Machine Vision Application For Automatic Defect Segmentation In Weld Radiographs

• Main Author: Soo , Say Leong
• Format: Thesis
• Language: English
• Published: 2006
• Subjects: TJ1-1570 Mechanical engineering and machinery
• Online Access: http://eprints.usm.my/31360/1/SOO_SAY_LEONG.pdf; http://eprints.usm.my/31360/
• Institution: Universiti Sains Malaysia

Objektif penyelidikan ini adalah untuk membangunkan satu kaedah peruasan kecacatan kimpalan automatik yang boleh meruas pelbagai jenis kecacatan kimpalan yang wujud dalam imej radiografi kimpalan. Kaedah segmentasi kecacatan automatik yang dibangunkan terdir:i daripada tiga algoritma utama, iaitu algoritma penyingkiran label, algoritma pengenalpastian bahagian kimpalan dan algoritma segmentasi kecacatan kimpalan. Algoritma penyingkiran label dibangunkan untuk mengenalpasti dan menyingkirkan label yang terdapat pada imej radiograf kimpalan secara automatik, sebelum algoritma pengenalpastian bahagian kimpalan dan algortima segmentasi kecacatan diaplikasikan ke atas imej radiografi. Satu algoritma pengenalpastian bahagian kimpalan juga dibangunkan dengan tujuan mengenalpasti bahagian kimpalan dalam imej radiogaf secara automatik dengan menggunakan profil keamatan yang diperoleh daripada imej radiografi. The objective of the research is to develop an automatic weld defect segmentation methodology to segment different types of defects in radiographic images of welds. The segmentation methodology consists of three main algorithms. namely label removal algorithm. weld extraction algorithm and defect segmentation algorithm. The label removal algorithm was developed to detect and remove labels that are printed on weld radiographs automatically before weld extraction algorithm and defect detection algorithm are applied. The weld extraction algorithm was developed to locate and extract welds automatically from the intensity profiles taken across the image by using graphical analysis. This algorithm was able to extract weld from a radiograph regardless of whether the intensity profile is Gaussian or otherwise. This method is an improvement compared to the previous weld extraction methods which are limited to weld image with Gaussian intensity profiles. Finally. a defect segmentation algorithm was developed to segment the defects automatically from the image using background subtraction and rank leveling method.