臺灣博碩士論文加值系統

　　軸承用於承受旋轉負荷，若發生無預期損壞，將造成機械系統停機。因此軸承實為機械系統之關鍵元件。若能快速有效診斷軸承的可能破壞狀況，以利故障排除及縮短維修時間，當可提高機械的運轉的效率。然而由於軸承的多樣性與不同破壞模式的組合，目前智慧診斷軸承破壞模式仍以建立該軸承特定模型為主，透過某些特定特徵來辨識破壞狀況；然這些特徵選擇仍有其不確定性，從而限制智慧型故障診斷方法的實用性。
　　因此，本研究提出二階段軸承破壞診斷架構，可藉由前階段的快速特徵自動選擇，與後階段的錯誤智慧診斷，可有效地自動分類不同軸承錯誤模式。在建立診斷模型期間，軸承破壞之時域振動值，將轉換成時域與頻域特徵以構成候選特徵。由於候選特徵眾多，因此前階段先利用最鄰近演算法並結合不同特徵選擇法，如主成份分析、基因演算法、或關鍵變異數等，來萃取關鍵性而有助於分類辨識的特徵。經萃取後特徵之關鍵特徵將輸入後階段的倒傳遞類神經網路，以建構軸承之振動破壞模型。在應用時，可僅輸入前階段的關鍵特徵，藉由後階段模型加以診斷分類軸承之可能破壞模式。
　　在研究成果上，以案例一而言，其七種多重錯誤模式與尺寸之訓練分類正確率為98.9%，測試正確率為94.5%，並可將時間收斂至純粹倒傳遞類神經網路的8.1%。綜合各種法所找出的10個關鍵特徵，訓練正確率96.15%，測試正確率平均91.79%；就案例二而言，其三種單一錯誤於不同負載與尺寸，綜合各種方法所找出的7個關鍵特徵，訓練正確率達98.3%，而測試正確率為92.0%。因此，所提出之軸承診斷架構，確可自動獲得關鍵特徵，並有效運用於診斷軸承之損壞，以診斷其破壞模式，從而提高診斷不同軸承條件與破壞組合的實用性。

　　A Bearing used to support rotating loading, when an unexpected bearing failure causes machining system shut-down, is a critical part of machining system. To improve the machining system utilization, we can shorten maintenance time if the bearing failures effectively could be diagnosed and fixed in time. However, due to combinations of bearing varieties and failures, current intelligent diagnosis of bearing failures is to build a specific bearing model with the specific features to identify the failures. The feature selection exists
uncertainty, which limits practicality of the intelligent diagnosis applications.
　　Hence, this study proposes a dual-stage structure, pre-stage automatic features selection and post-stage intelligent failure diagnosis, for effectively identifying bearing failure. In building diagnosis model, the failure features including time-domain and frequency-domain features are extracted from bearing vibration data to form the candidate features. To reduce the numbers of the input features, the proposed structure adopts K-NN method and different reduction methods such as Principle Component Analysis, Genetic Algorithm and variance analysis to automatically select the critical and identifiable features in the pre-phase. Then, a Back Propagation Neural Network is trained by the selected
features to build a diagnosis model in the post-stage.
　　In classification results, using the proposed dual-stage diagnosis structure to automatically select features, case 1 indicates that the mean accuracy of seven failure modes of training and testing are 98.8% and 94.5%, respectively. The training time is reduced to 8.1% of a pure Back Propagation Neural Network. When using top ten critical features identified from the proposed structure, the mean accuracy of training and testing are 96.15% and 91.79%, respectively. For three failures in various loadings and failure sizes, case 2 shows that using the top seven features mean accuracy of training and testing are 96.2% and 91.8%, respectively. Therefore, the proposed diagnosis structure can automatically extract the critical features and apply that to diagnose various bearing failures and improve the intelligent diagnosis practicality.

摘要 ................................................... i
ABSTRACT .............................................. ii
致謝 ................................................. iii
目錄 .................................................. iv
圖目錄 ................................................. v
表目錄 ................................................ vi
符號說明 ............................................. vii
第一章 緒論 ............................................ 1
1.1 研究背景 ........................................... 1
1.2 問題定義 ............................................6
1.3 方法重點 ........................................... 6
1.4 研究架構 ........................................... 7
第二章 研究設計與方法 .................................. 9
2.1 系統架構 ........................................... 9
2.2 時域分析法(Time Domain Analysis) .................. 10
2.3 軸承振動特徵頻率 .................................. 15
2.4 快速傅立葉轉換(Fast Fourier Transform, FFT)演算法 . 22
2.5 頻譜分段 .......................................... 25
2.6 取樣定理 .......................................... 25
2.7 主成份分析(principal component analysis) .......... 26
2.8 特徵變數組合方法 .................................. 28
2.9 基因演算法(GA)..................................... 29
2.10 關鍵變異分析 ..................................... 31
2.11 K-Nearest Neighbors(KNN)演算法 ................... 32
2.12 倒傳遞類神經網路(Back-propagation Neural Network) 33
第三章 案例分析 ....................................... 37
3.1 研究室實驗數據 .................................... 37
3.2 研究室實驗架構 .................................... 37
3.3 凱斯西儲大學軸承振動數據 .......................... 43
3.4 實驗結果分析-自測數據 ............................. 45
3.5 實驗結果分析-CWRU數據 ............................. 63
第四章 結論與未來方向 ................................ 69
4.1 結論 ............................................. 69
4.2 未來方向 ......................................... 70
參考文獻 .............................................. 71

