| This research presents a system to automate the inspection process of the nuclear fuel pellet before insertion to the fuel rod. To build such a system requires combination of various disciplines such as nuclear engineering, image processing, machine vision, fuzzy logic, data mining, and pattern recognition.;This system acquires the pellet image on site, improves the image quality by image processing, detects the defects, extracts the features of the defect, and classifies the defect patterns. During the process, two dynamic references, namely simulated reference and mean reference, are created to detect the defects. Simulated reference is based on the image irradiance equation to fit the brightness curve of the pellet image by using Phone's model, while construction of the mean reference is from the statistical point of view. Feature extraction improves the efficiency of the classifier and also keeps its high accuracy. Decision tree and fuzzy logic are two methods used to construct the classifier. Therefore, two types of classifiers are developed in this work. One is the rule-based classifier that is based on decision tree, and the other is the fuzzy logic classifier.;The rule-based and fuzzy logic classifiers perform very well for pellet inspection. This system reaches 100% accuracy in classification. The implementation of the system is through a software package designed with the object-oriented method using JAVA language. |
