A real-time process optimization system for injection molding.

INTRODUCTION

Injection molding is one of the most widely used techniques in the plastic industry, in which process parameters settings and optimization are recognized as important approaches to improve the quality of the molded parts. Many experimental works were carried out to investigate the influence of the injection molding parameters on the quality of molded parts and the occurrence of molding defects (1-6). The experimental findings show that the molding parameters have a significant effect on the quality of molded parts. For example, increasing both the packing pressure and packing time could reduce the sink marks, while decreasing the injection speed could eliminate the flow marks for the case of amorphous polymer.

However, the optimization of process parameters is a complex and difficult task, because it depends on many factors, such as the molding material, the part geometry, the mold structure, and the molding machine. Traditionally, process parameters of injection molding are determined by experienced molding personnel. Setting these parameters is a highly skilled job and based on the skilled operator's "know-how" and intuitive sense acquired through long-term experience rather than through a theoretical and analytical approach (7). Faced with global competition in injection molding industry, using this experience approach to determine the process parameters for injection molding is no longer good enough. Quite a few researchers have attempted various approaches to facilitating injection molding process setups to reduce the time to market and obtain consistent quality of molded parts (7-9).

Based on the modeling features, the existing approaches can be classified into two categories: hard computing and soft computing. The hard computing mainly involves numerical simulation. The numerical simulation of injection molding dates from the 1950s (10), (11). More practically valuable models started in the 1970s. After that, three sequential kinds of models--the mid-plane models, surface models, and solid models--were presented (12-15). Conventional numerical simulation could provide useful information in process design of injection molding. The process behavior predicted by the numerical simulation can help novice engineers overcome the lack of previous experience and assist experienced engineers in pinpointing factors that may otherwise be overlooked. Through the simulation packages, a number of process parameters for injection molding can be obtained (7). However, this approach involves creating a finite element model and running a number of simulations to obtain the acceptable molding parameters. As the time for running a simulation of a plastic part with a moderate complexity, such as the casing of mobile phones, could take an hour or more, it may not be practical to perform a number of conventional simulations in the real-time shop-floor production environment.

On the other hand, many researchers paid attention to another kind of computing technique, namely soft computing (16). The soft computing techniques used for determining process parameters of injection molding include expert systems, artificial neural networks (ANN), fuzzy logic, case-based reasoning (CBR), genetic algorithms (GA), evolutionary strategies, etc. (16-21). For example, Kwong (18) adopted the CBR technique to develop a case-based system for process design of injection molding, which aimed to derive a process solution for injection molding quickly and easily without relying on the experienced molding personnel. Chen et al. (19) transformed expert experiences into if-then rules to establish the knowledge base, which was applied to control the weld line position in the injection molding process. Chen et al. (20) developed an artificial neural network-based in-process mixed material-caused flash monitoring system (ANN-IPMFM) in the injection molding process. This proposed system integrated two subsystems: a vibration monitoring subsystem and a threshold prediction subsystem. And Zhu and Chen (21) developed a fuzzy neural network-based in-process mixed material-caused flash prediction system for injection molding processes. In these approaches, the effectiveness of the ANN, GA, and CBR techniques is dependent on many samples or cases, which decreases their practicability. The reason is that obtaining samples or cases needs molding trials. For the expert system technique, most experience of molding personnel may be found difficult to be codified into a set of explicit rules. Another problem is how to obtain a starting point.

In this article, an intelligent system combining a simplified simulation model and fuzzy inference is constructed for determining the process parameters for injection molding. The simplified simulation model is presented to quickly predict the melt temperature and pressure. This model is adopted to obtain the initial parameters by using a preliminary optimization. Next, fuzzy inference is used to adjust process parameters to correct for defects during the molding trials. The developed intelligent system integrates with the molding machine controller directly and can be used real time.

