Study Of The System For Welding Seam Defect Detection For Canisters Based On Computer Vision

Welding plays an important role in detecting the quality of canister. The traditionaldetection of manual welding lack not only efficiency but also stability. Recently, computervision has been utilized extensively in industrial detection. The automatic NDT system basedon computer vision can both avoid the shortcomings of manual detection and the send thedetection information to technicians for improvement. However, due to the diversity of weldingseam defects and some interfering factors, the automatic extraction and detection of weldingseam defect still remain difficult.In response to the requirements of enterprises, this paper delves into the various problemsin welding seam defect based on computer vision and studies the detecting system. By thepreprocessing over the original pictures collected by the camera, this system extracts the corearea of welding seam and carries out the defect detection in the forms of curve detectingalgorithm. Specifically it includes:Firstly, a hardware structure targeted at the requirements of the online system is designedand a software structure is established. The software covers the extraction of ROI and defectdetection. In addition, according to the features and causes, defects fall into four types: slag,perforation, lack of penetration and fusion welding.Secondly, the preprocessing on the original images. Rotate the images by graphicalgorithm. Get the horizontal position of welding seam by image centroid algorithm and thenrotate it. It will reduce most of the interfering factors after the extraction of ROI.Finally, based on the preprocessing over the original images, we proposed the curvedetecting algorithm for defect detection targeted at the features of defect types. By standarddeviation and first difference, this paper removes the interfering information and detects onends of seam. The experiment results show that, curve detecting algorithm has reached anaccuracy of over95%.Moreover, identifying and counting of the defects types can help enterprises improverelevant techniques while curve detecting algorithm cannot realize the identification of thedefects types. Therefore, this paper details in identifying the defects types in the chapter ofbackground difference algorithm. By background reconstruction algorithm based on possibilitythe background model is constructed. Then, the defects can be extracted by difference operation.By tree-classifier, the defects will be classified.