臺灣博碩士論文加值系統

在放電加工 (Electrical Discharge Machining, EDM) 製程中，前期研究已可因應其長加工歷程與高放電頻率的特性，有效的估測加工精度。然當發生加工異常時，由於放電製程包含多重加工參數，目前難以識別，從而增加事後的修補或損失成本。因此，如何即時監測放電製程之異常，實為本研究之主要探討目的。
本研究開發一放電加工製程異常診斷系統，首先，將資料透過環形陣列儲存以加速處理時間，並新增特徵間隔時間與變異係數特徵，已有效萃取異常特性。在診斷方法上，藉由類神經網路、隨機森林及XGB-RF(Extreme Gradient Boosting applying RF)所建立的複合投票模型，可提供穩健的診斷效果。最後經由Node-red網頁與MQTT協定，可提供即時監測機台異常的能力。
透過精進與優化前期研究成果，本研究以放電加工鑽石砂輪為例，開發出可於異常(如非導體沾黏或聚晶鑽石高突等異常模式)發生3分鐘內發出警示，並具有精準度達93%，誤判率小於0.2%的異常診斷能力。因此，本研究所開發之線上異常監測系統將可提供放電加工之線上異常診斷。

For EDM (Electrical Discharge Machining) processes, the previous studies have presented their effectiveness in estimation of machining accuracy in response to the process characteristics with long machining time and high discharge frequency. However, extra repair and loss cost will increase when there exist exceptions but failing in detecting from an EDM process with multi-parameter. Hence, the purpose of this study is how to monitor and diagnose EDM exceptions in real time.
This study develops a diagnosis system for EDM process. First, circular memory arrays are used to store data for reducing processing time, and features including spacing time and variation coefficient are added for extracting exception characteristics. Then, a hybrid voting model with NN (neural network), RF (Random Forest) and XGB-RF (Extreme Gradient Boosting applying RF) methods is proposed to diagnose various failure modes in robust. Finally, a real-time exception monitor is realized via Node-red web and MQTT protocol.
After improving and optimizing previous research, an EDM diamond grinding wheel case study shows that the exceptions, e.g., as non-conductor sticking and polycrystalline diamond bump, can be detected within three minutes with sensitivity more than 93% and false positive rate less than 0.2%. Therefore, the proposed diagnosis system can effectively diagnose and identify exceptions in an EDM process.

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
1.1 研究背景 1
1.2 研究目的 3
1.3 研究架構 4
第2章 放電加工理論方法 5
2.1 放電加工異常原因 5
2.2 特徵計算方法 7
2.3 異常檢測方法 10
2.3.1 類神經網路 10
2.3.2 隨機森林 11
2.3.3 XGBoost 16
2.3.4 KLD 18
2.3.5 混淆矩陣 18
第3章 放電加工監測系統開發 20
3.1 系統設計 20
3.1.1 系統概念 20
3.1.2 系統流程 21
3.2 感測資料蒐集 23
3.3 特徵計算 24
3.4 模型建立 26
3.4.1 類神經模型建立 26
3.4.2 隨機森林模型建立 26
3.4.3 XGB- RF模型建立 27
3.4.4 複合模型投票機制 28
3.5 系統介面 29
第4章 放電加工異常案例分析 33
4.1 放電加工異常一、異物 36
4.1.1 計算特徵之異常實驗結果 36
4.1.2 基於類神經網路之異常診斷 51
4.1.3 基於隨機森林之異常診斷 53
4.1.4 基於XGB- RF之異常診斷 54
4.1.5 實驗(一)小結 55
4.2 放電加工異常二、高點 56
4.2.1 計算特徵之異常實驗結果 57
4.2.2 基於類神經網路之異常診斷 64
4.2.3 基於隨機森林之異常診斷 66
4.2.4 基於XGB- RF之異常診斷 67
4.2.5 實驗(二)小結 70
4.3 放電加工異常三、混合樣本驗證 71
4.3.1 基於類神經網路之異常診斷 73
4.3.2 基於隨機森林之異常診斷 74
4.3.3 基於XGB- RF之異常診斷 75
4.3.4 複合模型投票結果 76
4.3.5 實驗(三)異常加工驗證 78
4.3.6 實驗(三)小結 80
第5章 結論與未來展望 81
5.1 結論 81
5.2 未來展望 82
參考文獻 83

