新一代CNC控制器趨勢朝向高品質加工，由於客製化產品及少量多樣的生產需求，快速調整CNC控制器參數以符合每一項產品之加工需求日漸重要。業界通常會以加工的週期時間、表面粗糙度以及精度作為產品的性能指標，然而由於加工特性的衝突，當加工速度不斷提高時，難以維持最高等級的加工精度與表面粗糙度要求。因此在不同的加工性能指標下，應有各自合適的CNC控制器參數組合。此外，CNC控制器內含大量的參數且某些參數又互相影響，如此將增加現場工程師調機的複雜度。
本研究利用數據驅動方式開發一CNC參數最佳化方法以達到高品質加工目標，此方法可以依據使用者給定的加工需求自動地設定CNC插補參數並預測加工性能。插補參數如速度、加速度、加加速度以及轉角平化化公差被選擇作為實驗因子，追蹤誤差、輪廓誤差與加工時間被視為加工性能指標。標準測試圖案程式KANINO路徑被使用以得到海德漢控制器產生的插補命令。在不同的插補參數組合下，實驗數據包含運動軸位置和工作台加速度可經由光學尺和加速規量測獲得，接著透過數據分析並收集成為訓練集。為了確保最終加工性能，反向傳播神經網絡的超參數使用Hyperband演算法進行優化調校，接著反向傳播神經網絡被使用來建立一個預測插補參數與加工性能關係的模型。此外，經過訓練的模型結合多目標函數以及非凌越排序基因演算法，可依據使用者設定的3種加工指標權重，求出最佳化的插補參數組合。最後，在一桌上型五軸雕刻機上執行實驗以驗證本研究所提出方法的可行性以及效能。實驗結果顯示反向傳播神經網絡模型的預測準確度可達90%以上，所有實驗的加工性能誤差皆低於11%。

The trend for next generation CNC machine tools is achieving high-quality machining. Due to the needs of customized products and small quantities of diverse production, it becomes more important to quickly adjust CNC controller parameters to meet machining requirements of each product during machining process. In machining industry, cycle time, accuracy, and surface roughness are usually regarded as performance indices of milling products. However, it is diffcult to keep machining accuracy and surface roughness at high levels while increasing machining speed due to the conflict in machining characteristics. Therefore, the performance indexes should have their own appropriate combinations of CNC controller parameters for each machining requirement. In addition, CNC controller has a large number of parameters and some parameters affect each other, which could increase the complexity of parameter adjustment for the field engineers.
In this study, a CNC parameter optimization based on data-driven approach is developed to achieve high-quality machining. The method can automatically set CNC interpolation parameters and predict machining performance according to machining requirements. CNC interpolation parameters including velocity, acceleration, jerk and position corner rolerance are selected as experimental factors. Tracking error, contouring error and cycle time are regarded as machining performance indices. The standard test pattern program such as KANINO toolpath is adopted to obtain the interpolation commands generated by Heihenhain controller. The data includes positions of motion axes and workpiece acceleration are measured by linear scales and accelerometer, respectively, in different combinations of interpolation parameters. The experiment data are analyzd and collected to the traning data set. To guarantee final machining performance, the hyperparameters of back propagation neural network (BPNN) is tunned by the Hyperband algorithm. Then the BPNN is ultilized to establish the predicted model between interpolation parameters and performance indices. Futhermore, the trained BPNN model and multi-objective function with non-dominated sorting genetic algorithm II (NSGA II) are combined to optimize the parameter combination according to three weighting values of performance indices set by end-users. Finally, the experimets are carried out on a desktop five-axis engraving machine to verify the feasibility and efficiency of the propsed method. The experimental results demonstrate that the prediction accuracy of the BPNN model achieve more than 90%, and the prediction error of machining performance are all less than 11%.

