- 學位論文
低對比及非均勻背景下玻璃基板瑕疵氣泡之自動化光學檢測
Under Low Contrast and inhomogeneous background Glass Substrate's Defect Bubble Using Automated Optical Inspection
摘要
摘 要 薄型平面顯示器在愈來愈朝向大尺寸及細緻化的兩極需求之下，其製程便面臨更難以掌控品質及良率的局面；然而，尺寸愈大則製造過程中發生瑕疵的機率就愈高；品質要求愈精細則瑕疵的容許度就愈低。因此所需要各製程的瑕疵自動化檢測能力也就隨之提高，同時更需要高速及影像解析能同時提升，但往往這兩項的要求卻是互相背離的，因為解析能愈高則檢測範圍就愈小，會導致整體檢測時間隨之增加；反之，要縮減整體檢測時間，就必須降低影像檢測的解析能。 製造一個完整的薄型平面顯示器所需要的材料零組件及製程是非常眾多且精密繁瑣的，而其中一項關鍵性材料–玻璃基板，是需要較高技術門檻的一項零組件材料，因為需要符合其基本特性要求之外，特別是要消除玻璃體中的不純物、不均質以及氣泡。 自動化光學檢測技術在工業界的應用已經非常的廣泛，諸如印刷電路板的孔徑及線寬線距量測、表面點膠均勻度檢測、自動影像選邊送料機、金屬表面電鍍均勻檢測等等。一般的光學檢測方法皆是利用其高對比的邊界且在均勻的背景下進行影像處理即可以計算或分析出邊界座標或相對距離等特徵，若影像的表現在非均勻的背景且特徵是低對比的情況下，則其上述的檢測方法則無法有效進行檢測或量測的。 本論文將提出一種影像辨識演算法，針對低對比且非均勻背景下玻璃基板瑕疵氣泡的檢測，以有效提升玻璃基板瑕疵氣泡的檢出率。並藉由本研究發展出一套可以安裝至一般倍率的工具顯微鏡下使用的影像自動化光學檢測暨辨識系統，以符合玻璃基板製造廠及薄型平面顯示器製造廠的需求。
並列摘要
Abstract Due to the fact that the sizes of flat panel displays (FPDs) are getting larger, the manufacturing processes are getting more complicated. Accordingly, it is more difficult to control the product yield. Consequently, the possibility that defects occur in the manufacturing is larger. To deal with this critical problem, it is necessary to develop automatic defect inspection systems which can not only detect defects from high-resolution images but also accomplish the detection task in a short time. However, the above two requirements cannot be fulfilled at the same time in general, because the higher the resolution of an image, the more the time needed to process the image. The FPD manufacturing consists of many complicated fabrication processes, and need various materials. One of the most critical materials is glass substrate. However, in some cases, the glass substrates would contain different kinds of defects, such as particles, non-homogeneous regions, and bubbles, which will makes the substrates useless, increasing the production cost. In recent years, automation optical inspection (AOI) has been widely adopted in industrial applications. Common AOI methods are designed to detect the defects which have significant contrast with respect to background. By using simple image processing techniques, the defects can be detected from the images, as well as their boundary coordinates and features. However, if the contrast between detects and their backgrounds is not high enough, then the usual AOI methods will be unable to accomplish the defect detection task effectively and efficiently. Unfortunately, the contrast between bubble defects and glass substrates is very low, which makes the defect detection problem intractable. In this thesis, a novel image recognition algorithm is proposed to deal with the critical problem mentioned above. This algorithm is able to detect bubble defects from non-homogeneous and low-contrast backgrounds. In addition, an image-based AOI system containing an automatic positioning motor system is also developed in this thesis. The system is designed based on the requirements from a real FPD manufacturer in Taiwan. More importantly, the developed AOI system has been successfully implemented into real FPD manufacturing line.