Vous êtes sur la page 1sur 6

Materials and Design 31 (2010) 21182123

Contents lists available at ScienceDirect

Materials and Design


journal homepage: www.elsevier.com/locate/matdes

Technical Report

A hybrid of mode-pursuing sampling method and genetic algorithm for minimization of injection molding warpage
Yi-Min Deng a,*, Yong Zhang a, Yee Cheong Lam b
a b

Faculty of Mechanical Engineering and Mechanics, Ningbo University, 818 Fenghua Rd., Ningbo City, Zhejiang Province 315211, PR China School of Mechanical and Aerospace Engineering, Nanyang Technological University, 71 Nanyang Drive, Singapore 639798, Singapore

a r t i c l e

i n f o

a b s t r a c t
This paper presents a hybrid optimization method for minimizing the warpage of injection molded plastic parts. This proposed method combines a mode-pursuing sampling (MPS) method with a conventional global optimization algorithm, i.e. genetic algorithm, to search for the optimal injection molding process parameters. During optimization, Kriging surrogate modeling strategy is also exploited to substitute the computationally intensive Computer-Aided Engineering (CAE) simulation of injection molding process. With the application of genetic algorithm, the likelihood-global optimums are identied; and the MPS method generates and chooses new sample points in the neighborhood of the current likelihood-global optimums. By integrating the two algorithms, a new sampling guidance function is proposed, which can divert the search process towards the relatively unexplored region resulting in less likelihood of being trapped at the local minima. A case study of a food tray plastic part is presented, with the injection time, mold temperature, melt temperature and packing pressure selected as the design variables. This case study demonstrates that the proposed optimization method can effectively reduce the warpage in a computationally efcient manner. 2009 Elsevier Ltd. All rights reserved.

Article history: Received 23 August 2009 Accepted 15 October 2009 Available online 20 October 2009

1. Introduction Plastic materials are commonly used in many areas of industry as they can meet various requirements due to their specic and wide variation of physical and chemical properties. These include lightness, resistance to corrosion, ease of shaping and forming, etc. Injection molding is a major processing route for the production of plastic parts. During the injection molding process, however, a number of defects may occur to the moldings, such as warpage, shrinkage, sink marks and weld lines. These are caused by many factors, which include the settings of the molding process parameters, the mold system, the part geometry as well as the plastic materials [1]. Among these defects, warpage is one of the most common and prominent, affecting both the part usage (function) and the appearance [2,3]. Hence, warpage minimization is one of the most critical considerations for the production of a quality molded part. Thus, it is not surprising that a lot of effort has already been devoted for injection molding warpage minimization. Section 2 will provide an overview of the existing effort, together with its limitations. To facilitate further discussion, a mode-pursuing sampling (MPS) method is introduced in Section 3, together with a brief
* Corresponding author. Tel.: +86 574 87600534; fax: +86 574 87608358. E-mail addresses: dengyimin@nbu.edu.cn, dym1000@hotmail.com (Y.-M. Deng). 0261-3069/$ - see front matter 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.matdes.2009.10.026

description of the Kriging surrogate modeling that will be used in conjunction with the MPS method. Subsequently, a hybrid usage of MPS method and genetic algorithm will be proposed to address the existing limitations. In Section 4, the procedure of the proposed hybrid optimization method is given. Section 5 provides a case study of a food tray plastic part to demonstrate the application of the proposed methodology. This is followed by conclusion in Section 6. 2. Overview of warpage minimization Warpage in injection molding is generally a function of the part geometry, the mold, the design of the runners and gates, and the process parameters. Many studies have already been conducted targeting at warpage minimization. For example, Lee and Kim [2] proposed to reduce warpage by optimizing the part wall thickness and process parameters. Subsequently, they [3] applied a two-step search method to optimize the gate location to improve the part quality, including warpage and some other quality aspects. The research group of Lam [49] had investigated various parameters and methodologies for the reduction of warpage. They show that by cavity balancing [46], gate location and molding condition optimization [7,8], and runner balancing for multi-cavity mold [9], signicant warpage reduction could be achieved. Methodologies employed include ow path optimization, injection pressure