摘要...i
Abstract...ii
誌謝...iv
目錄...v
表目錄...vii
圖目錄...viii
符號說明...x
第一章 緒論...1
1.1 研究動機與目的...1
1.2 文獻探討...2
1.3 研究方法...4
1.4 論文架構...6
第二章 伺服控制迴路...7
2.1 伺服控制迴路設計...7
2.1.1 速度控制器...8
2.1.2 位置控制器...8
2.1.3 速度前饋控制器...9
2.2 前饋補償迴路設計...9
2.3 伺服誤差定義與推導...10
2.3.1 位置到誤差閉迴路轉移函數...11
2.3.2 鐘型加減速命令規劃...12
2.3.3 追蹤誤差解析解...14
2.3.4 輪廓誤差解析解...16
第三章 類神經網路與最佳化方法...17
3.1 監督式學習...17
3.2 數據集...18
3.2.1 Z-Score標準化...18
3.2.2 K-Fold交叉驗證...18
3.3 迴歸分析...19
3.3.1 殘差分析...19
3.3.2 數據診斷...20
3.4 類神經網路...23
3.4.1 反向傳播網路...24
3.4.2 損失函數與優化器...24
3.5 超參數最佳化...27
3.6 最佳化方法...30
3.6.1 基因演算法...30
3.6.2 非凌越排序基因演算法...31
第四章 插補參數最佳化實驗...35
4.1 實驗平台...35
4.2 實驗設計...36
4.3 實驗架構...38
4.4 資料前處理...41
4.4.1 殘差分析...41
4.4.2 數據診斷...42
4.5 類神經網路模型...42
4.5.1 超參數優化...43
4.5.2 反向傳播網路...43
4.5.3 K-Fold交叉驗證...45
4.6 插補參數優化...45
第五章 實驗結果與探討...47
5.1 加工時間優先...47
5.2 精度優先...48
5.3 平滑度優先...49
5.4 使用者優化...50
5.5 實驗結果...51
5.6 加工性能比較...52
5.6.1 轉角輪廓...53
5.6.2 平台震動量...54
第六章 結論與未來方向...60
參考文獻...61
附錄一 追蹤誤差與輪廓誤差推導...63
附錄二 數據集之殘差分析與數據診斷...70

[1]FANUC公司，2018 FANUC技術連絡會，說明手冊，2018。
[2]Siemens Corporation, Milling with SINUMERIK: Mold making with 3 to 5-axis simultaneous milling, Manual, 2016.
[3]MAZAK Corporation, Smooth technology, Datasheet, 2016.
[4]J.-P. Besuchet, M. Etienne and R. Löttgen, A method and an operator support system for assisting an operator in setting machine parameters, European Publication Patent No. EP1308810B1, 2005.
[5]林建佑，五軸工具機雲端智能診斷與自動調機優化基於頻域系統鑑別，國立中興大學機械工程研究所，碩士論文，2017。
[6]H.-W. Chiu and C.-H. Lee, “Prediction of machining accuracy and surface quality for CNC machine tools using data driven approach,” Advances in Engineering Software, vol. 114, pp. 246-257, 2017.
[7]邱泓維，智慧型加工專家系統:基於CNC控制器參數選擇與優化，國立中興大學機械工程研究所，碩士論文，2017。
[8]郭維峰，工具機加工性能指標估測：使用實驗設計及數據驅動，國立中興大學機械工程研究所，碩士論文，2020。
[9]黃政豪，應用類神經網路預測CNC加工中心機之加工時間，國立中興大學機械工程研究所，碩士論文，2018。
[10]林忠毅，以加工指標為基準的CNC參數自動學習調整技術，國立中興大學機械工程研究所，碩士論文，2017。
[11]C. Ahilan, S. Kµmanan, N. Sivakµmaran and J. E. R. Dhas, “Modeling and prediction of machining quality in CNC turning process using intelligent hybrid decision making tools,” Applied Soft Computing, vol. 13, no. 3, pp. 1543-1551, 2013.
[12]B.-F. Yu and J.-S Chen, “Development of an analyzing and tuning methodology for the CNC parameters based on machining performance,” Applied Sciences, vol. 10, no. 8, pp. 1543-1551, 2020.
