電動馬達為智慧生產製造不可或缺的動力來源，如能於線上檢測出馬達故障及各項非即時錯誤，就能在確切時間完成機台保養及更換，縮短停工影響。馬達聯軸為馬達軸承連結外部機械動力模組間的動力傳動，聯軸受損往往造成轉子偏心現象，傳統的馬達聯軸偏心故障診斷，是利用單一感測器收集馬達訊號加以分析，不過單一訊號可分析的故障種類有限，使用之線上診斷儀器常屬昂貴之設備，本文即在討論僅藉由馬達驅動器已具備之電壓、電流、速度感知器之資訊，以及伺服馬達上層控制命令的結合，針對週期性變速度之電子凸輪應用-飛剪，利用機械學習中隨機森林演算法分類器作聯軸偏心程度分類及判斷，以設計線上機械動力模組聯軸偏心診斷系統。

Electric motors are the important power source for intelligent manufacturing, early detection of emerging problems or non-timely faults of motors can save invaluable time and cost. Motor coupling eccentricity is one of the non-timely faults which can cause damage to power modules. The traditional on-line motor coupling eccentricity diagnosis methods rely on analyzing the information from a single sensor set on the motor. However, the sensors are always expensive and the single-data analyzing is not precise thus can only detect few kinds of motor faults. This research proposes a design methodology of an on-line mechanical power module diagnosis system based on machine learning and real-time raw data with motor controlling signals from the servo driver. The main research focus is on the coupling eccentricity fault classification which is performed using random forests algorithm.

摘要 I
ABSTRACT II
誌謝 XII
目錄 XIII
表目錄 XVII
圖目錄 XVIII
符號表 XX
第一章 緒論 1
1.1研究動機 1
1.2研究背景與文獻回顧 2
1.3研究目的 4
1.4本文架構 5
第二章 資料收集平台與聯軸偏心 6
2.1 伺服馬達包裝機系統 6
2.1.1台達電子三伺服枕式包裝機 6
2.1.2台達電子ASD-M-1521-R伺服驅動器 8
2.1.3台達電子ASDA-Soft 10
2.2 週期性變速度飛剪聯軸偏心探討 11
2.2.1 週期性變速度飛剪 11
2.2.2 飛剪聯軸偏心問題 12
2.2.3 本研究飛剪各程度偏心狀態實驗設計說明 13
2.3 學習模型所需原始訓練及測試資料蒐集 15
2.3.1 監控變數選擇 15
2.3.2 原始訓練資料蒐集程序 17
2.3.3 原始測試資料蒐集程序 18
第三章 線上機械動力模組聯軸偏心診斷系統 19
3.1 隨機森林演算法 19
3.1.1 決策樹 19
3.1.2 隨機森林演算法 21
3.1.3 隨機森林與決策樹優劣比較 22
3.1.4 Scikit_Learn Python開源機械學習套件 23
3.1.5 Scikit-Learn OOB_Score 27
3.2 線上動力模組聯軸偏心診斷系統架構與設計 27
3.2.1 系統架構 27
3.2.2 狀態投票檢測法 28
第四章 實驗設計與結果分析及討論 30
4.1 實驗流程及背景說明 30
4.1.1 實驗流程 30
4.1.2 實驗使用軟硬體 31
4.1.3實驗背景 32
4.2 測試及訓練資料預處理及特徵工程 32
4.2.1 測試資料預處理 32
4.2.2 訓練資料預處理 33
4.2.3 特徵工程 34
4.3訓練及測試資料進階預處理-取脈波頻率峰值範圍下特徵值資料 36
4.3.1 訓練資料進階預處理 36
4.3.2 測試資料進階預處理 37
4.3.3 訓練資料進階預處理下學習模型各項實驗結果與比較 38
4.4 訓練資料量大小選擇 39
4.4.1 訓練資料量預處理 39
4.4.2 訓練資料量大小不同下學習模型各項實驗結果 40
4.5 訓練模型優化 42
4.5.1 以n_estimator做訓練模型優化 43
4.6 線上動力模組聯軸偏心診斷系統實現 44
第五章 結論與未來建議 45
5.1 結論 45
5.2 未來建議 45
參考文獻 47
附錄 49
附件一、台達電子ASDA-M-1521-R伺服驅動器規格表[16] 49
附件二、ASDA-SOFT軟件內建示波器可監視之狀態變數及其說明[22] 52
附件三、ASDA-SOFT搜集之原始資料TXT檔案樣本(部分) 56
附件四、訓練及測試資料CSV檔案樣本(部分) 57
附件五、進階訓練資料預處理PYTHON程式函式樣本 58
附件六、進階測試資料預處理PYTHON程式函式樣本 60
附件七、模型訓練PYTHON程式樣本 61
附件八、模型測試PYTHON程式樣本 62
附件九、線上動力模組聯軸偏心系統PYTHON程式樣本 63

