Detecting Shape of Weld Defect Image on X-ray Film by Image Processing Applied Genetic Algorithm

Several types of non-destructive testing methods are used for detecting weld defects. Because the X-ray radiographic testing method is particularly useful in inspecting the inside of a weld metal, it is often used in industry. However, since the number of skilled inspectors for X-ray radiographic testing has been gradually decreasing, recently, several methods to detect weld defects from films automatically have been investigated to improve the quality of the detection results. However, X-ray film images contain much noise, and defect images show very low contrast and various shapes in spite of the same kind of defect. Moreover, boundaries between a defect image and the background are unclear, making it difficult to automate the inspection of X-ray films. If the type of defect image were to be judged by an expert system or a neural network which learns the rules of professional inspectors, the boundaries of the defect image would have to be detected in a manner similar to recognition by a human's (or an inspector's) sense of vision. Therefore, in this study, a new image processing method applied genetic algorithms that were a method of optimization, was constructed and applied to the detection of defect boundaries in detail.