INTELLIGENT MODELING

The process determination procedure is illustrated in Fig. 1, using the injection pressure and injection temperature as examples. There is a feasible process zone for injection molding that is always referred as a process window. The process window is represented by a set of boundaries that define a window-like shape, as shown in Fig. 1. If the melt temperature is too low, higher injection pressure is required to deliver the melt polymer into the cavities. If the melt temperature is too high, material degradation may occur. On the other hand, if the injection pressure is too low, a short shot may be resulted. If the injection pressure is too high, flash may occur (7). Usually, the optimal setting of process parameters is the center of the process windows. However, the development of a full set of process windows would be very difficult because of the large number of process parameters involved and the time-consuming and costly process of test molding runs.

[FIGURE 1 OMITTED]

During molding trials, the personnel will set up an initial process point based on his experience, called [T.sub.0]. He or she will run a molding trial at that point and usually will encounter some molding defects. Accordingly, the personnel will adjust the processing parameters to a new point (called [T.sub.1]) and start another molding trial. The above adjustment will be repeated until the molding trial is fully successful, without any defect. Then, an optimal process point is obtained, call [T.sub.n]. Generally, this optimization procedure can be divided into two steps: initial process parameter setting and defect correction.

Corresponding to the above two steps, the proposed intelligent system mainly consists of the initial process parameter settings and real-time parameter corrections, as shown in Fig. 2. The initial parameters are first determined by a preliminary optimization program based on a simplified simulation model. Next, these initial parameters will be uploaded to the molding machine directly through the communication function and used for molding trials. The defects found during the molding trials will be fed back to the system by the molding personnel in a natural language. Then, the fuzzy inference module will produce adjusted process parameters accordingly, and these adjusted parameters are uploaded to the molding machine again for the next molding trial. The aforementioned adjustment procedure will be repeated until the quality of the molded part is found satisfactory; that is, a set of optimal or near-optimal process parameters are obtained.

[FIGURE 2 OMITTED]

INITIAL SETTING

The initial process parameter settings aim to find a starting point. Considering that many parameters, such as the injection temperature, mold temperature, etc., depend mainly on the material, a material database including the recommended parameters is used to initialize these parameters. Some other parameters, such as cooling time, packing time, etc., can be decided by empirical models. However, injection speed (injection time) and injection pressure are two important parameters that cannot be easily decided. A preliminary optimization model is established to obtain these parameters. The optimization objective is a lower required injection pressure, lower temperature difference, and a shorter injection time, written as

F(P,[DELTA]T,t) = [w.sub.1] X P + [w.sub.2] X [DELTA]T + [w.sub.3] X t (1)

where F is the objective function, P, [DELTA]T, t are the injection pressure, melt temperature difference at the end of filling, injection time, respectively, with [w.sub.1], [w.sub.2], [w.sub.3] being the corresponding weight values.

A GA is used to reach the optimization objective. The key problem during optimization is how to calculate the required injection pressure and melt temperature based on a given condition (including a given injection time). Conventional numerical simulation packages (CAE) have had success in predicting filling behaviors in extremely complicated geometries, but they are so time-consuming that they cannot be used real time. Thus, a simplified simulation model is used in this article to enable predicting the required pressure and temperature very quickly. This model is based on a geometric approximation of the original part by a rectangular edge-gated plate (22). For a given geometry and gate locations, the flow length L and averaged thickness 2b can be calculated. And the approximation is done, so that the length, thickness, and volume of the rectangular plate equal the maximum flow length, averaged thickness, and volume of the part, respectively. The approximated rectangular edge-gated plate can be illustrated in Fig. 3.