[1]MBAlib-智慧產業. Available: http://t.cn/Rr4KD4J，Feb. 16, 2017
[2]P. J. Tavner and J. Penman, Condition Monitoring of Electrical Machines, Herfordshine, England: Research Studies Press Ltd., 1987.
[3]P. Vas, Parameter Estimate, Condition Monitoring, and Diagnosis of Electrical Machines, Clarendon Press Oxford, 1993.
[4]葛世偉,李明遠, 馬達軸承故障線上診斷技術研究, 工程科技與教育學刊, vol. 12, pp. 1-7.
[5]O. Munoz-Ornelas, D. A. Elvira-Ortiz, R. A. Osornio-Rios, R. J. Romero-Troncoso, and L. A. Morales-Hernandez, Methodology for Thermal Analysis of Induction Motors with Infrared Thermography Considering Camera Location, 42nd Annual Conference of the IEEE Industrial Electronics Society (IECON 2016), pp. 7113-7118, Florence, Italy, 2016.
[6]A. F. Browne, S. Patel, S. A. Kaloti, S. M. Lopez, and P. Parikh, Area Acoustic and Electromagnetic Emissions Monitoring of 3-phase Motors, 2017 IEEE International Electric Machines and Drives Conference (IEMDC), pp. 1-6, Miami, FL, USA , 2017.
[7]I. Y. Önel and M. E. H. Benbouzid, Induction Motor Bearing Failure Detection and Diagnosis: Park and Concordia Transform Approaches Comparative Study, IEEE/ASME Transactions on Mechatronics, vol. 13, no. 2, pp. 257-262, 2008.
[8]M. E. H. Benbouzid, A Review of Induction Motors Signature Analysis as a Medium for Faults Detection, IECON '98. Proceedings of the 24th Annual Conference of the IEEE Industrial Electronics Society (Cat. No.98CH36200), vol. 4, pp. 1950-1955, 1998.
[9]X. Ren, W. Shao, and W. Z. Yang, Extraction of Gearbox Fault Features from Vibration Signal Using Wavelet Transform, Journal of Physics.F. Q. Su : Conference Series, vol. 48, pp. 490-494, 2006.
[10]G. Ferenc, M. Lutovac, V. Kvrgic, and P. Stepanic, A Proposed Approach to the Classification of Bearing Condition Using Wavelets and Random Forests, 2013 2nd Mediterranean Conference on Embedded Computing (MECO), pp. 140-143, 2013.
[11]Z. Wang, Q. Zhang, J. Xiong, M. Xiao, G. Sun, and J. He, Fault Diagnosis of a Rolling Bearing Using Wavelet Packet Denoising and Random Forests, IEEE Sensors Journal, vol. 17, no. 17, pp. 5581-5588, 2017.
[12]L. Wen, X. Li, L. Gao, and Y. Zhang, A New Convolutional Neural Network-Based Data-Driven Fault Diagnosis Method, IEEE Transactions on Industrial Electronics, vol. 65, no. 7, pp. 5990-5998, 2018.
[13]Y. Lei, F. Jia, J. Lin, S. Xing, and S. X. Ding, An Intelligent Fault Diagnosis Method Using Unsupervised Feature Learning Towards Mechanical Big Data, IEEE Transactions on Industrial Electronics, vol. 63, no. 5, pp. 3137-3147, 2016.
[14]TSE-3100振動計，儀立購有限公司. Available: http://t.cn/RBYsbA2
[15]Delta AC Servo Drive ＆ Motor ASDA-A2 Series. 2017.
[16]台達交流伺服系統ASDA-M系列產品說明書. 2014.
[17]C. Bulac and A. Bulac, Decision Trees, pp. 819-844, 2016.
[18]L. Breiman, Random forests, Machine Learning 45(1), pp. 5-32, 2001.
[19]L. Breiman, Machine Learning, Begging preditors, pp. 123-140., 1996.
[20]Y. Ang, Z.H. Zhou, and A. Zhou, Machine Learning and Its Application, Beijing:Tsinghua University Press, pp. l-40, 2006.
[21]Cherzhoucheer. (2017). Sklearn-RandomForest随机森林. Available: https://blog.csdn.net/CherDW/article/details/54971771
[22]ASDA-Soft軟體操作手冊. 2013.