臺灣博碩士論文加值系統

本研究乃是針對17-4PH不銹鋼之切削性能作一深入研究，其架構流程主要有三部份：實驗架構、刀具磨耗預測模式及車削參數最佳化模式。
在實驗架構方面，首先選擇切削速度、進給率、切削深度等三個實驗參數，利用直交表式的實驗計劃法設計實驗點，依照實驗點車削工件後再量測刀具加工後的刀腹磨耗量，進而求得倒傳遞網路所需之16組訓練範例與11組驗證數據。
刀腹磨耗預測模式乃是利用類神經網路中的倒傳遞網路原理，以田口法求得倒傳遞網路參數的最佳值後，並應用Fortran程式語言撰寫建構出一個刀腹磨耗預測模式。當輸入切削速度、進給率及切削深度等三個實驗參數，即可輸出刀腹磨耗的預測值，經驗證結果得知，此刀腹磨耗預測模式之平均誤差為6.64﹪。
接著利用遺傳學演算法程式，在應用田口法求得其程式參數之最佳值後，建構出一個車削參數最佳化模式，以求取在滿足所要求的刀腹磨耗限制條件下，獲得最大金屬移除率為目標函數時所對應的最佳車削加工參數。並且在不同刀腹磨耗限制條件；可得到不同的最大金屬移除率及其所對應的最佳切削速度、進給率及切削深度。
關鍵詞：不銹鋼、刀腹磨耗、最佳化。

In this study, the cutting performance was investigated on cutting 17-4PH stainless steel. The structure in this research contains three parts; that is, the experimental work, the predictive model for flank wear and the optimization of cutting parameters in turning operations.
Firstly, three cutting parameters including cutting speed, feed and depth of cut were selected. Then an orthogonal array for arranging the experimental points was used. According to the experimental points, turning operation was conducted and the flank wear of tools was measured. Thus the 16 training data and 11 verified data required in the predictive model were obtained.
The predictive model for flank wear of tool is built by the Back-propagation Network of artificial neural network theory. The Taguchi method was used to find the best values of parameter of Back-propagation Network and the Fortran language was corporated to construct the predictive model for flank wear. As the three cutting parameters； that is, cutting speed, feed and depth of cut, are chosen, the flank wear of tool can be predicted. According to the verification, the average error for the predictive model for the flank wear is 6.64﹪.
In addition, the genetic algorithm program was used in the optimized model. The Taguchi method was also applied to find the best values of parameter of genetic algorithm program, and the optimized model for the optimum cutting parameters in turning operations can be built. It can be used in which the constraint of flank wear is defined then the maximum metal removal rate is selected to be the objective function, thus the optimum cutting parameters in turning operation can be found. Forthermore, if different constraint of flank wear is used, the different value of maximum metal removal rate and the corresponding optimum values of cutting speed, feed and depth of cut can be found.
Keyword：Stainless Steel, Flank Wear, Optimized Model.

