本篇論文利用變動鄰居塔布搜尋法(VNTS)和決策樹演算法(Decision Tree Algorithm)去建構一個規則歸納系統(RIS)。將已經訓練完成的類神經模式，與變動鄰居塔布搜尋法相結合，來尋找射出成型機參數組合之最佳可行解，結果發現此演算法具有不錯的尋優能力。決策樹演算法則是利用一開始田口實驗中所收集的實驗數據加以分析與分類，從此數據中歸納出可以調控射出成型機台的規則。 在本論文實證案例中，發現到變動鄰居塔布搜尋法確實可以提供多組可生產射出成型良品的機台參數。再利用這多組中最佳的機台參數來修正原始田口實驗中的參數水準，最後便由決策樹演算法使用此修改後的資料來歸納出可供調機者日後調機的規則。實驗結果發現本論文所歸納出來的規則與過去相關射出成型文獻所歸納出來的規則幾乎相近，只有少數幾條規則例外，判斷應該是所使用的射出材料與機台類型各自不同有關。由此看來，本論文可針對射出成型機台，提供了實用的調機規則與最佳參數，以減少不良品的嚴重程度與數目，未來也將有效地幫助未來其他精密機械等產業，降低其調機之成本與時間。

In the thesis, variable neighborhood tabu search (VNTS) and Decision Tree Algorithm are used to implement the rule induction system (RIS). VNTS with the well-trained BPN criterion demonstrates good searching capability to find the best solution for parameter setting of injection molding machine. Decision Tree Algorithm is the tool for inducting rules from the experimental data. Taguchi method is adopted to perform task of experiment design. The illustrated examples in the empirical study show how this RIS works and sets up rules for machines. The rules inducted by Decision Tree Algorithm will be revised according to the best solution found by VNTS. The results show not only are the rules found by RIS pretty close to what the literature did, but also can RIS provide both the best parameters of injection molding machines and the appropriate machine operation rules to work staff in the workspace.

Table of Contents
摘要 I
Abstract II
Acknowledgements III
Table of Contents IV
List of Tables V
List of Figures VII
Chapter 1 Introduction 1
1.1 Research Background and Motivation 1
1.2 Research Purposes 6
1.3 Research Method 6
1.4 Research Process 7
Chapter 2 Literature Review 9
2.1 Artificial Intelligence 9
2.1.1 Neural Network 9
2.1.2 Type of Neural Network 10
2.1.3 Back-Propagation Network 12
2.1.4 Variable Neighborhood Tabu Search 15
2.1.5 Decision Tree Algorithm 17
2.2 Injection Molding Manufacturing 17
Chapter 3 Research Structure and Methods 21
3.1 Experimental Data 21
3.2 Combination Structure of Back-propagation network and Variable Neighborhood Tabu Search 21
3.3 Structure of Decision Tree Algorithm 29
3.4 Structure of the Entire System 32
Chapter 4 Empirical Study 35
4.1 Training of Back-Propagation Network Criterion 35
4.2 The experimental implementation of VNTS optimization 40
4.2.1 The comparison of VNTS and TS 48
4.2.2 The other discovery of thesis 49
4.3 The implementation of decision tree algorithm 51
Chapter 5 Conclusions and Contributions 67
5.1 Conclusions 67
5.2 Contribution 67
5.3 Future Research 68
Reference 69
Appendix The defect conditions of injection molding (IM) product 72









