Intelligent recognition of weld defects became more and more important in current large-scale production. Besides incomplete seam caused by deficient clamping pressure, some surface defects, such as pits, pores and convexes on circumferential seam caused by instable welding process were fatal for parts here. But the difference between normal lap spot and other defects were not obvious. An image collecting system based on linear CCD while keeping parts rotating was designed, which provided image with enough contrast and brightness. In order to recognize defects identically, an improved algorithm of image processing based on Edge Oriented Histogram (EOH) method. Two criterion parameters are proposed for on-line visual detection of concave pits. Normal lap spot and other defects showed difference in distribution characteristic of the two criterions.
展开▼
