fiber orientation is random. After this comparison, the
validation of the Matlab code is completed, the validation process builds up confidence for the
next part for the real fiber orientation mapping.
Fig.15 Validation process for the fiber orientation tensor results between the mapped process and non-mapped
process
Fig.16 Fiber orientation tensor results between the mapped process and non-mapped process
13
PUBLIC 公開
• CAE initial random: All fibers in the initial charge are random
• CAE initial mapped: Fiber orientation in the initial charge is mapped from the measured data
using the proposed method.
• CAE initial aligned in A11 direction: All fibers are assumed to be aligned along the length of
part
4.1. Fiber orientation tensor distribution comparison
Fig.17 Comparison of the fiber orientation tenor(A11) for Point B by using different CAE input with measurement
The predicted fiber orientation (CAE) in the finished part at location B, for different initial fiber
conditions, is compared in Fig.17. The results, based on the visual comparison, show that when
the initial fiber orientations are mapped, the prediction is closer to the actual measurement, while
other parameters such as fiber to fiber interaction parameter (Ci) remain the same at 0.01.
Fiber interaction coefficient (Ci) is an empirical constant that characterize the effect of fiber-
fiber interaction in the concentrated suspensions [7,8]. Variation in the Ci will result in different
level of fiber to fiber interaction and hence different fiber orientation tensor distribution in the
part. One of the approach used to improve simulation, is to change this empirical value of Ci
between 0 to 1, until a good match between the measured and estimated fiber orientation is
obtained. To understand the effect of Ci on the predicted orientation tensor, we also estimated
fiber orientations for two additional values of Ci i.e. 0.005 and 0.02
The fiber orientation tensor components A11, A22 and A33 for the three selected locations A, B
and C at different Ci are presented in Figure 18, 19 and 20. The improvement of using the initial
fiber orientation from measurement is quite evident from Location B and C, see fig 19 (a) and 20
(a). We observe a good match not only in terms of value but shape of the fiber orientation tensor
through the thickness. We believe that this is result of using initial fiber orientation based on the
measured values. Physics of the process also supports this claim. Typically, when resin flows
the fiber tends to align along the flow in a laminar flow pattern. The time or length of flow to
align all the fibers depend on the viscosity and other flow parameters. In our case of compression
14
PUBLIC 公開
molding, the time for fibers to align is very small and therefore the initial fibers in the charge are
playing important role in how the fibers are distributed in the charge.
For the clarity, the case when initial fibers are aligned along the length direction of the hat
section part is not included in these plots.
(a) (b)
(c)
Fig.18 Comparison of the fiber orientation results (A11, A22, A33) from point A at different Ci
15
PUBLIC 公開
(a) (b)
(c)
Fig.19 Comparison of the fiber orientation results (A11, A22, A33) from point B at different Ci
16
PUBLIC 公開
(a) (b)
(c)
Fig.20 Comparison of the fiber orientation results (A11, A22, A33) from point C at different Ci
The tensile modulus for A11 direction is estimated for 40GF Polypropylene material and
estimated orientation tensor using Mori-Tanaka homogenization [35] based on commercial
software Digimat [36], see Table 3. The modulus calculated are based on the fiber orientation
17
PUBLIC 公開
obtained by using Ci=0.01 for the baseline condition. We have also included the measured value
of elastic modulus at location B and C, as reference. The comparison shows that the predicted
elastic modulus from CAE using initial mapped tensor get closer match to the modulus from
measured modulus using MTS machine.
Table.3 Tensile modulus calculated at three locations for different fiber orientations assigned
Fig.21 Comparison of the calculated E11and measurement for these three locations and the difference between them
5. Conclusion
In summary, the effect of initial fiber orientation of the charge on the final fiber orientation in the
finished part in compression molding is investigated. The results show that for compression
molding, the initial fiber orientation in the charge is an important microstructure detail that must
be considered. A Matlab code is developed to map the fiber orientation measured from µCT scan
18
PUBLIC 公開
to the bulk charge used for compression molding, using radial basis function (RBF). The
compression molding simulation results show that by accounting for initial fiber orientation
mapped from the measurement, the predicted fiber orientation (A11, A22, A33) in the finished
part shows a closer agreement with the measured values.