List of Tables
Table 1.1 The causes and solution policies of the defects 3
Table 2.1 Summary of literature review 20
Table 3.1 The operational and its level parameters 21
Table 3.2 The experiment al combinations of L18 orthogonal array 21
Table 3.3 The grading and weighting procedure of defects 23
Table 4.1 The operational and its level parameters 36
Table 4.2 The experimental combinations of L18 orthogonal array 36
Table 4.3 The normalized experimental combinations of BPN inputs 37
Table 4.4 The normalized experimental combinations of BPN outputs 38
Table 4.5 The experimental results of VNTS with the level of iteration number of VNS is 10 and variable levels of iteration number of tabu 41
Table 4.6 The experimental results of VNTS with the level of iteration number of VNS is 50 and variable levels of iteration number of tabu 41
Table 4.7 The experimental results of VNTS with the level of iteration number of VNS is 100 and variable levels of iteration number of tabu 42
Table 4.8 The experimental results of VNTS with the level of iteration number of VNS is 200 and variable levels of iteration number of tabu 42
Table 4.9 The experimental results of VNTS with the level of iteration number of VNS is 500 and variable levels of iteration number of tabu 43
Table 4.10 The experimental results of VNTS with the level of iteration number of VNS is 1000 and variable levels of iteration number of tabu 43
Table 4.11 The experimental analysis of ANOVA to decide the best iteration combination of VNTS for the case 44
Table 4.12 The V50T200 experimental results of 10 execution times 47
Table 4.13 The experimental results of 10 execution times 48
Table 4.14 The solution replacement time among the neighborhood of VNTS 49
Table 4.15 The solution performance and experimental time after disusing the neighborhood of 2-swap with appointment. 50
Table 4.16 The original setting of nominal orthogonal array 51
Table 4.17 The new setting of the nominal orthogonal array based on the best solution found by VNTS 52
Table 4.18 The original nominal data sets 53
Table 4.17 Correct rate for the prediction of decision tree model for each kind of quality characteristic 54
Table 4.19 The table of comparison results with other past literatures 63
Table 4.20 The rearranged setting rules for the specific defect degree 65


















List of Figures
Fig. 1.1 The injection molding machine [16] 1
Fig1.2 Research framework 8
Fig 2.1 Network structure of back-propagation network 12
Fig 2.2 Neighborhood structures in VNTS 16
Fig 3.1 Procedure of BPN training 22
Fig 3.2 Parameter encoding solution 25
Fig 3.3 Maximum encoding range 25
Fig 3.4 two-swap with appointment 26
Fig 3.5 Random single point change 26
Fig 3.6 Random two point change 27
Fig 3.7 The optimization procedure of VNTS algorithm 28
Fig 3.8 The example of decision tree 31
Fig 3.9 The structure of the entire system 34
Fig 4.1. The structure of BPN criterion 37
Fig 4.2 The training result after 100 epochs 39
Fig 4.3 The training result after 5000 epochs 39
Fig 4.4 The training result after 10000 epochs 39
Fig 4.5 The training result after 20000 epoch 39
Fig. 4.6 The illustration of T with significant interaction respect to V 45
Fig. 4.7 The illustration of V with significant interaction respect to V 45
Fig 4.8 The illustration of V50T200 experimental results of 10 execution times 47
Fig 4.9 The data set for Clementine V6.5 53
Fig 4.10 The decision tree model of Clementine V6.5 53
Fig 4.11 The rules from decision tree of Clementine V6.5 for short shot 54
Fig 4.12 The rules from decision tree of Clementine V6.5 for sink mark 56
Fig 4.13 The rules from decision tree of Clementine V6.5 for Warp 58
Fig 4.14 The rules from decision tree of Clementine V6.5 for burn mark 59
Fig 4.15 The rules from decision tree of Clementine V6.5 for flash 60
Fig 4.16 The rules from decision tree of Clementine V6.5 for weld line 61
I. The normal condition of IM product 72
II. The short shot condition of IM product 72
III. The warp and flash condition of IM product with weld line problem 73
IV. The burn mark condition of IM product (see the place the arrow points) 73
V. The close-up picture of Burn mark condition of IM product (see the place the arrow points) 74
VI. The short shot and sink mark condition of IM product with weld line problem 74
VII. The warp condition of IM product 75
VIII. The serious short shot condition of IM product with deforming problem 75

