線放電加工(Wire Electrical Discharge Machining, WEDM)製程中，前期研究已可以從加工過程中透過電壓與電流之感測收集，並針對兩者之複合物理量進行特徵提取，進而監控加工製程的穩定度。然因線放電加工製程為多行程加工法，加工複雜、且耗費時間成本高，須從粗加工再精加工，而精加工又分為多重刀工，且各刀次之加工參數皆不同，欲藉由單一模型對於整個放電過程監控及預測著實困難。因此，如何針對線放電加工製程之監控，並提出有效之模型預測，實為本研究之主要探討目的。
本研究開發一線上WEDM異常診斷系統，利用高取樣率之類比數位擷取器以擷取電壓與電流訊號，並透過TCP/IP對線放電機控制器提取當下之加工座標位置。在基於已有特徵集下，本研究更提出一開路特徵用以分類資料，以及其他如電壓，線速度與目前加工刀次，從而獲得合適的刀次模型。在診斷方法上，藉由K近鄰(K Nearest Neighbor, KNN)以進行完整加工刀次資料中5類特性之分段，分別為正常、開路、異常、轉角與異常多切等分段，系統再將分類出之異常段以多項式迴歸(Polynomial Regression, PR)，預測其線極絲與工件壁間距(Deviation)，最後再藉由廣播將機台狀態提供戰情室決策使用。
透過精進與優化前其研究成果，本研究以某一線放電加工(WEDM)機為例，所開發的系統可即時判斷當刀次之異常，其準確度除某一刀次僅有73%，其於刀次皆90%以上，可知本研究所開發系統具有線上異常診斷之能力。

In the wire electrical discharge machining (WEDM) process, previous research has been collected from the voltage and current sensing process. Feature extraction for the compound physical quantity of voltage and current, and then monitoring the stability of the processing process. However, because the WEDM is multi-processing, starting with rough and then precision finish machining, the processing parameters of each cutting process are different. It is tough to monitor and predict the entire discharge process with a single model, and the time and cost are high due to the complexity of the processing. Therefore, how to monitor and propose an effective model for the monitoring of the wire discharge processing process is the primary purpose of this study.
This study develops an online WEDM abnormality diagnosis system, which uses a high sampling rate analog digital extractor to capture voltage and current signals, and extracts the current processing coordinate position through the TCP/IP. Based on the existing feature set, this study proposes an open feature to classify data, as well as other features such as voltage, line speed, and current machining tool number, so as to obtain a suitable tool number model. In terms of diagnostic methods, K Nearest Neighbor (KNN) is used to segment the five types of characteristics in the data, which are normal, open, abnormal, corner, and abnormal over-cut segments. The system then uses the polynomial regression to predict the distance between the wire electrode and the workpiece wall and finally broadcasts the status of the machine to the war room for decision-making.
Through refinement and optimization of the previous research results, this research takes a specific WEDM machine as an example. The developed system can immediately judge the abnormality of the current cutting process. The accuracy is only 73% except for a certain cutting time, and its accuracy is more than 90% for other cutting processes. It can be seen that the system developed by this research has the ability of online abnormal diagnoses.

摘要 i
Abstract ii
誌謝 iii
目錄 iv
表目錄 vi
圖目錄 vii
第1章 緒論
1.1 研究背景
1.2 研究目的
1.3 研究架構
第2章 放電加工理論方法
2.1 特徵計算方法
2.2 特徵挑選方法
2.2.1 F1-Score
2.2.2 Mean Square Error 均方差
2.3 異常檢測方法
2.3.1 KNN近鄰演算法
2.3.2 貝葉氏定理
2.3.3 多項式迴歸模型
2.3.4 線性迴歸
第3章 線切割放電加工監測系統開發
3.1 系統設計
3.1.1 系統概念
3.1.2系統流程
3.2 感測資料蒐集
3.3 模型建立
3.3.1 KNN近鄰演算法建立
3.3.2多項式迴歸模型建立
3.4 系統介面
第4章 放電加工異常案例分析
4.1 放電加工異常偵測、異物與凹陷
4.1.1 異常異物與凹陷實驗結果
4.1.2 實驗小結
4.2 放電加工異常偏移加工實驗
4.2.1 實驗目的與預期
4.2.2 異常偏移實驗結果
4.2.3 實驗小結
4.3 進階放電偏移加工
4.3.1 實驗目的與預期
4.3.2 進階偏移加工實驗結果
第5章 結論與未來展望
5.1 結論
5.2 未來展望
參考文獻