Y.-M. Deng et al. / Materials and Design 31 (2010) 21182123

2119

minimization, and genetic algorithms with pareto optimization strategy. Deng et al. [10] proposed a PSO (particle swarm optimization) algorithm for the optimization of multi-class design variables, such as the part thickness, process parameters (melt temperature, mold temperature, injection time) and gate location, targeting warpage reduction as well as improving other molding qualities. Despite the many achievements in warpage minimization, some practical difculties remain, one of which is the computational cost of the optimization process. Some researchers have attempted to address this problem through experiments, for example, Erzurumlu and Ozcelik [11] combined Taguchi method and Moldow simulation to study the effect of process parameters on part quality, and derived the optimal settings by choosing the ones from the limited set of experimental points derived from the Taguchi method. This kind of method has a limitation in that the search space is limited the obtained optimal process condition is only the best combination of the specied process parameters, not the optimal solution in the entire design space. A different approach is by using surrogate modeling technologies. Based on a quantity of experimental samples, various kinds of surrogate models, such as the polynomial models, radial basis functions (RBF), Kriging models, and support vector regression (SVR) [12] were employed to substitute the expensive injection molding simulation process for global optimization. Hitherto, quite a few researchers have exploited this methodology to optimize the process parameters in injection molding [13,14]. They show that surrogate modeling can be considered as a good approach to reduce intensive computation in injection molding warpage minimization. However, by employing the surrogate modeling approach alone, the optimality of the solution depends largely on the accuracy of the surrogate model. A good model generally requires a large number of sample points, and should be distributed as uniformly as possible in the design space. Unfortunately, these requirements conict with the requirement of minimizing the computational time. Therefore, some stepwise optimization methods are proposed in warpage minimization. For example, Gao and Wang [15] proposed a sequential optimization method based on Kriging surrogate model. This sequential optimization can improve the surrogate model using a current optimal solution until the convergence criteria are satised. However, it is not efcient to update the Kriging approximate model iteratively by only adding the current optimal point to the design set to reconstruct the Kriging surrogate model. Zhou and Turng [16] presented an integrated simulation-based optimization system based on a Gaussian process approach. This method could improve the surrogate models by nding the additional training points with greater variances through the optimization iterations. However, both methods are prone to converge prematurely, i.e. to converge to a local optimum, if it exists. Another approach is to generate directly new sample points towards the optimum with the guidance of a surrogate model. In contrast to the previous two methods, the surrogate model is not used as a surrogate in a typical global optimization process. Optimization is realized by adaptive sampling alone without a formal global optimization process. The surrogate model is used only as a guide for adaptive sampling and therefore the requirement on model accuracy is reduced. Gao and Wang [17] proposed an adaptive sampling method for injection molding optimization based on Kriging surrogate model, where an inlling sampling criterion named as Expected Improvement (EI) was introduced. This criterion can balance the local and global searches, even though the Design of Experiment (DOE) size is small. This method, however, presumes a continuous objective function and a correlation structure between the sample points, with a complicated optimization

process required for the identication of an updated point. In addition to this method, Wang et al. [18] have proposed a global optimization method based on a novel mode-pursuing sampling (MPS) method, which can generate more sample points in the neighborhood of a function mode (the term mode is explained in Section 3.1) while statistically covering the entire search space. In our previous work, we have attempted to apply this method in the determination of the appropriate process parameters for the reduction of injection molding warpage [19]. Our results show that this approach is efcient in comparison to the conventional simulationbased warpage optimization methods. However, there is also a limitation in that the optimization process may sometimes converge to a local optimum before sufcient exploratory points are generated. To further improve warpage optimization in the injection molding process, this investigation presents a hybrid optimization method. A hybrid of MPS method and genetic algorithm is proposed to minimize warpage effectively and efciently. Kriging surrogate modeling will be exploited to substitute the expensive injection molding CAE simulation analyses. To facilitate discussion, MPS and Kriging surrogate modeling will rst be introduced. 3. Introduction of MPS and Kriging surrogate modeling 3.1. The MPS method The mode-pursuing sampling (MPS) method was proposed by Wang et al. [18], which is an extension of the random-discretization based sampling method of Fu and Wang [20], a generalpurpose algorithm to draw a random sample from any given multivariate probability distribution. The word mode refers to a minimum (either local or global) of the objective function. Suppose an n-dimensional function f(x) is to be minimized over a compact set S(f) = [a, b]n (a < b), f x P 0 and it is continuous in the neighborhood of the global minimum, the MPS method may be summarized into the following steps: Step 1: Generate m count of uniformly distributed points x(i), i = 1, . . ., m on S(f) and calculate their function values f(x(i)). Step 2: Formulate an approximation function Sp(x) based on the above m points, for example, a linear spline function Sp(x) can be formulated as