摘要………………………………………………….….…..…….I
英文摘要…………………………………..………….………….Ⅱ
誌謝……………………………………..…………….………….Ⅲ
目錄…………………………………………..……….…………IV
圖目錄………………………………..……………….………..VIII
表目錄……………………………..………………….………....XI
第一章 緒論…………….……..………………………………....1
1.1 前言…………………………..………..………………...1
1.2 文獻回顧…………………………………………….…..1
1.2.1切削 ...…...……...…………………..………...…..1
1.2.2 田口法 ……..……….…………………….….…..2
1.2.3 類神經網路 ………...……………………….….. 3
1.2.4 遺傳學演算法 ………………..………………….4
1.3 研究範圍與目的 …………………………..………..….5
第二章 田口法…………….…….……………………………….6
2.1 前言…………………………….……………....…….….6
2.2 田口法原理……….….………………..……………..….6
2.2.1 損失函數………..….……...….……………..…...6
2.2.2 品質特性值種類………..….…..………….……..8
2.2.3 直交表…….…………….……….…...…………..9
2.3 參數之最佳化程序………………………………….….10
2.3.1 定義目標函數………..……..…..……..….…..…10
2.3.2 定義設計變數及水準表………….…...…..…….10
2.3.3 選擇直交表……………………...……..………..10
2.3.4 平均數分析…………………..……..…………...11
2.3.5 變異數分析…………………..……………..…...12
2.4 田口法的應用………………………..……….....……….14
第三章 類神經網路系統…………..………………....…………17
3.1前言……………………..….……….....…………….…..17
3.2 類神經網路………………………….……….………....18
3.2.1 類神經網路之學習演算法…….……..………...19
3.2.2 倒傳遞網路… ..…..……………………..….…..20
3.3 倒傳遞網路運算式…...……...…..…………….……….21
3.3.1網路學習過程……………………...….………...21
3.3.2網路回想過程……………...….………………...25
3.4 架構刀具磨耗預測模式……………...…...…………….26
3.4.1 定義目標函數…...……....…..…………………..27
3.4.2 定義預測參數、水準值及直交表……………...27
3.4.3 平均數分析……..………….……………………27
3.4.4預測ANN最佳參數………...……………….….30
3.4.5 變異數分析……….…..……………………….…31
3.4.6 ANN重要性參數之微調.……...………………..34
3.5 預測切削工件90cm的刀腹磨耗....…...………………35
3.6 刀具磨耗預測模式的應用範圍.………….…...……….36
第四章 遺傳學演算法……..………….…………..……………..37
4.1 前言……..………….…………………….……………...37
4.2 演算原理……………….………………………………..37
4.3最佳化數學模式…….…………………………………...42
4.4 架構遺傳學演算法參數最佳化……..………..………...43
4.4.1定義最佳化目標函數……..………..………….....43
4.4.2定義最佳化參數、水準值及直交表…………….44
4.4.3 平均數分析……….………………..…………….46
4.4.4 預測GA最佳參數……...…….………………….46
4.4.5 變異數分析……….………………..…………….48
4.4.6 GA重要性參數之微調……….…………..………51
4.5 車削工件90cm的最大金屬移除率…..…………………52
4.6 最佳化模式之應用範圍……………..……..……………53
第五章 實驗程序…………………………….………………...54
5.1 實驗材料與設備………………..……………………..54
5.2 實驗設計………………………..……………………..59
5.2.1 定義實驗目標函數………………..…………...63
5.2.2 定義實驗參數、水準值及直交表…..………...63
5.2.3 實驗參數分析……………….……..…………..66
5.3 實驗步驟………………..……………..……………….67
5.3.1 車削加工之實驗步驟………………..………....67
5.3.2 求得類神經網路之訓練範例與驗證數據…….70
第六章 結果與討論……………………….………..……..……72
6.1 ANN及GA 參數最佳化……...……………..………...72
6.1.1 ANN參數最佳化……….………...…………….72
6.1.2 GA參數最佳化…………….…………………...73
6.2刀腹磨耗預測模式之應用………………..…….……...75
6.2.1利用田口法改善預測誤差…………..…...……..75
6.2.2 刀腹磨耗預測模式之驗證…………...…..…….75
6.3 GA最佳化模式之應用………………...………...………81
6.3.1刀腹磨耗之限制…………..…...………………. 81
6.3.2搜尋最佳車削參數組合…………..…..………...82
6.4 討論………………..……..……...………………………83
6.4.1田口法參數最佳化………………..…..………...83
6.4.2刀腹磨耗預測模式………………..…..………...83
6.4.3 GA最佳化模式………………..…...…………...84
第七章 結論與建議……………………….…..………..………85
7.1 結論………………………………….…………….…...85
7.2 建議…………………………………………………….89
參考文獻…...………………………………….………….…..…91
符號索引…………………………………………...….…..…….98
Appendix A. Orthogonal array and S/N for BPN parameters…...100
Appendix B. Orthogonal array and S/N for GA parameters…….102
Appendix C. The specification of experimental equipments…....104
Appendix D. Level array、orthogonal array and experimental data
…………………………………………………...106
