Professional Documents
Culture Documents
cn
526 Int. J. Materials and Product Technology, Vol. 21, No. 6, 2004
Abstract: One of the important quality issues for injection moulded parts is
weld lines. The injection moulding simulation has been greatly developed
over a couple of decades. However, current simulation packages are
incompetent of detection and evaluation of weld lines location and
properties. In this paper, a detection model of weld lines location is
presented, which is based on the characteristics of the initial meeting node.
A quantitative evaluation model using the artificial neural network
approach is also established in this paper. For identifying the input of the
network, the affecting factors of weld line properties are discussed in detail
± the orientation coefficient of the material, the meeting angle, and the melt
mobility history coefficient. Comparison with experimental results shows
that these presented models are capable of predicting weld line location and
properties for engineering design.
1 Introduction
The interest in weld line analysis of injection moulded parts has increased
tremendously in the past few years. The processing of large and complex parts on
one hand, and the ever-increasing requirements on the performance of injection
moulded items on the other, are some of the most important reasons behind the
renewed interest in weld lines (Dairanieh et al., 1996). Weld lines are formed when
two melt fronts come into contact with each other. Separation or division of the melt
flow front can be caused by flow obstructions such as cores and holes, melt front
race-tracking due to variable wall thickness, jetting, unfavourable gate location or
runner branching for multi-gated parts. When these separated melt fronts rejoin at
some downstream location, weld lines are formed. Since the appearance as well as the
mechanical properties of the injection moulded part can be affected by weld line
formation, a strong need exists for understanding this phenomenon and solving the
problems associated with its occurrence.
Computer simulation packages of injection moulding have had success in
predicting filling behaviour in extremely complicated geometries. Although some
current packages can predict weld lines location, their accuracy and efficiency are
unsatisfactory, and none of them is capable of predicting weld line properties
quantitatively. In this paper, after analysing the characteristics of weld line
formation, the model for the initial meeting node is established based on the filling
simulation results. And correspondingly, a reliable and efficient algorithm to detect
weld line location is presented. On the other hand, based on a thorough analysis of
the factors affecting the weld line strength, an ANN (Artificial Neural Network)
approach is used to establish a prediction model for evaluating weld line properties
quantitatively.
There are two steps in the detection of weld lines. The first is to detect the initial
meeting point of two flow fronts and then to expand from the initial meeting point to
obtain the entire weld lines.
As for the initial meeting point, the detection model can be deduced as follows.
Premise: Because the detection of weld lines is based on filling simulation results,
the time of melt front reaching each finite-element node is known, which
is named as reaching time here.
Theorem 1: The first contact point of two flow fronts meeting is called the initial
meeting point. All initial meeting points must locate at the finite-element
nodes of the mesh, so the initial meeting point can also be called the
initial meeting node.
Because the reaching time of the filling simulation is based on finite-element nodes,
Theorem 1 is obviously true. Obviously, assuming the flow fronts meet at finite
element nodes is an approximation for the convenience of numerical treatment. The
precision of the prediction depends on the mesh size.
中国科技论文在线 http://www.paper.edu.cn
Theorem 2: For each initial meeting node (such as node C in Figure 1), when the
melt is very very near to, but still does not arrive at it, in the surrounding
polygon which is constituted by its adjacent nodes (such as Polygon
C1±C2±C3±C4±C5±C6±C7 in Figure 1), there are two un-intersectant
flow fronts (such as F1 and F2 in Figure 1).
Theorem 2 can be obtained by the definition of the initial meeting node.
Inference 1: According to Theorem 2, when the melt is very near to but still does
not arrive at the initial meeting node, there would be four points of
intersection between the flow fronts and the surrounding polygon (such
as P1, P2, P3 and P4 in Figure 1).
Theorem 3: For the surrounding triangle which includes a point of intersection of
Inference 1 (such as the shadow triangle T1 in Figure 1), between the two
lines linking to the initial meeting node (such as S1 and S2 in Figure 1),
one has a point of intersection with the flow fronts (such as S2 in Figure
1), the other does not.
The reason of Theorem 3 is that if the flow front gets through the triangle, there will
be two, and only two, points of intersection between the flow front and the triangle.