Sp x

X
i

ai kx xi k
1 i 1; . . . ; m

subject to Sp xi f xi ;

A non-negative function g(x) can then be constructed by dening gx C 0 Sp x P 0 on S(f) = [a, b]n, where C0 is a constant to make g(x) non-negative. Minimizing Sp(x) is equivalent to maximizing g(x), hence g(x) can be viewed as the probability density function to be used in the next step. Step 3: By applying the sampling algorithms of Fu and Wang [20], another m count of random sample points, denoted as y(i), i = 1, . . ., m, are selected from S(f) = [a, b]n according to g(x) (hence it is referred to as the sampling guidance function). These points tend to gather around the current maximum of g(x), which corresponds to the minimum of Sp(x). Step 4: Evaluate the derived sampling points and their corresponding objective function against the specied termination criterion. If the criterion is met, the search process is terminated; otherwise, combine the new m points, y(i), i = 1, . . ., m, with the old points x(i), i = 1, . . ., m, to obtain a new x = [x, y], then repeat steps 24. In step 3, to better control the sampling process (e.g. whether to generate more exploratory points or local points), a speed

2120

Y.-M. Deng et al. / Materials and Design 31 (2010) 21182123

control factor c can be introduced, details of which can be found in Wang et al. [18]. 3.2. Kriging surrogate modeling Kriging surrogate modeling is one of the methods for building a surrogate model of a full-edged process such as a CAE simulation process. In general, the Kriging models are more accurate for nonlinear problems, hence are suitable for the calculation of injection molding warpage to be addressed in this investigation. Given a set of m count of design samples S = [x1, x2, . . ., xm] xi = Rn and the responses Y = [y1, y2, . . ., ym]T (yi = Rq), we can construct a Kriging model as follows:

(2) Since the evaluation of the objective function requires intensive computation (by executing CAE simulation), an optimization algorithm that demands a large number of evaluations of the objective function (which often involves an iterative process) would be a computationally expensive process. These characteristics are especially suited for utilizing the MPS and surrogate modeling method, and will be further elaborated in the following section. 4.2. Hybrid optimization strategy As pointed out in Section 3, when applying the MPS method, more sample points will be generated around the current minimum point as the MPS progresses. This will increase the chances of nding a better minimum. However, a key issue to be addressed is the generation and selection of the sample points so as to avoid the optimization process from being trapped to a local minimum. According to Torn and Zilinskas [21], an optimization method must be able to probe the part of the search space that has been relatively unexplored, if it is to converge to the global optimum. Comparing with other surrogate modeling methods, Kriging modeling allows us to compute a measure of the possible error in the surrogate model by its mean squared error (MSE). The MSE goes to zero at the sample points, indicating that there is no uncertainty at the sample points. In between the sample points, the MSE rises. Intuitively, the further we are away from the nearest sample point, the more uncertain we are about the function value, and the higher is the MSE. As MSE provides a measure of uncertainty of the surrogate away from the sample points, a search method could be developed by combining the MSE measure with the sampling characteristics provided by the MPS method. This can be achieved easily by intentionally generate and select more sample points where the MSE is high, which represents relatively unexplored space. Later sampling should generate more points around those with small function values, yet are away from the current likelihood-global optimum, where likelihood-global optimum refers to the current optimum based on the current set of sample points and the current surrogate model. To materialize the above strategies, we propose to construct a new sampling guidance function, which is derived by adding two existing functions: the normal sampling guidance function, e.g. the linear spline function discussed in Section 3, and the newly proposed MSE function in the Kriging model. Normalization will have to be carried out when formulating the new guidance function. With this new guidance function, the conventional MPS method is to be modied. As the genetic algorithm is also employed to identify the likelihood-global optimum, hence the proposed method is a hybrid MPS/GA method. The detailed procedure of the MPS/GA hybrid algorithm could be given as follows: 4.2.1. Nomenclature (for those not explained in the algorithm) S_mode = 1: indicating a likelihood-global optimum has been found. S_mode = 0: indicating a likelihood-global optimum is yet to be found. Sp(x): normal sampling guidance function dened in Section 3.1. Sm(x): sampling guidance function dened by MSE function in Kriging model. Spm(x): the new sampling guidance function dened as Spm(x) = Sp(x) + Sm(x). Input. m, the number of design variables; X, design variable vector; S, [Xl, Xu]: compact set of X.

