Técnicas automáticas para detecção de cordões de solda e defeitos de soldagem em imagens radiográficas industriais

Abstract

This work proposes a new method for the automatic detection of the weld seam in radiographic images of pipeline welded joints. The proposed methodology is based on the optimization of parameters that are used to control and modify the position, size and shape of an image window, in order to enclose the image region that best matches the radiographic representation of a reinforced weld seam. The search for the best matching is managed by a genetic algorithm and uses an image similarity concept that is commonly applied on template matching procedures. The proposed technique solved weld seam detection problems for which no other automatic detection method was successful. The test results achieved an accuracy between 93% and 100%, regarding different tests circumstances, for a total of 478 radiographic images. The images data set included radiograph samples that cover all the recommended radiographic exposure techniques that are recommended, in the ASME V code, for inspection of pipeline welded joints. Meanwhile, a known image segmentation technique was modified and applied to perform the automatic detection of welding defects. For that test series, radiographic patterns from the International Institute of Welding (IIW) were used, including samples of the most popular classes of welding defects. After the used segmentation technique has been properly modified, defect detection samples were achieved for all the tested defect classes. Such results contribute with advancements in the automatic analysis of industrial radiographs and, as a final goal, aim at aggregating quality and efficiency to the radiographic inspection of welded joints.