[FIGURE 3 OMITTED]

The basic assumptions for the simplified simulation are (a) the material is incompressible and purely viscous; (b) inertia is neglected as compared with the viscous force; (c) the velocity components other than that in the main flow direction (x) are negligible; (d) longitudinal thermal conduction (in the x and y directions) and transverse convection (in the y and z directions) are neglected; and (e) the melt density, thermal conductivity, and specific heat are assumed constant. With these approximations, the governing equations can be written as

[[partial derivative]/[[partial derivative]z]]([eta][[partial derivative]u]/[[partial derivative]z]) - [[[partial derivative]P]/[[partial derivative]x]] = 0 (2a)

[rho][C.sub.p]([[partial derivative]T]/[[partial derivative]t] + u[[partial derivative]T]/[[partial derivative]x]) = K[[[[partial derivative].sup.2]T]/[[partial derivative][z.sup.2]]] + [eta][([[partial derivative]u]/[[partial derivative]z]).sup.2] (2b)

where u is the velocity component in the x direction; P, T, t are the pressure, temperature, and time, respectively; and [eta], [rho], [C.sub.p], and K denote the viscosity, density, specific heat, and thermal conductivity, respectively.

The volume flow rate Q corresponding to the velocity is

Q = W[[integral].sub.[ - b].sup.b] udz (3)

where W is the width of the plate.

In addition, boundary conditions are described as

T = [T.sub.w], u = 0 at z = [+ or -]b; (4a)

[[[partial derivative]t]/[[partial derivative]z]] = [[[partial derivative]u]/[[partial derivative]z]] = 0, at z = 0; and (4b)

T = [T.sub.0] at x = 0 (4c)

where [T.sub.w] and [T.sub.0] are the constant wall and injection temperature, respectively.

According to the above governing equations and boundary conditions, the fluidity S, pressure gradient [LAMBDA], shear rate [gamma], and velocity u can be expressed as

S = 2[[integral].sub.0.sup.b][[z.sup.2]/[eta]]dz (5)

[LAMBDA] = - [[[partial derivative]P]/[[partial derivative]x]] = [Q/[WS]] (6)

[gamma] = [LAMBDA][1/[eta]]z (7)

u = [[integral].sub.z.sup.b][gamma]([bar.z])d[bar.z] (8)

The Cross-WLF viscosity model has been used for the non-newtonian material, as

[eta] = [[[[eta].sub.0](T, P)]/[1 + [[([[eta].sub.0][gamma]/[tau]*)].sup.[1 - n]]]] (9a)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (9b)

where n, [tau]*,[A.sub.1], [A.sub.2], [D.sub.1], [D.sub.2], [D.sub.3] are material constants.

Equations 2-9 form a set of relationships between velocity, temperature, pressure, and viscosity, which clearly cannot be solved analytically but numerically by finite difference method. The finite difference grid in the half-gap thickness is sketched in Fig. 4. Let [T.sub.[i,j,k]] =T([x.sub.i],[z.sub.j], [t.sub.k]) denote the temperature of grid node ([x.sub.i], [z.sub.j],) at time step [t.sub.k], and it follows that

[FIGURE 4 OMITTED]

[[[partial derivative]T]/[[partial derivative]t]] = [[[T.sub.[i,j,k + 1]] - [T.sub.i,j,k]]/[[DELTA]t]] (10)

[[[partial derivative]T]/[[partial derivative]x]] = [[[T.sub.[i,j,k + 1]] - [T.sub.[i - 1,j,k + 1]]]/[[DELTA]x]] (11)

[[[[partial derivative].sup.2]T]/[[partial derivative][z.sup.2]]] = [[[T.sub.[i,j + 1,k + 1]] - 2[T.sub.[i,j,k + 1]] + [T.sub.[i,j - 1,k + 1]]]/[[DELTA][z.sup.2]]] (12)

Substituting Eqs. 10-12 into Eq. 2b results in

[[[T.sub.[i,j,k + 1]] - [T.sub.i,j,k]]/[[DELTA]t]] = [1/[[rho][C.sub.p]]](K[[[T.sub.[i,j + 1,k + 1]] - 2[T.sub.[i,j,k + 1]] + [T.sub.[i,j - 1,k + 1]]]/[[DELTA][z.sup.2]]] + [[eta].sub.[i,j,k + 1]][[gamma].sub.[i,j,k + 1].sup.2]) - ([[u.sub.[i,j,k + 1]] + [u.sub.[i - 1,j,k + 1]]]/2)([[T.sub.[i,j,k + 1]] - [T.sub.[i - 1,j,k + 1]]]/[[DELTA]x]) (13)