^ yx Fb; x zx f T xb zx

F(b, x) is a regression model which is generally a polynomial function. The coefcient b is a regression parameter to be estimated, z(x) is a random function, which has the following statistical characteristics:

Ezx 0 Varzx r2 Cov zxi ; zxj r Rh; xi ; xj


2

where xi, xj represent two design samples in the given set. R(h, xi, xj) is the correlation model with parameter h, where a Gauss function is generally used, which is dened as,

 Xn  dv Rh; xi ; xj exp k1 hk jxk xk j2 i j

ndv is the number of all design variables, xk , xk correspond to the kth i j component of the samples xi and xj, respectively. The parameters hk, 2 r , b could be determined via the maximum likelihood estimation (refer to references, e.g. Ref. [15], for details). 4. MPS/GA hybrid method for warpage minimization In this investigation, we employ the commercial injection molding CAE software Moldow to simulate the injection molding process and to retrieve warpage deections from the simulation results. Moldow can report the warpage deections (in the X, Y and Z directions) at each node of the part mesh model. In practice, the design engineers may only be interested in the overall deection or deection in a particular direction. Therefore, for ease of illustration but without loss of generality, we opt to optimize the warpage deection in only one specic dimension. The same methodology could easily be extended for all three directions. 4.1. The mathematical model The injection molding warpage minimization problem can be stated as follows:

Find X Minimize WX Subject to X l 6 X 6 X u 5

where X = {x1, x2, . . ., xm} is a vector of injection molding process parameters, and Xl, Xu are the lower and upper bounds of X respectively, W(X) is the maximum value of warpage deection. This optimization problem has the following characteristics [19]: (1) The objective function is not represented by an explicit mathematical expression (hence is a so-called black-box function). It may only be evaluated by retrieving results from Moldow or other CAE simulation.

Y.-M. Deng et al. / Materials and Design 31 (2010) 21182123

2121

Output. Xbest, the obtained global optimum; Wbest, the objective function value at Xbest; nit, the number of optimization iteration. 4.2.2. Algorithm procedure (1) Project the compact set S:[Xl, Xu] to S0 :[0, 1]m, randomly sample [(m + 1)(m + 2)/2] points; conduct CAE (i.e. Moldow) simulations to evaluate their corresponding objective function values (i.e. warpage deections), set nit = 1 and S_mode = 0. (2) If S_mode = 0, generate m sample points by applying the MPS algorithm based on the normal sampling guidance function Sp(x); otherwise, generate m/2 sample points by applying the MPS algorithm based on the sampling guidance function Sm(x) and Spm(x), respectively. Then evaluate their corresponding objective function values, and add them into the sample point set. (3) Construct a Kriging surrogate model based on the current point set, then apply GA to the surrogate model to determine the likelihood-global optimum Xg and the corresponding approxi^ mate function value WX g . Conduct CAE simulation to obtain the corresponding objective function value W(Xg) under the derived optimal condition (Xg), add (Xg, W(Xg))to the point set. (4) Determine the current optimum including Xbest and Wbest; If nit P 2, then evaluate the distance dis, between the current optimal Xbest and the last optimal X 0best , and the difference Df, between the current optimal function value Wbest and the optimal function value W 0best in the last iteration; as well as the absolute difference ^ between W(Xg) and WX g indicated by Dg. (5) If dis 6 ed ; Df 6 ef and Dg 6 eg , set S_mode = 1 and take (Xbest, Wbest) as a likelihood-global optimum; otherwise, set S_mode = 0. (6) Finally, if the convergence criterion is satised, terminate the program with all the output; otherwise, set nit = nit + 1 and go back to step 2. Note that in step 5, ef, eg should be specied according to the accuracy requirement of the function value. ed is normally set to a small value, such as 104. These parameters control the convergence speed of the algorithm, hence are referred to as the control parameters. The convergence criterion in step 6 may be a function of the last three likelihood-global optimums when there is no obvious difference between them, the solution could be considered as converged. 5. Case study 5.1. Problem description To demonstrate the proposed hybrid optimization approach, a food tray plastic part is used as a case study. Fig. 1 contains its geometry shape and its CAE analysis model in Moldow, with its overall external dimensions being 180 90 50 mm, and a maximum part thickness of 3 mm. The material employed is LDPE, manufactured by Eastman Chemical Products, and its properties are given in Table 1. Its warpage is quantied by the sum of the absolute value of the maximum upward deection and the maximum downward deection of all nodes of the part mesh model in the Z direction (only a specic direction is considered, as explained in the beginning of Section 4). We select four key process parameters as the design variables for warpage minimization, namely injection time (tinj), mold temperature (Tmd), melt temperature (Tmt) and packing pressure (Pp). For validation of the algorithm, we intentionally set a relatively large search space, making the initial ranges of the design variables larger than those usually encountered in practice, but avoiding short shot defect. The range of the mold temperature is based on the recommended values in Moldow for the specied part mate-