圖目錄
Figure 2-1 Process of searching optimum parameter for TM…….13
Figure 3-1 Effect of processing elements……………..……...……18
Figure 3-2 Structure of Back-Propagation Neural Network….…..20
Figure 3-3 Mathematic model of from input layer to hidden layer
……………………………………….…………………22
Figure 3-4 Mathematic model of from hidden layer to output
layer………………………………………………….…22
Figure 3-5 Transfer function…………………………………..….22
Figure 3-6 Training process of BPN…………………………..….25
Figure 3-7 Recalling process of BPN…..…………..……....…….26
Figure 3-8 Corresponding figure of ANN parameters and S/N
…...…………………………………………………….29
Figure 3-9 Comparison of predicted error……………….……….31
Figure 3-10 Confidence intervals of S/N for prediction and
experiment……………………………………….….34
Figure 4-1 Fitness scaling……………………………….………..39
Figure 4-2 The process of GA…………………………..………..41
Figure 4-3 Corresponding figure for GA parameters versus S/N
…...……………………………………..……….………46
Figure 4-4 Comparison for MMRR……………..…...……….…. 47
Figure 4-5 Confidence interval of S/N for prediction and
experiment………………………………………..……51
Figure 5-1 Valenite VN8（P10）coated TiN tool tip………….…55
Figure 5-2 Cutting state……………………………...…………...56
Figure 5-3 Yam Iron Lathe（YAM850-GH）…………………...56
Figure 5-4 Heat treatment furance（Great Dragon T600）……..57
Figure 5-5 Hardness tester（Wilson Rockwell 5JRaRB）………57
Figure 5-6 Computer servo hydraulic universal material testing
machine（Hung Ta HT-9102）……………………….57
Figure 5-7 Measuring microscope（Nikon MM-40/27）……….58
Figure 5-8 Measurement of the cutting tool flank wear………….58
Figure 5-9 Measured region B for flank wear……………………59
Figure 5-10 Testing minimum depth of cut and minimum feed
（V=44.839m/min, d=0.5mm, f=0.097 mm/rev）…...61
Figure 5-11 Testing minimum depth of cut and minimum feed
（V=44.839m/min, d=0.6mm, f=0.097 mm/rev）……61
Figure 5-12 Testing minimum depth of cut and minimum feed
（V=44.839m/min, d=0.7mm, f=0.097 mm/rev）……61
Figure 5-13 Testing minimum depth of cut and minimum feed（V=44.839m/min, d=0.8mm, f=0.097 mm/rev）……61
Figure 5-14 Testing minimum depth of cut and minimum feed（V=44.839m/min, d=0.8mm, f=0.105 mm/rev）……61
Figure 5-15 Testing maximum feed
（V=105.598m/min, d=2mm, f=0.359 mm/rev）…….62
Figure 5-16 Testing maximum feed
（V=105.598m/min, d=2mm, f=0.314 mm/rev）…….62
Figure 5-17 Testing maximum feed
（V=105.598m/min, d=2mm, f=0.279mm/rev）……..62
Figure 5-18 flank wear contour
（V=58.542m/min, d=1.1mm, f=0.132 mm/rev）……64
Figure 5-19 flank wear contour
（V=58.542m/min, d=1.4mm, f=0.179 mm/rev）….…64
Figure 5-20 flank wear contour
（V=58.542m/min, d=1.7mm, f=0.228 mm/rev）….…64
Figure 5-21 flank wear contour
（V=58.542m/min, d=1.1mm, f=0.179 mm/rev）….…64
Figure 5-22 flank wear contour
（V=74.845m/min, d=1.4mm, f=0.228 mm/rev）….…65
Figure 5-23 flank wear contour
（V=74.845m/min, d=1.7mm, f=0.132mm/rev）….…65
Figure 5-24 flank wear contour
（V=90.777m/min, d=1.1mm, f=0.228 mm/rev）….…65
Figure 5-25 flank wear contour
（V=90.777m/min, d=1.2mm, f=0.132 mm/rev）……65
Figure 5-26 flank wear contour
（V=90.777m/min, d=1.7mm, f=0.179 mm/rev）……65
Figure 5-27 Corresponding figure of experimental parameter…....67