Therefore, based on this study, we strongly recommend the use of initial fiber orientation for the
charge during the compression molding simulation. We do recognize that measurement of such
charges can be quite time consuming and expensive. One approach to address this may be the
development of a database of generic initial orientations for a given extruder for a range of
processing conditions. Also knowing repeatability of conditions in the charge the size of the
charge that needs to be scanned can be reduced.
Acknowledgements
The authors gratefully acknowledge the support provided by Magana Canada for the materials
/tool to manufacture the part, also for the support from Moldex3D Northern America, Inc. for the
technical support on CAE simulations.
Reference
1. Isayev, A. Injection and compression molding fundamentals. CRC Press, 1987.
2. Beardmore, P., et al. Fiber-reinforced composites- Engineered structural
materials. Science 1980; 208:832-840.
3. Biron, Michel. Thermoplastics and thermoplastic composites. William Andrew, 2012.
4. Ebewele, R. O. Polymer science and technology. Boca Raton, FL: CRC Press,2000.
5. Beardmore, P., & Johnson, C. F. The potential for composites in structural automotive
applications. Composites science and Technology,1986;26(4):251-281.
6. Schuh,T.G. Renewable materials for automotive applications. Daimler-Chrysler AG,
Stuttgart,1999.
7. Manual of Moldex3D software, CoreTech. Inc, Taiwan,2014.
8. Manual of Moldflow software, Autodesk.Inc. New York, USA,2014.
9. Hine, P. J., Lusti, H. R., & Gusev, A. Numerical simulation of the effects of volume fraction,
aspect ratio and fiber length distribution on the elastic and thermoplastic properties of short
fiber composites. Composites science and technology, 2002;62(10):1445-1453.
10. Londoño-Hurtado, A., Osswald, T. A., & Hernandez-Ortíz, J. P. Modeling the behavior of
fiber suspensions in the molding of polymer composites. Journal of Reinforced Plastics and
Composites, 2011;30(9):781-790.
19
PUBLIC 公開
20
PUBLIC 公開
29. Weber, Charles, and Chris Rannenberg. Method of determining an average length of
reinforcing fiber in a sample of reinforcing fibers. U.S. Patent No. 6,925,857. 9 Aug. 2005.
30. L.Liu, W. H. Yang, D. Hsu and V.Yang, ANTEC 02, San Francisco, CA, 5–9 May 2002.
31. Muzaferija S., Peric M., Sames P., Schelin T., Proc. Twenty-Second Symposium on Naval
Hydrodynamics, Washington, D.C, Aug.14,1998,
32. R. Y. Chang and W. H. Yang, Numerical simulation of mold filling in injection molding
using a three-dimensional finite volume approach. Int. J. Numerical Methods Fluids,2001;
37:125.
33. T.-H. Le, P.J.J. Dumont, L. Orgéas, D. Favier, L. Salvo, E. Boller, X-ray phase contrast
microtomography for the analysis of the fibrous microstructure of SMC composites,
Composites Part A,2008;39(1):91–103.
34. Huan-Chang Tseng, Yuan-Jung Chang, Chia-Hsiang Hsu. Prediction of Fiber Microstructure
for Injection Molding: Orientation, Degradation and Concentration, SPE-ACCE
Conference,2014, Novi, Mi.
35. T. Mori and K. Tanaka. Average stress in the matrix and average elastic energy of materials
with misfitting inclusions. Acta Metall. Mater.,1973;21:571–574.
36. Digimat User Manual, MSC Software corporation, Newport Beach, USA, 2015.
21