Master of Science in Industrial Engineering (MSIE)
The process of joining metals is called welding. At times, selecting a poor quality material or improper usage of welding technologies may cause defects in welded joints. Some of these welded joints have to be tested nondestructively, because their failure can cause lot of damage, for instance in power plants. Radiography is a very common method for non-destructive testing of welds. It is done by certified weld inspectors who have knowledge about weld flaws, looking at the radiograph of the welded joint with naked eye. The judgment of the weld inspector can be biased; subjective, because it is dependent on his/her experience. This manual method can also become very time consuming. Many researches were exploring computer aided examination of radiographic images in early 1990’s. With much advancement in computer vision and image processing technologies, they are being used to find more effective ways of automatic weld inspection. These days, fuzzy based methods are being widely used in this area too. The first step in automatic weld inspection is to locate the welds or find a Region of Interest (ROI) in the radiographic image . In this thesis, a Standard Hough Transform (SHT) based methodology is developed for weld extraction. Firstly, we have done binarization of image to remove the background and non-welds. For binarization, optimal binary threshold is found by a metaheuristic –Simulated annealing. Secondly, we use SHT to generate the Hough Transform matrix of all non-zero points in the binary image. Thirdly, we have explored two different paths to find a meaningful set of lines in the binarized image that are welds. Finally, these lines are verified as weld using a weld-peak detection procedure. Weld-peak detection is also helpful to remove any non-welds that were remaining. We have used 25 digitized radiographic images containing 100 welds to test the method in terms of true detection and false alarm rate.
Document Availability at the Time of Submission
Release the entire work immediately for access worldwide.
Gunaseelan, Rakesh, "A study of hough transform for weld extraction" (2010). LSU Master's Theses. 2167.
Liao, T Warren