Figure 5-28 The process of machining…………………………....69
Figure 5-29 The different machining length versus VB（V=100.22m/min, d=1.68mm, f=0.179mm/rev）……70
Figure 6-1 Change in error function while training the network
with the increase of number of iterations…………....76
Figure 6-2（a）Training degree of accuracy of VB estimation…..77
Figure 6-2（b）Recalling degree of accuracy of VB estimation…77
Figure 6-3 The different training patterns versus predicted error...78
Figure 6-4 The comparison between measured VB and predicted
VB（Machining 6cm length of workpiece）...……….80
Figure 6-5 The comparison between measured VB and predicted
VB（Machining 90cm length of workpiece）……....…80
表目錄
Table 3-1 Suggested value of ANN parameters….……..………..28
Table 3-2 ANN parameters and their levels……..………….……28
Table 3-3 Orthogonal array of ANN parameters….……………...29
Table 3-4 Optimum ANN parameters for ANOM………………..30
Table 3-5 ANOM of predicted error of flank wear for S/N ……...33
Table 3-6 ANOVA of predicted error of flank wear for S/N…..….33
Table 3-7 ANOVA summary for dummy treatment……………....34
Table 3-8 Error for No. of training patterns……………………....35
Table 3-9 Optimum ANN parameters ………………………...….35
Table 4-1 Reference value for parameters…….………………….44
Table 4-2 Level for parameters…………………………………...45
Table 4-3 Orthogonal array for parameters……………………….45
Table 4-4 Optimum GA parameters for ANOM………………….47
Table 4-5 ANOM of MMRR for S/N…….………………….……49
Table 4-6 ANOVA of MMRR for S/N……………………..……..49
Table 4-7 ANOVA summary for the first dummy treatment……..50
Table 4-8 ANOVA summary for the second dummy treatment…..50
Table 4-9 The different string length versus MMRR……………..52
Table 4-10 The different population size versus MMRR…………52
Table 5-1 Chemical composition of 17-4PH stainless steel………54
Table 5-2 Mechanical properties of 17-4PH stainless steel………54
Table 5-3 Shape and size for tool…………………………………55
Table 5-4 Planning cutting speed range…………………………..60
Table 5-5 Testing minimum depth of cut and minimum feed….....60
Table 5-6 Testing maximum feed……….….……………………..62
Table 5-7 27 Cutting condition Experiments………..……………67
Table 5-8 (a) 16 training patterns
（Machining 6cm length of workpiece）…………...….…70
Table 5-8 (b) 11 recalling data
（Machining 6cm length of workpiece）……………….71
Table 5-9 (a) 16 training patterns
（Machining 90cm length of workpiece）……………...71
Table 5-9 (b) 11 recalling data
（Machining 90cm length of workpiece）……………...71
Table 6-1 The comparison between measured and predicted VB
（Machining 6cm length of workpiece）……………….79
Table 6-2 The comparison between measured and predicted VB
（Machining 90cm length of workpiece）……………...79
Table 6-3 Optimum machining parameters combination
（Machining 6cm length of workpiece）………………..82
Table 6-4 Optimum machining parameters combination
（Machining 90cm length of workpiece）………………82

[1] 王繼敏編著，1997，不銹鋼與金屬腐蝕，科技圖書有限公司，台北縣，第29-44頁。
[2] Zdzislaw klim, Elmekki Ennajimi, Marek Balazinski, Clement Fortin, 1996, “Cutting Tool Reliability Analysis for Variable Feed Milling of 17-4PH Stainless Steel, ” Wear, Vol. 195, pp. 206-213.