Inference 2: According to Theorem 3, in the surrounding triangle which includes a
point of intersection of Inference 1, between the reaching time of the two
nodes except the initial meeting node (such as C1 and C2 in Figure 1),
one is earlier than that of the initial meeting node (such as C2 in Figure
1), and the other is later. Supposing that the reaching time of points C,
C1, and C2 are TC, TC1 and TC2 respectively, it follows that
TT
TC ÿ TC1
TC ÿ TC2 < 0
1
Conclusion: For each finite-element node of the mesh, if there are four surrounding
triangles satisfying Equation (1), the node is an initial meeting mode.
中国科技论文在线 http://www.paper.edu.cn
According to the above Conclusion, all initial meeting nodes can be found out by one
traversal of all finite-element nodes with Equation (1). The efficiency of this
algorithm is very high.
Starting from the initial meeting node, the weld line expands along two opposite
directions (such as V1 and V2 in Figure 1). Therefore, let the initial meeting node
being the current meeting node, the two nodes with the latest reaching time among
the nodes surrounding the current meeting node are the next two meeting nodes; and
then let these two nodes be the current meeting node and repeat the above process
until the entire weld line is built.
* Part and mould design. Modifying the part and/or mould structure to facilitate
the transmission of pressure and maintain a higher melt temperature is in favour
of improving weld lines properties. Placing a vent in the area of weld lines can
also increase the weld lines index, because the entrapped air would further
weaken the weld lines.
The above-mentioned microscopic causes are all related to the molecular mobility
during the creation of the weld lines, directly or indirectly, so the molecular mobility
is the most essential factor affecting the properties of weld lines. Of course, the
mobility in turn depends on the material characteristics and processing conditions.
Based on the above analysis, stressing the dominating factors and ignoring the minor
factors, the factors affecting weld line properties can be summed up as the following
three points.
Figure 3 The formation of a weld line (a) melt fronts approach, (b) weld line forms,
(c) meld line forms, (d) final results (1: meld line; 2: weld line)
中国科技论文在线 http://www.paper.edu.cn
where tfill and tvit are the moment of melt arriving and cooled to glass-transition
temperature respectively, P is the melt pressure, and m denotes the mean viscosity at
the interface, calculated by
Z b
1 1
dz=b;
5
m 0
z
where b is the half thickness of the cavity, and z means the coordinate in the gap-wise
direction.
mobility history coefficient. And the output is FWL . In order to reduce extra- or
over-fitting, a value of 0.9 is used for both momentum factor and rate of learning
with the generalised delta rule (GDR) method. No bias is employed for the network.
The network is trained using the MATLAB neural network toolbox. Using the
experimental data from open literature and the simulation results of the self-developed
CAE system HSCAE (Zhou and Li, 2001; Zhou et al., 2001), about 500 sets of inputs/
outputs are used for training the network. The training is achieved after 600 epochs.
4 Experimental verification
4.1 Verification of weld lines location
As shown in Figure 5, the experimental part is a plate of 100 100 mm. The thickness
of one half of the plate is 2 mm and the other is 1 mm. Two gates are placed at the
centre of one side of the two sections. The selected material is HDPE G201. The
parametric constants corresponding to the
n; ; B; Tb ; of the five-constant
Cross-type viscosity model (Chiang et al., 1991) are (0.35, 36540.44 Pa, 0.0123 Pa s,
5696.05 K, 4.50 10ÿ8 Paÿ1). The major processing conditions are a filling time of
1.0 s, a melt temperature of 200 C, and a mould temperature of 34 C. The flow front
advancements are illustrated in Figure 5. The solid line is the predicted weld line
location, and the dashed line is the test weld line location. It can be seen that the
prediction is in good agreement with the experimental result.
Another example is the front TV panel, as shown in Figure 6. Five gates are marked
by ` '. The predicted weld lines are the solid lines (the spherules represent air traps).
It is found that the prediction agrees well with the fact of the practical production.
5 Conclusion
With the prediction results of injection moulding simulation, a reliable and efficient
algorithm to detect the location of weld lines is presented. After analysing the factors
affecting the properties of weld lines in detail, an evaluation model based on the
artificial neural network method is established, which realises the quantitative
prediction of weld line properties. Finally, comparison with experimental data is
employed for the purpose of verification.
536
Table 1
中国科技论文在线
H. Zhou and D. Li
Acknowledgements
The authors would like to acknowledge financial support from the Specialised
Research Fund for the Doctoral Programme of Higher Education of the People's
Republic of China, under No. 20020487032.
References