The computation procedures are (1) based on the simulated temperature distribution at [t.sub.k], obtain [eta] by Eq. 9; (2) calculate the fluidity S by integral of Eq. 5; (3) calculate A and pressure P by Eq. 6; (4) calculate [gamma] and u by Eqs. 7 and 8, respectively; (5) calculate the temperature distribution at step [t.sub.[k+1]] based on Eq. 13; and (6) repeat above steps until the filling is complete.

Figure 5 shows the simulation results of the presented simplified model for a mouse cover, in comparison with those of the detailed CAE analysis. Generally, the simulation results of this simplified model agree well with those of the detailed CAE analysis, especially qualitatively, whereas it takes <1000th of the time of the detailed CAE analysis, only 0.03 s. Therefore, it is very suitable in the preliminary optimization stage for determining initial process parameter settings.

[FIGURE 5 OMITTED]

DEFECTS CORRECTION

Basic Idea and Framework

After obtaining the initial process parameters, the molding personnel will run the molding trial on the molding machine. Generally, he or she would encounter several defects, called [D.sub.i](i = 1,2, ..., m). Correcting each defect [D.sub.i] would require adjusting more than one process parameter called [P.sub.j](j = 1,2, ..., n). The adjustment ratios of these parameters relate to the defects and the current values of these parameters. Accordingly, the fuzzy inference system considers defects and current values of process parameters as its solution conditions and the adjustment ratios of the process parameters as its solution results. The fuzzy system deals with the defects one by one, and each defect will result in adjustments to its related process parameters. At last, the adjustments for the same parameter are integrated into its final adjustment by the method of composition. The main framework of the whole fuzzy inference system is shown in Fig. 6.

[FIGURE 6 OMITTED]

The adjustment of one of the relative process parameters [P.sub.j], which corresponds to a certain defect [D.sub.i], is written as [DELTA][P.sub.ji], which is composed of the adjustment direction and ratio. The former is decided by the type of [D.sub.i], and [P.sub.j], and can be obtained easily by traditional rule-based reasoning. The latter is decided by the seriousness of [D.sub.i] and the current value of [P.sub.j], and can be obtained based on the fuzzy inference.

As mention earlier, the system deals with the defects one at a time. For a given defect, [D.sub.i] its related process parameters are selected first, based on the expert knowledge, experience, and the sensibility of material properties to different process parameters. After that, the corresponding sets of fuzzy rules about defect [D.sub.i], and one of the selected parameters [P.sub.j] are then pushed into a fuzzy subsystem (namely, "adjustor ji" in Fig. 6) by the fuzzy rule assigner. This subsystem will be used to compute [DELTA][P.sub.ji]. Each subsystem has a different set of fuzzy rules but shares the same inference model. The adjustments by different fuzzy subsystems are finally integrated by the procedure of composition. Figure 7 shows the structure of the fuzzy subsystem, where the data dictionary contains the definitions of the membership functions used in the fuzzy system.

[FIGURE 7 OMITTED]

Fuzzy Rules

As mentioned earlier, the fuzzy subsystem is used to obtain the adjustment ratio of a certain process parameter [D.sub.i] that results from a certain defect [D.sub.i]. The adjustment ratio is selected as the output of fuzzy inference engine, whereas the seriousness of [D.sub.i] and the current value of [P.sub.j] are treated as the input. Two input modes of the defect seriousness--namely, the natural language and numerical values--are adopted in the fuzzy inference engine. The natural language mode takes values from the linguistic terms, which are composed of the primary linguistic set {slight, medium, serious} and an additional linguistic set {very slight, a little serious, very serious}, whereas the numerical value mode takes values from 0 to 1.