[3] G. H. Lim, 1995, “Tool-wear Monitoring in Machine Turning,” Journal of Materials Processing Technology, Vol. 51, pp. 25-36.
[4] 呂建宏，2001，「不銹鋼切削之表面粗糙度預測及參數最佳化模式之探討」，碩士論文，屏東科技大學，屏東。
[5] Jang, D. Y. and A. Seire, “Machining Parameter Optimization for Specified Surface Conditions,” Journal of Engineering for Industry, 114(2), May 1992, pp. 254-257.
[6] Yellowey, I. and D. Adey, “A New Approach to the Real Time Optimization of Turning Operation,” Int. J. March. Tools Manufacture., 32(3), 1992, pp. 453-466.
[7] 溫程雄，1998，「工具機最佳加工參數的設定」，碩士論文，中興大學，台中。
[8] 橫山巽子，1999，品質設計的實驗計畫法，中國生產力中心，台北。
[9] 羅錦興，1999，田口品質工程指引，中國生產力中心，台北市，第1-37頁。
[10] Phillip J. Ross, 1996, Taguchi Techniques for Quality Engineering, McGraw-Hill, Inc., New York, pp. 1-20.
[11] Su, Y. L., S. H. Yao, C. S. Wei, and C.T. Wu, 1998, “Analyses and Design of A WC Milling Cutter with TiCN Coating,” Wear, Vol. 215, pp. 59-66.
[12] Yang, W. H., and Y. S. Tarng, 1998, “Design Optimization of Cutting Parameters for Turning Operations Based on the Taguchi Method,” Journal of Materials Processing Technology, Vol. 84, pp. 122-129.
[13] 陳興忠、王柏村，1998，「田口法應用於最佳化設計問題」，屏東科技大學學報，第七卷，第一期，第13-20頁。
[14] Yang, S. M., and G. S. Lee, 1999, “Neural Network Design by Using Taguchi Method,” Journal of Dynamic System, Measurement, And Control, Vol. 121, pp. 560-563.
[15] 葉怡成，1997，應用類神經網路，儒林圖書有限公司，台北。
[16] Tamas Szecsi, 1999, “Cutting Force Modeling Using Artificial Neural Networks,” Journal of Material Processing Technology ,Vol .92-93, pp. 344-349.
[17] T. Obikawa., C. Kaseda, T. Matsumura, W. G. Going, T. Shirakashi, 1996, “Tool Wear Monitoring for Optimizing Cutting Conditions,” Journal of Materials Processing Technology, Vol. 62, pp. 374-379
[18] Santanu Das, A. B. Chattopadhyay, and A. S. R. Murthy, 1996, “Force Parameter for On-line Tool Wear Estimation： A Neural Netw Approach,” Neural Network, Vol. 9, pp. 1639-1645.
[19] Tamas Szecsi, 1998, “Automatic Cutting-Tool Condition Monitoring on CNC Lathe,” Journal of Materials Processing Technology, Vol. 77, pp. 64-69.
[20] L. H. S. Luong, T.A. Spedding, 1995, “A Neural-Network System for Predicting Machining Behaviour,” Journal of Materials Processing Technology, Vol. 52, pp. 585-591.
[21] David, E. G.,1997, Genetic Algorithms in Search Optimization and Machine Learing, Addison Wesley, Alabama, pp. 54-119.
[22] Rajeev, S., and Krishnamoorthy, C. S., 1997, “Genetic Algorithms-Based Methodologies for Design Optimization of Trusses,” Journal of Structural Engineering.