[1]Chang, F.J., et al., 2003, Theory and practice of neural network, Tung-Hwa Pub. Co., Taipei.
[2]Glover, F., et al., 1997, Tabu Search, Kluwer Academic, Boston.
[3]Li, Y.C., et al., 2004, Fuzzy theory and its application, Cheng-Hwa Pub. Co., Taipei.
[4]Mehmed Kantardzic, Data Mining-Concepts, Models, Methods, and Algorithms”, A John Wiley & Sons, INC., U.S.A.
[5]Choi, G. H., Lee, K. D., Chang N., Kim, S.G., 1994, “Optimization of Process Parameters of Injection Molding with Neural Network Application in a Process Simulation Environment”, Annals of the CIRP, Vol. 43, pp.449-452.
[6]Glover, F., 1989, “Tabu search – Part I”, Operations Research Society of America Journal on Computing, Vol 1, No.3, pp.190-206.
[7]Glover, F., 1990, “Tabu search – Part II”, Operations Research Society of America Journal on Computing, Vol 2, No.1, pp.4-32.
[8]Hansen, P., Mladenovic, N., 1997, “Variable neighborhood search”, Computers Ops Res., Vol. 24, No. 11, pp.1097-1100.
[9]Hansen, P., Mladenovic, N., 1999, “Variable neighborhood search: Principles and applications”, European Journal of Operation Research 130, pp.449-467.
[10]Kenig, S., Ben-David, A., Omer, M., Sadeh, A., 2001, “Control of properties in injection molding by neural networks”, Engineering Application of Artificial Intelligence 14, pp.819-823.
[11]Kim, D., Rhee, S., 2002, “Design of a robust fuzzy controller for the arc stability of CO2 welding process using the Taguchi method”, IEEE, Trans. on Systems, Man, and Cybernetics-Part B: Cybernetics, Vol. 32, No. 2, pp.157-162.
[12]Lau, H.C.W., Ning, A., Pun, K.F., Chin, K.S., 2000, “Neural networks for the dimensional control of molded parts based on a reverse process model”, Journal of Materials Processing Technology 117, pp.89-96.
[13]Lau, H.C.W., Wong, T.T., Pun, K.F., 1999, “Neural-fuzzy modeling of plastic injection molding machine for intelligent control”, Expert Systems with Application 17, pp.33-43.
[14]Mok, S.L., Kwong, C.K., 2002, “Application of artificial neural network and fuzzy logic in a case-based system for initial process parameter setting of injection molding”, Journal of Intelligent Manufacturing 13, pp.165-176.
[15]Perez, J., Marcos, J., Martin, I., 2003, “Variable neighborhood tabu search and its application to the median cycle problem”, European Journal of Operation Research 151, pp.365-378.
[16]Vagelatos, G.A., Rigatos, G.G., Tzafestas, S.G., 2001, “Incremental fuzzy supervisory controller design for optimizing the injection molding process”, Expert Systems with Application 20, pp.207-216.
[17]Chen, J.S., 1995, The optimal manufacturing process design of injection molds using expert system technology, National Cheng Kung University, Master Thesis.
[18]Chou, T., 1996, The fault diagnosis fuzzy expert system on an injection molding machine, National Chung Cheng University, Master Thesis.
[19]Hsu, H.S., 1996, Injection molding expert system, Chung Yuan Christian University, Master Thesis.
[20]Kuo, J.T., 1995, The development of the fault diagnosis and monitoring expert system for injection molding process, National Chung Cheng University, Master Thesis.
[21]Li, Y.C., 2002, Structuring and analysis of the injection molding process for the neural network-using the Taguchi method, National Taiwan University of Science and Technology, Master Thesis.
[22]Tzeng, K.H., Lin, N., 2003, Data mining technology application to injection molding, TamKang University, Master Thesis.
[23]Yang, J.C., 2002, Prediction of optimal machining parameters of injection molding machine, National Taiwan University of Science and Technology, Master Thesis.