[1]蘇品書，線切割放電加工，復漢出版社，2000.
[2]L. Selvarajan, J. Rajavel, V.Prabakaran, B. Sivakunar, and G.jeeva, “A Review Paper on EDM Parameter of Composite material and Industrial Demand Material Machining,” materialstoday proceedings, Vol. 5, Issue 2, Part1, pp. 5513, 2018.
[3]A. Caggiano, F. Napolitano , R. Teti, S. Bonini and U. Maradia,”Advanced die sinking EDM process monitoring based on anomaly detection for online identification of improper process conditions,” Procedia CIRP Vol. 88, pp. 381-386, 2020.
[4]A. Caggianoa, R. Tetia, R. Perezb, and P. Xirouchakisc, “Wire EDM Monitoring for Zero-Defect Manufacturing based on Advanced Sensor Signal Processing,” 9th CIRP Conference on Intelligent Computation in Manufacturing Engineering Procedia CIRP, Vol. 33, pp. 315-320, 2015.
[5]J. Gan, G. Wen, H. Yu, W. Zheng and C. Lei, “Supervised feature selection by self-paced learning regression,” Pattern Recognition Letters, August 2018.
[6]R. Dwaraka and N. Arunachalam, “Investigation on non-invasive process monitoring of Die Sinking EDM using Acoustic Emission signals,” Procedia Manufacturing Vol. 26, pp. 1471-1482, 2018.
[7]張廷彥，“微放電加工效率與能力提昇之研究”，國立臺灣大學機械工程研究所碩士論文，2010.
[8]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.
[9]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.
[10]W. König, R. Wertheim, A. Wei ,”Funkenerosive Bearbeitung von Hartmetall,” VS Verlag fiir Sozialwissenschaften, pp. 123 1974.
[11]T. Bergs, M. Mohammadnejad, M. Witteler, L. Heidemanns and A. Klink, “Experimental Investigation on Process Signature for EDM Processes,” Procedia CIRP Vol. 87, pp. 273-278, 2020.
[12]J. Hodge and J. Austin, “A Survey of Outlier Detection Methodologies,” Artificial Intelligence Review Vol. 22, pp. 85-126, 2004.
[13]J. Francis, D. Addison, S. Wermter, and J. MacIntyre, “Effectiveness of Feature Extraction in Neural Network Architectures for Novelty Detection,” Proceedings of the ICANN Conference, 1999.
[14]M. Kopp, T. Pevný and M. Holeňa, “Anomaly explanation with random forests,” Expert Systems with Applications Vol. 149, July 2020.
[15]Z. Zhang, Z.Yang, W. Ren and G. Wen, “Random forest-based real-time defect detection of Al alloy in robotic arc welding using optical spectrum,” Journal of Manufacturing Processes. Vol. 42, pp. 51-59, June 2019.
[16]T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” University of Washington, 2016.
[17]D. Chakraborty and H. Elzarka, “Early detection of faults in HVAC systems using an XGBoost model with a dynamic threshold,” Energy and Buildings Vol. 185, pp. 326-344, Feb 2019.
[18]P. K. Bossom, ” Trends and recent developments in the material manufacture and cutting tool application of polycrystalline diamond and polycrystalline cubic boron nitride,” International Journal of Refractory Metals & Hard Materials Vol. 18 pp. 147-151, 2000.
[19]Y. Jia and J. Li, “Impact Analysis of Electrode Material on Electrical Discharge Grinding Polycrystalline Diamond Cutting Tools,” 19th CIRP Conference on Electro Physical and Chemical Machining, pp. 23-27, April 2018.
[20]X. Wang, S. Yi, M. Easton and S. Ding, “Active gap capacitance electrical discharge machining of polycrystalline diamond,” Journal of Materials Processing Tech, Vol. 280, pp. 2-5, 2020.
[21]L. Breiman, “Classification and Regression Trees,” Wadsworth and Brooks, Pacific Grove CA, 1984.
[22]L. Breiman, “Random Forests,” Machine Learning, pp. 5-32, 2001.
[23]J. Pei, L. Zhang, J. Du, X. Zhuang, Z. Zhou, S. Wu, Y. Zhu, “A model of tool wear in electrical discharge machining process based on electromagnetic theory,” Int. J.Mach. Tools Manuf. Vol. 117, pp. 31–41, 2017.
[24]S. Chen, “Evaluating the Effectiveness of Random Forest Mosel,” Mater Thesis of NCTU, pp. 12-20, 2015.
[25]I. Wu, “Network Synchronization in the Presence of Random Routing Delays,” Mater Thesis of NCTU, pp. 41-43, 2016.