[23] Jenkins, W. M., 1991, “Towards Structural Optimization Via The Genetic Algorithms,” Computers and Structures, Vol. 40, No. 5, pp. 1321-1327.
[24] 林世豐等，1998，「基因演算法於船體結構件之最佳化設計應用」，碩士論文，成功大學，台南。
[25] 曹以正、陳家豪，1991，「遺傳演算法於機械元件最佳化設計之應用」，中國機械工程學會第八屆學術研討會，第715-720頁。
[26] 洪彥欽等，1994，「基因演算法於結構最佳化設計之應用」，碩士論文，交通大學，新竹。
[27] Ko, T. J., and H. S. Kim, 1998, “Autonomous Cutting Parameter Regulation Using Adaptive Modeling and Genetic Algorithms,” Precision Engineering, Vol. 22, pp. 243-251.
[28] 姚成福，1997，「基於類神經網路及遺傳學演算法之金屬切削參數預測模式」，碩士論文，屏東科技大學，屏東。
[29] 周宗易，1998，「304不銹鋼切削性預測模式之建立」，碩士論文，屏東科技大學，屏東。
[30] 洪敬祥，1999，「面銑削加工參數預測模式之建立」，碩士論文，屏東科技大學，屏東。
[31] 謝政宏，2000，「建立AISI 304不銹鋼之鑽孔品質預測及參數最佳化模式之探討」，碩士論文，屏東科技大學，屏東。
[32] 張永宗，1996，「工具機最大金屬切除率基於銑削振動之探討」，碩士論文，屏東科技大學，屏東。
[33] 關錦穗，1996，「工具機最大金屬切除率基於工件表面粗糙之探討」碩士論文，屏東科技大學，屏東。
[34] 洪國欽，1997，「鈦合金切削表面粗糙度基於切削參數之關係探討」，碩士論文，屏東科技大學，屏東。
[35] 焦成李，1991，神經網路系統理論，儒林圖書有限公司，台北。
[36] 盧炳勳等編譯，1992，類神經網路理論與應用，全華科技圖書股份有限公司，台北。
[37] 胡玉成，1997，暢談類神經網路，倚天資訊股份有限公司，台北。
[38] 葉怡成，1999，應用類神經網路，儒林圖書有限公司，台北。
[39] 葉怡成，1999，類神經網路模式應用與實作，儒林圖書有限公司，台北。
[40] Santanu Das, A. B., Chattopadhyay, and A. S. R. Murthy, 1996, ”Force Parameter for On-line Tool Wear Estimation：A Neural Approach ”, Neural Networks , Vo. 9, pp. 1639-1645.
[41] 蘇木春等，1999，機器學習：類神經網路、模糊系統以及基因演算法，全華科技圖書股份有限公司，台北。
[42] 王柏村、林義勝、彭耀哲，1994，「遺傳學演算法在螺栓組合結構最佳化之應用」，屏東技術學院學報，第三卷，第一期，第1-9頁。
[43] 劉惟信，1996，機械最佳化設計，金屬科技圖書股份有限公司，台北。
[44] 劉克琪，1994，田口式品質工程與實驗設計，華泰文化事業有限公司。
[45] 李輝煌，2000，田口方法-品質設計的原理與實務，高利圖書股份有限公司，台北。
[46] 李鈞澤，1997，切削刀具學，京元圖書股份有限公司，台北。
[47] 傅光華，1986，切削刀具學，高立圖書股份有限公司，台北。
[48] 張渭川譯，1987，切削加工技術資料集，全華科技圖書股份有限公司，台北。
[49] 陳永信譯，1994，「不銹鋼的加工技術」，機械月刊，第20卷，第5期，第236-241頁。
[50] Shie, K. H., 1986, Total Capacity Tooling（Valenite）, pp.1-20. （Tel：（07）3324337）
[51] 彭國倫，1997，精通Fortran 90程式設計，碁峰資訊股份有限公司，台北。