臺灣博碩士論文加值系統

車用單向軸承為汽車變速箱內重要零件之一，車用單向軸承的瑕疵不僅影響了汽車的性能亦有安全的疑慮，早期對於車用單向軸承之品檢方式多由檢測人員以目視進行檢測，此方法常因目視時間過長造成眼睛疲勞進而發生漏檢之情況，因此本研究開發出一套自動化光學檢測系統用於取代人工目視檢測。
車用單向軸承表面具有反光之特性，加上製造中產生的油汙，會增加影像處理、邏輯法則和訂定條件的困難度，為了降低程式的繁雜處理程序，本研究導入人工智慧技術中的深度學習，以神經網路模型和卷積神經網路模型分別對影像處理數據及影像資料進行訓練學習，最後做出決策。
在自動化光學檢測系統的部份，該系統能準確的辨識出車用單向軸承中缺件的位置並標示出來，其檢測準確率、過檢率與漏檢率分別為99.96%、0.04%與0%。在類神經網路的部分，利用自動化光學檢測系統取得應有軸針位置處之影像資訊並將其匯入模型進行訓練，此方法不受邏輯法則與檢測參數條件的影響，其模型預測準確率為99.97%。在卷積神經網路的部分，是透過自動化光學檢測系統裁剪出應有軸針位置之影像，並直接利用該影像進行模型之訓練，其模型預測準確率達99.98%。

The unidirectional bearing for the automobile is one of the important parts in the automobile transmission. The quality of the unidirectional bearing of the vehicle not only affects the performance of the vehicle but also has safety concerns. For the inspection method of unidirectional bearing, most of them are visually inspected by the inspectors. However, visual fatigue will lead to false detection. Therefore, an automated optical inspection system was proposed in this study to replace manual visual inspection.
The surface of the unidirectional bearing has reflective characteristics, and the oil generated in the manufacturing process will increase the difficulty of detection. In order to reduce the influence of reflection and environment, this study uses the artificial neural network and convolutional neural network to establish the prediction model.
The inspection results will be divided into three parts, namely automatic optical detection system, artificial neural network and convolutional neural network. In the part of the automatic optical inspection system, the system can accurately identify the location of the defect in the unidirectional bearing and mark it out. The detection accuracy and over-detection rate are 99.96% and 0.04%, respectively.
In the artificial neural network and the convolutional neural network, the neural network model use image information or images to train the model. These methods are not affected by the environment and detection parameters, and the prediction accuracy of the artificial neural network and the convolutional neural network is 99.97 % and 99.98%, respectively.

中文摘要 I
Abstract II
致謝 IV
目錄 V
圖目錄 VIII
表目錄 XI
符號說明 XII
第一章 緒論 1
1-1 前言 1
1-2 研究動機 2
1-3 文獻回顧 3
1-4 論文架構 12
第二章 機械視覺與基礎概論 14
2-1 光源照明系統 15
2-2 影像形成裝置 17
2-3 數位影像處理 18
2-3-1 灰階 18
2-3-2 二值化 19
2-3-3 圖塊分析(Blob Analysis) 21
2-4 類神經網路 22
2-4-1 人工神經元 22
2-4-2 激活函數 23
2-4-3 損失函數 25
2-4-4 優化器 26
2-5 卷積神經網路 27
2-5-1 卷積層 27
2-5-2 池化層 28
2-5-3 全連接層 30
第三章 實驗設備與軟體 31
3-1 硬體設備介紹 31
3-2 軟體工具 36
3-2-1 C# 36
3-2-2 Visual Basic 6.0 36
3-2-3 Matrox Image Library 5.12 36
3-2-4 Python3.5 37
3-2-5 Keras 37
3-2-6 MATLAB R2019a 37
3-3 研究載具簡述 38
第四章 研究方法 39
4-1 影像擷取系統 39
4-2 光學檢測系統 40
4-2-1 待測件中心定位 41
4-2-2 以參考位置(支架鐵)預測軸針位置 42
4-2-3 圖塊分析(Blob analysis) 43
4-2-4 檢測軸針數量 44
4-2-5 軸針進階辨識 45
4-3 類神經網路架構 49
4-2-1 數據資料預處理 49
4-2-2 建立模型 49
4-2-3 訓練與驗證 49
4-4 卷積神經網路架構 50
4-3-1 圖像資料取得 50
4-3-2 圖像資料預處理 51
4-3-3 卷積神經網路模型 51
4-5 田口方法 53
第五章 結果與討論 54
5-1 自動化光學檢測結果 54
5-1-1 系統檢測 55
5-1-2 檢測性能 57
5-1-3 誤判分類 58
5-2 類神經網路檢測結果 61
5-2-1 自開發類神經網路模型之驗證 61
5-2-2 神經網路模型優化 62
5-3 卷積神經網路檢測結果 68
5-4 三種不同檢測方法之比較 72
第六章 結論與未來展望 74
6-1 結論 74
6-2 未來展望 75
參考文獻 76

[1]H. Zheng, L.X. Kong, and S. Nahavandi, 2002, “Automatic inspection of metallic surface defects using genetic algorithms,” Journal of Materials Processing Technology, Vol. 125-126, pp. 427-433.