[13]楊倍驊，深層類神經網路於CNC參數與加工指標之動態性能預測與優化應用，國立中興大學機械工程研究所，碩士論文，2018。
[14]余瑋軒，基於深度類神經網路於彈性約束多目標優化CNC控制器參數之應用，國立中興大學機械工程研究所，碩士論文，2020。
[15]廖學佑，智慧化加工表面粗糙度預測模型及材料移除率最佳化研究-以塑膠射出成型模銑削為例，國立中興大學機械工程研究所，碩士論文，2015。
[16]曾翔駿，CNC工具機虛實整合系統開發及插補參數最佳化之應用，國立中正大學機械工程研究所，博士論文，2021。
[17]Y. Koren and C.-C. Lo, “Variable-gain cross-coupling controller for contouring,” CIRP Annals - Manufacturing Technology, vol. 40, no. 1, pp. 371-374, 1991.
[18]林明宗，開發具有預視及學習功能之運動控制器，國立中正大學機械工程研究所，博士論文，2007。
[19]M.-T. Lin, C.-L. Yen, M.-S. Tsai, and H.-T. Yau, “Application of robust iterative learning algorithm in motion control system,” Mechatronics, vol. 23, no. 5, pp. 530-540, 2013.
[20]劉德葳，PCB機台之伺服模型最佳化與自動調校，國立虎尾科技大學機械設計工程研究所，碩士論文，2017。
[21]賴涵餘，五軸工具機適應性負荷控制與伺服參數優化技術開發，國立虎尾科技大學機械設計工程研究所，碩士論文，2020。
[22]M.-T. Lin and S.-K. Wu, “Modeling and improvement of dynamic contour errors for five-axis machine tools under synchronous measuring paths,” International Journal of Machine Tools and Manufacture, vol. 72, pp. 58-72, 2013.
[23]I. Goodfellow, Y. Bengio and A. Courville. Deep Learning. The MIT Press. 2016.
[24]R. H. Myers. Classical and Modern Regression with Applications. Duxbury Press 1st ed., 1986.
[25]S. Ruder, An overview of gradient descent optimization algorithms, 2016. arXiv preprint 1609.04747. https://doi.org/10.48550/arXiv.1609.04747
[26]D. P. Kingma and J. Ba, Adam: A method for stochastic optimization, 2014. arXiv preprint arXiv:1412.6980. https://doi.org/10.48550/arXiv.1412.6980
[27]J. Snoek, L. Hugo and R. P. Adams, “Practical bayesian optimization of machine learning algorithms.” Advances in Neural Information Processing Systems, 2012.
[28]J. Kevin and A. Talwalkar, “Non-stochastic best arm identification and hyperparameter optimization,” Artificial Intelligence and Statistics, 2016.
[29]J. H. Holland, Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control, and Artificial Intelligence. MIT Press, 1992.
[30]K. Deb, A. Pratap, S. Agarwal and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,” IEEE Transactions on Evolutionary Computation, vol. 6, no. 2, pp. 182-197, 2002.
[31]李孟哲，五軸刀具中心點路徑即時預讀與插補技術之研究，國立虎尾科技大學機械設計工程研究所，碩士論文，2016。
[32]黃建祐，CNC工具機之伺服調機軟體開發，國立虎尾科技大學機械設計工程研究所，碩士論文，2016。
[33]K. Lee, S. Ibarakil, A. Matsubara, Y. Kakino, Y. suzuki, S. Arai and J. Braasch, “A servo parameter tuning method for highspeed NC machine tools based on contouring error measurement,” WIT Transactions on Engineering Sciences, vol.12, 2003.
[34]Weisberg and Stanford. Applied Linear Regression. John Wiley & Sons, Inc. 4th ed., 2013.
[35]Heidenhain Corporation, iTNC530 NC Software 340 420-xx User’s Manual, 2002.
[36]Heidenhain Corporation, iTNC530 HSCI NC Software 606 420-02, 606 421-02 – Technique Manual, 2012.
[37]Google Inc. API Documentation | TensorFlow Core v2.9.1, https://www.tensorflow.org/api_docs
[38]J. Blank and K. Deb, “Pymoo: Multi-objective Optimization in Python,” IEEE Access, vol. 8, pp 89497-89509, 2020.