Fig. 1. CAE food tray model.

Table 1 Material properties. Melt density Solid density Eject temperature Maximum shear stress Maximum shear rate Thermal conductivity Elastic module Poissons ratio 0.73537 g/cm3 0.94781 g/cm3 80 C 0.11 MPa 40,000 (1/s) 0.31 W/m C 124 MPa 0.41

rial. The lower bound of the melt temperature range is 10 C higher than the minimum values recommended by Moldow to avoid short shot. The range of injection time is selected according to the recommended value by the Molding Windows analysis module from the Moldow software. Further, the packing prole is considered as a constant pressure process, with the packing time set as 5 s. The range of packing pressure is usually specied based on the experience of the manufacturer, where the percentage of maximum machine pressure is used to measure the packing pressure. For this case study, these ranges are summarized below:     Mold temperature (C): 2070. Melt temperature (C): 180280. Injection time (s): 0.40.8. Packing pressure (%): 7090.

With these specied ranges, the optimization problem can be formulated as:

Find X t inj ; T md ; T mt ; Pp Minimum WX Subject to 0:4 6 t inj 6 0:8 20 6 T md 6 70 180 6 T mt 6 280 70 6 Pp 6 90

5.2. Implementation of optimization procedure During each optimization iteration, a Kriging surrogate model has to be built based on the current sample points. We choose quadratic function for the regression model and Gauss function for the correlation model, as discussed in Section 3. In the GA optimization process, the commonly used GA operation parameters are adopted, namely the population size, the

2122

Y.-M. Deng et al. / Materials and Design 31 (2010) 21182123

Table 2 Optimized results (10 runs of MPS/GA hybrid algorithm). Run no. Injection time (tinj/s) Mold temperature (Tmd/C) Melt temperature (Tmt/C) Packing pressure (Pp/%) Minimized warpage (W/mm) No. of CAE simulations 1 2 3 4 5 6 7 8 9 10 0.74 0.74 0.795 0.786 0.785 0.80 0.79 0.79 0.80 0.78 68.64 68.60 67.00 69.33 69.57 69.06 69.66 69.43 68.90 69.45 279.17 279.32 279.13 279.60 277.64 280.00 279.12 278.96 377.23 280.00 71.33 70.23 71.30 71.45 70.04 70.34 71.18 70.00 72.00 70.47 1.0808 0.9506 1.0343 0.9465 0.8742 0.8088 0.9006 0.8145 1.1196 0.8287 135 135 105 125 115 105 100 135 115 100

Fig. 3. Minimized warpage obtained by the proposed MPS/GA method. Fig. 2. Warpage deection using Moldow recommended processing parameters.

crossover rate, the mutation rate and the generation size are set as 20, 0.9, 0.005 and 100, respectively. The control parameters of the MPS/GA hybrid algorithm are set as follows:

Table 3 Some data generated from the MPS/GA optimization process. All sample points and corresponding warpage values (mm) once a likelihoodglobal optimum was found [0.77 [0.72 [0.67 [0.78 [0.60 [0.79 [0.44 [0.42 [0.40 [0.40 [0.74 [0.79 [0.41 [0.40 [0.47 [0.41 [0.42 [0.42 [0.40 [0.40 67.89 23.58 22.49 66.13 69.74 28.25 22.13 69.25 67.86 67.50 49.58 69.19 67.47 20.17 20.06 20.19 31.78 21.18 31.53 22.56 266.67 84.19]/3.9136 267.22 78.96]/4.8916 213.02 86.83]/2.2127 183.02 70.04]/4.0855 275.90 89.26]/4.5418 275.95 71.40]/5.2846 264.81 70.74]/5.0052 185.62 70.27]/5.0425 180.67 70.67]/4.7556 183.90 89.05]/5.0085 235.36 70.25]/5.0172 180.05 86.14]/4.1236 278.64 89.39]/6.8916 277.77 70.87]/5.2002 279.51 70.21]/5.2003 187.30 70.37]/1.5847 180.34 78.20]/1.6956 271.87 89.78]/4.9075 278.05 89.85]/5.3918 184.14 70.76]/1.4631 All the likelihood-global optimums [0.40 20.01 189.09 89.89]/1.4743 [0.70 27.98 187.80 89.15]/1.3705 [0.79 69.52 277.54 70.67]/0.9382 [0.79 69.52 277.54 70.67]/0.9382 [0.79 69.52 277.54 70.67]/0.9382 [0.77 68.55 278.58 70.08]/0.8934 [0.78 69.45 280.00 70.47]/0.8287

ed 0:0001; ef 0:01; eg 0:01


Since GA is probabilistic, we ran the proposed MPS/GA optimization algorithm 10 times. The optimization results produced from all these runs are shown in Table 2. To validate and compare the optimization results shown in Table 2, we also obtain the benchmark warpage value by employing the Moldow recommended values for the processing parameters in the simulation, namely:

t inj 0:66;

T md 23;

T mt 240;

P p 80%

The corresponding warpage, as shown in Fig. 2, is

WXrec 2:03 mm:


5.3. Discussion Results in Table 2 shows that,  The proposed MPS/GA hybrid method can reduce warpage with a limited number of CAE simulation runs (which largely determine the amount of computation of the optimization process). Considering that a general simulation-based optimization process could easily require thousands of full-edged CAE simulation runs, obvious efciency is obtained.  The warpage deection has been reduced by approximately 45 60% when, comparing with the warpage deection obtained using Moldow recommended processing parameters, i.e. reduced to 0.811.12 mm as compared to the benchmark value of 2.03 mm.

The sample points are put within squared brackets, in sequence of injection time (s), mold temperature (C), melt temperature (C), and packing pressure (%).

 The optimal process parameters are: injection time: 0.8 s, mold temperature: 69 C, melt temperature: 280 C, and packing pressure: 70%. Fig. 3 shows the corresponding warpage deection of 0.81 mm. To further illustrate the characteristics of the proposed optimization method, we select one of the optimization runs (the last of the 10 runs), and record some of the data generated from this run in Table 3.