[2]H. Shen, S. X. Li, D. Y. Gu, and H. X. Chang, 2012, “Bearing defect inspection based on machine vision,” Measurement, Vol. 45, pp. 719-733.
[3]G. L. Peng, Z. J. Zhang, and W. Q. Li, 2016, “Computer vision algorithm for measurement and inspection of O-rings,” Measurement, Vol. 94, pp. 828-836.
[4]R. Shanmugamani, M. Sadique, and B. Ramamoorthy, 2015, “Detection and classification of surface defects of gun barrels using computer vision and machine learning,” Measurement, Vol. 60, pp. 222-230.
[5]F. Zhong, C. N. Dai , T. He, and Q. H. Wu, 2007, “One Method of Bearing Outside-Diameter Detection Based on Hough Transform,” IEEE International Conference on Automation and Logistics, Jinan, INSPEC Accession No. 9880714, 2007/08/18~21.
[6]M. Ferguson, R. Ak, Y. T. T. Lee, and K. H. Law, 2017, “Automatic Localization of Casting Defects with Convolutional Neural Networks,” IEEE International Conference on Big Data, pp. 1726-1735.
[7]L. Shang, Q. Yang, J. Wang, S. Li, and W. Lei, 2018, “Detection of Rail Surface Defects Based on CNN Image Recognition and Classification,” International Conference on Advanced Communications Technology, pp. 45-51.
[8]X. Tao, D. Zhang, W. Ma, X. Liu, and D. Xu, 2018, “Automatic Metallic Surface Defect Detection and Recognition with Convolutional Neural Networks,” applied sciences, Vol. 8, Issue. 9, 1575.
[9]M. Zhang, J. Wu, H. Lin, P. Yuan, and Y. Song, 2017, “The Application of One-Class Classifier Based on CNN in Image Defect Detection,” Procedia Computer Science, Vol. 114, pp. 341-348.
[10]S. Zhou, Y. Chen, D. Zhang, J. Xie, and Y. Zhou, 2017, “Classification of Surface Defects on Steel Sheetusing Convolutional Neural Networks,” Materials and technology, Vol. 51, pp. 123-131,.
[11]黃成凱、黃緒哲、嚴健榮、古有彬、楊淳宜、陳俊皓，2018，“基於深度學習之工件自動辨識與取料之應用” ，機械工業雜誌，工業技術研究院，第428期，85-95頁。
[12]L. Song, X. Li, Y. Yang, X. Zhu, Q. Guo, and H. Yang, 2018, “Detection of Micro-Defects on Metal Screw Surfaces Based on Deep Convolutional Neural Networks,” sensors, Vol. 18, Issue. 11, 3709.
[13]S. Indolia, A. K. Goswami, S. P. Mishra, and P. Asopa, 2018, “Conceptual Understanding of Convolutional Neural Network- A Deep Learning Approach,” Procedia Computer Science, Vol. 132, pp. 679-688.
[14]W. Chen, Y. Gao, L. Gao, and X. Li, 2018, “A New Ensemble Approach based on Deep Convolutional Neural Networks for Steel Surface Defect classification,” Procedia CIRP, Vol. 72, pp. 1069-1072.
