Multichannel high-performance optoelectronic control of the surface of fuel pellets

Optoelectronic methods for controlling the appearance of nuclear reactor fuel pellets are considered. In the proposed methods, reflected images of pellet surfaces are taken by digital cameras, which provides high contrast of defective areas against the frame background. Image processing algorithms for identification of defective products are given. An experimental model of a system with a capacity of up to 10 pellets per second was designed. A database containing surface images of simulators and real pellets was designed to train the system. Results of real time processing of the obtained images show that the probability of detection of defective pellets is not less than 95%.