臺灣博碩士論文加值系統

本論文開發一套應用於鑽石磨粒線的光學檢測系統，可進行即時檢測並設計介面操作與觀察，可取得磨粒分布、線徑和鍍層厚度等關鍵參數進行製程統計。檢測表面磨粒時，最大的挑戰就是電鍍的鑽石磨粒線之圓弧表面不同段落可能具有不同的反射係數，且在圓弧表面的頂端易曝光過度，但邊緣曝光量仍然較少。無法固定曝光條件檢測，是表面磨粒檢測不準確的主因，對此使用曝光融合方法結合深度學習，曝光融合方法不需要使用相機響應曲線校準，拍攝不同曝光時間的照片，並以指標為疊合權重，藉此可得到曝光程度均勻且清晰的圖像，再以瑕疵檢測深度學習網路分析，比較其不同反射條件、不同的光源以及不同的曝光時間對深度學習模型的影響，並採預測標記與人工標記面積重疊率、溢出的比例及分析召回率、精準度。實驗結果得知，鑽石磨粒線的光學檢測系統具有準確度且重現性高，而深度學習檢測召回率與精確度達93%，且透過曝光融合，辨識的特徵接近人為判斷結果。

This thesis develops a set of optical inspection system applying on diamond wire saw, which can test and observe with graphical user interface and acquire key parameters such as abrasive distribution, abrasive protrusion, wire diameter and coating thickness by analyzing collected data. During detecting surface abrasives, the most challenging factor is the surface of the plated diamond wire may have different reflection coefficients in different sections. Moreover, the top of the surface of arc is tend to over-exposed, while the exposure intensity at the wire edge is rather low. Hence, the none-constant exposure condition is the main cause of inaccurate surface abrasive detection. Therefore, this system employs the exposure fusion method with combining deep learning method. Therefore, camera response to curve calibration is no longer required. By taking photos with different exposure time, and superimpose the images with different indicators to get a clear and evenly exposed image of diamond wire saw. And then the flaw detection results were compared with the influence of different reflection condition, different source of light and different exposure time based on deep learning model. And further comparison between the area of prediction mark and the manual mark overlaps, the ratio of the overflow, the recall rate, and the accuracy were conducted. The experimental result shows that the proposed optical inspection system of the diamond abrasive grain line has high accuracy and reproducibility. The recall and precision rate of deep learning detection result achieve 93%. Through the exposure fusion, the identified features are similar to human judgment.

摘要 i
ABSTRACT ii
致謝 iv
目錄 v
圖目錄 vii
表目錄 x
第一章 緒論 1
1.1 前言 1
1.2 研究動機與目的 3
1.3 文獻回顧 4
1.3.1 光學檢測領域 4
1.3.2 影像處理領域 6
1.3.3 深度學習領域 8
1.4 研究方法與架構 11
第二章 檢測方法與原理 12
2.1 線上自動光學檢測系統 12
2.1.1 線徑檢測影像處理 14
2.1.2 顆粒檢測影像處理 23
2.2 深度學習離線檢測 24
2.2.1 曝光融合演算法 24
2.2.2 深度學習檢測架構 27
2.2.3 網路架構 27
2.2.4 資料集 30
2.2.5 訓練方法 32
2.2.6 預測方法 33
2.2.7 評估方法 34
第三章 鑽石磨粒線的檢測架構 36
3.1 取像設備 36
3.2 鏡頭組 37
3.3 光源系統 38
3.4 線上檢測光機架構 40
3.5 深度學習檢測光機架構 42
第四章 實驗結果與探討 43
4.1 線上檢測結果 43
4.1.1 系統測試結果 43
4.1.2 驗證與重現性試驗 47
4.1.3 誤差討論 51
4.2 深度學習離線檢測結果 55
4.2.1 曝光融合取像結果 55
4.2.2 曝光時間對深度學習檢測結果影響比較 57
4.2.3 光源與反射性質對深度學習檢測結果影響比較 59
4.2.4 深度學習參數對檢測之影響 64
第五章 結論與未來展望 66
5.1 結論 66
5.2 未來展望 67
參考文獻 68
附錄:口試Q&A 70

[1]M. Nakai et.al, “Development of a fixed diamond wire saw for electronics application,” Osaka Diamond Co., Ltd Japan, pp. 233~240, 2001.
[2]李志士等人，「鑽石顆粒電鍍於切割線之研究」，2013綠色科技工程與應用研討會(GTEA)，2013年。
[3]K. Ishikawa, H. Suwabe, K. Kanayama, “Study on machining characteristics of wire tool with electrodeposited diamond grains,” JSPE, vol.60, no. 3, pp. 329~334, May 1994.
[4]A. Mital, M. Govindaraju, B. Subramani, “A comparison between manual and hybrid methods in parts inspection,” Integrated Manufacturing Systems, vol. 9, pp. 344-349, 1998.
[5]P. Kopardekar, A. Mital, S. Anand “Manual, hybrid and automated inspection literature and current research, ” Integrated Manufacturing Systems, vol. 4, pp. 18-29, 1993.
[6]張振堶、金美敬，「自動光學檢測(AOI)市場及技術發展趨勢調查」，工研院IEK系統能源組技術報告，2002年。
[7]Krizhevsky, I. Sutskever, G. Hinton, “ImageNet classification with deep convolutional neural networks,” Proceedings of the 25th International Conference on Neural Information Processing Systems, vol. 60, pp. 1097-1105, 2012.
[8]K. He, X. Zhang, S. Ren, J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778, 2016.
[9]Byrne, G & Sheahan, Con. “Comparative analysis of vision systems for electroplating surface quality inspection,” International Journal of Production Research, vol. 43, pp. 3787-3801, 2005.
[10]鄭絢陽，「基於機械視覺之鑽石切割線製程優化」。國立臺北科技大學機械工程系機電整合碩士班碩士論文，2018年。
[11]T. Mertens, J. Kautz and F. V. Reeth, "Exposure Fusion," 15th Pacific Conference on Computer Graphics and Applications (PG'07), Maui, HI, pp. 382-390, 2007.
[12]M. Bevilacqua, G. Di Leo, M. Landi and A. Paolillo, "Measurements on metal-rubber profiles with HDR and EF imaging techniques," 2012 IEEE International Conference on Imaging Systems and Techniques Proceedings, pp. 403-408, 2012.
[13]G. Huang, Z. Liu, K.Q. Weinberger, L. van der Maaten, “Densely connected convolutional networks,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, vol.1, pp. 3,2017.
[14]蘇詠靖、游原瑋、何昭慶，「結合機器視覺與深度學習之金屬圓柱表面缺陷檢測系統」，儀科新知季刊，第兩百一十五期，第46-58頁，2018年。
[15]Nobuyuki Otsu,” A threshold selection method from gray-level histograms,” IEEE Trans. Sys., Man., Cyber., vol.SMC-9, no.1, pp. 62–66, 1979.
[16]T. Y. Zhang and C. Y. Suen. , “A fast parallel algorithm for thinning digital patterns” , Commun. ACM 27, vol.3, pp. 236-239, 1986.
[17]Basics lighting . (n.d.). Retrieved July 24, 2019, from https://www.opto-e.tw/ media/downloads/docs/Basics_Lighting_TW.pdf/
[18]P. Burt and T. Adelson. “The laplacian pyramid as a compact image code,” IEEE Transactions on Communication, pp. 532-540, 1983.
[19]D. Weimer, H. Thamer, and B. Scholz-Reiter,” Learning defect classifiers for textured surfaces using neural networks and statistical feature representations,” Procedia CIRP, vol.7, pp. 347-352, 2013.