[1]. S. A. McInerny and Y. Dai,“Basic Vibration Signal Processing for Bearing Fault Detection,”IEEE Transactions on Semiconductor Manufacturing, vol. 46, no. 1, pp. 149-156, 2003.
[2]. T. Doguer and J. Strackeljan,“Vibration Analysis using Time Domain Methods for the Detection of small Roller Bearing Defects,” SIRM 2009 - 8th International
Conference on Vibrations in Rotating Machines, Vienna, Austria, Paper ID-16, February 23–25, 2009.
[3]. H. Ahmadi and K. Mollazade, “A practical approach to electromotor fault diagnosis of Imam Khomaynei silo by vibration condition monitoring,” African Journal of
Agricultural Research, vol. 4, no. 4, pp. 383-388, April 2008.
[4]. B. Li, M.-Y. Chow, Y. Tipsuwan, and James C. Hung, “Neural-Network-Based Motor Rolling Bearing Fault Diagnosis,” IEEE Transactions on Semiconductor
Manufacturing, vol. 47, no. 5, pp. 1060-1069, 2000.
[5]. C. S. Tyagi, “A Comparative Study of SVM Classifiers and Artificial Neural Networks Application for Rolling Element Bearing Fault Diagnosis using Wavelet Transform
Preprocessing,” Proceedings of World Academy of Science, Engineering and Technology, vol. 33, pp. 319-227, 2008.
[6]. B. Samanta, Khamis R. Al-Balushi and Saeed A. Al-Araimi, “Bearing Fault Detection Using Artificial Neural Networks and Genetic Algorithm,” Journal on Applied Signal Processing, p.366–377, January 2004.
[7]. B. Sreejith, A.K. Verma, and A. Srividya, “Fault diagnosis of rolling element bearing using time-domain features and neural networks,” IEEE Region 10 Colloquium and the Third ICIIS, Kharagpur, PaperID.409, December 8-10, 2008.
[8]. http://www.eecs.case.edu/laboratory/bearing/
[9]. J. C. García-Prada, C. Castejón, O. J. Lara, Incipient bearing fault diagnosis using DWT for feature extraction,” 12th IFToMM World Congress, Besançon (France), June18-21, 2007.
[10]. J. Chebil, G. Noel, M. Mesbah and M. Deriche, “Wavelet Decomposition for the Detection and Diagnosis of Faults in Rolling Element Bearings,” Journal of Mechanical and Industrial Engineering, vol. 3, no. 4, pp.260-267, 2009.
[11]. Z.-Y. Yang, “A Study of Rolling-Element Bearing Fault Diagnosis Using Motor’s Vibration and Current Signatures,” the 7th IFAC Symposium on Fault Detection,
Supervision and Safety of Technical Processes Barcelona Spain, pp.354-359, June 30 - July 3, 2009.
[12]. M. Analoui, and M. Fadavi Amiri,“Feature Reduction of Nearest Neighbor Classifiers using Genetic Algorithm,”World Academy of Science, Engineering and Technology,
pp.36-39, vol. 17, 2006.
[13]. 蘇民揚 等，“使用改良型基因演算法於網路入侵偵測系統之特徵選取”，銘傳大學資訊工程學系，TANET2007臺灣網際網路研討會論文集［一］ 。
[14]. 王進德、蕭大全，2003，“類神經網路與模糊控制理論入門”，全華科技圖書股份有限公司。
[15]. 連國珍，1995，“數位信號處理簡介”，茂昌圖書有限公司。
[16]. 張思揚、匡芳君及徐蔚鴻，“基于 WPD 和模糊神經網路的軸承故障診斷”，湖南科技大學學報(自然科學版)，第 28至 31頁，第 25卷第 2期，2010年 6月。
[17]. V. Wyk, B. J, M. A. van Wyk, J. J. Naude, and F. Perrier, “A variable kernel classifier for bearing faults diagnosis using simple statistical features,” Proc. of the 16th Annual Symposium of the Pattern Recognition Association of South Africa, Langebaan, South
Africa, pp.39-45, 2005.
[18]. E. Y. Kim, Andy C. C. Tan, B.-S. Yang, and V. Kosse, “Experimental Study on Condition Monitoring of Low Speed Bearings: Time Domain Analysis,” 5th Australasian Congress on Applied Mechanics, ACAM 2007, Brisbane, Australia, pp.108-113, 10-12 December 2007.