Purpose: To detect and identify all kinds of visible defection on floating glass production line in real time, such as
bubbles, tin ash, impurities, stone, craterlet, thread, distortion, etc., which is extremely important for glass process
management, grading and stowing. Methods: A set of machine vision system is used to detect the defects. In this system,
a novel illumination technique and sensor unit based on time-variale LED raster is developed to obtain both the
distortion and deformation features together. The resulting defects are determined by a RLE-based image processing
algorithm and transferred to subsequent marking or cutting devices. Results: Real experiments illustrated the stability
and effectiveness of this system, by which most of the main defects are inspected at real time under the speed of 30
m/min. With 5 m glass width, the inspection precisions are 0.1 mm/pixel both in direction of width and length.
Applications verify the speed, reliability and accuracy of the proposed method. Conclusions: Quality inspection of
floating glass at real time requires multiple linear cameras to construct distributed data processing system. Also material
characters of the glass should be stressed to design proper optical structure, so that the glass defects will be inspected
successfully. Using this system, the quality of floating glass can be improved effectively.