An Expert System to Analyze Homogeneity in Fuel Element Plates for Research Reactors

In the manufacturing control of Fuel Element Plates for Research Reactors, one of the problems to be addressed is how to determine the U-density homogeneity in a fuel plate and how to obtain qualitative and quantitative information in order to establish acceptance or rejection criteria for such, as well as carrying out the quality follow-up. This paper is aimed at developing computing software which implements an Unsupervised Competitive Learning Neural Network for the acknowledgment of regions belonging to a digitalized gray scale image. This program is applied to x-ray images. These images are generated when the x-ray beams go through a fuel plate of approximately 60 cm x 8 cm x 0.1 cm thick. A Nuclear Fuel Element for Research Reactors usually consists of 18 to 22 of these plates, positioned in parallel, in an arrangement of 8 x 7 cm. Carrying out the inspection of the digitalized x-ray image, the neural network detects regions with different luminous densities corresponding to U-densities in the fuel plate. This is used in quality control to detect failures and verify acceptance criteria depending on the homogeneity of the plate. This modality of inspection is important as it allows the performance of non-destructive measurements and the automatic generation of the map of U-relative densities of the fuel plate