鋼珠（Steel ball）的應用普遍存在於多種日常用品中，而在精密工業中更是一個不可或缺的元件。目前鋼珠主要應用於精密軸承（Bearing）之製造，鋼珠的表面品質與精密度將主宰軸承運轉的磨耗情形以及使用壽命，進而影響整個機械設備的功能，因而顯示鋼珠品質檢測的重要性。鋼珠表面瑕疵包含表面凹陷、過度粗糙以及因金屬疲勞產生的碎裂等，在傳統的檢測方法中多採用人眼檢視方式，容易因疲勞及個人主觀判斷造成誤判或漏檢，因此本研究提出以自動視覺檢測方法改善人眼檢視所造成之缺失。
鋼珠具金屬表面的高反射性，以及形狀為一球面的特性，在本研究中應用機器視覺方法取像時，將探討最佳光源、鏡頭的選擇以及環境的設定，以適合檢測。本研究使用二種方法檢測鋼珠表面之瑕疵，其一為灰階對比度擴展法（Contrast stretching），並以熵函數法（Entropy）進行二值化分割，以區別正常區域與瑕疵；另一方法為灰階切片法（Gray-level slicing）保留瑕疵範圍灰階資訊，再以視窗檢測方法計算視窗內灰階的平均值與變異數，訂定一瑕疵分割臨界值，便可得到瑕疵區塊的面積以及個數。判斷鋼珠好壞的方式，採計算最大瑕疵面積以及平均瑕疵面積兩量測標準進行判定。本研究在適當的影像尺寸與解析度之下，可取得一張包含四顆鋼珠之影像進行檢測，使減少重複取像次數，提高檢測工作之效率。

Steels ball are important elements in a precision machine, especially as components of stainless steel bearings and automotive parts. The quality of steel balls makes a great impact on the operating condition and the life cycle of the composed product. Traditionally, the inspection of steel balls in the manufacturing process requires highly experienced operators with their naked eyes; this method is out of time nowadays. In this research, we propose a machine vision method to tackle the deficiency of manual vision inspection and to achieve high effectiveness and efficiency of automatic inspection.
Owing to the high reflectance and the spherical surface of steel balls, the inspection lighting must be selected appropriately to acquire an optimum image. A ring-shaped LED light along with a dome-shaped unite reflective board is employed to eliminate the reflectance of light and enhance surface texture of the inspection steel ball. The resulting inspection region is a ring-like area under the experiment framework. In this thesis, we use two image enhancement techniques (contrast stretching and gray-level slicing) to stress the difference of gray-level information between the defect and normal areas. According to the contrast stretching and gray-level slicing results, we use Entropy method and gray-level statistics in a neighborhood window, respectively, to compute a threshold value for distinguishing defect areas from the background. Since the surface roughness consists of irregularities such as large scaled crack, dent, cuts and highly concentrated small defects, we compute the maximum defect blob area and average defect blob area as two quantitative measures to determine whether the steel ball under inspection is good or bad. In order to improve the inspection rate, multiple steel balls inspected in the same image is also evaluated using the proposed method. Experimental results have shown that the proposed two inspection methods are able to detect the defects of steel ball effectively for both single and multiple steel ball inspections.