[1]李奇，模具構造與制造，中國: 清華大學出版社，2024.
[2]X. Wu, S. Li, Z. Jia, B. Xin, and X. Yin, “Using WEDM to remove the recast layer and reduce the surface roughness of WEDM surface,” Journal of Materials Processing Technology, vol. 268, pp. 140-148, 2019.
[3]P. S. Rao, K. Ramji, and B. Satyanarayana, “Effect of WEDM conditions on surface roughness: A parametric optimization using Taguchi method,” International journal of advanced engineering sciences and technologies, vol. 6, no. 1, pp. 41-48, 2011.
[4]A. Caggianoa, R. Tetia, R. Perezb, and P. Xirouchakisc, “Wire EDM Monitoring for Zero-Defect Manufacturing based on Advanced Sensor Signal Processing,” Procedia CIRP, Vol. 33, pp. 315-320, 2015.
[5]S. Kwon, M. Y. Yang, “The benefits of using instantaneous energy to monitor the transient state of the wire EDM process,” The International Journal of Advanced Manufacturing Technology, Vol 27, pp. 930-938, 2006.
[6]S. S. Kumar, M. Uthayakumar, S. T. Kumaran, P. Parameswaran, T. K. Haneef, C. K. Mukhopadhyay, and B. P. C. Rao, “Performance monitoring of WEDM using online acoustic emission technique,” Silicon, Vol 10, pp. 2635-2642, 2018.
[7]A. Caggiano, R. Perez, T. Segreto, R. Teti, and P. Xirouchakis, “Advanced sensor signal feature extraction and pattern recognition for wire EDM process monitoring,” Procedia CIRP, Vol 42, pp. 34-39, 2016.
[8]S. S. Nain, D. Garg, and S. Kumar, “Prediction of the performance characteristics of WEDM on Udimet-L605 using different modelling techniques,” Materials Today: Proceedings, Vol 4, no. 2, pp. 546-556, 2017
[9]X. H. Zheng, Z. Q. Liu, Q. L. An, M. Chen, and C. Y. Wang, “Experimental research on the high speed drilling of PCB using microdrill,” in Advanced Materials Research, Vol. 426, pp. 223-226, 2012.
[10]V. S. Jatti, R. B. Dhabale, A. Mishra, N. K. Khedkar, V. S. Jatti, and A. V. Jatti, “Machine Learning Based Predictive Modeling of Electrical Discharge Machining of Cryo-Treated NiTi, NiCu and BeCu Alloys,” Applied System Innovation, Vol. 5, no. 6, pp.107, 2022.
 
[11]R. Chalisgaonkar, J. Kumar, and P. Pant, “Prediction of machining characteristics of finish cut WEDM process for pure titanium using feed forward back propagation neural network,” Materials Today: Proceedings, Vol. 25, pp. 592-601, 2020.
[12]B. Choudhuri, and R. Sen, “Optimization of WEDM parameters for Machining Inconel 800 by ANN based Bayesian hybrid algorithm,” Materials Today: Proceedings, Vol. 62, pp. 1098-1101, 2022.
[13]J. K. Kruschke, “What to believe: Bayesian methods for data analysis,” Trends in cognitive sciences, Vol. 14, no. 7, pp.293-300, 2010.
[14]G. Guo, H. Wang, D. Bell, Y. Bi, and K. Greer, “KNN model-based approach in classification,” in On The Move to Meaningful Internet Systems 2003: CoopIS, DOA, and ODBASE: OTM Confederated International Conferences, CoopIS, DOA, and ODBASE 2003, Catania, Sicily, Italy, November, 3-7, 2003. Proceedings, pp. 986-996, 2003.
[15]J. Joyce, “Bayes’ theorem,” Publisher, 2003, Available: http://seop.illc.uva.nl/entrie s/bayes-theorem/. Accessed on: July, 24, 2023.
[16]F-score維基百科，來源：https://zh.wikipedia.org/wiki/F-score. 索引：2023/7/24.
[17]均方誤差維基百科，來源：https://zh.wikipedia.org/zh-tw/%E5%9D%87%E6%9 6%B9%E8%AF%AF%E5%B7%AE. 索引：2023/7/24.
[18]多項式迴歸維基百科，來源：https://zh.wikipedia.org/zh-tw/%E5%A4%9A%E9% A1%B9%E5%BC%8F%E5%9B%9E%E5%BD%92. 索引：2023/7/24.
[19]M. Akaish(章奇煒，王心薇譯)，深度學習的數學地圖：用 Python 實作神經網路的數學模，旗標出版社，2020。
[20]蘇庭煒，”開發放電加工之智慧型關鍵特徵萃取系統”，高雄第一科技大學電機工程研究所碩士論文，2017。
[21]謝慶嶧，”智慧製造分析工具箱之開發”，國立高雄科技大學電機工程研究所碩士論文，2019。
[22]莊閔鈞，”開發一放電加工線上異常監測系統”，國立高雄科技大學電機工程研究所碩士論文，2020。