The linguistic rules derived from experts and/or experiments for defects correction are converted to fuzzy rules that are described as where [R.sup.(i)] is a fuzzy if-then rule; the antecedents x and y are linguistic variables of the seriousness of the defect and the current degree of the process parameter, respectively; the consequence z is the linguistic variable of the degree of adjustment; and A, B, and C are the corresponding linguistic values of x, y, and z, which take values from the linguistic term sets {serious, medium, slight}, {high, mid, low}, and (significant, great, big, moderate, small, minor, tiny}, respectively.

According to the above linguistic values taken by A, B, and C, the formation of fuzzy rules about a certain process parameter (e.g., injection pressure) adjusted by a certain defect (e.g., short shot) is illustrated in Table 1.

TABLE 1. Fuzzy rules about injection pressure modified by short shot.

                                        Fuzzy if-then rule

Defect  Process    Adjustment  Seriousness   Current    Degree of
        parameter  direction   of defect    degree of  adjustment
                                               the
                                             process
                                            parameter

                               Serious        Low      Significant
                               Serious        Mid      Great
                               Serious        High     Big
Short   Injection              Medium         Low      Big
                   Increase    Medium         Mid      Moderate
Shot    Pressure               Medium         High     Small
                               Slight         Low      Small
                               Slight         Mid      Minor
                               Slight         High     Tiny

[R.sup.(i)]: if x is A and y is B then z is C.

The parameterized triangular membership function (23) is used to fuzzify the linguistic values of each linguistic variable, including the linguistic values in fuzzy if-then rules and input-output parameters, which can be expressed as

triangle(u,a,b,c) = max(min([u - a]/[b - a],[c - u]/[c - b]),0) (14)

Conforming to human thinking, the interval (0,1) is defined as the universe of discourse of the linguistic variables. For example, the membership functions of the primary linguistic values of A' ("the seriousness of defect") can be given as

[[mu].sub.slight](x) = triangle(x, - 0.5,0,0.5) (15a)

[[mu].sub.medium](x) = triangle(x,0,0.5,1.0) (15b)

[[mu].sub.serious](x) = triangle(x,0.5,1.0,1.5) (15c)

The input parameters (the degree of current value of the process parameter), and output parameters (the degree of adjustment) are normalized to the interval (0, 1). The input values are determined according to its range of possibilities that is estimated by the simplified simulation model. The absolute output values are determined according to the permitted adjustment range, which is also estimated by the simplified simulation model.

Fuzzy Inference

The mode of "Multiple Rules with Multiple Antecedents" and the Mamdani fuzzy model are used in the fuzzy inference engine (23), with the block diagram shown in Fig. 8.

[FIGURE 8 OMITTED]

In this model, C' is the linguistic value of the final adjustment of the process parameter, z is the crisp output, x = (A' or [x.sub.0]) and y = [y.sub.0] are facts and the input of fuzzy model. For the mode of crisp input of [x.sub.0] and [y.sub.0], x = [x.sub.0], and y = [y.sub.0] can be converted to the fuzzy singletons A' = {l.0/[x.sub.0]} and B' = {1.0/[y.sub.0]}, respectively. Therefore, two modes of fuzzy and crisp input can be uniformed to the mode of fuzzy input.

Every fuzzy rule can be transformed into a ternary fuzzy relation [R.sub.k] based on Mamdani's fuzzy implication function. If maximum and minimum are adopted as T-norm and T-conorm operators, respectively, [R.sub.k] can be expressed as

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (16)

Thus,

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (17)