[15]蘇詠靖、游原瑋、何昭慶，2018，“結合機器視覺與深度學習之金屬圓柱
表面缺陷檢測系統，”，科儀新知，第215期，財團法人國家實驗研究院。
[16]D. Weimer, Bernd Scholz-Reiter, and M. Shpitalni, 2016, “Design of deep convolutional neural network architectures for automated feature extraction in industrial inspection,” CIRP Annals - Manufacturing Technology, Vol. 65, pp. 417-420.
[17]程冠倫、梁碩芃，2018， “應用類神經網路於刀具磨耗估測，”，機械工業雜誌，第428期，36-42頁，工研院。
[18]張傳育、黃振珂、王薇鈞，2017，“應用深度學習檢查輪胎缺陷，”，全國計算機會議論文集，400-404頁，國立雲林科技大學。
[19]龚丁禧，2014，“基於卷積神經網路的植物葉片分類，” 計算機與現代化，pp.12-15。
[20]石明于、湯燦泰、黃茂裕、黃泰惠、簡珮珊，2018，“人工智慧於PCB 瑕疵覆判之應用，”電工通訊季刊，81-95頁。
[21]李兆健，2017，卷積神經網路應用於中文字手寫風格辨識，碩士論文，國立成功大學工程科學學系。
[22]馬朝勛，2019，以深度學習作工業檢測應用，碩士論文，國立雲林科技大學電機工程系。
[23]W. Liua, Z. Wanga, X. Liua, N. Zengb, Y. Liuc, and F. E. Alsaadid, 2017, “A survey of deep neural network architectures and their applications,” Neurocomputing, Vol. 234, pp. 11-26.
[24]Y. Guo, Y. Liu, A. Oerlemans, S. Lao, S. Wu, and M. S. Lew, 2016, “Deep learning for visual understanding : Areview,” Neurocomputing, Vol. 187, pp. 27-48.
[25]V. Natarajan, T. Y. Hung, S. Vaikundam, and L. T. Chia, 2017, “Convolutional Networks for Voting-based Anomaly Classification in Metal Surface Inspection” IEEE International Conference on Industrial Technology, pp. 986-991.
[26]M. Yurtsever, and U. Yurtsever, 2019, “Use of a convolutional neural network for the classification of microbeads in urban wastewater,” Chemosphere, Vol. 216. pp.271-280.
[27]K. Kamnitsas, C. Ledig, V. F. J. Newcombe, J. P. Simpson, A. D. Kane, D. K. Menon, D. Rueckert, and B. Glocker, 2017, “Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation,” Medical Image Analysis, Vol. 36, pp. 61-78.
[28]Q. Li, W. Cai, X. Wang, Y. Zhou, D. D. Feng, and M. Chen, 2014, “Medical Image Classification with Convolutional Neural Network,” International Conference on Control, Automation, Robotics and Vision, Singapore, 2014/12/10~12.
[29]王健新、楊慧珍、白福忠編著，2013，視覺測量技術基礎，電子工業出版社，pp. 61-80
[30]彩色與數位影位概念，http://163.27.124.4/~chang/graphic/1-2.pdf
[31]A. C. Bovik, 2009, The Essential Guide to Image Processing (Second Edition), Academic Press, pp. 69-95.
[32]Matrox Electronic Systems Ltd., 2005, Matrox Imaging Library User Guide, Matrox Electronic Systems Ltd..
[33]斎藤康毅編著，2017，用Python進行深度學習的基礎理論實作，碁峰資訊股份有限公司，pp.78-82。
[34]林大貴編著，2017，深度學習人工智慧實務應用，博碩文化股份有限公司，pp.20-21。
[35]Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, 1998 “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, Vol. 86, pp. 2278-2324.
[36]吳男爵，2017，具自動化校正與光點引導之三維量測系統應用於金屬製品量測之研究，碩士論文，國立高雄應用科技大學機械工程系研究所。
[37]張祐慈，2013，非接觸式表面瑕疵檢測精密度提升之研究，碩士論文，國立高雄應用科技大學機械與精密工程研究所。
[38]葉難編著，2015，Python程式設計入門，博碩文化股份有限公司。
[39]周品、何正風編著，2012，MATLAB數值分析，五南圖書出版股份有限公司。