中文摘要 I
英文摘要 II
誌謝 III
目錄 IV
表目錄 VI
圖目錄 VII
第一章 緒論 1
1.1 研究動機與目的 1
1.2 研究方法簡介 3
1.3 論文架構 3
第二章 文獻回顧 4
2.1鋼珠製造程序簡介 4
2.2相關自動檢測技術 5
第三章 研究方法 8
3.1研究方法概述 8
3.2鋼珠環形區域之擷取 16
3.3決定環形檢測區域 18
3.4檢測方法 20
3.4.1灰階對比度擴展結合熵函數二值化法 21
3.4.2灰階切片結合視窗檢測法 27
3.4.3檢測結果分析 32
第四章 實驗結果 34
4.1系統架構 34
4.2實驗環境及實驗參數設定 35
4.2.1光源設定 36
4.2.2鏡頭與取像系統之設定 39
4.2.3解析度訂定 40
4.2.4實驗參數設定 41
4.3單顆鋼珠之實驗結果 42
4.3.1灰階對比度擴展法檢測結果 42
4.3.2灰階切片法檢測結果 49
4.4實驗結果討論 55
4.5一影像中包含多顆鋼珠之檢測結果 68
第五章 結論與建議 80
參考文獻 82
附錄A 影像未做強化之結果比較 84
附錄B 瑕疵檢測結果疊回原圖 86
附錄C 不同光源亮度之檢測結果 91
附錄D 程式說明 96

1. 進口救濟相關案例資料庫，鋼珠鋼礫最後調查報告，http://moeaitc.tier.org.tw/itc/NewDoc.nsf/0/483152237ace96e6482567b7000aca1e?OpenDocument&ExpandSection=3
2. 台灣金圓工業股份有限公司，鋼珠製造程序，http://www.nisuma.com/processC.htm
3. 天工鋼珠股份有限公司，鋼珠製造程序，http://www.tankong.com/en_flow.htm
4. Kang, J., M. Hadfield and R. Ahmed, 2003, “The effects of material combination and surface roughness in lubricated silicon nitride/steel rolling contact fatigue,” Materials and Design, Vol.24, pp.1-13.
5. Svahn, F., A. Kassman-Rudophi and E. Wallen, 2003, “The influence of surface roughness on friction and wear of machine element coatings,” Wear, Vol.254, pp.1092-1098.
6. O’Brien, M. J., N. Presser, E. Y. Robinson, 2003, “Failure analysis of three Si3N4 balls used in hybrid bearings,” Engineering Failure Analysis, Vol. 10, pp.453-473.
7. Schofer, J., P. Rehbein, U. Stolz, D. Lohe and K. —H. Zum Gahr, 2001, “Formation of tribochemical films and white layers on self-mated bearing steel surfaces in boundary lubricated sliding contact,” Wear, Vol. 248, pp.7-15.
8. Wang, L., R. J. K. Wood, T. J. Harvey, S. Morris, H. E. G. Powrie and I. Care, 2003, “Wear Performance of Oil Lubricated Silicon Nitride Sliding Against Various Bearing Steels, ” Wear, Vol. 255, pp.657-668.
9. Garcia, M., D. F. Garcia, J. L. Diaz and F. Suarez, 1994, “Flatness Defects Detection in Rolling Products with Real-Time Vision System,” Intelligent Systems Engineering, Second International Conference, pp.407-412.
10. Zoughi, R., C. Huber, N. Qaddoumi, E. Ranu, V. Otashevich, R. Mirshahi, S. Ganchev and T. Johnson, 1997, “Real-Time and On-line Near-Field Microwave Inspection of Surface Defects in Rolled Steel,” Asia Pacific Microwave Conference, Vol.3, pp.1081-1084.
11. Kakimoto, A., 1996, “Detection of Surface Defects on Steel Ball Bearings in Production Process using a Capacitive Sensor,” Measurement, Vol.17, pp.51-57.
12. 孫小龍，民國88年，“球研磨過程之運動特性研究”，碩士論文，國立中山大學機械工程研究所。
13. Kapur, J., P. Sahoo, and A. Wang, 1985, “A-new method for gray-level picture thresholding using the entropy of the histogram,” Comput. Vision, Graphics, Image Processing, Vol.29, pp.273-285.
14. Gonzalez, R. C. and R. E. Woods, 1992, Digital Image Processing, Addison-Wesley, New York, NY.
15. シーシーエス株式会社，CCS Inc.，光源介紹，http://www.ccs-inc.co.jp/cgi-bin/hp.cgi?menu=102-104-02-01e