It follows that

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (18)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (19)

where [[omega].sub.xk] and [[omega].sub.yk] are the maxima of the membership functions of [A.sub.k] [intersection] A' and [B.sub.k] [intersection] B', respectively. In general, [[omega].sub.xk], denotes the degree of compatibility between [A.sub.k] and A'; similarly for [[omega].sub.yk]. Specially, if A' = {[1.0/[x.sub.o]]}, [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and if B'= {[1.0/[y.sub.0]]}, [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] Because the antecedent part of the fuzzy rule is constructed by the connective "and," [[omega].sub.xk] [conjunction] [[omega].sub.yk] is called the firing strength of the fuzzy rule, which represents the degree to which the antecedent part of the rule is satisfied. The MF of the resulting [C'.sub.k] is equal to the membership function of [C.sub.k] clipped by the firing strength [[omega].sub.k], [[omega].sub.k] = [[omega].sub.xk] [conjunction] [[omega].sub.yk].

For multiple rules, there is

C' = (A' X B')[??][r.[union] (k=1)][R.sub.k] = [r.[union] (k=1)][(A' X B')[??][R.sub.r]] = [r.[union] (k=1)][C'.sub.k] (20)

And therefore,

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (21)

That is, a fuzzy set that expresses the adjustment degree can be obtained by aggregating the reasoning results of each fuzzy rule and can be defuzzified to a crisp value.

Composition

Several types of defects are often encountered simultaneously during a molding test run. The correction of each defect involves more than one process parameter. A conflict would appear if several defects adjust the same process parameter in opposite directions or in the same direction but to different degrees.

According to the molding experience, the strategy of "taking the maximal adjustment" is adopted when a process parameter is adjusted by two different defects in the same direction. That is, the maximal adjustment ratio is the final decision. When a process parameter is adjusted by two different defects in opposite directions, the problem is resolved based on each defect's priority in the knowledge base. Their priorities decide the order of defect correction. That is, the final adjustment is the one resulted from the defect with a higher priority.

INTEGRATION WITH THE MOLDING MACHINE

Nowadays, the controller of the injection molding machine usually provides a standard Transmission Control Protocol/Internet Protocol (TCP/IP) communication interface and the addresses of all the data in its memory. The developed intelligent system integrates with the molding machine controller through this communication interface.

The system can upload the resulted data to the controller automatically for molding trials and can download the current data of all the process parameters from the controller. Based on the standard TCP/IP protocol and Microsoft Visual C++, the communication function with the controller is developed by using Winsock (24).

In this article, the platform of the injection machine is HAITAI 3800A with an EST controller P10CY. The developed apparatus and its interface are shown in Fig. 9.

[FIGURE 9 OMITTED]

CASE STUDY

For the verification of the developed system, two parts in the practical production were used as examples, as shown in Table 2. One was a cover with a single-cavity mold and four gates; the other was a flashboard with a two-cavity mold that was treated as an integrated part. The molding material and calculated cavity data are listed in Table 2.

TABLE 2. The molding material and basic geometrical data of the cavity.

         Properly                  The cover         The flashboard

Cavity geometry              [??]                  [??]
Molding material             ABS                   PP
Overall size (mm X mm X mm)  335.3 X 481.0 X 37.2  162.0 X 71.5 X 22.8
Cavity volume ([mm.sup.3])   431.6 X [10.sup.3]    13.9 X [10.sup.3]
Flow length (mm)             274                   180.55
Average thickness (mm)       2.1                   1.7

[ILLUSTRATION OMITTED]

For the cover, the main initial process parameters were determined by the preliminary optimization and simplified model, as shown in Table 3. These parameters were then uploaded to the injection molding machine for the molding trial. The molded part of the first trial run is shown in Fig. 10a, which had a serious short shot defect. This defect and its degree were fed back to the system by the operator. The process parameters were then adjusted by the fuzzy system for the first time, and these adjusted parameters were uploaded to the injection molding machine for the next molding trial. This time, the degree of short shot lessened, as shown in Fig. 10b. Similarly, the second adjustment and the third molding trial were carried out, in which the short shot became slight, as shown in Fig. 10c. Then, the third adjustment and the fourth molding trial were completed, which resulted in a part without a short shot but that had a flash defect, as shown in Fig. 10d. Next, the fourth adjustment and the fifth molding trial were performed, in which the defect of short shot occurred again, but its degree was very slight, as shown in Fig. 10e. Finally, the fifth adjustment and the sixth molding trial were completed, which resulted in a fully successful part, without any defects, as shown in Fig. 10f. Thus, the optimized process parameters were obtained in a very short time. The main process parameters and their adjustments in these trials are listed in Table 3.

