FEATURE-EXTRACTION FOR ARTIFICIAL NEURAL-NETWORK APPLICATION TO FABRICATED NUCLEAR-FUEL PELLET INSPECTION

Nuclear fuel must be of high qualify before being placed into service in a reactor. Nuclear fuel vendors currently use manual inspection for quality control of the nuclear fuel pellets before they are inserted into the zirconium fuel rods and bundled into assemblies. The feasibil ity of automating the pellet inspection process using artificial neura l networks is examined to improve accuracy, speed, and cost; to reduce employee radiation doses; and to provide defect statistics to the fue l manufacturer. Sample nuclear fuel pellets (252 pellets) are photogra phed and scanned, and appropriate feature extraction techniques are de veloped and applied to the scanned images. The extracted features are then used as inputs to a backpropagation neural network. The results i ndicate that a backpropagation neural network is capable of classifyin g pellets as good (passing) or bad (failing) with high accuracy.