Machine vision has been widely used for automated inspection of defects in industrial parts. In this thesis, a new physics based machine vision method is proposed to detect defects on CRT glass panel.
Defects on CRT glass panel can be classified into four groups, check, scratch, blister and mark, by their physical characteristics. From those physical characteristics such as reflection and refraction of light, a geometric optical model of each group is developed for defect detection. Check type defects can be modeled by an interreflection with Torrance-Sparrow model. Mark type defects can be modeled by a concave or convex lens.
Furthermore, several illumination methods and configurations of an illumination and a camera are developed for an inspection system based on the proposed geometric optical models of defects. For inspecting check type defects, bar-type diffuse illumination method is proposed and it makes system configuration more flexible. A slit-type illumination method is proposed for detecting mark type defects.
Finally the proposed geometric optical models of defects are validated by experiments. Experimental results of detecting various defects on CRT glass panel show the performance of developed inspection methods and confirm the possibility of their applications to CRT manufacturing processes.