[FIGURE 10 OMITTED]

TABLE 3. The main process parameters and their adjustments in the
molding trials for the cover.

                                                   Adjustments

Process parameters                  Initial   First   Second   Third
                                    setting  (second  (third  (fourth
                                    (first    trial)  trial)   trial)
                                    molding
                                    trial)

Injection temperature ([degrees]C)   225.0     +5.0    +3.0     +4.0
Mold temperature ([degrees]C)         40.0     +3.0    +2.0     +2.0
Injection speed (%)                   25.0    +10.0    +6.0     +7.0
Injection pressure (Mpa)              50.0     +9.0    +5.0     +3.0
Packing pressure (Mpa)                48.0    +10.0    +9.0     +3.0
Packing time (s)                       8.0     +1.0    +0.2     +0.3
Cooling time (s)                      20.0      0.0     0.0      0.0

                                                   Adjustments

Process parameters                  Fourth (fifth  Fifth (sixth   Final
                                        trial)         trial)    values

Injection temperature ([degrees]C)      -6.0           +2.0       233.0
Mold temperature ([degrees]C)           -5.0           +3.0        45.0
Injection speed (%)                    -13.0          +10.0        45.0
Injection pressure (Mpa)               -10.0           +2.0        59.0
Packing pressure (Mpa)                 -12.5           +7.5        65.0
Packing time (s)                        -1.0           +0.5         9.0
Cooling time (s)                         0.0            0.0        20.0

For the flashboard, only three molding trials were needed to lead to a fully successful part. The initial process parameters and their adjustments are listed in Table 4. The defects during the first and second molding trials were a serious short shot and a slight flash, respectively. The molded parts of the trial runs are shown in Fig. 11.

[FIGURE 11 OMITTED]

TABLE 4. The main process parameters and their adjustments in the
molding trials for the dashboard.

Process parameters                    First      Second        Third
                                     molding     molding      molding
                                      trial   trial (first    trial
                                    (initial   adjustment)   (second
                                    setting)                adjustment)

Injection temperature ([degrees]C)   230.0       236.0         233.0
Injection speed (%)                   30.0        40.0          35.0
Injection pressure (Mpa)              15.0        27.0          21.0
Packing pressure (Mpa)                15.0        15.0          15.0
Packing time (s)                      14.0        15.0          15.0
Cooling time (s)                       5.0         5.0           5.0

CONCLUSIONS

In this article, an integrated intelligent system for determining the process parameters for injection molding has been constructed, which combines the advantages of simplified numerical simulation and fuzzy theory. The process determination procedure was divided into two steps: initial process parameter setting and defect correction. These two steps have different features and therefore different solution methods. It is realized that there is no single technique to be used in solving the whole problem. A practical system for parameter setting of injection molding should combine and mutually strengthen the different artificial intelligence technologies.

Conventional numerical simulation cannot be used in the shop-floor production environment because it is too time consuming. The presented simplified simulation model takes <1000th of the time of the conventional numerical simulation and their simulation results agree well qualitatively. So, this simplified model is suitable in the stage of initial process parameter setting. Fuzzy theory would facilitate the practical use of the molding personnel's experience that is usually not very explicit. It should be considered that several defects would be encountered in a molding trial and each defect would relate to several process parameters.

The traditional approach for injection molding process setting is through design of experiments (DOE). Using this simple model, an optimal or suboptimal operating point can be determined. But if many process parameters are involved, large number of experiments are required and the experiments should be carefully designed. Compared with DOE, the developed method needs much less molding trials and is independent on operators' knowledge.