Y.-M. Deng et al. / Materials and Design 31 (2010) 21182123

2123

Table 3 shows that,  Once the optimization process has determined a point as a likelihood-global optimum, the following sample points will not spread near this point.  The warpage values of the above sample points show that the proposed sampling method has the characteristic of spreading sample points around those regions of relatively small function values, and yet are away from the neighborhood of all the likelihood-global optimums.  Although many sample points correspond to a large warpage value, and that they may not be useful for nding the global optimum directly, they increase the reliability of regarding the likelihood-global optimum in the last generation as the global optimum. 6. Conclusion This investigation integrates the injection molding simulation with a global optimization algorithm, which is based on the mode-pursuing sampling (MPS) method and the traditional global optimization method of genetic algorithm (GA), for the search of the optimal process parameters for a minimum injection molding warpage. This proposed optimization algorithm can reduce the amount of computation required while determining a global optimum with good reliability. A plastic food tray case study, where injection time, mold temperature, melt temperature and packing pressure are selected as the design variables, demonstrates that the proposed hybrid optimization method can effectively reduce the warpage deection in a computationally efcient manner. This methodology, although targeted at injection molding warpage minimization at this investigation, could well be equally effective for other similar engineering optimization problems, which involve iterative runs of a computation-intensive black-box function (such as the time-consuming CAE simulations). Acknowledgements This research is funded by Zhejiang Provincial Natural Science Foundation of China, Grant No. R104247. It is also supported by K.C. Wong Magna Fund in Ningbo University. The authors would like to thank Dr. Gaofeng Gary Wang from the Simon Fraser University of Canada, for his helpful advice on the MPS method.

References
[1] Bryce DM. Thermoplastic troubleshooting for injection molders. Society of Plastics Engineers, Connecticut; 1991. [2] Lee BH, Kim BH. Optimization of part wall thickness to reduce warpage of injection molded part based on the modied complex method. Polym-Plast Technol Eng 1995;34(5):793811. [3] Lee BH, Kim BH. Automated selection of gate location based on desired quality of injection-molded part. Polym-Plast Technol Eng 1996;35(2):25368. [4] Seow LW, Lam YC. Flow optimisation in plastic injection molding. J Mater Process Technol 1997;72:33341. [5] Lam YC, Seow LW. Cavity balance for plastic injection molding. Polym Eng Sci 2000;40(6):127380. [6] Jin S, Lam YC. 2.5D cavity balancing. J Inject Mold Technol 2002;6(4):28496. [7] Zhai M, Lam YC, Au CK. Automated selection of gate location for plastic injection molding processing. Polym-Plast Technol Eng 2005;44(2):22942. [8] Deng Y-M, Lam YC, Britton GA. Optimization of injection moulding conditions with user-denable objective functions based on a genetic algorithm. Int J Prod Res 2004;42:136590. [9] Zhai M, Lam YC, Au CK. Runner sizing in multiple cavity injection mould by non-dominated sorting genetic algorithm. Eng Comput 2009;25(3):23745. [10] Deng Y-M, Zheng D, Lu X-J. Injection moulding optimisation of multi-class design variables using a PSO algorithm. Int J Adv Manuf Technol 2008;39(7 8):6908. [11] Erzurumlu T, Ozcelik B. Minimization of warpage and sink index in injectionmolded thermoplastic parts using Taguchi optimization method. Mater Des 2006;27:85361. [12] Wang GG, Shan S. Review of metamodeling techniques in support of engineering design optimization. J Mech Des, Trans ASME 2007;129(4): 37080. [13] Ozcelik B, Erzurumlu T. Determination of effecting dimensional parameters on warpage of thin shell plastic parts using integrated response surface method and genetic algorithm. Int Commun Heat Mass Transfer 2005;32:108594. [14] Shen C, Wang L, Li Q. Optimization of injection molding process parameters using combination of articial neural network and genetic algorithm method. J Mater Process Technol 2007;183(23):4128. [15] Gao Y, Wang X. An effective warpage optimization method in injection molding based on the Kriging model. Int J Adv Manuf Technol 2008;37(9 10):95360. [16] Zhou J, Turng L-S. Adaptive multi-objective optimization of process conditions for injection molding using a Gaussian process approach. Adv Polym Technol 2007;26(2):7185. [17] Gao Y, Wang X. Surrogate-based process optimization for reducing warpage in injection molding. J Mater Process Technol 2009;209:13029. [18] Wang L, Shan S, Wang GG. Mode-pursuing sampling method for global optimization on expensive black-box functions. Eng Optimiz 2004;36(4): 41938. [19] Zhang Y, Deng Y-M, Sun B-S. Injection molding warpage optimization based on a mode pursuing sampling method. Polym-Plast Technol Eng 2009;48(7): 76774. [20] Fu JC, Wang L. A random-discretization based Monte Carlo sampling method and its applications. Methodol Comput Appl Probab 2002;4:525. [21] Torn A, Zilinskas A. Global optimization. Berlin: Springer; 1987.

Vous aimerez peut-être aussi