| Inspection of nuclear fuel pellets is a complex and time-consuming process. At present, quality control in the fuel fabrication field mainly relies on human manual inspection, which is essentially a judgement call. Considering the high quality requirement of fuel pellets in the nuclear industry, pellet inspection systems must have a high accuracy rate in addition to a high inspection speed. Furthermore, any inspection process should have a low rejection rate of good pellets from the manufacturer point of view.;It is very difficult to use traditional techniques, such as simple image comparison, to adequately perform the inspection process of the nuclear fuel pellet. Knowledge-based inspection and a defect-recognition algorithm, which maps the human inspection knowledge, is more robust and effective. A novel method is introduced here for pellet image processing. Three artificial intelligence techniques are studied and applied for fuel pellet inspection in this research. They are an artificial neural network, fuzzy logic, and the decision tree method. A dynamic reference model is located on each input fuel pellet image. Then, those pixels that belong to the abnormal defect are enhanced with high speed and high accuracy. Next, the content-based features for the defect are extracted from those abno1mal pixels and used in the inspection algorithm. Finally, an automated inspection prototype system---Visual Inspection Studio---which combines machine vision and these three AI techniques, is developed and tested. The experimental results indicate a very successful system with a high potential for on-line automatic inspection process. |