The developed system integrates with the molding machine directly and can be effectively used to set up the initial process parameters and correct them real time, which greatly reduces the trial-and-error cycle of molding as well as dependency on human experts. Despite the good results from the practical application, additional case studies are needed to further evaluate the performance of the intelligent system.

REFERENCES

(1.) T.H. Lee and N.J. Mills, J. Mater. Sci., 29, 2704 (1994).

(2.) Y. Yang and F. Gao, Polym. Eng. Sci., 46, 540 (2006).

(3.) H. Hamada and H. Tsunasawa, J. Appl. Polym. Sci., 60, 353 (1996).

(4.) S.C. Chen, H.S. Peng, L.T. Huang, and M.S. Chung, J. Reinforc. Plast. Compos., 22, 479 (2003).

(5.) J. Pomerleau and B. Sanschagrin, Polym. Eng. Sci., 46, 1275 (2006).

(6.) S. Dowlatshahi, J. Manuf. Technol. Manage., 15, 445 (2004).

(7.) S.L. Mok, C.K. Kwong, and W.S. Lau, Adv. Polym. Technol., 18, 225 (1999).

(8.) Z.B. Chen and L.S. Turng, Adv. Polym. Technol., 24, 165 (2005).

(9.) J. Zhou and L.S. Turng, Polym. Eng. Sci., 47, 684 (2007).

(10.) G.D. Gilmore and R.S. Spencer, Mod. Plast., 37, 143 (1950).

(11.) R.S. Spencer and G.D. Gilmore, J. Appl. Phys., 21, 523 (1950).

(12.) C.A. Hieber and S.F. Shen, J. Nonnewtonian Fluid Mech., 7, 1 (1980).

(13.) H.H. Chiang. C.A. Hieber, and K.K. Wang. Polym. Eng. Sci., 31, 116 (1991).

(14.) R.Y. Chang and W.H. Yang, Int. J. Numer. Methods Fluids, 37, 125 (2001).

(15.) H.M. Zhou and D.Q. Li, Simul. Modell. Pract. Theory, 13, 273 (2005).

(16.) V.R. Delia Julieta, Soft Computing Technologies in Quality Control with Applications to Injection Molding, Ph.D. Thesis, New Mexico State University, New Mexico (2001).

(17.) W. He, Y.F. Zhang, K.S. Lee, J.Y. Fuh, and Y.C. Nee, J. Intell. Manuf., 9, 17 (1998).

(18.) C.K. Kwong, Int. J. Comput. Appl. Technol., 14, 40 (2001).

(19.) M.Y. Chen, H.W. Tzeng, Y.C. Chen, and S.C. Chen, ISA Trans., 47, 119 (2008).

(20.) J. Chen, M. Savage, and J.J. Zhu, Int. J. Adv. Manuf. Technol., 36, 43 (2008).

(21.) J. Zhu and J.C. Chen, Int. J. Adv. Manuf. Technol., 29, 308 (2006).

(22.) B. Mathias, A Model to Predict the Flow Length in Injection Molding, Master Thesis, University of Wisconsin-Madison, Madison (1993).

(23.) J.S.R. Jang, C.T. Sun, and E. Mizutani, Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence, Prentice Hall, Inc., New Jersey (1997).

(24.) D.J. Kruglinski, Inside Visual C++, 4th ed., Microsoft Press, Washington (1997).

Dequn Li, Huamin Zhou, Peng Zhao, Yang Li

State Key Laboratory of Mold and Die Technology, Huazhong University of Science and Technology, Wuhan, Hubei 430074, People's Republic of China

Correspondence to: Huamin Zhou; e-mail: [e-mail]

Contract grant sponsor: The National Natural Science Foundation Council of China; contract grant number: 506750180; contract grant sponsor: Research Fund for the Doctoral Program of Higher Education of China; contract grant number: 20060487056.

DOI 10.1002/pen.21444